Episode 33 - with Moritz Heiden and Moritz Seibert

[Music] Welcome to the StrongSource Commodity Podcast. We are your host, Metambeon, and Alexander Stairk. Welcome at another fantastic episode of the StrongSource Commodity Podcast. And I'm very happy with the two guests that we have today, Moritz Heiden and Moritz Cybert. Both are co-founder of Takahe, capital-acquantative investment manager. And they are also very active in commodities. In my career as a commodity trader, I have seen quite an important development among CTAs. When I started trading in the late 90s, hedge funds were companies that you could visit and that you could speak to. And over time, I've seen CTAs evolve into quantitative hedge funds. And I always had the idea that you could not talk to these people, that they were purely algorithms. And here's the exception. We have here a quantitative investment manager, CTA. And now we can talk to them. And I'm very excited about it. I met both gentlemen once in Amsterdam, which was a very nice experience, very nice guys. And I was a bit humbled because when I went through the resumes, I thought, "Shit, these guys, hey, I'm much smarter than me. I'm a simple economics student. These guys, they have, you know, I saw one of the mortises, the Moris Heiden has a PhD in statistics. And quite often when I come across profiles like that, also people who do, you know, theoretic mathematics, astrophysics, I'm thinking, you know, what the hell, who am I? Can I have a normal conversation with these guys? Let's figure out today whether we can have an insightful conversation about what they do, that we are a commodity podcast. They do many asset classes that's also interesting. And what I also like about them is that they are active on LinkedIn. So they are not hiding somewhere in the bunker. They are sharing what they do. And I find that super interesting. So I think for our audience, this is the perfect, these are the perfect guests, very intelligent, entrepreneurial, traders, active in commodities. So welcome. Thank you, Martijn. Thank you for the kind of introductory words. Your correct clever Moritz has the PhD. I'm like you, I'm the simple economist without the PhD. And I guess it's very easy to confuse us because we shared the same name. So, yeah, Moritz, Sebert, Moritz Heiden. Thank you for the invitation. It's going to be fun, I hope. Thank you. And it was very lovely meeting you in Amsterdam, which is probably a year or so ago now in this. I'm not sure what it was, the popcorn cafe for lunch. And I remember you were surprised that we were in Amsterdam to meet with investors and friends. You thought that I was there for the cocoa conference. And I was, I wasn't even aware that there was a cocoa conference. But obviously you were because that's your background. But, yeah, we, we do trade these markets. I guess we'll get into this in a bit. But we're not really that interested or let me take that back. We're not interested at all from a trading point of view about any fundamentals or opinions or what market participants are saying about supply and demand. We have algorithms, as you say correctly, which respond to price action. And that triggers our trades, which is the key difference to a discretionary commodity or physical commercial commodity trader. Yeah. But it's then, so that's super interesting. And because you're both Moritz, should we say Heiden and Sebert? Or how, how do we, because then we know we're targeting the question too. Maybe, I don't know, Moritz asked or Moritz H or. Well, we'll figure it out. So what if you, you say like, yeah, we have an algorithm that triggers. So does that mean? So for my point of view with AI is that you guys can just go skiing in the mountains and enjoy that algorithm do a thing. And you, uh, you let a trade right whenever it hits. So you don't have to do any human effort anymore. Or you are. And I'd like to explain why that is. First of all, AI, um, I mean, AI is a big thing. It is a big buzzword. It may be, it may become incredibly important in our lifetime. So I think it will be, uh, but I'm really not the expert to opine on that. Uh, it's out of my zip code, out of my circle of confidence to be quite honest, where it's Heiden can speak about this more competently than I, I guess. But we're not currently using AI. I think it is very difficult, very hard for us to have artificial intelligence or genetic algorithms or machine learning techniques applied to financial market data, which is incredibly noisy. 99.X percent of what you see happening on a day to day, second by second, millisecond by millisecond, microsecond by microsecond basis in the financial market is, is noise and randomness. And the signal to noise ratio of the thing is extremely poor. So there is a very high risk that when you run these algorithms, that they are teaching them to learn noise and inconsistencies and silliness. Now, that does not mean that AI cannot be powerful in contrast. I think it is very powerful. You know, we're using chat GPT and where it's using cloud and we're using all these things. I personally use it to sometimes enhance email or have a better search function than Google. But we're not currently using it to design trading systems. If you, and let me give you the counter example, if you ran a say market making firm, you guys have optifer for example, the Netherlands or flow traders. You can look at an order book and the rules of an order book. And probably more time, you still know the rules of the cocoa order book on the i6 change. The rules of that order book are very, very clearly defined. It's kind of like boundaries to the left and to the right. This is when the market is open. This is the pre-opening time. This is when it closes. This is the settlement time. This is auction. This is blah, blah, blah, blah. This is when the options expire. Now, all of this is then very defined. And you can feed this to an AI algorithm. And within this very clearly defined circle and boundaries, that algorithm can learn a lot of useful things, right? Kind of like how does how to operate within a market making environment. But in a kind of like free flowing market environment where all sorts of things can happen. You know, there's discontinuities, there's price gaps. There's all of a sudden there's a war happening in Iran. There's just not enough data to be quite honest, even though there is a lot. But there's not enough instances of these massively impactful events, which are happening all the time for any of these algorithms that I think to learn from in a meaningful and consistent way. And what I critique and I cannot let me make this the final word. But we've seen this time and time again in the financial markets. And not everything in the financial markets is proper and fine and honest. A lot of people use the financial markets to enrich themselves. Or kind of like just create the latest product, which they can actively sell and promote to their clients for a fee, rather than really helping them. And what we've seen time and time again is that whenever these new things pop up, and it's been blockchain, it's artificial intelligence, whatever the bus word is of the time, some firm will come up and create a product around it, create a fund that has artificial intelligence on its label, right? Or has blockchain on its label. And people get carried away with it. Or has data center infrastructure, whatever a hyperscaler in its label. And people, you know, because it fits the sign of the times, they will buy it. But so far, and I'm on the record saying this again, please show me the one track record or the few track records that are real traded with real and meaningful money that are purely driven by an AI algorithm. And you can demonstrate that. Say, have an auditor kind of like check it and say, yes, I mean, that has been pure AI fully automated. No human intervention, nothing. Just, you know, machine learning, whatever. And show me the track record that's good. And I have not seen a single one. Now, it might be that I'm not seeing them because they're not shown to me. You know, think about Renaissance and Citadel. Maybe there's a millennium to think about some of these firms who may actually run them successfully because they're smarter than us. And they, as business people, have absolutely no incentive showing their good results to us because they're not in the business of sharing. They're not in the business of running a charity. They want to have the PNL for themselves, right? But the funds out there that are marketed to people that say that they're using artificial intelligence or machine learning techniques. I still need to see the track record that is convincingly positive, and I haven't seen it yet. So that means basically a short answer, you cannot go skiing with algorithm, and you have sorry, if you work on it, then you have to be aware of it constantly, but that's that's I dodged your question, I forgot, and then I let, let's do what you said. But basically, it sounds really like when you have an algorithm, but it's what you say, we have an algorithm, but you're just so I understand, it's also that you have human interaction with the algorithm that makes you guys decide a trade or not. And it's very important. Well, not decide a trade or not, because we don't want to discretionarily intervene in our trading system, but I'll answer your question now and then we'll move it over to where it's H, we have an algorithm, and it's not AI, and the algorithm runs, it's essentially once a day, we're not, you know, intraday high speed type of traders. So, but we have systems, not just one algorithm, but an ensemble of algorithms or systems that run and that respond to price action, they will produce or trigger, suggest, trades and positions in the market, say, cocoa, we're short cocoa, right, we're long crude oil, we're long gold. Okay, fine. But when we get a new position or a new trade or we need to roll the market, because we're moving from one expiration to another on the futures curve, then we're not automating that process. We could, it's technically possible, you could program that and say, hey, we're rolling cocoa from, you know, May to July or whatever, and you could absolutely automate that, but we don't. We use it as a very important touch point for us to go to the market. So we're using the mouse and the keyboard and we're producing orders and we're pointing and clicking, but this process of doing this forces us to have a look at the market. So we see, hey, what is the, in my example, May, July, cocoa spread, how volatile is it, how liquid is it, what's the volume in that thing, what is bit offer, you know, and these are incredibly valuable feedback points for us because we're putting together a portfolio of markets and it's about a hundred or slightly more than a hundred. You cannot be an expert in all of these 100 markets. But what it at least gives us is, hey, you know, when we're putting cocoa in the portfolio, this is probably a good, a good enough period where we should be rolling it, where the spread market becomes liquid, right? This is how the task market functions if we want to use that. You know, this is when the task market is liquid. All of this helps us to reduce slippage, to reduce bit offer costs and to improve the overall trading results. And that is important to us and it's also important to our clients, if we naively told the algorithm to go to market, consume liquidity, maybe place a market order and force, in my example, the May, July, cocoa spread trade to go through, we may be paying five points bit offer, you know, skip through the trade, half slippage. That's just not necessary if you, I don't want to say if you're smart about it, but if you put in the work to understand where liquidity is and how these markets function, you can use this information as a feedback and improve your trading system and you can really only get this, if you as a human being, intervene with these markets and get involved. That's super insightful. Look, I'm not sure whether Alex is still with us, but I'm here. Oh, okay, good. Yeah, so there's already so many, so many interesting things I've heard that I was not aware of. But, so for example, the difference between market making and proprietary trading, I found it an interesting topic. Why? Because, you know, I recently spoke at what is called the trading fair in Amsterdam. I actually made this morning just before the podcast, I made a post about it. And companies like Optiver and IMC were there, flow traders. They are seen originally as market makers and I always thought or in the old days that maybe they were called high frequency traders and I always thought, okay, what these guys do is, you know, they, they make money of the offer bid spreads and then there was always a fight to be as close to the exchange as possible, okay. Now I also understand though that over time these players have been moving towards proprietary trading. Now, and, and for example, you mentioned Renaissance, now Renaissance is what is, I believe one of the most successful quantitative trading companies in the world. Jim Simon's, there's a great book about him, but I always got the impression that what they did has never been replicated by anyone. Now, and then you have recent news about the company called Jane Street. Now, I tried to, they apparently made the air, it was in the news the last couple of weeks and insane amount of money everyone is talking about it. Can you explain what it is that these companies are really doing? So in terms of what is the market making that they do compare to proprietary trading, how do they do it and how does it compare to what you guys do? Because it seems all a bit similar, but clearly you guys are not doing market making. You do proprietary trading, mostly trend following, that's what I understand. Now, what is it that these, maybe we can we can focus on Optiver, maybe we can focus on Jane Street, but what is it that they do? And how does it compare to you guys do? And I'm especially interested in how does the engine, the engine work. So from a fundamental point of view, that's not what you look at, but I understand it because that's what I've done all my years is really looking at an SMD. And then thinking about okay, how does that in SMD lead to the behavior of farmers and consumers and how does price need to move in order to sustain a balance in the SMD. As an essence. Now, that's not what you guys are looking at, you guys are looking at mathematics. Now, so it's a long answer, but it is because I've heard so many things and we mentioned a couple of names and this is really something that I would like to understand how do these companies work, how do you guys work, and how does the engine work, how do you build that algorithm in order to make sense of the noise. Because there's also a super interesting thing I totally agree with you. Most of what you see is noise and randomness on a certain basis and that is why I always say to people, look, day trading, it makes no sense because it's just noise, what you see, right? Now, anyways, look, could you could you provide insights in what I said would be very happy with that? That's an easy question, that's for you Moritz. Yeah. Sure. I mean, let's kind of like break it down in the that first of all, of course, we're humble to be even put into one bag with chain string or up to go all these things because they are much bigger and they are doing much more complicated things that we do. Maybe I just give you a little bit of background what we do and then you will be the difference and we can focus on the difference of what we do is essentially as you mentioned trend following it focuses on we are buying things that went up in the price and we are certain things that went down and prized that we hope in both ways that it either goes up a little bit when we are long or then it goes down a little bit or even not a little bit but a lot if it goes down. So either way, we have positioned we just hope for the trend to continue, right? And this is kind of like the end of the whole strategy probably of financial markets, right? This is not super smart. You don't need a pitch key for that. It's actually something people have been doing for years and years and years. It's very simple and we try to not be able to enter the noise by different methods. One method is we don't look at, for example, intraday prices. I mean, we look at it from an execution point of view, but the models don't know about for the priceful family when it comes to the market. to make the decision which is already basically thrown away all of that zigzagging intraday that that you see which helps the model to denoise and then we focus on very simple models so we do not as much as mentioned we do not do AI not for the reason that we do not like AI for certain tasks up because it has this problem of overfitting in many cases of kind of like taking the noise too seriously so we deliberately focus on very simple models though that can be breakdown that can be momentum very simple trying detection models acting also on a very long time horizon so we are not even interested in these many many small trends that are emerging like look at the intraday chart if you for example take an intraday cocoa chart or all chart and look at for example breakouts 20 20 minute breakouts whatever it is for some people doing the intraday trading world you will have signals all the time right you will trade back and forth but in our case we have a very long term some of our traits for example cocoa in a good example or these traits have been with us for years so we sometimes end trends for more than two years three years we have been in the gold position way before really kind of like ramped up last year in 10th of October so we have been in that for more than a year already so that the positions are really long term so we give them a lot of space to develop because that's the essential thing we do we try to catch the outlier traits we try to cut losses very very early so kind of like things go against us we out but once it runs into our in our favor we basically stay in the trend as long as it might carry us because one of the things and that's why trend falling works is that people underestimate how positive or negative a trend and go or have how long it can develop a lot of people they are very prone to taking profits once their position has moved into the favor but the essence of trend falling is really staying for the outlier traits and really keeping the losses very small and being very disciplined about it and that's also one of the things why it's so hard to do because if you do it manually without a system and if you're not sticking to the system and most cases you will be prone to taking profits and waiting for losses to come back so kind of like we all know that we're in a losing position it's down a little bit and you're like now I just wait for it until it comes back and runs into our profit position but in most cases it will not so that's how the system helps us so we are very very simple and I think like the market previously the market making from those as a proc trading firms that you mentioned left up to budget straight of course a variety of different frequencies most of them will be very active in the intraday frequency they will not hold a lot of risk usually on the bulk overnight so kind of like more focusing on both market making and also arbitrage in that sense so really kind of like making a riskless profit in our world what we are doing is definitely not riskless profit it is bound in terms of losses to the downside but if you look at volatility for example as a measure of risk and say that's in convenience we have a lot of inconvenience in the system by design so we are acting on a different time frame we are acting in a different way how we evaluate risk think that's the big difference is of course we could also trade some of the systems late trade or some of the methods late price or like there's a retrash somewhere that is interesting and that fits into our system and we can actually execute it then well we probably would do it but in most cases it's just so far apart it's kind of like we won't even get into their way and they would be now always that me Martijn I'd like to maybe at least try and answer your question in a slightly different way and maybe that is helpful because I think what you were observing or what you are observing is that a lot of the trading world today is using quantitative methods and algorithms and James Street is one example of that and Renaissance is another example and we are yet another example but when you categorize this you have say these high frequency traders or market making firms and it's you know it's James Street it's Citadel it's jump it's Optiver it's flow traders go down the list now they're using algorithms but in a very very different way they have a much different and much higher technology spend they you know improve their systems down to the hardware level where they have maybe custom built jobs to improve the speed of their execution and their access to the order book it is a very very different business their aim obviously is to provide liquidity and get paid for it to realize the bid offer spread and maybe in the case of James Street over time that has more of the little bit more into a directional prop trading that they're doing on top of it that is something that I cannot talk about I just don't know anything about this Renaissance is a yeah probably the most successful at least from a risk adjusted returns perspective trading firm that has ever existed and continues to exist today even without the late Jim Simon's who's been an absolutely fantastic mathematician and the way they're trading the markets with algorithms and quantitative techniques is very different to we're doing it they use speed speed detection techniques Jim Simon's used to be focusing on geometry and multi-dimensional spheres so many falls and they're using just different methods to find trades in the markets now go down the list there's now firms like us and you put us into the CTA category which is a regulatory term that is actually quite historic it means commodity trading advisor we're actually not a commodity trading advisor we're a commodity pool operator we're regulated by the CFTC but what a CTA essentially means is a firm that's regulated by the CFTC reports to the NFA and trades futures contracts that are under the supervision of the CFTC that not only includes commodities it also includes financial futures contracts but it's kind of like it became the catch all phrase for firms that trade in the futures markets but firms operating in the futures markets I mean especially with respect to commodities there's so many different techniques you know there might be CTAs or commodity hedge funds that run quantitative systems but they're executing carry strategies you know exploiting a essential structural risk premium that exists in the commodity markets to keep supply chains running and you know prevent stockout risk or priceful stockout risk these type of things there's firms like us that trade trend following strategies there's firms that trade mean reversion in commodities especially in commodity spreads where it exists there's firms that take commodity markets into the delivery window after the first notice date where all of a sudden different economics and price behavior takes place and they're looking to exploit this so it goes down the list right I mean it and then at the end of that spectrum you have the discretionary trader that also exists and maybe people looking at SMD supply and demand that go like you know what I just want to be full-blown long deck 20s crude oil because the market is incredibly backwardated and maybe the back end of the curve is going to be catching up with current price dynamics and I'm just going to put on a big risk position and I'm long oil and that's it all of that exists and but I guess the main point is the majority probably of today's trading activity is by in some shape or form informed or influenced by an algorithm or statistics or some quantitative method or like an automated technique it's just it has become that way because computers gave us the possibility to do that you know 20 25 years ago it wasn't that easily to do but does that mean then that it becomes more predictable because if if it's less influenced by you and more by computers you would expect that it's better predictable and you know what Alex in some way you are absolutely correct when you think about things such as by the dip in the S&P 500 or like you know you have risk parity strategies in a way you have predictable flows now is the end of the month so the first today is the first of May so the first business day of the months kind of like this window dressing month ends beginning of a month also has some you know specific flow dynamics in different markets and equities and effects and so forth right there's this continuous bit on us equities from 401k plants which is a thing right so yes in a way it does become more predictable but that does not mean predictability does not mean that there is profit opportunity profit opportunity usually correlates with uncertainty with you know people not having the same position you going against the stream you viewing things differently because as every bit you know things that are predictable are usually baked into the price and it's price in the market if the market is half way efficient right and that means that the opportunity to profit from it goes away or becomes very small and if I may add this as a final [BLANK_AUDIO] point, trend following, as Moritz has explained, this is, this is not a strategy that only the night template knows about, everybody can know about it. It's on the internet, right? How you implement a trend following strategy? That is a different thing. Like what methods do you use? What systems do you use? How do you combine them? What markets do you trade? Maybe Martijn and Alex at some point, we can speak about the markets that we trade because we don't just trade the largest markets, we trade some of the smaller commodity markets, which is why we have a size cap of 500 million. We're operating in markets such as oats and orange juice and things that aren't scalable into the billions. So the rules for trend following are out there. But why do trend following strategies continue to make money? Not every year of course, it is very noisy. It is because it's very hard to follow them. It's very hard to like them. It's very difficult, very counter to your natural behavior as a human being because it forces you to buy the high. If the SMPE or whatever market goes and makes new highs, as a trend following, you will buy it. And most human beings have a tendency to fake the move and say, no, no, no, the price has gone too far too quickly. I'll get out or I'll take the other side. And this inconvenience, which is, it results in volatility. It means that the return stream that we're generating is not smooth. It's not even, it doesn't have the highest shop ratio. It's very episodic. It has, has bursts of high performance and then nothing. But because this is so difficult for people to follow and stick to, I guess that is the main reason why there is an edge. And Coco, two years ago or two and a half years ago, when we had this massive run, I forgot when it was probably two years ago, is a great example of that because this market has been a dog for decades. No, no systematic trader made money with Coco. It was very difficult at least. And it's kind of like a $2,500 per ton. And then all of a sudden, it goes to $3,500. Right. And you go like, wow, finally, I make a thousand bucks. Right. As a discretionary trader, you have this urge and this impulse to say, this thing hasn't made me money for 20 years. I finally made 10% with it or whatever it was or 20. Let's call it a day and get out of the market. But this is this strange thing like you have tails in the markets on both sides of the distribution and the case of Coco, depending on the contract you looked at, it went to 12 or 13,000. So it did another four X from the level that you got out, which is completely, it is so, so wacky to even think about this, that it's very difficult for any discretionary trader to follow that move. It causes people such as Andrew Rand, who has nothing to do with Coco to all of a sudden, like Coco is completely out of the, we, you know, I've never been like, unlike you, Martijn, I've never been to a Coco plantation. You know, I drink it, I enjoy it, I like eating chocolate. But as a systematic trader, we go like, well, the thing is on a ride. Let's follow it. Let's go. And of course, it goes to 13, not of course, but you know, we follow it all the way to 13,000 because we don't have any reason to get out. But we also take it on the gin when it gets down. You know, we can never sell the top. We have to wait for the market to signal to us that it wants to stop moving higher, which means it has to go from 13,000 to whatever the level was to like 8,500 or 9,000. We go like, OK, well, probably that trend is now, it's not done. And you know, what we did, we then went short. And it's been one of our best short positions. So, you know, we're just completely unemotional about these things. But we recognize that these price moves in these markets can go much longer. They can be much more fat tailed than people expect. And sticking to these systems and following these moves, this is the very hard part. And this is why trend fully works. And because we cannot forecast which market it's going to be earlier this year was gold and silver. And who knows what the next market is going to be? It might be propane gas or palm oil or whatever. We really don't know that. But we will position ourselves to be in a position to participate in it. Now, it's very interesting that you say that there was a very, I thought inside for part of in Jim Simon's book that he writes that he wanted to override his systems in 2008 when everything was crashing. So even he was at some stage tempted to override the system. And I recall that we on the trading desk, so at some stage, we said, look, we also need to develop a quantitative trading strategies in cocoa. And so I hired a quantitative analyst. Very smart young lady. And she started to analyze all kinds of hypothesis that we had from a fundamental point of view. And we set up a, let's try to see whether there is merit in designing a quantitative trading strategy around it. And we did. And then we had times where our algorithm, if you will, gave a certain signal. And then we on the desk said, yeah, but it doesn't make any sense. Because we just had someone on the phone from origin. So, for example, we knew as proprietary traders that, for example, Ivory Coast or Ghana were selling the shit out of it. And they algorithm said to divide the market. And then we said, yeah, but look, the moment we start interfering with this algorithm, it is no longer purely quantitative. So we just need to let it, let it run on its own. But that was very difficult. But, you know, I think you guys have the experience. That's exactly. Not my time, but why was it difficult? Because it was that your team was not capable of letting it. It was the control, the human aspect of it that you couldn't get. Do you trust what you built? We trusted what we built, but there were times that the system said, look, you need to buy the market. And we just had five minutes ago, we had Ivory Coast on the phone. We wanted to sell the shit out of it. And we knew it would pressure the market. So yeah, what do you do? But the point is, if you override that algorithm, it's no longer purely quantitative or systematic. And that is what you need to get used to. And then the nice thing is in cargo, there was also a trading desk that traded all the commodities that everyone else was trading in the company. But they traded purely quantitative. They didn't give a shit about fundamentals. And I, yeah, I think the way it should be done. Now, could you guys provide some insights on how the engineering works? So people talk a lot about the algorithms. And I follow on LinkedIn, for example, companies like Citadel, we mentioned, but also point 72, Bariasni. And I see all those smart people. And what they talk about a lot because they need to obviously attract intelligent people who want to have an interesting day. So what they talk about a lot is solving problems. Okay, my impression then is, okay, they need to solve a math problem. Or they are engineering and algorithm. But can you guys give us some insights on you start your day or maybe the fund with a blank sheet of paper? How then and do you engineer the system that gives you the signals to do to execute the trades? I'll give this to Moritz. Happy for I'd like to just say one thing. I, by the way, Martina, once gave a presentation to your former employer, Cargo, speaking about how we trade soybeans. And they found this also very interesting because we have this system that trades soybeans and soybean meal and soybean oil and whatever. Yeah. And we're not looking at S&D. So I'm speaking to people with your background who were in the room and they were trading soybeans for Cargo. And it's like, yeah, well, it's kind of like, so I told them what our position was and we were probably short at that point in time. And they were kind of like, I silly, you know, we got to be long. It's like, I don't know. But the thing is, there is not one, that's the point I want to make. There is not one system. You can design an infinite number of systems that trade the same market in my example soybeans. And what you find and you ask, is it math? My answer to that is, no, it's not math because it's not an equation. It's not like A plus B equals C. There's not like an equation that produces a very precise output. It all has statistical probabilities and noise around it. But what we do, what we find, and this is what Moritz will tell you, is what is very important. Once you have a system, and once you have reason to believe that the system that you're running has a statistical edge, it is robust and resilient. That is something that is very, very difficult to do. Moritz will tell you that in a second. But once you have that system, you need to stick with it for such a long time. Come, hello, high water. That is incredibly inconvenient because it goes against your intellect. It tries to outsmart you. You think you're smarter than the system because ivory coast called, right? And that is all stuff that you then need to throw into the trash bin, and you need to follow the system. And you need to understand that all these systems go into long periods of drawdown. They go into underperformance. They're not there for you every day, right? They make you feel bad. But over time, if they are robust and if they have edge, they will come out ahead. Now, one way of mitigating that pain and then inconvenience is by not trading just one system on soybeans, but trading an ensemble of systems on soybeans. Some quicker, some slower, some using one method, some using another method, but all robust and statistically with a positive expectation. And that reduces some of that volatility and timing risk and episodicness of your return stream because there's always, not always, but there's then greater probability of at least one of these systems in the ensemble working at that point in time while others lay you down. So now, Moritz, over to you, how we designed them. Yeah, I think I will start talking about the jobs or what it comes down to because Marpine said specifically, and I mean, it's like with every job, you have to be honest, of course, the job adds to what's always emphasizing the very exotic and very interesting things like trying to attract maybe academics or sort of equations. Like Moritz already mentioned, nothing early anything in our cases in equate. And most of the things I would say 98% of the coding or the engineering are all that is plumbing. So it's go wrong. You have to make these faster. The code doesn't run. You have reconciliation errors. All of the argument stuff is in the day to day business, right? That we as a more company have to take care of ourselves. And in bigger companies, of course, the roles are usually split up a little bit. And we combine it all in kind of like two, three people essentially. So it's a little bit different, I would say. But in general, the job variety, very high. I mean, I will start the day. The systems are designed. So if you do not develop a new system, the systems are designed to run in the morning. They produce orders for us. We look at the orders. We look at the market. We trade different markets at different times, of course, because there's opening hours. There's different liquidity windows. So we start in the morning usually, but it's relatively, most of our liquidity or the trading is center-owned. Yes, open. So in the afternoon, usually it's good for us to write and can take the mornings to look at the models. Look what has been happening in the markets. Anyway, even though we are systematic traders, we still like to feel the heartbeat of the market. Read the news. Even if we are not influenced, we have an opinion, right? We try to keep that away from the systems, but it gives us an idea what's coming up. And that's essentially from the day-to-day work, the thing we start trading in the afternoon. And in between, we have a lot of free time to hop on calls with clients that's most of our day-to-day work. We do some reporting. All the other stuff that comes with having a business that goes along with it as well. So that's in our case, the job, right? From a purely-quant perspective, usually in the bigger firms, what you will encounter is a much more diversified role. So kind of like there will be people who will have to deal with the data. There's engineers, of course, who deal with the systems. There's usually a trading desk, who's maybe doing the execution. There's a portfolio manager who looks at other things in the portfolio, kind of like risk maybe as an additional function. So it's more nuanced as well. But still, I would say for people who are interested in this, it's a very interesting job, even if, like I said, before I leave, I said, it's probably kind of like bombing to some extent. Now that's kind of like the overview of what we do and how the job looks like. When it comes to designing the systems, that's a whole different story like Marx, how do you make a robust trade system mutely? Yeah. Not a big fan of backtesting, to be honest. Like over the years, you see it all backtests, and we've never seen the backtests that go down, right? You only see the PNL code, which was up nicely, and people show it on Twitter or LinkedIn or whatever, and they advertise it, and it's great. And of course, we also look at PNL curves in the trades, but as Moritz said, they need to be robust, they need to be a significant amount of signals as well, and they need to be a lot of markets for us to make sense. So for example, what we do, we usually apply one system to our whole universe of markets. We do not kind of like fit it to each market in the Virginia, because we say, well, this is meant to capture a trend. It should not care if the trend is in cocoa, if the trend is in gold, or the S&P essentially. We apply this kind of like system to all of the markets, because we have such a small sample size of one of the markets that we will have to basically treat the whole subset of a hundred markets that we have as one realized position of this is how a trend can look, not only if we can catch it. And that's what we're using this system for. And then we do that with different systems, maybe that are tuned to different frequencies, some of them might do different varieties of trend, but that's how we treat them. And then after that, we will try to evaluate what's the outcome and not touch it too much. That's also one of the rules we try not to look for, let's say we found that a hundred day break out, and then we find that the 98 day break out is blended better, we will not switch, we will not go to 98% break at all, introduce some new parameters that are better in there, like people do that all the time. And like the the bettas always lie, it's always, even if you try to include maybe slippage and trading costs on all of that, in the end the bettas will always lie to you, so the only thing you can do is actually take the thing live. Preferably we do that even with our own money, so previously we have been trading a few strategies, which we kind of just test about ourselves, for example, and sometimes if I find something interesting, I just trade one lot of one market to kind of like better feeling for others and work, for example, and then is it something we can do in the fund, maybe, or even in the fund, we usually start very small and see where it goes and see if we really traded that way. For us, the bettas is only a very rough indication of what's happening out there, and we have a huge collection of strategies that we are ready to deploy, but then in the end, it comes down to kind of like how does this system really behave in the real world, what could be the edge cases and what could actually go wrong. In our case, good news is there's a lot of things that can go wrong, because we are usually not very time sensitive when it comes to execution, for example, even if an exchange, for example, it's offline, happens in some markets more than another, we do not care too much if we have a signal, and if it happens today or tomorrow, usually with the long-term finalizing that we are on, it's not an issue, so we can still take the trend today, tomorrow, the day after, it over time doesn't make a big difference, but you usually have to test for that, and that's why we have our own stack kind of over the years, we built the stack ourselves in various roles, always extending it, always putting some effort into customizing a little bit, and that is a fun part for me, right? I like building that stuff, and also like trading it may be less than more it's for me, it's more, I would say, I'm mostly even more remote to the trading than he is, I'm more into the development and the building of the infrastructure, happy if something goes through, if the process is work, if I don't get an error message. In most trading companies, Excel is still the core of everything they do, how is that with you guys? Is that very different? It's a hybrid, so the trading systems, they all in Python, the infrastructures in Python, we have stuff in the cloud that is running, the databases are using the cloud, but to be honest, it wouldn't make a big difference, we could also run this locally, it's not that problematic to run in the small robustness thing, to have it in the cloud, and to have it in Python, the systems we do, you could also run in Excel. I mean, it takes some time and make precious and there's some drawbacks to using Excel, and we also have Excel sheets, because we're getting those by the administrators, for example, we're getting those by the broker, sometimes you need to check things very quickly, so you're making a back-of-the-end calculation, or you have to check something, you want to see it side by side in that heavy layer to be. which is just for us as humans nicer than me going into the code and kind of like looking at numbers and then trying to kind of like explain it. So we live in both worlds and there's a benefit of having both things but if you're systematic trading I would definitely recommend not maybe at some point moving it away from excel to something that is easier to maintain that is not as error thrown because a big drawback of having these systems hard coded in excel is usually that there is no versioning so you don't know what the changes are and that you have a trip of error so over time usually there's a smaller error that might treat and it's not easy to replicate things and also things get lost. The thing with having this in a solid code base and in an environment which is inspired by software development is actually robustness so you have a very robust system and I think it's how all of the big firms in space actually work they usually have a very big percentage of software engineers of people who are prone to engineering who like it who like the structure who like the whole life cycle of things how software is deployed and how it's implemented because it just trades a lot of robustness and it's also very I would say very satisfying because for example now not an algorithm but I've worked on systems that are doing intraday trading as well and it's extremely satisfying to be able to for example deploy a new strategy to a market in under five minutes so kind of like from the ideation of a strategy to the goal of actually trading it takes five minutes and this is just with the whole thing with getting the trades in a repeatable fashion to getting the reconciliation so having everything when the book everything is showing up nicely in whatever back office it's 5 minutes and that's kind of like extremely satisfying from a core perspective to have the ability to do something like that and that's only possible to be infrastructure because that's the that's the thing for me is like when I was trading you know I found it always interesting that you had actual containers fiscal product no currencies the world problems what is for you guys the fun part of what you're doing right is that exactly what you just said you know that's you know launching a new strategy within five minutes what's this you talk about satisfying what is the fun part about what you're doing at the moment I'd say research is the the most interesting part that never stops you can always come up with new ideas figure out new things design new systems I mean that is the challenging intellectually challenging and enjoyable part really at some point and probably Martan you agree with that like putting an order into the audiobook and getting an order executed that's like breathing in breathing out it creates excitement the first you know a couple of times you do it I mean the first year but after a while it's kind of like a manufacturing process you just you just do it right obviously as a trader you can never step away from watching your peak now and watching the risk and running the trading book and you know looking after the fund I mean that is obviously a daily task that that I do that were it's involved in as well and I must say I enjoyed that too that is not something that I kind of like don't look forward to in the day it's something that I'll actually be doing after our podcast which we're recording in the European morning and I'll do it with a cup of coffee in hand and I'll enjoy going through that it's it's not a big deal but the research is the the fun part and let me say there is at least in my experience even with quantitative traders experience plays a massive role you sometimes hear this or you often hear this for discretionary traders and go like well you know you have a junior trader but to become a really good solid trader you need to have years in the trenches and ideally you have worked in the pits and you know and understand the market and the market participants and like after 30 or 40 years you're this accomplished trader that you know has all the battle bruises and you know still say yeah and this look I mean the markets teaches painful lessons important lessons and I think experience is incredibly important but it's also important for quantitative traders because when you like I started designing trading systems in the late 90s which is about 30 years or so ago now but obviously this wasn't my late teens you're just getting into university this is kind of like when you're intellectually building as a human being and you think you know there's some treasure at the end of the rainbow you just need to get there and you can you know work with all sorts of complexities because they teach you that at university and you actually very motivated and excited to use them so whatever you're where there's a great propensity and likelihood that in your younger age you'll be designing quantitative trading systems that are overly complex because you're using all the techniques that you're learning at uni right this thing this overlay this filter this new parameter this technique yada yada yada and obviously you will create something that looks absolutely fabulous in the back test in the rearview mirror right but what you've created is very unlikely to repeat itself you know it has sample size issues and has all sorts of problems you've essentially overfit a system and you're not yet at a point in your career where you can understand what it is that you've done and you can identify the mistake that you have made because you're lacking that experience so then over time and I really think there's only one way of learning this and this is with your own money your own losses your own P&L and not with a paper trading demo account because that doesn't give you the feedback the feedback needs to be painful otherwise you're not learning it is kind of like Neanderthal logic so over time you start using this feedback if you're halfway intelligent and a hopefully good way and you change the way that you're designing systems and one of the things that you'll do is you'll rip out a lot of the overlays and filters and additional parameters that you've previously put in now you don't like that process because what you now see in the back test is something that has a lower shop ratio that has more inconsistencies that has bigger drawdowns that has more volatility that is no longer as smooth right that is more the real world unless you are gym simons or you know some of these towering intellectual people that we're looking up to but they're just in a different you know they're playing this in a different leak and who knows what it is that they're doing we're just not there so really over time and this comes with the experience factor also for quantitative CTA whatever trend falling type of traders I think you need quite a few years decades of being around the block to figure out how to build these systems and how to combine these systems in such way that you can still crystallize the edge that a trend falling system gives us but you're staying away from the perils of using overfit systems which will end in disaster they just do and that is something that if you're 20 years of age you have a very high chance of going with the most complex stuff and you think that's great because you're proud of yourself that you were so smart to design these things but you don't know yet the what you've designed is very has a very high likelihood of letting you down that makes me bring to something that I that I thought about I saw a guy win a Citadel hackathon and he was saying that he was gonna work at some company and I replied on LinkedIn to him and I said look you need to start your own business because I thought about Kang Griffin who started from his dorm at Harvard and I think there are plenty of people very smart people now and with the access to technology who think well you know what I can do this as well so what do you say to those people and something related to that what everyone in the podcast always wants to know especially the youngsters how do I break into this so would they need to apply to you you're you're obviously as much smaller company and the don't have like a traineeship like like the large ones have but what do you what do you advise to people who listen to this podcast are excited about it because what I think is super excited exciting about what you guys do it has both the intellectual part of engineering but still being genuinely interested in in markets as you as you discussed super interesting how do you what does your advice to to those youngsters that listen to this how do I break into maybe a company like yours some of the names that we mentioned the Quantrators or start your own firm now you've you we now we now need to add a another and maybe that is the third component to this which is the business component. Being interested in markets is one thing. being interested in finance is another thing. Building a business and running a business and managing that business and applying for regulatory approvals is yet another thing. And you're not enjoying all of these parts equally. So if you want to start your own business, make sure that you have a passion for it. You need to have, I think, this entrepreneurial spirit. You need to be prepared to fail, which happens most businesses that start fail. So the statistics or the probabilities are actually set against you when you start. Obviously, that is very difficult to recognize when you start a business because you think, well, if you're starting a business, clearly, that's the one that's going to be successful, but, you know, the statistics will tell you a different story. So just make sure that you're prepared for that. But really, if you are passionate about this, if you want to run your own business, you don't want to report to a boss. You want to be your own boss. You want to operate in your own motors up around the, there's just no better way. Go for it. Do it. Give it a try. The younger you are, the more chances you have to give it another try and yet another try, right? Now, but also be prepared that starting a hedge fund. Let's just call it a hedge fund or any financial services or trading firm has become increasingly more difficult time intensive and cost-intensive over time. And that is because regulations have only gone up. You need to apply for approvals if you're on Europe with your European home regulator. We're all still regulated in the US. It's very hard to get away from that. And once you are, it's not just the financial commitment but you have to have reporting commitments. You have to have compliance. You have to run your business in a certain way. It's not just, you know, it's all just the Wild West and let's just go and do whatever you want. There's all of a sudden a bunch of rules that you need to adhere to. And none of that, as far as I'm concerned, is fun. It is a necessity of our business. Something that we have to do is essentially a cost function that sits on top of our business as an umbrella. It's not a driver of P&L, you know, to the contrary. It is something that detracts from P&L and it's not necessarily enjoyable, but we understand that it has to be done. Okay, fine. I would advise, I mean, this is just me personally, for all the reasons that I've spoken about with experience just five minutes earlier. I think it makes a lot of sense as a trainee or a younger adult to go into one of these firms that you've mentioned and get some experience there. Because you'll see things that are going to be invaluable and very important that you can take away and at some point maybe translate to your own business if you still want to start it. But you don't have to start your own business as I trade at 18 years of age because I think you'll be lacking experience. There's nothing wrong. Absolutely, I did the same thing. That doesn't mean it's right, but I think it's absolutely cool to go and work for these firms for 10, 15, 20 years. Create a network, get to know people because guess what? When you start your own business, you need to know people. If you don't know anybody, if you don't have a network, nobody will give you any money. Nobody will be there to help you. Nobody will be there to give you an introduction to somebody that could help you do something. It is, I think, a good idea to do this. Whether that is a market-making firm or a hedge fund or a bank or a commodity trading firm like Cargo, find what's interesting to you. If you're into physical commodities, by all means, go to drive, go to cargo, go to VTile. If you're into derivatives and risk management and hedging large books, go to an investment bank and see if you can trade fixed income or equity derivatives for them and create structure products. If you're more into, well, I want to have my old book and create P&L, go to a hedge fund. See if you can start as a junior trading trainee and more PM and maybe at some point, get your own book and produce P&L. I mean, there's different ways, but they're all valuable. I think all of these things will teach you invaluable lessons. I sometimes, personally, I traded derivatives and structured derivatives. I always love the commodity markets. I just think there's the most fascinating ones because it's actually stuff happening that we depend on and molecules moving. You can become extremely good at arbitraging the forward curve of the S&P 500 and the repo and expectations, but it's never going to be the thing that you get physically delivered. It's never going to be the thing that's impacted by weather. It's never going to be seasonal or anything. These commodities have something to them that makes them incredibly. They make them just, I think, personally, just my opinion. More interesting. I think if people go out, as they start trading commodities and get into these markets, there's a lot of things that can't be learned. Well said, well said, sir. I think we need to look a bit at the time. Alex, do you have a final question? No, I think it's right. I think for the young guys, I think it's especially like why is this an interesting one? Would you say as well? If you have experience, this is an more interesting place, I think, then when you really start from scratch with all the experience. I think also what you guys have to go through in the beginning by keeping your shit together, basically not acting on everything you hear in the market from origins or whatever, but keep following your algorithm should be super, super tough. That should be really, really hard somehow and also maybe super easy if you trust what you're doing where you believe in, but on the other hand, maybe also really tough. I think that's for me really interesting. I think a lot of commodity trading houses, also the smaller ones, are trying to find out what's the best way, what's good algorithm, what's the forecast of trading the physical product or at least helping them in the positions they're taking. Maybe you guys have an edge that a lot of the companies would love to know about and to help them trading. It's super interesting in your story what you guys do and thanks for sharing that. You're more than welcome. Thank you for inviting us. I love this. Thanks a lot.