“When I was in 2nd grade I wrote Portfolio Manager when asked what I wanted to do. I knew quite early what I wanted to do.” – Oliver Steinki (Tweet)
Our next guest has started and invested in many companies, and is from an entrepreneurial background. First and foremost, he is concerned with running a business, but he also knew he wanted to be a portfolio manager from an early age. Hear the fascinating story of Oliver Steinki on this episode, as we dive into his new firm and the details behind his models and strategy.
Thanks for listening and please welcome our guest, Oliver Steinki.
In This Episode, You’ll Learn:
- The entrepreneurial background that Oliver comes from.
- How he studied in Germany, Madrid, and Manchester.
- About his childhood and when he started to realize he was interested in mathematics and finance.
- What he likes doing when he is not managing his own firm.
- What algorithmic trading is all about.
“We are really very open about our strategy and what happens.” – Oliver Steinki (Tweet)
- How the scientific trading model generates signals.
- What the trade implementation phase entails and how it works.
“I put a lot of emphasis in my teaching on the common mistakes people make when they backtest strategies.” – Oliver Steinki (Tweet)
- What it is that Oliver is trying to deliver to his investors.
- About the other companies he started before his current firm, and all the different industries those companies are in.
“I’ve co-founded or invested in almost 10 companies, all in Europe.” – Oliver Steinki (Tweet)
- Why he is a business man first and a trader second.
- Who is on his team and the roles they fill.
- Why the early phase of his business is important and what you can learn from how Oliver tackles this phase.
“We only trade very liquid futures on the main asset classes.” – Oliver Steinki (Tweet)
- How he would scale his current business model.
- How he creates a culture and keeps a partnership with his cofounders.
- His track record so far.
- The details of how his strategy is created.
- What a levy-process model is.
- How many models he is testing and trading and how different they are.
- Why they need to be right 54% of the time with their models.
“In finance your product is very transparent and comparable.” – Oliver Steinki (Tweet)
Resources & Links Mentioned in this Episode:
- Learn more about Evolutiq from the presentation slides that Oliver mentions.
- Read about the French mathematician that Oliver discusses, Paul Lévy.
- Niels mentions his earlier conversations with Dave Sanderson, which you can listen to here and here.
This episode was sponsored by Eurex Exchange:
Connect with Evolutiq:
Visit the Website: www.Evolutiq.com
Call Evolutiq: +41 55 410 7373
E-Mail Evolutiq: firstname.lastname@example.org
Follow Oliver Steinki on Linkedin
“A very weak model can be turned into a very good one because if you have something that is only 30% right you just trade the inverse.” – Oliver Steinki (Tweet)
Oliver: You’ll see a lot of traders who come out of big firms that are definitely great traders, but they don’t know how to do a cash flow statement on a company level, and they might not be able to estimate the real cost of wanting business. Plus to get the revenues right it’s pretty easy in the alternative investment strategy. You have your average management fee and your average performance fee, and you need to make an estimate of what is the performance you can realistically achieve, and then you come up with the revenue. That’s pretty easy. But what’s pretty tough is the correct costs estimation in my opinion.
Niels: Picture this for a second: a small group of research savvy PhDs who have spent a large portion of their adult lives studying and developing complex financial models with the intention of being able to make accurate forecasts of where to buy and sell financial instruments, decides to venture into the world of entrepreneurship. You would expect the model to outperform a simple rule based strategy like a classical trend following approach.
As amazing as technology and sophisticated systems are, they don’t always end up performing better than their simpler cousins. But perhaps they perform differently, and that may be just what you as an investor is looking for. If algorithmic trading seems overwhelming to you, today’s guest not only practices it, he teaches it as well to aspiring students. So take a deep breath and listen to today’s episode.
Introduction: Imagine spending an hour with the world's greatest traders. Imagine learning from their experiences their successes and their failures, imagine no more. Welcome to Top Traders Unplugged. The place where you can learn from the best hedge fund managers in the world, so you can take your manager due diligence, or investment career to the next level. Here's your host, veteran hedge fund manager, Niels Kaastrup-Larsen.
Welcome to Top Traders Unplugged, where my goal is to give you the clarity, confidence and courage you need to invest like or invest with one of the top traders in the world. It is the stories that you never get to hear set out in the most honest and transparent account that I can make of what goes on inside the minds of some of the best investors in the world. Today you're listening to episode 81. If this is your first episode you've heard, you might want to go back and listen to all the earlier conversations. Before we find out who’s on today’s show, I want to mention that today’s podcast episode is brought to you by the Eurex Exchange, which, of course, is the home of the European Yield Curve.
Oliver: Hi, this is Oliver Steinki, CEO and Founder of Evolutiq and you’re listening to Top Traders Unplugged.
Niels: Thanks for doing that, Oliver. And by the way, if you want to read the full transcript of today’s episode, just visit the TOPTRADERSUNPLUGGED.COM website and sign up by clicking “I’m In” on the button in the right-hand top corner. It’s really that simple. Now let’s get started with part one of my conversation. I hope you will enjoy it.
Oliver, thank you so much for being with us today, I really appreciate your time.
Oliver: Thank you, Niels, for having me.
Niels: You’re very welcome. Now our conversation today will be a bit different, but in an exciting way I’m sure, since today we’re not only going to learn about your journey as an entrepreneur and your entry into the systematic trading world, but we’re also going to make use of your experience as a teacher, an educator, when it comes to algorithmic trading in general. So I’m sure we’ll have a lot of fun, and we’ll all learn a great deal. Before we jump in, I just want to ask you a simple question that I try to ask all my guests in order to appreciate the many different answers there is to this question and it’s basically, how do you respond when a person whom you haven’t met before asks you what you do? How do you explain that Oliver?
Oliver: Usually I say I’m running a quantitative trading firm based in Switzerland, and the trading idea is based on my doctoral research, which I did in financial mathematics. That’s usually if you just want to put it in one line, that’s what I would say.
Niels: Sure, absolutely. But anyways, we’re going to stay with you for a little while longer, so tell me your story. How did you get to where you are today? Perhaps in order to put some extra color on, tell me a little bit about what you were like as a kid, or a young man growing up.
Oliver: Yeah, so I come from an entrepreneurial background. My parents, my grandparents, they have all been entrepreneurs. So for me it was pretty clear from the beginning that I would like to start my own company at one point in time in the future. So after school I studied finance, I used to work with Commerzbank and Morgan Stanley, and a hedge fund in Geneva – Stigma partners, and I studied a bit internationally, so I did my Bachelor’s degree in Germany, then my Master’s degree in Madrid, and my Doctoral degree in Manchester. So I like to travel a bit. Now I’m living close to Zurich, where we run Evolutiq out of Freienbach; it’s a small village roughly half an hour south of Zurich.
Niels: Where did you actually grow up? That was obviously a very, very quick recap of your life, but where did you, in fact, grow up?
Oliver: I grew up close to Cologne in Germany, like roughly half an hour north of Cologne in a small village close to Düsseldorf.
Niels: Sure. What were your interests back then? What did you do when you were hanging out with your friends?
Oliver: I think pretty much the normal stuff. I played football, although without success. I was always interested in mathematics, so I did a bit of mathematics next to school, and as a kid I would say nothing extraordinary.
Niels: Sure, now there is a bit of a gap between mathematics and finance. Is there a time that you can think back on when you started to realize that maybe it’s not just mathematics I’m interested in but applying it in the financial world? Or maybe you didn’t think about applying it in the financial world straight away, but the finance part, how did that sneak into your life?
Oliver: Actually I was interested in finance as well from an early age. One of the best friends of my parents is a portfolio manager, so I always found it very interesting to listen to his conversations why he would buy a certain stock or he would see the markets in general. So I had an interest in managing money quite early I would say.
Niels: You know, often when we’re kids, we have some kind of childhood dream. Some people want to be a fireman, a policeman, whatever. I don’t know if you can think back to that time but did you know back then what you wanted to be if not a football player?
Oliver: Actually people make fun of me because when I was in second grade I wrote portfolio manager in a friend’s book we have in Germany. My friends are still making fun of me as I knew it quite early what I would like to do. I never wanted to become a fireman, because I’m probably not good at it.
Niels: I have to say that just the fact that you know the world portfolio manager in the second grade. I’m impressed already Oliver. It’s going to be good today. Now, just rounding off your background and for people to get a chance to know you a little bit. As a kid, were you sort of the curious type that needed to know answers to everything, or how did the interest in these topics, how did that play out?
Oliver: Yeah, I’m a pretty curious guy, so I remember that I was reading even under my bed, under the duvet when I wasn’t allows to read anymore officially. So yeah, I would say I’m quite curious.
Niels: Fantastic, great stuff. Anyway, we fast forward, you’re now the CEO of a boutique firm if I can put it like this, and that plays a big part of your life, but what do you do when you’re not working? What do you like spending your time doing?
Oliver: Actually I really like cycling and skiing and with a couple of friends we will be cycling from Geneva to Monaco in July. So right now I’m just back from a little road trip on my bike because I have to get in a bit better shape for July. Niels: Sure, fantastic, excellent. Now you’ve worked a few places and feel free to go back and talk a little bit more about the various places. I wanted to ask you, when you’ve been working and when you’ve been educating yourself, which has obviously been a bit part of your life, where do you feel you’ve learned the most that has prepared you for the journey you started with your own business?
Oliver: It’s a difficult question because I grew up in an entrepreneurial family, so I knew from the beginning how it is to run a company.
Niels: May I ask what your parents did?
Oliver: They trade herbs.
Oliver: A different area, but it was quite interesting, and it was more like my curiosity for financial mathematics, which led me first to do the Masters, and then the Ph.D. afterward. Because I did my Ph.D. next to work, for me it was very nice actually really to apply the research to live trading, and also to research a field that is very practical. So if it would have been a Ph.D. in Theoretical Physics, probably I wouldn’t be able to do it.
Niels: Did any of the places that you worked at that you would say have a big influence? You worked both at the hedge fund, but you also worked at a bank. That’s obviously big contrast between those two places.
Oliver: MSCI was quite good because we worked in research, and we were 70 people all of them PhDs and it was a very international firm. I spent some time in London, some time in Switzerland, some time in Hong Kong, so it was very international. Then at the hedge fund I was basically the right hand of a guy who used to run the prop desk at Solomon Brothers in the 90s. So from him I could really improve my trading skills, but it was a smaller place. It was less than 20 people, so it was not as international as Morgan Stanley, of course.
Niels: Sure, absolutely. Anyways, before we jump into the next topic, I want to ask you a much broader question. Now, algorithmic trading can be said to be a discipline at the intersection of finance, computer science, and mathematics. You happen, not only to practice it, but also teach it. So let’s for a moment imagine that I’m a student attending your class, and take us into the classroom and teach us the general overview, perhaps in a condensed version of this unique trading style. Hopefully do this in a sort of down to earth way so that the broad range of listeners that we have can appreciate what algorithmic trading is all about.
Oliver: I think, in general, algorithmic trading describes a process of using algorithms to generate and execute orders in financial markets. We have three different areas or applications of algorithmic trading. The first one is algorithmic execution, so you can use algorithms to search some dark liquidity pools, to optimize execution costs. You have all these things like Iceberg orders and this kind of stuff. On the other hand you have market making, which is also another application of algorithmic trading, which is basically supplying the market with ask quotes for financial securities, and then the last one is trade signal generation that is what we focus on at Evolutiq. This is basically designing proprietary strategies to generate profits by betting on market directions, and hopefully you’re more often right than wrong.
Niels: Sure, absolutely. Tell us about that branch of the algorithmic trading world that you focus on. Tell us about that from a professorial point of view when you teach this. How do you explain it to your students?
Oliver: So we came up with what we call the algorithmic trading framework, which basically consists of three different steps. The first one is signal generation, where you decide when and how to trade. The second step is trade implementation, where you size and execute your orders including the exit orders, and then the last step is performance analysis where you calculate different return risk and risk-return ratios to measure the success of this strategy. I introduce this framework for students already in the first session, and then usually we go step by step. We have a couple of sessions on just the signal generation, what are the different approaches people use in the financial world to trigger trades.
We have the second step that is a trade implementation, so you came up with your signal… OK, you want to go long in the Euro/dollar. How do you size your position relative to your portfolio of 1 million dollars? What should be your position size? How do you define the stop loss distance? How do you define your exit level? So you have a collection of very interesting question here also, how do you take into account cross correlation with other portfolio holdings and potential portfolio constraints, because you might only be allowed to invest X% of the total portfolio in a certain industry or sector or asset class.
Then the last step is performance analysis. After you have done your trades, you have to come up with a smart way to judge if the strategies are actually successful. What I see in finance is still that a lot of people are looking at only returns first, so then volatility is coming a bit more in vogue nowadays. But people still use sharp ratio, which honestly I cannot really understand because it punishes upside deviation, and I guess no one would be unhappy if suddenly you have a very good day. That’s why I believe that the Sortino ratio is a much better basis for judgment to decide if a trading strategy is actually good or not.
Niels: Sure, no absolutely. Since we have time, and we have no constraints, I would love for you to take us deeper into those three main areas. Because I think it’s so important for people to understand these things because it’s part of the demystification, if I can put it that way, of what we do. The black box isn’t black though many people believe that it is. So please continue and take us into the first part of it, namely the signal generation, and how you, from and teacher’s point of view, would go about explaining that?
Oliver: So we have a couple of… All the lectures I’m talking about are available on our website Evolutiq.com where you can download all these slides and then it might be a bit easier to follow, but in the first step about signal generation, I introduce a couple of mathematical tools and attributes of scientific trading. For example mathematical tools, mark of models, cointegration, you have to look for where there’s non-stationary: Is it a mean reverting process? If so, how do you check for that? How can you do boot-strapping signal processing tools? How do you estimate return distributions properly? How can levy processes help in doing so? What are the things that you need to look out for when you do time series modeling? How can ensemble methods help you to improve the prediction accuracy of your model?
So we look into all these different mathematical tools, and we also create a framework where we make it clear what the attributes are of scientific trading models. For example, it has to be best and logical arguments. You have to be able to specify all of the assumptions underlying your strategy. You have to be able to quantify these assumptions. The model properties can then be deduced from the assumptions you have defined, and you can also backtest your strategy based on these quantified assumptions. So when you have this framework, you can then go through iterative strategy improvement by just changing the specifications of your models and just see what works well and what doesn’t work well.
Niels: Now that’s obviously the professor talking to a group of students who are really trying to get into every nitty-gritty detail of these things. You mentioned quite a lot of technical terms if I can put it that way. These terms for many people, and certainly for many on the investor side of the table may not be something that they were taught or familiar with. How do you deduce all of those technical terms into something that, if we now change the audience to the investors perhaps? How do you explain those things in perhaps a simpler way without losing any of the value of it? Because I think that’s a big problem, not just for us as an industry, but for many people we struggle with demystifying algorithmic trading, so do you have a sense of how we could do that?
Oliver: No, we tried in our Evolutiq presentation to really… one of our core values is transparency. So we are really very open about our strategy, what happens and what steps we define, what’s the data input, what comes after that, the analysis and prediction phase where we create levy process based market predictions, and how we generate the trades. So basically we apply this general framework from my teaching world to real life trading strategy.
Niels: Sure. Now you talked about the initial signal generation. So signal generation, and a lot of people, I think are of the impression at least that the buy and the sell signal in an algorithmic trading strategy or let’s call it a systematic trading strategy, is very, very important. I think if you ask a lot of the practitioners, they would probably say, “Well, the portfolio construction, for example, and other things that go into it are equally or maybe even more important.” Is that part of the next steps you take in your teaching? Or where do you take it from there, once you’ve done the signal generation part?
Oliver: Yeah, that would be step number two, once we come up with the trade signal. Basically, all systematic strategies have some form of the underlying model which based on whatever can be a very simple moving average. It would say, “OK, we go long or short underlying A,” but then you have the trade implementation phase where. I agree with you, it’s very difficult and, in my opinion, as important as trade signal generation, is the trade implementation. How much money do you actually bet on this trade signal? Where do you put your stop loss? Where do you exit the trade? What I find extremely challenging is the correct correlation estimation. So these kinds of question, in my opinion, every portfolio manager is asking of himself, and it just varies to the degree of sophistication, how to deal with these problems - it varies among portfolio managers.
Niels: Sure, no, absolutely. Of course the third step you mentioned, if I called that correctly was how you do you test your idea or your hypothesis? What do you teach your students to be aware of when it comes to testing their own rules and ideas?
Oliver: So we put a lot of emphasis in my teaching on common mistakes people make when they backtest strategies. A lot of them, they don’t take into account transaction costs, bid/ask spreads, regime shifts. There are so many small things that might not seem that important in the beginning but which are very important in my opinion. We have to really look at all of them, also in a backtesting context. For example, some people they backtest strategies with stop losses, but they don’t have intra-day data. So they can only check based on the open high/low/close if there might have been executed the stop loss or the take profit on that day, if you only have open/high/low/close you don’t know if the high has been achieved before the low or the other way around.
We spend the second part of the course; we do a lot of research into which backtest errors should you really avoid. It’s a bit of a fortification approach. It’s not like it has to be like that. It’s more like, OK don’t do the big mistakes.
Niels: This is a topic for many, many people who want to get involved in the systemic trading world. Obviously they have to do the work themselves first. The second challenge that they face is that when they go out and talk to investors with only maybe a backtest in their hand to get the initial funding, that is a tall order to say the least. What’s the biggest, just in your point of view, what’s the biggest challenge in order to convince someone that the validity of the backtest you’re presenting, actually that it has real value and that they don’t have to wait three years to consider perhaps making an investment?
Oliver: I think here it’s very important to outline in detail what are all the underlying assumptions in your backtest, and how have you done this backtest to come up with the results, presenting to the investors.
Niels: Oliver, thanks for that overview. I think that was very helpful, and I certainly would encourage people to go and check out your website and the full course. But anyway, let’s go back to the normal way that I conduct these conversations. We’re talking about today for the purpose of our conversation of what you call the Pred-X Multi-Asset Class Model. Just tell me from a 38,000-foot point of view what it is you’re trying to deliver to the investors. Then we’ll get into more of the details.
Oliver: So the objective of our Pred-X Model is pretty basic: absolute return, systematic direction, multi-asset class strategy, where we have a target Sortino ratio of 1.5 and a target return of 10% per year, which implies annualized volatility of roughly 7%.
Niels: Excellent, now the first thing I want to talk to you about is your organization. You are a relatively new firm, I know. There was actually one thing, now that I think about it, maybe I do want to jump back a little bit and ask a little more about your background because this is not the first company you started. I want to, before we go into the organization, I want to ask you about if you want to share, what else you’ve done in the entrepreneurial world before getting to set up Evolutiq?
Oliver: Yes, so I’ve co-founded or invested in almost ten companies in Germany, Switzerland, and the UK and totally unrelated industries. For example one of them does backpacks, another company they sell trollies, another company is an online beacon hub like a Bluetooth center. We believe in the future basically Google Adwords will come to real life. So if you go to a fashion store and you stand for a long time in front of certain jeans, let’s say, when you have your Bluetooth on and you open another app probably the app will tell you hey these jeans are now on sale at 10% lower. It’s basically cookies in real life. Another company we try to revolutionize the parking market. Basically it’s an app where you don’t have to deal any more with all the hassles of going into a big garage, wait in line to get the ticket, wait in line to pay, so streamline all this. I have a couple of other different businesses, investments, and my experience in co-founding them and taking care of the financial side just allow me to estimate how to run a business.
You will see a lot of traders that come out of big firms that are definitely great traders, but they don’t know how to do a cash flow statement on a company level and they might not be able to estimate the real cost of running a business, hiring people, getting an office, travel costs, regulation costs, so there're a lot of costs involved in running a business and you need to manage a certain amount of money because to get the revenues right it’s pretty easy in the alternative investment strategy. You have your average management fee and your average performance fee. You need to make an estimate of what is the performance you can realistically achieve, and then you come up with your revenues and that’s pretty easy. But what’s pretty tough is the correct cost estimation in my opinion.
Niels: Absolutely. I completely agree, Oliver, and I’m glad that we just touched on that. That is a big thing, and it’s very rare to meet someone who has started or has been part of starting several businesses, and then you start your own CTA if I can call it that. Maybe that’s also why, when I looked at your information, it looks to me that you’ve actually got quite a sizable infrastructure for the size of the firm or the AUM at this time at least. Bring us up to date where you are in terms of AUM and talk about your organization, how it all fits in, why they’re important and how it all came about. Maybe you were all football friends back then or maybe not. I see some commonality in terms of schools and other things so maybe you could bring us a little bit closer to that.
Oliver: So I started Evolutiq in 2013 and together with Peter Miko, my partner, and we used to work before together for roughly three years at Stigma Partners in Geneva. So he is the hardcore quant, and I’m the medium quant I would say. So I think we have a strong team because we only launched the multi-asset class strategy in March, but we are already four Ph.D. guys and one former McKenzie guy, so next to me it’s Peter Miko, he did his Ph.D. where he analyzed the use of artificial intelligence in decision making and automatic classification of high dimensional complex spaces, so something very fancy. I’m always joking that he can code faster than I write an email to my family. So he’s a very smart guy. He takes care of the research and the IT operations.
Then we have also Dr. Agnes Antal working with us. Like Peter, she did her Ph.D. at EPFL in Lausanne. So they’re both Hungarian, so they are old friends even. Next to Aggy, we also have Dr. Francesco Comandrè. He is also a Ph.D. from EPFL and basically he’s in contact with Peter and Aggy for several months before he started with us earlier this years. Then we also have Ziad Mohammad who used to be my student in Madrid, actually. So he was one of the best students in my class, and he used to work for McKenzie before, but he wanted to get more into this entrepreneurial alternative systematic trading world. So I’m very glad to have all four of them on board. We have quite complimentary skill sets I would say.
Niels: Sure, a lot of brain power, that’s for sure. Now in terms of the function that you do in-house and functions that you may outsource, how does that look from sort of your point of view?
Oliver: We don’t do any outsourcing. We trade via VIX, so VIX means that basically our server talks directly to the server from the broker, so all the orders are implemented automatically. Then on our side we test them and Francesco, he supports Peter on the trading infrastructure side. Then Aggy, Peter, and myself, we also focus on the improvement of the general strategy and then Ziad and myself, we also focus on the sales side, so meeting potential investors and these things. As we just started in March with five million, for me it was important, first to create a top class infrastructure and then to raise the assets.
I think in finance you cannot blow up, so you have only one chance, and you have to get it right. So it was important for me to finance a bit longer the process of getting the right infrastructure set up. Making money comes later because first we want to have a very good infrastructure. I think we have achieved that, and now it’s about building a track record, getting to know potential investors. I think we have a few years peace, which differentiate us from other CTAs. I think most CTAs that are relatively small in terms a year, like us, they probably have smaller teams.
What is also unique is that we focus on Sortino and not return. I think an awful lot of CTAs have this explicit focus on Sortino. Then we use ensembles of Levy processes.
Niels: We’re going to get into that for sure, definitely.
Oliver: We’ll explain that a little bit later, but basically we are not the 50th CTA, who does long/short equity based on S&P 500. Then we also have business and entrepreneurial backgrounds, also, Peter, he was the one who has his own company already a couple of years back, so that is also a part of it that we have this entrepreneurial experience already.
Niels: Now I want to just ask you just a couple of questions about this early phase, because I think there are a lot of interested people listening to us today who maybe are thinking about starting their own investment management firm, or maybe another business. Certainly there are also, as we know, is it 80% or more of the assets managed today, I should say actually, maybe it’s 90% of the assets managed today are managed only by a few firms in our industry. That means that a lot of smaller firms are really competing for not so much of the assets. So I would say that thinking about the challenges of being small is quite important. I wanted to ask you, have you thought about, or in your own mind, how does the growth phase look like for a firm like yourself and when do you need to scale the business so to speak?
Oliver: I think now what we need to achieve is a good six to twelve months track record, or maybe longer. Of course, we have these early stage investors. So for example we have one institutional investor, he basically invested based on the academic merits, which he sees in our approach, but most other investors, they say, OK it all sounds very fancy what you are explaining but, to be honest, I don’t really fully understand the mathematics behind it so show me some results, what is the track record of this strategy? So I believe that it will be at least another half a year to a year until we can really raise assets significantly. Of course, now we are started to dialog with investors because especially with institutional investors it can take half a year from first contact until they would actually invest. They want to follow. They want to be on your monthly return distribution list. So they want to see a bit how do you behave in different market environments and so on, which is totally fair. I would do the same if I were on their side. I think the massive scaling will come probably the end of the year, early next year, but we are ready now. Because of the ultimate transition so far we have only one managed account. For our service, it doesn’t matter. I think right now we have something like twelve or thirteen different accounts we are managing. If we would have a hundred, it’s not a lot more effort for the computer. That’s the beauty of the scalability of such a business model.
Niels: Absolutely. Just out of curiosity, based on your initial research and looking into the strategy as it stands now, and we’ll talk about which markets you trade and all of that later on, but what do you think this strategy, in terms of AUM, what’s the capacity as it stands right now?
Oliver: OK, so we did some analysis and even if we would manage 750 million, we would trade less than 1% of the daily liquidity in any of the analyzed markets. As we will mention later, we only trade very liquid futures on the main asset classes. So in terms of equity indices it’s like the German DAX, Hang Seng Index, NIKKEI, so really this kind of very liquid futures markets where there’s quite a bit of capacity for firms like us.
Niels: Final question on the organization, Oliver. I’m just curious about these things, have you got a vision as to what kind of culture that you’re firm is driving towards? How do you keep a good partnership between the people that have been your co-founders because these are not necessarily easy things in the long run? It requires work, just like a marriage, and so have you thought about these things? Or are you just sort of happy where you are now and not thinking too far ahead?
Oliver: No, no, I think we have… what make Peter and myself start this company is that we have very similar values. We really believe in an academic approach, excellence in research, focus on producing unique investment strategies and innovation, so we believe that in order to run a successful long-term systematic asset manager you need to stay on top of your game and that’s why we are both involved in the academic world in order to really stay on the edge in terms of research in order to stay innovative and come up with attractive investment strategies. In finance your product, it’s very transparent and comparable. If you are in fashion, it’s very difficult to compare short A with Short B. In finance it’s very easy to compare the Sortino ratio of a hundred different CTAs.
Niels: Sure, yeah, no absolutely. Let’s jump to the next area that is about your track record. It might seem a bit odd when it’s only three months old, but let’s give it a go anyway. First of all, just really out of curiosity, March, April, May of 2015 has been somewhat different. March, at least in the trend following the space, March was the continuation of a great start to the year. Many people made a lot of money. April comes along, and a lot of trends changed at the same time and there was quite a bit of a give-back by many managers. May, as far as I can tell, and you and I are talking at the end of May before any numbers have really been released, May seems to me, my hunch is it’s going to be a bit of a mixed bag. So just out of curiosity, how did you do in March, April, and May?
Oliver: Unfortunately we are down roughly 1.5% since March, but this is in line with our expectations, so it’s in line with the backtest. Of course, we would have preferred to start positive right away, but it’s really in line with what the backtest has shown. It’s very difficult to… A lot of investors, they ask you, are you a trend following guy of more mean reversion guy? We don’t really fit in one of these baskets. I would say we are a bit more trend following then mean reverting, but it’s difficult for me to really say, “OK we are a clear cut trend follower,” because we base our predictions based on basically the return distribution forecast of Levy processes that can be, on some underlying… sometimes it can be five days in a row long, so typical of trend following. On other days, we can be long/short, long/short. We change the direction quite often. So it’s difficult for me to classify us in one strategy. Performance so far is not great, but it’s also not alarming for us because it’s in line with what we have seen in our backtest.
Niels: Sure, sure, absolutely. Not at all. It’s a funny period because there were sort of three different environments, if I can characterize it like that. Did you behave like the industry? Meaning March was strong, April was weak and May was mixed or did you even?
Oliver: No, no, we didn’t have a… April we were down minimally. No, sorry, in March we adjusted and lost a little bit. Then we had a not so good April. I think that’s more in line, and this month we also had a small loss so far.
Niels: OK. It’s a short period of time as you correctly say. Now, in terms of the environment, you already mentioned that you’re not really a trend follower; you’re not really something else, and so sometimes when you look at a manager, if you’re an investor at least, it’s about getting a feel for, “In what environment can I expect the manager to deliver a certain result?” My initial understanding of what you’re saying is that it’s going to be difficult. Is there any way to describe a certain kind of environment where you know you’re going to do well, and another environment where you know you’re going to do badly? Is that possible at all?
Oliver: No, we don’t have that very clear cut and it’s also because we trade so many different asset classes. So let’s say you were to have now… Currently, we have a situation with a potential GREGXIT because we trade VIX, we trade Euro/dollar, and we trade rates. It’s like you might lose on the equities, but you might win on volatility, for example. Then you might also win on rates, or lose. So for me, I would be, I don’t think I would be serious if I would say we would always perform good in this kind of environment and bad in this kind of environment.
Niels: Sure, absolutely. Now you’re coming into the CTA world at a time where it’s becoming perhaps a little bit easier to have conversations about the strategy, after a couple of years where many investors didn’t really want to consider CTAs. Of course, the period prior to 2014 was probably one of the most difficult periods that CTAs ever had. Now has this data, if I can describe it like that, has this data helped you, do you think, knowing what happened in ’11, ’12, ’13 in designing your models and maybe even ’14, has this been a help in designing your systems or is it not really that relevant as such in the process that you’re applying?
Oliver: No, we didn’t really look at other strategies that much before. Of course now we have a presentation where we calculate the correlation to the main CTA indices and how we can improve a potential global equity portfolio by adding our strategy, and this kind of stuff. I think even the CTA industry, in the smallest subset of the alternative investment area, it’s so diverse.
For me it’s always very difficult to answer the question what is the appropriate benchmark, because we have such a unique academic approach, if we would do as well as, let’s say, some strategy based on Bollinger Bands, then we can compare as to other guys using Bollinger Bands. But we use ensembles of Levy processes that, to my knowledge, aren’t used by anyone else. So for us it’s very difficult to compare or look at other CTAs and what they’re doing because we’re doing something different.
Niels: Let’s move on to that. Let’s move on to the heart of our conversation which is the program itself. I will confess as the first one, that I’m not familiar with a lot of the intellectual methods that you’re referring to here. So for me it will be, certainly a learning experience. Obviously quantitative investment strategies are kind of driven or there are four key factors. You have your trade frequency, your success ratio, your return distributions, and your leverage ratio. Tell me how, maybe from that point of view, your process of designing your strategy, how do these four factors fit into that? Feel free then to go into the actual process itself.
Oliver: OK, yeah, so what we do is every quantitative investment strategy is driven by four success factors. Trade frequency, success ratio, return distribution - when you’re right or wrong, and your leverage ratio. However, this is more like a performance analysis tool. It’s like when you have a given strategy, for example, trade frequency, all the high-frequency guys, if you trade much more often, it can be OK for you to adjust to a super mini-edge, but you do one thousand trades a day, it might be sufficient.
The success ratio, if you have a strategy that is, let’s say 80% wrong but 20% of the time where you are right you make ten times more money than what you’re losing when you’re wrong, it can also be a very good strategy. That’s why you have to look at the combination of success ratio and the return distribution when you’re right and wrong.
Then the last one, using leverage, for me leverage is a very misunderstood concept because it also depends a lot on the volatility of the underlying. For example, if you compare a strategy on VIX or a strategy on Euro/dollar, Euro/dollar moves something like 0.5% a day roughly, VIX moves something like 3.5% on average each day. So again if you have a one million dollar portfolio and you put 100% of your notional in VIX, with that strategy you’re not leveraged, but it can be much more dangerous, this strategy, compared to four times leverage the Euro/dollar position on the one million portfolio, so your trading a notion of four million but your expected daily move will be much smaller. What we do in our strategy is we use a so-called ensembles of levy process models. Let me get a bit more into detail what we are doing and why we are doing it.
So first, why do we use levy process models?
Niels: What is a levy process model?
Oliver: A levy process models go back to a French mathematician called Paul Levy, and they are just more flexible distributions than the normal distributions. So the financial theory that we know, like Scholes, Markowitz, Sharpe, this kind of theory was invented around the ‘70s, and they rely on three core assumptions. The first one is that financial assets behave according to a normal distribution. However, empirical evidence shows that return the distributions exhibit jumps, execute to the left, have higher peaks, and heavier tails than those with normal distribution and they display skewness and kurtosis. So we all know that financial security returns have fat tails. The second assumption, of classic financial theory which is wrong, is that volatilities are not constant, so they vary over time, especially when you look at the VIX, although it measures perception of volatility and not realized volatility, you see that volatility really moves over time.
Then the third one is that returns and volatilities they are correlated, usually negatively for equities and commodities. So, for example, if markets really turn down massively, volatility usually shoots up. So my PhD was in the analysis of levy process option processing models where, in the option processing world people have been using these more flexible distributions and have been able to improve or overcome these three shortcomings that I’ve just mentioned. So the base of our process is levy process models because they are better able to adjust for these three shortcomings.
What ensemble methods are is basically a fancy word for saying that you combine a lot of models. What you can show mathematically, if you have say ten models, all of them are better than 50%, so better than just a random coin toss and they have uncorrelated errors that you can improve that overall prediction accuracy. So it’s the same mathematics like in Markowitz, the covariance term drops out, and, therefore, the prediction accuracy increases of the overall ensemble.
So what do we combine? We combine a lot of different levy processes because there are a lot of different flexible distributions we can use, and we combine them with so-called ensemble methods. What are ensemble methods and how do they actually work?
So usually you can break them down into three stages. The first one is so-called ensemble generation, where you apply a certain algorithm or mathematical, statistical function to training data to extract the most promising set of models to solve a certain problem. In our context, you calibrate this distribution, this flexible distribution to historical return distributions.
Niels: Can we make… I just want to make sure that everyone is following this. I could be completely wrong here, but is this the same as if a, and I think most people are at least somewhat familiar with trend following systems, is this the same as picking a number of parameters such as different kinds of moving averages, you want to combine to find which ones are performing best based on certain criteria?
Oliver: No, what we do, we have a different sets of models so of a different time period, of a different distributions, let’s say we calibrate based on the last one hundred trading days. Then this more flexible distribution, based on the last one hundred trading days, the most likely outcome if you have seen these patterns in the last one hundred trading days is that tomorrow underlying A goes up. This we do like one million times. So that’s the ensemble generation process.
Then step number two of the ensemble process is so-called ensemble pruning. So what you do is basically kick out the bet models. For us, the bet models are those which are very close to 50%, or very close to random. Because if you have a model that is only 30% right, for us it’s a very good model because you just trade the inverse. So a very weak model can be turned into a very good one.
The next and last of the ensemble processes is called ensemble integration. So let’s say you start with one hundred models out of the pooling step you’re down to twenty. If you have these twenty models, each of them will have a different prediction, so one will say up, the other one down, and so on. It’s how you combine them. The simple version would be just building the average of all predictions, and that’s the prediction. You can get a bit more fancy in it by for example taking into account the error of the Sortino ratio of the different models past, and overweight very strong models and underweight a little bit the not so strong models. So that’s what we do. We don’t adjust and take the simple average; we have some form of weighting based on Sortino as I tried to explain earlier. Sertino is what we’re really aiming for. We over and underweight those models that are left over after the ensemble pruning step.
Niels: Sure, OK, that’s fine, and I’m glad you clarified that. So again, I just want to make double sure that I also understand it. So essentially how many different models do you have that you run these tests on before you start pruning and combining them? What kind of models are they? They must be a little bit different, what you come up with different predictions.
Oliver: We have roughly ten different levy process distributions, and they have differences in the number of parameters you need to use in order to calibrate them. Then we have different, what we call look-back periods. So we go from looking back only ten days, only very recent observations, up to more than one thousand days. Then we also have different return distribution smoothing applications, so some of the returns we smooth them down before feeding them into the levy process. That’s why it goes quite exponentially. You start with ten models, but then you have twenty different time periods you applied, then you have different smoothing periods that you applied. We end up with more than one million different models.
Niels: Interesting, now you mentioned something, and again I’m a little bit out of my depth with these details, but in one of my previous conversations that also is regarding a trading strategy that uses artificial intelligence. It’s KFL out of Canada, one of the things that Dave Sanderson told me was that it was very important for them in their forecast to get to a number of 54% in order to be profitable. In the long run, they needed to be right 54% of the time in their forecasts. Does this ring a bell in your research and is this something even that you have as part of the goal of your design?
Oliver: Yeah, exactly. We found that it’s between 53% and 54% for us because of transaction costs and so on. It, or course, depends on your success ratio when you’re right or wrong. There could be even a strategy, but for us the distribution, when we are right or wrong, is pretty similar. So you could even have a strategy where you’re only, as I mentioned earlier, 20% of the time right, but when you’re right you’re right big time, and when you’re wrong you’re just losing a little bit. But we are more or less, we lose the same money then what we gain, so we target something around 55% prediction accuracy, and that’s the beauty of the ensemble methods by combining a lot of weaker models, this 51%, 52% prediction accuracy. By combining them and because they have uncorrelated errors, you can improve over the prediction accuracy to somewhere around that level you mentioned earlier around 54%, 55%.
Niels: Now, your predictions, how long do they last for? Is that a 24-hour period, or?
Oliver: Yeah, so we always, for every underlying we get daily trading signals. We trade for every underlying we trade at a different point in time. Some of them trade at the same time, but in general we predict the move over the next 24 hours and then, let’s say we’re long right now, tomorrow there comes another prediction for the underlying if it stays long, we stay long, if it goes to neutral we go out of the market. If it goes to short, we turn the position.
Niels: Sure, and how many underlying markets do you run this on today?
Oliver: we started with only five underlyings, and we are extending it now to something like twenty underlyings for your investment universe.
Niels: The process, regardless of what underlying it is, is the same.
Oliver: Yeah, so we always have the same investment process. Step one we have the ensemble generation, so we generate a lot of levy process based market models. Step two is always independent of the underlying, the ensemble pruning, so basically…
Ending: Ready to learn more about the world's top traders? Go to TOPTRADERSUNPLUGGED.COM and sign up to receive the full transcripts of the first 10 episodes of the show and visit the show notes where you can find useful links to other amazing resources. Thanks for listening and we'll see you on the next episode of Top Traders Unplugged.
Become An Insider
Subscribe for free and be the first to receive new and exclusive interviews with the world's top traders. As an insider we'll also send insightful bonus content direct to your inbox.Free Instant Access »
You might also like:
Date posted: 02 Jun 2015no comments