AI & Machine Learning in Finance: AI Applications in the Financial Industry - Panel Discussion

can you hear me it's a stupid question if you can't hear me you can't answer it well come back from the coffee break my name is parvisian I work here at the Swedish House of finance and I will moderate this session we are going to do basically two things first we're going to present to you three case studies studies of two Swedish and one U.S firm that is successfully working with AI in their business and after those 10 minute presentations we will bring the three presenters up on stage together with Brian Kelly for a panel discussion and we want to discuss sort of how will all of this we've heard about Ai and machine learning how might that affect the financial sector will it be a total will nothing ever be the same as before or are these better tools but it will basically be the same industry as it has always been those kinds of questions and I strongly encourage you to take the opportunity to ask whatever questions you may have all priorities goes to questions from you on whatever you have that the panel might answer uh and our three presenters I think they appear on on pictures behind me is Katie Kaminsky Chief research strategies at Alpha Simplex I will introduce you a little bit more thoroughly when you come up on stage Center and East head of eqt digital and so on the Backstrom very responsible co-founder of links Asset Management and you all successfully use AI in your respective businesses and as I said after that Brian Kelly will join us on stage Katie will start followed by from the Amsterdam um and if there are short questions on the on the cases you may ask them right after it's just sort of clarifying questions otherwise we save the the discussion for the panel um and questions we might ask is as I said is this revolutionary revolutionary wheel AI lower costs or will it create new products sort of What kinds of changes what might we expect okay so our first presenter is Katie Kaminski kts Chief research strategist as I said this one portfolio manager ref Alpha simplexgroup she has a PhD from MIT she has written multiple industry white papers co-authored a book on Trend following um she is on the list of top leading women in hedge funds and I can go on and on and on but she's also a great friend who I was here at the Swedish House of Finance she has worked here for many years and taught here speaks fluent Swedish we are so happy to have you back Katie please the floor is yours [Applause] thank you it's so nice to be here to be back in Stockholm haven't been here for quite a while I moved about eight years ago so coming back and seeing some familiar faces really exciting today I'm going to talk just a little bit about how we think about machine learning and some of the experience that someone who's managing money in the industry myself personally I've also been affiliated with the MIT laboratory for financial engineering so I've also done some research in machine learning as well with my thesis advisor Professor Andrew Lowe and so maybe it can give a little bit of perspective of different applications versus the trading perspective versus say other applications where you have to think about where machine learning and AI makes sense so just to start um I want to just move to what Alpha Simplex is so you know a little bit about who we are alpha Simplex was a firm is a firm that was founded by Professor Andrew Lowe who's now actually moved much more into biotech but the whole entire premise of the the strategies we trade is to think about adaptive markets to think about the idea that markets are ever changing and we need to develop systematic approaches so that we can change with the markets and provide returns for investors so our entire approach is really about how do we create systematic models that will actually benefit from markets changing over time in such that we can actually provide interesting returns to institutional and Retail investors so we're based in Boston and we manage roughly 9.4 billion dollars in the Futures markets so one of the core cornerstones of our approach as I said was to apply models systematically that find Opportunities and markets as they change since the beginning we've been applying about 40 percent of our risk to what we call adaptive models so in that bucket it's a range of different machine learning techniques we're trying to determine how to trade the trends that we see across Global markets and this is actually extremely difficult for several reasons if you imagine thinking about a back test for me as somebody who trades the markets on a day-to-day basis my out of sample is today it's not rolling windows it's today and thus it's you have to really think very clearly about what you're willing to actually put at risk going forward so I think I just want to make one more Point here there's a very good paper by Andrew that's Andrew Lowe that's called warning physics Envy may be hazardous to your wealth and in this particular paper he talks about this idea of irreducible uncertainty versus reducible uncertainty and so I think one of the challenges that we can have all the tools in the world but we need to find situations where we're actually able to reduce that uncertainty in a way that actually makes sense on a walk forward basis so in that sense when you're talking about trading you have to think very clearly about what actually can be traded going forward versus other applications so let's give a simple example of demographics right so we also had a research paper we looked at Panic selling so in this we use one of the largest brokerage um brokerage databases out there to examine who Panic felt in financial markets in that case think about it demographics behavior is pretty persistent so if you have more data you can actually really do a good job at explaining this so this paper was very very controversial because it came out and said demographics in the U.S that if you were older married male and you said that you had good exp investment experience you're actually more likely to panic still that was very catchy the news reporters loved it anyways so there's a good example of if you have demographics those don't change you can actually use more data and better forecasts actually determine relationships in the data on the other hand when you're thinking about trading the markets you have to think a little bit more about and I think this goes back to already what Brian was talking about earlier is what are the trade-offs between the two of these so you have an over specified model and sort of a simplified model so in practice for us what we're trying to do let me just explain this graph this would be my Trend signal so if it's up then I would be buying if it's down then I would be selling and this is the data that I can observe to try and fit that particular signal so should I be long or short the US tenure as an example now what we try to do in practice is try to think about explaining this to our clients in the following way classic approaches like we talked about earlier have a linear linear signal strength relative to data that you bring in so you want to use learning techniques to try and learn and sense oops wrong the response curve of markets to your signal so what is your signal as a function of the data that you're bringing in so if we look at this picture right here let's just assume this is the U.S tenure so interest rate contracts what this would show is that if I learned to trade like this when my signal when my data is very positive then I'm actually going to buy more than the linear model it's going to slow down and then I'm going to buy less than the linear model on the downside so these models actually work very well for markets that tend to trade at different frequencies so if you use an example of the US tenure note the US 10-year note tends to accelerate at certain points and decelerate at different frequencies so with learning models what we're trying to do is in some sense create a response curve a non-linear response curve of data to the market so that we can trade the market there's one little trick to this that I think is really important as well in the last two or three years and this is something I've heard from a lot of institutional investors a lot of machine learning techniques have struggled in the last two to three years because there's been a structural shift and this is about that uncertainty so what do I mean by that if this is your signal for trading bonds this year what would you do bonds are down you'd be under allocated to that Trend because you learned as you should have that bonds aren't good that bonds you know are always good they never go down right so let me just try this in this audience for a second you thought rates would go up this year most people how many of you are short bonds the Infante okay good okay so the point is that you can have bias just like we have bias as individuals you can have bias in your machine learning models as well so that's something that we always have to think about is whether or not this response curve is then implicitly creating a bias should the world change so just to kind of give an idea of how this works is you can use different techniques this is an example of a random Forest type model where you use different data you apply those to create filters and that filter creates a response function which then generates a signal so the goal is to use all that data in such a way so that you still have an interpretable forecast but at the same time you're allowing your system to have non-linear features so this is just I wanted to give this as a simple example of how we think about trading the markets so just one last point on that um before I finish I think this year has been a very interesting year for us in the systematic space the biggest trade this year has been short bonds and if you think about it from a historical perspective your entire back test said don't do it but this year it worked that's where you have to really think how do you make sure that those models can change that they're not stuck in the past so I think that's one of the key things we spend a lot of time thinking about we get that question from investors all the time making sure that we actually are willing to remove the bias in our systems so thank you is CEO and one of the founding fathers of links Asset Management hedge fund located in Stockholm erlinks was founded in 1991 then you utilize machine learning techniques Please Santa pure is yours thanks so much sounds sounds good I'll start so just to give you some background on me and and uh and and put me in the context of this conference is that I I worked as a stock broker back in the mid 80s at the stock and Stock Exchange I I did I have always had a bias towards Quant uh strategies I did the Arbitrage between the investment companies and underlying stocks and I did I traded the lottery bonds premium black home there against the the forwards and stuff but but it was you know it was not machine learning it was back in the days where you before Excel basically so and then I I moved to North Bank and Nuria and met my two uh becoming a founding Partners illness Benton and Martin sanquiz and they are engineers so so and they are the smart guys uh that's kind of how we divide things in the company so I I run the company I I I am on the board on the investment committee uh I allocate risk to machine learning models but I'm not Hands-On I mean I I went to the stock on School of economics and uh and we can we learn to to do some maths but not uh in the degree that you may need nowadays so so I I my focus is to hire very smart people uh some of them are are here today and and they do the nitty-gritty stuff so so I will focus here on kind of how to allocate risk to machine learning models how we see it from a business case but I would not be able to give you kind of the technical answers to to what we do okay so uh links then what's about the firm is this yeah there you go okay so uh we are 100 systematic we're pretty much like uh Alpha Simplex so that's a close uh competitor or or fellow in in the market we're highly correlated and do a lot of things that are similar uh we trade long and shorts we have no bias to be long in in markets and we trade only future markets we don't do any single stocks at all so it's purely Futures but in different sectors both fixed income Commodities and and FX and and stock illnesses and we trade roughly 100 different markets uh and and the thing that we compared to traditional funds we allocate risk so we don't allocate Capital at all I mean since Futures you you put in a margin and then you can buy High leverage uh positions so that's the way we do it we don't care that much about the capital uh that we have on the management it's a risk game in terms of the size of the company we are 8 billion U.S so we are one of the probably top five Trend following ctas in the world and if you look at the clients that we have we are we have a heavy till towards Big institutions International institutions not that many Swedish clients actually but more the clients that go more for risk mitigation strategies they put this in in their portfolios to to having in difficult Market environments stressed markets uh crisis Alpha and kts written a good book on on the kind of the case for this kind of investments in the portfolio and we have a team of 80 people in Stockholm we have a small New York office as well but most of the people are in Stockholm I would say 30 each people are researchers and then we have uh 30 people working with system development and and trading so so it's heavy uh tilted towards research and and building the models that we use so going to uh the uh how we Define machine learning and and because that is not uh self-explaining I would say we look at it as when models that learn and improve from from data and also that they adapt and change over time when new data came available and and that's kind of blurry when you compare it with the other models that we have in the portfolio we have totally like between 40 and 50 models in the portfolio where we allocate risk to and I would say a third of them maybe 10 to 15 models or machine learning models and they adapt over time and and then we have a more liner models as well on top of that and also in terms of risk I would say 30 percent of the risk is uh is in machine learning um so how do we use machine learning and why would you use it I would say that if you back when we started we had more traditional models that are more liner and then it was much easier to build models when they weren't that complex but when you try to look at a lot of different markets at the same time in a multivariate way and also adapt over time then machine learning is something that that is really really good so especially the the adaptiveness of these where since we know that markets change over time we know that the ml models helps us to to you know stay in tune with with Market developments and another question would be that are they better than our traditional models do they deliver higher sharp ratio or or do you find the kind of Holy Grail here and I would argue that probably not uh they in our portfolio they have delivered slightly higher sharp ratio and that's good but if you look at the industry as a whole there are a few managers that are you know totally focused on machine learning models in their portfolios and they they don't uh you know over deliver compared to to us for example so it's not kind of an easy thing just to say that machine learning will will improve your your returns uh and that there are a couple of different scenarios where a machine learning struggle and one would be like we had in the covet crisis the pandemic where you get totally new data coming in and markets get stressed and and the the machine learning models they haven't seen this before so then they struggle and and lose money probably more money than than traditional models in that environment and also in very strongly trending markets where when markets are stressed then normally the the simpler the better model basically you don't need fancy stuff then you just need simple Trend following and they tend to perform best in that environment okay so there are a lot of challenges with the with the machine learning and I will not dig into all of this but I would say that the one problem we have in the financial sector is that we have a very low signal to noise ratio uh compared to other Industries when they use machine learning so that's a big challenge also the data is not synchronized so for example you have uh different time zones that you have to take into account when you build these models also if you have macro data that we use in some models that they you need to take into account if it's point in time data if if it is revised over over time and so on so that's a challenge uh also the trading cost implementation cost is something that if you're new in the industry that's a big challenge to really understand how expensive it can be to execute your trades especially if you have a more that adopts over time and change then then you have to really know how to calculate those costs over time but we use machine learning when it comes to to execution we use a lot of machine learning in the execution process when we use the execution all goes but then we already know what to do in the markets and then we don't have any trading costs basically we just try to improve the execution but the big the Big Challenge I would say with machine learning compared to traditional models is is really that as long as you make money and and then everybody is happy and you you trust your models but as soon as you start to lose money and you get into a drawdown then it's if you have a very complex model and really don't really understand why that model gives the signals that you get then that's a challenge and and then that's I would say that's the biggest challenge with with the machine learning for us it's a black box to some extent to some extent you can't find you know reasons for the positions but but the more complex model the the harder I would say and that's why I think it's I wouldn't really like a mall that is totally you know too many parameters and too complicated um but but I really like the presentation but it's a challenging challenging thing to to implement meant okay just a couple of words on our our journey then the machine learning uh we we started one of the managers the founding Partners even as Banks and started playing with with the neural networks back in 1994 we bought the floppy disks from the US the software but you know the computed power and and the date that we had and so on it wasn't really useful back then to to get anything uh that could could be useful for us so it was uh in 2009 when we hired our first ml expert PhD from from Google and he started to build malls for us and since 2011 we have implemented a handful of of models or a dozen uh of different features and and uh and uh and different ways we have built those models partly depending on who has built the models we have a big research team as I said and and over time we have looked at different uh type of of ways to build these models so uh we're really bullish about this I think uh you know a lot of data comes available we also look at the alternative data uh sometimes it's encouraging sometimes you don't find anything but it's it's really nice to play around with it I think it's it's being a big manager and have a lot of resources is nice because this is an investment game you have to you know buy a lot of of computers and service and and uh and pay high salaries to people digging into into these data and try to build new models but I think it's it's really uh a very nice uh future we have in front of us thank you so much [Applause] kiss the head of eqt digital you have a master from the Stockholm School of Economics I noted if I've worked for Google in Sweden and you have seven years almost experience at eqt and long experience of working with technology and we're very happy to have you here great to be here thank you so yeah I I do my best to disguise the fact that I'm actually an SSC Alum by you know I'd work I work work with data scientists and data Engineers so so I I have to wear sneakers it's part of my job here to present mother brain and now Shifting the perspective on on the private Equity markets uh a bit uh for those who I I'm not going to introduce eqt I assume it's well understood what we do we're sizable player currently the third largest private Equity house in in the world so we've grown significantly over the past years the reason for that or you know a big part of that is is our digital enablement so not limited to Mother Brain but also to our investment on on the ekt platform so so far before long before digital transfer information was was the term we were actually investing in and what are the most modern tools and and how can we make sure that our team are are enabled so when we we you know something like the pandemic hit we were quite resilient we were able to actually navigate through that and and uh act on opportunities when when people were maybe a little bit uh less inclined to do so Mother Brain so mother brain is a sizable team it's a startup within eqt and we've treated it as such we insulated it and let it grow uh we you know these guys and uh girls they are not natural immigrants to private Equity right so you have to you actually in fact we lead with the the brand Mother Brain not eqt because that's a repellent whereas Mother Brain if you're a data engineer or data scientist in Europe you've heard of it and you might be attracted to it um and as with any startup we were built on on the most modern tools and and the Technologies uh very and you know even even a young uh company like or or startup like like mother brain has actually gone through a few generational shifts and and thrown out uh technology that is not cutting edge anymore to adopt new platforms um we're quite proud of the the third image here to the right it's it's actually uh we're the only alternative investment firm uh in the world who has been published in one of the most prestigious AI conferences so so there's there's real deep science behind what we're doing the other companies uh typically here are Alibaba Google and those guys but what is it done what do we do so uh so our approach is a very integrated one right so we don't as as many companies who invest in in mother in in machine learning I think are go wrong a little bit this is you know you hire really brilliant people and they put them in the basement and treat them as code monkeys that's not what we do we we have a very integrated approach in in many ways right so so in order to uh have private Equity investment professionals adopt the toll it's very really important to think about things like how do you get it in front of their eyes right so if it's a hard to log in if it's you know lots of of friction in using the Tool uh they're not going to get there but even if you make it really sexy it looks good and and it's easy you know web interface and all that you still have to have to have some form of active active decision to go into it and and the re the way we cracked this was actually tomorrow charge deal management software or you know think CRM so this is a basic tool for any investment Professionals in their day-to-day work everyone goes into the tool on every level right partner or director or associate Dr analyst everyone touches these tools some are more reading data and others are are also writing data but but this is this is how we how we make use of it because we can be super smart and have all the data in the world but if that data is not put in front of the right pair of eyeballs it's not actionable so we thought very much about that um also physically this manifests itself that I'm actually not very differently dressed from most of my investment professional counterparts in ekiti you can see very very you know over six years how that the cultures had merged between digital folks and and the investment professionals so so most of us actually address me I'm slightly overdressed if you compare to to Howard data engineering the team dressed us uh I look like an investment professionals basically but also in our offices we actually we we have the data scientists the data engineers and the full stack developers sitting next to their their counterparts in the investment uh professional uh so so there's no there are no silos whatsoever but it's very if you come up and visit the equity only ex got done you'll see uh how how data scientists and investment professionals working side by side and that is crucial because otherwise these guys will go in and dive into a problem maybe not worth solving maybe interesting for data scientists but not that worth worthwhile for for eqt as a farm um and and similar to or rather I should say a little bit The Challenge in private markets is of course that the data is a lot less structured it's a lot less real time we're dealing with private companies and and in Europe for the most part there's good availability we can get financial data easily in the US it's a very different ball game private companies are like just that private and Asia is is a is a mixed bag you could say but clearly uh you know if there's not one single data source we have to look at many data sources in order to apply machine learning and get something out of it so so clearly people is one one element Founders individuals we use uh our mother brain tools to actually do senior recruitment uh finding boards advisors Etc uh attract certain types of Founders is really interesting for us companies financial metrics that we buy a lot of data and most of that is third-party data but then what's really cool is that we also use a news so so for instance uh anyone in private Equity at any given time will be working on two three Deals uh one or two portfolio companies and well at the same time be responsible for monitoring maybe 80 listed companies or private companies right and there's there's of course that's you know you don't do that you don't go through 80 companies every week and look at something happening rather you want that data to come to you so we we've set up our algorithm so that you know when when a cxo is leaving or there's a there's a transaction going on M A or there's a you know something drastic happening to to the the the the the the the the the share price or or you know other news that we find are interesting we can calibrate those and then this is the trigger for me to go and look at that specific company something happening is this a trigger for for an opportunity in our growth and Ventures business we also look at Social data this can give us an early warning where where Mother Brain will any given day mother will give us about 30 signals here's an interesting startup you you should look at and and it's used this a combination of everything might be download app downloads or or you know social media mentions on on Twitter or uh Google queries there's something happening this there there's a this trend is not linear anymore it's starting to explode then we get an early warning signal go look at this company it could be interesting and that gives us about a six months Advantage so so we can spot our radar is better than the most uh Venture capitals uh VCS radar because we get we get an early warning signal a signal we can Target the company we can build conviction we can establish relationships with the founder team and potentially act on that as an opportunity um and sourcing if the primary use case uh not the only but it's at it's the one big proven uh use case that that has uh paid for this entire team uh you know in Spades so so um uh what you see here for a few examples of the ultimate score that we use uh and something has happened so it's an aggregate number of all the things I mentioned and and many many other signals aggregated in one big ultimate score telling telling you something right there's there's an absolute number to it and then of course there's there's uh the Dynamics behind it something is happening here go look at this company uh it could be an interesting opportunity um this is what it looks like from in in where where Mother Brain was born in our Ventures team uh We've we've so far done 15 algorithmically sourced deals uh where we're an investor one of those uh has uh is an exit and three are deemed unicorns so this is you know astounding really good results so we're super happy about what Mother Brain could do and and you know you'll find we might have done that deal anyway yes maybe but if we didn't have that Head Start six months then you know what are the likelihood that of us actually coming in and winning that deal a lot less right so so um you know we always discuss how much should we attribute to to Mother Brain there's a lot more companies that we could attribute to Mother Brain these are the ones where in inequivocally we wouldn't have done this this would not have happened without mother brains so so these are the ones we put on on slides but then there's a lot where you could say mother brain assisted deals that we've done and then what we're super excited about now and probably more excited now than we were only six months ago is the add-on opportunity and that's twofold right so any given you know the PE game it's rolling up companies we can come in uh in you know on a very early stage before we invest Bankers or or uh consultant time or anything we can build are there add-on opportunities here and boom get a long list of of add-on opportunities that's bread and butter in in private equity and and we can just produce that list very very quickly and and at zero cost um and the other aspect of add-ons are digital add-ons where we actually look to accelerate digital transformation of traditional companies think anti-semics with with digital startups right so we buy digital startups that can in some way technology or go to market or Talent aspects whatever it might be accelerate the digital transformation of that said traditional private Equity investment and make it a lot better and of course given how the markets have developed this is a much more actionable and attractive opportunity now as valuations have come down drastically over the past six months and and we are super excited about that I think I will stop there over to you Pat thank you well uh Brian let's start we heard sorry our well this morning showing how the use of uh AI was growing rapidly or almost explosively you know all kinds of the economy is this technology is it a revolution or is it Evolution I've heard you used those terms before and I like this what a nice way of thinking about it right now it's a good question I'm going to answer this from the from the perspective of an asset manager like aqr where I think it's important to recognize the type of business that a lot of asset managers are in which is uh High AUM relatively low frequency forecasting business right so if you think about that as the problem that we're trying to solve you know the role for machine learning is highly specialized it's a small data problem because you can't have a lot of turnover right turnover dictates The Horizon at which you're forecasting and that dictates the number of observations you have in a Time series setting so when I think about the asset management industry I want to sort of focus on that because I think Market making and hft those are sort of different animals and in this context I think I'd be very careful I mean I sort of take the the other stance which is I like to think about ml as a very natural evolutionary development in the asset management industry so the types of things that I talked about this morning the types of things that we're doing now where we're trying to genuinely use complex models even for these low frequency prediction problems this is Cutting Edge but there was there wants us to get here and it wasn't a short process right so if you go back to sort of the the foundation of the Quant industry we're trying to do what ml is fundamentally trying to do which is use data use technology put them together build informed investment decisions and over time as tools have improved as computation has improved those tools that immediately get put to use right so if it's 20 years ago Quant investing becoming more estimation driven even if it's just using regression models I mean that's a that's a big advancement that was the underpinning for current investment advancements and so I I do think about this as fundamentally an evolutionary shift we need to recognize that we're never going to transition into the types of environments where ml has really been revolutionary because those are all fundamentally large data environments we are fundamentally a small data environment and I think the last point I'd make is that we need to recognize this Evolution Revolution question what is the what is the starting point right when we start using ML and asset management we're not starting from Ground Zero we have a model in place we have a maybe a linear model but it's a good model right you might have a couple dozen or a couple hundred signals you're using them in fairly simple ways linear that has attractive features it's robust it's transparent but you also have to recognize that a linear model is going back to my talk that's the first order approximation the first order approximation is the important one the second order one can be incremental it can have Alpha relative to the first but it's probably not going to be as big as the first third order a little bit less fourth order less so so I think this is important just in terms of building expectations about what machine learning can do in Asset Management incremental evolutionary it's not the pitch you get from I think a lot of people are that are in the business uh but I think that's the reality as as I would present it Sven would you give another answer from a private Equity point of view or yeah I think I think Roy Amara said we tend to overestimate the impact of technology and in the short term and underestimate it in the long term so I think there's there's a there's a Time perspective I I I I do agree but I think in private markets especially Venture Capital that's a millions of startups are coming online every year right and they spew data it's it's just it's very rich it's very dense it's just hard to capture those signals I I think actually one of the biggest problems we've been facing is adoption and and you know investment professionals adopting these tools and doing things differently so so that's that but everything else is sort of coming together now we've got data we've got a low-cost uh technology to to actually analyze that the talent Market is developing so so I think I think there are there are um parts of the market where where it'll be revolutionary and parts of the market will it's more evolutionary real estate data you can do so much right looking combining demographics data looking at what kind kind of of uh people are moving into any given area and and what that will pre you know what will happen to uh rentals uh over you know the next 10 years you can do a lot right so so applying applying machine learning Technologies can have massive impacts on some in some categories and less so in others okay do I see any hands you don't have to ask a question you can have views you can sort of differ with what speaker said from please yes thank you Urban film red I'm from the switch Securities markets Association I mean I think early this morning I had this fear of Hell ninth house and taking over you know from the humans but it seems from I mean what you've all been saying that humans still have a very important role to play here I mean be taking over you know from the autopilot when when the plane is about to crash or you know just checking the models Etc but in the long term I mean where do you think will there still be room for you know the kind of more normal human Trader Etc or will all um all Traders and asset managers be more or less robots will we see Algos computer human beings compete I mean if you look at our trading desk nowadays that we have still have Traders 24 hours a day but they do a very different thing compared to what they did 20 years back right so they they overview and Overlook the process make sure that we are connected to all the exchanges and so on but we have the Algos that that trades but I don't expect uh to have an environment where we don't have any traders in place but they will have you know new things to to do and do things in a different way same thing with a ml all those that we use I mean we the financial markets have always changed over time they will continue to change we will need to adjust the the models and and there will be plenty of things to do for our research team I'm pretty sure anybody else yeah I'd add to that so coming back to this small data point right one fundamental fact just statistical understanding of how we approach models and how humans play a role in those models data and structure our substitutes when you have a lot of data you don't need structure from a human but when you're in a small data environment any structure that you can inject wherever it comes from it can come from logic intuition it could come from a theoretical model any structure you can inject it takes pressure off of the data right you ha I mean I like to think about data as a resource allocation problem right I have a certain number of observations I can learn from to the extent that I can supplement my model with some Theory well then I get to use those observations and spend them elsewhere to learn something else that I don't know right so to the extent this is not about the trader this is about the researcher right to the extent that we can have humans that can understand economic mechanisms and build those into models that can be complemented with learning about things that the humans don't understand via the data that's a great opportunity that's never going to go away in the asset management industry okay yes um so of course all these uh new tools are as useful as as Brian was saying the more data you have the more of an advantage you can take from these tools and so my question is how much investment you think you have put in the Gathering of data and how much do you of that you do internally versus Outsourcing your data Gathering and how much of a strive is there on like proprietary data relative to your investment in the tools who would like to cut Katie I can start um I think the challenge is depending on what your use case is so in our state in our space we're treating Global Futures so it is actually very difficult to find the you know data that will give you a better predictive value than something else whereas it's really exciting as you move farther down the curve there's more and more opportunities to use data so for example as you move even until agricultural Commodities there's more data out there that might be useful for more idiosyncratic market and that actually continues down the spread to private Equity where suddenly there's this data that not everybody has that has some information that you can process so from my perspective it's it's your use case that matters in terms of what data will actually help you and it's measuring things with that data that that is where the advantage comes from these techniques somebody else I'll go I think the way we look at it it's decision intelligence right so so and also touching on the on the question before um it's very much in the private equities Market uh people game still you have to you can't do it uh you will never see or at least I can't Envision a world where robust will be doing the trades we we look at this in two ways right we're making investments in data and technology and and building a tool which is which is Mother Brain but we're also building a capability that goes much further which is which is something different right and I think you can call it decision intelligence you can call it augmented investing if you will what we do is is to to accelerate uh decision making to walk away from this deal because you know you don't need to spend three weeks on this because data upfront will tell you it's not worth it right or wow this looks really promising can we accelerate our conviction here by by applying data science Advanced analytics submission learning yes boom and then you know then you double down and then you you accelerate all those decision and budget approvals what what have you uh in order to get to to uh to the right to spend spend your time wisely that's that's one big part of what we're trying to to achieve and and in that game you can't really Outsource it we we've made the um call that we need to build this uh uh this capability internally right and that there's no uh we haven't found any shortcuts uh so so we buy a lot of data we produce a lot of data and we have a big team okay it's so dark guys oh boy it looks like a nightclub it's hard but I think it was harmful in um Eric okay Hans please yeah so I would like to comment to something I just want to say that really hit home um for me you know so I think uh one of the most difficult situations if you're managing like a large quantitative investment operation is when the models are not doing well right uh you have to explain to the board or to your CEO or whatever like you know this is why things are not working and you said you know a challenge with your ml models is that you can really explain very well and when they get more complex it's even more difficult to explain so from like an organizational behavior point of view I think it's a question I I would go for for Brian first because you're you know like the complexity man here in this meeting but maybe maybe Katie you you also want to comment that I mean I I think that sounds like a very challenging situation and also like objectively how do you judge is this the modern you know not working right now because there's a problem or is it you know fundamentally now mispecified it hasn't adapted to to the world how do you judge that from the situation yeah so uh this is a great question and uh I hear this one a lot so I think the the interesting thing to keep in mind is that we have to have some perspective on what it means to be interpretable or to understand a model right we have fluctuations in asset markets we don't understand where those come from right asset pricing Theory doesn't understand it financial markets participants don't understand it all right we may give some sort of local explanations for why things happen that's very different from understanding what's going on right so suppose I have a trend signal in my model and I lost some money and I can track it back to a trend signal do I understand why I lost money I might be able to explain to an investor that I lost money because this particular Trend signal didn't do what I hoped it would do but I would say that that's an extremely shallow explanation so when I think about the potential loss of interpretability from moving to complex models I feel like we're sacrificing a lot less than we might make it sound right because we have these situations where we have the ability to expand a model and improve our predictions improve our performance on average I might not be able to explain where that's coming from but if I can believe that I've built a model that has reason to be more predictive for example for some of the complexity arguments that I gave and it realizes on average to be more predictive I'm going to lose money sometimes markets are risky right and I'm going to be willing to take those bets even if I don't have a clear interpretability so I mean in terms of I think this is also closely related to The Evolution versus Revolution point where I see really the evolution coming from in the next three five years in the asset management industry so I actually see a softening up in this perspective of what flexible models can do for us right this idea that maybe we didn't have such a great interpretation of what they were doing already and this gain even if it's a relatively smaller incremental gain from having more flexibility in our models maybe it's not costing us us all that much in terms of interpretability Kitty so I would have to say that from our perspective we've seen a shift in the industry people have asked for you to be very transparent we go into Pension funds and we show them our machine learning models and we explain exactly how they work and we try to maintain some structure so they understand we create reports and things that explain our models are speeding up or slowing down and we have to be accountable for them the reason is that we're seeing that you know if you're working with Pension funds who are managing money for teachers and governments you know there is a reason where they need to understand exactly why they made money and lost money and yes we want to use those techniques but we have to at the end of the day we need to be able to explain that and that's why our space is actually bifurcated in that there are people who are much more pure in their approach and some of the Pension funds have really appreciated that in the sense that they want to understand exactly they'd rather know exactly how you do that so that when things don't work they understand why and so I'd say that that's why the evolution is so important because at the end of the day if you're down an extra 10 percent on a pension fund's account um that you know that's a really hard thing to explain so from my experience it's much more about being consistent like the models I showed you there where you can understand versus the trade-off of sort of that little incremental sharp ratio that may or may not be there looking forward yeah I agree with Katie and I think the big challenge is really the time frame of our clients uh in the industry because if you look at Trend following that we work with you need you most of the time you're in a drawdown so now we are lucky we have a fantastic year but normally like 80 percent of the time we already drawdown and people will question you know that's Trend flowing really work anymore or has Market changed or and and the clients their time frame is normally a bit too short we need to educate them all the time and to explain you know why you have to have a time frame of maybe five or seven years and why we don't deliver in in some Market environments so so it's it's a really education game and a Confidence Game both for us when we look at our models but also from the client side you know it's easy to invest in something that has said Jordan historically but then after three months you will call your manager nice yeah I know that you're not making money what's what's happening really so it's it's uh it's tough okay we have six more minutes and I've seen three hands in the air let's do it like this we take those three questions as short as possible and then we give them to the to the panel erected them your first oh thank you very good presentations I just want to kind of bring up another aspect of this and that's Financial stability would you think and maybe this especially to just front and to Catrine that this will kind of increase the risk of herd Behavior so that if you get many of funds like you run that in a potential crash of the markets that that will actually cannot be people even further by by your strategies and hence it could be a concern of of uh for people like me thanks okay is her behavior more likely yes you're next to each other you sort this out okay thanks uh what is your opinion about academic research is it too simple too naive short example we collect data on Twitter and financial data we combine it and we think we have a good paper but if you compare it to Mother Brain then it's not even one percent what you are doing so are we too naive too simple okay my name is Giovanni billios I'm a background my background is in Quant management I've seen the space wobble to put it like mildly it personally 98 2007 2008 and more recently part of that wobbling certainly in 2008 was due to the fact that some of the models being used were the same I would argue with responders with the same um what do you expect to happen when AEI becomes dominant we've seen an effort now in a major swf in the world to implement very large scale programs based on AI we're talking about trillion dollar sized so what do you expect to happen are we prone to the same sub sort of things happening or this time is going to be different okay so herd behavior and over simplistic models or something like it who wants to go first I think I can start with the models please so I think it's a and going to the herding Point as well I think we talk a lot more in the US about simple models and passive investing is actually active investing with simple models and we have more and more of those type of rules in the way that people invest in every aspect of our business so I think the hurting issue and sort of the the issue of potentially having Cascades and prices is a real issue because the way that we trade today is a little different than the way that we traded before and so I think that's something we think about but we trade only in the most highly liquid markets which is very very deep so I I think the biggest joke we have in our industry is reporters will always call us and say did you move the s p and I usually laugh because I say there's no way so in certain markets where you have enough liquidity and they're large enough sort of simple complex models like we're using they're not as detectable as say simple simple rules so I'm actually more worried about simple rules than I am about complex ones okay Brian maybe I can take the uh asking a question so I I mean I think we need to recognize what academic research is which is you know trying to make some progress on difficult questions given the resources that we have at our disposal and academic research has made some extraordinary strides in these questions and then other times it attacks problems from very naive perspectives as well I mean the number of papers that are written that are you know agnostic to real world frictions like trading costs are through the roof right um but that's just one aspect of what what what Finance research is trying to accomplish um so I would kind of try and tie that to some of these earlier comments that were coming off Simona's question and the other question as well about you know what is the role of humans in the in the future research process in asset management and I think doing things like trying to better understand what these more sophisticated models are doing that's again a human role that's not going to go away so in some sense thinking about how do we actually solve problems like interpretability those are those are jobs for humans right um so in any case so yeah I'll build on that so so I I again I think it's about augmented investing right in our industry certainly the private markets uh uh it's it's a there's a lot of art still and that will remain uh you you might be present with a host of opportunities but if you don't have people who are convincing and can put you in a position to actually action take action on those opportunities you there's zero value produced in terms of of uh uh you know the models and the complexity it's quite remarkable uh if you look behind the uh you know if you dive into these algorithms and and what they can do right so so um the the ultimate score I mentioned in my presentation is is uh it's it's uh it's rarely you know super indicative you have to go in and if it's a if it's a B2B company there's a lot less signals right because you know people don't tweet about the great example of whatever you know the software tool they were using or there's a lot less bus around this you have to look at other factors and and it's all relative right so you have to you have to look at it really some some categories are just there's more noise uh and there are more signals because people are more engaged in it and and uh so you have to factor in that uh so so you know only then that's this this produce value but again um for us uh decision intelligence is I think the word I want to lead with because uh what it what it helps us is is to make humans apply humans where humans should be applied and and machine learning where that applies right so so that's um that's how it's some sum it up Monty you get the last word okay so I'll take Eric's question regarding if we are a threat to the system if it's uh in general yes there have been a few uh cases but I I'm not concerned about our small part of the hedge fund industry because what we do is directional so we buy when things go up we sell when things go down what you historically have seen is that it's the funds that work with Arbitrage going long and short making spreads and that type of where you build up a huge leverage positions that's where you can have the problem but as Katie said we trade very liquid markets we will not be your problem I promise [Laughter] I'm I did very much Katie Brian Smith thank you so much and so I leave the word to Nick classier club uh chairman of the board of the Swedish House of finance and also CEO of Hope before please thank you bear and wow what a spectacular day and and well organized day and and many thought-provoking discussions and and presentations during the day and yeah I was amazed when our at events like this when we have a mix of practitioners and academics for many reasons one is how close the distance is between the research conducted at universities and what the practitioners are doing out in the financial industry and how fast research results are implemented into the industry I can't hardly think of any secretary industry where this closeness is is that is so close like in the financial industry and that always amazes me when I when I'm at occasions like this one and another Reflection from the discussions today is as always when you discussed AI big date I'm actually learning how fast the development is and what type of impact it has on Industries like the financial industry but also the society overall and and today I don't believe there are any Financial organization that is not working with a AI machine learning in one way or another as a small example I work at an Institutional investors like April Forum we put quite a lot of efforts in our internal work in our in try to understand how we can use the Am Machine learning to improve our investment decisions and investment process but we spend even more time in understanding to understand how AI machine learning might create new businesses might impact value chains and the companies we invest into and also on the macro level the potential for am actually learning to in to to give productive gains and and actually increasing the economic growth overalls there are a lot of things to think about when we talk about Am Machine learning and has been a great day and a lot of good discussions and Reflections regarding that and of course there's been very many persons involved in organizing a successful day like this one in particular I would like to extend a great thank you to the Swedish secures markets Association for providing funding for without that founding this day wouldn't have been possible then of course we also have some persons that have worked many parts of work to do to organize these states but in fact particular Professor like Alexander youngquist I don't know where you are there you are and and it has been responsible for planning and organizing this day together with his colleagues regardless about turkey and and Angie Shang and and they have done a tremendous work as you can see it's not doing today it's also day tomorrow which has a little bit more academic touch so so a great thank you to Alexander and your colleagues then of course we have had a lot of great speakers today and a great thank you to all of you that have contributed with your thoughts and and research and ideas and and discussions there so and you come from all almost all over the world to to come here to Stockholm and and have presentation discussions a big thank you to all your speakers and and lastly and not the least a big thank you to all of you sitting in in this room and all the good questions the energy and dynamic environment we have in here it's a big thank you to the audience without you it wouldn't have been this successful day either so with that as a closing remark it might be so that we should give the Swedish security Market Association the speakers Alexander his team and not the least all of you and ourselves a big hand to end this day [Applause]

Channel: Swedish House of Finance

Views: 24,967

Keywords: AI implementations, Sven Törnkvist, Kathryn M. Kaminski, EQT, AlphaSimplex, Svante Bergström, Lynx Asset Management)., Finance, Research, AI research, AI in finance, Machine learning, Artificial intelligence, Swedish House of Finance, Finance Sweden, financial research, finance research, financial economics, AI asset management, AI asset pricing, AI investment

Id: AONZoaWC9v4

Length: 65min 33sec (3933 seconds)

Published: Fri Sep 02 2022