My guest is Artur Sepp, Director of Research at Quantica Capital AG in Zurich.

In 2008, Artur was working on structured credit products for Merrill Lynch, giving him a front-row seat to the ensuing credit crisis.  We use this experience as a jumping off point for our conversation, with Artur providing both pragmatic and philosophical lessons learned.

One of those key lessons was the role of liquidity, which Artur argues is the key factor behind many premia we see in the market.

Artur’s focus on liquidity grew as he transitioned to London in as an equity derivatives quant, where he was responsible for building models to hedge options on illiquid underlying assets.  Here we get into the nitty gritty, discussing a paper Artur wrote about the practical realities of delta-hedging options under a framework of discrete hedging and transaction costs.

In 2015 Artur moved to Julius Baer’s advisory solutions group in Switzerland where he served as a client-facing advocate for alternative risk premia strategies.  Here Artur had to learn how to translate his deep quantitative knowledge into client understanding.  He shares with us some techniques and tricks he learned for effectively communicating what can be rather complex ideas.

Today Artur works at Quantica Capital, whose flagship product is a Managed Futures strategy.  I ask Artur for his opinion on recent struggles in the managed futures space and what he thinks the future for trend following managers will look like.  You definitely won’t want to miss his answer.

Artur is a fountain of quant knowledge and offers the unique perspective of someone who has both spent time deep in the weeds and time trying to explain the esoteric.  There are lots of gems in this one, so stay tuned.

Transcript

Corey Hoffstein  00:00

All right, are we ready to go?

Artur Sepp  00:02

Let’s go. Let’s go.

Corey Hoffstein  00:03

All right 321 Let’s dance. Hello and welcome everyone. I’m Corey Hoffstein. And this is flirting with models, the podcast that pulls back the curtain to discover the human factor behind the quantitative strategy.

Narrator  00:25

Corey Hoffstein Is the co founder and chief investment officer of new found research due to industry regulations, he will not discuss any of new found researches funds on this podcast all opinions expressed by podcast participants are solely their own opinion and do not reflect the opinion of newfound research. This podcast is for informational purposes only and should not be relied upon as a basis for investment decisions. Clients of newfound research may maintain positions and securities discussed in this podcast for more information is it think newfound.com.

Corey Hoffstein  00:56

My guest is arct herself, director of research at quantica capital Ag in Zurich. In 2008. Arthur was working on structured credit products for Merrill Lynch, giving him a front row seat to the ensuing credit crisis. We use this experience as a jumping off point for our conversation with Arthur providing both pragmatic and philosophical lessons learned. One of those key lessons was the role of liquidity, which Artur argues is the key factor behind many premia we see in the market. Arcturus focus on liquidity grew as he transitioned to London as an equity derivatives quant where he was responsible for building models to hedge options on illiquid underlying assets. Here we get into the nitty gritty discussing a paper Artur wrote about the practical realities of delta hedging options under a framework of discrete hedging and transaction costs. In 2015, Artur moved to Julius Baer’s advisory Solutions Group in Switzerland, where he served as a client facing advocate for alternative risk premia strategies. Here arctor had to learn how to translate his deep quantitative knowledge in decline understanding. He shares with us some techniques and tricks he learned for effectively communicating what can be rather complex ideas. Today, Arthur works at quantica capital, whose flagship product is a managed futures strategy. I asked Arthur for his opinion on recent struggles in the Managed futures space, and what he thinks the future for trend following managers will look like. You definitely won’t want to miss his answer. Arthur is a fountain of quant knowledge and offers the unique perspective of someone who has both spent time deep in the weeds and time trying to explain the esoteric. There are lots of gems in this one. So stay tuned. Artur, thank you for joining me on the show today, I got a lot of elcome really excited to have you here. And this is one of those times where I feel like this podcast could probably run it for hours, there’s so many interesting things to talk to you about so many different ways in which we overlap in our research. Normally, I’d like to give guests a chance to sort of give some background, but I’m just going to dive right in to your story sort of Midway, try to pick up the thread based on conversations that we’ve had in the past because I think that’s the best way to sort of kick off in this conversation. And so where I want to begin is talking to you about your 2008 experience. I know that in 2008, you were working at Merrill Lynch credit derivatives desk, you were doing structured credit work. And that gave you a real insider’s view as to how the 2008 crisis manifested really within the bank. And so what I was hoping you could do is sort of start the conversation exploring some of the things that you saw at that time, how the crisis unfolded and what you learned or what your experience was like.

Artur Sepp  03:54

Yes, definitely. It was, I think it was great time. It’s a start the career, enormous experience that you can only get in lifetime. I think, going precrisis there is a sub know that you can read the books. And from insider point of view, I think what happened is greed, of course. So I think we’re going back to say 2000s After the economic slowdown, investment banking, we’re looking for new opportunities to enter the field to start creating new products that also were demanded by investors. At that time, we also had a period of low interest rates. So everyone was looking for risk to take risk. And of course, investment banking saw an opportunity and again, it’s simple. If you take what is portfolio theory tells us all about if you take something that uncorrelated and you can remove I do syncretic risk, I do syncretic risk can be done linearly reduced in a number of years instrument. That was a all thesis about creating credit default tranches or mortgage bonds. And then, of course, what happened, it was still not enough. So people, I mean, if you save your PayPal, normally, credit default displayed on the corporate bonds is maybe 100 basis points. In normal times it goes to 50 basis points. So, at certain point, it was not enough. People wanted to have say, two 3% of our labor, investment banking come up with a solution. So effectively you you take, say mortgages, started with, say, tripled out mortgages, and then went to more subprime mortgages, but it is simple. So you take few mortgages, you bundle them into special utility vehicle, enter, the top bar will get like triple A rating, because it’s very rare that every single default in your pool. So what kind of that was big demand from, say, institutional in the banker fields. And yet, at certain point, though, it’s theory, you need to do origination of mortgages of asset backed securities. And at certain point it was replaced just by synthetic securities. So instead of say, giving someone a loan, and by structuring nature, mortgage bonds, you just reference and this reference the problems that you can reference multiple securities. So one mortgage could be referenced by hundreds of synthetic bonds, and so on bank inside, then it became a machine that, unfortunately, you you need in with banking data and inventory. And of course, was when everything goes up. And if you keep an inventory of risky stuff, it’s okay, you can make extra money, but the problem started with liquidity, once liquidity dries out, you cannot sell your inventory, it starts draining your balance sheet it what happens in to seven, and then in the end, you have too much inventory. On top of that example was with several banks, that their treasury department invested in the very same triple A securities that started to lose in value, it was both on liability international side that you get double hit, but going to specifically to say more than me, because my role was always a structure strategist, I developed models, I was responsible for giving the accurate prices for managing risk, and to given supported and what happened actually interesting was also oversight. So imagine a simple credit derivatives index with reference. A simple way to understand these securities is like a T if you have a TF on cash bonds. So if we will references several underlying bonds. And for risk perspective, if you want to say computer, some credit duration or some sensitivity to credit, you need to price individual securities. And of course, if they are deportable securities, it can happen that one of the bond defaults, and it should not affect your pricing. It just is nothing. But actually, the whole derivatives. One of the biggest broker dealers was not able to account for defaulted names. So in effect, one could default a company could default, if you still have bonds, right? And you know that there must be some pre recovery before the resetting before the market participants settles recovery, maybe few months is really important. This gives Zahra but you cannot price it. That was oversight. We think if you want to put say going back to over a year, for example, you want to put value of zero, but you cannot because it’s a contract. So, another thing to try you want to put duration to zero, then you still you can compute your risk and to duration to zero. That is one way you just above the spreads, you can put spreads, maybe two or 3000s. If you don’t you have zero value. It was rigid. So to tell you this, but it seems like the biggest learning experience is always be susceptible to models. The model is only as good as the assumptions that you assume. And the second is market say market prices. If you assume that you can say you can You’re also more than one this applies for more illiquid securities. But what we’re talking say cash quantity, liquid security. If you say, I can interpret it, if I use a liquid securities like credit default swap and Treasury curve, in practice, I can replicate, I can replicate the fair value of this bond. But given that it’s equal, the market is liquid enough. And what actually another example what’s happening to your weight, exactly as this token, convertible arbitrage, or even cash arbitrage, if you say, if I’m doing more anti by protection, I’m neutralize them. So I can also crash my Treasury for instance, what I’m after is a spread between yield that I get on cash bond, which has some several components, it may be callable bond. So why yes, it’s typically higher than I would pay to protect it. But what actually happened into your wait, there was wide wide spread almost five 6%, or several bonds, meaning that if you owe cash bond, you would be at a disadvantage, even though you would be interested years. But because it is priced swabs were more liquid underlying bonds. You could come here with a female, even though you did this. So to summarize, always be critical or modal and always be critical of what you fit into your model. That is the recovery plan. If say, the market was frozen, and I still need to price, why securities is your model allows for that.

Corey Hoffstein  11:46

So I want to pick up on that idea of what happens when the markets frozen, right? You mentioned this idea. In price via replication. I know that in less liquid markets, there’s often this sort of idea of price via interpolation that you can look at similar enough securities and, and try to come up with a price for something that maybe hasn’t traded for a while. But the fundamental assumption there is that liquidity exists somewhere in the market. And I know that in 2008, that assumption was really challenged, right that if there’s no liquidity anywhere in the market, or the securities that do have liquidity are not similar at all, to what you’re pricing, ultimately, your model breaks for pricing. So what is the solution look like for a market environment like that, when there is no liquidity? How do you handle an environment where you can’t price, necessarily via interpolation of other like securities?

Artur Sepp  12:41

I think, to answer shortly, if you say we have different agents, right? If you’re an investor, and you know that some of your liabilities, you wouldn’t be able to add value correctly, you need to introduce this period of liquidity where your liabilities cannot be priced accurately. And the only way to say hey aged is to bring something liquid that can cover your risk. So for example, say cash bonds, if you’re long Caribbean, say you can overlay with credit default source. Again, there’s all this since US bases, but usually in dissertations, first of all models a period of how under what conditions you can cover this bureau. So for unfavorable pricing for you, and second overlay with Samsung, that is still liquid and where you can get at least some protection is on the investment side. Or say on banking side. It’s usually why all business was kilter since your weight, any business that would rely on the model and this exactly the model that can get it wrong model always assumes some continuity of price discovery between liquid instruments. And since market learn or regulators loans, that is actually impossible, because we are not talking about theory, we’re talking about practice, say in practice, in the sense I think it’s like a spread trade almost. So this type of arbitrage trades, they’re always a spread trade that you assume that deviations from fair value will mean revert at certain point. The problem is before they mean revert, they can diverge even more effectively, you can end up also the problems that you have long dated stuff say 10 year yes exam you can make money if you call for this position 10 year, but if you’re highly leverage, like Lehman Brothers or beyond stores is impossible. So therefore, and I think why his businesses eventually died. And right now, similar stuff is maybe like, I think, annuities, they will move to insurances. And these people, they are relatively conservative in their capital requirements. So chances are for institutional, if you want to do it, just make sure you have enough capital in your reserves to go,

Corey Hoffstein  15:23

Thomas a little bit like pricing for model risk, you got to self insure assuming the model is going to fail. I want to stay on this topic of liquidity. In a prior conversation that we had you actually mentioned a little bit offhand that you thought that liquidity was really a driving factor of pricing in the market, and you even viewed things like the credit and volatility premiums as really being a function of liquidity. And that liquidity is sort of the key factor from your experience. Can you walk me through that line of thinking and how that sort of plays out in the market,

Artur Sepp  16:02

actually, I was a Intuos 13 to 14, I was asked to do like a small project, I will not the name of bank where I work, but at that time, they wanted to start paying dividends. So it was already a recovery period after the financial crisis. So they actually wanted to say in the matter is very practical topic, right? How much how much do you budget the bank should pay? And they asked me to like just present some more arguments, given my background. And actually, when I looked at it, I always thought credit is always when you look at any company, the average company, it’s always a trade off between the bondholders and equity holders. And each of them have contrary objectives for credit holders, you want to console for as much cash as possible equity holders want you to pay out cash. But now, what happens liquidity means if you know credit literature, there are two very interesting pricing models. One model is called Milton model of default, it effectively say that shareholders equity is the residual value of company minus obligations. So, in fact, equity holders they called call options, because your liabilities are capped at stock price and the model is good is but what people found the research and found what Samson’s that is called I think credit puzzle, that the model is not explained by fundamentals, you can only explain part of the credit spread and the credit spread is consistently higher, that would be assumed by the model. So, to paraphrase that also, if say empirically, if we take say AAA securities, and we say we look at realized default or losses due to rate in transition, it may be 1% annualized, and Mike repays you say, 1.5% annualized. So, you have 50% basis points that you cannot explain by structure model, but since it does mean the series three, and the answer is cyclical risk, where there is another layer, credit spreads, which come from systemic risk, and how it translates to a company, because you release the company also you don’t really repay the debt. So, now imagine in good year, say in two or three, a company raised five year there, I’d say whatever 1% Anything to wait it needs to refine us. And because it’s nice to ask again, the it needs to ask is debt holders to give it cash, but he’s got it, everyone wants his money back into your way, it’s actually a good thing and I just recently heard in a crisis, you sell more what you want, you see what you can and it was through enjoy it any possibility to get your money back, you will embrace it instead of four. So for a company, so, there is a part of say depth risk management that actually does not depend on you, no matter how good you are conservative, if you want to roll your debt in a favorable time, implicitly you need to account for your higher borrowing costs. And now so going back to this project, is what I did my views some simple model of actually what I like about it was meeting say real So, for fundamental model, you always need to know say how much cash you generate. And fundamental model tells you okay, this is your rate equity is, the more equity you have, it’s easier for you to raise the depth on one hand. So if you want to pay dividend is like a function that maybe you can put increase your reserves. But then when the crisis comes, there is a cyclical widening of spreads, that you can maybe not raise it right now you can wait for it, but the inside of this model will work exactly the same. It is saying there is always cyclical risk premia. So part of our question is liquidity. Liquidity manifests in periods, when money is very expensive when to get money is or to save to finance your liabilities, it becomes very expensive.

Corey Hoffstein  20:47

So there’s this idea of cyclical liquidity that you’re talking about that sounds like market participants will price in that there will be this extra part of the carry that they demand for that potential absence of liquidity. But there’s also these assets out there that are just structurally less liquid. And I know you had a lot of experience with some of those more illiquid assets in your role. After 2008. I think in 2009, you moved to London, where you were working on an equity derivatives desk, and you were building models for hedging options of assets that were just more illiquid in nature. I think you mentioned potentially emerging markets was one of those at that time that you were working on. As you look at more structurally illiquid assets and trying to build models for hedging derivatives. And that sort of stuff. Again, going back to this idea of less liquidity means that there’s less market implied information, what are some of the challenges that you’ve faced in dealing with those types of assets and building sort of derivatives and hedging models on top of them? And and how do you try to account for those problems

Artur Sepp  21:59

in liquidity? So imagine, for example, we still option or know, emerging market equities, say, some Chinese stock or say Alibaba, Alibaba, probably trace in Hong Kong is traced in New York, but also in Hong Kong. So the problem is that if the size of your threat is very big, and if there are say, imagine, we just saw struggle, so we saw put, and we saw coal. So initially, our density is about zero, but then something say happen. So just for example, it’s a car. It trades in New York. But imagine overnight that there was some tweet from Trump that Hong Kong is deeply Dow and given the size of your position, if you want to hedge rebalance your delta, which say the stock tanked, you need to say buy whatever 10 millions of shares, you know that if you put your order in the market, you’re going to move with, it’s already moved out, and then you are going to proportionally to your hedge needs, you’re going to move it out in this crazy, cyclical feedback loop. And it’s a big problem for less liquid markets, where to trade in big nationals is a problem. And if you can retell you that say Originally, the trade could be good say you sold a template, that’s your TNT, you expected realize this at 20. And in option trading is relatively good trade. But if during the Heejun, you’re say, volatility cost is 20%. So you lost money, and you lost money, not because they realize volatility can still realize at 20. But because you cost it was you who created this volatility, it costs you a lot to hit. This is one example. Another example is gap risk that, for example, let’s say again, the CDR maybe you just don’t you don’t have big notional. But imagine end of day you’re Delta neutral. But every time there is something going on in China, and every night, the stock gaps 4% all say to 3%, which you cannot hit so every time you lose this two 3% gap is so insane. It’s again close to close volatility can be relatively small and will not cause your problem. But if you hit every gap, you will bleed to gamma to say that and therefore from beginning is not only enough to look at say what is my expected spread? You need to look also what was my contingency plan if something goes wrong, or I need to catch a big chunk of it, what is my expected cost? So therefore, what I like in this state A huge in a small quantity in the output of your model is not only this a fair value in ideal conditions, but also adverse, you need to account for adverse conditions.

Corey Hoffstein  25:13

You actually wrote a paper on this right? I think I read it, it’s on SSRN. about sort of the idea of okay, in theory, we have all these option pricing models, and you can assume continuous delta hedging, you actually wrote a piece where you talk about what optimal hedging might look like, I think it’s for maximizing Sharpe ratios I recall correctly, can you talk us through some of the big muscle movements of what the paper was about and sort of what the big driving factors are for making those decisions around frequency of hedging option exposures? And what ultimately are the big sort of cost benefit analysis you need to make

Artur Sepp  25:57

actually talking about this before it’s your thing at the time when I was looking at it, traditionally, oh, hedging was done intuitively. So, the trader would say, compute the price or the delta of his book and he would have an intuition Okay. This one I know that I should not Ah right now, I should wait in the day and so on. So, actually, I started working with one very smart guy, he saw that there must be like a program that he had good ideas, but in February, he said there is a bond. So you cannot hit your residual risk and you need to model the visa of your bond. So, for example, if it’s expensive stock, you cannot trade very often you need to increase your bonds or in the opposite, if stock has a lot of gaps, then also your your bank must be relatively wide. What that means for you is that you have more Delta risk your position needs to be associated with Delta risk in the only way is okay we are back into market risk framework, you have say cost reward, your reward will be always say spread between important realized now, minus transaction costs, which are say, quadratic function of your occasion frequency and your risk and then you have risk which is volatility of your portfolio, which you can reduce by hedging frequently. But then, if you hedge frequently you have more transaction costs. So, it’s a perfect currency risk reward balance, and we can actually optimize so we have our Sharpe ratio, we can optimize and we can solve explicitly given say, estimated spread, transaction cost and volatility of stock we can actually design say, optimal hedging frequency. So just to give you an example, you really notion trading you try to reduce the oh, there is a spread this is a rock from say from vix your expected spread will be a row row yield the difference between say, one month future contract in this spot, that you would if nothing happens, this is your roll yield that you would get if futures are in contango. So, if you say thinking more relative value, it will be your expected implied say implied volatilities that you get from selling the stock and realize that you accrue by dedication. So these will be say 5%. Now, if my say transaction cost, it cost me say 1%. For my data, I will lose say on spread, say in one underlined, it will cost me 1% In another 3%. So, for 3%, I should yes frequently, that’s a goal that you need to have higher residual risk is is pretty high. That’s it and then of course, if you say great thing also you can look at cross section. Now say I can compare different types of cheese I can compare different options. And given my forecast I actually can see I can sport virus my optimization given all this input, so why I think it’s a good thing then it’s very intuitive. And it’s actually also intuitively what I like say for each option chain, you can compute what is optimal say kg and frequency and what is my expected Sharpe. Actually what you can see is sort of equilibrium risk reward analysis. That in terms of if you look into sale options, it will be something between two and three months. That is from risk reward is the best part of option chain because longer dated options. They have too much risk. Your risk volatility is typically very high. On the short end, also the opposite, not the opposite of longer dated, volatility is high because of implied volatility on short, dated, is high because gamma risk. And if you look in so say, options are different strikes, actually volatility out of the money puts is higher than what you have the money the gap is. So if you actually account for all these effects, not only you can derive your optimal frequency, but you can also derive what is where I should look in terms of what changes I should trade or what maturities were teeners. So, so exempt is very good. And I know that some people apply this, we use it out for relative value. So,

Corey Hoffstein  30:51

when I read the paper, there’s this and I know you were going to jump away ahead now and go off track here. But I know you currently do a lot of work on trend following and one of my comments to you, when I read the paper was there’s this natural connection between option theory and trend following were sort of the Delta hedge of being short, put in a call replicates being sort of long and put in a call. And that delta hedging strategy is almost identical to a very naive trend following strategy. It strikes me that there’s this interesting relationship potentially, between this problem of delta hedging discreetly, and adopting your trend following strategies discreetly. I wanted to get your thoughts on, is there a connection there, you’re talking about the most profitable part of the curve for trading options was sort of your two to three months out? Is there any connection that can be made to the trend following space, and what that might imply around frequency of looking at trend following signals, which is, in many ways very similar to frequency of delta hedging your options position

Artur Sepp  32:01

is indeed, there’s a deep connection, I was not the first one to discover it was CFM. I think there are a couple of good papers from CFM. But in fact, it’s true. If you say, on one hand, if you buy options, so in options pay, if you buy options, you pay implied volatility, if you didn’t take care of this options, you get realized volatility, your net p&l is a spread between realized volatility and deployed voltage. So in a sense, to make money, you need a big volatility where say a lot of things happens. But interestingly, in volatility service, you’re linked to the strike. So the only way to make in volatility you need to Delta hedge. And you need to have big gamma. Otherwise, for example, if you bought at the money put, and something bad happened, and it dropped suddenly. So even though you made Manyata initial gap, then after that, you need to cover all your data because there is no gamma see, it went deeply out of the money. So you’re not making any money on what happens next. If you’re strictly data hit, and this option train is the same, the option to understand the options trading. So imagine it’s again, maybe it will be a little bit technical. But imagine we sold the call option, or we bought a call option, and we need to dedicate it. And as a money Delta hedge, I’d say options as the money. So we need to sell underlying underlying goes up, our value of our call increases, which will be say, which compensated by loss on our delta, our negative Delta. But now say, we’re sure that say, stock goes up to say initial price was 100. And it goes 10% up to 110. The date of call increases, so now it will be 60. So initially, it was 50% that we needed to be short. It now it’s 60%. So we need actually extra sale it again. But we let it run. So in the fact that we were letting our detrimental to just integrity time. So our data will grow. So we want the the neutral. So in any subsequent rise, our p&l will increase. So it’s like a ladder, every time we get bigger and bigger exposure. Similarly in Gen Z, what we get is the final price say stock went to 200 and we hit 50% at 100. So we made 150 That will be well compensated by by our cost, strictly speaking We won’t be trading volatility, we will trade underline, and restraints only the same. So imagine at the beginning, we started with Delta, we say our position is zero, stock when 10% up to 110. We opened up, say 10%. And now we’re stuck again, it continued to go up or say 10 bucks more to 120, we have a little bit of female recruit on our 10%. So around, say two bucks, but then we increase so we think, Ah, okay, it started to go up. Now we increase our train signal, say to 20%, and so on. So it’s, again, we’re trying to build up a step function, the higher it goes, the more we accumulate, and each subsequent move will produce a healthy meal. In other way, mathematically, we create a parabola every time like we try to build up our position in the step function. And if everything goes right in our direction, right, each step will be multiplied by increments of x. So in the end, we get spotted return. So that this is some reason I find that is, and this exactly as you said, in a special conditions, when there is like traitor Street is exactly equivalent being say, Cajun options with a little bit of say, delta with skipping a little bit of Delta, and building a trend following where you’re not trying to be with a binary system you’re trying to leverage the more it goes up, the more you try to leverage and then in this circumstances, you get exactly the same profiles. Nevertheless, it is what is interesting in options, so now of course, nothing is ideal in Tracfone there will be dips up and down so we will be spending on rebalancing. In option space, you rebalance your cost or implied volatility in Tracfone, is this like, say volatility of this small term dips that you need to rebalance? And what typically occurs what empirical evidence is that implied volatilities typically is more expensive than the cost of short term range. And for CTA,

Corey Hoffstein  37:29

I want to make a bit of a conversational jump here because I know your career took a big switch in 2015. Moving from the banks to you move over to Julius Baer’s advisory Solutions Group in Switzerland, really to serve as an advocate for alternative risk premia. And this was unlike a lot of your desk work where you had been working on models and pricing structures. This was a largely outward facing role for you, where you had to communicate a lot of these somewhat difficult quantitative concepts. And I was hoping you could spend some time talking about some of the lessons you learned and some of the tools you put in your toolkit to help better describe these quant topics. And then topics of alternative risk premia to some of the advisory solutions clients you were working with.

Artur Sepp  38:23

So yeah, in chapter 15, I think I left I think it was one of the it was great environment, one of the best people that I met through professional career. But in the end, as we discuss over the regulation Kyoto investment back in, so it’s just people started to move out. Natural it is, if say you have money and you have connections to open up a hedge fund, if you don’t have deep pockets, they don’t have connections, you try to go to asset management. So for me it was when I noticed opportunities. Where I thought it was good is, say opportunities or the genre. Of course, culturally speaking, it’s different cultures, I say. It’s more conservative first people are very concerned with you. They don’t trust quants at all. It was actually I discovered it was a mistake to talk about modeling hotels that you are one to one strategist. So I try to reduce this as neat as possible. On the other hand, there’s definitely demand say that Tran to the hero what the centrality of is female factor investing, like volatility. I think they use a lot of fields. Like good example with short volatility, a lot of our clients who started to ask it into 70 right it was a lot of demand actually a lot of questions. So in a sense, people are more say performance chasing Don’t like technical details, if you start talking about like, say kind of freeze budgeting is too difficult. So I think for me, the top experience that alone never shows the main equation always illustrate with provides him intuition. So intuition would be more like say, visual, have a good visual sense that makes sense. So say a lot of what we talk and a lot of these conditional details, all the stuff, it was actually through those years that it’s a relatively simple to screen in one figure, what happens in bear market Visio portfolio, what happens in bull markets, and what happens in normal. So, in this sense, you can communicate and you can communicate quite complicated stuff, you can say, for sure volatility, because a lot of the stuff is so cyclical, that they see that short volatility return into your 17 return was 100%. But then if you show that there is a good be another regime, you can do some sort of one. And then in this way is that period. Some people who have time series, which is time series will hide the compounding effect, which you understand time series will. Of course, if you have a drawdown in Xanthi, recovers the time to review K somehow, it will hide the compounding effect that you stated also a lot of people what I’m always trying to tell them in risky products, you get your say long term, expected rate of return only if you stay in it. Only if you think of market timing, forget is not for you never try to say do market timing invoice again, what we started with illiquid stuff. Therefore it’s important on portfolio level. If you want to do a risky trade, make sure that your portfolio is well suited if say, for example, if you’re long stocks or mutual funds, don’t sell volatility if you create NASA, to your portfolio, NASA. And usually when people say get hit, they always close the seal. It’s easily the most recent stuff, or the scenes, they don’t understand it most of the time, they do exactly at the wrong time. And I saw like many examples people know just stick to it, just sell delta or structure products or these type of strategies, you lose much more if you liquidate at the wrong time with delta one products, okay, you can say with you can buy, say in one month, you won’t lose much. But then you need to communicate and then you need to have clear say visual tools explaining all this risk premia. And I think for me, the bottom line communication aspect is important in communication. Last few if not using formulas, you are not allowed to use formula. Otherwise, they think that you’re either don’t understand, or you’re trying to hide something, you need to

Corey Hoffstein  43:29

create a story that is appealing. It’s sort of a two fold problem from a conversational aspect. And you hinted at both of them, right? There’s the conversational problem of what does this strategy do? And how do I develop intuition about this strategy. And then there’s the conversational problem of not only how the strategy behaves, but how it behaves relative to an existing portfolio, a client might have that these things don’t exist in isolation, they exist relative to what else the clients already allocated to. And you need to get an understanding of how it’s going to fit in a lot of your writing and recent presentations. You reference this conditional beta model, and you started to hint at it a little more, you’re talking about how to things do in different regimes, bull markets, bear markets, and I know that you found that to be a very effective way to communicate these concepts. Can you talk a little bit about how that model works and why you like it both from an analytical as well as a conversational perspective?

Artur Sepp  44:29

Well, so so first I history, let’s say I went to the cycle in the cycle yo was thinking that the world becomes two dimensional, you either survive or not. So these kinds of binary outcomes, I think, more or less, they’re important to think of this. And then of course, you have if you model if you think in this way that there will be periods when risk aversion is sky and Piros when it’s moderate, and of course, there are I’m like, I worked on more complicated models like how to say implicitly model this and, but in the end, what we want to do is some kind of classification. If you look at in sample analysis, I want you to tell that okay, statistically, these periods correspond to what I call beer regime. These periods is normal regime where there is mood issue. And what happens with these more models, there are more as I said, there are more complicated, more advanced models, statistical models to do it, to identify these regimes. But then it’s again, you don’t want to people may think that you’re like, inventing something, or you’re doing like data mining, here is a simple example, you just look at a standard deviation contest. So statistically, you know, that say, standard deviation quantiles is bottom 16%. Observations below the 16% percentile corresponds to realisations say returns that are below one standard deviation. And then opposite, say MoodGYM with 84%. quantiles corresponds to realizations that about one standard deviation advantage of this framework. Also, I said about one standard deviation. I didn’t specify what period. So effectively, we can apply this model for any period, we can look at weekly, monthly, quarterly annual returns. So we have a very simple sample realization. And as I said, a lot of stuff is conditional conditional. These regimes say the simplest examples, say it’s selling goods. Of course, in battery dreams, CBOE has a very simple strategy to index which sells as the money puts. So initially, there are 50 Delta. Of course, a, during a bad quarter, when s&p goes down, say, average probably 10%, Your food will be in Somalia. So initially, it was 50%, Delta, Delta was point five in there, that was one meter was one. So you increase it conditionally, that we went to bear market, you have large exposure in your short putt. And in a way, this now it’s appealing, you can say, Look, this quarter, it corresponds to say, not outlined, but it corresponds to a regime that is outside of the normal range that we expected 6% annual sorry, over say, longer term horizon, we expect 6% of observation to become to be a ratio, which also corresponds to more or less what we observe, say empirically, if you use maybe more say, intuitive, like economic regimes or different, maybe some moving averages of stuff like that. So in St. I found this model is it’s intuitive. In fact, it’s again, is it simply explanatory model. So, of course, the great advantage. What is also a great advantage is that stationarity conditional on these betas, says a strategy that sells at the money puts, will more or less half always be the one in any better resume. Of course, we then know maybe s&p goes down 10%, maybe it’s 20, maybe 30%. But in which one, if we have this model, we can project we can project what would be the actual p&l,

Corey Hoffstein  48:41

I find the model very appealing, having looked at a couple of times, both visually, it develops a lot of intuition. And it makes a good deal of sense for most people who have operated in markets, not only when you’re talking about exposures that have nonlinear payoffs, you can very clearly map those nonlinear payoffs. But a lot of market portfolios have changing correlation profiles in different market regimes. And there’s been plenty of ink spilled about how you can get crashing correlations when equity markets start to decline. Or to the opposite side, you can get flight to safety characteristics in different asset classes, depending on what’s happening with equity markets. So I find it a very intuitive and appealing visualization of some of these ideas. One critique might be it does inherently rely on the realization of past information. And when you’re talking about both bull and bear regimes, almost by definition, when you’re looking at a rare event, you have less data to work with. How do you respond to the potential critique that while you’re looking at an environment where there’s perhaps not enough data for you to develop an intuition or a model, or we haven’t seen a realization that’s as bad as it could be? And therefore the model understates the potential risk associated with that regime.

Artur Sepp  50:00

So I think when people talk to this, there are a lot of different scenes going on. So first is Taylor realisation. So for example, crash of 1987 in October, I believe went down minus 23%. This realization is so busy from TableView. Is it for castable? No, the answer is no, it gives you a point in, say distribution of returns, it gives you one point. Now, if we talk of regimes, actually, what I refer to is more say, period. If we look into time series, and we look into periods, there are obviously periods of high volatility and strongly negative returns. So, and my goal is not accurate to more what can happen. So, within this period, of course, within this period, you can have very much positive negative returns. What I’m after to describe, both since I want you to describe the frequency of these regimes, and average distributions within this regime. So, in a fog, what I’m after is not really to describe the whole potential outcomes, that can be events like rash, that has inherent uncertainty over realizations that potentially can happen. I’m more after identification of erodes when St. lightsheer in absolute value, are more likely to talk to your daddy dotting N. Importantly, if I can identify these museums that I can forecast, then it costs implication for my portfolio minute, probably if I believe that we are due to go into periods of high volatility regime, I would cut my meetings for I would cut my volatilities for, I would build the risk management around scenes that are, say, identifiable that at least I’m not making go to describe all the possible tail events, what we can do is to describe, say, frequency of these periods of larger statistical returns, and then plan accordingly. We need to have some history. But we are not saying that this is exclusive analysis that it gives you all the potential ranges, or what things can happen.

Corey Hoffstein  52:46

And I think one of the appealing things about a lot of the alternative risk premia that people are evaluating is the idea that they potentially can be hopefully uncorrelated in those negative equity regimes. I know that’s one of the reasons a lot of people initially adopted a lot of trend following strategies was the empirical evidence in a market environment like 2008, that it could potentially not only be uncorrelated, but even negatively correlated and create profit in those types of environments. You have since left Julius Baer, you work at quantica capital now, which offers trend following mandates. As you look at trend following as a space may let’s start off with is trend following. does it fall under the category of alternative risk premia as you look at it? Or do you think it’s sort of categorically unique in what it is and the type of strategy and p&l that it creates?

Artur Sepp  53:48

Well, we published an article in the Journal of hedge funds, which is actually trying to answer the same question. In my opinion, alternative risk premia, it always has to do with liquidity. So, there are different categories, but says the most basic one is risk seeking strategies where you you can sell say boots, you can dedicate them, you can say, Buy cash bonds, you can sell credit protection, in the sense that you are swapping that most of the time you will be making money. Most of say traditional risk premia may be a part of a factor investment trial, but the most typical say carry volatility created all this time, they make you money 80% of time. And then 20% You can lose a lot. So, and then it derives. The best way to classify assess from statistical perspective is in actually to see again A token of our resume model. So we’re not trying to model the worst possible outcomes of these risk premia. We just look at how they behave in bearish regimes. And that’s exactly what we see they have zero correlation to stock market in normal or injurious regime. They also they benefit they, because of compression spread compression in the regime, they suffer because of spread, widening, spread, widening and delta widen. So in a way, given my experience, I always thought that for me, risk premia assumptions that you swap, say, it’s like selling insurances, you swap a stream of cash flow for sudden gaps. And in a sales day, then from this perspective, cities are different, say traditional cities that only tried to capture trends. In fact, it’s an opposite the strategy that it doesn’t perform well most of the time. But what we’re after is we are trying to capture big changes, and therefore, it inapposite So, when it seems to happen in market, and when things don’t happen, we are not making money, but we are not losing that much. And actually if since happens, and one also what we discussed before Tracfone benefits greatly from our auto correlated market, when the margins as a market went down so much that there is a feedback effect that people face margin calls leverage place praise margin calls, they need to sell sell sell. In this type of environment, it creates conditional predictability, where trend followed is keeps increasing the exposure is a falling market. And therefore it gets this kind of quadratic leverage return. And when we look at say how telphone performs in bearish regime, we actually see negative correlation to the stock market saying negative betas. So it’s a defensive strategy. And in this say classification, for meats is not alternative risk premia also, if you look at cross section of funds, tenfold in this a different is very difficult say to stay over since again, at my feeling that a lot of peers, they miss it somehow whiskery trade, whiskery is or through options or through cures, and then you lose on this defensive, you perform better, but you become a mix between alternative risk premia and trend follower.

Corey Hoffstein  57:56

The introduction of carry signals into trend is certainly something I’ve seen in looking at the trends space over the last couple of years, strikes me as potentially something folks are doing as a way to help protect their portfolios better from some of the negative performance that trend has seen, right? I think when you look at trends, strategies, they did incredibly well in 2008. They struggled for a little while when oil prices sold off, and there was a bit of a commodity crash trend strategies seem to do really well again. And then since I’ve been sort of bleeding a little bit of money. I think there’s been a lot of people trying to explain, maybe why trend hasn’t done as well. In the last decade. I’ve seen some people saying that the inflow of capital has caused sort of the broad index when you look at trend followers as an index has to slow down their signals. So prior to 2008, trend followers tended to be a lot faster in their signals. Now post with the influx of capital they’ve had to slow down to prevent higher trading costs. AQR actually recently published a piece talking about that it’s really just been that magnitude, the absolute size of trends. And the post crisis period has largely been subdued compared to the pre crisis period. I wanted to get your perspective, obviously, you work at a trend following firm. Now, my guess is you don’t think trend following is inherently broken. But do you think that there have been significant changes in the market environment that have caused trend following to no longer work? And do you think that there’s a risk that there has been an inherent shift in market dynamics that will prevent trend following from working in the future?

Artur Sepp  59:31

So since a contract of strategy, we need to always understand the conditions under what conditions trend for will do work. And the condition is simple, as I said, simple say advanced and follow you always we are trying to leverage continuous trains. So since it goes down, we leverage it. We are just not sticking to the positions. Even if we were to use binary signals. Imagine that If see start to go down more and more assets, say treasuries start going up, dollar up, commodities down. So even we use some kind of crossovers binary crossovers, we start accumulating the short exposure. And in this sense, what conditions do we need to perform when he tries to continue. In other words, we need positive autocorrelation. If sins go down today, it means probabilistically, they are more likely to go down tomorrow, or on the way up. It is not enough for this to go up. Like in my recent presentation, I offered the simple market that number one is dynamics where you say, you toss a faulty coin, where probability of heads is say two swords. And for each hit, you get $1. So you are bias to go up if you’re alone on the VISTA. And you say, by is this, you play this game, you expect it to go up your p&l trend for a row. So in this environment, of course, most of it’s a trend following the ones that who looks at the most recent outcome. If say today was tail, I bet on the table next round, in the sense that we’re trying to follow the most recent moves. And of course, if coin is biased, we also were capturing it a little bit not strongly not as good as the one on the register. What matters for us is if the coin is biased in a way that if it shows here to allow, on the next round, it conditionally will say to sort of probability of Thrones, the hit again, if it shows tale now, then conditional here to source will detail tomorrow. In TV, just a trend for the weights of the tale. So conditionally still realistic, if you can be negative for us. But in this way, for all non investor in this game, you get zero in this condition or game because then the expected value is zero. But actually for Tracfone, we get the benefit. So to paraphrase, trend, following conditions for trend following to perform is when markets are auto correlated, in a positive way that trends are continuous. It is not enough just to seems to go up, it should be conditionally that they go up and what happens during the last decade, since go up but for a different reason. It’s just a dream. It’s a dream that is not say conditionally not predictable. And moreover, we have gaps. So, information disseminates very fast that we have gaps over very short periods of times that no trend system is able to capture. And we have recovery in a normal say environment where trend phone will benefit if these gaps continue to go down. Now, what comes up as it they every deep is both and when deep is both it means there is a mean reversion. And mean reversion is environment where Printful mathematically we can show of course, the answer is not such as it’s satisfactory for a quant for a person, maybe for a quadro for person who understands the game is not satisfactory for investors. So now what causes then we need to look into what causes this mean reversion. Of course, people refer to central bank policy, which is somewhat true that I think there’s some way complacency is that any gift needs to be bought. Then on the other can say in commodities market. I think also there’s more money arbitrage wars. Now, a lot of commodity producers have their own trading desks that are trying to eliminate any kind of rare opportunity or any type of mismatch between different contracts. Traditionally, maybe trend forming was much more diversified towards having some offers in many markets across many markets, offers that we all see uncorrelated. It’s either part of positive autocorrelation or part of carrier. Nowadays they don’t exist. So to answer what’s next, and I think the industry as you rightly mentioned, I think most of the guys they just write you in to crease the window, so less trade less frequently in tray in this, and then question comes how I’m able to adapt to a fast changing environment? If since the start changing again, am I able to go transition to fast and forward? So that’s a question that investors need to ask

Corey Hoffstein  1:05:25

what I think is really interesting about the whole moving more slowly is if you take it to the extreme, if you take a Managed futures strategy, and you say, let me use really, really, really slow signals. In fact, they’re so slow, they’re always long, you’ve effectively created risk parity. So on one end, you have risk parity. And in that case, you would say, well, it doesn’t necessarily always make sense for me to be long, all these commodity contracts, I would want to look into the carry, it might make sense to be long equities and long rates, but it might not make sense to be long commodities all the time. So in that sense, carry might make sense. But as you start to come down the curve a little bit in the speed sense, as you start to have trends that are actually going to make trades, that whole carry aspect can create potentially conflicting signals. So it’s this interesting aspect of where you fall sort of creates this barbell of how much you start to approach this risk parity type, extreme. But you are seeing folks like Winton, for example, I think is a perfect example, who are saying, Look, we’re throwing in the towel a little bit on trend, we’re looking for new signals, new strategies, they still are going to have trend as a core component, but reduce its allocations and some of their core funds. As you look towards the future, where do you think it lies for trend following managers? Is it simply that they need to start looking for other signals? Is it should they be looking to introduce things like carry to make a trend following strategy more robust? Or is there still benefit in the purity of a trend only mandate?

Artur Sepp  1:07:00

Well, I think in terms of say, point of view of Manager, you can do two ways. I think, more or less this crisis Alpha. It’s already people don’t believe so it’s better not use it anyways. So what Vinton is doing, they’re trying to do become just multi risk premia. And of course, you can do like lon short. And t what is great thing about trend following we’ll protect you in certain circumstances, there is cash benefits, tenfold are very big, especially on portfolio of alternative. So just to step out why. So, if you can think your way, you probably know what principal component analysis says, say in market we have 1000s of instruments, what we see tells you that effectively out of the south factors, you can use some statistical analysis. And at certain periods of time, one or two first principal components will explain variability of all your 1000s of instruments. And these will be periods like your weight and trend following works perfectly fine in those conditions. Because there is no need to diversify, you’re just go long risk of his illness. Well over the past decade, in more normal markets, secondary components dominate and secondary components they have more predictable statistical dynamics like mean reversion and maybe some training in what is long short or cross sectional trade time series is just trying to capture this extra effect. So in effect, you are becoming more sort of closer to risk premia. You still say example of Intel, they just try to, they know that most likely so you cannot say rely on that trend follower makes you breach performance that you need from a manager point of view. So, you go more into risk premia you diversify, even keeping your trend signal alive. So, on the other hand, your question was, if your trend following can exist, or you become more like a risk parity, I think the longer you make your window, the closer you become to risk parity. In the fog, you’re trying to be lonely, everything. You just try to balance the ways in these ways is risk parity is again, it’s a bad concept. Probably people also don’t like it right now. Fewer trend following. I think the case for models in my perspective, one of the problems are talking of this PCA analysis In turn phone signals are very coveted. If you make money, you make money across all instruments, almost everything if you lose you lose a consultant Smith, there is no say signal diversification. So, I think the next step what most people are trying to do is to try to diversify signals in a way that you may be to work with signals that on their own, they are not say exploitable, they can have Sharpe of say, maybe point two or point three that you cannot do them on a standalone basis. But if the signals are uncorrelated, and especially to non correlated, by building a portfolio of signals, you can get much better Sharpe. So, the goal is more say there was occation of signal. This brings us to systematic macro, that to diversify signals, you need to use more data, not only price data, and what else drives the markets apart from prices? Its economy, right? It’s economic data is one of the factors that determines the policy market reaction. And therefore, I believe there will be evolution of trend following is your more say market driven? Where do you stretch your view, create independent signals. That in a sense, and I know some good systematic pattern of funds that had very good year to wait. Because in the end, if you have big trends, all your macro trends, they will converge into one into the same as trend following the user wants to be short equities or $1, or long treasuries, or short commodity insane tails of your signal, which I think is quite exciting. But I think going forward, I think that there is still the industry will evolve. And there’s a lot of say interesting work to be done on quantitative side.

Corey Hoffstein  1:12:06

Last question for you here, Artur. And it’s one I’m asking everyone in this season plus question. And the question is this, if you had to sell all of your investable assets today, so you had to liquidate all your investable holdings and could only buy one thing for the rest of your life, whether it’s an investment strategy, or individual asset class is up to you. What would you buy? And why?

Artur Sepp  1:12:35

I would buy a piece of flood this house, I would have said that piece of land somewhere in mountains. Is it in Switzerland? Or actually in Catalonia? My wife is Catalan. And the answer is simple. I mean, they both kind of go, right. If I go for my piece of work my house, I can always go there. I can not ever retire, I can lose my job. Or I can get rid of the scenes. If it can’t say nice environment like mountains, river, whatever lake I can spend the rest of my life. Find it that

Corey Hoffstein  1:13:14

sounds like a good choice to me. Well, Archer, this has been a lot of fun, deep dive on a lot of topics. Very fascinating stuff. If people want to find more of your research. I know you’ve started publishing more a couple of your papers are in SSRN you get in a little bit more active on Twitter, but where can people find you

Artur Sepp  1:13:30

out of those tip.com I try to keep a regular blog, and also people can find all my research and my presentations that are

Corey Hoffstein  1:13:39

wonderful. Well, thank you for joining me today.

Artur Sepp  1:13:42

Thank you. Thank you a lot for your good questions. was nice talking to you.