In this episode I speak with the anonymous twitter user @macrocephalopod.

The arc of our conversation follows the arc of his career: beginning with slow-frequency style premia in a hedge fund to building a prop desk that trades mid-to-high frequency strategies in crypto.

A large part of the conversation can be characterized as comparing and contrasting the roles through the lenses of research, operations, and risk management.  For example, in what ways is long/short equity meaningfully different than long/short crypto?  Or, how important are topics like market impact, fill ratios, and borrow fails in mid- versus slow frequency strategies?

While crypto is the venue, I believe the wisdom imparted in this episode spans all markets.

Please enjoy my conversation with @macrocephalopod.



All right. Are you ready?

Macrocephalopod  00:01

Yeah, absolutely.

Corey Hoffstein  00:02

All right 321 Let’s do it. 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  00:22

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 new found 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 in securities discussed in this podcast for more information is it think

Corey Hoffstein  00:53

In this episode, I speak with the anonymous Twitter user macro cephalopod. The arc of our conversation follows the arc of his career, beginning with slow frequency style premium and a hedge fund to building a prop desk that trades mid to high frequency strategies in crypto. A large part of the conversation can be characterized as comparing and contrasting the roles through the lenses of research, operations and risk management. For example, in what ways is long short equity meaningfully different than long short crypto. Or how important are topics like market impact fill ratios and borrow fails in nted versus slow frequency strategies? While crypto is the venue, I believe the wisdom imparted in this episode spans all markets. Please enjoy my conversation with macro cephalopod. Macro cephalopod.

Macrocephalopod  01:50

Hi, Corey, great to be on the podcast.

Corey Hoffstein  01:52

Thank you for coming really excited to have this one. It’s gonna be a fun conversation, though. I have to dance a little nimbly here to make sure that we’re we’re keeping you anonymous as you like to be. So I can’t do the usual what’s your background starter? So let’s start with something a little different. What made you join Twitter? And if you can disclose where did the macro cephalopod handle come from?

Macrocephalopod  02:17

Sure. So I joined Twitter in late 2020. But obviously, I’ve been using Twitter for a long time before that website had a real name account from very shortly after Twitter was first founded and you could make an account. And I use that for a very long time, mostly to look very occasionally to ask a question or someone. And then during 2020 during the pandemic, obviously, there was a lot of work from home, I found myself with more time on my hands and wanted some entertainment. And also more and more felt like there were people on Twitter who I respect whose opinion I value, I want to be able to speak to them more discussed ideas that I had, especially ideas around finance, but also general banter. So because of my job where I work, I felt the best way to do that was with an anonymous account. Obviously, you can see that there are many, many people on Twitter using anonymous accounts in finance for a variety of different reasons. But for me, it was mainly that I wanted to be able to speak a little bit more freely than I thought I’d be able to, under my real name because of my job. So I created an account, started posting a little bit of finance content, trading content, quantitative finance, which a few people seem to find interesting. And yeah, it went from there found a really great community of people there. A lot of people posting really interesting content and who seemed to be at least half the time interested in what I had to say as well. And the user name, there’s no great story behind it. I saw a lot of people had animal themed accounts, I thought it was great. What animals grow like octopuses are kinda cool. And macro cephalopod. Well, I was trading mostly futures and currencies at the time. So I thought I’d stick a macro in front of it as cephalopod was already taken. So sadly, no exciting story for you about handle.

Corey Hoffstein  04:09

Well, I understand it at least. So let’s go pre what you were doing. When you launched the Twitter account? You’re talking about trading some futures? We’ll get to that. I know earlier in the 2000 10s. From your background, you were actually implementing longer frequency futures strategies, a lot of them in the vein of style premium. In our prior conversations, one of the things you said was that looking back, they were very naive strategies, and you’re surprised that they worked in the first place. Why do you think they worked? And why do you think the Alpha has decayed so much since then?

Macrocephalopod  04:48

Yeah, good question. So I guess I should clarify that. We did make money with these strategies, and the kinds of strategies I’m talking about are primarily daily strategies. So you rebalance your portfolio once per day, you typically hold it for weeks or months at a time, you’re mostly trading very liquid futures and currency markets, stock indices, bonds and interest rates, commodities. And you are not doing a whole ton of modeling around the details and the microstructure of the market, to some extent, not thinking about it that much. So why did some of this stuff work? And I can give a little flavor of the kinds of strategies that I was thinking about. So CTA trend following was one example. I spent quite a bit of my early career as a quant researcher working on that kind of strategy, carry strategies, strategies, which now people like to call macro momentum or economic momentum. So looking at the cadence of economic data releases, and whether they’re surprising to the upside of the downside, and trying to build long, short portfolios around these ideas. So why do I think they worked in the first place? Well, you know, there’s a big academic literature on these kinds of strategies. I think there’s pretty compelling evidence that, at least over the long term, if you’re looking back 30 years, they have worked, at least in backtest. And there’s also quite a long out of sample period of people live trading these strategies, and they have made money. Look at many, many CPAs with 30 year track records, you can look at futures trading firms, AQR Winton who’ve been around a long time and making money with these kinds of strategies. And I think they were for two reasons, one of which is that many of them have a genuine risk premia. There, you are being compensated for taking some kind of risk that other people don’t like to take. The classic example of this is a carry strategy, where you are just subject to big left tail crashes in essentially every asset class where you’re trying to run carry, particularly in effects, but definitely in race as well. But also, I think, historically, these strategies had a big component of inefficiency in them as well. And that inefficiency came about, because there were large barriers to access for the strategies. So it was much more difficult in the past, to just get the data, you needed to even do the research to see if these kinds of systematic strategies worked. Even if you could get the data, it was more difficult to access the market. And there were no knife wrappers for non institutional players to access the strategies. If you’re an individual who wanted to access some momentum strategy, you basically have zero options, but you can either try and do it yourself, which is hard. And that was it, there was no ETF, which wrapped up on momentum strategy for you or give you exposure to value or something like that. And because of those market access difficulties, it meant that people were not paying as close attention, as they should have been to the fact that if you just went long currencies with high interest rates and shorter currencies with low interest rates, then you made money with a sharp of like one and a half to two for 10 to 15 year period, which really only came to an end during the financial crisis in 2008. And this kind of explains why the Australia has performed less well now. I think like now there is much better access to data access to markets, these strategies now have the kind of returns which are commensurate with the risks that are being taken. And the inefficiency component has gone away a little. And so I think we were a little lucky, we may be caught the tail end of some of the period where there was still an inefficiency here. But it also coincided with this big explosion of whatever you want to call it smart betta liquid alternatives around 2016 17. Suddenly, a lot more capital came flooding into these strategies and competed away a lot of the inefficiencies and what you’re left with is, is the risk premia a risk premium good. A lot of people should invest in this premium and should probably have more risk premium exposure than they currently have. Yeah, there is a notable decline, I think in in the profitability of these strategies over the last 10 to 15 years.

Corey Hoffstein  08:56

So I want to contrast that though, with another comment you made to me, which was that some of the longer term strategies in equities still seem to work, despite the fact that maybe they’re parallel concept strategy applied in the futures space doesn’t. I’m curious as to why you think they’ve maintained their edge in equities, but not in futures.

Macrocephalopod  09:22

Yeah, so I guess the first thing is, is that statement true at all. And it’s very difficult to make any concrete claims that is true or not true. But if you look at what big multi strategy hedge funds are doing, they are generally putting more of their capital or their risk to work in equity markets than they are in futures markets, implying they think there’s a great opportunity there. And you know, if you’re in the position to have access to data and run simulations and run back tests, you can see that the performance has held up for some of these market neutral quant strategies in equities more than it has in the macro asset classes and And part of that, I think, is that equity trading is a technically more difficult problem than futures trading. So if you want to do it at scale, you need access to leverage, which means you need financing and prime brokerage, you need to be able to do stuff on Swat. And you need to be able to handle executing, you know, what is a pretty complex dispersed market, especially in the US with futures trading, if what you’re trying to do is take some positions and hold them for a few weeks or months. That is much more straightforward than an equities. But then another reason, I think, is that there’s a much broader range of names to trade, and equities, so many more data points. And also, it’s generally much more expensive to trade in equities, no one who elaborate on those a little bit. So if you’ve got a market, which has a small number of names, let’s talk about currencies like jute and currencies, you’ve basically got nine pairs, you can trade, and it’s very cheap to trade. So typically spreads of less than a basis point to try to use and currencies, and there’s a lot of volume, which there isn’t essentially every currency pair, then you’re really never going to expect to see a lot of alpha in that market, because it’s so easy to access it and you can put on so much size, and it’s so cheap to trade that anyone who can find alpha is going to deploy as much capacity into that as they can, and it’s going to compete away the advantages. Whereas equities, you know, typically, for at least the mid and small cap segment of the market, multiple basis points, bid offer spread, generally, really low volumes, except for a handful of the biggest names, maybe something trade 1% of his market cap per day. Most s&p 500 futures like probably trade multiples of the open interest every day. And this combination of lower liquidity and higher transaction costs means that and this is kind of subtle, you can find good edges in equities, if you temporarily suspend reality and get to ignore trading costs. And you can find really strong, persistent predictive edges, where you can predict how the price of some stock is going to move. And because there’s such a broad universe of names, you’ve got a lot of data you can use to validate this, you can really do stuff like taking half your universe, doing your research there, and then saying Does this still hold and the other half of the universe in something that you really can’t do, if you’re trading currencies, see, if a lot more tools to prevent yourself from finding fake effects, like strategies, which don’t really work, they’re just data mined, and you’ve got this kind of cushion of the trading cost is very high, so you can more easily find predictability, the game in equities is to then monetize that predictability. So you know, it’s not enough to say, Great, I can predict that this stock is going to move on an average of five basis points, whenever x happens, you actually, if that stock cost you six basis points to trade, then there’s no p&l to be had there. But by combining lots of lots of these weak edges, which maybe only predict within the bid offer spread, you can get something which looks good after transaction costs in equities. Whereas in currencies, you’re weak edges. If you’re trying to predict something over the course of weeks or months, they’re very, very weak indeed, and they’re so weak that actually, it’s it becomes much easier to fool yourself and just find defects that are not really that

Corey Hoffstein  13:13

this conversation starts to lean into the realm of simulations and research and testing, right, that concept of transaction costs and equities being a key feature as to maybe the limits of arbitrage as to why some predictability continues to exist and persist. This is an area I know you’ve thought a lot about not just transaction costs, but just back testing and research and simulation in general, you’ve written a lot about it on Twitter, even did a whole series called the 24 days of back tests errors, which was a great series, I highly recommend people look it up. My guess is we could do an entire podcast just about back testing. Maybe we’ll do that certainly not next season. Given how much you’ve written about it, presuming you think it is a useful exercise in the first place? How do you think it’s best employed within the research process? And where do you think most folks go wrong?

Macrocephalopod  14:08

Yeah, absolutely. I always do think back testing is a useful tool. And I guess what I would say is that, I don’t think of back testing as a research tool. So much like it’s not something which is important to me in the research process. For me, the research is what you do before you run the back test. And then you do the back test to validate the idea and get some ideas about well, what is performance gonna look like? What a drawdown is going to look like? Maybe you want to test some different portfolio construction ideas and use a back test to do that. But the majority of the research that I do and then I think a lot of people working professionally in this industry do is that you spend a lot of time looking at data and building models and testing hypotheses and trying to understand effects before you even think about simulating at all because the first question is, is there something predictable here? And you can go a long way with buildings or features and doing some linear regressions and getting a prediction out and saying, Great, what is the correlation of my prediction with the return over the next two days? And as long as you do those things? Well, there are a million ways to mess up doing a linear regression. Before we even get on to how can you mess things up if you use some very modern nonlinear machine learning tool, but you can do a lot before you get to the back test, including which features are going to be useful? Yeah, once I build a feature, do I need to transform that feature? Before it’s useful, then a prediction? How do I combine multiple features into a final Alpha? How do I make forecasts or volatilities and correlations of the assets that I’m trading? What are sensible limits to have on position sizes or market beta exposure, all this stuff you can do before you ever think about your back test. And then once you’ve got that, you then take all these inputs that you’ve created your alphas your volatility, forecast, your your base model, your credit constraints, and then you put those through a back test to say, Okay, what kind of performance? Can I expect with this model, but I’ve created, your back test is always going to be optimistic, right? Because your back test has a sharp have to you absolutely do not expect to have a sharp or two, when you start reading this thing. Apart from all the data mining, you’ve done, you’ve probably miss modelled a lot of things in your back test. And almost always, these errors go against you. But it’s very rare that you make an error in your back test. And then it turns out when you start trading, you do better than you expected. So the kinds of things I would use a back test for, like I say this final check that the strategy works, given a set of assumptions, I would use it to compare different simulation assumptions. So if I have different assumptions around, what is my market impact? What am I fill rates? What is my bolo availability? How much do I pay for my bios, or what is my latency, you can use a back test to, for example, make a chart of how does my p&l vary as a function of my latency? Assuming you’ve you’ve modeled everything else correctly, you can change your latency assumption and see how latency is sensitive, your that’s something I think a backtest is useful for, or potentially, to compare two different variants of the same strategy. So two different ways of constructing your alpha, the given set of features, I build two different models for getting a final alpha out of those features. And then I run them both through the same back test. And you hope that even if your back test has some assumptions, which are not quite true, in reality, at least the delta between those two factors is going to be something which is meaningful. And then finally, a very important thing which we use back testing for which I think is a little underappreciated is reconciliation versus live trading. So at the end of every day, or week, or month, or whatever your frequency is, you run your simulated trading over the day that you just traded, and you say, how well does my back test match reality? And if there are differences, you really want to understand those differences? That’s something we use back testing for a lot.

Corey Hoffstein  17:52

Another area you’ve written quite a bit about is factor models in equity factor hedging. Can you talk a little bit about why factor hedging is so critical in a multi manager, equity hedge fund? And how these types of concepts might actually extend to hedging macro factors?

Macrocephalopod  18:12

Yeah, absolutely. I guess I’m gonna be able to take it as an assumption that your audience understands what an equity factor model is, and what equity factors are and how you talk about it a lot. So

Corey Hoffstein  18:22

we’ll assume that if you don’t understand hit pause, go read about it come back.

Macrocephalopod  18:27

Great. So the idea really of an equity factor model, though, there are two parts to it, there’s a part that helps you describe risk, and there’s a part that helps you model returns. And they’re both important. And you know, quant equity manager will quite likely have a factor model somewhere, even if their strategy is not entirely built as a factor model. There’ll be some factor assumptions or factor modeling somewhere within it. And they’ll use that both as part of the return modeling process and as part of the risk modeling process. But yeah, I had a Twitter thread, some years ago now where I talked about how this can be useful, even if you’re not a quant equity, market neutral manager. And the situation here is assume you’re a big multi manager, hedge fund. And you’ve got a large number of stock pickers who work for you. So I’m trading teams who maybe are sector specialists or country specialists or something like that. And they’re quite discretionary. There’s some discretionary Alpha you’re trying to capture with all your in house stock pickers, and they put genuine alpha, you can look at the returns or some of the big multi manager funds. And you can see there is genuine alpha that, but there’s a problem as well, which is that they tend to have, for example, biases to particular kinds of stocks. So the classic thing are that discretionary stock pickers or hedge funds, like momentum names, they like quality names, they like strong earnings, they like sales growth. I wouldn’t say they like short interest, but empirically, they tend to pick stocks that have a lot of short interest to be in their short book partly because that’s how stocks get a lot of short interest is that pods or hedge funds short them and Because of this, as the manager running the fund, you end up with a lot of incidental exposures that you don’t necessarily want to be except your currency or country exposures, market exposures, because project manager left low end devices will go long, some high beta name and just do a simple one for one equity index short against it. So yeah, we’ve actually got positive data after their rough hedge, plus all the factor exposures from momentum and earnings growth and things that maybe you want some of that, but you don’t want as much as you’re getting from the individual stock picks. And so you can put a quant model on top of this and say, What factor exposures do I have? And given my exposures and where I’d like to be, how can I find some cheap hedges for this, this exposure that I don’t want, and what I mean, when I say you don’t want it? Well, a really important thing here is that if you’ve got, say 100 different pods, you really care about how correlated they are. And to give a counter example of this secret 100 different managers building a portfolio and they’ve all got a Sharpe of one say, and they’ve all got a 20% correlation with each other because they have some similar exposures or similar stocks in their portfolio. And 20% doesn’t sound like very much. But that 20% correlation means that if you’re diversifying among those 100 managers, the best you can do is wo sharp from one to two. By doing that, this is sort of very simplistic model here. And so say you could reduce that correlation by hedging away some of the factors which are making them correlated. So you can reduce that correlation down to 10%. Then instead of being able to double the Sharpe, you can triple the Sharpe I just by getting rid of some of the correlation and then leveraging up a bit, that’s a 1.5x boost to your shop from be able to apply some smart hedging, if you can get that correlation down really low. If you can get it down to 5%, you can quadruple the shock of a single manager in the portfolio. And if you can get it down to zero, then the Sharpe your portfolio is 10x the Sharpe of the individuals around in reality, you can never get this down to zero. But this gives you an idea of even just being able to reduce a correlation between two different trading groups that you have a little bit can be a real boost to your performance. You asked the end about how does this apply to hedging macro factors? And I said, That’s difficult. Yeah. Ideally, you’d like to be able to hedge their your exposure to GDP growth or your exposure to the next inflation print. And if you try and do this in a classic Quantway, then you’ll get the time series of economic data releases and surprises and it’s okay great which stocks move a lot when CPI surprises to the upside, which stocks move a lot when GDP growth surprises to the upside and cloud computing thesis to that this is an incredibly difficult problem, just because there are so few data points. So GDP is quarterly CPI as monthly. So you’re really not getting enough data to try and estimate this. A better approach, I think, is to try and find proxies for these macro factors, which are higher frequency. So things like the US dollar, the yield on the 10 year, the price of crude oil, and look at betas to these. And this is something you can estimate with a lot more fidelity. And then you can apply similar Id try and hedge out your exposure to these factors. If you don’t want your energy managers just to be taking a lot of positive oil data. For example,

Corey Hoffstein  23:08

at one point in your career, you move from operating in a hedge fund to operating in a prop shop. We often think of those as similar worlds. But I think there are some distinct differences that I’d love for you to talk about, specifically, how does the change in the structure of a hedge fund versus a prop shop affect the incentives of the people within the firm? And how does that change? Incentives affect things like how infrastructure is built? Or what sort of research goes on?

Macrocephalopod  23:41

Yeah, so the biggest difference, I think, is that in a prop trading firm, you have more freedom in how trading Revenues can be used. So if you think of this fund is having some claims on that stream of trading revenue that comes in. So you’ve got your baseline costs, you pay rent, you have infrastructure, and data costs, have legal costs, and they’ve got to be paid for no matter what kind of firm you are. And then you have stuff, the you have your your traders who probably expect to the center of p&l or something like that. But then you have a lot of non trading staff, as well, who also need to be paid competitively, and who also expect some better to the firm’s performance. And then you’ve got the providers of capital. So in a hedge fund, that’d be the investors who expect a return on their investment, obviously, that’s why they’re invested in the first place. And then you’ve got the owners slash managers of the firm, who are trying to realize some profit, either by a dividend or because they’re eventually gonna sell the firm or something like that. And you’ve got all these claims. And in particular, there’s tension between all of these because there’s only a fixed pie to go around. And one big thing that you resolve if your profit rather than a hedge fund, is that the providers of capital and the owners of the firm are generally the same people. So a lot of the arguments or tensions that would happen around stuff like how well do we pay our traders, or how much of our trading revenue do we reinvest into tech or infrastructure improvements go away, because those two groups are the same person. So for example, if you say, we want to have a big tech spend this year, because we want to buy a lot of GPU clusters, or we want to spend a lot on Cloud Compute, or something like that, this is work that is presumably only going to benefit many years down the line. And you’re, if you’re a hedge fund your your current investors have no real idea whether they’re still going to be invested 10 years down the line in time to see these benefits. So they’re going to prefer you to do less of that reinvestment and for you to pay out more of the profits as as return to the fund. So that’s the biggest difference, and it creates a real culture change inside the firm, because I think you end up with everyone feeling a lot more ownership in what’s going on inside a prop firm. And viewing things a little bit more for the long term, there’s one difference. And then a second difference is that if you’re a hedge fund, you are incentivized a little bit always think about scalability, you’ve probably got a couple of different revenue sources, you’ve got your management fee, and you’ve got your performance fee. And if you make the fund to x figure, okay, your performance is probably not going to be quite as good, because you’re larger, and you have market impact and harder to put the positions on you need. But if your fund gets twice as big, your management fees just going to double. And that’s nice if you’re the manager of a hedge fund. So you’re always incentivized to think about how can I make things bigger? How can I raise more money? And that means that for the people working there, they’re constantly being pushed to say, right? How can you do what you’re currently doing, but five times bigger, even if it sacrifices the quality of the p&l for that, and that can be nice, because that can be bigger paydays. But operating a sharp One strategy is a very different feeling to operating a shop to strategy, if you’re the person managing that strategy, right in terms of the size of your drawdowns and your level of confidence that the strategy is continuing to work. And so if you’re a prop firm, and you have more of an emphasis on Okay, let’s build a quality stream of revenues, you know, some high quality strategies, focus more on the long term, I think it gives a nice environment for your traders to work in.

Corey Hoffstein  27:16

One of the things you didn’t talk about was mandate flexibility, which is, I think, going to lead us nicely into the next part of the conversation I want to go into because at the prop shop, one of the things that you decided to do in 2021, was helped set up a crypto desk, which if you’re operating within a large hedge fund with a defined mandate is certainly not something you’re going to have the opportunity to do. So talking about the crypto desk, I want to dive into the weeds there. I want to understand when you think about setting up something new, like a crypto desk, how does the infrastructure and operational risk of crypto specifically differ from equities and futures markets? I’d love to know like just generally speaking, what lessons did you learn building this desk from scratch?

Macrocephalopod  28:03

Yeah, so yeah, we started building our crypto desk in the first half of 2021, just in time to see a big crash in 2021, which was great, obviously, in time to see many, many more crashes in crypto in 2020. But yeah, this is still something which we’re still committed to. And I’m continuing to run a crypto desk today. So there is a huge difference in how crypto markets operate, versus how trade fight futures and equity markets operate. The biggest one, I think, is that traditional markets are much more intermediated than crypto markets are. And you will frequently find that you’ll have the exchange and clearing house and the prime broker will all be different firms. Whereas in crypto, typically, this would all be the same firm. And this was, to some extent still is sold as an advantage of crypto that the crypto exchange is your counterparty clearing house and your prime broker and your financing provider all at the same time. I think it is now recognized that that creates interconnectedness and fragility that you don’t necessarily want. I think the exchanges would still love to be providing all these facilities, but I’m now seeing more of a push to have more intermediation here. Another big difference is that crypto is much more real time. So if you are taking some losses in your portfolio, there’s no prime broker who calls you up towards the end of the day and says, hey, you need to top up your account put in more margin before the open tomorrow. You just get liquidated straightaway in crypto. So that’s something you need to be much more on top of and it means that your treasury management which for a Trad fi trading firm is kind of middle office role becomes much closer to the trading desk in crypto, because it’s you know, there’s real edge we call it operational alpha in being able to efficiently manage where your assets are at any point of time to essentially maximize your return on capital. Another difference is that accessing financing in crypto is much harder, particularly post FTX. And your internal cost of capital for holding assets on a crypto exchange is much, much higher than it is for holding capital at a prime broker, for example, you really get concerned about assets which are held illiquid to exchange. So it affects the kind of strategies you want to run as well, that you, you really only want to run stuff, which is high return on capital, because you’re worried about that capital, and then infrastructure wise, crypto exchanges, and I think this is a good thing, typically designed to be accessible for much smaller traders than traditional markets are. So it is absolutely possible. And there are many examples of this of one person sitting in their bedroom, coding up a market making algorithm and going to trade it on finance, or buy their Tokyo or something and making money. But then firms like this have expanded into big teams and become really significant players in crypto in a way which doesn’t really happen in traditional markets anymore. But as a result of that, the tech stack on the crypto exchanges looks very different. So for example, the exchange typically runs in the cloud, rather than in a data center somewhere, their message scripting market data, is to send it over a web socket, encoded as text, basically, which is an extremely inefficient way to encode market data that has very high latency. Because of this, there’s lots of randomness and what we call jitter. In the latency because of this. So if you are working in an institution, and you care about stuff, like latency being low, or you know, having relatively predictable order entry times, you do a lot of work, which you don’t really have to do in or is a different kind of work to the work that you have to do in in traditional markets to get an acceptable trading experience. So there’s a very particular skill set, which looks extremely different from the skill set of a trading software developer. In traditional markets,

Corey Hoffstein  31:55

a lot of those points you touched upon were very much centered around risk. And I’d love to dive into that area a little bit more. You mentioned, for example, that thinking about money on different exchanges with a cost of capital, I’ve heard of firms very explicitly having to think about it as a loan. So whatever money they move to an exchange, there’s an implicit hurdle rate of that they have to exceed based upon some sort of risk assumption about that exchange, I’d love for you to just talk about risk management more, maybe contrast it within crypto to more traditional markets, touching on things a little bit more deeply, like exchange risk. And then another one that comes to mind for me is sort of that managing the fat tail of line items. Crypto is a space where you get all of these new currencies, and derivatives coming to market all the time thinking about managing that fat tail, I’d be certainly interested in hearing more about.

Macrocephalopod  32:51

Yeah, sure. So the first thing to say is that since November 2022, and the FTX blob, everyone, every institution, at least, I think just everyone takes exchange risk much more seriously. Now. So I think this is something where people have kind of had it at the back of their mind, or maybe didn’t think about it that much before FTX. And now, it’s a real primary concern. And this idea of a hurdle rate is something which I’ve heard many, many places, trading crypto, if you have a strategy, and you think I can make 20% a year on my capital deployed or the exchange for that, even if you can make that with very high Sharpe, what you’re saying is, I need to have five years of my assets sitting on that exchange before I’ve made enough p&l to cover the assets that are at risk. And what do you think the chances are of that particular crypto exchange blowing up at some point in the next five years, so you really care about things like your return on capital. And even more than that, there are some exchanges where I personally, or we, as a firm, and I know many other firms just wouldn’t trade on at all, almost no matter what you thought you could get from p&l from trading there, because you have things like enormous lack of transparency around the corporate structure, or the ultimate beneficial owners of the Exchange, or concerns around very inflated volumes on the exchange, wash trading, self dealing, the possibility of there being internal trading teams are the exchange are gonna have advantages which are not available to external trading teams. So there are definitely some exchanges where we would not trade, which shouldn’t be surprising what there are hundreds of crypto exchanges and really, maybe only 10 or so with ease and volumes where an institution would feel comfortable trading, but then even for exchanges where you aren’t comfortable trading, you still have to assign some probability the bad stuff is gonna happen at some point there. So that could be fraud. That could be the exchange gets hacked. It could just be some incompetence, but yeah, even for a top tier, gold plated crypto exchange, there’s a chance that something bad is gonna happen at some point. Yeah, why don’t we just come up? Guess More recently, the first few months of this year is regulatory risk. What happens As the exchange gets sued by the SEC or the CFTC, and all these events could result in loss of funds. So you have to mitigate this as best you can, you can never fully remove this risk, you monitor news flow, you monitor wallet movements, you look at volumes on the exchange, you try and spot anything which looks out of the ordinary. And you make your threshold for getting out just saying I am not happy here, I’m going to remove all my assets from the exchange, you make that threshold as low as you can, whilst being consistent with not constantly having to pop your access on and off the exchange. And then secondly, you just try to have sensible limits on what is the maximum amount of my trading capital that I’m going to put on any one exchange, and that thinking, no one, sensible, whatever now have 100%, or even 50, or 25% of their capital, all on one exchange, suddenly, not as they thought they couldn’t afford to lose it. And then yeah, the other dimension of this is the very large number of line items that you accumulate in your book, especially if you’re running some kind of systematic strategy or doing any kind of fast trading. So we, for example, run a long short market neutral crypto book with hundreds of different underlyings in the book, and any underlying can have multiple different derivatives contracts referencing or you can have the same token held at different exchanges, which they can trade at different prices. So in some sense, even though they are to some extent, fungible, to some extent, they’re not as well, you can’t always move a crypto asset between two different exchanges smoothly. And the other thing is that crypto assets are very volatile, right? So you’ve got this big long, short book, you’ve got a lot of names. And these things have annualized volatility is in the triple digits normally, so high that when we’re talking about this, we normally just quote daily volatilities looking like daily volatility is in the range of five to 15%, which means these things can easily move 25 or 50%, over the course of a day, like that’s not uncommon. And it’s definitely not unheard of, for something to go to zero, or to go up in price 10 times within a very short span of time. And this is kind of a cow on the long side of your book. You know, if you have 1%, your book in a column, which goes to zero, then you’re down 1%. And that sucks. But it’s very survivable for you. But if you have 1% of your short book, and an asset, which goes up 10 times in value over the space of two hours, that’s a 10% loss in your book, and you’re probably trading this thing with leverage, as well. So it sounds slightly insane to trade, incredibly volatile asset with a lot of leverage, but you’re counterbalancing it against your desire to not hold a lot of collateral on the exchange, because you’re worried about something bad happening at the exchange. So you maybe have three or four or five times leverage in your book. And so you’ll have a short position, which is one potential book and it goes up 10 times in value in your trading with five times leverage, that’s half of your trading capital on the exchange gone in a single market movement. So we worry a lot about this. And we have pretty tight limits overall on how big we’re ever going to allow a single position to be in our book, and we’ll aggressively start getting out of stuff, if we think there’s a big danger, individual nine.

Corey Hoffstein  38:01

In our pre call you said one of the primary differences between crypto and equities is that equities, even when you’re talking about higher frequency strategies, can still be characterized by their fundamentals. And that’s something that’s not necessarily the case in crypto. Can you explain what you meant by that comment and why you think it’s an important insight?

Macrocephalopod  38:23

Yeah, it comes down to how you want to measure and characterize risk. So in equities, you have, as well as market data, you’ve got a lot of fundamental data on every name, that you trade, you know, there’s part of a specific sector or or sub sector, you know, trades in a particular country in a particular currency, you have balance sheets and income statements, which you can analyze, you probably have quite a large amount of analyst coverage of the stock. And that’s on top of all of the market data, trading data positioning data that you have for anything with a price. And you can synthesize this data to try and understand which stocks are likely to move together, even before you start looking at any data or any prices. Whereas in crypto, you largely don’t have fundamentals. I don’t think this is a controversial statement to make it apart from stuff like this is a dog coin. The narratives which cause different assets to move together are largely invisible to you if you’re just looking at prices. Often these narratives are just manufactured narratives. A few months ago, there was a lot of talk about potential deregulation of crypto in Hong Kong or people called the China narrative. And this meant that a small subset of coins which are associated with this started to move together very strongly, but there’s nothing really which could have predicted that just by analyzing prices beforehand. And yeah, crypto So mostly, all you have is priced in if you’re trying to understand what kinds of assets are going to move together in the future. More or less, the best thing you can do is look back at what happened in the past so your ability to build models from now. hedging risk is much lower and capitalizing than an equities.

Corey Hoffstein  40:03

To that point, sort of as a follow on statement, you said, when you’re forming baskets on momentum, there’s often some fundamental thing that’s underlying those moves that you haven’t captured in your other factors. True and both traditional financing, and crypto. And I think the China example you gave is probably a very pertinent case where there was a fundamental narrative that all of a sudden cause these cryptocurrencies to all start moving together. So maybe maybe you could talk a little bit more about that idea of why you think momentum is a really important risk factor to consider whether you’re trading momentum or not. And then I’d love if you can even tie it back to your comments about managing the short side talking about that short risk, when all of a sudden, a large number of that fat tail that perhaps your short could start behaving in a very correlated manner.

Macrocephalopod  40:57

Yeah, so as he said, this is an idea which applies to all asset classes. And it’s definitely not specific to crypto or even equities, it kind of applies everywhere. And the fundamental insight is that assets always move together for a reason, even if that reason is something which is invisible to you. And that reason is likely to persist into the future, as well. So it may be mysterious to us why 10 seemingly unconnected stocks have all suddenly started to move together, but maybe someone who understands that they are all stocks held with enormous leverage by a hedge fund or family office, who is now unwinding their positions, for example, like to pick something not quite entirely at random. So if you’re a Call Manager, before, we have some kind of model, fundamental model factor model, which describes the exposures of the stocks, you trade to some more underlying fundamental factors. And the ideal situation is that after accounting for the college moves due to these factors, the residual component of those stock returns to the idiosyncratic component is completely uncorrelated between individual names. And in practice, that is almost never the case. Because your model for what exposures you have and how assets move together is not complete your gaps in other things you haven’t considered. So when you form baskets, based on just a very simple idea of stuff that has moved together in the past momentum basket is an example of this. It’s a basket which is long as stock which is up a lot and short stuff, which is download your to some extent grouping based on these kind of latent hidden explanations of why prices are moving. And this is particularly true if you are neutralizing the understood factors first. So for example, if you just kind of naively go out and say okay, I’m gonna buy the stocks that went up, and I’m going to short stocks that went down. Well, this is going to be driven primarily by things like high beta stocks versus low beta stocks or cyclicals versus defensives or if it’s the last month or so like whether a stock is a bank or not. And that’s not going to be maybe super informative. But if you could have removed the effects of sectors, the sector initialized before you start forming as baskets, or you remove the country exposure, you remove the value or quality exposure first, and then you form momentum particles on the residuals of that you’re capturing more of the hidden factor. And that factor even though something you didn’t know about is likely going to exist in the future. And it’s likely going to continue moving prices in the future. And those stocks which move together are likely going to continue moving together into the future. And you can see this in the data, right? So baskets which you form on past momentum, past price moves have much higher realized volatility in the future, then baskets you form randomly, even when you’ve removed all the sector and factor exposure. Exactly, because you’re picking up a loading on some meaningful risk that drove past returns. So that’s that’s something which you really want to know about. If you’re doing any kind of quantitative trading.

Corey Hoffstein  43:51

One of the things that changed the arc of your career is that you’ve gone from focusing on longer term strategies to what you would call mid frequency strategies. So first, I would love for you to define what you think of mid frequency strategies I often find in talking to quants, the definitions of high mid and low frequency actually aren’t aren’t consistent as an industry. And then so define what mid frequency is and then what you think the key differentiators are between mid frequency, high frequency and low frequency alphas.

Macrocephalopod  44:22

Sure, I’m going to be a little annoying and define mid frequency for what it isn’t what it is. So I’d say high frequency is where you care about every tick. And whether or not some liquidity was added at the fifth level of the order book is a real meaningful thing to you and you can read a lot and low frequency is where you are executing daily or perhaps twice a day in the opening closing auction or something like that. And the mid frequency basically captures everything in between. So mid frequency trading, you’re not caring about every tick likely using some form of bins data, second bins or minute bit In our been something like that, but you are trading continually throughout the trading day. And you’re typically holding positions for somewhere between a few minutes with a very fast end to a few weeks, okay, but typically not seconds and typically not months. And it’s a different skill sets from either high frequency or low frequency trading and different infrastructure, as well. So you typically have a lot more data than you would ever have with a daily diary. And that normally means, unless you’re willing to sit for hours waiting for your simulations to run normally means you probably want some kind of parallelization infrastructure to make your research go faster, you possibly want some form of specialized data storage so that you can load up data quite quickly. So it’s a more technological component than with low frequency trading. But it’s way less technologically intensive, then high vacancy trading is where just to manage a single day of market data for HFT, you’re probably talking about very specialized data storage structures, which are not quite necessarily frequency.

Corey Hoffstein  46:07

How does the research process differ when you’re talking mid frequency versus slower frequency signals? More specifically, is it a completely unique and differentiated set of strategies that emerge at the higher frequency? Or do you find that it’s largely the same signals just somewhat faster? So for example, instead of trading momentum over several days, it’s now intraday momentum?

Macrocephalopod  46:34

Yeah, there’s a lot that is in common between mid frequency and either slow or faster styles of trading. So an example you gave, which is great is exactly like you will measure momentum, instead of measuring it over five months, give one month that often gets used in momentum studies. For daily models, we’ll look at it intraday, or maybe over a few days, we’ll look at reversion on the order of minutes rather than reversion on the order of days, but a lot of the same ideas can still be applied, I guess, it’s maybe more interesting to talk about what doesn’t really translate across. So anything that has a low frequency of data updates is something which doesn’t really translate from low frequency trading to mid frequency. So if you are building a strategy around quarterly earnings announcements, that’s unlikely to be that relevant mid frequency just because you know, most days there isn’t an earning an earnings announcement, or if you’re building around economic data releases, or building it around Alice revisions to a stock, maybe it’s kind of in between most of the time, there isn’t analysts revising the rating on the stock. But when it does happen, sometimes it does happen in the middle of a trading day. And then I’d say the other style of trading, which or the other kind of research you would do for low frequency, which doesn’t translate over so well to mid frequency is anything which uses an extremely broad universe, okay, so anything where you’re looking at the very long tail of very illiquid stocks, which, you know, maybe you can build a slow equity strategy, which builds into those positions very slowly, over time, and you hold in your book, many multiples of the ADV, of that stock, you’re okay with that, because you’re gonna hold it for six months anyway, that doesn’t really work. For me frequency, you mostly want to be able to get out of your positions intraday, if you can, so you end up looking at a narrow universe. And that means some longtail strategies or strategies really only work in the niches most illiquid stuff you don’t really look at from a frequency. And then things which you would look at a mid frequency, which you wouldn’t look at so much in low frequency is anything to do with microstructure. So what is the order look look like? What is the short term trade flow look like? And yeah, these are all things where you would care a lot about them. For high frequency trading, as well. I think the difference is probably that in high frequency you care about the individual events, a lot more that you care a lot about, okay, a trade just happened this price level of this exact time. And that’s going to affect my forecast in the following way. Whereas mid frequency is more about what is the general trend of order flow at this time.

Corey Hoffstein  49:14

As frequency goes up. One of the things that people tend to focus on as being a more important contributing factor to p&l is transaction costs, right, those transaction costs often increase. I’d love to get your thoughts and how you think about modeling. Things like slippage, particularly in crypto where you’re talking about a variety of exchanges and crypto prices tend to be impacted by block time. Right congestion in the network is going to impact both decentralized and potentially the ability to arbitrage on centralized exchanges. Seems like that would have a more profound impact on potential slippage during extreme market environments than perhaps traditional markets.

Macrocephalopod  50:00

Yeah, so you’re absolutely right the your focus on transaction costs and in particular, market impact slash slippage becomes very important, as you’re looking at faster and faster trading. In fact, I’d say even more extreme with that, in some markets or for some strategies, slippage completely dominates your trading costs where it’s a bigger by an order of magnitude compared to the fee you’re paying or the bid offer spread. And naive modeling of market impact leads to assumptions like well, you know, I’ve got a large trade to do, but I can just split it into 10 small chunks and execute them equally over the next 10 minutes with a very small amount of market impact in niche mutual been that I trade them and still have that leads you to write back tests and simulations, which are not very reflective of reality. And market impact. It’s kind of nice, it’s something where there is actually, I think, a pretty good academic literature on it, because people tend to view it not so much a component of their alpha as part of their edge. So they’re much more willing to publish on it. So you can go read this literature, and there are really interesting puzzles to resolve, like, Okay, we know trading causes price impact. And we know that after large trades, prices tend to revert a little bit as the market absorbs some of that flow. And we also know that order flow tends to be very correlated in time so that if you see a lot of aggressive buying, in one time period, you’re likely to see aggressive buying in future time periods. And the time lag on this can be extremely long, it can be weeks or months can observe some strong blood pressure today and give me a prediction that like two months from now, it’s going to be slightly stronger by pressure than average. But even though that order flows is correlated, and is extremely predictable, prices themselves don’t seem to be that predictable, even though the order flow impacts the price. So yeah, how do you resolve this? And if you’re trying to model this yourself, you’ve got the, and you’re working at a hedge fund or prop firm, and you’ve got your own trades, you know, saying what is the market impact of my trades, you’ve got the double problem that you traded, presumably because you had some alpha, and your alpha was going to predict that the market would move in the direction of your trade anyway, even if you didn’t trade and alpha is good, then indeed, it would have moved that way. So how do you disentangle the effects of the market impact you had by trading versus how much the market was gonna move anyway, just because you predicted that it would. So the it’s a fascinating area to study, it’s interesting to try and incorporate it into a simulation as well, because acute are factors that empirically, we see that the market impact of a trade of a certain size follows what people normally call a square root law, so that the size of your market impact is proportional to the square root of your participation in the market over the time period that you traded. And then if you go and implement this in backtest, and you say, Okay, there’s some market impact, which is proportional to the square root of the trade, and then it decays slowly, over time, you will quickly find that this is very, very arbitrage trouble, in the sense that you can just write a strategy, which if simulation does lots and lots of small trades in the same direction, and builds up some market impact over time. So you buy buy, buy, buy, buy successively, you push the price up, and then just do a big sell trade into all that price impact you’ve created and your big sell trade, because of the square root law doesn’t move the market down as much as your previous buys moved it up. And you can just watch something which is simulation is profitable. So it’s a like, it’s a weird example of what seems to empirically fit the best, in some sense cannot be correct. There’s some missing factor. And I’m trying to untangle this is something that I’ve personally spent a lot of time on and found very interesting.

Corey Hoffstein  53:35

I love that example. And it highlights how modeling has many inherent limitations, right, you can find these edge cases where the model might work in aggregate. But certainly, if you take it to its weird, logical conclusion, you can come up with this example where you can create your own arbitrage over time. And I think it brings our conversation full circle back to the beginning about conversations around using back testing, not as a research tool itself, but as a testing tool. Other things that seemed to be more important in the mid frequency range. And I, I want to sort of talk about these and get your thoughts through the lens of how important they are in the simulation. And the back tests are things like estimates of market impact, your fill ratios and your trades, borrow fails on the short side versus your expected alpha, How accurate do you need to be in your estimates of those sorts of details to really make sure that your strategy has legs?

Macrocephalopod  54:37

The short answer is that you end up needing to model these things very accurately. And I say end up because you can often start with some relatively poor assumptions here. And as long as your strategy is small compared to the market, and maybe you can get away with that so you can get away with assuming that you don’t have that much market. to impact and that your trades always get fulfilled, and that your locates always get filled. And you can short as much as you like at reasonable fees. But then, if that strategy ends up being profitable, you’re inevitably going to want to scale it up. And as you try and scale things up, the failures in your modeling become more and more apparent and become much more important. And you’ll notice this because your, your simulation will be saying Great, this strategy should be really performing right now. And then in your life, you’re just losing money hand over fist. So while you can, in some circumstances, get away without modeling that stuff, if you’re going to be successful, and you’re going to try and push your profitable ideas to the limit of how much you can extract out of them, you end up having to be much better at all this kind of modeling. But it’s nice because you get more data which you can use for your modeling, as you push the size of your portfolio. And as you trade more, it’s a little bit of a chicken egg problem, because you want to do simulations, you want to say, Well, how would my strategy look, if I sliced everything up two times and traded double the volume I’m doing now. And you don’t have the data to fit a model that you have to extrapolate from a different regime. And it’s not until you actually push to that larger size of trading that you get the data you need to make your model more accurate. So you’re always in this process of trying to push your assumptions right to the edge of what you can do with the tools and the data you’ve got available. To

Corey Hoffstein  56:29

go into maybe the other side of the coin, we’ve spent a little bit of time talking about maybe the fixed costs and risks, I want to return for the end of the conversation to some thoughts on alpha. Where did new ideas come from? And how has that changed as you’ve changed the time horizon of the alphas that you’ve been working on?

Macrocephalopod  56:50

I used to predominantly get my ideas from reading papers, whether that meant academic papers, or industry papers, or sell side research, which I think it’s still a great way to get ideas when you’re starting out. But then when you’ve been doing things for a little while, you’ve realized that there’s a lag between industry practice and the paper getting written. And that lag is something like five to 10 years, which makes a lot of sense, when you consider that someone discovers an idea, that idea then has to percolate within industry. And then it has to escape into academia or onto the sell side. And then someone has to collect data and write the paper and the paper has to get published. But you can easily get five to 10 years of lag by doing this. So once you’re at the level where your own research is at or has surpassed the level of what you can read about the papers, you’ve then more or less, I think exhausted those as a source of new ideas. And you have to find other places to get good new ideas. And there’s a few places where I go, there’s also one is new datasets. So yeah, whenever I hear about a new dataset becoming available in an area that’s relevant to me, I want jump on that, like, try and get as much as I do, because I can understand it, see if there’s any value in it. Because sometimes hearing about a new dataset, like a piece of data you didn’t know existed gives you an idea for a new piece of research, you can can do just watching the market is another decent source of nowadays, particularly in HFT, I think, yeah, just having a really good Orderbook visualizer. So you can see how the market evolves over time can be a useful source of edge, where you just get to watch how prices form, see a new orders entry in the book and then try and construct your own understanding of who put this order in, why were they doing and why did they trade at this price and build a model of how other market participants act. And that’s something you only really get from what I’m gonna call watching the market, which doesn’t necessarily mean watching the market in real time, it might mean going over historic data and replaying it and zooming in and trying to understand what happened. And then a third really great source of new ideas is by thinking about our own trading and talking to other traders. So what I mean by this is that say we have some problem to solve, like, we want to trade 2% of stocks ADV we want to do that in a way, which is not going to cause too much market impact. And well, we will execute that flow in some particular way. And we will have our proprietary market making algorithm which tries to get into a passively or will have some spread crossing algorithm which tries to get it aggressively. And then you go one step further. And you think about what effect is that going to have on the market? And how could someone who’s on the other side, try and take advantage of that. You can say well, if I’m if I’m thinking to execute my trades in this way, there’s probably someone else who’s thinking along similar lines, and how can I try and identify their trading pattern in the market and use that in a predictive way. And as well as thinking about your own trading and talk to other people about the problems they have trading. And you know, when you’re talking to other traders, sometimes someone will just drop some legitimate alpha on you out of the blue because they’re trying to brag or they want to impress you or they think you already know it and that can be great, but mostly what happens is that they’re they’ll have some offhand comments that triggers Some interesting thought process on your side that after much research leads to a new idea. So I think it’s really worth spending the time to talk to a lot of people about their own trading about the problems they have about how they try and solve those problems. Because that’s for me now a big source of new ideas.

Corey Hoffstein  1:00:16

Last question of the podcast for you. Listeners at this point in the season know that our cover art is inspired by tarot cards, and I’m having every guest pick their own tarot card to design their cover, and you chose the card for strength. Again, I think like most guests, and like me going into this season, you had no experience with Tarot cards before, but my question to you is what drew you to choosing that card?

Macrocephalopod  1:00:46

Strength? Yeah, mostly because it’s card number eight in the tarot card deck. I thought card number eight was appropriate for anonymous internet. Octopuses.

Corey Hoffstein  1:00:55

Does good an answer as there’s going to be. Well, Mr. Octopus, thank you for joining me. I think folks will get a ton out of this interview. I appreciate you breaching your anonymity a little for me. I think this would be a fantastic episode.

Macrocephalopod  1:01:07

Thanks very much, Cory. It’s great to be on.