Andrew Lo discusses systemic risk

I'd like to start by thanking Chester for that very generous introduction and thanking the program committee for inviting me to participate in this terrific conference on systemic risk and data issues and I'm particularly pleased to be part of the session on the shadow banking system because I think that that really is at the heart of systemic risk and data issues and I'm going to try in the next 15 or 20 minutes to illustrate to you why that is particularly for the hedge fund industry which I was asked to speak about now before I begin let me ask how many people here are actually familiar with the hedge fund industry yeah not not everybody so I have to admit that you know until 1998 I really didn't know much about the hedge fund industry either but LTCM really put that industry on the map particularly for academics and when I asked one of my friends at the time you know how to define a hedge fund he gave me an answer that it didn't quite understand at the time but I think I do now and the definition was kind of interesting he said a hedge fund is a partnership it's a private partnership that has limited partners and general partners and it lasts for a finite amount of time and at the start of this partnership the general partner brings the investment experience and the limited partners bring the money and at the end of the partnership the general partner leaves with all the money and the limited partners leave with all the experience so you know there's a lot of skepticism about hedge funds but the fact of the matter is they're big and popular and more active now than ever and we don't know a lot about them so I want to describe to you just how much we don't know and to do that let me first start by describing what we do know about save the economy so let's just go down the list and think about the top five or six statistics that we cite when we talk about how the economy is doing we know that inflation as of last month is 3.8 percent we know that unemployment is at nine point one percent serious problem we know that GDP growth is at 1.3 percent we know that non-farm payroll is a zero change from last month we know that housing starts at three point two percent and we know that the Fed balance sheet is at two point eight trillion dollars we know a lot about the economy although I think that all the economists in the audience would agree that that's not nearly enough for us to really get a good sense of how to manage the system now let me show you what we know about the hedge fund industry this is what we know about the hedge fund industry now when I say that I'm using the word no in the legal sense you know can you actually base a regulatory action on information that you can actually verify in the hedge fund industry and the answer is no because hedge funds aren't obligated to report now there's a lot of things that we think we know about the hedge fund industry so let me tell you about that but after most of these so-called facts I have to qualify them by saying that we don't really know so for example we think we know that the hedge fund industry is larger than it's ever been the estimates are that the hedge fund industry today is at about 2.3 trillion dollars now Russ just before I mentioned that the money market funds are about three trillion dollars so it seems like money market funds are you know bigger but that neglects the fact that this 2.3 trillion is without leverage and it also neglects the fact that these 2.3 trillion dollars are not invested in triple-a bonds with credit protection second the industry we think is becoming more concentrated thanks to the financial crisis a lot of investors now no longer want to invest in the small start-up hedge funds they want to invest in big stable institutional quality hedge funds and that means that there's a much higher concentration I'd love to be able to show you a graph of the - dollar index but I don't have one because we don't have data for that the biggest source of new inflows we think our pension funds and that's a source of concern because pension funds as most of you know are mostly underwater right now they're trying to boost their asset growth because their assets are below their liabilities and by ERISA laws they are actually in violation I'm going to have to make contributions so they're looking for higher yielding assets and the only place to go when you're looking at higher yields if you go through the historical data as misleading as it is is hedge funds traditional strategies are underperforming as far as we know we're not sure because we don't have regular reports but things like convertible arbitrage statistical arbitrage global macro all sorts of strategies have really faltered over the last couple of years there are pockets of good performance but some of the traditional strategies have really had a hard time the returns are becoming more highly correlated that is the returns that we observe and we only observe the returns that hedge funds voluntarily report so the task database the Morningstar database those are databases of hedge funds that have voluntarily agreed to share with us the returns but for the most part we don't have access to what's really going on in the larger and broader industry we think that leverage is more dynamic meaning that now hedge funds are moving their leverages around much more rapidly as a function of market conditions we think we know this because we talked to prime brokers and brokers seem to suggest that that's what hedge funds are doing but of course there are a number of hedge funds that use multiple prime brokers and they spread their trades in ways that can be very difficult to tell what's going on and finally it seems as if hedge funds are becoming more short horizon oriented for a variety of reasons that make sense but in fact we don't really know that it just looks that way from the data that we have access to so that's a lot of hemming and hawing and hand-waving but that's where we are right now and so let me just share with you a little bit about size because you know Rob angle at lunch today gave his list of top 10 and Russ had his list so I've got to come up with a list as well so my list is the list of top 25 hedge funds as of the end of last year as surveyed by institutional investor and you probably can't see in the back so I'll just read off a couple of names none of these are real household names except for maybe one or two that you know our old-time hedge funds but the largest hedge fund today is a company called Bridgewater associates they're based in Greenwich Connecticut they've got fifty eight point nine billion dollars that's one institution one hedge fund has fifty eight point nine billion dollars what do they do we don't know multi-strategy currency trading futures that they could be doing any number of things they're offering memorandum is pretty broad JP Morgan Asset Management is fifty four point two billion man investments forty point six Paulson and company the fellow who bet against the mortgage market years ago Paulson and company has thirty five point nine billion dollars in in in 1999 you know in the aftermath of LTCM Paulson was an analyst he hadn't started his hedge fund until 2003 he started a hedge fund with 300 million dollars he's now at thirty five point nine actually I think it's a little lower because he as some of you may know he lost thirty percent over the last two months now the smallest hedge fund ESL investments on this top 25 list is 14 billion and 14 billion may not seem like much in the grand scheme of things but remember that these assets are highly leveraged and very dynamic and invested in high volatility assets and you might say well geez you know there's been a lot of inflation you know maybe these assets are really not as much as they were you know years ago so let me show you by comparison in 1998 dollars during the days of LTCM what these assets would look like in 1998 dollars Bridgewater has forty two point four billion in fact the smallest hedge fund on this list ESL has 10.1 billion and just by way of comparison LTCM you remember that fund they sort of gotten a little bit of trouble 1998 they had four point seven billion that's right but you know they're at their peak they were at something like 8 billion right which is still smaller than you know the the hedge funds listed here and the list actually goes on quite a ways obviously so you know this should be of some concern the fact that we have very large very active investors that are basically opaque so to give you one example of the kinds of things that we might be able to do to get some understanding of what's happening let me just quickly review some research that I've done with Aamir Khan Doni Amir is a former student of mine who's now at Morgan Stanley and actually he's here today sitting in the back so I may call on him to to comment on some of these things from his current perspective but in August 2007 something really strange happened and when I asked this audience what happened in August 2007 my guess is the most common response is oh the LIBOR Oh is spread blue out during that second week of August and the Fed and other central banks had to inject liquidity on that Thursday and Friday but if you ask the hedge fund manager what happened in August 2007 they probably wouldn't start with that what they would start with is the quant meltdown the fact that all quantitative equity market neutral managers seemed to have lost money exactly at the same time during the second week of August for no apparent reason at all and it was so strange that the Wall Street Journal wrote an article about this how market turmoil the turmoil waylaid the quads and so going through this you know I got curious because several of my former students who went into this industry called me up and asked me how things were going what was going on and did I hear from any of my other former students that were in the quant equity fund and so finally I called on Amir who is a graduate student at the time they said Amir we ought to you know look into this and try to figure out what's going on but of course nobody's talking because this is the hedge fund industry there's no transparency so Amir and I decided to simulate a very simple strategy mean reversion strategy today you by yesterday's losers and today you CEL yesterday's winners so you're basically simulating a market making strategy and let me show you what the simulations look like now we simulated it originally for daily but then we got a hold of transactions data so we were able to simulate it for 5 minute 10 minute 20 minutes our returns and so this is a graph that shows you the cumulative profits of this mean reversion or market making strategy during that second week of August we actually simulated from July 2nd to September 30th 2007 if you simulate a one-hour mean reversion strategy the cumulative profits look like that it's you know slightly positive a little bit of a dip right around that second week of August now if you go half an hour a little bit more of a dip and if you go 15 minutes a bit more of a dip if you go 10 minutes a bit more of a dip and now if you go at 5-minute intervals you see something interesting a higher frequency at the time it was high frequency now it's pretty slow and plodding but back then at 5-minute mean reversion strategy started losing money the beginning of the week and ultimately it lost quite a bit of money and then effectively recovered and became profitable thereafter from this kind of simulation we can piece together some conjectures we conjecture that there was a massive deleveraging that occurred during the first two weeks of August it actually started August 1st at 10:45 a.m. and it stopped at 11:30 you can actually see that from the tick data on August the 6th the second wave began at 9:30 and it went until one o'clock and it affected particularly earnings momentum and book to market portfolios in the financial sector on August the 7th price momentum and cash flow to price portfolios got hit August 8th and 9th all groups lost money and there was a sharp reversal in August 10th all of which seems to suggest that there was some kind of a massive unwinding of a stat our portfolio during that week now when we wrote this and published the paper we felt a little bit odd about this because I don't think we've ever written a paper I certainly haven't where you're actually writing about something that where you're making a guess and you know for a fact that there are people out there that know what actually happened but they're not talking so in fact this entire paper could be science fiction or it could be dead on we have no idea to this day we don't know because nobody's talking they're not allowed to talk because that would disadvantage their shareholders just out of curiosity you might wonder how that strategy is doing well this is the one-minute cumulative return strategy starting in 2006 and the strategy's been working pretty well at a one-minute interval until just about a year ago and it's actually having a very hard time right now maybe we have to go to the sub millisecond level that's possible but the point is that this industry is extraordinarily dynamic hedge fund managers and investors go where the money is and if the money's not here they're on the move and they're doing something else right now they're doing something else we don't know what that is a second example is some work that I've been doing on Granger causality networks with Monica B Leo Milligan Manske and loriana peláez on here we're trying to understand just how interconnected the financial system is and so we use standard Greene Granger causality tests to measure the relationship between the monthly returns of hedge funds broker dealers banks and insurance companies and what we find is that when you graph these networks these are relationships that have a 5% level of statistical significance between one institution at time T and another institution at time t plus one it looks like a bit of a yarn ball here all right this is a three year period from 1994 to 1996 where we're graphing all of these connections the color coding tells you what the institution is red for hedge funds green for broker dealers black for insurance company and blue for banks and you can see that there are definitely connections among these institutions back between 94 and 96 let me show you what the graph looks like from 2006 to 2008 so that network is a lot denser and denser still even today and you notice that it's not just the banks and broker dealers that are responsible but the hedge funds have tremendous connections to these and that's a sector that nobody knows anything about in fact we don't even know how many hedge funds there are I can tell you right now that there are three thousand nine hundred and ninety stocks trading on the New York Stock Exchange I can't give you a number for hedge funds I don't think anybody knows how many hedge funds there are and I suspect that that's an issue so the point is that indirect measures are suggested but they're not conclusive and think about the National Weather Service or the Geological Survey or the Census Bureau and about how they do their work with the data that they have and now imagine if you don't allow them to have the data how can they do what they are supposed to do we need data in order to talk about systemic risk and that's why this conference I think is so terrific because it focuses on exactly those two issues data and systemic risk now the problem with data is that there are privacy issues particularly with financial institutions because you want to be able to allow financial institutions to develop proprietary technologies without having to give away their secret sauce to their competitors and so there's a tension then between privacy the need to have confidentiality and the need for there to be some kind of disclosure so the question is is there some compromise that we can strike between the two because this is a really serious problem and hedge funds in particular are being very recalcitrant about trying to give up any of their highly valuable information and so I want to tell you and conclude with the answer yes there is a compromise in fact it may not even be a compromise we may be able to have our cake in either two and it has to do with some research that's been going on in the computer science field known as secure multi-party computation Amir and I are working with a third co-author Emanuel Abbe a a computer scientist who's here as well in the back of the and what we're trying to do is to develop methods that will allow us to compute systemic risk but in a completely private way and in order for me to illustrate that I have to give you an example so let me give you an example about something that all of us feel rather private about let's say our salaries it'd be interesting to know what the average salary is in this room and so let's just start with a quick survey shall we so let me let me ask up lemma lemma would you mind telling us what your salary is oh well I say so my guess is that most people despite these kinds of publications would not be very comfortable talking about salary so let's try to think about whether we can come up with a method for calculating the average salary in this room without anybody having to reveal their salaries now how could that be that seems ridiculous well I'll give you an example of how we can do this this is using an idea that was developed in the computer science literature about 30 years ago I'm gonna start I'm gonna give lemma my salary but I'm gonna add to it a random number that I select so only I know what that random number is so the number lemma is three hundred and fifteen dollars that number is my salary plus a random number that I picked and I'm gonna give that number to you and no one else so only you know that number you're gonna do the same thing you're gonna take my number y1 and it's close to your salary Marilyn is not a state school is it all right you're gonna take my number which is my salary plus a random number and you're gonna add to it your salary plus your own random number and then you're going to take that number which is y2 and you're gonna give it to Richard and then Richard's gonna give his number plus a random number to Nancy Nancy's going to give her number plus a random over two so and so on and so forth until finally Chester's the last person we give Chester the number plus the random numbers for everybody and Chester adds his salary plus his random number so now what Chester is in possession of is the sum of everybody salaries and the sum of everybody's random numbers which means nothing it's garbage right but then Chester now gives me his that number subtracting his own random number and then I take it and pass it to lemma and lemma you subtract your random number and lemma you give it to Richard and he subtracts his number and passes it to Nancy and so on and so forth by the time it goes back to Chester what does chester have when chester subtracts his random number he has the sum of everybody salaries and when chester divides by the number of people in the room we then have the average salary of the room nowhere and at no point did anybody have to reveal their salaries now this is a pretty simple idea but you see the power of it and there are ways to break this by colluding for example if i cluded with nancy i could probably figure out a way if i can measure other people's incremental changes to try to eke out information about people's salary Frisch and if enough people collude then we can figure it out so if everybody in the room but chester is in on it we can figure out chester salary but it requires all of us to collude which is pretty hard to do I don't know how many of you have ever dealt with investment banks and financial institutions but you know getting them to talk to each other is a challenge never mind getting them to collude so the point is that in a paper that Emanuel Amir and I are working on we actually show that using cryptographic methods you can compute securely privately means standard deviations correlations Kovar marginal expected shortfall all of the statistics that you would like to have for systemic risk measures securely with multi-party privacy and the reason we did this is because our view is that you don't need all the data in fact I think that that's a ruse that the industry is put up which is to say oh well we can't possibly do this until you ofr guys you get all the data and then we'll talk you don't need all the data you can compute systemic risk statistics in a secure and efficient manner using technology that we have available today so this is just an illustration of some of the things that we do in the paper I'm gonna let Emmanuel explain this if any of you ask me about it but these are encryption algorithms that basically take simple 0/1 computations and show that you cannot reverse-engineer floating-point operations in any meaningful way there's no more excuses in our view once we show how this is done so let me summarize by saying that you know there's lots of applications for this one application that is related to some work that I've done with Amir and another co-author adler kim is looking at consumer credit risk we were able to get 1% of customer data from a major bank with all of their credit card transactions all of their banking con transactions as well as their credit bureau history and using this 1% data we were actually able to increase the forecast power of consumer credit delinquencies and defaults by a factor of 10 and this is extraordinarily proprietary data in fact one of the things that we were not allowed to get from this data was zip codes and we under wondered why the heck can you not give us zip codes we just want to know which part of the region of the country was affected the most well apparently there's a law against that because of redlining so data issues are tremendously important and let me summarize by saying that because the financial system is much more complex really measurement is the first step in addressing that complexity and the ofr is absolutely central to this effort in terms of its data emission standards analytics and research but financial innovation does require privacy and we think that multi-party computations can resolve this conflict so you know let me conclude by saying that you know these privacy issues may be new to economists although the mechanism design folks I think I've been working on this for years now but for most of us privacy is relatively new but I came across a story that somebody told me recently that brought this home to me it was a story about a 11 year old boy who at school learned that you could get people to tell you all sorts of secrets to give up their privacy if you simply told them I know everything I know existed you know went to his mother and he says mom I know everything it's not a secret anymore and she said listen son here's $20 do not tell your father so the next morning the boy says to his father dad I know it all I know the secret and the father said son here's $40 do not let your mother know that we talked and so the boy said this is pretty good he's on his way out to school and he sees the mailman's he decides I tried to the mailman you go to the mailman he says excuse me but I want to tell you that I know the secret I know it all the mailman you know shock drops his back and says son come to Daddy that's the best reason for a multi-party security I could think of thank you well thanks for that thanks for that terrific talk I'd like to terrific talk about betting on many different levels I'd like to open open it up to questions well I think it's fascinating the secure multi-party computation I'm sure many of the audience does as well my question is this though we all know that averages are dominated by the largest salaries so maybe limos salaries dominating the average and in the case of the long-term capital in 1998 you made a point to note that long-term capital was much smaller than the dominant hedge funds so as this is an interesting way to measure systemic risk on an aggregate scale how do we pinpoint those smaller hedge funds which may actually be the risk triggers but may be dominated in your metric by the larger hedge funds that's actually not as hard as you might think the first step in any kind of regulatory action is to first document that there's an issue because if you don't that there's no point to being able to start any kind of an action but once you document there's an issue so let me give you an example one of the things that we show in our paper you can compute securely is the - dollar index for a group of parties if the her friend out index shows that there's extraordinary concentration that's the smoking gun that a regulator can then use to say now it's time for me to have a conversation with the top ten prime brokers and very quickly you will find out from talking to those brokers exactly who has the very biggest exposures but the point is that that conversation doesn't even happen because no regulator is gonna feel like they want to go out and talk to prime brokers on a regular basis and just say tell me tell me all of your secrets let me know what's going on right now you've got to have a reason to go into their offices and what we're trying to do is to create that reason yeah it's Pete Kyle again how do you prevent Chester from manipulating the statistic by lying convince everybody else that finance professors are grossly underpaid very simple you Institute random verification checks so that you know you draw a take all of the names in here draw the name out of an urn and you verify and the person that lies you shoot them that's all you need to do this is what I thought you're gonna say and you see this in the SCC proposals you know for various types of disclosure requires that you have a name address and a token number a place of business and that you have records that you can inspect and that you'd be able to send in some kind of auditor to collect the information that is a fairly burdensome requirement that's true if you have or is any that's true if you have to do this in person but you know the cool thing is we now have computers and we have this thing called the internet and we have servers and even the SEC has servers and the servers can actually tie with the Internet to these various institutions and the institutions can actually you know send messages and the messages can be verified and you can actually check every once in a while that the address is correct the phone numbers working and so on I think that technology is the key to dealing with this in a manageable way by using technology and leveraging it I think there's tremendous things that we can do that we're not doing now mark photo if our encryption technologies are a silver bullet if the only problem you're trying to solve is a privacy problem at the more general case of the issue that Pete just raised is data quality and if data quality issues are pervasive which I think they are how do you reconcile if every day you're you're discovering strange results coming through and the aggregates so garbage in garbage out you're absolutely right and we're not commenting on the fact that this is gonna solve all the problems obviously data quality is a very important issue globalization being able to in you know integrate different kinds of of data fields and ontology z' is a challenge but i think it's less of a challenge when you can guarantee to various different counterparties that their data will be secure and that you don't need all of their data you only need those pieces of data that'll be relevant for systemic risk computations so I think what it does is it lowers the bar or it reduces the threshold by which you can actually do something useful and still be able to demonstrate to various parties that they're not going to be giving up their golden eggs to their competitors alright Pete Axelrod DTCC just some of the horses left a bar and a full disclosure our company is a strategic partner of the managed funds association but we've been publicly taken to task a lot for handing out hedge fund data at any regulator that asks not quite any regulator with a material interest in fact they've all signed something that said we could do that I think the dodd-frank act is going to make all hedge fund swap data in a mirror available to to regulators worldwide and I think I think they are gradually coming around to the fact that regulators are going to see what they're doing regardless of how hard they fight because really regulators concerned a more market manipulation and systemic risk and they they want to see the trade level data right but I think that well the best way to respond is that you know my when my 11 year old son gets asked to clean his room somehow he drags that into like a 2 a 3 day process on the other hand when his mother tells him that the only way he can go get to see a movie is he cleans his room somehow it happens in half an hour it's funny how that works so I think that we're gonna see the industry dragging their feet until and unless we actually show that there's a credible mechanism by which not only can we get their data securely but that they themselves will benefit from the data so in the example of banking transactions you can actually lower the cost to banks for engaging in credit if you show them a more accurate way of measuring did you oh sorry I know you are Joe okay good you said that you could you could verify by essentially a sample of them but all you have is the presumably the best you can get is their information plus the random number so unless you can actually dig back down and get the cumulative sums all the way around and then oh sure sure yeah this uh bridge you're gonna have everything you have everything absolutely what I gave was just a very simple example in the few minutes that I had it was enough to illustrate you know that it could be done but in fact the algorithms that are used in you know for example the RSA technology that all of you rely on when you enter your credit card number on that Amazon website that's quite a bit more sophisticated and so those algorithms are the ones that we use and where you can actually engage in verification as well that's right absolutely well no no you verify you verify you know on a random basis that they're not lying and but the the random basis has to come with extreme penalties so what you've done is to reduce the verification problem from verifying a thousand participants to verifying one every few years ahead of time you've identified what they said with their salary is so you now know for each individual what they're no know where you have their audit trail for what they submitted you don't have to know it at the time you're actually measuring it but at the time that you engage in a verification they display audit trails that are actually stored by the central server so you can actually connect the dots when you need to the point is that you're not connecting the dots every single day to answer your marks question can't you just do it if you use a vector instead of a scatter so instead of you giving me your salary plus a random number give me a million numbers of which one of them's your salary ask everyone to do it in the winning subtract it out then have a set of scalars at the end yeah stead of the sum which case Otto Marx you can do statistics whatever you want still grab a through the secure word yeah sure for example but there are even stronger methods that are robust to collusion so I think you convinced me that the hedge funds that are going to lie are the ones that are in dire straits and a great deal of trouble where fraud is going to be occurring anyway and they don't have any assets to seize so that you can punish them when your policeman comes around to check their integrity that works for me let me thank the the audience of a special especially the two presenters for a very interesting session

Channel: Maryland Smith

Views: 11,936

Rating: 4.8133335 out of 5

Keywords: Andrew Lo, systemic risk

Id: nuDIoBeNwD0

Length: 35min 41sec (2141 seconds)

Published: Fri Oct 14 2011