13th Annual IRS/TPC Joint Research Conference on Tax Administration

gorgeous okay connected I know um foreign foreign foreign [Music] good morning everyone we're about to get started if folks could take seats so good morning I am Wendy edelberg I am director of the Hamilton project at Brookings and it is my great honor and pleasure to have the task of welcoming you welcome you here on behalf of Brookings and I think gaveling in this conference I have long been a fan and consumer of the work coming out of the tax policy Center so this is a really a privilege and looking over today's agenda just fills me with optimism the work being discussed today reflects the efforts of scores of excellent researchers dedicating their experience to making government function better the day will be filled with actionable ideas and findings and recent legislative events notwithstanding there is clearly momentum and additional resources for the IRS to invest in better implementation even under current law so now with that I'm going to hand things over to Eric toter Institute fellow and co-director at the tax policy center before joining TPC Eric Ed stands at the office of tax analysis of Treasury the IRS and CBO and Eric is going to also take a few minutes to welcome you [Applause] well thank you Wendy and welcome everyone to the 13th annual IRS TPC research conference that's one thing that changes every year the the number for front of that it's really great to be back after four years of being online and to see so many old friends again um I'd like to thank the conference committee headed by Brett Collins and Alan Plumley at IRS and Rob McClellan at TPC for putting together this year a really great program and this is a very interesting time we're living through in tax policy and administration in addition to all the usual concerns we always have about the tax system and the need for reform we also have to face how to deal with the exploration of the individual income tax cuts at the end of in tcja at the end of 2025 and the ongoing expiration of some of the and increases in taxes on some of the business base items so um there's a there's a lot going on and of course on the enforcement side you've had this increase in the budget that makes up for you years of cuts hopefully that will be sustained there are some concerns about that so it is a very uh challenging time and I think it makes the research at this conference on compliance and how that can be done effectively even even more relevant than it than it usually is um today's program will address major questions in Tax Administration and that this is just some of them how can we reduce compliance costs and more effectively deliver tax benefits what are the sources of racial differences in audit rates um how changes in the quantity and do changes in the quantity and the composition of audits affect voluntary compliance uh how is the composition of taxpayers and their activities changed and how does that affect compliance and how much assets are hidden in foreign accounts and how we Ida do we identify them our keynote speaker today is uh Catherine Ron Powell an opinion columnist of the post who has written a lot about tax issues and tax compliance funding recently so I think you'll find her her comments uh interesting uh let me now turn the microphone over to uh Barry Johnson who is the Deputy head of um Ras and thankfully also the director of SLI and we appreciate all the help you've given to the tax community over the years foreign thanks Eric so Melanie Krauss the irs's chief data and analytics officer and the director of research applied analytics and statistics or as couldn't be here with us today so you've got me instead but she will be joining us online later today I know if she were here she would like to thank all of you for attending this 13th annual conference so I'll start by thanking our co-sponsor at the tax policy Center thank you Eric for your continued generosity in supporting Tax Administration related research this is such an important important performance form for IRS researchers for our peers and other agencies and from other tax administrators to come together to get feedback on our work and to learn from each other thanks also to Bill Gale and the Brookings Institution for sharing their beautiful facilities with us so that we can resume meeting in person after such a long hiatus before going any further I just want to note how proud I am to be associated with the really outstanding folks who make up Raz and those in the IRS research functions that are embedded in the operating divisions these talented staff daily undertake Innovative highly technical work that is already impacting Tax Administration activities across the IRS and globally through our work with the oecd tax form impacts included probably applying behavioral insights to improve notices developing helpful nudges and even helping to improve the voice props taxpayers receive when they call us staff are exploring uses of modern machine learning and automations to reduce the time staff spend on routine tasks and to better leverage all the information available to the IRS they've developed innovative ways of linking tax data to provide Fuller pictures of taxpayers experiences through a wide range of interactions this approach will help us to identify pain points and measure impact of steps taken to mitigate them staff are also making progress combating identity theft improving audit planning and selection methods improving estimates of the tax Gap and developing helpful statistics to help taxpayers and researchers better understand the full scope and impact of our tax system I could think of no better time to be coming together to discuss ways that data and analytics can impact Tax Administration than now has already been noted the recently released the the IRA invest investment and and the IRS strategic and operating plan details that have been released in the most recent plan uh show how we're going to leverage the historic investment um which will transform Tax Administration on behalf of of of of American taxpayers even a cursory review of that plan makes clear that data and analytics are going to be at the heart of a modernized effective taxpayer-focused IRS every initiative envisions a role for the type of research analytics you'll see today these include improving services to taxpayers who reach out to us for assistance new services to help taxpayers identify credits they may have overlooked designing new treatment streams to reduce time it takes to resolve an issue and developing new on selection mechanisms to Target the most egregious cases of non-compliance and reduce the number of no change audits it's exciting to see that so many of the operating plan priorities are featured on today's program I have the honor of leading one of the transformational initiatives in the Strategic operating plan a set of projects that recognizes the value to the IRS from partnering with academics to tackle difficult research questions one of those projects is revitalizing our joint statistical research program which is a program that supports that invites external research proposals to to benefit Tax Administration we're currently accepting applications so the OMB standard application portal that window will close on June 30th but we are committed to annual cultural proposals moving forward in addition the operating plan notes the value to the American public from safely using aggregated data to assist with external evidence building projects and we're working on several ways to increase our ability to meet those demands within the restrictions of the Internal Revenue code some of this work is done in partnership with the tax policy Center and I'm particularly grateful to Lynn Berman Claire Bowman and Claire Bowen and the whole team who are keeping that work moving forward and then finally I'll note that the additional Ira investment will have a significant impact on all the IRS research functions including my organization the statistics of income division we're actively hiring new data scientists social scientists statisticians computer Specialists and others to support the many transformational activities going on across the service so we would greatly appreciate any referrals anybody attending here today would like to send our way we do have some direct hiring Authority we are actively recruiting on college campuses and we have numerous job postings out on USA Jobs which is the federal government's hiring website so if you know anybody that's looking for a job please send them to us so it is a great time to be discussing Tax Administration research and we at the IRS are very grateful for the support and interest from you who are gathered here today and for the academic and research institutions who have been helping us move forward both during the lean times and and in the future without the work that you have done with us over the last few years we would not be where we are now especially given the tight resources over the last decade or so so thank you um so without further Ado I'm going to invite our speakers and our chair up to kick off the first session thank you for being here I'm looking forward to learning from everybody [Applause] I'll go upstairs there's limited space up there yep hello hi I'm Dina Opperman at the office of tax analysis in the U.S treasury and I'm here to introduce and moderate the first session which is service is our surname and we have four great papers about Tax Administration uh we've got some great data to share with you and lots to think about there and one paper that's basically more about the background about who's getting the tax credits and what they're experiences would be and which would be very necessary in order to make the right policy calls for things about that so um I want to introduce the four papers first we'll hear from Jan Jan Millard with a cast of thousands on using behavioral insights to improve taxpayer experience the authors there are all at IRS the next paper is how do we reduce tax costs and taxpayer burden for resolving balanced due accounts which will be presented by Howard Rossi who's also at IRS and has many colleagues on that paper as well third would be understanding yearly changes in family structure and income and the impact on tax credits with which will be presented by Nikita irie and that is a Brookings paper Brookings Urban paper and finally racial disparity in audit rates presented by Tom Hertz with his IRS colleagues and so the session will go uh that there'll be 20 minutes for the authors to present one at a time standing here then we'll have each of the discussions gets there 10 minutes and then we will all come back I will be here and you will be at your signed seat and we will take questions and uh we invite you at that time to to bring us questions so without further words uh Jan Mellark [Applause] well good morning it's such a fun way to start the day right I gotta set the scene for everybody so I get to do the fun uh and I call this fun because this isn't work to me this is actually something I'm very passionate about so what I am going to talk about today is we're going to talk about looking at behavioral insights right how we can begin to improve on the taxpayers experience so make sure I'm doing this right yes I saw you okay so I have done numerous papers over the years I've presented a couple here at some of these conferences in the past using behavioral insights to try and the big word is nudge right but really what we're trying to do is improve our correspondence to taxpayers to say hey this is what we really want you to do in an easy way of doing it so we thought hey here's a great experience it's not Her Majesty's Revenue customers I think it's now his right his majesties but they presented a paper and they talked to us a little bit about what they did is they took the voice props when people call in and they said we kind of did a little bit of that nudge idea in there and we saw some great results so we said hey here's an opportunity to expand our research so we decided to look at our voice portals so when taxpayers call in we call it a voice portal because what they do is they call in and depending on the choices they make right so they pick one to go here pick two to go here they put them in what we call applications so those applications allow them to sit now let's be honest how many of us have ever had to call into the IRS anybody even me and I work for the IRS so um if you're anything like me I'm a prepared individual I get out my coffee I make sure my phone is plugged in and I make sure I'm multitasking as I'm sitting on hold um it's not on purpose it's just because we have a very large customer base that we have to deal with and we have a very small Staffing to deal with them but we decided let's take a look at one particular application we call this application 75. application 75 is basically like the balance due where taxpayers really need to get into us to talk so we said okay what can we start to look at when the taxpayer calls in they get into a holding position and in that holding position they get to hear five repeatable messages but they actually don't repeat they're just five messages that are heard in 30 right there's like a 30 second increment between each once that fifth message is gone they're going to be on hold for the next you know 80 90 120 minutes I'm kidding it's not that bad so we said okay let's think about what is a way since we know that they're only going to hear five messages how can we position these messages in a way that's going to make sense to the taxpayer so we took the five messages and currently one message if you look at it the first message is hey make sure you're prepared so when your call is answered okay well what does prepared mean make sure I'm prepared it's like I don't want to be prepared I just want to talk to somebody right that's that's the mentality the second original one was like oh here are a few online options but they were not very clear the third original message is hey go to irs.gov and use the Searchers right search search well okay what am I searching for because I don't even know what I'm trying to do and then last but not least it's just that one more prompt to say hey again go to irs.gov so we went to the call sites and we said what are some of the calls that you get that are more frustrating for you and the taxpayer one of the major calls was okay people call in and they just want to make a payment over the phone guess what we do not accept payments over the phone it's not in our purview so they waited for a good hour and a half just to be told you have to go into irs.gov and look at our sort and then you had to pay that way so we said let's start to look at these messages in a unique order that's going to bring the taxpayers what they need right up front instead of making them wait so we redesigned the first message to say did you know that we do not accept payments over the phone however you can go to irs.gov and then we put that backslash payments we directed them to the exact location on irs.gov they would have to go in order to present that the second message we said hey let's take a look at what is it the taxpayers want to do most of the time they want to set up an installment agreement so we said let's let them know if you go on to irs.gov online payment agreements did you know you could save money because basically if they go in and input the installment agreement themselves or set it up themselves and using our online tools they'll save money versus if the person on the phone does it for them it costs them a little bit more so not only we encourage them to where they should go to resolve it with our channels then so on and so forth then we kind of get into the hey use your online account you don't know what's going on take a look at your account on online accounts on irs.gov right so we fastened our messages to kind of adhere to that that flow what is it the taxpayers want to do how do we want to start to filter out those people that really can handle using our online resources versus anything else so that's how we decided to do that wonderful process now we went in with the goal of okay what is it that we really want to do here well we want to increase Channel shifting Channel shifting means it's like those taxpayers that are tech savvy and especially don't want to really talk to anybody they just want to take an online activity so we wanted to increase that channel shifting to say there are other options other than the phone we wanted to increase the use of online services over the years we have developed a lot of unique tools now that are available on irs.gov that allow taxpayers and the customers to you know self-service so we wanted to improve on that process as well and then also we wanted to like save on the resources of allocation what that meant is there are taxpayers that truly do need to talk to us they have a reason there's very there's some complexity behind it so we want to kind of open up the lines so that they get to us faster so those were the goals that we had in mind when we decided to do this project foreign so here's the fun part how do you actually do a pilot and then take all of that information in and like analyze it right so we had to kind of structure it in a way where we looked at the data because as we started to look at the data there's like three specific groups that we saw a pattern in the first group is is those callers that called in and they heard the message at least one message when they you know got through they heard one message and that was one unique and they only heard one message the other group was the taxpayers who called in and maybe they didn't get to the messages like they just kind of you know hung up and then they said oh I'll call back because I was in the right location and then they finally did hear at least one message and then of course we have our last group of callers which they call multiple times and within those multiple times they've heard at least one or more of the announcements in in the realm something to keep in mind is we kind of we have those outliers one of the outliers we had is those taxpayers who called in on multiple days so they could have heard both the control and then also they could have heard the redesign so we kind of pulled those out of the database just to make sure that as we were looking at the data it was a little bit cleaner so that we saw some statistical you know significance in the data so as you can see just looking at the um population as a whole we saw at least a 12 percent increase in Channel shifting on the redesigned prompts versus the control prompts and when we say Channel shift they like hung up and they took an activity somewhere else online or they resolved their issue without even having to talk to a caller we broke down the analysis you know in those specific groups so that you could kind of get an idea how they perform but with the whole population a 12 increase in in providing customer service I think is exciting it's exciting to me another thing that we noticed in the data is that you're looking at um the access among callers write message increased online access and we say online access it's not just oh they're going to set up an installment agreement they actually went online and did various things you have online you have direct pay you have the credit card payments you also have what we call online payment agreements so there's multiple areas online and we noticed that in that whole sample based on just using that hey going to irs.gov and giving them the various options we saw that with the whole sampling population we saw at least an eight percent increase in Channel shifting with the redesigned messages than we did with the control messages that were currently there another fun fact is that Opa we really wanted to track how many taxpayers are really jumping off or Channel shifting to go on and set up their own installment agreement the reason why we thought this one was really important as you can see there was like almost a 27 increase in the the redesign versus the control and the reason why that is significant is because we stress the importance of it's going to save them money and it will save them money in the long run because if if they go on and set up their own installment agreements and they can do that it can save the taxpayer an average of 76 to 95 dollars just by handling or doing it on their own so we were kind of excited about these numbers where taxpayers are going online and setting up installment agreements and actually using the products that we've started to develop for their use okay let's talk about you know among the callers abandoning EQ so a large portion of these redesigned messages um we saw that there was an abandon after the second message so remember I said the first message is kind of we want to filter out those taxpayers who just want to make a payment over the phone we know they can't make a payment so of course we saw you know a significant as far as the redesign over the control of people actually Channel shifting where it really gets interesting is you see the larger uh people starting in the second one the second message was the one that said hey you know you can save money by going online and setting up an installment agreement so um we we see that we really start to see a significant Channel shifting within that second population as we get further down the line we asked ourselves it's like are they Channel shifting because of the messages as we go further down the line or is it because there's only that 30 second gap between each one and they were still busy writing down what it said and then they're they're jumping off so again the data right now it's it just it's it's gets exciting for me because it's like ooh here's another Avenue that I can start to explore where we can even improve more on our customer service and how we start to introduce these messages within that queue process okay so as the um as the callers on the redesign were more likely to abandon their call spent time waiting to connect right so the interesting thing about this as you see the taxpayers who are starting to abandon the calls and do the channel shifting that really began to open up a little bit more space in time for those taxpayers who really wanted to talk to a customer service representative and I think that's probably the most exciting part is like even five minutes may not seem a lot to you but for the person that's on hold it it's a lot it's like oh Hallelujah I got him faster and I thought I do see that it is saving um we are saving some time the more and that less leads us to believe the more we start to even look at these redesigns and start to Fashion them even better using those behavioral insights right how does the taxpayer goal then we can start to see even more decreases in in wait time for taxpayers who actually want to talk to the the customers so what does all this mean right we go we spend all this time we do all this studies we do all this research so what does that really mean it allows us to start to take a look at truly what is happening to the taxpayer right what is that taxpayer's Journey why did they call us what was the purpose of calling it is it because they just decided one day it's like a just for you know Giggles I want to get on the phone and see what the IRS has to say no most of the time it's because they've been introduced to us in like oh I know I owe maybe I should call and find out what I should do or they receive multiple notices right or they receive notices that they don't even understand so that's usually the taxpayer's journey so we wanted to take you know really start to understand why it is so that we can refine our data and our analysis one thing that we learned as we started to really delve into that data right why did they call us what caused that call to prompt that we were looking at our analysis um it was interesting to see that over 60 percent of the pilot taxpayers were sent to at least were sent at least one notice within 30 days of the time that they called us so 60 got at least one notice so then that makes you ask why did the other 40 call right are they like me where they go on a trip for a month and they never go to their mailbox and then they come home and they see the big stack of letters that's another question that we wanted to ask ourselves one of the largest contributors to the phone calls in our um in our inventory that we were looking at they received at least that final notice right so there's this final balance due notice it's a little bit of a harsher language um and then the interesting the second most uh notice that was sent out would be the very first notice of balance due right so it's like okay the last one was the largest inventory but the first one was a large inventory too so um that kind of like you hit him the first time of course they're going to call you hit them the last time of course they're going to call what's happening in between right those are the questions that I like to ask myself so when we were looking at these multiple notices right so we asked ourselves more than 50 of these taxpayers that were issued multiple notices prior to calling remained in the queue wouldn't you I mean if you had like five notices from the IRS wouldn't you wonder why did I get five notices so you might be more likely just to stay on and talk to the customer service right so um we also thought that's another interesting Avenue of information that we can start to pursue it's like why are multiple notices going out so it not only leads us to improve on the process of the phones but it also leads us to start thinking about how can we improve on our correspondence and how we communicate with our customer base so then that leads us to the next question everyone always says that the level of service that the IRS has is really low okay well here's the formula that we use and you as you can see in the formula what is kind of bad toward us as far as a level of service are those abandoned calls will those abandoned calls are what we consider those Channel shifters so really should we be you know asking ourselves should we really be dinging us if we're providing them with an opportunity to self-correct are we not providing a service to that taxpayer so we kind of looked at this analysis and said you know there are some limitations to how we currently look at at level service that we provide to the taxpayers and what's another approach maybe that we can start to think about when we look now as we're designing these new voice prompts and we're encouraging taxpayers to self-correct and we're developing all these new tools that allow taxpayers to do online application process buses so we started looking at different types of ways that we can really improve on our level of service right so you have the level of access right this kind of measures this proportion of calls received um the average speed to answer that's another way again we can look how long did it take us because that's a common one that it should always be considered um the other one is the first Contact of resolution right so did we help them when they first contacted us and then of course effort to serve effort to serve is kind of where you start to get that taxpayer effort and that effort to serve it's like the taxpayer they're taking their own action but we have provided them the ways to do it and the tools that they need to do it as well as if we're looking at our level of service are we providing the tools and are we providing the appropriate avenues for them to use those tools so that just leads us to this great opportunity if we were to take our database that and all the information and apply to some of these you know um unique ways of looking at level of service you can see that if we took the estimated taxpayer efforts we do see that there is a decrease in uh so our level of service there's a lift right we're we're servicing them better than if we just use the plain old level of service as well as the same with the estimated effort to serve and I know that's a lot but there's a lot of information here but it's really interesting and fascinating to me but all of this that we do we can start to build upon that so again we hope to apply more principles of behavior and enhancing how our customers and how we can serve our customers better um we want to continue to develop we want to do even more testing in this field to see what we learn so the more we learn about that taxpayer's Journey the better we can start to refine our information toward it and then of course um we also want to play a lot around a little bit more with that the metrics um so that we can get her a better picture of how we truly are servicing the taxpayer and the various avenues that we begin to service them in the future well thank you guys [Applause] hello my name is Howard racy and I work for wni strategies and solutions I think this is the forward yes and today I'm going to be talking about the balanced due taxpayer really with the goal of reducing IRS costs and taxpayer burden and when we talk about balance to taxpayers we're talking about people that owe at the time of filing and really we think about filing over all there's really kind of three outcomes we have people that have an even balance so we they don't know us and we don't know them and we have people with the balance due that owe something or people that get a refund so the vast majority and today I'm talking about individual 1040 filers the vast majority of taxpayers 80 of this group they don't have a balance due um but within that population the 20 that do have a balance due again you have 13 of those folks that pay it they either pay all of it or they do the installment agreement but they resolve that issue but then we have these seven percent of people that receive a notice and of those people about 87 percent of those people receive the cp-14 which is that first balance due notice you owe us money you didn't pay you did file but you didn't pay so it costs the IRS a significant amount of money to handle balance due notices and specifically the cp-14 now every year we send out about 7.5 million so this is 7.5 million taxpayers that are not paying at the time of filing or not paying full and so at a cost about 51 cents per notice these numbers start to add up and so then outcomes when we think of Downstream costs here is really what we're talking about we have people that pay once they receive that notice and kind of this Jan was alluding to that first the first time we they see something they do pay or set up an installment agreement installment agreement still costs money because you have and you have monthly reminders there is a cost also um of setting that up in terms of a customer service representative then we have those people that ignore us and then we send them depending on the balance deal we most likely send them 501 a CP 501 we may send them a three based on the amount they owe but then here's our call folks and but Jim was just talking about all those people that call well it costs us about 65 million dollars a year just for the people for the cp-14 not including all of the other people to call for other notices they forget now some of the other outcomes here even when we think about costs per people that visit a taxpayer Assistance Center people that have a written response we didn't have good numbers on these it's hard to delineate out people that specifically went to taxpayer Assistance Center specifically for 14. but overall we're spending a lot of effort we're spending a lot of time and the taxpayer is spending a lot of time on the phone as we were pointing out so a lot of our intervention and again I keep looking at Jan because her group actually did a lot of the work here that I'm referencing here with a lot of our work goes towards fixing the notice fixing that cp-14 what can we do to get people to contact us to get people to pay and we do a lot of things and we've done a lot of great work around changing the format changing the fonts changing the language all with the idea of getting more people to pay but with all of this effort and kind of working on this idea it's the balance due has already occurred and so we're kind of remediating our reacting our to the balance do that's happened so what if we shift that and we put it before of what if we could prevent the balance due from occurring in the first place what if we could identify causes the balance do and then develop strategies to mitigate that well we kind of have a lot of good outcomes there we decrease the balance of notice we decrease Downstream costs that I was just talking about and in many ways when I've talked about this before it's in it's kind of like a hypertension analogy here it's like we can treat hypertension in one of two ways we can wait get hypertension and then take drugs and then change diets start to exercise stop smoking and all those things or we can start earlier and we can don't start smoking maintain a healthy weight exercise eat right so you avoid the issue in the first place so that's really what our research is going to talk about today so in order to do that we really need to identify populations so who are the folks that really are at risk for having a balance due and so what we're doing here is we're looking at balance do change and now I have we looked at tax year 16 to 17. so why do we look at 16 and 17. well this is the last year before the tax cut and jobs act changed withholding so that you doubled withholding so that the number of schedule a is significantly dropped you also had following that we had this little thing called Global pandemic that resulted in filing changes we had fewer schedule C's also you had differences in marriage and divorce rates during that time so either you couldn't get married because nobody was around and you couldn't travel or you couldn't get divorced not sure I think that would be the worst of the two um but really what we're looking here is there's three main categories here you have no change somebody had in 16 and 17 had the same occurrence they had a refund one year they had a refund the next you have a favorable shift so this would be a shift away from balance due or specifically a balance due with a cp-14 and then unfortunately you have the unfortunate shift people that are going from a balance due to a balance due with a cp-14 or going from a refund to a balance due with a cp-14 and this really this unfavorable shift is the folks that we wanted to really Target and so using chi-squared and kind of in Kramer's B to really kind of look at effect sizes we start to identify certain populations that are having these balance do these unfavorable balance do changes and the first one appears with our friends that couldn't get divorced and so when you get divorced now specifically we're defining the divorce as moving from married filing jointly to single filing status and here with that divorce we're seeing an unfavorable balance do change so people from 16 to 17 that got divorced moved to from either refund or just a balanced without a 14-2 a balance due with a CPU 14. with schedule a folks removed schedule a again moving towards an unfavorable balance to change and schedule C now Schedule C was interesting here because either adding schedule specifically here adding Schedule C and then our next analysis having Schedule C in general um led to this unfavorable balanced new change and then we also looked at some other schedules we looked at age and looked at positive total positive income and I point out income here because I'm going to talk about income later but with income you had both unfavorable and favorable so it kind of was was all over the place so then what we did we had these populations the three that we targeted here we targeted divorce we targeted removing schedule a and we targeted Schedule C and using a logistic regression we're really trying to look at risk how more likely is it that one of these populations is going to end up in a balanced due situation or an unfavorable balancing situation so with divorce we found that there is triple the risk compared to the standard taxpayer population of a divorced taxpayer moving into an unfavorable balanced new situation now for this group between 16 and 17 when we're talking about impact here there's really only 500 I say only but because I guess I shouldn't say only 578 000 individuals moved from this marriage filing uh jointly to single um and really that only accounts for about 7.8 percent of this favorable unfavorable change but when we moved when we moved to schedule a and schedule C a much bigger impact here so removing schedule a twice the risk so remove having it one year removing it the next year you have twice the risk nearly almost twice the risk of moving into unfavorable balance do and the same with Schedule C now Schedule C if you either had it in 16 and 17 or you added it you had double the risk so we looked at income we tried to look at income and so what we did we created a debt ratio where we took the balance due or the refund and then divided that by total positive income to get a percentage so it basically was your what was your refund and what was your balance due what was the percentage of your total positive income and not surprisingly it would it kind of turned out as you would expect that the more difference that you had there in terms of positive numbers the more likely or you're going to have a risk of an unfavorable shift and so here like if we look on this kind of this bout this this bottom line for 2017 you have the refund folks the refund even balance going to a balance due with a cp-14 and so you had a different tier of 5.67 percent so the greater difference between the amount that you owed and your total positive income the more likely you were to have a balance do that you couldn't that you received a cp14 with but the issue here and when we talk about kind of how do we address the issue income and debt is kind of hard to do it's like we can't like have a different I mean you can't tell people to have a different debt ratio or make less money or make more money or anything like that so when we think about how to prevent the balance due from occurring well first we start looking at what are those activities like so debt ratio doesn't really give us a good indication of how to do that but as we can with divorce well what happens with divorce you may lose dependence that you can claim you may have to withdraw 401K because lawyers are very expensive you may lose schedule a because you no longer have mortgage interest our mortgage deduction and you may have to add Schedule C and so again when we think about divorce it's kind of that Trifecta of horribleness where we already saw that taking schedule a away is an issue adding Schedule C can be an issue and so again but these are the things that can happen there removing schedule a it's usually around mortgage deductions and medical expenses that you're no longer able to take and then adding Schedule C was seeing so many more people with the gig economy and kind of the side hustles that they do and one of the things that we believe that happens with the schedule C is that if you are a wage in investment taxpayer where you're used to getting a W-2 specifically they're taking withholding out at the source however when you start driving for Uber or Lyft or deliveries or things like that as a taxpayer that may not know those things you don't know to kind of withhold or hold money back to account for that income and so at the when tax time comes around it's like oh I made all that income but nothing was withheld so then you end up in this situation here so one of the things that like if we know these things are happening can we inform taxpayers about these risks these Behavior changes that happen and does messaging really work well according to Raz and there are 21 comprehensive tax taxpayer attitude survey it does 86 percent in that survey said The more information and guidance the IRS provides the more likely people are to correctly file so we conducted a gap analysis we looked at what information is out there around divorce around schedule a and around the gig economy and so our specifically here yeah divorce starting small business in the big economy and we looked at three sources we looked at irs.gov we looked at Google and we looked at kind of in super in-person support now irs.gov is a has a great amount of information unfortunately it's finding that information as in a sense a novice taxpayer it's like you are not aware sometimes of the information that you're trying to find and so here I mean a lot of the Publications are very technical in nature and we know that we write in legalese we light in IRS speak as we all have our pages and pages of acronyms and so in a sense we have the information but being able to locate that information Google really as great as it is we use it all the time but unless you know the specific terms that you want to add like divorce and taxes are divorce and something or starting a business where they're not taking out withholding it's like well you're probably going to get an article then but otherwise not now surprisingly not so much with Google but when we talk about gig economy we also looked into like Lyft and Uber and their pages to see if there's anything in there about informing taxpayers of the impact of making that money without withholding not surprising there's no information there and we looked at in-person support and again it's it's links but we don't have that that in that good quality information so what we did we paired or partnered with the taxpayer experience office to address kind of we have these populations and for our purpose now it's like in a sense we've just done this and so we focused in on divorce the one that was three times the risk obviously we're not dealing with as many people that have issues with Schedule C but we're making interventions here for those folks and so the first thing we did with their help and working with communication and liaison we developed irs.gov divorce and this is a live page that currently exists out there as of about a month ago and so again this is a landing page for divorce taxpayers it has the one stop that can get all of the information that they would need that in many cases they don't know they need and that's kind of really this idea is like what you don't know often is what gets you in trouble and then again developing new material for that page like how not to owe taxes after divorce five things to know and then also developing this kind of external communication campaign and this is where we are right now in terms of the focus here on divorce this external campaign is like how else can we get this information out there working with communication liaison taxpayer experience office so what do those populations is it divorce attorneys is it a way to get the information to them to say hey I know you're getting divorced I know that's horrible but don't forget that you don't want to come around tax time and end up in having even a worse situation or maybe a better situation than you are right now so I want to give just when we talked before just about this divorce and chat and taxes checklist this is again kind of what it looks like in terms of this is information that's on that site it's like will your will your taxes get up I can't see from over here um yeah we'll be able to claim dependence those things remember the things that were in the box before about those things that happen can you claim this can you what's going to happen there did you withdraw from 401k which again it's a lot of people don't realize there's a penalty for 401k withdrawal as everybody knows but that counts as income in the end which is devastating because then you end up owing lots of money on that and so really kind of overarchingly what we're talking about today and what we've kind of would have tried to show here is this this idea of an evidence-based intervention where you're identifying an area of concern so people getting moving shifting into this unfavorable balance do and then using a series of research techniques or analytical techniques um like likely logistic regression where you're trying to determine the amount of risk like How likely is it for your population of interest to end up in this situation that we don't want them at because we don't want them to get we don't want them to have a balance due and we specifically don't want them to have a balance to do with a cp14 and then partnering with stakeholders who have that experience who have that knowledge and expertise to develop media campaigns develop websites and so forth thank you [Applause] hi everyone uh my name is Nikita I'm from the tax policy Center uh today I'll be presenting a paper I wrote with Elaine mogg and Lillian Hunter of the tax policy Center we're also grateful to Kevin Werner Joyce Morton and Laura Wheaton Who provided assistance and advice in creating our data sets and Performing micro simulation runs and thanks also to the Intuit Financial Freedom foundation for funding this work so our goal was to understand how families economic and household Arrangements change from year to year and what difficulties they or the IRS might face in predicting refundable tax credits based on prior year's tax information our analysis given the renewed interest in advanced tax credits following the 2021 CTC expansion can help inform the design of Advanced Tax Credit programs and Provisions to protect low-income families in the event that their eligibility changes from one year to the next and they end up having received an overpayment of their Advanced tax credits so for background is the earned income tax credit and the child tax credit together make up around 17 percent of low-income families income versus only three percent for families overall and so how families receive such a large chunk of their income how and when they receive it and matters a lot among low-income families with kids uh the eitc and ctcr are a relatively larger share of incomes for families with multiple children which makes sense because the credit increases with the number of children you have and then uh another interesting thing to note is for women heads of household this has less to do with the ATC and CTC but earnings are a relatively smaller share of their total income relative to other income which is largely dominated for this group by transfers so let's look at what determines uh someone's eitc amount and could therefore influence year-to-year swings and credits so families are first eligible for different maximum Credits based on how many eligible dependents they have so uh starting from Noel eligible dependence up to three or more for Simplicity we'll refer to these dependencies as children though there are older dependents who are eligible whether you receive the maximum credit a partial credit or none depends on your income and then those income thresholds vary for married versus unmarried filers and then some of you will have noticed that this graph shows eitc amounts from 2018 eitc amounts our indexed to inflation and it increase slightly every year so these are a little bit below what current amounts are and I'll go into the reason for 2018 in a little bit and then next for the CTC this graph shows uh child tax credits mount for head of household in other words an unmarried parent with a dependents the dark blue line shows the child tax credit amounts by income for a filer with one child up to a maximum of two thousand dollars uh unlike the eitc CTC recipients need to earn at least twenty five hundred dollars in order to be eligible for the credit in 2018 with additional earnings above that starting point of twenty five hundred dollars you could then qualify for up to fourteen hundred dollars per child as a tax refund and then those with enough uh income to owe taxes could qualify for up to two thousand dollars total per child uh including that fourteen hundred dollar friend refundable portion and then for reference I'm also showing uh the increased CDC amounts from 2021 and gray Above So 3 600 and uh three thousand dollars per child so in qualitative research with eitc and CTC recipients you'll see responses from some of them who are pretty Savvy about understanding what their total tax refund is likely to be and why and they plan around receiving it they plan their finances around it some understand how their income affects their credit and how to strategize like who claims which child in the case where they're overlapping households and caregivers um but more respondents particularly like low-income respondents on TANF who were interviewed in recent work by Urban Institute colleagues had very little understanding of what goes into determining their total tax refund uh and thought of the more as surprise bonuses rather than a reliable recurring source of income and that has to do with you know tax credits aren't the only thing that determine your total tax refund withholding can also go into it um and so low-income families also have more volatile incomes than families overall around 60 percent of them will have one month of the year that's uh significantly above their average income for the year and they also tend to have more ambiguous household Arrangements that can complicate their tax filing decisions such as uh on children living with multiple unmarried parents in the household so then you throw into the mix the question of whether to deliver tax credits all at once as a lump sum at tax time or in advance of tax time split up into periodic payments like monthly payments like we did in 2021 with the child tax credit research by Zachary paralyn and colleagues demonstrated that each type of receipt of each type of payment was associated with alleviating different Financial stressors so large one-time amounts helped families catch up with housing payments while the monthly recurring smaller amounts helped families afford food so there's benefits to both and receiving eitc refunds on the other hand has been shown to be associated with positive outcomes like deciding to enroll in college and go to the doctor while Advanced payments can smooth incomes month to month and helping families with low-income families with that income volatility we that we discussed and cover day-to-day expenses we also saw families surveyed by in the well-being and basic needs survey at the urban Institute reported preferring monthly payments to lump sums 45 to 27 percent so the question we wanted to answer was how accurately can advance payments be delivered and how can considering how families income and household changes might lead to overpayments so in order to do that we analyzed data from the CPS outgoing rotation group I know it says esek in which around a quarter of households in the CPS are interviewed twice a year apart they're interviewed for four months they cycle out for eight months and then they're interviewed for four months again and so then we have income data from those families from one year apart we use that to create a data set of households with kids in at least one of the two years in which they were interviewed uh and so we paired 2015 and 16 2016 and 17 and 2017 and 18. uh using that data set we estimated eitc and CTC amounts using the urban institute's trim micro simulation model uh applying 2018 tax law to all years with some caveats but the decision to use 2018 was motivated by that was the last year before uh or in order to minimize the effect of the tax credit tax cuts and jobs acts changes to the CTC but we did allow for example the ertc to inflate from one year to the next uh so those changes are reflected in here uh we counted changes in tax credits of at least five hundred dollars and then in the next few slides I'll focus on results for households with income under 200 of the federal poverty level in the first year observed all right so uh this graph breaks down the frequency of each eitc changes for all families with Inc and all families with income below twice poverty uh the blue bar the blue section of the bar represents increases the black section represents no change and the yellow section represents decreases with gradients within each color breaking down the magnitude or type of each change so we saw uh overall 69 of families saw no change in their eitc compared to only 39 percent of families under twice poverty but for many of the uh 69 with no change in the overall group that's because they just didn't qualify for an eitc in either year and so you'll see that's represented by the gray section of the bar labeled zero both years for families under twice poverty 39 saw their eitc decrease and 22 percent saw their eitc increase around half of all increases and decreases in this low-income group were changes of two thousand dollars or more we paid particular attention to decreases because this is the situation that would trigger needing to repay an overpaid Advanced payment and in a similar analysis of an anonymized Turbo Tax return showed very similar results in years not affected by covid uh what drives these changes for those low-income families whose eitc's decreased and increase in earnings explains the eitc dropped for nearly three quarters of them so 28 out of the 39 so that happens when a family's income moves into or past the phase downrange of the eitc if you'll recall the trapezoid graph from earlier that's like that right most portion sloping downwards um so then accounting for income changes uh decreases in the number of eligible children and increases in earnings explain around half of the remaining ertc decreases income changes in either direction caused eitc increase increases with about equal frequency but relatively few households saw their eitc's increase as a result of having additional qualifying children in the tax unit and again this is just focusing on low-income households and then lastly for the eitc among all low-income households with kids we saw some variation in ertc changes by a household head characteristics a few interesting differences that we observed uh were black non-hispanic and white non-hispanic households saw similar rates of eitc's decreasing staying the same and increasing but Hispanic households saw similar raids of eit's uh sorry at 43 of Hispanic headed low-income families would see their eitc drop compared to lower portions of black and not white non-hispanic families uh unmarried household heads had more stable eitc amounts than their married counterparts we can speculate as to why maybe this has to do with the way a tax unit with two earners can experience income volatility differently or more acutely than a household with one earner and then for married household heads eitc's decreased more often than they increased so like 48 to 20 percent and lastly changes were distributed pretty similarly among household heads under age 35 and between 35 and 50. but a much greater share of households with heads age 50 and above were not eligible for the eitc again this shows households with kids some of this has to do with the underlying distribution of Ages among ertc recipients moving on to the CTC first of all changes among all families with kids and families with kids under twice poverty were more similarly distributed than we saw with the eitc and this has to do with the CTC going to a much wider group of families over 90 of all families with kids received a credit in either year compared to 82 percent of families under twice poverty so like a greater share of families uh low-income families were ineligible for the CTC we'd expect to see this with the Faison and the 2500 earnings threshold for CTC eligibility families under this cutoff make up a greater share of the twice poverty subset uh 21 of all family ctc's increased compared to 31 percent of low-income families for low-income families smaller and larger changes happened with similar frequency whereas for All Families changes were dominated by increases of at least two thousand dollars for example so CTC decreases were about evenly split between income decreases or decreases in the number of children at home so these are changes going in the same direction but CTC increases for low-income families were mostly explained by increases in income moving up the phase and range at the beginning of the the credit if we remember that graph relatively few increases for low-income families were explained by the household gaining children what that was actually the much more common explanation for families overall about half of CTC increases in the overall group were caused by additional children and half of those specifically by a child being born or the household adopting a child I haven't shown on this slide a few other options uh a few other possible reasons for changes including changes in the number of other dependents other older dependents I should say income changes in the opposite direction of the credit changes or changes in marital status those make up the remainders not shown here and then compared to the eatc we see relatively few differences in CTC changes by race ethnicity marital status and the age of household heads among low-income families with kids in general low-income families are more likely to see their credit increase than decrease and that again we can look at where along the phase-in range for the ERT for the CTC those families incomes are distributed we see minor differences by age for example households with heads under age 35 are almost twice as likely to see their CTC increase as households with heads age 50 or over and you can see that in the bottom three blue bars so to wrap up our results show that pretty big shares of low-income families would be at risk of needing to repay Advanced tax credits because so many many families credits decrease from one year to the next uh specifically we saw 39 of low-income families with a decrease in their eitc and 20 with a decrease in their CTC overall most families overall credits remain stable within five hundred dollars from year to year which suggests a safe harbor of that amount would protect most families from needing to reconcile repay excess payments but for low-income families further safeguards might be needed this might point to uh only advancing part of the advance payments maybe higher Safe Harbor thresholds for those lower incomes or perhaps only advancing the CTC and not the eatc because we see that those higher rates of decreases in the eitc at the same time Outreach efforts can help families plan for some foreseeable changes in their tax filing situation for example you can anticipate a kid coming into the tax unit uh in advance to some extent and you can also anticipate a child aging out of eligibility for the eitc and CTC um though what our analysis shows is that this accounts for a relatively small portion of changes overall which are dominated by those income changes So yeah thank you so much [Applause] okay thanks everyone I'm pleased to be able to present some new research today on differences in audit rates by Race So with my colleagues uh Brian sartin Carol ibel and Mark Payne all of the IRS and this work is really an extension of the work originally published by the Stanford team uh uh cited as El Zane at all we sometimes call that the Stanford paper and we're trying to extend that work and really draw the lessons from it that we can for operational uh impact on IRS so some important disclaimers here more important than usual perhaps this document reflects the views of the authors one of whom me is also an author of The alzained all paper so this work is preliminary and we're sharing it in hopes of uh elicit eliciting some constructive feedback to improve our understanding of what's going on here and the perspectives and findings expressed here should not be taken to represent IRS or Treasury Department policy so uh in this presentation I'll use we and our when I say we an hour I'm referring to the four people listed there I'm not referring to the other Stanford paper co-authors and I'm not speaking for IRS or treasury policy makers second important disclaimer as many of you know the IRS does not collect data on taxpayer race so all of these estimates are based on race ethnicity probabilities which are based on first uh last name and address matching against published distributions of the race and ethnicity frequencies for those attributes and importantly these estimated race data are then used for research purposes only they are not used in audit selection and they will not be so the first thing I'll do is look at some of the key findings in Stanford paper and then try to figure out what we can really take away from them and then I want to step back a little bit and look at some of the other sources of audit rate differences by race which we'll touch on some of the uh work of the last paper actually um and then to really stress the point that Stanford stressed which is that audit enforcement objectives matter okay then we'll look at some evidence for algorithmic bias and eitc audit selection and some evidence that unscrupulous paid preparers contribute to the audit rate Gap okay so here are some of the key findings from the alzained all paper which I trust many people in this room are somewhat familiar with uh the headline finding was that using the imputed race data they found that black taxpayers were audited at between three and five times the rate of non-black taxpayers looking at tax year 2014. they find that the bulk of this Gap reflects differences in audit rates by race among eitc claimants not simply that eitc is audited at higher rates than on eitc and that eitc has a disproportionately black clientele as it were but that within eitc there is a race Gap in audit rates so that's an interesting finding so looking within the ITC claimants again they built alternative audit selection models using the representative audit data from the NRP to try to infer what might be creating a race Gap in the operational outcomes and they looked at the following questions what is the objective of the audit selection model all right and what they find is that models that try to find claimants with the highest total tax understatements will pick non-black taxpayers at higher rates whereas models that tried to find claimants with the highest overclaimed refundable credits picked black taxpayers at higher rates so to be clear overclaimed refundable credits are one form of tax understatement but they are not the only form of tax understatement so that latter objective is a subset of the former and it turns out that the demographics of who you pick depends on whether you focus on the total or whether you focus on the overclaim refundable credits sub subset they also found that looking for the highest expected value of tax understatements picks non-black taxpayers at higher rates whereas selecting those with the highest probability of any understatement picks black taxpayers at higher rates so expected values versus probabilities another factor that matters foreign last they figured out that large schedule C's make a difference in particular if you constrain your model to audit a limited number of eitc returns that also have a large schedule seat above 25 000 in Gross Schedule C income that will audit black taxpayers at higher rates than if you do not constrain the model in that way so it was hypothesized that maybe there is some Schedule C constraint perhaps related to Resource constraints on the IRS side that might be driving down the schedule C share and driving up the share of black who are audited so these are three hypotheses based on simulations in research data they're not actually sort of forensic analyzes of the operational models so what do we make of all this um first of all our subsequent research has basically replicated the headline finding uh and confirmed it and confirmed that it's relatively stable over time with some interesting changes uh but fundamentally relatively stable over time we've also extended the work to look at Hispanic Asian Pacific Islander white and all other slash multiple race taxpayers who were aggregated as non-black in the Stanford paper then we turn to the question of these sort of uh decomposition results that the paper presented um in particular uh a decomposition tried to break the total disparity in Black non-black audit rates into the differences in audit rates among eitc claimants among non-eitc claimants and in this between effect which is the one I referred to earlier that if eitc is audited at a higher rate and is disproportionately black then that will mechanically have a an impact on black taxpayers so what what is the relative weight of those three factors and the results in the paper come up with uh the Lion's Share uh being the within eitc piece and relatively small amount 14 being the between effect the components C there and this would seem to imply that to reduce the overall race Gap we really need to focus on part a okay but in fact there are other ways to do that decomposition and many of them assign a somewhat larger share to component C the between piece in particular there's a thought experiment you can imagine which is what if we equated black and non-black audit rates within eitc just equate them mechanically recalculate the overall disparity what do you get and then what if we did the opposite experiment equate the eitc a non-eitc audit rates weaving disparities in each of those sectors as they are what would that do to overall disparity and you might think that from these numbers that one would have a much bigger effect than the other but in fact they have about the same effect both of those thought experiments reduce overall disparity by about 60 percent so that kind of tells you that there's two ways into this problem that are both quantitatively important okay so then how about these three modeling exercises what can we take home from what can we take away from that uh the first conclusion is really very important that among eitc Returns the choice of audit objective really does matter okay fundamentally the demographics of the distribution of non-compliance of total tax understatements are different for the demographics of the distribution of overclaimed refundable credits those underlying distributions of non-compliance measured in two different ways are just different and that means that models that pursue those different objectives will select different populations that's a fundamental Insight of the paper now is it relevant well yes because historically the refundable credit audit program has focused on refundable credits so we do in fact fall into the overclaimed refundable credits side of that dichotomy okay so we are in that uh we're on that side of the equation however the paper Stanford Paper Source seems to suggest that a change in objective would reduce the race Gap and raise total revenue but it's important to be clear that the Stanford Revenue estimates only hold if all audits are NRP style audits which are full scope Audits and they take about 18 hours a piece okay whereas the vast majority of eitc audits operationally are correspondence audits with a limited scope they take about an hour and a half okay so that Gap is big enough to be able to conclude that the trajectory in the Stanford paper and this sort of approach to reducing disparity is not a feasible trajectory just simply cut and pasted to operational reality okay what we don't yet know is whether a feasible version of it is possible and what is disparity in Revenue properties would be the expected values versus probabilities question does not appear to be the problem okay we are uh it is the case that eitc audit selection models generally have the primary objective of minimizing the probability of of selecting compliant taxpayers with secondary consideration given to revenue so we are more on the probability side than the expected value side and there's a reason for that right I mean avoiding selecting compliant taxpayers is really important for pre-refund audits of low-income taxpayers holding their refund check for six nine months whatever it takes for them to resolve their correct claim and incorrect audit status is a big deal so there's a reason for for focusing on minimizing the no change rate or the false positive rate um furthermore subsequent analysis by the Stanford team actually has confirmed that when the objective is refundable credits uh the models of probability actually select fewer blacks taxpayers than do models of expected value so it flips that conclusion the previous conclusion was about total tax understatements saying ah you get more black taxpayers if you go with a probability model that actually flips when you switch the objective function to overclaim for fundable credits which is more like what our objective function actually is so this does not appear to be the problem um uh with our current audit modality next the issue of large schedule C's uh Stanford folks found that the eitc returns were the larger Schedule C businesses are audited at lower rates than would be optimal than an unconstrained model would choose and that selecting more such returns would reduce the black non-black audit rate Gap so that sort of points a wave points to a way forward but again this conclusion applies to models that are based on total tax understatements and audit the full return so it's the same issue we have before we don't really know what a feasible version of that would do and whether it would reach the same conclusion about the importance of larger small schedule C's okay so stepping back a little bit to look at some other potential sources of body rate differences by race and this really ties into the previous paper about eitc eligibility and volatility and eligibility think about the way that the eligibility rules themselves might have a disparate impact by race there's a lot of examples one example is this married couples can claim stepchildren but unmarried taxpayers cannot claim their partner's children even if they support them even even if they're co-resident and they run the risk of being audited if they do so this could potentially have a disparate impact on black taxpayers who have lower marriage rates so there are many other ways that you can think about the eitc eligibility criteria having disparate impact by race and that that might filter into the question of non-compliance and audit secondly the unscrupulous paid preparers who those who submit lots of false claims for eitc and are spotted by our preparer strategy program what we have shown is that they do in fact draw clients disproportionately from Minority communities so if your preparer is submitting a bad return on your behalf and that's getting you audited that might be part of the problem next is the question that the Stanford paper really prioritized the whole question of exam objectives are we trying to minimize no change rates find largest over claims largest total tax dollars and more generally are we focused on single issue Audits and doing a lot of them or are we focused on fewer longer multi-issue audits um these are policy choices and they have demographic implications uh and then comes what I would consider algorithmic bias these first one two three are not algorithmic bias in any data science sense of the term at least in my book let's talk about actual algorithmic bias what that is in my book is that when a particular model for a particular audit selection work stream generates demographic differences in audit rates that cannot be explained by underlying differences in non-compliance where those underlying differences in non-compliance are defined in relation to the existing tax code and in pursuit of a chosen enforcement objective right so how you define your audit objective dictates how you measure and classify compliant versus non-compliant and only after you condition on all that stuff can you start talking about whether the model is doing what you want it to do and if it's not and if it's generating demographic differences that are not justified by uh by what it should be doing that's algorithmic bias or that's what I consider algorithm algorithmic bias so having stressed the importance of enforcement objectives and having stressed that there are many different ways to measure non-compliance that correspond to different enforcement objectives let's look at what some of those objectives might be and people in this room know more than I about the the multiplicity of ways that you might conceive the irs's overall objective one that comes readily to mind is to maximize revenue for a given budget right and to do this you would want to audit according to marginal revenue for dollar of cost subject to current Staffing levels and sales constraints that approach will get you a particular mix of of audit clients demographically a different approach is a little bit closer to what the Stanford paper uh documents or illustrates is what I call proportionality which is the principle that audit risk should rise sharply with the value of the total tax understatement even if that does not maximize enforcement Revenue so it may not be the most cost effective way to audit but it's a sort of vertically Equitable way to audit in the sense that you're auditing the largest non-compliers most heavily and that is not the same as vertical Equity as normally conceived which is to say that you audit the richest taxpayers most heavily this is saying you audit the most compliant most heavily that has a different demographic implication or we could try to minimize audits of compliant taxpayers for reasons that we mentioned but that sacrifices some revenue or we could focus on minimizing improper payments of refundable credits but that will tend to lead towards higher audit rates for black taxpayers or we could focus on the question of minimum coverage or and or focus on the question of minimum coverage and here we have to recognize the variety of audit work streams there is no single audit model there is no master list that we go down there are many many many different work streams each driven by different models and the proportions of work drawn from those different work streams are an important policy choice that can affect disparity as well uh maximizing total federal revenue is another worthy objective and there we would want to emphasize the deterrent properties of enforcement which is difficult to do because they're not as well understood as some of the other properties of enforcement so bottom line is that once you have a set of objectives it's possible to estimate the expected demographics of taxpayers who meet the corresponding audit criteria and then if your models are not reproducing those demographics and are deviating for them from them in significant ways that is a sign of algorithmic bias so here is some evidence that that is part of what's going on here first of all this this comparison that I just said it's like compare the demographics of who you expect to audit and who who you're actually auditing we've done that comparison for many many different definition definitions of non-compliance different thresholds different uh sort of objectives and we do find that the current black to non-black eitc audit rate ratio is higher than the corresponding non-compliance ratio so that suggests that something like algorithmic bias is at Play and to date three mechanisms that contribute to this bias have been found we have not found a Magic Bullet there is no magic bullet that we know of but we found three little pieces of the puzzle first as many of you know residency and relationship status of dependents are a crucial eitc eligibility criteria and they we those we must impute from incomplete information we don't know the relationship status and we don't know the residency status in particular of all taxpayers we impute that and that process is not error free moreover it looks like imputation errors raised the audit risk for black eitc claimants relative to others so hopefully better models and better data supplementing existing data sources might be able to mitigate that problem uh the ddb model dependent database model that many of you are aware of is uh not brand new it is an aging model and updating the existing eitc scoring model could potentially increase exam revenue and reduce racial bias right uh algorithmic bias in my definition is the kind of thing that if you get rid of it you have a win-win you make more revenue and you have less disparity and that's what we're hoping might be possible with better machine learning applications to the problem and then in the past uh weekly audit selection targets led to the over selection in some weeks uh led to over selection in some weeks going too far down the scoring ranking in some weeks and that drives down the quality of the Audits and appears to also have generated some racial bias this problem has largely been resolved with better scheduling through the year uh next to last slide um the unscrupulous paid preparers that I referred to previously we've been tracking them since 2005. there are penalties that can be applied treasury is proposing more such penalties um and the numbers are interesting in 2019 17 million returns out of 160 million came in from these preparers who were on our list as past or present unscrupulous providers uh and six million eitc claims out of about 26 million so it's a big chunk of returns coming from preparers who we don't think are exercising due diligence on behalf of their clients um and what we have shown is that those clients are drawn disproportionately from Minority communities so it matters and what the uh one way of quantifying how much it matters is just to exclude all the returns from all of those preparers to exclude those 17 million or those 6 million and if we do that the overall black to non-black Gap in audit rates Falls by about 21 percent so that's a non-trivial chunk of the issue right there to wrap up um the emphasis on preventing over claims of eitc credits is reflected in both the number of eitc audits conducted and in the way eitc claimants are selected for audits that was uh an important lesson of the Stanford analysis is that in both cases this emphasis serves to raise the audit rate for black taxpayers relative to others our ongoing research is evaluating the hypothesis that a change in audit objectives to focus on top dollar tax understatements is feasible in a pre-refund correspondence audit environment where we can't spend 18 hours but we can spend an hour or two per audit or maybe three this includes estimating outcomes in terms of differences and audit rates by race but by burden on the compliant taxpayer and enforcement Revenue and secondly we have found evidence of algorithmic bias uh it's not the full story but it seems to be there and preliminary research has identified potential updates to algorithms that may be able to lower the Gap in audit rates while improving audit outcomes improvements to audit selection algorithms take time to test and Implement and are critically dependent on the funding made available through the IRA that's it thanks [Applause] okay yeah the right arrow right there right arrow I always want to go up here I did the same thing thank you I don't know what else what it just seems intuitively it should be up not to the right uh I did it excellent I don't want to do handheld because I screw up okay I want to thank the organizers for um giving me the opportunity to comment on these two really interesting papers I'm going to start with my comments on Jan's paper common name it's also a very common experience that we can all relate to what makes all of you hang up and switch to the online website uh is it the how long the wait the messaging the number of messages the music your tolerance for pain for me it's the music that always drives me yeah but that's not what's been tested here I'll be curious to find out what you do play there um how you get into their sample you've got to be the type of taxpayer who receives a collection notice from the IRS and then follows up with a telephone call and listens to the first pre-recorded message the first pre-recorded message was followed by four or more in between music and then after the messages are over it's just straightforward music the metrics to measure this are first does they call or hang up the abandon rate does a call or go to the website the channel shift and does a call caller actually use the website application the access rate I'm going to throw in a cribble here which is there's a lot of emphasis in this report on percent differences between the two groups I think percent differences overstate all the impacts because for example on channel shifting you will see a 13 increase in Channel shifting but that's associated with going from 12.5 percentage uh points of those people who use it to 14.1 percent um cribble now in terms of the emphasis not the music again it's the message and a lot happens in terms of how each message changes there's a lot of variables that are changing the five that were most alien to me was they began each of the five messages with a question and then they provided an answer in action in each of the messages and particularly in the first measures which they don't do Under the current situation they refer the caller to an IRS website and thankfully for people like me they do it at least twice in the same message if not more than twice and they give a link to the website that corresponds to the suggested action in the message they order the messages by the most frequently as and recognizing the intelligence of the taxpayer they don't keep telling the caller that the call volume is high another irritant um so what is the impact of messaging and what is it that we're interested in um for me what my interest would have been speaking personally is what is it in the wording of the messaging that causes a change in behavior um and that's really hard to quantify because as I say there's a lot going on in changes in each message and if I look at the results I'm a little doubtful as where we draw the line here if you look at among those who hang up it's the first the second message we see a higher percent change percentage Point change percentage Point change in listening to that second message and then hanging up but if you look at the percent change for those who take action after hanging up it's the first message which leads you to a more General website and also in terms of seeing action taken um it's the people who are the biggest action is in terms of going and setting up an online payment system so although it's the second message that tells you that it's a lot cheaper to set up a to make your it's a lot cheaper to set up an online payment plan online rather than doing it through the customer service those people who do tend to be motivated by the first message which just gets you to a generic IRS website um and then in terms of the numbers and I may be misunderstanding the numbers here because I pulled but it seemed to me from my reading that a lot of people were eligible to use the online payment system or and so forth to set up a plan online but when you actually look at the numbers there's not a lot of people relatively speaking who do it under the current message and then while you get an increase when you go to the on to the new messages maybe it's only 400 people so um that raises the question that my niece the user experience expert pointed me to which is is the issue the messaging or is it the issue the website and we're not there yet I like this paper because it was fun to read and think about our you know our all experience it made two important observations which are really relevant to stuff I'm thinking about which is the issue how do you Rec how do you measure performance they point out that the IRS current metric for looking at the effectiveness of telephone calls probably understates or ignores the effect of people hanging up and actually then transferring or Channel shifting to the the website and it also indicates that not one metric is ever enough to really get to what the user experience is for a researcher perspective I'm very interested as is a lot of people in terms of the return on investment on customer service in this specific example they were able to Trace through from the beginning of getting a letter to taking an action by having the social security number to 10 and I'm curious as to whether we can in other forms of customer service be able to track taxpayers through the use of their tin that would be great turning to Tom's paper Tom paper follows the Stanford study that was released a few months ago which even though is called the Stanford study Tom is a co-author um and that study confirmed our worst fear is that there are racial disparities in the audit rates uh in part because most of the co-authors were not affiliated with the IRS they couldn't really delve into why there are those racial disparities and that falls on Tom and his colleagues to undertake but there's going to be a long and difficult task I don't envy you but it is interesting but one of the things that you could get out of the Stanford study was the objective of the audits mattered why were you doing that and I had flashbacks then to my life as a child in the 90s and getting involved don't laugh Barry and getting involved with eitc compliance at the very beginning and when I think of what we were looking at at that time in terms of the objective of compliance it was purely political well maybe not purely political the goal was to avoid the huge cuts in the eitc that were being proposed by Congress at the time it was was also to open the door on refundability for the child tax credit and an agreement that was worked out between the White House and Treasury and then Speaker of the House skiing Bridge there were no cuts to the eitc and it didn't open the door for refundability so if you look at the political objective What followed next really gained the objective the at least the commitment to improving compliance in the eitc and what was worked out in this particular letter was that the IRS would get an additional 700 million dollars in Appropriations over the course of five years and use that strictly for eitc compliance and then yield out of that a certain some billions dollars of revenues that may sound familiar to those of you involved in the current debate we were leery of using the error rate even though we wanted actually to improve eitc compliance we were Leary of using the error rate because the problem with error rates and tax scabs measures is that on the one hand they may show a reduction over payments but they may also show an increase in overpayments but the same thing that may lead to reduction Also may increase the amount of overpayments that are detected that kind of desired not to go too fixed on the error rate we started changed in 2002 with the enactment of the improper payment Act and OMB putting the eitc on that list and now annual measures of the error rate are published in terms of We're treasuring the IRS were in terms of how you're going to achieve those objectives the goals were to minimize the burden on both the IRS and taxpayers and the result was this emphasis on correspondence audits which would be dealt with through the mail not in person interviews and would also be focused on several issues and those were informed by the compliance studies and in addition they were accompanied by other statutory changes um and those statutory changes one of which led to trying to get greater access to third-party data in the 1977 act one there was this increased access that was given to the IRS to data on child support Arrangements through the national registered child support registry also there was a requirement that SSA would obtain the SSN of the parent and Link it to the SSN the child that they were currently applying for those were going to be a long-term strategy okay I got it down to a minute and a half overtime oh yeah we knew there were cracks we knew that neither was going to get us all the way in terms of uh being able to detect The Hope was that data analytics would fill in the remaining hole and some of what Tom has identified suggests that either we haven't kept up with that or that there were holes there I think in terms of the largest crack in terms of the data was that inability to find any data on marriage this was a particular concern because one of the issues that we found was that there were married couples who were filing as head of house cells in order to avoid head of household status but we were told that marriage data might be available in the shoe boxes of many Town City Halls we couldn't get that that could have effect on the racial disparities because we know that the marriage rate for a black couples is black adults is about half that in white and so if you're targeting ahead of houses you may be missing the people that we were most concerned with going forward there's a lot of things that need to go forward what I'm interested in personally is how data after the audit plays into the audit selection we know from the work that John Guyton day menoli and others did there's a high correlation between uh the eitcd Niles and people not responding to an audit if people don't respond to an audit letter because they don't speak English well or because they're have good reason to be fearful of government that may be and that then fact is back into the race into the audience election that historical data that may be compounding bias in the system and frankly that's going to be a problem the extent to which the IRS moves forward in between learning which often is based on historical data thank you for your patience and moving on to the next one can I do that Emily I got you up there [Applause] um it's my pleasure to discuss these two very interesting papers the two papers have a common theme on life events and their tax consequences I would discuss the balance due paper first and then the credit paper uh balance due taxpayer paper is very interesting and uh very important for tax administration I especially like the focus I placed by this paper how to prevent taxpayers from being in the vulnerable situation I have a few comments some about data analysis and some about the administrative strategies first the paper identifies life events by looking at changes in 1040 character heuristics from one year to the next such as changes in filing status or adding or removal tax Schedule I wonder if the authors also look into other situations more generally for example you talk about adding Schedule C so you may be able to calculate the share of taxpayers income subject to withholding in general so that's the result will not limited to the schedule C but you may be able to see maybe capital gains or dividends also play a role in explaining the results and the paper said that the regressions were performed on taxpayers who have an unfavorable shift in the balance due status it may make sense to narrow it down to those taxpayers who also have scp-14 issued so that we can like rule out taxpayers who are able to pay anyway regardless of status shift because I think from table two you can see that among taxpayers who have this unfavorable shift in balance due status majority of them did not have a CPE 14 involved and um so I agree with the papers strategy of identifying the Live Events but I think the focus of the paper is taxpayers who need a notch in order to pay the balance due so maybe you have already imposed that data restriction but it's not clear to me from reading the paper uh you talk about the income effect and I'm interested in seeing some of the income effect on the CPE 14 issuance and some of the resulting outcomes um and next the paper points out a communication gap between the IRS and taxpayer and one of the Strategic goals covered in the paper which I found particularly intriguing is about the external communication campaign and I would encourage the authors to look into behavioral insights to see or to consider how the rs2 could make the message or communication with taxpayers in a more proactive way or to shape or to influence the taxpayer Behavior um also given the new infrastructure um underway as laid out in the recent strategic upper strategic operating plan are there opportunities that us could think of to more fundamentally change how it interacts with taxpayers um and as many of you may remember for tcha changes uh in taxi year uh 2018. RS automatically waive the estimated tax penalty for eligible taxpayers because many taxpayers fail to pay such payments uh in correspond to the new law so I understand the papers want to avoid an outlier like 2018 and focus on 2016 to 2017 changes but I also think that for all our situations like that it is even more important for the IRS to have a effective communication mechanism in place to help taxpayers to avoid that vulnerable situation thank you last comment um the paper is emphasis is an upstream prevention but it looks like the artist may also want to identify the downstream activity to see why some taxpayers need um additional contacts with the IRS as the paper mentioned the phone calls visits at the taxpayer Assistance Center are all very expensive options and burdensome to taxpayers so maybe you could see uh or learn from this additional context to see if there are any improvements to be done in terms of the clarity or the timing of the initial notice so that we could increase the successful rate of cp14 on the full pay or even installment agreements let's turn to the next paper yearly chance to credits this paper estimated that a large decline in the year to year eitc and CTC occurs for a significant fraction of low-income taxpayers with children and the impact is a greater amount certain minority households this paper also has important implications for the design of Advanced Credit I only have a few comments uh first I would like to see more analysis and discussion about the scenario in which a child may be claimed by more than one taxpayer the paper mentioned multi-generational households cohabiting couples um so as far as a child's well-being is concerned it may be interesting to evaluate the credit changes from a child's perspective in addition to from the taxpayers perspective when more than one taxpayer could claim the same kids second on page seven and Page 8 of the paper for the discussion about changes in marital status I'm a little bit surprised that the paper talks mostly about the income differences between the single and marriage dates but not so much about the potential change in the child's residency maybe it affects a small fraction of taxpayers but I suspect the impact could be large when the kids move away from the household relatedly in the few places the paper talks about long-term Trend in the U.S household living situation for example the decline in the number of over the share of married couples living only with biological children or the increases and had the household filers or joint custody um the authors may want to connect this discussions more closely to the papers focused on the yearly shift in credits I would love to see one implication we could learn from that kind of analysis uh the paper shows that undercurrent law in the car under current tax administration if the advanced eitc and CTC were to base the solely on the taxpayers prior year income in family structure a significant fraction of taxpayers would need to be repay the credit back at the end of the year or they or some for some they may not be able to receive the full eligible amount during the tax year and we also learned that there are a few conditions under which this risk could be minimized for example if the live event is anticipated or it can be reported timely to the IRS to be reflected in advance payments and also the credit design also matters for example facing the credits or face out the credits at the relatively low income level certainly will make the advance payment less predictable for a larger fraction of um taxpayers and I think these are all very great points we should keep in mind in a related discussion however regardless of whether there is an advanced option or not um I think it is desirable from policy and tax administrative point of views to reduce the risk of unexpected and unfavorable change in the credit amount from one year to another and one implication alluded to but not elaborated in the paper is the importance of RS taxpayer service I guess the the discussion in the previous paper about how to improve IRS communication in the more proactive way can also apply and certainly will help with the current case thank you [Applause] because foreign okay um before taking questions I just want to thank everyone for presenting and your interesting work and I also want to give special credit because never before in history of presentations have I seen people behave so well regarding time so thank you very much yeah yeah so like except for Janet thank you everyone okay so uh questions yes got a question for Howard um you you notice or you uh indicated that one of the drivers perhaps of um reducing baldu situations was the filing of schedule a itemizing and I wonder well that that sort of makes sense right so if someone has at the end of the year a certain amount of tax withheld and then they figure out their tax and it it's not met by the amount that they've had withheld they could reduce their maldu by making up some itemized deductions maybe things like terrible contributions that aren't really tracked all that well by the IRS um and so I'm wondering that that the the outcome of um let's say not having a bow do or having a smaller bow do sounds like a good thing but it may actually be caused by a bad thing if it's non-compliance reporting non-compliance and I'm particularly concerned about that because our our random audits historically have said that the compliance of people who report claim that they have a refund due is less compliant than those who admit that they have a baldu they're more honest so I'm just wondering if you've been able to um check into that that maybe the the reduction of baldus may actually be caused by a bad thing I see you like a foot in front of my face that's not something that we that we've looked at um through really talking about kind of mitigating or using that baldu in a so you know it's not something that we've considered or used but it's really with the schedule a it's the removal of that so it's like you had those in one year then when you remove schedule a that's when someone's more likely are twice as likely in terms of the risk to move into a Badu situation does that answer um yeah but removing it if the prior year is not the right Baseline if the right year they had fake itemized deductions they take that away that could be a good thing but it may result in about due this year okay I see what you mean yeah so it's it's really would be looking at more of a trend overall so you'd have to look at the pre previous year so did they use that um yeah did they use like things that weren't true um on one year and then removed it so yeah we didn't we only looked at the two years so we didn't consider the just are the if it was legal what they did in 16 compared to what they did in 17. but yeah something that we can look at kind of more of a trend over time okay and question over here yes a couple of quick things for John for the first first paper so one thing that I assume so please please confirm this you all guys ask for the identification number at the beginning of the call because you you'll you know you are following these people over time or you are supposed to be following these people all the time so I assume that that's different things you the second thing is that I didn't understand very well because you said that you drop some people that call multiple times but you will assume that these people originally would assigned to a three minute control now the first time they call so if they call multiple times then your treatment is not working so you should be considering these people in the in the analysis now yeah so um the reason why we chose the application that we did is because in order to be in the application the taxpayer did have to provide a tin so we were able to follow them now that was the nice thing um one of the reasons why we did the multiple is because we were trying to test the validity of the messages right so is the nudge possibly doing it is the control so those that called multiple times they heard both the control and the nudge and and so it's like okay we can't apply the logic to did we see a channel shift because of the message change right is because that's one of the things but right yeah so but the interesting thing about that data is now we have another little subset over here that we can say okay why are they doing this it's like you know we've tried multiple ways why are they doing it so what type of path are they taking which again now we've got all these cool ways like oh how can we even do a better next time or how can we do other designs how can we start to address those taxpayers that we kind of put on our outlier for the moment so we didn't quite ignore them we did for the analysis but we still have them in the back of our mind based on the analysis we took on them on how we can start to approach in different Avenues and and also we couldn't assign taxpayers to a control or a treatment group we had to do it by day so taxpayers who called in on certain days heard the control message and taxpayers who called in on other days heard the revised message yeah question okay well did I miss anybody I see a hand over there and this question is for Tom I was really struck by the unscrupulous preparer impact and whether you're examining the presence of a tax time bank product with those non-compliant returns and the question is more about ability to improve and whether other treasury agencies like the financial bank Regulators would want to be involved if the underlying return flowing into a bank is in fact an improper payment then um you know positive impact would be broader than IRS tax time bank product you want to Rack row refund transfer product all the ones where they'll um enter into a contract with the preparer at the time of preparation to have fees deducted from the forthcoming refund by having a limited purpose depository account opened in a third party banking institution um that that tends to be something that I've seen in doj press releases tied to civil injunctions and criminal cases wow well um I would I don't know anything about that I would first the first thing I would want to know is whether we can track that information uh and I can certainly relay that question to the people who do know something about that yeah I think Iraq and a row indicator there's a GAO paper from from that goes into the ability and accuracy for IRS to track racks and Rouse and then I think they also use um there's only probably like six or seven Banks where there's a unique routing number and you can identify consecutive consecutive prepaid debit card trackers okay account number yes yes so this is for uh is Howard and Nikita kind of jointly um and I'm wondering you know you kind of talked a lot about this exposed or like a website but given that the IRS does have at least some information like for wages how much they're changing kind of during the year or Nikita for your paper you know like we know when the child turns 18 right they're likely to drop off I'm wondering if you've kind of thought about how we can use these kind of ex-anti predictive measures to maybe send out a letter you know in the year before this happens and say hey your child is going to drop off that's going to change your eitc or that might change your filing status here are some things you might want to consider and use that as kind of a preemptive measure as well to reduce kind of this non-compliance or shocks to taxes okay well I'll leave the designing tax uh pair customer service to the experts but I will say I've done some research on a program where nurses did home visiting with like new mothers for example and they did that uh and there was a pilot in which the nurses would refer the new expecting mothers to tax filing services and so and with Kim who's here actually and so you know that might be one venue maybe not directly through the IRS but through trusted Messengers and third parties uh you know people who work with vulnerable populations who could be you know trusted you know who could overlay that information and help people plan around those life events um so not quite the IRS but like another idea for how to do that and I think the idea of messaging in terms of letters I think you run into obviously there's a cost associated with that um just as we saw with the cp-14 but also do those letters then generate calls and so when you're looking at kind of like that return on doing that but that's one of the things in terms of partnering with when I was talking about for divorce can you partner with divorce attorney groups to get the information there that's really they're dealing with that taxpayer can you partner with schools so you have at part of the a letter that goes home potentially from the school for the graduating senior that that information can then like kind of spur that it's like hey I just graduated in May it's like well you may not be able to claim me next year or but you might have aotc to help you out or something like that so yeah it's it's trying to put it in those because as we as we know many people we could send we send millions of letters that sit in piles that don't get opened so partnering with these other organizations that are directly impacting with that individual I think is a strategy that we wouldn't want to look at all right we have a few questions online and to the people who've submitted them whose questions I will not ask I will make sure they find their way to the right people this question is mostly for Jan and it comes from Nina Olson I I know I know I know um Nino Nina wants to know if there's been any investigation into the quality of the plans that people the payment plans people end up in whether they go online versus whether they do it with a human and are there good are there real differences in the process or is the caller relatively following a script so that's always a really good question um we find that there's this I couldn't I shouldn't call it generational Gap because I'm sitting up here with gray hair but you do find that we have a certain group in our demographics that we consider like hand holders where they kind of have an idea yeah I can do it but while I'm doing it I really want to have an individual to step me through it to make sure I'm doing it right because I might make a mistake and there's nothing worse than making mistake with the IRS and having them send you another notice telling you you did something wrong so we haven't seen like a difference in the value of a taxpayers who use Opa versus those who call and set it up online we have um we still get the same results they're still making the payments they're still doing everything that they need to do it's just really the preference of the end user or the individual that's calling that on whether which approach they're going to take all right thank you and lastly um it's a general comment from Harper Sutherland at HUD who'd like to hear the panelists make enough or to focus on when relevant the difference between the tax situations and what you're discussing as they may apply to W-2 workers versus gig workers because their tax situation and their difficulty filing is quite different and with that we are out of time um so I thank you all very very much [Applause] thank you you have a newsletter emails or the IRS doing this with me um [Laughter] foreign for example [Music] greatness hi I'm John Gordon I'm with IRS research applied analytics and statistics I'll be moderating the second session here uh which is on estimating audit aftershocks uh part of the the broader context here is having a is developing a better sense of the uh the full range of impacts of our of our actions as an agency and the the broader uh the broader business impacts there we had a nice setup for a couple of those themes and Janet's discussions of some of the uh Jan's paper in particular talking about the ability to use some of the information discussed in those Pilots as part of a return on investment kind of analysis so I think we can think in a similar kind of frame the the types of work we're going to be looking at here are part of the evidence space for giving us a broader impact uh a broader broader measure of the impact of of our actions so we've got three authors here first up is Alan Partington from Australian tax office and then we have India Lindsay from miter and then Alan Plumley from the IRS and will bonding from treasury will be the discussion so let's get started Alan you want to start you want to start here foreign good morning everyone uh my name is Alan Partington um I'm a director with the Australian taxation office so let's thank you for the opportunity to talk to you this morning um I'm presenting a paper that I co-authored with Mira at besnik in relation to changes to voluntary compliance following random audits on income tax returns um so this paper sort of follows on from previous research that we conducted um there's a few differences in relation to the previous research and this research so previously the ATO didn't have a random audit program and well until relatively recently we started in 2017 starting with returns for 2015. so previously our analysis on audit Effectiveness was limited to using risk-based operational audit data which obviously provides us with a biased sample of of taxpayers true reported or true liabilities compared to what they report so we were good operating in that quasi-experimental space using a Hickman correction and propensity score matching techniques and The Perennial question we would often get asked from those sort of having a look at those papers is how do you know the extent to which those techniques are effective in in de-biasing the data in effect how do we know our results from those sort of earlier estimates weren't biased and probably the honest answer to that was well in the absence of a random order program we can't be 100 sure so when we were able to Avail ourselves of the random audit data we were very pleased because it sort of provided us an opportunity to to answer that question that had been unresolved until that point so yeah so the first difference I guess with this paper is that it uses newly available random order data um we also had some challenges with some of the data distributions that we were working with um typically um the techniques I mentioned earlier Heckman uses a logistic regression and a Tobit regression in a two-stage setup we sort of found though as we work through it that the normality assumptions underpinning the use of uh of OLS based or censored regressions it didn't really fit the the data that we were working with so we looked at other other um other estimators that were more robust to um to non-normal data and so we used a technique called poisson pseudo maximum likelihood to try and deal with some of those issues around non-normality the bit I'll sort of come to that later before I get too far into the approach I'll sort of go through a few of the concepts that we're sort of looking at in relation to um uh that sort of underpinned the analysis so um so I mentioned the random inquiry program that we have um through a series of comprehensive audits uh based on a random sample of taxpayers um we determine whether a taxpayer is compliant or non-compliant essentially whether there's an adjustment that needs to be made to their return um uh that sort of definition is useful for the purposes of our paper it does ignore the possibility though that sometimes Auditors when reviewing a taxpayer may make it a decision that they're compliant when they're indeed non-compliant or alternatively but less likely that they find that the taxpayer is um non-compliant when they're compliant but we sort of abstract from that a little bit for our analysis but it it can be important I think in explaining some of the outcomes that we see so it's worth bearing in mind for later we worked with three populations under that random inquiry program so we have a look at taxpayers who are in our individual tax Gap population these are taxpayers who predominantly salary and wage earners they may have um some pieces or I'm sorry they might have some investment income but it's a passive income we also have two subpopulations within our small business Gap population um so they're individuals who have some active business income so they um what we call Sole Traders and also small businesses that operate through a company entity structure the the paper attempts to um there's three sort of Concepts I guess that we look at in in terms of the effect of um of our review and audit programs um the first is the audit yield so they're there the direct liabilities that are raised as a result of the uh the reviewer audit um and then we look at what we term direct flow on which is the the change in voluntary Revenue reported in subsequent years after the audit which provides us with some indication as to how taxpayers have responded to the to the audit in the previous year or previous years um there's also another important component where we don't measure this as part of the paper that relates to spillover effects or the indirect flow on from the audit so that's the behavioral change that we see in taxpayers not directly affect by the audit but may change their probabilistic assessment of being audited based on what they know of what we're doing in the market um so uh in terms of uh the the role of audits um uh I guess this is probably subject to um some contention but we see that the audits underpin uh the system uh in in an Australian context at least our order coverage is fairly small it's probably one to two percent of the overall population but we think it plays a very important General deterrence role in underpinning um voluntary compliance I guess are assisted compliance um so we sort of suggest there that um that taxpayers do tend to factor in uh at least subconsciously that um uh what their probability of being ordered it is and um that the like or the consequences of being audited um uh the other argument I guess is that they're intrinsically altruistic and and very keen to to pay their fair share and so forth um we see some of the responses in our later results perhaps not being consistent with altruism being the predominant driver of the results that we see but uh yeah so um I guess uh the bottom line there probably for us is that the more credible the threat of an audit the lower the payoffs for non-compliance um making it more beneficial to contribute um so what it's basically underpin that dynamic as we see it um so the the actual analysis itself it looked at um random uh random selection of taxpayers from those three populations I referenced earlier um at the commencement of this study there was uh data available for 2015 income year uh 16 and 17. so we had three years of data available the first year of our program um uh it um some of the data produced by that was interesting in terms of we think it was subject to a higher than usual level of uh of ordered a non-detection um given that they were new to the process they were engaging in full scope or comprehensive audits for the first time many of them and it's an interesting year to have a look look at with respect to the data but we've failed that the three years looked as a whole provides us with a a good basis for drawing some conclusions so we we analyze each year separately but we also do a joined estimate as well we only include taxpayers that have been contacted directly by the ATO so the reason for that was that we were very keen to have a degree of intensity in the interaction between the tax administration and the taxpayer to support the notion that there's been some sort of Behavioral change and we felt that a correspondence review or audit probably didn't provide sufficient visibility or impact to Warrant using that as a as a product in this context to to estimate behavioral change um the intention of the study was basically to try and establish an average treatment effect for uh in effect what would happen if we audited everyone using the properties of the random sample to approximate that um we uh in order to set a baseline for the analysis we selected a control group this was constructed after the event um as part of the the random audit program design um uh no one really thought to retainer a control group per se so everyone that was initially selected uh as part of the sample was treated so we had to undertake some efforts to following the same criteria as the original sample selection in effect sample or select a control group of of taxpayers that hadn't been treated we had to go through both the treated and the control to identify potentially confounding interventions so this is where people treated under the random order program may have been subsequently treated under our risk-based program in in later years um or similarly for controls where within the year within the time series that we're looking at um may have may have been treated uh in in that time period which could potentially confound the results so any any taxpayers that were treated subsequently or any in the control that were treated were removed from the study we also undertook a a test to see whether or not um the pre-intervention uh Trend was a parallel Trend and probably not an entirely necessary step but it did confirm that um that both the treated and control were similar in Trend terms prior to the Intervention which was just a useful way of confirming the the quality of that control group that we had to construct after after the treatment group had been selected um as part of the analysis we removed the net tax amounts from the um the year that the random inquiry was undertaken or the random order was undertaken um so for instance taxpayers who were part of the 2015 random order program we removed that 2015 returns from the data set we do that for the control that's to remove that direct audit yield value from the from the analysis um so just in terms of what we found um this this just relates to the uh audit yield that we uh derived though from uh those people who were treated um so this isn't the treatment effect per se it's the it's that it's the treated on the treated basically um so we find that um across our populations uh the audits were effective at raising uh direct audit liabilities as indicated in in that table on the left there so not in substantial sort of adjustments um uh relating to the actual random audits themselves um we also noted that the um the incidence rates of non-compliance in some of our populations was significantly more than we'd previously thought um so for example the Orange Line there in our individuals not in business population uh uh roughly 95 of the population had some issue that was detected with a comprehensive audit um which uh we we thought was incredibly High um the some of the issues there is uh with our self-assessment system uh in relation to work-related expenses we don't have a standard deduction in Australia and there does tend to be a lot of non-compliance in relation to work related expenses also there's very high incidence rates around people who've got rental properties or investment properties which pushes pushes it up significantly um that's important for our study as well because for individuals at least when we went to calculate the effect of the order with reference to a control the individuals who were compliant or were found to be compliant with the audit there were too few in number to actually derive credible estimates of the impact of the audit with reference to a control group for those so we only we only estimated the effect for the non-compliant taxpayers for individuals uh that gives you some indication of the sample sizes that we have so our total program the rep program uh is about 1250 taxpayers per per year so very small compared to the IRS their NRP program and the like unfortunately that limits our ability to um to disaggregate our results so in terms of moving forward we'd certainly like to to increase our sample sizes to help us stratify some of our results uh and narrow in on some of the causes of what we're seeing um uh I mentioned some of the I'll get to the results in a minute I'll just uh take a step back in terms of some of our data challenges um when we reviewed the literature we found uh that um looking at our data we found uh issues with um the presence of zero values in our data and a non-normal distribution uh the literature tends to suggest two approaches that are commonly used so one is the use of a log linear model we found though that um because of Jenny's inequality that back transforming those results uh led to biased estimates so we we um we sort of stepped away from using the log linear approach another commonly used approach is to add a small constant to the zeros and then logging both sides of the equation it can create issues around heteroscedasticity across the regressors by doing so um yeah so so we sort of stepped away from that as well um where we arrived uh and this was a recommendation from my uh my colleague Murat um uh who um was having a very good look at the literature and uh came across uh this poisson pseudo maximum likelihood estimator that was being picked up in some of the literature I was quite surprised actually I'd only come across that technique used for count data so my initial reaction was to suggest that um maybe it wasn't appropriate but the more we looked at it the more we found that it had some very desirable characteristics for the sorts of distributions that we were dealing with um so it's uh ppml is basically a Persona regression with some robust standard errors the estimator is based on the conditional mean typically the the mean is equal to the variance with a poisson regression however for continuous data that sort of rarely holds in practice so we have to estimate our standard errors using Echo Hoover white robust covariance estimators what does that get us it gets us a bit of precision around our coefficient estimates and it also gets us the ability to report our results using fairly standard tests of statistical significance which we think is fairly desirable so we did go through with our results though sorry I'll just uh that as I sort of touched on before ppml seems to be increasing in popularity in the literature I haven't seen too much application of it in tax analysis but um it's certainly more popular in other other fields the the part of the race and is it's a fairly robust sort of estimate um with the only condition required for consistency um is the correct specification of the conditional mean um the and its ability to handle the zeros um the uh so when we when we analyzed our results we're used to pretty standard difference of difference estimator or two minutes um sure um so we estimated it with respect to a difference in difference estimator we take the the values after the audit versus the the values before the audit and take off the control group to derive the treatment effect um you'll see various specifications of it uh one uh one thing we did with ours we divided our population between compliant and non-compliant which necessitated the need for individual fixed effects so these are our results we found for populations that were based on individuals the random orders had a negative effect on compliance in subsequent years um so in this case minus 475 on a pulled basis refine Divergent uh impacts based on whether the taxpayer was found to be compliant or non-compliant during the audit so for those that were compliant in following periods here their reported liabilities dropped off relative to the control for people who were found to be non-compliant they responded to the audit by improving their reported liability since subsequent years relative to the control and for populations that were Incorporated for both the compliant and non-compliant population they improved their reported liabilities compared to the control um so that's it really so we hope that this adds to the sort of or certainly added to our understanding of the issue we I think we got a lot out of using the PPA ml estimator it gave us a lot more confidence around the results that we were seeing um uh thank you very much [Applause] all right well hello everyone my name is India Lindsay and I'm with the miter Corporation I'll be presenting our research on the long-term impacts of audits on non-filing taxpayers starting off with some backgrounds so non-filers are responsible for 32 billion or 9 of the individual income tax cap and the specific population that we're focused on in this research consists of non-filers with at least a hundred thousand dollars in reported income and the those populations of key interests because these higher earning non-filers make up a significant proportion of or sorry a small number of the total number of non-filers but they're responsible for a significant portion of the tax cap it's estimated that they owe greater than 73 of the non-filing gat and there's been an increase in the number of non-filers identified every year there are seven and a half million identified in 2010 and that number increased to 10.7 million in 2016. alongside that increase there's been simultaneous decrease in IRS resources to audit these individuals the IRS was able to initiate about three and a half million cases in 2010 and that number fell to 800 000 cases in 2018. so the IRS recognizes the need to strategically pursue these individuals and they've released several policy initiatives pertaining to that cause in 2020 the non-filer enforcement initiative promised stronger pursuit of non-violers with a specific focus on higher earning individuals in 2023 the Ira Ira strategic operating plan um emphasized a continuation of the strategy with a focus on addressing high dollar compliance issues and overall audits are the primary mechanism that the IRS has to encourage compliance within this population and while the direct revenue from an audit is understood there's need to evaluate the indirect impact of these audits in order to get a proper Roi metric so this leads us to a research question what is the effect of an audit on the long-term filing behavior of a non-filing taxpayer so considering the literature on this topic first there are conflicting theories on how enforcement affects how a taxpayer will adjust their future filing Behavior and regarding the specific population of higher earning non-filers there's no set theory on how they will respond to audits second literature has primarily focused on understanding the factors that influence a non-filer's future following Behavior key findings from this literature suggest that how visible the taxpayer's income is to tax authorities the Persistence of filing Behavior and the taxpayer's perception of government and send some moral duty play a prominent role the majority of this analysis has focused on the general non-filing population and to the best of our knowledge Gerard was one of the first to extend this analysis to consider higher earning non-filers third there are studies looking at the indirect effects of enforcement on the non-filing population but these studies are limited however two one conducted by taglakas and the other by data were able to confirm the presence of an indirect effect when looking at non-filing taxpayers but overall there's a gap in literature analyzing both the behavior of higher earning non-filers and the role of IRS enforcement activities on their future filing Behavior okay and so to conduct this analysis we had to identify how non-filers are selected for audits and in this research we are specifically focused on non-filers undergoing field audits conducted by sbse and I also want to note that this selection process details the procedure used to identify the majority of non-filers for this type of audit but non-filers can be identified for this audit by alternate processes such as referrals all right and so it starts off when employers entities and financial institutions report taxpayer compensation to the IRS the IRS then compiles this information with prior year return information then individuals that have not filed would have reported income or a prior return are identified by the case creation non-filer identification process then selection filters are applied to these taxpayers and dependent on their characteristics they can be distributed to various enforcement functions within the IRS and like I mentioned we're focused on sbse field audits so then the iris is able to assign taxpayers for audit by estimating their tax liability assigning a priority score and then assigning eligible non-filers to feel a Personnel based upon available resources and so to conduct this analysis we used a two-sample group comparison comparing the filing behavior of individuals in a treatment group with the filing behavior of individuals within a control group so we obtained our taxpayer data from the irs's compliance data warehouse and in this research the term Baseline year refers to the tax year the taxpayer entered the sample as the return for that year is either audited or is eligible for an audit and our tax years of Interest range from 2009 through 2014. so the treatment group consists of non-filers audited under field exam that were identified from examination records and we're only considering non-followers audited under this work stream that were identified via the selection process I just detailed and that's that we can construct a representative control group we also excluded any pickups for my treatment group and Pickups refer to returns that were audited under this work stream as a result of a separate but ongoing audit covering another year's return so then our control group consists of non-filers that were eligible but unaudited for this type of field exam they are identified by replicating the audit selection process just detailed and to ensure this group consists of non-filers we excluded any late filers secondary filers or individuals that filed in response to any type of IRS notices okay and on the slide you can observe our sample size Broken Out by Baseline year with the treatment or audited group shown in red and the control or not audited group shown in blue so our treatment group consisted of about 5 700 taxpayers and our control group consisted of about 4 300 taxpayers payers you can observe A variation in the sample size by Baseline year specifically regarding the number of taxpayers that were audited this or that number increases leading up to 2011 and then decreases afterwards and that variation may be the result of changes in audit resources leading to potentially fewer audits being conducted we also performed some sample cleaning and we dropped any taxpayers from our sample if they were deceased within the eight years after Baseline year if they were identified for audit by another procedure like I'd mentioned if they are in our treatment group and had any missing or incomplete examination record data or they were in our control group and underwent any kind of audit within the six years surrounding Baseline we then wanted to visualize the distribution of their priority score across groups and so this shows that score with the audited group in red and the control group in blue and so this priority score is an IRS internal metric that ranks taxpayers for audit selection it's based upon their amount of balance due and the likelihood of being able to actually secure that balance due and so we were able to observe significant overlap between the two groups for this priority score and that helps confirm that our samples were constructed representatively you can observe that a higher number of audited taxpayers have a higher priority score and that's expected as the IRS is pursuing the higher priority taxpayers all right so moving on to our dependent variable so we're interested in monitoring or examining the impact of an audit on a taxpayers filing Behavior so we want to monitor this filing behavior for the five tax years prior to Baseline and eight tax years after Baseline I will say this analysis was constrained by the availability of data on these individuals so individuals were able to be identified if they were non-filers with at least a hundred thousand dollars in reported income within uh between tax year 2009 and 2014 but given the Persistence of non-filing behavior a lot of these individuals would disappear from Iris records in years outside of those Baseline years and so because of that our outcome variable fact of filing has quite a nuanced definition so it's set to one if a taxpayer filed a return in this research that includes any taxpayers that filed timely filed late or secondary filers mean that they were married filing jointly and listed as a secondary filer on their spouse's return a or the fact of filing is set to zero if a taxpayer did not file a return so this includes taxpayers that are non-filers that are identifiable by the IRS because they have income being reported to the IRS but it also includes ghosts and ghost here is a term used within non-filing literature to refer to individuals that are not present within tax records and so a taxpayer could become a ghost via two mechanisms they could be a ghost because they don't have a filing requirement meaning they earn below the filing threshold which in 2023 was around thirteen thousand dollars or they could be a ghost because they're actually a non-filer but their income is not being reported to the IRS and so the IRS is not able to detect them and this is where the research got a little bit tricky is because we cannot distinguish between these two types of taxpayers because they're simply not present within Iris records and a taxpayer that does not have a filing requirement is compliant but a taxpayer without a or a taxpayer that's a non-filer but but just does not have reported income is non-compliant so we had to determine how to categorize these with our outcome variable and from conversations with subject matter experts at the IRS and given the Persistence of income and Persistence of non-filing behavior we chose to categorize them as factofiling equals zero and in essence this assumption is saying that if an individual is earning or has at least a hundred thousand dollars in reported income for their Baseline year and is a non-filer it's more likely in later years that they continue to be a non-filer but their income just changes to a concealable source as opposed to them losing the majority of their income and not having a filing requirement but that assumption is one that we plan to conduct sensitivity analysis on so then we sought to visualize the timing of these audits in order to understand when we may begin to observe this indirect effect so shown in red audits began two to five years after the Baseline year and as shown in blue they tend to end about three to six years after the Baseline year so we hypothesized that an indirect effect will not be observed until at least two years have passed from the Baseline year as that aligns when the individual is being informed of their audit we then also wanted to look at the proportion of taxpayers filing over time with it within each group with the audited group shown in blue and the not audited group shown in red for each year from Baseline so you can observe that in year zero all individuals in our sample were non-filers and when looking at the years leading up to Baseline specifically here is negative five through negative two you can't observe a difference in the proportion of taxpayers that are filing between groups and so despite our matching efforts that difference is likely the result of inherent differences between groups where audited taxpayers tend to have more suspicious returns so that could be one reason another cause could also be our ghost assumption where we are assuming that all taxpayers in our sample have a filing obligation however in the Years surrounding Baseline specifically here is negative two through two we do observe parallel Trends in their filing Behavior where the proportion of taxpayer is filing in both groups decreases in the Years leading up to year zero and then increases afterwards considering years after a year for Baseline two we do observe that a higher proportion of taxpayers in the audit group are filing all right so moving on to our model so to estimate the fact of filing variable we used a linear probability model this model was chosen because it does not require any assumption or any assumptions regarding the distribution of the data and for ease of interpretability in the future we will employ other forms of models in order to conduct analysis on the robustness of our results but in this model we control for several characteristics the audit variable captures the difference in average filing Behavior across groups for all years the year from Baseline variables capture the filing behavior for each of the 13 years surrounding Baseline and then our primary regressor of interest is this interaction term between audit and you're from Baseline as this is the term that captures the indirect effect of an audit as it informs us of the relationship between an audit and each year from Baseline on that individual's filing Behavior we also control for several characteristics of these taxpayers and like I mentioned our research is constrained by the availability of data on these individuals so on this first pass through the research our control variables were a time invariant only being sourced from the third party reported information or the prior return information for that individuals Baseline here we identified these characteristics or from any variables that are considered significant in past non-filing literature and we sought to control for any of the taxpayers demographic or financial characteristics or their past filing Behavior then tax year is also a set of fixed effects that captures the yearly fluctuations across all taxpayers okay and so in looking at the results of this model we were able to confirm the presence of an interactive fact seeing that the audited group is 2.9 to 5.3 percentage points more likely to file in the 47 years after treatment we did observe a negative effect in the Years surrounding Baseline specifically is negative one through two for the audit group and and as within those years the or it takes until about year two or after year two for the audited individual to be informed of their audit so this helps inform us that there's a Persistence of filing or sorry a Persistence of non-filing behavior for audited individuals around the years of um non-compliance or Theory from Baseline near zero and when considering the control variables we also had several interesting findings we controlled for the presence of several visible income sources and we found that if those income sources were present it did increase the taxpayers likelihood of filing specifically investment income had the strongest effect at increasing the taxpayer's likelihood of filing by 9.6 percentage points we also found that taxpayers that reside in a state that taxes individual income increase their likelihood of filing by 17.9 percentage points and as is consistent with past non-filing literature we did observe a Persistence of filing Behavior where taxpayers that filed a returned in the year prior to baseline or in general 20.3 percentage points are more likely to file and taxpayers that were ghosts or not present in Iris records in the year prior to Baseline were in general 10.7 percentage points less likely to file okay so what can we make of these results so these results support the value of audits as a tool for encouraging future filing in non-filers like I mentioned we found that audited taxpayers were 2.9 to 5.3 percentage points more likely to file and the 47 years after treatment we also observed that the impact of an audit on future filing Peaks about five years after an audit and then Fades seven years after that we wanted to compare these results to another analysis looking at the indirect effects of an IRS enforcement activity on the non-filing population so a study conducted by data looked at the automated substitute for return program and interestingly the interactive fact of a field audit was smaller than the indirect effect of this program about less than half that of the asfr program however the automated substitute for return program looks at taxpayers that tend to have lower income and simpler returns so that difference in estimates could be indicative of higher compliance rates in lower income populations initially I know will bonding had conducted a study looking at the impacts of audits across the income Spectrum finding that audits on higher income taxpayers resulted in greater Revenue per dollar spent so this does just inform us that income plays a significant role when looking at the impact of audits so that concludes the majority of This research we have quite a variety of future research areas we want to explore going forward um it's important for us to obtain the indirect effects in terms of Revenue so we'd like to estimate the total tax model in order to obtain those dollar valued estimates additionally a good majority of our audited group experienced multiple audits or within this work stream throughout the years that we were interested in and so we would like to separate our analysis by looking at the indirect effects on taxpayers that are audited once by the sbsc field audits of non-filers and then also looking at the indirect effects of taxpayers that are audited multiple times within this work stream like I mentioned we also have this assumption that ghost taxpayers have a filing obligation so we would like to conduct sensitivity analysis on that ideally we'd like to be able to verify whether these ghosts actually have a tax liability in off Baseline years and that would likely require using statistical models to impute their income for years they're not present within our data and last our analysis was constrained by the availability of third-party reported data only being available for the taxpayers Baseline year and so ideally we'd also like to be able to obtain a richer set of control variables that vary with time and that would help us or by attaining that we would likewise be able to obtain data for these goes and that may require sourcing information from other government agencies or imputing the information using statistical models all right so that concludes This research so thank you [Applause] a lot of appendix slides my uh paper is estimating the compliance response to declining audit coverage this is Joint research with some IRS colleagues and miter colleagues and it begins with the observation that at least since 2008 we have had a rather steady decline in audit coverage rates caused by decline or certainly not increases in IRS budgets and that would normally be a dark cloud over Tax Administration and for for the most part it is but we got to thinking that well maybe there is a silver lining to that dark cloud in that it might present us with a natural experiment to see if there's a compliance response to the audits and so for example if you take the largest category of 1040s you find that there is indeed a downward Trend in the audit rate but a simultaneous somewhat upward Trend in non-compliance well the question is is there any connection between those two well maybe maybe not it's not clear like that suggests in other categories of returns other segments of the population and taxpayers don't necessarily form their judgments about their their own probability of being audited contemporaneous with what the actual audit rate is in that year it's probably some lag in the in the linkage there but of course also correlation doesn't mean causation and we're really trying to find causation there could be other IRS actions in fact there always are a complex set of enforcement and service kind of actions tax policy changes during the time could have an influence on taxpayer Behavior societal Trends maybe influencing behavioral choices so um I want to talk a little bit about what's been done so far in this area but before I do that I wanted to define a couple terms so we've sort of used similar terms here already in this panel one is what we would call the specific indirect effect and that is the effect of the audit on the specific taxpayer who was audited in some subsequent years so both of the earlier papers here focused on that in contrast there is also a general or at least we think there's a general indirect effect which is the effect on of the audit on the general population and getting signals perhaps from from some kind of network or news source but even that General effect can be defined and maybe explored in a couple of different ways one is what I would call a demonstration study can we even see whether this is happening anywhere and so these demonstration studies tend to focus on a particular population maybe a particular kind of mechanism by which the information flows from the audit to the unaudited people and so there are a number of studies that have taken that tack uh looking uh for example at tax preparer Networks or supply chain networks or Geographic networks as avenues for the flow of information about the audits to the unaudited people there's another type though that I would call a comprehensive study of indirect effects which isn't constrained to a subset of the population and it tends to be in fact explicitly agnostic about the mechanism for the information flow whether you know it's it's not constrained to being a particular kind of network like among preparers um so there have been a number of studies along these lines using State panel data zip code panel data or even Micro Data and so our approach is in its very infancy but here's some some initial results that we're getting our purpose is to isolate the general indirect effect on the compliance of people who are not audited in that comprehensive way not in a more focused way we use the national research program of Representative audits from the entire individual taxpayer population for tax years 2006 through 2014. and how do we do that well we apply econometric techniques to the micro NRP data to model the dollar amount that's misreported on the return as some function of the amount that should have been reported on that return to and things like the audit rate and other factors to control for now both the missed reported amount and the true amount come from the NRP audit the audit rate that we use is not the audit rate of the NRP taxpayer but recognizing that that taxpayer is representative of quite a large number in the population who are similarly situated and so it's the audit rate that they collectively face on average for that year and it's evaluated for each category of return there are like a dozen categories I'll speak to in a minute and so it's the audit rate faced by that category the category of return that that taxpayer was in but it's averaged over all the taxpayers that taxpayer represents and we lag that in most of our analysis by two years expecting that their their perceptions will be formed roughly two years after IRS makes some adjustments to Auto raids our main dependent variable or non-compliance measure is a net misreported amount and it's defined in the taxpayer's favor so if it's misreporting that reduces their tax it's positive and we can derive that for any single line item on the return or for any group of line items and so for income and tax line items it's defined as the amount that should have been reported minus what was reported but for offsets to income or offsets to tax it's defined the other way it's the amount reported minus the amount that should have been reported so for our Baseline analysis we evaluate that net misreported amount at the bottom line of the tax return essentially which we call tax after refundable credits Tark so that reflects any non-compliance that's happening anywhere in the return it all comes down to the bottom line and we also introduce a a category of of line items on returns actually six different categories and separate them by the extent to which those line item amounts are visible to the IRS and I'll explain that in a minute but we can establish the net misreported amount and the amount that should have been reported for each of those groups of line items well if you look at the population the IRS tends to for particularly for audit purposes categorize returns in a dozen categories that are mutually exclusive and exhaustive and as you can see for for all of them particularly that top one the auto rate was declining during that period that top one just represents three tenths of a percent of the population so just zooming in a little bit on that you can see the other ones likewise have the general Trend downward in audit coverage rate I've highlighted the ones in bold the ones that have at least five percent of the population in them so our Baseline econometric specification is to model the log of the net misreported amount modeled at that bottom line of the tax return tax after refundable credits and model that as a function of the audit rate for the the activity code the the group that the taxpayer was in now the correct amount various taxpayer controls and fixed effects for tax year and the taxpayer group the activity code and what do we find we get the surprising result that there's a positive relationship rather than a negative relationship between audit coverage and non-compliance so that it would suggest that the more you audit the more non-compliance you're going to get sounds counter-intuitive right well this is a just a high level essentially it's the entire population bottom line tax and it caused us to wonder are there things going on under the hood for categories of taxpayers or categories of line items where they're being non-compliant and so we did that same analysis for each category of tax return these activity codes that IRS exam uses and sure enough three of them had significant positive relationship with audit rates the counter-intuitive result one of them had a marginally negative relationship which was a bit more intuitive but most of them it's just a mixed bag it's not nothing clear when you do this by type of return so we looked at it by type of line item where where is The non-compliance Happening and so there are six visibility categories here the first four are representing income line items where category one is the the basically it's it's wages so it has a very high visibility to the IRS with third party information reporting and withholding ranging all the way to a category four which is the group of line items income line items that have very little if any third party information reporting to the IRS and then category five is line items representing offsets to income things like adjustments deductions exemptions and then category six is offsets to tax so what do we find when we analyze those groups of line items separately well it's basically the same specification except now it's the net misreported amount for the visibility group that's our dependent variable and then the corresponding correct amount on the right hand side again we get mixed results for the first visibility group we do get a negative response very significant but for the fourth one we get a slightly significant positive relationship and every the other ones are kind of all the way in between what does that tell us well this is a mixed bag and there are probably lots of things going on here the population is not homogeneous it'd be real nice wouldn't that just say you've got this and you've got that and it's it's a cut and dry deal there's a lot going on so we did some sensitivity analyzes we asked do taxpayers respond to a different lag in the audit rate than we were assuming we tried a three-year lag four-year lag and we found that our bottom line estimate did reverse its sign when you use the four year lag a longer ramp up period but it wasn't all that significant then we asked do only certain taxpayers adjust certain line items so we're breaking it down by type of return and type of line item at the same time and we found that our expected negative effect did happen for higher income taxpayers another question was do taxpayers respond to more aggregated audit rates in other words not just for their category of return but possibly similarly defined categories and that had a little bit of of a response or a fact um reversing some of the counter-intuitive results we saw earlier but the last question is do taxpayers respond to spending spending on audits the actual dollar amounts of IRS budgets spent on audits and um as opposed to expressing it as audit rates and interestingly that does seem to show the more expected negative effect more often well by way of summary we're just starting this research this is kind of fresh off the the research press if you will we're still learning a lot one finding is that misreporting on high visibility income seems to the misreporting tends to drop by 3.6 to 6.1 percent with a one percent increase in audit rates not a lot else is all that clear but for other line items it may be that for certain taxpayers and certain line items [Music] um we we do have an effect but we're still early in trying to narrow that down and likewise our results are mixed on misreporting by taxpayers earning more than two hundred thousand dollars uh who earn business income but we're thinking part of the problem there is that that's a group that's hard to detect their extent of non-compliance anyway and so we haven't yet corrected for non-compliance and we're hoping maybe we can do that somehow uh non-detection I mean we'll see so our next step is to disaggregate some of our our definitions or groups that we're looking at see if there's any sensitivity there we do want to convert our estimates to dollar values or more useful that way and we've got a number of econometric extensions that have come to mind that we're still in process of executing so that's it I'll turn it over to Will [Applause] all right excited to discuss these uh very interesting papers uh my opinions are my own and not those of the treasury uh so these are some really neat experiments uh there were some commonalities uh in some of the challenges that they're facing um so we'll talk about those first and then we'll go paper by paper uh for a couple of the papers it's not always clear what the bottom line is which is totally fine for papers at an earlier stage you know we're kind of on this road from what we tried to what you as the reader need to know and I think that's something to keep in mind as you advance these analyzes is to try and understand what are the things I really want to convey to the reader and then build towards that um just sort of an advice more for the writing I know this is something I try to do in my own papers and then in general I have concerns about statistical power so Theory would expect similar things for similar subgroups if we're finding really different results for people who live in town a and town B and they're right next to each other probably it's not that the model of the world you know how people respond to audits is different in town a in town B probably it's noise because the power is pretty low sample sizes are pretty low when you're working with random audit populations for example um one thing to keep in mind is that tests might be over rejecting here so you could try bootstrapping your standard errors and you could also think about corrections for multiple hypothesis testing especially if you're going to do multiple subgroups just because uh you know when you run 10 tests with a 10 rejection rate you're going to reject one one out of ten just from from noise um and finally you know being humble about the fact that we're in a world with low power here you know if you don't find an effect that doesn't mean the effect is zero means you just we don't know right like there's a lot we don't know here uh so on the first paper a big clarification question and also a piece of advice is that you start out with a randomly selected sample but then there's a bunch of profiling income matching issues are reviewed um and it sounds like there may be some correspondence there and then finally we get to the audits and it sounds like the analysis sample is the audited returns at the end of this process rather than the whole sample at the beginning of this process but the the beginning piece is the one that's randomly selected from the whole population and is also the one that's comparable to the controls right if you took the controls and did the same funnel to them the ones left at the bottom would look really different than the average across all the controls so the solution to this is to just use everybody and do the averages that way um and I think you can definitely do that and you'll probably get uh somewhat different results as a result um a second thing is I think one thing you could do especially if you want to think about how you're positioned relative to the other papers that have been written in this literature is what are the new things that you can speak to that other people haven't spoken to as much um and I think it's interesting to know about Australia for its own sake but it's also interesting to know things like uh what's the effect of being audited on the future behavior of small corporations for which there have been relatively few studies in other countries that I'm aware of um another one is that the Australian tax system is is kind of this neat natural experiment relative to a lot of the other countries we're used to thinking about because it has different features than the us and we could learn about how compliance is related to those features and whether those features are desirable or not and probably things like pre-filling which there's been a lot of conversations about lately Australia can sort of talk about well what does the compliance pattern look like for an income item that's pre-filled and do people like not fill in extra wages if they see that the the wages amount that's been pre-filled here is only this much um and the the final thought is there's a power and simplicity so uh for the whole sample uh you it's better to to pool especially because you have so few people right um and pulling across audit outcomes too also gets you that that size but also that comparability with the whole universe of people um as I was saying before differences across years there's probably a noise going on there um and then the cost benefit analysis for the app the audits is going to depend on the overall average not just on whether uh you are compliant so actually the reason people have split on compliance in the past is sort of to think about confirming that the audits are driving what you're finding but actually you have other ways you can do that by looking at like the most common kinds of non-compliance detected and whether there are changes on those particular income items in future years that's something that's really would be really neat to see oh I guess I have one last comment which is that like in some of the other papers it would be really nice to see year by year because that gives us the beforehand as the placebo test we can actually see it visually I'm sure it looks great because you've got great random audits uh and the trends over time are going to tell us how quickly the effect Fades which is something we really are really interested in this literature um and also you can say something about the discounted return over time okay on the second paper uh you have a really nice clear question and then bottom line this paper is sort of further along than the other two um but I think the place to go next is to really try to build up that contribution relative to the most similar paper by data at all in this area and so what they did was automated substitute for return which is sort of like we send you a letter that says hey we think you owe this much tax that kind of a flavor to it and I don't know the details of that in as much as I should but actually that's somewhere the paper should fill out because audits are more intense treatment than that and you're also looking at uh you know this High income population which you say drives most of this non-filing tax Gap actually you answered one of my questions which is great but then also you can probably improve on their methods and you should talk about the fact that you're improving on their methods and how you're doing that to sort of build out that contribution and say well here are all the other new things we're learning from this paper um my biggest concern is looking at this this graph of the trends over time if you look at the pre-period sort of your audited group is going down and your non-audited group is going up and that gives me a lot of concerns that what would happen if you hadn't had the audits here wouldn't be that similar and that's just like it really makes me very nervous um luckily you're in an amazing context for matching so you don't actually do much matching I think you do control but but really you have audits that have been selected based on a set of criteria you get to observe right and for non-filers you know we don't have you know it's not like we have a huge book of information on them we have very limited information you even see things like the risk scores and the amount of tax liability estimated from information returns to match on those so you could do basically really great matching on the Baseline and prior year characteristics get some propensity scores How likely are they to be audited based on all the characteristics we can see and then match on those or do a similar method called inverse probability weighting and basically then look at the trends for the years before the stuff you use for matching to make sure that those look good as a validation of your matching method and so actually then you you could sort of say well look the pre-trends look really really flat in the years before the stuff we used okay so uh moving on to the the final paper uh I think this paper could benefit a lot from leaning on some theory for some guidance um you know Alan did a great discussion of this but like thinking about it it's really about the perceived probability of detection and there are a lot of competing hypotheses for how people are reacting to to changes in audit rates like you know one story is that people are totally ignorant like I could believe that for at least some people uh another is that people have a hazy lagged idea about the change in audit rates like you know over Decades of experience from you and your friends you sort of learn How likely people are to be audited that lag could be decades not just a couple years and also it's very hazy because you you get so little information about it right like you only talk to a small number of people about whether they've been audited people are not always like giving you a table of audit rates by year by ZIP code by income level right and so the perfect information world at the bottom there seems to me somewhat implausible and some of the prior studies really leaned hard on that if you're doing by ZIP code right you're saying that like people are knowledgeable about the audit rate in zip code a versus zip code B and I just don't think that that that's that plausible um and another related point is that the probability of detection for wages and salaries is probably driven more by changes in document matching and what kinds of information returns report those kinds of income rather than uh by audit rates so if you could think about like you know stuff like uh 1099 income being an important uh component of long-term uh compliance there and in general like that motivates this this point about well you know if you're thinking about longer term Trends then you've got to start to think about longer term changes in technology and document matching Etc um so with that big picture uh supposing that people have these hazy-lagged perceptions uh the story is sort of that audit rates have fallen a lot and especially for very top taxpayers right um and you find that aggregate non-compliance in the amounts has been pretty much flat but one would expect Aggregates to grow over time with population and with you know just the the size of the economy and so actually like one one question I had was whether non-compliance as a percentage has actually Fallen even as audit rates have fallen um and that's a hard thing to get at but I think it's an interesting question that you can you're well placed to dig into so one one thing you could do is a relatively simple comparison between High income people for whom the audit rate has been much larger audit rate decline has been much larger levels at least in percentages the audit rate declines are actually uh comparable it's like you know half the audits across the board were cut or 60 or something but because the level was so much higher at the beginning for the high income people the the absolute percentage has fallen a lot more and you could maximize Power by splitting into fewer groups that are more matched up to what people think about rather than activity codes so like high income versus the rest or even like you know instead of 12 activity codes you could have four groups you could have business eitc High income other I know it's hard because the activity codes mean a lot for audit rates but actually like trying to convert them into something that means something to the reader is pretty important here and then actually I I just have a question I don't know what the nmp over time is for each grouping and like even just the trends over time are very interesting here so like you could do a lot more to talk about those uh one thing I I did inspired by having some of our Australian friends here was I looked at some tax Gap studies from various countries over time so I just grabbed the UK US Canada and some of the Australia ones and threw them on this graph and so the US this is one minus the voluntary compliance rate so this is like roughly an Azure measure of the percent uh non-compliant and I know it's going to be hard for some people to see this uh because the levels are so different uh but like you know the US level estimate matters less than the fact that the U.S Trend over time has been towards more compliance or less non-compliance here and if we look at all these other countries the trends are either less non-compliance or flat so you know maybe there's some story that's common across countries here rather than it being driven by whatever the change in U.S audit rates is and so that motivates me into this sort of concluding slide here which uh is directions for future research in this area I think it would be great to see some cross-country studies I think we're just starting to get to the point where we have enough things to be able to say something about relating Trends over time across countries to changes in technology adoption across countries and things like e-filing and document matching I think uh one piece of this literature that I always admire is that Raz has done some studies on the monetary costs of being audited but I would love to see more surveys on what are the psychological costs to taxpayers of being audited because everybody's story is that the audit was the worst year of my life and I would love to know like okay how bad is it really like in dollar value um another piece here is you could do surveys on that per CHP right like the perception of audit rates perception changes over time perception of changes over income instead of me speculating about well it's probably pretty hazy you could actually bring data to bear on that question I would love to see that and that wraps me up all right [Applause] foreign okay so let's take some questions here Janet these are great papers oh God that sounds so weird um these are these are great papers and I learned a lot and they're all related to of questions that we are all facing right now in this resource constrained world but my particular questions are for Alan Plumley um and I was particularly interested in the fact that you didn't just look at audit rates you also looked at the resources and the effect because uh that may be even more relevant in terms of the world of which we live there's an element that's connected to resources and connected to order rates and that's the productivity of the audits so part one of my question is to what account could you take into account the no change because we've seen particularly likely corporations in particular but perhaps other groups that during this period we've seen more audits that result in the taxpayers favor um I mean you got the pain of the audit but then you have the ultimate outcome and that may be driven by the declining resources and the impact of the IRS being outgunned the second related question is artists are not the only thing that has declined in the enforcement world we also saw a decline and then follow up to discrepancies information reporting so you see that there are fewer automated Under reporting cases that reach resolution as the declining resources have led to an increase in the threshold that triggers that so it's twofold taking into account unproductive Audits and also taking into account other forms of enforcement that are also funded by it also have suffered because of declining resources good ideas okay Alan let's work on that bro yeah good ideas okay next go ahead okay thanks so this is a question for India um I'm wondering if you can split so you have these very high income non-filers if you can split by the type of income just at a high level if it's like types of income that are more likely to be one time like large capital gain versus other types of income like wage or labor that you might think is ongoing so partly that's just like curiosity if you have a sense of the relative size of those but also relating to your assumption about future do they have a filing requirement or not you might have a different prediction for those two groups yes that's a that's a great idea and I think that's wonderful like add to our list to plan to explore in the future I know we were controlling for whether taxpayers had the presence of various income sources in their Baseline year at least so we controlled four if taxpayers had wage income versus retirement income versus investment versus the few types of self-employment income that are required to be reported to the IRS but we didn't look at the number or the proportion of taxpayers in each sample falling into each category and that would be interesting to explore over time especially the it would be interesting to look at the taxpayers that have wage income because I would assume that they may be in IRS records um or they may be more present in IRS records for each year of interest that we're looking at like they may be less likely to be ghosts so that's a good point so thank you other questions yes I have a question for Alan Partington uh the Australian Allen so one thing that you know we often think about with this type of research is it's really interesting knowledge gained you know what is the impact of our audits but to then translate from that to what do we do about that information in an operational context you know what are the decisions that we can make to improve our Roi on audits Etc how do we use that information so I'm curious to the extent that you can talk to us about that if you might share a little bit about what ATO is thinking about with this research and what you all might do with this information operationally uh thank you for your question um I think the the use to widget support um or the degree that to which we can use it depends upon some of the questions that remain unanswered from the analysis um uh sort of looking at our results you would perhaps conclude that audit said and not terribly effective at changing compliance but the general deterrence effect we don't estimate as part of our study and it may be reason alone to to continue with an audit program that we have or expanded even so I think until we can sort of shed more light on on that effect we'd be very cautious with making recommendations one way or the other about significant changes to what we currently do one other thing that motivated the study was to perhaps Benchmark the effectiveness of our risk-based program so to provide some sort of Baseline for determining the value add from that which uh which sort of supports Arguments for for engaging in that sort of risk-based audit activity and and sort of roughly quantifies the value ad from doing so one other thing we've been doing with the the results that I didn't put them in the slide deck but it was we used that to um uh also check the validity of other estimates that we make in relation to those wider Revenue effects and total revenue effects that we that we do report on so um included in those estimates is an effect using um a statistical processes and we find that we get quite different results for the random audits depending upon whether they're compared to a control or subject to the econometric estimation so that's something that we're very keen to get a handle on as well because we think it um it goes to the reliability of those estimates and we think we can fine tune it with this analysis hi Brian Callie Georgetown this is also a question for the Ozzie Allen so I was pretty surprised about the finding that non-compliant individuals complied less after audit so I wonder if you could say more about that I mean certainly there's Will's concern about you know this is all a selection story and the people who get through that process of actually getting to audit are more reactant in some way or otherwise just less likely to comply long term so you know is is that your story do you have kind of long run effects is this a bomb crater effect that does only persistent for a year or two anything you could tell me about that population I'd like to hear yeah what we observe is that the um the effect is more pronounced in the year immediately after the audit which is sort of consistent with the literature around that bomb crater effect um what motivates the drop-off in non-compliance we're still having a look at um whether it's uh you know an attempt to get back to some sort of permanent income or or that their expectations that we won't be back for quite some time uh might inform their compliance in in the year or two after the audit but yeah that's something we're still looking at okay yep go ahead um this question is more goes to some confusion in my mind about like high income definition just because I think about it more complexly sometimes about who might be a high income individual when they have Upstream entities and and defining at least the context of what the high income individuals for the study just because a lot of non-compliance may happen in a an S corp or a partnership where the income is reported on a higher level entity and expenses are taken maybe improperly upstream and just an understanding and I don't know if it's out of scope how to orient when looking at the study about what population it may be not addressing in terms of high income individuals yeah so in this research I guess it's better to refer to it as higher earning individuals because our income threshold was just individuals earning above 100 000 dollars and So within our research we didn't we kind of use that as the threshold for considering a taxpayer to be part of our sample and I think it will be interesting to explore like going forward the number of taxpayers that fall within each higher income bracket and see if we could do separate models for taxpayers that earn above like different thresholds or in different TPI classes and when you mean earning you're saying you're looking at wages like a W-2 because you know I'm just thinking like are you pulling in k1s or looking Upstream about what like earning that's just where my mind is going is the definition and maybe it's in your paper just wasn't crystal clear in my mind no important clarification so this is just an income that or compensation that's being reported to the IRS that's all that we can control for so so this includes um income from wages investment income retirement income various types of self-employment income and there's a lot of smaller income categories such as agricultural income or phishing income but the only income that we could consider was income that is reported by these third parties which are like various financial institutions and entities and employers I maybe would add to that that um the focus there of course is non-filers and so we don't have a return to to work with we could construct a pseudo return perhaps but when those taxpayers were audited they're probably closer I mean the Auditors are looking at more than just the the W-2s and 1099s and so um but they're classified by what's you know the the kinds of income that India talked about Alan if I could let me see if I can get to maybe what's at the heart of the question here k1's uh are there k1s that you're able to identify associated with uh non-filers and and how does that fact around I think that's was kind of at the heart of the question yeah we should be able to but I don't know that we have yeah I think the dollars in our study did have some income from k1s and so yeah I think the IRS is able to identify K1 income and I think that capability has been increasing over the past few years okay we're we're about a time here I'd like to thank the presenters and or discussing and everybody for the questions [Applause] foreign to introduce one of my favorite columnists Catherine rampal Catherine is a opinion columnist for the Washington Post where she covers economics public policy immigration and politics with a special emphasis on data-driven journalism and recently she's written a number of pieces on the tax system and compliance she's also a commentator on CNN a special correspondence for PBS NewsHour and a contributed to Marketplace before joining the post she wrote about economics and theater for the New York Times she's won many awards and I'm sure you'll find her presentation interesting so thank you Catherine [Applause] thank you thanks so much for that lovely introduction um whenever I speak before a tax crowd I always like to specify You Are My People um quite literally because I come from a long line of accountants my dad was a CPA my mom was a CPA my grandfather was a CPA my grandmother was a bookkeeper um if that counts and my uncle is sort of a tax attorney so in any event um I feel like it's it's somewhat in my blood and I love writing about taxes I write about a lot of other kinds of things too including immigration policy as you just heard budgets more generally safety nets various other kinds of economic public policy issues and I'm often asked how I choose the kinds of topics that I write about because I do have a lot of freedom in my job as a columnist and I think my usual response is something like I tend to gravitate towards areas um that present a lot of complex issues where I think journalists can add value by helping explain things and unfortunately well fortunately or unfortunately there are a lot of those kinds of areas right now um and there are you know tons of government policies and procedures that are needlessly complex uh and one of my most deeply held beliefs I think as a journalist and as a citizen is that complexity tends to reward demagogues right the more non-transparent to policy is the easier it is to lie about or to misconstrue in some respect and I think that's often true of tax policy for Better or For Worse certainly gives me plenty to write about um so I am going to uh talk a little bit today about things that lots of people get wrong including myself um particularly when it comes to tax issues and I I say everyone is wrong about taxes I'm sure that doesn't apply to this audience but you know just for the sake of argument so as I'm sure you are well aware Americans have very strong often very wrong views about uh how the tax code works or does it work and um some of that is born from genuine confusion and some of that is born from the narratives that they hear from politicians from journalists and others and some things are genuinely tricky right like explaining bass lines to re my readers is the bane of my existence uh it is so difficult to explain without like having a chalkboard behind me which obviously I do not have the luxury of doing when I'm writing for a newspaper um but baselines counterfactuals very counter-intuitive uh or at least not I would say not intuitive to understand and I think it's reasonable that people have legitimate levels of confusion about how those work and and those are understanding baselines is obviously relevant to understanding whether policy has delivered on the objectives that politicians have laid out and how voters respond to that so things like you know if tax revenues Rose after a tax cut does that mean that the tax cut led to the higher revenues um or for that matter did it lead to them for the mecca the the reasons that were laid out by The Advocates of those tax cuts I.E sort of supply side things as opposed to demand side um explanations and it's very hard to explain that to the public like what would have happened in the absence of a policy so I get why people might be confused and and look I think reasonable people can make different assumptions about questions like that as well then there are other things that are I think a little bit sillier that are less forgivable that people or especially people in my business journalists get wrong so things like um you know the complexity of the tax code is about the number of tax brackets we have um as opposed to how difficult it is to Define income like I get emails all the time from readers saying like the tax code is so complicated why doesn't why don't we just have a flat text I'm like okay we could have a flat tax we could have one bracket but what's income and it's it's hard to have those kinds of discussions um and then there are other silly things like people don't really understand how marginal tax rates work in general they think I have seen this mistake made by journalists that they think that when you hit a particular level of income it's not the marginal dollar that gets gets the marginal tax rate the higher tax rate but like your entire income so like every tax year every year on tax time you see these stories about like people trying to avoid making whatever the you know 250 000 or whatever the the ceiling is because they're afraid that their taxes are going to go up um uh on all of the dollars that came before that and then um there are other kinds of issues that I think people tend to get wrong in part because there there are vested political interests in people misunderstanding them so that might be things like um whether the tcga only benefited the rich so it is true if you look at the tax policy centers data in fact and I cite it all the time that the 2017 tax bill was heavily weighted toward higher income people but almost everybody got a tax cut and most Americans do not understand this and I think that's partly because you know the tax code is very non-transparent there are a lot of moving parts and those moving Parts interact in different ways and they determine how your tax bill changes by the end of the year um so you end up seeing people say well I think that you know my my taxes must have gone up because my refund shrank no um or you know a lot of people think that because they're no longer deducting the the full amount of their state and local tax uh that that must mean by itself that their taxes went up and in fact another uh frequently cited tax policy Center report that I often use I don't know if this looks familiar if anybody in this room is actually responsible for creating this but basically it shows that even in blue in even in like the deep blue States the more democratic High tax states um a very small subset of the population actually paid higher taxes on net even when you account for the fact that they can't deduct their full salt because again rates went down Etc et cetera and the state with the highest share of people who saw a net tax increase so again once you take into account all of the different things that happen in the tcga was New Jersey we're only 10 percent of people saw their taxes go up but if you've met anyone from New Jersey they all unilaterally think that their taxes rose right and so there so there's this misperception that's partly born of the complexity of the tax code partly born of the the narratives um that get cast upon it um and more broadly one of the things that I have been trying to focus on in my writing that I think affects uh how people both how people interpret the tax system as well as what they demand of their political leaders and how this filters into electoral decisions is this idea of whether or not we are a high tax Nation and whether or not the tax burden is too high relative to the services that Americans themselves consume and this seems like you know it should be a relatively simple question and it's not um objectively speaking I would say that we are not a high tax Nation relative to other oecd countries for example um but then there's this this question about like uh whether people have already paid enough into the system to account for what they're taking out of it I get emails all the time from people whenever I write about Medicare for example saying I paid for my medicare now I don't know I don't think Gene Sterling is here but he puts out this annual report every is he here no he puts out this annual report with some colleagues at Urban um that basically shows on average you know for different for people at different points in the income distribution how much they pay into Medicare and Social Security and how much they are expected to take out even when you account for you know the time value of money and all sorts of other things how they could have how they could have alternatively invested their um payroll tax dollars and the Medicare numbers like will blow your mind that people consume much more in health care than they pay into the system and have no idea about any of that again that's partly because the tax system is not transparent and that's also obviously because the Health Care system is not particularly transparent and I think part of the reason additionally is that for many years conservatives conservative politicians have made the argument that tax cuts are too high and they need to come down um to some extent they said it could be taxes are too high and they need to come down to some extent they've also said that spending is too high but lately they've been less committed to that to that particular principle but I think that there was really a change like around the 80s when Republicans um became sort of anointed themselves as the tax cut Santa whereas Democrats were the spending Santa I don't know if any of you know this this famous Wall Street Journal op-ed in the 80s but it basically talked about it was by this guy Jude wininski basically said that um Republicans needed to stop demanding fiscal responsibility all the time and and reward voters with tax cuts the same way that Democrats were rewarding voters with spending and I think it used to be the case until relatively recently that Democrats in service of arguing for a broader social safety net may also made the case for higher taxes uh taxes being the price we pay for civilization as the saying goes and that has changed it certainly has changed over the course of my career as a journalist and I've been doing this for 15 years let's say um it feels like in more recent years Democrats have been sort of conceding the argument that most Americans are over taxed as as they are they still deserve more services but they are currently over taxed and the only people who deserve to or who can afford to pay higher taxes are quote unquote the wealthy which turns out to be a vanishingly small slice of the population that gets redefined higher and higher and higher in terms of where it is in the income distribution Obama said that the only people who could afford to pay higher taxes or who should pay higher taxes were those earning over 250 000 Biden said over four hundred thousand dollars um more recently I don't know if you any of you remember this a very glamorous dress from the Met Gala a couple of years ago Alexandria representative Alexandria ocasio-cortez where the stress that said tax the rich she was subsequently asked about what she meant by rich people and she said um that she meant quote excuse me people with quote hundreds of millions of dollars if not billions of dollars unquote not quote a doctor or a lawyer unquote um who are I'm sorry to say objectively speaking at the higher end of the income distribution you know if we're talking about high income people they are certainly in the U.S and in the world uh you know maybe not as rich as some of their neighbors who run hedge funds but they're still high income um and I I have different theories about why this might be that Democrats seem to have sort of conceded the issue and stopped making the case for higher taxes more broadly again I think it's partly about this sort of bombardment of anti-tax messaging from conservatives I think it's also partly that the Democratic voter base has changed quite a bit in the last few decades to become more Highly Educated higher income people who don't necessarily want to think of themselves as Rich they're neighbors who are the hedge fund Partners they're the rich ones you know I'm I'm a lawyer um I or whatever uh doctor accountant you know some professional some other professional class person um I'm you know comfortable but I'm not rich and when I have argued with gently people in the Biden Administration about this pledge that has been made about four hundred thousand dollars being the threshold for Rich which is depending on what definition of income you use like usually around the top five percent or top two percent of the income distribution um they basically these members of the administration who shall remain nameless basically counter that they have no choice that you can't win elections by telling people that their taxes are too low you have to you have to tell them that they are overtaxed um and my view is I get that like you don't want to tell people boohoo like you're not paying enough but I think if you're making the case for essentially a more of a European style welfare state you need to make the case also for European style tax system whether that's a vat or otherwise to help pay for it and now that there is relative unanimity among the political parties that the public is over text unsurprisingly the public increasingly thinks that they are over taxed um no I I can't say for sure in which direction causality runs um but I think because there's been no pushback to this narrative that um Americans are not paying enough taxes to sustain the size of the the welfare state that we have now it is probably unsurprising um that gallop in its most recent poll finds that the highest share of Americans in about two decades think that their own taxes are too high that's I don't know if you can see this but it's the the green line at the top so about 60 of Americans now say that their taxes are too high and if this is not an outlier Pew has also asked similar questions slightly different wording um over time and uh uh and by the way you know people's tax burdens have generally been falling in this period in these recent years particularly in 2020 2021 because there were a lot of temporary pandemic era tax related policies like the stimulus payments which were technically tax cuts um but anyway so Pew has a similar poll uh where they asked people pay more than their fair share less than their fair share about their fair share of about the right amount of taxes and I had them I asked them to break it down for me by um demographics so again I hope this is legible to you but even about half of Democrats now say that they are paying more than their fair share of taxes so I think these kinds of narratives do matter and uh it is true for views of the tax burden today as well as in the past that people really mediate their views of the tax code their views of their own tax liabilities through these narratives that they hear so again this is not entirely a new phenomenon for example if you look back at um the tax cuts that took place during the Obama era which Americans do not remember um and certainly didn't notice at the time here's an article from 2010 uh where people were asked do you think the Obama Administration has increased taxes for most Americans decreased them or kept them the same and the administration actually had cut taxes this was part of the first the response to the economic crisis the Great Recession and the financial crisis only 12 percent say knew that taxes had been lowered about double that said that they had been raised so you know uh again what happens in the real world is sometimes different from people's perception in part because the tax code is so opaque and it wasn't true only for Obama this was uh 2004. bush very similar poll uh only about one in five people said that their their own tax burden had been reduced under George W bush and that was like part of his whole shtick right was was cutting taxes um but there was a broader um pushback to that narrative that was only going to rich people and again disproportionately to higher income people but not exclusively to them but that's how people saw it and it has been argued to me that maybe there was an asymmetry here that maybe people would notice more if their taxes go up than if they go down I'm not aware of any research on that but if those in the audience are working on it tell me because I'd love to see it I think we don't know for sure in part because again people don't really understand what's happening to their um to their tax burden over time um and I think there's also an open question of who Americans would blame if in fact they notice that their taxes are going up and they're actually going up certainly the Biden Administration has made the calculation that Democrats are going to be blamed no matter what um and I think that partly explains their decision to interpret the that pledge that I talked about that you know no no tax increases for people making under four hundred thousand dollars uh to apply to the current policy Baseline as opposed to current law Baseline by which I mean there had been some debate within the administration about whether to support an extension of the Trump the tcga tax cuts which are set to expire in a couple of years less than a couple of years I guess at this point um will would the expiration of a tax cut be the equivalent of a tax increase and I know they're probably people in this room who are pushing for the administration to say no like that we are not considering that a tax increase we are going to allow those tax cuts to lapse um they ultimately decided against that in their budget this year they said that they are in favor of extending the Trump tax cuts for you know the bottom whatever it is 95 percent of the income distribution even though no no Democrats were in support of that when it passed as you may recall um but that's the that's the way that they've staked out the issue for themselves I guess because they believe that's that's how the politics would play out that they would get blamed if the Trump tax cuts lapsed in taxes Rose um so in any event a lot of what I do as you can tell is sort of explain I write an opinion column but it's sort of explanatory I'm trying to push back against what I view as false or at least confused narratives who's actually upper income whether people have already paid for their benefits that they might be receiving um whether tax cuts pay for themselves and lately as Eric mentioned I have been um really interested in compliance issues that even if tax cuts don't fully pay for themselves tax audits might and in fact they might more than pay for themselves so this was a a fun piece that I did last week based on the research of a group of economists one of whom may be here I don't know if hi we haven't met because he's not allowed to talk to the Press um but I've talked I've talked with your co-authors uh looking at some very cool data are you presenting it here okay not today well it's it's awesome you should you all should check it out but basically he and his co-authors um looked at internal accounting data from the IRS to see how much it costs to conduct audits for various kinds of taxpayers uh when you account for the wages paid to the IRS employees how many hours it takes the overhead cost Etc and then how much money comes back in return so this was one of the charts we put together based on that work um that basically shows that for the bottom half of the income distribution the IRS only about breaks even in terms of how much money it gets back from the initial you know from whatever's recovered from the audit relative to how much it costs to conduct the audit whereas at the very top of the income distribution it's like six to one and then um that doesn't take into account deterrence effects I think there was a paper presented today on deterrence maybe or will be um and that shows even higher um bang for the buck and then additionally on the mo there's also been this long debate about the marginal dollar of spending for uh tax enforcement does that have more or less bang for the buck than the average dollar because maybe the IRS has already gone after the really high value Audits and whatever's been left in The Cutting Room floor is less valuable or less you know less likely to have a high Roi and in fact they they do not find evidence of diminishing marginal returns they show that uh because of the cost structure you know because most the costs are fixed and um and because the IRS has not been sort of maximizing Revenue in terms of who it's auditing that's not the case um so anyway I find that kind of research super interesting and a lot of it is somewhat counter-intuitive and I think readers are really into it um and uh you know not to mention that other kinds of investments in the IRS can also pay off including things like investing in in better I.T and customer service another thing that I love writing about um I don't know if there's any evidence about how that affects voluntary compliance rates but I'd love to see it if there is because that would be interesting to write about but at the very least it does seem like processing returns more quickly um is is a valuable thing to taxpayers and in terms of getting money in to the government's coffers faster that's that's helpful I took a little field trip to the IRS to a processing center Last Summer where I got to see the cafeteria this is the cafeteria um because the agency was so backlogged with paper returns partly because their I.T is bad you know that factor colliding with the pandemic disrupting operations they just had every like every square inch of this this processing center in Austin was just covered with paper and this was their cafeteria um I am told that the cafeteria has happily returned to its state as a cafeteria and the papers are gone the stacks of papers are gone um but it was still kind of wild to see and I also got to see some of what the I.T system looked like pre-investment in from from the inflation reduction act last year including the fact that they did not have scanning technology and instead there I don't know if you can tell but this is an IRS employee who is uh manually typing in digit by digit the materials from a paper tax return into the computer system because they did not have a way to scan it they have since bought scanners so that's very exciting we've joined the late 20th century finally um and then there are a bunch of other improvements that have happened already even though we're relatively early in the period of this this reinvestment in the IRS um the agency was able to answer literally millions more phone calls this year than it did last year cut down on its long waiting times from I think an average of 27 minutes in 2022 to four minutes this year and these triumphs are the result of getting more resources and one of my favorite things to write about is investing in administrative capacity and I think the IRS is a great example of why um better government is not necessarily cheaper government that you need to actually have the basic infrastructure to get stuff done then I think so here I'm getting to like the stuff that I think I've gotten wrong um or that I'm I'm working to do better on I'll put it that way I think there's an Untold Story that I've been trying to figure out how to write on how investing and enforcement can also improve the customer experience which is sort of I think not an intuitive narrative for for most taxpayers um I think among the biggest mistakes that the administration made and those of us who were supporting uh more funding for the IRS made was not sufficiently explaining how better enforcement is actually how how stronger enforcement or more sophisticated enforcement is actually good for taxpayers as opposed to just being more likely to harass people and um you know there had been a lot of focus in my own work I did this other as well talked about how you know cracking down on tax sheets Etc I think that fed into this narrative about the 87 000 gun toting IRS agents who are going to you know knock down Joe the Plumber's door or whatever and you know and seize everything all of his hard-earned income even when he was honest it fed into that narrative and it's very hard to convince people like no they're not going to go after you they're going to go after these these other people and frankly I have asked the agency repeatedly like so how do you actually determine how this is going to work and I don't you know how you're going to direct these dollars and I don't always get a satisfactory answer that I can at least that that I understand well enough to be able to explain to my readers um and so I think that subsequent to the focus on the Crackdown on tax sheets narrative there had been the bite Administration Democrats and others had sort of switched to talking more about the service elements of the uh of the investment the IRA investment in the agency and less about the enforcement element but again I think that they are related in that better enforcement in the long run protects against tax rates Rising right because if you're collecting more of the tax liabilities that are already owed then you don't have to make up the shortfall by raising tax rates on the people who are already paying the correct amount as we have seen happen in lots of other countries where tax morale is worse and and there's more rampant tax cheating but the other thing is that if you invest in better enforcement capacity and better I.T and you can better Target whom you want it um you can leave the people who are paying their taxes honestly and correctly alone um and I think this is a this is part of the reason why Americans are so skeptical about the way that these dollars will be used because like if you look at um the percent of audits that results in no change and I I think this might be your chart yes um uh it's it's quite high and it's been quite high like about 10 of audits don't result in any um in any change in the uh in in the taxes owed um so if you get better data if you make partnership returns auditable you know digitized and accessible not only by manual combing through of the returns but like have a have a system where you can search things and the you know maybe someday you use AI to figure out this like science fiction I realize um given the state of the it right now but you could use AI to figure out like what are what are the actual red flags that are more useful for selecting whom to audit I think that would be that's a good story to tell it's a complicated story to tell but it's a good story to tell to taxpayers and and in fact I think there is a universe in which better I.T doesn't necessarily result in more audits but fewer ones right and yet still more money being collected because the agency is focusing its resources where they are most likely to pay off and and not bother not not taking honest taxpayers people paying their taxes correctly anyway dragging them into this this uh auditing Dragnet but right now you know we're not there um and so I think the the rhetoric on these kinds of policy initiatives has distorted the goals somewhat and again I'm partly to blame for this I think because I've been talking so much about cracking down on on tax cheats and doing more audits as opposed to smarter audits um and I think you end up with more support for kind of punitively oriented measures um both in tax enforcement and the design of taxes themselves I think that's part of the reason you see a lot of public support for windfall taxes for example it's the idea that you know taxes are not necessarily about raising revenue efficiently or correcting for market failures it's about like paint getting payback um from from either individuals or companies that you think are making too much money um and I think that uh you know that's that's a bad place for our discourse to be moving it's true not just for taxes I think there's a lot of more punitively oriented policy discussion as opposed to problem fixing policy discussion that I find somewhat distressing and the goal should be about again raising revenue efficiently to fund the size of the government we want not about punishment for punishment's sake because we're resentful about people making money you know find ways to to get the money we need as fairly as possible however we Define um anyway those are my comments uh I'm happy to take questions if folks have any um scale back I'm not aware offhand of pull data on on the the recent debt limit Bill if that's what you're referring to where there was a deal or there was a some language in the bill and then some side deals about rescinding money from the IRS I'm not aware of anything on that it may exist my recollection is that when the IRA the inflation reduction Act was passed I believe the one there was polling done about the different elements of the bill so like tax credits for electric vehicles and the prescription drug negotiation price prescription price negotiations and things like that um those things were relatively popular and the funding for the IRS was not but I'd have to go back and check to confirm it would not surprise me particularly given the rhetoric uh you started out by saying that a lot of Democrats don't think they're among the rich do you have any polling data that they're changing their mind um I think I have looked into this before but let me get back to you on that because I don't want to I don't wanna make something up but I I think if you look at the demographics of the democratic party they have certainly changed people um it used to be that the party relied more on you know this the the shorthand is like non-college whites right um and though that contingent has become more Republican and those are people I'm sure who would not consider themselves High income it is the case that over time um very few Americans consider themselves High income so it's just a matter of which party that they're they're voting for like everybody thinks they're middle class it's another one of my favorite hobby horses is to like see what is middle class and it's whatever I am and you know anybody richer than me is not and um so press the button okay assuming that um Biden wins the primary and there's some Republican candidate who emerges do you anticipate any discussion of taxes in the coming election year uh let me rephrase it any good discussion any good discussion a bad discussion but anyway any mention of it uh well good and bad I guess are in the eye of the beholder I can tell you that I think that the bide Administration has absolutely committed itself to we will not raise taxes um we will assume away any fiscal problems that currently exist and certainly that would come from expanding the safety net and we will say that it can all be paid for by taxing the rich whatever that means now again if you look at their budgets like when I was saying that the I don't know if I mentioned this but when the Bible Administration said that it acknowledged that it planned to support extending the Trump tax cuts they did not account for the cost of that in their budget estimates I don't know if I already mentioned that so I think the discourse is not likely to be great the one area in which I am somewhat hopeful and maybe I'm wish casting here is on the trial tax credit so the child tax credit has historically been supported by both parties expansions of it have been supported by both parties over the years there was a major expansion of it as you all may recall in 2021 as part of the American Rescue plan it became a monthly allotment and it was fully refundable it was much larger there were aspects of that that I'm very supportive of including making it available to the lowest income families I don't know that families making four hundred thousand dollars need to get a bigger tax cut but that was that was the version that went through there has been periodic discussion as you probably know in the last couple of years about Reviving some version of that now the devil's in the details oops sorry it's rebelling against me um so I don't know uh but I I hope that there is a version of that that could be revived that Republicans and Democrats could get behind yeah you had pointed out earlier that um the complexity of the tax code is not um really derived from the multiple tax brackets and so on but it turns out that does greatly um harm or at least make more complex withholding particularly for multiple incomes and this kind of thing it makes it really complicated even though it's a fairly simple concept so food for thought that's fair but I don't think that's the main reason why the tax code is complex and why it takes people so many hours to prepare their taxes every year did you write about the underfunding of social social security and Medicare and did you give your own prescription I have written about that periodically um my I'm trying to remember for what I've said about my own prescriptions I have sort of a mixed bag of views I I think it is unhelpful that both parties at this point seem to be dismissing the idea to some extent that there's a problem at all um initially look like Republicans were going to lead the charge to do something about entitlement reform Democrats opposed it you may recall at the State of the Union um Biden got ever you know said they all want to cut Social Security and Medicare and then the Republicans booed or you know had some kind of some kind of reaction and he said okay so we all agree we're not going to do that and the consensus seems to have been that nothing needs to change uh if you look at the Republican study committee report that just came out of the house they do say that some things need to change I think it's probably going to be a combination of um of taxes and spending fixes like increasing the wage base for Social Security I think is it kind of a no-brainer some of the other things I think are going to be trickier to figure out like raising the retirement age or curbing benefits or making them more Progressive or or what have you um and I don't know exactly where I come down on that it's a really hard issue all right we time for one more question yes so I'll change the topic a little bit um you know I think we're all interested in this room in taxes obviously but I'm wondering if you have any tips you know we want to get our research out we I think all of us would love for more people to write about taxes do you have any tips for us as academics people doing research the kind of better inform and get better opinions out to the media or you know uh reporters who maybe don't have a family tree Laden with debits and credits uh so you know write about taxes and kind of a smart and informed way um obviously reaching out to journalists is helpful um there are a few journalists who cover taxes full-time I mean as I said I do not cover taxes full-time but there's some people who are really excellent Rich Ruben is great I I think he's like in an invaluable reporter on these issues so making connections with those people is useful now the thing that I think a lot of academics are not as good about understanding is how to put their research in the context of ongoing policy debates or some in the journalism business we call it a newspeg like what is the what is the what is the news relevance to your research and pitching your work in that context I think is much more likely to get coverage so even when I wrote about the the um policy impacts research uh on audit rates I found out about that months ago because one of the co-authors had uh presented a version of it at the American economic Association and it was super interesting um and I could have found like almost any news Peg but I ended up writing about it in the context of this debt limit bill this debt limit agreement just passed how much does it cost which was not directly addressed in the research but that was like a nice hook for talking about it so thinking about those kinds of things I think will be useful all right all right thanks everyone appreciate your questions [Applause] thank you if you had something all right good afternoon everyone we're gonna go ahead and get started my name is Russell James and I will be the moderator for session three uh this afternoon session will focus on understanding contemporary taxpayers we have a dynamic presentations for you guys today and outstanding speakers our topics are in the form of questions today and what we're going to do is Chase those questions uh with the presentation you'll find the answers from there so at this time I'll go ahead and present the presenters we have Emily Lynn who are married father separately and why should we care we have Carlos santizini willing but unable to pay the role of gender and tax compliance and then finally we have Tyler menzar who sells cryptocurrency and then last but not least we have my colleague from the IRS Yang's son who's going to be the discuss it for today so at this time we'll go ahead and turn it over to Emily to start us off thank you guys so much foreign thank you for having me um the paper I'm going to present today is a joint work with my treasury colleague navodias American raccoon in this paper we investigated understudy Topic in federal income tax specifically we analyze returns claiming Mary filing separately status and the exam the associate tax issues the usual disclaimer and disclosure review apply to the paper into today's presentation so many individuals can choose between filing jointly and separately and we know that filing separately generally leads to a higher federal income tax liabilities however probably because there is no single formula or condition to determine when to file separately the extent and the magnitude of the tax disadvantage associated with this filing status is not well known many online posts are available to inform taxpayers when it makes sense to file separately IRS on the other hand are devices taxpayers to calculate tax both way and pick the filing status that gives you the lower liability but the reality is that very few returns were filed as Mary filing separately in taxi year 2020 2020 2.4 percent of federal income tax returns were memory filing separately or mbfs returns compared to 34 percent of joint returns my refunding jointly if we count each joint return as filed by two taxpayers then 3.4 percent of Mary filers file separately in taxi year 2020. so although MFS filers make a very small share of total taxpayers as you will find out in this presentation the filing status is associated some complexity equity and compliance issues worth about attention uh let's begin uh understanding this filing status a little bit so of course this filing status can only be used by married individuals and um memory status is determined based on the individual status on the last day of the tax year for the vast majority of taxpayers this rule or definition is very easy to understand and straightforward to apply but for individuals going through a divorce or separation it may require a closer look at the definition to determine the appropriate filing status so under the tax codes specific Court actions are needed for an individual to be legally separated for tax purpose maybe contrary to some common sense some legally binding agreements or certain court documents does not count and the law also provides an exception for a married person to be considered and married when the so-called abandoned spouse rules are met so the rules apply to married individuals who have qualifying children who pay for more than half the cost of maintaining the household in which they live with their children and the spouse is not the member of the households during the last six months of the taxi years when all of these rules are met the taxpayer can claim the head of household even though they are married are many tax provision contribute to the unfavorable tax treatment of MFS status first off the length of a tax brackets for MFS status or half of those for the joint status so as far as the brackets or counselor and filing jointly will never result in the higher federal income tax liability than filing separately and also MFS filers have limited eligibility for tax credits they are they cannot claim the education credits or adoption credit and they are only eligible for the earn income tax credit and premium tax credit under very specific conditions Ms filers also face restrictions with spec with with respect to their claims of exclusions and deductions and all this rule together also makes MFS filers less disadvantages relative to single or head of household filer all else equal but for the marital status so if this status is so unfavorable why do taxpayers still use it um in limited situations filing separately can actually reduce couples tax liability this occurs when a spouse has low income but significant itemized deduction that is subject to an income flow so in this case the couple may be able to claim a higher amount of itemized deduction and reduce their liability and on the separately filed return there are other tax reasons related to non-resident spouse or when the couple could have a lower federal and state liability combined by filing separate returns and our paper focus on only different federal income tax also according to our review of online post married person may consider filing separately for other reasons despite having a higher tax liability for example filing jointly may not be an option in the case of domestic abuse or spousal abandonment or can be difficult for those in the prolonged separation some merry individuals may want to protect themselves from being liable for the spouse tax debt or a spouse refund offset so they file separately and also couples May simply want to keep financially independent so they file separate returns it is also uncommon for tax professionals to advise couples in the process of divorce to file separately so that they may avoid the hassle of potential audits on their previously filed joint return after they get divorced and finally if the couple has large student loan expenses subject to an income-based repayment plan filing separately might reduce the payment thank you so in the past decade there has been gradual but slow increase in the relative and absolute presence of MFS returns but the share of returns following this status is still very small as I mentioned earlier in 2020 less than three percent of federal income tax returns to our MFS returns filed by just under 4 million taxpayers probably because of the assorted reasons why taxpayer chose to file as MFS Ms filers income is quite diverse compared to other filers mfmfs filers are slightly over represented in the middle income ranges but a non-trivial proportion have income below thirty thousand dollars or above a hundred thousand dollars Ms filers are also diverse in terms of a duration for how long they claim this filing status so in this analysis we pull the population of MFS returns filed for tax years between 2013 and 2021 resulting in nearly 32 million returns filed by 13.4 million unique taxpayers we found that over this nine-year period of those who ever claimed this filing status half of them use it for just one year 80 percent use it for no more than three years and five percent use it for more than six years and the estimated uh duration gets a bit longer if we account for the sensor left sensor the right sensor data sensor sensory means that some observation in our sample actually begin or continue outside our observation window so for example in our file we have taxpayers who began filing separately before 2013 or continue filing separately after 2021 and we can adjust for that in a duration model so when the model accounts for this data sensory of course as expected we get more longer spells of MFS claims but the diverse patterns remain still half of MFS claims and after just one year more than 70 percent and after three years and 12 percent last after eight years so on an MS return taxpayer is asked to enter The spouse's tin or taxpayer identity identification number so in principle we could match the two spouses MFS return and then calculate the potential joint liability had the capital filed jointly however in our file we have 17 percent of Ms returns have missing 10 spouses 10 and 11 percent um have spouses filing a MFS return so for the rest of the 70 to 72 percent of Ms returns in our file we could identify The spouse's return or when the spouse is now filed we assume the income is zero and then we could simulate The Joint return as a couple level and compare it with the combined liability on the two Ms returns to determine whether the taxpayer has a separate filing penalty or separate filing bonus uh we use the five dollars as the threshold so the tax difference needs to be over five dollars in order to be considered penalty or bonus and this slide uh shows the simulation outcome 63 percent of uh the MS filers in our calculation have separate filing penalty meaning that they pay more by filing separately then pelling then and filing jointly the average amount is twenty two six dollars about a quarter of Ms filers would pay the same federal income tax between filing separately and filing jointly and 13 it should be issued with 13.3 about 13 percent of Ms filers have separate filing bonus within average of 11 30 meaning that they pay less by filing separately than uh filing jointly so the majority of taxpayers claiming this filing status actually pay more by doing so even when there is a tax benefit to do so the tax bonus is modest on average so next we associate the penalty and bonus status with the taxpayer and return characteristics we found that the bonus status is associated with itemized deduction uh taxpayer income and taxpayer age and the penalty status associated with a receipt of eitc on the joint returns uh Ms filers generally cannot claim the eitc except for very limited situations that were only made possible under American Rescue plan Act of 2021. so as you can see from the table 15 percent of Ms MFS filers who have a separate filing penalty would claim the eitc on the joint return and this percentage is much higher than the other filers who do not have the separate filing penalty we then link the separate filing penalty to the decision to claim this filing status we model the decision on whether an MS filer chooses to continue to file separately given that they have claimed this filing status for a certain number of years so the terminology for this probability is the survival rate of survival probability at time t we estimated the survival probability by income quintile in general the higher the taxpayer income the more likely that the taxpayer will continue filing jointly and when you compare the two charts side by side given income you could see that having the separate filing penalty significantly reduced the survival rate at any given time meaning that taxpayers are more likely to shift to another filing status when they have the separate filing penalty I want to emphasize that this result is a correlation not not necessarily causation it is possible that taxpayers who use this filing status uh for a very short time or temporarily um tend to be those who are more likely to have the filing uh panel separate filing penalty however it is also possible that in the absence of this penalty taxpayers may be encouraged or have the incentive to continue using it so what are the tax issues associated with this filing status as I mentioned earlier taxpayers going through a divorce or separation may have the living situation or circumstances similar to those of an a married person however they cannot file as a marriage if they don't have the um require the core actions or they do not meet the abandoned spouse rules this leads to the equity concern for several reasons low-income taxpayers may not have the resources to get the tax advice on the appropriate and most favorable filing status or the resources to obtain the required call Action to get the legal separation and for those who receive outside support it may be difficult for them to meet the household maintenance test in order to qualify for abandoned spouse rules MSS MFS filers are only limited or only eligible for the eitc and premium tax credits under very limited consider condition with rules that are not verifiable by the IRS this rules are about leaving Arrangements separation agreements or separation duration or even the condition of The spouse's relationship so this certainly would lead to difficulty for targeted Outreach education and enforcement for the IRS and another tax administrative challenge is about the compliance with filing status um individuals going through a separation or divorce are susceptible to making filing state status errors analyzing the audit data in the IRS National research program we found that about 1.7 percent of returns claim the MFS status between 2006 and 2014 each year however the audit result determined that 2.7 percent of returns should have used this filing stash status so while the only a small number of taxpayers were affected the net misreporting rate is quite High um to conclude I think the concerns raised in this paper about increasing awareness for credit eligibility and providing assistance to taxpayers in assisting a correct in determining correct filing status have broad implication for low-income population your thoughts and suggestions on the paper as well as the administrative strategies in this regard are welcome thank you [Applause] thank you you know this guys okay thanks for the invitation it's a pleasure in here I had the opportunity to present another paper with Andrea a few years ago that you know got published very well so I hope that you know this is a this is a pattern I work at the inter-american development bank so you know some some blogs from from here uh you know as a as a employee of an international organization I'm a you know I have a non-resident status so I'm one of those people filing you know MFS that Emily was uh was mentioning I have been doing that for more than 20 years so you know I'm one of those long-lasting people in the tale of the of the distribution so let me tell you a little bit about the about the paper this is a relatively simple paper where we build in in a previous paper that we published I published many years ago where we are looking at the role at the role of gender in tax compliance and let me be clear when we are talking about gender here in the in in the paper we are talking about sex a sign at Birth and I'm going to be talking more about that so so you know when we are talking about women men we are always talking about sex assignment at Birth but the idea the idea here is trying to trying to recover so we know and we're going to be talking about that women are more compliant than men you know that they tend to evade less and we're talking about about that the question is how do women react to interventions uh of the tax the tax Authority so do they also react more by paying more and they sure you know that we're trying to look at here is that we know that there are pre-tax gender differences in Pay uh you know women tend to make less than than men but also if they pay more taxes than men than they post you know they post task tax gender gap is going is going to increase okay and and we know that again we know that compliance but women is higher than compliance is is high performance and then if women react more to the actions of the tax Authority then you know the Gap gets you know larger and larger over time so that's a little bit of what we are trying to look at it here again so we know that there is that there is plenty of evidence that you know says that women are more likely to comply with their taxes than men there are plenty of papers out there showing showing that we are going to show you some data and why why is that well there are two hypotheses one is that women are more risk-averse than men and so they they fear you know they are not as willing to take risks and us as men as men are you know we know that you know men tend to do more stupid things than women and you know and full of things and you know you know get on higher altitudes and minutes so we know that men are a little more uh risk love in that than women but we also you know there is also some evidence that women have a higher tax moral okay so they believe that paying taxes you know is more of the of the right thing to do than than men tend to tend to think okay so the question is that when faced with an intervention of a tax Authority are women going to react more than men here when when we say more sensitive we are talking about more sensitive to the intervention not that you know you're more sensitive you know in general so please don't so it's more sensitive to the intervention uh but also men have more room for improvement so if if men tend to cheat more than women then you know there is higher room for for men to improve when faced with the tax intervention so it is okay what happens in practice you know what does the theory uh Theory there so so look it's an intervention that affects you know I know it all population and we're going to look at where men and women react differently uh one key of the intervention is that it's going to provide a treatment of deterrence and also three million tax morals so you know we know again that these are the two things that that may be affecting differentially the compliance and we're going to in the context of a very simple tax which is which is the property tax and and and and for this it's important for you all of you to move from the you know the world that you know it that you know of the US of the developer and move into the developing World environment okay just to give you an idea if you think about Argentina where I'm from or actually from Latin American General you pick a tax you pick a country and if you think that tax compliance is about 50 you're going to be right okay just whatever attacks whatever country more or less compliance is about is about 50 percent in the property tax compliance tends to be even lower which for you may be quite you know quite weird because you know most of you just you know pay your property taxes don't even think of the property tax because it's you know comes in the scroll with your mortgage Etc it's not even an action that you make but that's not the case in Latin America in the case of Argentina for example less than one percent of the properties are financed through a mortgage okay so basically you either inherit inherit your house or you say for 20 years you know and then you go and buy your house okay but then once you're on a property then every month or every two months you are going to receive a tax bill from the from the authority and then you are going to make the decision do I pay the tax or do or don't I pay this tax and one of the interesting things about about the property tax in this case is that then it's very easy because there is no uncertainty about the tax you have to pay okay there are no reporting requirements you are told by the government that you have to pay 100 and you decide to pay or not okay so we we you know we don't have a problem which we don't know how much people you know were supposed to to pay so so again we build in this in this other experiment that we did you know you know more than 10 years ago uh uh and basically it's you know it's a city you know it's a city of about you know 250 000 people and basically you know we get 23 000 taxpayers and we did an experiment with this uh with these people and again property tax you know this is a very simple context in which to do an experiment again the tax is built by the city okay the the city sends you a bill in this case every two months okay and basically they compute what that your tax is going to be so they they know the size of your house let's say there are the improvements they know what type of public goods they provide in most of cases in in Latin America the tax accounts for the services you provide the government employees so if they collect your trash you have you know you pay you pay some amount if if they give you you know Street lighting you pay an amount Etc and then you know the size of your house Etc is going to determine how much you have to you have to pay so again compliance is very simple you pay your tax or you don't pay your tax that's that's that's the only decision that you have so basically what we do is that we send these people uh three types you know so we had you know these 33 000 people we divide them into four groups you know one group did not receive any any type of messages and then the other three you know the other there are three groups you know each one of them receive a different message one was all deterrence you know what happened if you don't pay another was about you know reciprocity with the government so we informed people what the government does with the money they receive so we say okay look in the last six months the government collected this amount of money this is all the works the public works that the government did with that money and the last one was about what other people are doing where people are paying their tax or not so what we do is that we really signed the tax bill this is you know these are the three examples of tax bills uh but this this is what you receive on in in in the mail okay so you receive one of these every two months telling you how much you have to pay and then you go to the bank and you pay you pay what you are being uh what you are being taught so basically we know the owner so we know who receives this bill which is usually the owner or in some cases it's the tenant you know the the in the in the in the property you know in the in the leasing contract they can they can stipulate who is going to pay for a property tax if the tenant or the owner of the or the landlord of the owner of the house so we know the name and one characteristic of Argentina is that there is a list of officially approved names so you cannot name your your child like you know like Elon Musk said did you know like X Y whatever so there is a there is a list of officially approved names for you know for people who are assigned you know the gender of a woman at Birth and then at least that is separated performance so there cannot be you know there is no uh uncertainty about when somebody is named Carlos as my guess is it can only be somebody who was born as a male okay and if it's you know the name Maria can only be assigned to a person who is assigned women at Birth so there is no uncertainty about you know who is you know what was the gender of the of the person at Birth so we can so we could recover you know the the the the gender of 92 percent of the of the sample okay so everything works well in terms of intermar validity you know we have the you know it's very it's well balanced etc etc so basically again what you know what what it happens you know the government sense that the tax bill in this case you know during the during the second part of the month the month of August then you receive by the first of September people receive their will so they have you know 15 days to pay the reveal so that will be the first due date then there is another more week that they can use to pay to pay the tax so that will be the second due date uh and then you know we we look at what happened you know after um after another another two months which is when they are going to receive the event so we are going to be looking where people pay their tax by the due date but the first you rate by the second you date or after after a period of two of two months okay basically if you pay the your your your tax within that period of two months nothing happens the government you know forgives your your your your interest rate and everything okay so what happened so basically in the original in the original paper what we find is that there is an increasing compliance for the deterrence message well that is that is sizeable is like 10 percent increase effects for the other for the other two but there are you know very interesting heterogeneous effect that I will not go in you know go into that if you want go and go and read the previous paper so just to give you an idea so what what they what we did in the in the original in the original paper was that you know the message will be something like this did you know that if you don't pay the tax on time you know and you have a bit of a thousand pesos you will have to disburse additional 268 pesos in fines at the end of the year and the municipality can take take administrative and legal actions basically what happened is that a lot of people thought that there were no consequence for not paying because the cost of not pain was a monthly compound interest of two percent okay and when you tell that you know the average taxpayer that they have to pay a monthly compound interest of two percent they have no clue what you're talking about they listen they hear two percent they think that is very you know something small it doesn't that's it doesn't count okay uh so what we do is just basically try to explain that a monthly compound interest of two percent amounts to real money under the government you know can can pursue legal legal action so what is that we find so here you have women in the treatment in the treatment group you know in the in the boulder line and women in the control group so what you see is that there is you know there is an increase in the share of women that when they receive this this treatment you know increase their payment all about you know four percentage four percentage Point okay you see that you know there is also a significant difference by the second due date but then there are no differences basically at the in the end period so what is happening is basically you have stock of of women let's put it this way who are paying their tax now they receive this message from the government telling them that you know that you know that there are penalties for not paying so what they do is that they rush to make sure that they pay early okay but the overall number of women who pay their tax remains basically the same okay about you know 47 47 percent okay so let's say you know 50 of women pay their tax both in the treatment and in the control Europe but you know the women who receive the notice now they make sure that they they are you know they pay they pay on time what happened with men so as you can see first is that compliance is much lower okay it's about 10 percentage Point lower in in in in the Europe of men than women what you see is that they don't rush you know they couldn't care less about you know being being early you know either for the first and the second day but then there is an increase towards the second period so so now what is happening is is men they are keeping the same behavior but there are new men who decide now to pay okay so there is there is an increase in this extensive margin in terms of increasing the the number of men uh that that pay which is something interesting because then you see well now you see both things both groups in the in the same group so basically you see that there is higher compliance by women and now there is a high you know overall there is higher compliance by women by the first due date which basically disappears towards the end but now men are are compliment so so this is kind of an expected result so we expected that you know more women will end up also paying pay more the the behavior of the men is the same that you know you will find in most in most experiments but you know we were surprised by the fact that women the number of women that would actually paying the tax remain more or less the same so we try to explain what may be going on why is that women are not you know are we here in this way and one of the things that is interesting and probably you don't you don't see here is that while men are not reacting to the size of the attacks okay so basically all of the men you know the average the average effect is the same across across the different levels of tax women who are paying relatively low taxes react a lot so you know they increase pain a lot but the women who have to pay a very large tax are not are not paying are basically not paying that much where are reducing how much how much they pay so the first idea that we we started to look at is okay maybe a case that there is a virtual constraint okay clear these are you know household you know led by women we know that income is going to be to be to be lower there may be the case that they really want to pay more okay but they are cash constraint okay in order to pay because again one thing that is important to consider is again that the case in Argentina or in most Latin America is different than the US in the U.S the value of your property is very correlated with your income okay because basically you are paying with your mortgage the size of your mortgage depends on your income so the type of house you are going to buy is dependent on your income here in Latin America the the wealth that you have in terms of housing is not going to be correlated with your income okay because in most cases you may have inherited your house you know you maybe you you know you're a woman and you're divorce now you get the house but you know your income is lower Etc so there is a kind of a disconnect between your income and and the wealth that you have and that's why the property tax is not well correlated with your uh with your income so we have a model in which basically the traditional the traditional tax compliance model but we in introduce you know Cash Cash constraint in the model and we show that once you introduce these Bachelor constraints then what you you may have this possibility of corner Solutions in which women may be willing to comply but they cannot because because of that so we look at some complementary survey data that we that we collected and then we look at census data that seems to correlate with information with the with the hypothesis that we that we that we have so basically when you when you ask women if the government you know they tend to be more um answer that you know the government is going to is going to detectivations if they don't pay and they also think you know more than men that women are going to this there's going to be consequences if they don't comply so there is this idea like clearly you know women think that the government can catch them and the consequences are going to happen so clearly the fact that we are finding that they react to the notices is correlated with information but we also find that they are less willing to buy a higher tax and they are also more likely to answer that the tax is too high okay and then when we look at the income distribution basically we also see that you know women tend to have a much lower income than men uh in in in in monthly income you know for the household you know in per capita so so how do we interpret the results so basically what we see is that women tend to have a stronger reaction to intervention but they are battery constrained so that's why we don't see the overall the overall effect for women so why why again why does that that we that we care about so we are here in a context in which touch compliance is low as I showed you know only 50 percent only 50 percent of the people are are paying their tax however women are paying their tax you know you know 10 percentage more 10 percentage points more than more than men and we also see that there is you know that clearly you know the income distribution is a skew against against women so women are making less money but women are paid you know are more likely to pay their tax and now there may be the case that they even react more to the actions of the tax Authority so what we are basically bringing to a table in the in the uh with this paper is the idea that not only we have to think about tax policy and the effect that tax policy can have in the post tax gender equality but we also also have to think I will have the actions of the tax administration may also affect gender equality thank you [Applause] all right I'm Tyler uh this is who sells cryptocurrency hopefully the title is pretty self-explanatory uh with co-authored work with uh Jeff Hoops who is here and Jaren Wilde who is at Iowa with me not here today um I'm sure everybody is sick of these uh by this point but obviously this is IRS data I'm not speaking for the IRS and the views that I express are my own so just overview what am I going to be talking about today if you couldn't tell from the title first you know cryptocurrency who owns them what am I already looking at why should we bother with cryptocurrency right isn't it dying right now and ultimately how do we actually identify cryptocurrency owners so I'm going to start with the why this presentation and paper has been super interesting to present over the last year or so as we've gone from bull uh to Bear markets you know the framing kind of changes so you guys are getting the bear presentation today but we see market failures right and my example up here of perhaps the one fortune that was always going to come true FTX the now disgraced kind of cryptocurrency exchange that blew up last year advertised in fortune cookies for some reason and in retrospect it was inevitable right so we've seen these huge market failures and we know you know there's a lot of retail investors that are investing in cryptocurrency this is no longer just kind of a niche investment it's no longer just you know Tech Pros in Silicon Valley that are investing so it pays you know from both a tax administration but also a public policy standpoint to understand who owns cryptocurrency who might be hurt by these market failures who do we want to protect with new policy and regulations obviously cryptocurrency also offers kind of these unique enforcement it's a technology that is inherently kind of anonymous very associated with crime and illegal activity and so are these people even reporting right uh a lot of papers here today so far about enforcement about how audits affect that and this is maybe a little bit of a different area where the reactions to some of those interventions might be a little bit different so it pays to kind of think about how are these individuals and investors different from the general tax population also I think something that we've seen a lot today is you know kind of these racial inequalities and just inequality in general how tax laws and regulations affect different people cryptocurrency is kind of unique so a few articles pulled up here kind of from the last two years and the earliest articles are talking about how cryptocurrency is going to provide black investors you know a path to creating wealth and how this is going to be a progressive kind of investment and then we saw what the market kind of collapse right and ultimately and I just looked up today uh the most recent articles are still talking about how this is actually bad for inequality and kind of the racial wealth Gap because you had investors if African-Americans were more likely to invest they might be more likely to lose the money as well and so we wanna consider that and kind of provide at least some initial evidence on what these differential effects might be more than just the anecdotal articles so how do we actually identify cryptocurrency transactions right as I mentioned before you know the reporting and the cryptocurrency kind of environment is pretty unique um prior to 2017 um basically there was no third party reporting even from a lot of the major exchanges the IRS eventually did the John Doe summons some exchanges like coinbase and over time more of them have decided to voluntarily report cryptocurrency sales of their users and so we kind of take a two-sided approach we look at third party third party reporting for 1099bs and we also go to schedule D where taxpayers self-report because there's going to be a lot of cryptocurrency exchanges a lot of cryptocurrency owners who aren't getting third party reported so we want to capture them as well we ultimately use textual analysis of descriptions so for the 1099bs it's a little bit easier unsurprisingly large companies are a lot more standardized and have a lot less spelling mistakes in their descriptions it's a little bit more noisy when we go to schedule D obviously we have people who are saying you know see attached or just cryptocurrency sales in general we try to pick up a lot of that but certainly there's noise in our measure we're going to be missing some things we focus primarily on bitcoin and ethereum and this was done one because the size of those two cryptocurrencies encompasses a very large portion of the market and two because we're using kind of textual analysis it's hard to separate certain types of cryptocurrency from just regular Investments and so we want to keep our kind of false positive rate as low as we can one example of that is there's a cryptocurrency called polka dot the ticker for that is Dot and as you can imagine when we run a search right for DOT there's a lot of Department of Transportation bonds that show up and so taking those out from the sample is a lot of manual work and so we stick to bitcoin ethereum which are both large and a little bit easier to kind of separate from your traditional stocks and bonds that we don't want to look at so obviously as I mentioned before we have concerns with tax avoiders we've talked about non-reporters in some of the presentations today that is definitely an issue um we hope that you know the sample we have is fairly representative of kind of the universe of at least cryptocurrency investors who are not doing illegal things I firmly believe that if you're buying or selling drugs with cryptocurrency you're probably not reporting it on your taxes but that should be a small percentage especially in the more recent sample period and also we have the issue that identification is contingent on you actually selling luckily we find in some of our supplemental tests that the holding period that we do observe is pretty short but if you bought in 2013 and you have never sold your cryptocurrency first why haven't you retired yet and second you know we will not see you in our data so who are the people who sell cryptocurrency our results so first a little bit of background after I tease you there um our paper 2013 to 2020 you can see that we get over 2 million cryptocurrency users so taxpayers who report cryptocurrency and our total sample is about 200 million and one of the benefits that really IRS data has provided us is compared to Prior studies that have been looking at cryptocurrency owners we have a huge sample we have a ton of cryptocurrency owners we're a little bit less Limited in kind of some of these self-selection concerns so again hoping that our results are a little bit more representative of the broader population so when we look at cryptocurrency Sellers this graph is showing two things the yellow line is just the number of cryptocurrency taxpayers over the course of our sample period as you can see 2016 to 2017 is kind of a turning point prior to 2016 almost nobody reported after that the growth has been exponential one of the reasons for that is again IRS enforcement actions the IRS you know really came out 2017 was when the coinbase John Doe summons kind of got litigated through Court you can see the black bars are where we have the number of taxpayers that are receiving a 1099-b so again increasing you know the importance of third-party reporting in individuals complying even though these 1099 B's are voluntary by the exchanges themselves and since then it has again just kind of blown up exponentially to uh I believe almost 1.6 million taxpayers in 2020. and how has that affected who is actually selling so William made a good point in his discussion that maybe we want to do a whole group sample because the year by year comparison is not great I'm going to do a year for you comparison because I think there's a lot of change that's happening again that 2016 to 2017 seems to kind of be an inflection point here we're comparing cryptocurrency sellers so people we identify as cryptocurrency to non-crypto investors so look there's obviously a difference if you've never invested in anything so we want to compare people who invest in crypto versus people who have investable assets maybe they have stocks or bonds or interest to make the groups a little bit more comparable as the starting point and then look at how that changes over time so we can see you know investors who don't invest in crypto are a lot older and crypto investors were younger than other investors but since 2016 they've gotten quite a bit younger and then an income right the non-crypto investors income has been slowly kind of increasing over time that's the gray line at the bottom and we can see the cryptocurrency investor again with this just explosion of more people coming into the space more third-party reporting potentially more tax reporting right we've just seen the average income of cryptocurrency investors plummet since 2016. and giving credence to kind of our hypothesis that you know the enforcement of cryptocurrency needs to fit with the individuals who are actually selling and these characteristics have been changing pretty significantly over time so next uh I'm are going to go a little bit into are there racial disparities right we saw these news articles about how you know black investors may be doing better maybe they're being doing worse the problem as other presenters have been pointed out the IRS does not have racial data we are hoping to use your measure in the future we've taken a different approach so far so interested to hear thoughts on that um we start just kind of with a little bit more of a course measure and we use census level zip code data to look at the racial characteristics of each census code and then we aggregate our reporting statistics by those zip codes to see all right if in a particular zip code there are a lot more African Americans what does that do to you know the percentage of those people that are reporting cryptocurrency and their gains an additional note for the next graphs I do standardize all of these percentages to mean zero standard deviation of one so the coefficients should be comparable between groups even if you have a small percentage from African-Americans and a larger percentage of Asians they should be comparable so first question does racial characteristics matter for the number the proportion of individuals selling and we see that it does and there are again changes over time uh certain groups so Asians much more likely to be a cryptocurrency seller throughout time still increasing African-Americans kind of in line with the news articles originally were actually less likely in 2017 and over time they've increased to be more likely than our Baseline to sell cryptocurrency no but that's just one part right we want to know not only whether they're more likely to own it but what are the outcomes of that so we also look at cryptocurrency gains over time and here I do on the bottom uh have the yearly Buy and Hold Bitcoin return as a baseline for how much did the market go up or down and we can see you know racial minorities they were more likely to hold and they seem to perform better when the market is up they perform worse when the market is down and so given they're more likely that maybe is a little bit obvious but we think it's interesting and it kind of provides evidence that from a policy standpoint these groups are a little bit different in their investing characteristics and habits so finally I'm not going to go into all of our other results uh just due to time we have a lot of them so I encourage you to look at the the paper When It ultimately comes out for the conference but we find Geographic dispersion has also increased over time back in 2013 as you might expect it was East Coast West Coast you know the Silicon valleys is where the cryptocurrency sellers were and over time it spread to be almost in every county in the United States we look at the profession of individuals who are reporting cryptocurrency again in our early part of our sample period we find it's Tech professionals Finance professionals that are really reporting whereas by the end you know there's a lot of restaurant workers or Hospitality workers who are reporting again giving this broader adoption of cryptocurrency potentially bringing in people who might not have otherwise invested in other assets and I think one of our interesting tests we also look at you know people who had huge gains so some other news stories that we've kind of seen is you know that oh so and so bought cryptocurrency or mine cryptocurrency in 2012 and now they're have a billion dollars right because they just held it so we look at individuals who in a single year had over a million dollars of cryptocurrency gain and we find that for the top income percentiles that yeah we see the big spike they have a big gain and then their income kind of goes back down to the Baseline within two years but for lower income individuals who have these a really life-changing amounts when your average income might only be fifty thousand dollars and you have a million dollar gain at least two years out their average incomes are still significantly higher than they were in the pre-period so again giving this you know maybe there's some silver lining uh to cryptocurrency for at least a small number of people so ultimately takeaways uh the cryptocurrency user base is not stable over time I think that presents interesting opportunities for research when we're thinking about compliance and maybe even on other non-cryptocurrency topics how to use this as a setting but also for other cryptocurrency research demographics continue to rapidly change and even things like geography profession and racial composition also continue to change and in case you were wondering that is hercu the hawk the Iowa mascot so we have to have him on the slide for us as well and that is all I have so thank you very much I'm looking forward to the discussion thank you hi um my name is Yen I'm Economist um we service of IRS I'm honored to have opportunity to contribute to the conference as a discussion so today I will review three papers I think these three papers have brought us the research expertise from the treasury from the academics and from International communities um the research in this session highlights the importance of understanding our taxpayers and they are empirical papers and even um they they focus on tax compliance but they from the different angles from the um fighting status and the intro household resource allocation from the racial disparity and the in response to the intervention and from the cryptocurrency um so the first paper is uh married filing separately so this is the interesting paper examine the MFS fighting status it used the individual return for nine years and the key finding is most MFS filer face a separate filing penalty and so um I like this paper a lot because I think this it quantifies the penalty and the bonus um the paper also I think did a good job brings that inequality the angle into the big picture so because this MFS filer faced an unfavorable favorable on treatment of tax um with with respect to tax credits and for the low-income household they sometimes they cannot file and marry the fighting jointly um so uh yeah so they kind of like uh have the penalty also I think this paper make me think um how could IRS help taxpayers to understand the consequence of different filing status what's the advantage what's the disadvantage brought by the tax as well as the tax um credit because MF as a filer they have limited eligibility for their tax credits they are they are not able to um claims that eitc until 2021 and they cannot claim education credit so I think the paper would be more interesting if it discussed the scenario under which MFS are better off or are worse off so author does exp does explain why using MFS but in in the sample like a is a rational household or why I choose this elect to choose these fighting status um by maximum maximizing my expected utility so I think from the data if you can tell us what information that would be um great like is it because of the student loan payment or is it because of the medical expense deduction or because of I have big disparity in the couple's income respectively also because of like uh something I need to file because I'm in um pending The Voice or separation or I just have lack of information the reason I'm saying that because different scenario for IRS we have different um intervention policy so yeah also I think it would be great if you exam the heterogeneity in the MFS sample also look at the heterogeneity over time because you do have the data it's nine years data and um uh is a minor question what's the consequence or impact of msfs these fighting status on the um on the household structure family structure for example the this incentive for the marriage does it have any consequence I just come across this paper and from tax policy Center hopefully it would be um uh useful so the second paper is by Carlos so this is um uh experimental method to inform the policy it is in municipality in Argentina exam how taxpayer responded to the uh intervention message from the city government on the property tax payment it has three messages three types of message so the main finding is women would be more motivated to pay but they face significant household Financial constraint so I like the paper because um along with the empirical model it also developed a very nice analytical framework like you pay now or I like don't pay but I face probability of being penalized in later and um I also like it it brings a lot of domains and approaches from development economics for example beyond the risk of version and tax moral we see in the tax compliance literature it also talked about household budget constraint and income inequality of and poverty because the female household had it that household have lower income also the disparity in the labor Market outcome so we all know tax compliance is very complex complexible because it it cannot like use just a one one approach so I think that all these approach make that paper very rich of the context a very interesting to read um also I think it's an empirical analysis is conducted very um carefully for example for the uh for the regression um it it analyzes the text compliance along the gender line and across the gender line so along the gender line is a regression for men and women respectively and then combine them together and introduce the interaction of the gender variables so I think it is um very nicely done um so uh the main challenge I think for the experiment is scale up so I think that finding is very um interesting but how can we learn from that experiment so this is the city in Latin America as I also already said the tax structure in that country is very different from from the other side of the country from the developed country or from oecd country and it is a property tax and versus the income tax so does it matter does the finding would differ because of that and also the local government versus central government government if I want to learn from this experiment like what should I do like a can I take all of them or like a it's different um inside um also I think like a full experiment if there are others seeing affect your findings like for example I have the property joint owner how do you define the uh gender for the joint ownership and um uh for this experiment uh it is in 2011. uh do you have another round of most recent data can give us more updated um information um yeah so the next paper is very exciting the cryptocurrency um so it exams the cryptocurrency sellers who report cryptocurrency sales to um IRS so it'll use 80-year data 2013 to 2020 and it gives a general information on the characteristic of cryptocurrency users um so I think I think also did a really smart way to utilize the administrative data to understand the relatively new Financial product when the survey data is not available and it offers the first broad um descriptive statistics uh on this U.S taxpayers who sell cryptocurrency and it also provides a empirical analysis uh so the main challenge is the Under reporting because it's a subset of the of the population so I think IRS we have uh strengthened the digital assets reporting I think um and also I think the fight is very interesting because you have uh the cryptocurrency user has increased and income has declined um but like uh but is this like a because of Under reporting or like is there any like other explanation so I think it seems uh tax year 2020 IRS has required all the taxpayers to answer the digital asset question in form 1040. so do you receive sell send exchange any virtual currency so for all taxpayers even if you do not have so you have to answer no so I'm thinking uh when you face this Under reporting challenge do you think this um this update off from iisesi would uh uh address this Under reporting issue so are you considering to use like a another rounded data like a tax year 2021 um also that I think investing behavior is very complex like um but our tax data since it's tax purpose we don't have a lot of variables there like for example education we don't have that variable so are you considering to to use additional like a survey data also I think we don't have the the gender as a race and like the arrest co-residency for example I think all these variables are they are very relevant for the investing behavior and um another factor is Network externality and complementarity because as you said now more and more people use cryptocurrency and um I think now we have more and more financial exchange firms and the payment um payment uh payment firms all these players they create as a positive externality and I also have minor data issues and um for example for your sample selection you pick up the first observation but do you think you think alternative way like using the TC code to pick up the um amended a return also you define your student as if they have 10 98 T form but I think for some student populations they are students but they don't receive that time 98t form if they have the more scholarship than their tuition also the income I think probably you have to clarify like I think in the paper you uh we have tax return I think you put very clear why you pick up the taxable income not as an AGI but but then here you use per capita income so in your paper you have to be very consistent and also the wage income they are usually uh is endogenous variable how do you um like a say something address this and that's I think probably one factor your explain uh the R square is very low um yeah so I think this is a wonderful research they just make me distinct um you know uh how can we make our tax intervention more effective why are you considering the disparity in gender household structure and labor market outcome how can we make our administrative Authority a more trusted institution provide more better guidance for taxpayers and especially the disadvantaged people like low-income people female headed households how can we update um address The Challenge from the um from the fast growing like technology Advance all of this we don't have the definite answer for this but now IRS work with the treasury with the support from the academic Think Tank and international communities so yeah so um I would like to thank authors for sharing these great papers in this conference with us thank you [Applause] thank you please [Music] but it was phenomenal job they did a really good job and we even had a college mascot to join the group today so let's open it up for questions and answers anyone have any questions yes sir I had a question for Emily did you see any evidence in your data when you're analyzing things that married filing separate really serves kind of as a transitionary filing status where you have it really is kind of going from it's like pre-divorce it's like you're you're separated and you're not going to get you're but you're on your way to divorce so like how many of those people didn't just end up in a divorce state or filing single I guess would be that next year um yes that's kind of our next step we haven't linked to any um the multi-year that we we have the data only on the MS returns and we should have like no Inc uh link the data into the before and after to see the transition but as one of uh the reason we think people file separately is in the transition from Marriage to uh divorce so we and many of them use only one year so we suspect that that is what's happening yes sir yeah uh hi it's a question for uh the cryptocurrency paper says for Tyler um there's Dynamic patterns are really interesting and the types of people that you're seeing reported sales of cryptocurrency I maybe I don't understand the reporting environment very well I understand you showed us there are these very large changes in the reporting that's happening over the same set of years so how much of the dynamic the demographic pattern is just a change in who gets reported on and how much of it is a change in who's dealing in cryptocurrency in real terms yeah so uh I wish that I could provide you with a concrete answer um it's definitely something that we've been thinking about and I think one of the parts of our research is just kind of showing you know a lot of papers will use kind of 2017 as the shock to IRS enforcement of cryptocurrency and to be honest at this stage we're kind of doing some more tests but I think it's both I think it's there's a lot of new people coming in that's changing the income down there's people who had been using it before and are now newly reporting as well and yeah teasing that out is hard given we only see one half we don't see the non-reporters but it's definitely something that we're thinking about so if you have any suggestions yes ma'am she does also for Tyler you notice that big spike in the number of reported sales as the volatility grew and more people came into the market could that possibly be because there's no wash rule you know you can't sell your your stock and buy it back the next day and lock in that loss but you can sell your Bitcoin and buy it back the next day uh yeah definitely so I think one of the interesting things that gives us a little bit more uh confidence for like eliminating the Buy and Hold people who you might not see in traditional Securities is that exact fact that you know there's no penalty to selling and then kind of buying back right away because of the the lack of the wash sale rule I know some of the additional tests that we're running is kind of looking at both the volatility right around 2017 and then also we're going to expand the sample out to 2021 to look at what's called the crypto bowl and the ads are on the Super Bowl and see how trading kind of changes around there but I suspect that there is a lot of short-term Trading yes sir so two questions the first one for Tyler um you mentioned that a lot of the 2017 increases particular cryptocurrency exchanges and actually it would be really cool even if you just give them each a color and like you take your black bar over time and you split it out by you know this is color green this is color yellow this is color blue whatever exchange and just see oh how many of the increase in our particular exchanges and so you might say oh you know here are the people who are reporting who are also reported on a 1099-b from one of the exchanges that was summonsed and here are the people who actually increased the reporting even though their exchange wasn't summoned or they weren't on one of those and that would give you some leverage on the what was driving that 2017 increase in over time and also like is there a spillover of that enforcement yeah we definitely do that is a great idea um to split it out by exchange we do see both increases in the 1099 B reporting but also individuals who don't have 1099bs who are reporting so that does I think slightly answer that that it's increasing in both groups it's not entirely driven by just the 1099bs and my second question is for Carlos so on the um do you have to what extent do you have demographics or even like you know you have something about the value of the property and it would be interesting to know how much that varies with gender and just like to sort of link that in even if you can't see income you could at least see well the women have properties that are on average this percentage as large as the men's or vice versa yeah no so we don't have detailed demographics just because you know we are we're using the the tax data but we can see the properties and and women tend to have smaller properties and they are more centrally located in the city so they tend to live you know more in the center of the city rather than you know in in the outskirts so then the then the you know lots are going to be smaller and so the property tax is going to be to be a relatively smaller than the than the men's because the property tax is not really there is no there is no it's not like in the U.S that you have the value of the house and then you have basically it's just a combination of the size of the Law plus the services you receive et cetera to the services they received very differently for women and for men because they're more centrally located women tend to receive a relatively more services than dunman yeah it's interesting yeah anyone else I have a question for Tyler yes I'm curious if there's any research done on um the type of person who tends to own cryptocurrency and their overall tax compliance um so there's definitely a lot of kind of surveys and stuff at least on just the general characteristics [Music] um I cannot think offhand of the general tax compliance on like non-cryptocurrency uh aspects I think is what you're getting at maybe I know compliance and kind of the enforcement um Jan mentioned the cryptocurrency check box is one thing that we do also have on our list to kind of continue on with the research and look at further yes ma'am following up on that I've done a little bit of research into the crypto space and um the last data I looked at from the IRS I think it was around three percent of filers who report cryptocurrency um and if you look at the Pew Research Center survey that came out recently I think around 16 of Americans say they've invested so I just wonder if you've looked into that Gap um and also a second part of that is do you anticipate that it will become easier or harder to identify those filers over time uh so we have looked at kind of like the survey evidence um I'm always a little bit skeptical of some of the survey since Pew does pretty good stuff I know a lot of the cryptocurrency firms also kind of put out their own surveys that you're usually pretty high because they're all online and they're for their customers so um I don't quite know you know how we might get tie that kind of one to one to you know zero in on what that Gap is um and as far as will it get easier harder I honestly I don't know I mean some of the IRS folk here probably have a better idea of what they're thinking I think one of the big challenges right now is like the SEC enforcement actions if they shut down the big cryptocurrency exchanges that are already deciding to voluntarily report 1099-bs and we move to all decentralized exchanges right it probably gets harder uh there's probably gonna be less people reporting if those core cases go differently and the IRS requires a broader 1099-b requirement you know then it probably more cryptocurrency people will report because they'll get more third-party reporting but I don't know I don't have the fortune cookie to tell me which way that'll go yes sir that was kind of my question um in regards to efforts and uh regulatory enforcement by other agencies such as the SEC cotcrinson uh did your study take into account any of those efforts um so in our sample period we're missing a lot of that because a lot of those actions have kind of been more recent so we kind of look around that 2017 period but again it's so correlated with the huge increase in price that happens in 2017 the huge increase in hype that it is really hard to just kind of disentangle what's enforcement what's General economics what's kind of like the advertising effect um in this really rapidly shifting asset yes ma'am in the back so your results imply that here in the Echo here so the results imply that women are much more responsive to treatment but is that the response you wanted because you wanted the people to pay more taxes not to pay them quicker yeah no so we were kind of surprised you know so we if we assume that women are more likely to pay than men again because of risk aversion and tax moral Etc but we still have you know half of the women not paying we assume that with the intervention what we will see was a you know higher impact also on on women and we didn't find that we found this weird kind of weird result in terms of you know moving women that were paying but pain probably late just to pain pain on time and the traditional you know result for men that you know they they you know after after an intervention they pay they pay more so yeah we're not expecting then then you know we started to do this like forensic analysis why is that you know it may be the case that we are not bringing more women more women in and we find this plausible result which is this issue of of you know cash constrain or you know budget constraint but yeah we want to inspect it just let me be clear when this intervention took place we hadn't considered this you know these gender differences you know we were thinking about just taxpayers and then you know this is something that we you know we did 10 years after after we ran the experiment we say okay we can recover the gender so given that we can recover the gender we can go back and answer a question that basically nobody has looked at is whether you know women and men react differently to the actions of the tax Authority yes sir question for Emily following up on the earlier question though the transitory kind of use case versus some of the other uh you were discussing around the kind of more permanent uh folks who are long time married filing separate and it reflects maybe different needs for how they organize their lives were you looking to take this I mean it seems like that's going in the direction of some kind of segmentation of understanding different uses of the policy what are where are you looking to take this in terms of uh trying to take this apart a little more yeah that's a good question at first that we started this uh just because we feel like we didn't know much about this and but we heard about filing status issues and then we look at an RP data to know that you know a lot of like head of household um memory filing separately and singles those type of issues so that's where we began but also when we started Gathering all these reasons to file separately and we think that by looking at data we could link to the reasons I think that's also related to the discussions comment to link the the length of using it to the reason using it so you know we don't have a specific direction as to where we are going to take this project to and it's so right now we just kind of our beginning of our work and we want to see like you know which area will require additional work and we would just go from there I'd like to thank you all for all your hard work regarding this important Endeavor we really do appreciate it and we have time thank you guys so much [Applause] thank you but he could take a seat we can start the next session for our final session today we have three our final session today hidden assets and hidden networks we have three papers on some of the less visible parts of the tax system okay our first paper in this session following k1's considering foreign accounts in context is being presented by Tomas wind a data scientist at the irs's research and applied analytics and statistics division nice to talk to you our second paper is application of network analysis to identify likely ghost repairer networks it's being presented by Joshua King who works in the emerging risks lab also at the irs's research and applied analytics and statistics Division and our final paper of the session and the conference is the offshore world according to fatca new evidence on the foreign wealth of U.S households it's being presented by Daniel wreck who is an assistant professor of Economics at the University of Maryland and a faculty research fellow of the national blear of economic research discussing for today's session is Paul organ who's a finance Economist at the office of tax analysis in the treasury Department hello my name is Tomas wind and I am at Raz in the IRS and I'll bring some results from my work here following k1s considering foreign accounts and contacts before completed with my colleagues David Bratt Alyssa Graf and Ann hurlash and as always the usual disclosure applies this work does not reflect the views of the Department's treasury or the IRS okay so some of the motivation uh for this work so it grew out of a couple different work streams um first digging into what we know about taxpayers with overseas accounts so as you'll hear about a little bit in much more detail a little bit later in this session recent estimates have uh have shown that in 2018 for example about four trillion dollars in assets from U.S taxpayers are overseas and there's also earlier work that goes into great detail tracking has so much partnership in Clinton flows to owners and tax Havens and then the lack of transparency in following these funds so I think we're we're beginning to build a better understanding of how much money has been tracked for U.S taxpayers overseas and then also some of the steps taken by some taxpayers to make that money difficult to track then we also are buildings off of previous work looking at using network analytics in Tax Administration so for example the agaw wallpaper from a couple years ago that is that ties the presence of pass-through entities and corporate structures to tax avoidance and uncertainty so here we're looking at in the individual context but in the also K1 networks so we're really trying to understand how using K1 Networks rather than just individual returns can help us understand whether non-compliance or in this case looking at the uh the population of taxpayers with offshore accounts just bring us to our research question which is to what extent are taxpayers K1 Network characteristics predictive of their disclosing a foreign account so I'll just walk through a quick overview of our methodology of how we went around doing this so first we identified all individual taxpayers that did a report holding a foreign account so I will note here that we are only looking at individual taxpayers we are not looking at Partnerships or other business entities then we then filtered down that population to individuals that also received a K1 that is because we are interested in looking at K1 Network so they would have to overseas K1 and we then took a sample of that population can you turn on your microphone oh okay better all right um okay uh right so if we took a sample so this population uh referred to this as the RFA taxpayers or reported for an account taxpayers uh then we also took a sample of individuals that also received k-1s but never reported foreign accounts uh so to act as a comparison group so I'll refer to these as nrfa taxpayers or non-reported foreign account taxpayers then we created a graph database just depicting the K1 networks and also the spousal relationships among taxpayers in these groups and then finally specified a series of models on whether a taxpayer reported account uh reported for an account in a specific year so very briefly just here are the sources that we used to identify taxpayers with foreign accounts so we did look at tax rate for fbar we also looked at form 8938 as acquired by fatca just another note here about form 8966 which is third party reporting by Foreign financial institutions so we did not use that to identify taxpayers because we were specifically interested in reporting Behavior so we were interested in seeing whether or not a taxpayer with a foreign account self-reported we did use the 8966 for one purpose and that was just when we looked at our non-foreign account population we want to make sure they were not mentioned in an 8966 filene just so we could be sure to the best of our ability that they did not have a foreign account they should not have been reported or reporting having a foreign account either um and then see the last two we also included taxpayers that participated in various IRS voluntary disclosure initiatives like the offshore volunteers closure programs and then the streamlined filing compliance procedures so you're just uh clarifying A little bit to look at the the two different groups uh so the rfas with k1s and the nrfas K1 so just again the the main difference here is that one reported it for an account and the other one did not report a foreign account um however both of them are comprised of only individuals that receives at least K 1 K1 between 2006 and 2017. um and we did limit this to make sure that they received a significant share from K1 income so we defined a significant share here rather simply as just uh that the individuals have directly received at least a 30 percent of the income that was reported out by the K1 payer so uh construction of the graph database I won't go into too much detail here but we essentially just got the K1 Network for each of the taxpayer um got the spouses and then added additional data 1040 data and some other taxpayer characteristics so uh this is a graphical depiction of a sample network of a taxpayer so see we would start with our that blue rectangle they are starting node so this is an RFA taxpayer Network it is an ovdp participant so Ultra voluntary disclosure so you can see this taxpayer would have received two k ones one from entity B one from entity l in our Network just uh just to keep things manageable in terms of sizes and networks we would only keep entity B because we could see entity L he just received a one percent stake so for this point we would not include that in our Network and then going up from NP we would just follow the K ones through looking at different payers and and pay leaves again always make sure that they are above that threshold sure this it was uh a one percent threshold after that initial identification stage and then you can see on the bottom there's just a spousal relationship here as well so we're just noting that and then also taking a note that the spouse did not have a reported foreign account okay so now some uh insights are from the graph so a few of the RFA taxpayers did report a foreign account in all of the 12 years so what we did notice though is that tax taxpayers that did report a phone account in multiple years uh were more likely to report a foreign account in the Years directly following the first report so what this shows here is these are taxpayers who first receipt who uh reported a foreign account for the first time between 2006 and 2012 and then it just tracks uh how many of them continued reporting a foreign account to five years later so you will notice that the y-axis begins at 50 there just go to that so about 65 percent of taxpayers reported a for an account in the year directly following the first disclosure and about half of taxpayers uh continued to report a cow uh five years later so now turning to looking a little bit at the differences in their networks and network characteristics so on this left-hand side here so this is just the number of taxpayers in a in the network in the K1 Network so the uh the lighter blue are the nrfas and NF RFA networks and darker blue are the RFA networks so kind of right off the bat you can see that the uh never RFA networks tend to be smaller in size so the normal number of taxpayers they're also smaller size than just the dollars that are in the network it's not shown here um so in fact 55 of nrfa taxpayer contain less than four taxpayers and that's compared to about 45 for RFA taxpayers so we're going to really start seeing some difference in the kind of in the and the types of net worth that they have and then the right hand plot uh so just show some different some different metrics and again the percentage of uh RFA or nrfa taxpayers networks that meet that so the first one is uh received K1 from a multi-tiered pass-through entity so here multi-tiered password entity defining as a password entity who itself received um a K1 from another path for entity friendly uh so this is just kind of could be one uh proxy for Network complexity and you can see here a much higher percentage of RFA taxpayers did receive a a K1 from multi-tiered password entity and then the other two just show other RFA taxpayers in their Network so that first one RFA owners would be um mostly okay individuals so um other RFA K1 recipients who are also reported for an account and then uh they own the RFA payer so K1 issuers who are in the network that reported a foreign account and again both of those were much more likely to happen in RFA Networks and then one last slide show and the difference between two so here we're looking at different 1040 measurements so different measure of income and tax so right off the bat you can see that the the dark blue line is once again RFA taxpayers so the median RFA income so AGI for example is considerably higher than the uh than the non-rfa the median for non-rfa taxpayers um that kind of holds across the board and now looking along the x-axis so for the RFA taxpayers uh this is so these are RFA taxpayers who had their first reported foreign account between 2009 and 2014 and then it shows a little bit how their reported income or whatever 1046 we're looking at changes in the Years prior to first report and the years after um so one interesting note here you can see that the capital income uh actually increases slightly and then while wage decreases and then for the non-rfa taxpayers of course they don't have a foreign account so uh they don't have a first year of foreign account so we did we just we randomly assigned a a year zero for them between 2009 2014 so just to get a comparison group uh to see um just make sure there's no kind of larger economic trends that are going on here um we do have plan to kind of take a deeper look into into this it's really seeing the differences in behavior and seeing kind of maybe what might be going on that capital income and wage um so we're still a little early on especially on this stage of the analysis okay so now digging into some of the results so we ran a series of logit models with year and individual fixed effects in each of the models we had the same outcome the outcome was whether the taxpayer reported a foreign account in a given year so we run four separate models of doing this again same outcome but we just changed the grouping of covariates to describe a different element of the taxpayer so the four elements that we run we're using network variables using K1 taxpayer variables using 1040 variables and then a combined model using all of them so on on this slide you can see the results for the first three set of models we're using just the subset of the variables so just a little bit how you read this so these are odds ratio for each of the covariates so exponentiated coefficients so anything greater than one would be a positive relationship um below one would be a negative relationship so you you so starting with the network model so this is intention just to give us insights on the ability of network characteristics to inform us of the likelihood of reporting a foreign account and then also which network measures were associated reporter for an account so you can see in our top uh kind of that top one right there so that is um whether the taxpayer Network had another had a K1 payer who reported a foreign account uh so let's say you could see that with all the variables content the odds of a taxpayer whose network contains a K1 issuer in the same year uh is about 1.1 Times Higher or 10 more likely to report a foreign account so it's intuitive and then it's also the same kind of see similar results for also if there's an another RFA owner in the network uh and then moved on to that second model the taxpayer K1 model so these are variables that are directly associated with the k1s that received by the taxpayers instead of looking at the network either directly the k1s so this actually tells a similar story to uh the network variable where here we see that top variable there is looking at the odds of the taxpayer reporting a foreign account increases by about 25 percent when that taxpayer receives a K1 from an entity that also reported a foreign account so again totally intuitive but it's good to check and then the last Model is looking at 1040 variables um so just to to note here that these all the different incomes those dividend Insurance AGI et cetera has been broken down to deciles and it's decile from our our population um and yeah so future work we might also uh break it down a little bit finer than Dev file so seeing some of the other previous work that's come out that looking at there's there's really a difference between the five percent the top one percent top point one percent um so we might definitely want to look at that in some some finer granularity uh now moving on so this is our combined model where we of course combine all the previous models uh so unsurprisingly we see the top variables from the previous slide are showing up once again um so you know this this is showing us that there definitely is something there just by instead of just looking at for example 1040 uh variables there is something that seemed proximity to other taxpayers for an account does seem to have at least some relationship with also report account the the right plot the right hand plot here so we then uh compared all of our different models um using the akaike information Criterion or AIC so the AIC takes into account the trade-off between model accuracy and simplicity so a lower AIC indicates a better balance between model fit and complexity so it's showing here is the difference uh the percentage difference in AIC from our worst performing model which was the taxpayer K1 model to the best model which is uh the K the combined model so it is interesting note though that even though you know we couldn't it might be intuitive think that the combined model would form the best but also this is even one to account that the model has been penalized for the additional complexity and the additional parameters so we do you believe that our results uh demonstrate the promise and considering the taxpayers K1 Network when assessing their likelihood to report a foreign account but this is also I think in combination the previous slide shows that with additional taxpayer uh characteristics um such as you know 1040. so moving moving forward we do plan on the couple steps we do plan on taking such as updating data right now for different data issues um we were only able to select taxpayers through 2017 so we do hope to get that resolved soon adding additional variables so exploring other types of taxpayer characteristics such as you know taxpayer had a foreign a foreign address whether it had an amended return we do hope to potentially incorporate the form 8966 so we can incorporate some third-party reporting somehow into our results uh we also again talked a little bit about how we're only using individual data but you know we kind of from previous research we also know that a lot of the um the high dollar accounts are in the business and so trying to see if we can you know include some partnership data and then some of the previous work linking partnership data with individuals and see how we can make use of of some of those insights and then again try different models especially we are plan on moving into a different um potentially like a predictive modeling space so different random forest model even a graph neural network so definitely exploring what that could take us thank you [Applause] all right um hi my name is Joshua King I work I'll be presenting on uh the application of network and analytics to identify like the ghost prepares I work not sure my notes I'll move closer to it um so I work in IRS research and applied analytics and statistics while I'm presenting this we do have a much larger team that worked on this analysis and uh it's good to hear you okay um I'll work on speaking up um we did have a much larger group that contributed to this analysis and uh and writing this paper so I I wanted to acknowledge that at the outset all right so um paper pairs are an important IRS partner the service depends on them to help taxpayers comply with uh with tax law um ghost prepare is a compensated tax return repair who does not provide a a prepared tax identification number or a P10 on the returns they prepare um as required by IRS rules and regulations ghost preparers May intentionally or unintentionally hide their identity for parents who are unaware of the rules and requirements around P10s there's concern that they may not be qualified to provide assistance and advice to the clients to their clients and may unknowingly put taxpayers at risk of not meeting their tax obligations or may put them at risk of IRS audit hmm for preparers who knowingly hide their identity there's a range of reasons to do so you know because preparers May prey on taxpayers stealing refunds or engaging in potentially illegal or fraudulent preparation strategies and uh even where a taxpayer is collaborating with it goes with with it goes to falsify returns or Mac to maximize maximize refunds or claim unearned credits um the ghost repair enables this illegal Behavior Additionally you know we've we've heard of cases where ghost repairs May lie about their activities to avoid their own tax liabilities or because they've already been identified as a problematic repairer and then lastly we recognize that ghost pairs are very likely engaged in schemes that we've not yet identified compounding these risks an individual goes preparer may be responsible for tens if not hundreds of um of or thousands in some cases of returns so they very much have an outsize impact I like ghost repairers disrupt and stabilize and established IRS preparer and taxpayer ecosystem um The Innovation lab is an initiative at the IRS to encourage collaboration across the service on a specific administrative or compliance challenge in fiscal year 2021 The Innovation lab sponsored research to apply network analysis to detect and identify ghost preparers the objectives of that Innovation lab were to explore multiple approaches to identify ghost prepared returns and to develop a tool for compliance staff to access and investigate these networks for treatment over the course of the Innovation lab analysts delivered two clustering approaches a data set of 1040 returns networked across a range of filing characteristics for three filing years um and then going forward those results should be updated with more recent data and additional data sources to build build out context of the cluster networks and additional inputs to our clustering approaches and the lab delivered a ghost preparer specific tool designed to deliver results to users and facilitate investigation and Lead development all right so the value of network analysis to identify ghosts a ghost repair is inherently difficult to identify because they don't identify themselves in the returns they repair so from the IRS perspective you know 100 returns completed by a ghost preparer might look like 100 individually filed self-prepared returns um behind a ghost repair however is a web of relationships between all and all individuals involved and in theory a network model can help capture some of these relationships and by analyzing the structure of the network we can identify patterns across returns which suggests ghost prepare involvement in addition to providing groups or what we call clusters in this analysis of interconnected self-prepared returns having the data in a network format lends itself to seeing second and third order connections between the returns which facilitates connecting these clusters that we've of returns we've identified to a potential lead or ghost preparer and practice how this process works is we take available 1040 data and we use it to build large networks of self-prepared returns and then we search in those networks for sub networks or clusters we have two general approaches for doing this one's risk-based and then the risk-based approach we first limit the returns that are include that are that are um we prioritize uh returns which have suspicious filing characteristics before building a network and searching for sub networks and then the second approach um sort of which we've also termed sort of a top-down approach is we first build a network and then work to limit um points of connection based on degree limits and um and then where we still have uh large sub networks we apply a label provocation algorithm to break them apart um uh and then um just generally we feel that having these uh these data sets stored in a network format provides insights that might otherwise go unobserved if we were working with tabular data and you know while this is one application of network analytics there are there's potential for other use cases and applications all right so on to the analysis um you know while we we did connect we did conduct analysis in some of the Clusters we identified our main research aims at this point are to understand how suspected ghost repaired clusters appear early during the filing season and how they how they evolve through the filing deadline and then also to understand the effect that ghost repairers have on the returns they prepare and the um and eventually we'd like to know the larger impact they have on Tax Administration before we go into those that that analysis a few pieces of context um we only use results from one clustering approach for this which is the risk-based clustering approach and uh you know that approach also has a lower bound of 50 interconnected returns that that that that get returned um uh and another big limitation is that at this point we do not have labeled data so we assume the clusters of interconnected self-repaired returns we detect uh using our risk-based clustering represent individual ghost preparers and that all returns in each cluster are goes prepared but we do recognize that there may be false associations or returns incorrectly identified as being ghost prepared as well as returns where we do not detect ghost repair involvement and we do not currently have a measure of the extent to which we misidentify ghost preparers and then something else to be aware of is the analysis does span the coveted pandemic where which was a period of where many taxpayers tax situations changed all right so first uh cluster Evolution during the filing season for this analysis we ran the clustering approach um at various intervals during filing season 2021 to provide snapshots of what clusters would have looked like at that point in time so these snapshots were in February March and April we then looked at the returns identified at each snapshot and checked to see if at the end of the filing season they would have fallen into a cluster of 50 or more returns which would have been detected by the model and if they did we sort of labeled them as suspicious um so you can see uh on the bottom of this chart in the bars that's sort of the volume of returns that fall into each of the cluster sizes and then on the top we have some percentages that that indicate you know how many of those returns that would be that were that would have been detected at that size would have ended up being in a suspicious cluster at the end of the filing season so for example if you were to look um at the February snapshot to the left of the slide um you would see that 80 percent of self-prepared returns appearing in a cluster of 20 or more returns in February would at the end of the filing season be in a cluster of 50 or more returns um and you know again looking at the the volume of actual filings at that point you can see that they're relatively small compared to later points in the filing season suggesting that there might be opportunities for interventions early on that um all right um to the impact analysis the larger aim of this analysis I mentioned before is to eventually quantify the risks that ghost repairs posed to Tax Administration however at this point in our research we're working to understand the effect of being in a suspected ghost cluster has on a given return and we're doing this by looking at tax year over tax year changes an important consideration is that anecdotal evidence backed up by some of our analysis conducted earlier on during The Innovation lab suggests that ghost repairs do not work with a group of taxpayers who are representative of the entire tax paying population so to address this we only look at primary filers who appeared in a ghost preparer or suspected ghost repair cluster at some point and we think this should limit the risk of attributing filing behavior of a specific Community or demographic to a ghost preparer so the process is that we ran that clustering approach for three tax years 2019 2020 and 2021 and then for those tax years we selected a sample of primary filers and then considered their returns for those three tax years and then from that data set we were able to to identify two groups of returns those where the filer appeared to be transitioning into a ghost preparer cluster and those were a filer was remaining and it goes to prayer cluster so that means that you know at this point in time both returns were in a ghost cluster however one set you know the the the the primary filer did file in the previous year but um did not but wasn't picked up by the clustering approach and then for these two groups we looked at how the returns changed from the previous tax year our hypothesis is that um you know returns where the individual appears to be transitioning to a ghost cluster should have a more significant year-to-year change in their returns when compared to individuals who appear to be a returning ghost client and we think these changes are due to the involvement of the suspected ghost preparer um we did these comparisons across sort of the returns themselves as well as existing IRS risk metrics so looking to the table at the left there are two groups as I mentioned before those are joining ghost prepare clusters and those who have stayed in the red box the the returns themselves you know are close suggesting that it's a similar population and that um if there's a ghost preparer involved first-year clients are receiving the same treatment as returning clients then to the uh to the I guess to the right of this uh to this chart um you can see that the year-over-year changes are are more significant for um the individuals who or the the returns or the individual appears to be transitioning into that ghost cluster and this is especially true for the refund amounts um and then looking to the table to the right we consider sort of an existing IRS risk metric the discriminant function or the diff score and it's been mentioned you know before in this in this conference but it's an algorithm or technique that predicts How likely a return is to have a significant adjustment scoring is done uh specific to an activity code um just defined by the by by elements on the return and so as a result the scores themselves can't really be compared to one another necessarily or across processing years so to allow for a comparison we just really considered counts of returns which falled in the top five percent of disc scores indicating that they're some of the riskiest returns irrespective of processing here our activity code sort of allowing us to to do a comparison across um these years and in the chart you can see that both groups had similar percentages of returns in the top five percent of riskiest returns as calculated using diff um I mean uh so if you can see here it's significant that you know that that it appears that returns that fell into this cluster are appear to be three times as risky as returns over overall um and in the bottom of the chart we can see that primary filers who appear to be transitioning into suspicious clusters or you know first time you know first-time filers with a ghost preparer transitioned into the riskiest diff um categories at a higher rate than than those uh those return clients it's also worth noting that you know just um the population of filers um regardless of whether they they appeared in the in a ghost cluster or not for this group were just report where like we're we're in the um in the you know at 13 of the returns ended up in the top five percent of the diff distribution so it you know there are risk here then um returns overall but uh and that's not something that we anticipated so um to conclude briefly um you know we do see ghost preparers as a significant challenge we I think we've demonstrated that Network analytics can identify specific ghost prepared returns at least and while results are promising we you know we as we receive feedback and refine these approaches we hope that our results will improve and concludes my presentation [Applause] all right great um so thank you all for uh sticking around my name is Daniel reck um for some reason this this keeps automatically flipping forward so uh let me try and keep it on the title slide um okay so this is a paper called the offshore world according to fatca um it's joint work through the jsrp between an academic team consisting of Niels Johannes and Max ritual slimrod and myself um one with John Guyton and Pat langatigue at the IRS and um one more thing and then that this will stop happening I mean I should note we have the usual disclaimer that these are just the views of the authors um and uh the the paper is called the the world the offshore world according to fatca so you should be bearing in mind when I show you all this that might be different from the real world and how what those differences are you know we're still trying to sort out the paper is a working paper and so you know if you have institutional some insights that would help us make sense of some of these patterns then I hope we can we can talk about that all right so um globally there's an estimate that about 7 trillion dollars in wealth is held offshore that's a global estimate estimate for the U.S a little bit more uncertain this is a great concerns globally about the loss of tax revenue from concealed offshore assets and from what data we have these assets appear to be highly concentrated among the uh wealthiest individuals and so um sort of a regressive loss of tax revenue fat cut is part of a global Crackdown on concealed offshore wealth the intent of fatca is to make it impossible to for an American to hide wealth offshore um in financial assets at least by implementing third-party information reporting so we now have um foreign financial institutions reporting on U.S owned accounts to the IRS um which if you like it conceptually as an extension of third-party information reporting to uh Financial income and assets uh in foreign foreign Source uh Financial income and assets um okay so what we're doing in this paper is to use the administrative data especially from the forms that are filed as part of fatcus I mean by that the form 8966 is where we'll pull the most data to try and understand uh what foreign financial institutions that are doing all this new third-party information reporting are saying you know Americans own um so to construct some some high quality the best descriptive statistics that we can from this data um the second question which I'm not going to tell you anything about today but the team is working on which is sort of how does this relate to the compliance effects of the fact guys what effect did did fact you have on on compliance um so I'm going to focus entirely on a descriptive analysis today on the Micro Data is there data from linked form 8966s or tell you things about you know how much assets reported who holds these assets individuals or individuals holding um assets through pass-through entities um and uh where the individual owners are located in the income distribution um so just by way before I dive into a lot of data on that type of thing by way of background um the factory reporting regime um works as follows so um foreign financial institutions are required to uh identify Accounts at those institutions um that are been or the beneficial owner is a U.S taxpayer or you're going to be more specific a specified U.S person um and one of the one of the important features of this is that it's it's focused on beneficial ownership um but in the US the way the the fat kid is written uh you know a beneficial owner could be a U.S company so it could be a U.S partnership and then U.S Partnerships have owners their partners and so so what you might think of as a beneficial owner from a very kind of economic perspective is not exactly the same as the way things are reported that's going to turn out to be very important because we're going to see a lot of wealth assigned to Partnerships on these forms um okay so it's enforced by a threat of withholding penalties there are some exceptions where an American could have an offshore account that is not reportable under fatca the main exception to know about is that if the account is less than fifty thousand dollars then it's not required to be reported on although it turns out many ffis do report on on those types of accounts they can forego that exception to the rules and a lot of them do um okay so uh in between us we did a lot of data cleaning before we got to this table there's more on how we cleaned the data looking for things like duplicate accounts and so on um in the paper itself um I'm not going to talk too much about some of the data quality issues there are a lot of interesting data quality issues in the first few years of fat reporting as institutions are finding out what they're supposed to report they're trying to track down U.S owners there are a few things about data quality that are going to show up in the mecca Aggregates that I'm about to show you though um okay so we end up I'm going to focus mainly on 2018 where it looks like the reporting is is more or less complete the numbers are starting to stabilize in terms of the assets that are being reported over time and the number of reporting financial institutions so we've got about four trillion dollars in in offshore assets that are reported on the form 8966 in that year that's coming from about 45 000 uh foreign financial institutions um it corresponds to about 4.6 million accounts owned by 1.5 million or so U.S owners along with another about a third of the accounts 1.6 million accounts that we cannot link to an identified owner uh mainly due to missing tens because the tax per identification number was not on the form more about that uh as we go um so this is uh decomposing the assets that are reported on form 8966 according to um the 10 of the account owner that's on the form um and so something that's going to come out of a lot of these is there's a contrast between a typical uh typical account holder a typical account in the data and a typical dollar of wealth so a typical account um more than half of accounts are is owned by an individual so this is directly the 10 on the 1040 matches the 10 on the 89 and 66 it's a directly individually held account uh the next largest category is missing tens I suspect many in account weighted terms many of these are individuals but we didn't get the 10 from the ffi that's that's a sort of a known issue with 8966 reporting that's expected to improve um as the 10 requirements get more strict in more recent years um okay uh meanwhile the typical dollar of wealth or dollars of wealth are much more likely to be owned uh by Partnerships Partnerships have just 1.2 percent of the accounts but uh fully a third of the wealth so that's a sign that a lot of the very large dollar accounts are in Partnerships and other types of entities uh there's a chunk of wealth in C corporations a relatively small amount in other types of entities and then about a quarter of the wealth is attributed to those with missing tens there are a few other reasons why we don't uh allocate wealth um a lot of that is um sort of ambiguous matching issues or issues with eins that show up on that that appear to be valid eins but don't match to any tax returns um so uh we wanted to show you quite some some stuff about sort of where the wealth is located there are some restrictions that that that keep us from showing by country data so I'm going to split uh into two groups of countries here following an academic literature I'm going to call one group of countries tax Havens the other non-taxavers there's a evolving policy landscape so some of these tax Havens May no longer be considered tax Havens but this is just a group of two countries that we kind of understand from the academic literature um so uh the majority of accounts are in non-havens a typical account is in a non-tax Haven Country and owned by an individual so a picture is starting to emerge here meanwhile uh dollars of wealth are much more likely to be located uh in tax Havens about 50 percent of the of the wealth is located um in these countries were designated designating tax Havens this is now going to split the location of accounts and wealth by Haven status and the type of owner that we're talking about um so if you like a typical account uh this is showing you that you know a typical account once again is owned by individuals and that's sort of true in in Havens and non-havens although you're starting to see accounts owned by Partnerships or uh more frequently observed uh in tax Havens uh whereas it counts with uh with where we can't find the owner the owner's unknown mainly because of missing tens those are mostly in non-tax-saving countries um meanwhile for dollars of wealth you see the Partnerships uh the orange piece of this bar are are far more important uh so partnership wealth in tax Havens is you know it appears to be especially important so we're there's the story of the data are telling here that large accounts are disproportionately located in tax Havens and they're disproportionately owned by Partnerships I also want you to notice just how small the the missing piece is uh once we do it things in dollar weighted terms um within tax Havens so there's a few signals here that a lot of what's going on with missing tins is more sort of small dollar accounts are likely to be held by NBA individuals and Havens and so on um but not all uh okay so um let me back up for a second and compare what I just showed you to uh to what we would have expected from prior literature on this type of thing um so the number in Haven specifically is about two trillion dollars so about half of the total um that's ten percent of GDP in 2018 um the best estimate that we know of for the U.S was about seven percent of GDP in tax Havens uh Circa 2007s that was a long time ago how you want to think about comparing those uh you know is is up to you but um it's a pretty large number suggesting that at least you know offshore wealth has not entirely been repatriated because of uh because of fatca um okay now the next question we want to come to is what can we learn about the income generated in these uh in these accounts because that's ultimately what's going to be taxable uh there's been prior work trying to understand that the rate of return is in in a offshore account um the the faculty data are not perfect for this because we see a lot of missing uh data on the income reports on the form 8966 so the 8966 has rows for a total account balance and four types of uh of flows and those flows are very often uh missing so what what we'll do is uh to the the best thing we can we can sort of come up with here is uh first we'll condition on some of the incoming items being present and then compute a rate of return within that sample where we do see income being reported uh so for this table it's just interest in dividends so I'm taking total interest in dividends reported on this 89.66 divided by the uh the total account balance conditional on our observing interests or dividends so this is a rate of return that wouldn't include anything from say realized capital gains um because that's where the uh some issues with the way information is reported get really messy um okay and so uh what we what we see here is you do get a sizeable a decent rate of return in in that type of a sample overall around three percent it's quite heterogeneous by the type of account or who owns it and where it's located um so we're seeing larger returns um in tax Haven countries and for accounts owned by either individuals or Partnerships relatively lower rates of returns for the more exotic types of entity owners uh or in non-haven countries and those rates are returning you know five and six percent I mean those are significant and right in the ballpark of what one might have guessed from uh from previous research on this stuff um okay so so Partnerships hold the plurality of assets and so the next thing that we would be natural to do is to link the Partnerships to their owners and to learn about you know within the you know U.S tax environment who owns this wealth um so we uh have done that we take the uh the 1065s that match to 8966s and then the k1s attached to those and then find you know distribute foreign assets and income to the owners of Partnerships um and we do that based on the share of total income distributed on k1s um there's some issues there with sort of the the relationship between realized and unrealized capital gains and sort of special partnership allocations that may not be perfect you might take it with a grain of salt um but we do look through you know Partnerships that are owned by other Partnerships and so on to get to uh as best we can an ultimate owner um this is just a descriptive fact about the types of Partnerships that we're talking about here this is the the next code that is um assigned to these Partnerships so this is uh this and it's all in dollar weighted terms it's like the share of assets what we see is about 70 percent uh of partnership wealth is in a single next code uh four digit next code was 52.39 that's other financial services this is the next code you expect to see around private Equity offshore Investments uh the financial Partnerships that facilitate that type of activity um the there's this other Financial category that accounts for most of the uh remainder um so there's the so-called Financial Partnerships maybe not so surprisingly um and when we allocate the the wealth that Partnerships own to the owners of the Partnerships this is how things break down in dollars in dollar terms the red is the share of assigned wealth the blue is the share of all partnership income just to kind of give you a comparison um and indeed things break down kind of similarly for fat Co wealth compared to say all partnership income uh about 10 percent a little more than 10 percent matches to entities like Texas impedities and trusts there's a share that it belongs to foreign corporations but most of this wealth uh just over 55 percent we can get to uh we can allocate to a U.S individual like a 1040 filer um and so everything that I'm about to show you in the next few slides is just based on these individuals who own wealth through a partnership who own offshore wealth through a partnership and individuals who directly own offshore wealth and we're not going to focus going forward on any of the other categories uh uh and you might expect that to be pretty limiting but it turns out that we we do have pretty good coverage of at least one segment of the income distribution despite all of those limitations and sort of what you know what we're able to focus on um and that is the very top of the income distribution so despite the fact that this linkage is only based on um you know a portion of the wealth we're ending up with over 60 percent of individuals in the top 0.01 percent of the income distribution who are showing up in these data as owners of offshore wealth according to form 8966 uh much of that through Partnerships that they own um so the the blue is uh individuals who only own offshore wealth through a partnership the gray are individuals who own some wealth directly as an individual and some well through a partnership and the black is individual uh only so you can see that a lot of this a very large fraction of individuals have at least some stake in an offshore account most of that through pass-throughs and most of it also um in tax Havens so this is now adding the dimension of whether the excuse me whether the account is located in a tax Haven you can see of that 60 percent um in the in the top 0.01 percent of the distribution almost all of those individuals have accounts in tax Havens um now even within the top one percent of the distribution that gradient the probability that you show up as an owner of offshore uh assets is steeply increasing suggesting there's meaningful heterogeneity within the top one percent of the income distribution here uh the losses bucket also pop out is somewhat important which because that doesn't include a number of uh High income high wealth persons so it shouldn't be totally shocking um okay so um next we've looked at the uh the dollars of wealth I guess before I before I show you that let me tell you that uh that steep increase although it the absolute level of it the fact that we get to two-thirds of individuals at the very top of the income distribution is a somewhat surprising fact the steepness of this gradient and then the fact that you have uh people at the very top are much more likely to own offshore wealth than people even a little bit further down in the income distribution that that steepness should not exactly surprise you based on prior work from uh you know based on the offshore voluntary disclosure program or amnesty programs in other countries um finally in dollar weighted terms what share of the assets is owned uh by individuals at different parts of the income distribution continuing to focus on uh directly individually held wealth and individually individuals owning wealth through Partnerships and indeed the wealth is quite concentrated so um for wealth owned through Partnerships specifically almost half of the wealth is owned by the top 0.01 percent of the income distribution uh the Lion's Share of the rest is belonging to uh at least the top ten percent of the income distribution the wealth that's in that's owned by Partnerships is more concentrated suggesting you're owning a welfare partnership is a sign of sophistication that stuff is more concentrated at the top than the directly held uh individual wealth um and these numbers uh are once again if you there's a more detailed uh laying out of all of this in the paper but they're roughly comparable to what you might have expected based on the partial data sets of owners of offshore wealth that one studies in Prior work um we have a in in summary of that about 46 of reported off uh offshore partnership assets are owned by the top 0.01 and 80 percent of those by the top one percent um so you know given what we know about the concentration of offshore income and also the concentration of partnership income specifically so about 70 percent of all partnership income uh ends up in the top one percent of the income distribution what we're seeing in the fact get data sort of gels with all of that but um this is the first time that we're seeing it um and then finally this is about how the dollar totals break down between Havens and non-havens we're seeing that uh non-haven wealth is actually offshore wealth is actually rather concentrated so it's not the story that that um that all of the the uh the concentration is in tax Haven owned wealth uh nevertheless uh you know you see comparable rates of extremely concentrated ownership in Havens and non-havens a little bit more concentrated uh in tax Havens okay so putting it all together uh including Havens non-havens pastors individuals it's about 30 percent of the total wealth that's owned by the top 0.01 percent and uh about oh I don't know if I have this one in a picture it's it's about 60 of the wealth at the top one percent um okay that I'm gonna skip over that that's a if you want to see a really detailed comparison of this to Prior work on that uses things like the Nordic amnesty the uh there was a Colombian amnesty program you see quantitative comparisons to all of that then that's um in the paper okay so what do we what do we take away from this um fatca is providing new evidence on how much uh wealth Americans hold abroad uh about four trillion dollars of financial wealth is reported on these forms uh about half of that or about two trillion dollars um is located in tax Havens that's on the large side of what you would have expected from prior work but it's not in you know not to an insane degree um and uh a large share of this wealth is held indirectly through entity set at least 46 of the wealth is held indirectly an important implication of this for our ongoing efforts to understand the compliance effects of fatca is what was the compliance effect for Partnerships is you know about 50 of the answer to the question of what was the compliance effect of fatca overall uh because about half of the wealth is done by Partnerships um and so we really need to drill into the Partnerships and embrace the complexity that that entails in order to understand uh the effects of fatca um so if you're thinking about the incidence of the the revenue effects or the compliance effects of fat good those are likely to be highly concentrated because a typical dollar of wealth and thus a typical dollar of income generated by offshore wealth it is falling pretty high up in the income distribution although uh you know there are a large number of individually held small dollar accounts where we might worry about things like the compliance the cost of fatca that are further down uh in the income distribution um so these these are just some of the uh some of the more striking figures on the the extent of the concentration uh of the fedco wealth so that's um 62 percent of households in the top 0.01 percent of the income distribution um 64 of the wealth owned by the top one percent uh 77 percent of uh the top 0.01 foreign assets are held through pass-through businesses especially important pass-throughs are especially important at the top um okay and so uh those are important distributional facts to be keeping in mind as we evaluate fatca and it's pointing to you know what are the important populations to really understand if you're trying to understand the compliance effects of fatca um so with four trillion dollars in wealth and a decent rate of return on that wealth there is a substantial potential compliance effect of fatca but uh we still don't know really how tax compliant these populations where the data are telling us is especially important to understand like were those populations tax compliant after fat could in fact actually make them comply and were they in fact non-compliant before factor to what extent were they uh non-compliant before fatca um and then you know everything that I'm showing you is sort of after any repatriation effect of fat because there's been some some recent work on that suggesting that there's a substantial there is a significant compliance effect coming through that channel um as well it's a challenging question yep I sure can uh that's a challenging question to answer the compliance effects so maybe uh maybe next time thanks everyone [Applause] let's get through my appendix for you all right thank you very much I'm Paul Oregon I'm an economist in the office of tax analysis at the US Treasury very happy to be here closing us out with a brief discussion of these these three papers and I'll just note these are my opinions not those of the treasury so I'll be discussing these three papers and I was trying to think what what connects these three papers what brings them together and I came up with two two things so one is these are papers that are either they're applying new methods to existing data or they're thinking about our existing methods and applying that to new data and they're also highly operationally relevant and also of academic interests so I think this is very much in the spirit of this conference and of what Barry was talking about this morning about bringing academics and the IRS together to work on things together so starting with the the K1 paper following the k1s so I think there's quite a bit to like about this paper so I think it's a clever idea to look Beyond just an individual taxpayer and understand that taxpayer sits in a network of all kinds of other actors they're tax preparer potentially their Partnership if they're in a partnership Network other taxpayers Allen alluded this to this a little bit in thinking about spillover effects and that there's some research in this and so these two sub bullets here I think you could think there's sort of a predictive or a correlative question so if you look at the people around a taxpayer what can we learn from looking at them about that taxpayer the second subtle there is sort of about a causal question what what does how does what other people in a taxpayers network how they act how does that influence the focal taxpayers behavior and I think that this research could speak to both at the moment I think it's mostly speaking to that first question and that's the main goal and I think that that on its own is that on its own is interesting but I would also encourage the authors to consider that second question and maybe push in that direction whether in this project or another and then I think it also shows uh something interesting about how you can observe one group of taxpayers about whom you might have a lot more information or just particular information here that they're reporting foreign accounts and then see how carefully analyzing that group you can then learn about a different group of taxpayers and also I think they as you saw carefully thought about what variables should be included in the model what are the costs and benefits of that so I'll keep my comments or suggestions for this paper are threefold so one is trying to take advantage of the richness of information in that taxpayer Network then trying to understand how well the prediction is working so back testing it and then I'll talk a little bit about maybe a next next steps timing and mechanisms um so Tomas presented that or showed that Network and you saw how complicated it could be there could be the focal taxpayer who is receiving a K1 from that they've got a 40 share of and they're also receiving a K1 with a one percent and there's all these there's so much information contained in that graph and at the moment that's all getting distilled down into one binary variable does this taxpayer I guess there's a little bit of a timing Dimension but it's like is there uh reported for an account in that Network or not and I think um I would just encourage the authors I think you could use that Network and collapse that with a little more richness thinking about how many accounts are in that of the dollar value in this network how much is represented by the foreign accounts versus not um and I think that that would help the information then contained in those variables would be a lot more precise than what is right now just getting condensed and flattened into a 1-0 and so I think that would help lead to tighter predictions and then another sort of side note but would be maybe making that Network even more complicated you know adding the taxpayer and thinking about how the tax preparer the accountant they're involved here too I think that would speak to the mechanisms too so I'll bring that back up okay my second comment is sort of about so the goal of this paper maybe the main goal is we want to use what we observe about taxpayers who are reporting foreign accounts and see what does that tell us about the taxpayers who aren't might they be might they have accounts that they're not reporting and I think a key question then is does it work like we build this algorithm we build this prediction approach is it actually successfully predicting the taxpayers who we should maybe go go look for and I think there's two approaches you could take that one would be we have our predictions now we go forward we do some Audits and we evaluate maybe that's in the cards but I think you could also try to find are there examples of existing historical data that you could use to back test this algorithm and then maybe improve it and tweak it um some ideas there where you note there were these programs over the 20 tons where there was increasing disclosure of foreign accounts you might be able to use some of that information where you observe taxpayers newly reporting accounts so if you roll back a couple years apply your approach of following the k1s and say is it flagging the taxpayers who you then expect that you think should or should be reporting an account do they actually show up reporting an account maybe the NRP would give you something here where you have some Randomness because of course the reporting of accounts during the 20 tons was endogenous and not random but the NRP might give you something um and then I do think uh very related to Daniel's paper what would this approach you have you observe the report of foreign accounts it tells you something about the potential number of unreported accounts what would that say about the size of undisclosed accounts and how would that relate to what we see in the facca data all right and then finally sort of an idea for future research here this is more on the causal question about how like what is causing taxpayers to report a foreign account and so my idea here is something about one mechanism for that might be other taxpayers in a focal taxpayers Network starting to report and that information just kind of permeating throughout that Network hey I just reported my account maybe you should consider doing that or I just got a letter you should do this you could imagine it flowing through this network um and I think you could test for that so you have these data You observe the first year in which the focal taxpayer has someone in their Network reporting I think that's a nice setup for an event study approach I think you're pushing in that direction which is great and you actually have some evidence of this so you do test current versus prior year and you see a different effect there so I think that suggests this is worth considering okay so moving to the ghost prepared I couldn't show a ghost prepared but this is a prepared ghost uh with with the first aid kit there um so again I think this is another great example of a paper that's applying to me very exciting new methods um and I think it's a really great example of often those methods are exceedingly complex and Technical and so it's not necessarily a method that that I could apply or that most people can apply but what I like about this paper is they are applying that method and Distilling it down and explaining it in a way that practitioners and others at the IRS can then use that tool so we are all we all benefit from the use of those methods and that's just great like that's a thing that I think we should all strive for and finally I think this they're they're trying to build a tool that is adaptable so as new data comes in they can build that into this clustering approach and it's it's not just a fixed tool that is then done so two big comments on this paper pretty similar to the first paper one is how effective are the algorithms and then the second is just thinking about how to explain the different clustering approaches and and the costs and benefits or pros and cons of those approaches so my comment here is going to sound pretty similar to the last one it's we developed this approach and where we are trying to predict potential clusters potential ghost prepared clusters and what we really want to know is did it work did we actually prepare a cluster or not and so again you could do that forward-looking and that may be happening but I think there might be some um just I think it's worth getting creative to think about a way that you can find some some past data so the paper notes that they're at present basically the way that ghost preparer networks are noticed or identified is kind of ad hoc referrals people get a tip that it might be going on or it comes up in another compliance engagement that they're checking something else and they discover that it's a ghost repair so of course those are totally not random but they are there they exist and you that gives you a flag of this was do you have some historical information about what were truly identified ghost preparer networks that might allow you to then apply the algorithm in the Years leading up to that and see if it would predict that and then I think you could iterate on that um and so Josh also noted there are a couple of different cluster approaches and a key piece of that is the information that is going into the algorithm so there's a risk based on a top-down with the label propagation I think it'd be really helpful for the reader to see a table of like here are the approaches what are what do we gain by doing this approach and what what's the cost of that well so that we could decide like what is the correct route to go down and then thinking about this as sort of a repeated game with the ghost preparers that they are adapting to what the IRS is doing and changing their behavior to not get identified so thinking about the information that goes into each of these approaches one of that information is a fixed characteristic that the taxpayer can't change versus um for instance one of the pieces of information is about the phone number that they put in that's something they can play around with but there might be other characteristics that are fixed and we might want to build a tool that relies more on those fixed characteristics that the tax the ghost preparers are less able to game okay and so finally here we have the uh the offshore world according to facca I'll have a comment on the subtitle of this paper shortly so stay tuned for two minutes from now um so things like about this paper really exciting new data I think we all just learned a lot in the last 15 minutes so it's really exciting to see this data finally being being analyzed really careful detailed linking I know it probably took a ton of time to take those Partnerships and actually get them done individuals we just get to see on slide two or whatever the table but you could probably have a thousand slides before that so Props on that and I think of course this is going to spark some some follow-on research of course so I've also had the the fortune of discussing this paper at length with the authors and so I was trying to think of what are new comments that I can that I can tell them that will help improve the paper from where it is and so I'm going to talk about one mostly which is about what what we can learn or what could we learn about those unmatched 40 and then what that would inform about the overall the universe and I'll talk a little bit about U.S citizens abroad um right so in that in that table of kind of splitting down by types of owners that Daniel showed about 40 of the accounts you know 38 of the wealth a little more of the of the accounts uh is not able to be matched and there were a couple reasons the big one was no tens but there's some other things and I think partly inspired by the earlier two papers in this session I was thinking like what what could we do what could we learn about that 40 unmatched to predict something about if we had more information or you know what what bucket should it be going into and I think we do have some information about those Council like the 8966 that's filed by the ffis well first I think it would just be useful to know like is it that whole ffis are reporting on accounts and none of them are matching or is it that an ffi reports 100 accounts and most of them match but it's a few that don't those are different situations and we might think differently about this um but also even for the the accounts that we do get we know some things like what currency is the account reported in um understanding we don't always have every field but we know some things and I think that the approaches of the first two papers are thinking about prediction and so you could take some of those methods and try and do the same thing here given the information that you have predict do your best to get a probabilistic sense of what bucket it would fall into and then that would help us learn more about what we could say about the universe okay just finally I'll note so some of this is going to be U.S citizens who are living abroad and so calling this offshore wealth foreign wealth of U.S households I think is mostly true but it's certainly the case that some of it is what what they would call local wealth of foreign resident U.S citizens and that is a mouthful and I would not add it as a subtitle you know new evidence on foreign worth of U.S households and a little bit of local wealth of some foreign yeah that's not good but I do think you could quickly get a a lower bound if you look at the addresses are they foreign filers and just right off the bat of the individuals that's a quick statistic to get I also think of the unmatched the 25 no 10 probably a huge chunk of those are accidental Americans or foreign residents and maybe the predictive method would pick that up potentially if they looked similar to the Matched tins who were foreign resident that would give you something so I'll I think I'll leave it there I'll leave it with a teaser which is I'm very excited about the causal work I know that it's ongoing and I'm excited to see that so thanks very much [Applause] foreign we have some time for a few questions from the audience if there are any questions yes um this is a question for Josh um so um Paul mentioned an idea to um you mentioned you don't have any label data on who's actually a ghost preparer and Paul gave my idea of looking at you know referrals maybe another idea could be looking at tax promoters to see because those are identified by the IRS so maybe looking at tax promoters to see what types of networks around them as a way to infer something about Shady repairs in general that's a good idea thank you questions for the first and third paper so the first one are you grouping together k1s from Partnerships and from trusts because those seem like very different stories so like in particular I would think that the partnership k1's right these are mostly private Equity firms where's the trusts you're probably going to have relatively small networks that mostly just consist of you know a few beneficiaries that sort of thing yeah so we are including um s Corps Partnerships and trusts that is something that is available that we did look at for the model so whether or not that K1 taxpayer model we did look at whether they received the K1 and who the the payer was I mean I think it would be interesting to see coefficients separately for those entity types I've moved my hypothesis is that there's a very different combination but and briefly for for Daniel so I guess one puzzle is why is this money in in a Haven at all we have a like for individuals if you're a U.S citizen we have a worldwide tax system not for the c-corps but you know for the other taxpayers like what are they getting out of it and I guess one hypothesis is they are Sheltering uh trust income from their state and so it'd be interesting to know whether there's a difference between the proportion that's held in Havens from high tax States versus low tax states relative to the overall you know expected wealth from that state okay so thanks for the suggestion I I had not thought of of doing that and that's something that we can certainly look into um to the broader question about why people own uh wealth in Havens uh I think um there could be some types some tax motivated reasons like the one you suggest um the a big tax motivated reason that's supposed to have just changed is the enforcement environment um so why people still hold wealth and Havens uh you know I don't fully know the answer to that question I think um from from what I have read you know there are investment vehicles that open up to you if you invest through uh uh through offshore companies because that are dodging some rules about how U.S companies can invest so there's some potential there I I don't pretend to be an expert on all that stuff and if anyone in the room wants to talk about it after then I'm all ears foreign quick question for either Paul or Daniel just that last Point she made about foreign residents um isn't there income still taxable uh by the U.S and so why exclude them from the calculation uh yeah so if your only goal is to understand tax compliance for the reason you've just mentioned you you wouldn't want to throw them out of your sample if you're trying to compare to Nipa data on say all U.S households then those people are not considered U.S households According to some of those types of economic statistics and so you might want to to exclude them it's a it's a fair point by the way that that's that's something we can we can look at a little closer and I'll just clarify I'm not suggesting to exclude them at all just uh like I think it's interesting to know how much is one versus the other and that there's different policy implications of what we might do about one of the other one uh yes I have kind of two questions the first one for Josh and I think more of just a comment one thing that seemed really interesting was your statistics kind of on what the returns of these ghost repairs look like that they increase income but also increase the eitc credit and I think it'd be super interesting if you could show and if it does happen you know because you have a phase-in and a phase out for the eitc if if you're on the lower phase-in threshold your income goes up to get you the max credit and if you were phasing out the ghost preparer brings your income back and lowers it uh to again get you the max credit and if that's more likely with cost preparers than not that is a really good idea it would be yeah that's a good idea that's something we should do um and then or uh Daniel the law during your presentation you mentioned wealth I might be wrong on this but right you only have kind of the monetary or stock accounts if the partnership say partnership is in Bermuda and it owns an apartment building in Germany uh do you have like the value you don't have the value of that building right from the factory reporting you just have the monetary assets uh okay so this gets a little complicated um so my unders if you haven't say an offshore company that owns the what is it some real estate and pick a country wherever um then the shares that you own in that offshore company can be an offshore asset that's reportable under fatca depending on how you own it who does the reporting like who's the ffi um if it's extremely closely held you can end up in a situation where it's you um so whether that stuff gets reported on whether they comply is sort of dubious but you can end up with a lot of situations like that where you know wealth is held in a way that there is a financial asset that should be reportable under fat even though the what you might think of is the more primitive underlying asset is real estate or yeah okay this question is just about one of the tables um Daniel that you had on the ffis where it was like 45 000 is that controlled for ones that issue 89 66's or just ones that are registered and could issue a form uh that is a count based on the 8966 data so that is the number of ffis that file uh an 8966 for at least one account holder that is not a useless record that we scrub from our data so literally that's what the number is my question is for Tomas you mentioned in passing that you had used diff scores to try to identify new clusters of ghost preparers can you elaborate on that a little bit hi Josh sorry I'm sorry Josh um sure I think um outside the work in the paper we definitely did look at um diff scores for you know returns that ended up in suspicious clusters versus returns that were you know prepared by uh preparers or self-prepared and um and then within preparers we tried to like do some sample matching with so that we could like try to try to replicate the population of of individuals who appear to be having their returns prepared by ghost repairs and we did find that the returns that ended up in these sort of suspicious ghost prayer clusters um were at least according to um you know the diff calculations you know um more risky than um than any other population we considered including those of um those returns prepared by the riskiest repairs that are as identified by another risk metric so um we did that was one analysis that we did that didn't end up in the paper um but yes but we are not including we are not diff is not part of the clustering approach other questions no so I had a question um respect for Tomas and uh Daniel but especially for Daniel so we've been thinking about compliance and moving trying to do more in compliance and quantifying and I'm wondering if we've also heard a little bit about astinal Americans and um people who were caught up in this system I was wondering if you had any thoughts about trying to work more on the taxpayer compliance burden well I should mention there's an excellent study related to this by Paul um on people who renounce their citizenship uh in response to fatca um so there's some research on that and I think it's important I think you know if we ever get to the stage where we have enough of uh attraction on compliance effects to sort of think about you know evaluating factual systematically you would you really would want to bring in some numbers um and try and get some numbers onto those sort of compliance costs um because it is important I agree yeah I guess that a little bit differs between um kind of our work and what Daniel is doing is that the majority of our population were F Bar filers and so fbar the threshold is ten thousand dollars per phone reports uh for an account so it is most likely going to be and individuals so kind of with Daniel show so it is definitely going to be more like um potentially kind of accent of Americans or um probably not that point zero one percent um that they don't try to get out there recent U.S citizens too all right and a few others you got me in there okay any other questions no okay [Applause] wow I hope you all have as much fun today as I have but this has just been a really great day thank you to all of our speakers for taking time to share their research with us it takes a lot of work to put together these kinds of presentations and to prepare and we really appreciate all the effort that went in it really showed in the quality of what we saw thanks so much to our discussions for taking for reading the papers and and offering such thoughtful and helpful suggestions for how we can improve the research uh everyone should look forward to the the published volume coming out sometime next year uh with with these uh papers in them thanks also to the conference organizers that's a big job sifting through all the different papers and putting together coherent sessions that are interesting and hang together and and you did it just a marvelous job every year we think uh while the what a great conference how's it going to get better but every year it seems to get better so so thank you so much for that thank you again to the tax policy Center and to Brookings for co-sponsoring and for hosting um to make today so special and then finally thanks to our audience both those of you who are here in the room and for those of you who are virtually with us I hope that you all really enjoyed this and learned something today and I hope that we see you all back together again next year so thank you all [Applause] foreign

Channel: Brookings Institution

Views: 2,963

Keywords:

Id: A7HaVQomH_0

Length: 491min 54sec (29514 seconds)

Published: Thu Jun 22 2023