Angus Cameron is the Founder and CIO of Liminal Capital, a machine-learning focused investment manager.
But Angus does not come to markets with a computer science PhD. Rather, his career arc took him through the prop desks and buyside of Asia, trading global fixed income, FX, equity markets, and arbitrage strategies on a discretionary basis.
A machine-learning driven approach is a new endeavor, but one informed by the wisdom of experience. Angus would consider himself a quantitative trader, not a quantitative investor, and his approach reflects that. Like many systematic investors, Angus breaks the broad investment problem down into data ingestion, idea generation, position management, and risk management. Where he differs from many past guests is in the latter two pieces.
Informed by his trading experience, Angus places a strong emphasis on trade structuring and on-going position management. Liminal automates this philosophy by using swarms of systematic trading agents to place and manage different trades based upon the same underlying signals.
From market structure to machine learning: we cover the full range. I hope you enjoy my conversation with Angus Cameron.
Corey Hoffstein 00:00
Okay 321 Let’s have some fun. 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.
Corey Hoffstein Is the co founder and chief investment officer of new found research due to industry regulations. He will not discuss any of newfound 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 and securities discussed in this podcast for more information is it think newfound.com.
Corey Hoffstein 00:51
This season is sponsored by simplify ETFs simplify seeks to help you modernize your portfolio with its innovative set of options based strategies. Full disclosure prior to simplify sponsoring the season, we had incorporated some of simplifies ETFs into our ETF model mandates here at New Found if you’re interested in reading a brief case study about why and how visit simplified.us/flirting with models and stick around after the episode for an ongoing conversation about markets and convexity with the convexity Maven himself simplifies own Harley Bassman. Angus Cameron is the founder and CIO of liminal capital, a machine learning focused investment manager. But Angus doesn’t come to markets with a computer science PhD. Rather, his career arc took him to the prop desks and by side of Asia, trading global fixed income FX equity markets and arbitrage strategies on a discretionary basis. A machine learning driven approach is a new endeavor, but one informed by the wisdom of experience. Angus would consider himself a quantitative trader, not a quantitative investor. And his approach reflects this. Like many systematic investors, Angus breaks the broad problem down into data ingestion, idea generation, position management and risk management, where he differs from many past guests is in the latter two pieces. Informed by his trading experience, Angus places a strong emphasis on trade structuring and ongoing position management. liminal automates this philosophy by using swarms of systematic trading agents to place and manage different trades based upon the same underlying signals. From market structure to machine learning. We cover the full range. I hope you enjoy my conversation with Angus Cameron. And Angus, welcome to the podcast. Very excited to have you here. I think this is gonna be a really fun episode for everyone to listen to. I hope so. It will be I know it will be. I want to start with something though, that has nothing to do with finance whatsoever. I think it’s just a fascinating part of your history as a person. And so the question to you is, how did you come to spend a year in a Buddhist Zen monastery near Mount Fuji?
Angus Cameron 03:11
I grew up in really the far north of Australia. So I grew up in Darwin, which is right at the very top, and I went to boarding school, an intended what you would consider to be a small liberal arts university in Queensland, where I’d studied economics and international relations. And so when I was at college, I studied microeconomics and macroeconomics. And I took an interest in commodity markets, particularly will and wheat markets. And the more I looked into it, the more I thought becoming a commodity analyst would be an interesting job. I really wasn’t aware that traders existed. And when I was at college, the largest commodity traders in the world were in Japan. They were big organizations called Sogo chaussure. And there was Mitsubishi Maribel and Mitsui Itochu. So when I finished college, I had this idea that I would go and get a job working for a large Japanese commodity trader. So I sent a letter to a friend of mine in Tokyo, and I said, I’ve got this idea. What do you think he said, it was a great idea. Call me when you get to the airport. So I spent some time traveling throughout Asia. And when I finally got to Tokyo, I called up my Japanese friend. And he said to me that he’d thought about my problem thought it was a good idea, but he and his mum thought it would be a good idea that I spoke Japanese before I apply for these jobs. And Tokyo is quite a metropolitan town so they both felt it would be a better idea for me to go somewhere when nobody spoke any English at all, in order to be Learn Japanese properly. And he suggested to me the idea of going and staying in this temple in Yamanashi Prefecture. And guaranteed there was not a person who spoke English around there. And I thought, yeah, I’ll give it a go. So within a week of arriving in Tokyo, I was in this 500 year old Zen temple with a small number of monks. I spoke, basically, one semesters worth of Japanese from college. But I spent a year there really kind of learning to speak Japanese, working on various projects around the template self. I did Zen every day. And over time, my Japanese became possible. And it got to a point where I could pretty much have a conversation in Japanese. And at that point of time, my priest basically said, if you want to go and work, Your Japanese is ready, you should go to Tokyo, and see if you can get a job in one of these organizations. So that’s what I did. I applied like everybody else, I got a whole stack of rejection letters, and eventually got a job working at Bankers Trust, which at the time in the early 1990s, was a pretty formidable trading house. And they were kind of interested in people with unusual backgrounds,
Corey Hoffstein 06:36
I have to ask the way you phrase that the choice to go to this monastery seemed like maybe the most casual decision I’ve ever heard. Was there anything else that drew you to the monastery other than wanting to learn Japanese? Because it is, it seems like a massive life choice.
Angus Cameron 06:53
I was always curious about different ways of thinking and different religions. So to me, it was really kind of a question, is it would I regret this decision later on. And I kind of felt that it was certainly going to be interesting, certainly going to be fun. And, you know, I was curious about Buddhism. And I was curious about Zen as an idea. And when the opportunity is put in front of you, I think as a young person, you should take opportunities when they come. And I don’t regret it for a second is a form of kind of bucolic bliss, living in that little mountain. And I’m still great friends with people them. And every time I go to Japan, and I’ve got time, I’ll take the train up to the small town and catch up with a bunch of friends and stuff like that. But it was something that I thought would be fun to do. And it certainly was, and I got a lot out of it. It’s one of those things, though, where the payback, you kind of feel over a lifetime. So I didn’t really know it. But studying Zen and meditation was quite useful in periods of my life where stress was a bigger issue. And from a philosophical perspective, it was compatible with the way that I looked at the world around me.
Corey Hoffstein 08:14
I’m almost reluctant to move on to the finance related questions, but this is a quant finance podcast. So we do have to take our road back to the more trodden path here. So let’s pick up with your career in the 1990s where I know you worked on a couple of Prop desks in Tokyo in Singapore. As someone whose career really started in the mid 2000s. When I hear prop desk, I tend to think things like risk inventory, or even revenue source for a bank with a bunch of my friends having worked on those desks. In one of our first conversations, though, you described the original prop desk model is something more akin to an insurance policy. I was hoping you could expand on that and why you think the prop model has changed over time.
Angus Cameron 08:58
So I started working at Bankers Trust in Japan in 1991. I was hired by the head of asset management at the time, a guy called Vernon barback. And Bt had a hiring policy where 90% of their graduates came from your traditional business school backgrounds. But they had a program that they called the wild card program, where they deliberately hired 10% of their new graduates from non conventional backgrounds. So I was hired under the wild card program at BT. It was me it was a professional blackjack player, there was a guy who broke his back driving NASCAR. So it was kind of a strange, eclectic group of people that was attached to your traditional MBA intake stream. But I guess getting back to your question, the ecosystem, the financial ecosystem, the 80s and 90s was considerably different to what it is now. There was a legal separation In between retail banks and investment banks, and they had been separated in the 1930s, under Glass Steagall. So, the retail banks took deposits and made loans. And they were guaranteed to a great degree by the government and the investment banks who really focused on sales and trading. And derivatives and m&a were almost run as partnerships. There was no explicit or implicit guarantee. And so the investment banks had balance sheet and they loved that balance sheet. And in the 90s, a lot of them either were partnerships in terms of structure or run effectively as partnerships. And that model had kind of formed over a number of years, where markets being volatile, went through multiple boom bust cycles. And the investment banks have learned a mechanism by which they could maintain their earnings stability over those boom bust cycles without the need of government intervention. So Bankers Trust was a dominant player in the kind of 80s and 90s. It was run by two very interesting guys, Charlie Stanford and Jean shanks. And they had really brought an innovative model to the investment banking world. It was really based on understanding risk and market cycles. They were as an organization obsessed with risk, they effectively invented the VAR methodology that we use today. It was a really innovative organization. They pioneered a lot of ideas in the world of derivatives. And it was a fascinating place because it was a meritocracy. If you walked on any of the trading desks, compared to today, you’d be astounded by how young the people were, the Global Head of foreign exchange was 28 years old, when he took that position. And so this was kind of a different world to where we are now, these institutions have gone through multiple boom bust cycles. And they developed effectively a business model, which had been revised over the oil crisis, the Latin American debt crisis, Black Monday savings and loans where the business effectively ran with a barbell structure. That’s the way to think about it today. And on one side of the barbell, you had your franchise business, which was procyclical, which consisted of market making sales, derivatives, structured products, m&a, and then on the other side of the barbell, you had a proprietary trading business. And the job of the proprietary trading business was really, firstly, to monetize whatever edge the bank had. And secondly, to run a long volatility position, which we would today call a tail risk position as a counter cyclical position to the cyclicality of that franchise business. And it worked in a worked relatively reliably over multiple decades, we during the boom phase, when markets were in a bullish condition, volumes were high, the prop team would make money from arbitrage and our V strategies. And in the bust phase, we had this kind of long volatility backstop. And that would kind of not just make enough money to compensate risk within the prop desk alone, but make enough money to compensate a decline in revenue in the franchise business. Now, BT wasn’t alone in doing this. Solomon was very good at it. Goldman was very good at it. Credit Suisse, Lehman and UBS will had similar approaches. But in the 80s and 90s, that basic investment bank model or that investment partnership started to be globalized. So what Wall Street investment banks were taking that particular model and starting up similar operations all across emerging Asia, developed Asia, Europe, etc. So when I worked at bankers, I worked for two teams, I worked in Tokyo for a global macro team, under two really clever senior prop traders, one of them or both of them made probably the best macro call in my career, the Nikkei 93. They were full on long Japanese government bonds, which at the time was just considered to be nonsense, but it worked out to be the best macro treatment scene for 20 years. And then I went to work for a different team. I worked for an emerging market team, which was run by two very experienced Israeli guys who basically focused on emerging market opportunities. And that was just a collection of very young, ambitious traders who were trading everything we did carry volatility bond basis. equity Arbitrage Fund arbitrage and basically trying to eke out edge that existed within the bank structure. So, to answer your second question, what really changed, there was a change in the model in the early 2000s. The big change was the repeal of Glass Steagall. So that separation between investment banks and commercial banks had remained in place from, I think, 1930 onwards. But in 99, it was decided that it would be a good idea to allow commercial banks to participate in the same activities as the investment banks. So all of a sudden, you had a bunch of new entrants in the space, such as Citi, Bank of America, and they started pivoting their model towards this levered trading business. I guess the second thing that you have was the Goldman IPO. And that was fairly significant. Because up until that point in time, there had been the assumption that any revenue the bank generated from trading had a lower multiple than revenue generated from traditional banking activities. And so when Goldman IPO traded at a premium, on a p basis to the other banks, it basically flipped a switch. And a lot of banks started to pursue this kind of levered trading model. I guess the other factor that was involved was that hedge funds started to emerge. And we started to see a lot of the really solid risk takers that previously had been on the bank prop desks, move to hedge funds or start their own hedge funds. So we ended up having a lot of banks with procyclical risk appetites, going into an environment where there was a huge structural bull market in housing. And we don’t need to get into much detail on how that ended. I want
Corey Hoffstein 17:01
to pull on this thread about monetizing bank edge a little bit. You mentioned that a couple of times there, I know that you were eventually approached by tutor and moved back to Australia. And in this transition, one of the comments you made to me in a prior conversation was that quote, you don’t know you’ve got edge until you’ve lost it? What do you mean by that? Edge is
Angus Cameron 17:24
a really interesting concept. Anyone who wants to trade has to understand ah, it’s really describes how one generates positive expectancy in a trading strategy. And so it’s to some degree, how one generates alpha. And the way I used to explain it to people, and I don’t have that many people to explain it to these days. Feel free since we’ll go into I used to use an acronym Asia. So there’s different four types of edge. There’s market access, there’s speed, there’s information, and there’s analytics. So if I look at market access, I may have better access to certain liquidity, or I may have access to markets that other players don’t have. In terms of speed. It’s really how quickly can I react to information? How quickly can I respond to new information in the marketplace? As far as information is concerned? It’s really data news, opinion and flow data that you get. And then there’s this kind of fourth form of edge, which is really more of an analytical edge, am I better at understanding the implication of events? And I’m better at executing a trade to take advantage of that. So the question that you have to ask yourself as a trader is that in this particular seat, do I have an edge and what is my edge look like? If you go to the typical kind of bank trading room, or even if you go to a pit, which is an open outcry pit, that physical infrastructure has been designed to maximize informational edge, and that’s to some degree, it’s quite sad that a lot of those things have closed down, because to some degree, they were the training wheels of becoming a trader. And a lot of those kind of easy kind of hot edge environments have been taken away from you. But you’ve got to have edge, you’ve got to have a process to identify and execute that edge. And you’ve got to have the temperament to kind of deal with that, that actually underlying risk. So at the core of it, you can break it down into a certain formula. If you’ve got edge, you’re going to have positive expectancy on your trading strategies. And positive expectancy is really a function of your hit rate, which is your probability of winning your risk reward and your win loss ratio, and also the number of trades that you’re looking at. So when I left Bankers Trust I wanted to CUDA I was, this is 1997, I was a cocky 26 year old, had been making money at Bankers all this time. And I was hired as a junior Portfolio Manager and given a small amount of capital to trade in one, arguably one of the greatest hedge funds in the world. It’s so expecting to be kind of validated for this, I went into that environment. And I was shellacked for two years, I wasn’t making money on my short term trading. By the time information hit me, it had basically exhausted its power. So I had no informational edge had no speed edge. And, you know, I wasn’t making any money on my longer term trading, I thought I could be clever by putting on fancy options structures. But I didn’t really have a framework to identify ideas, I was very kind of reactive. So for me as a kind of a young guy, it was a really dark place. I’d always made money at Bankers Trust. But I’d realized that that was probably less because of any skill that I had, and more because some very intelligent people had set up an operation for me to work in. So for me, it was a very kind of difficult period as a trader, I was lucky. in two respects, one was, my boss, or tutor was very patient, far more than I deserved. And secondly, I worked with somebody who comes from your world, who was a systematic trader, a guy called grant Christensen, who also worked at Bankers Trust, and he was a trend follower. And he had been watching me for a couple of years and basically said, my idea generation was reasonably good. I was right 60% of the time. But my trade entry and my trade and management was terrible. So his recommendation to me and this is, after my second year at Tudor was kind of counterintuitive to what anyone would suggest. He said, instead of kind of widening your stops, why don’t you tighten your stops. So I went from kind of being the guy with an idea to basically having an idea and looking to put that trade on multiple times with really, really tight stops. So that forces you into this market timing exercise, though, it forces you into finding optimal entry. And it shifts your focus as a trader on risk reward or the payoff of that trade rather than hit rate, which is counterintuitive for a lot of people going into the space. And that was kind of the turning point. For me as a trader. On the buy side, once I realized that, if I focus on the risk reward or the strategies or generating asymmetry, or convexity in the trade itself, I didn’t need to have a super high hit ratio. So that was kind of a big turning point for me. But it took me like two years to work that fact out.
Corey Hoffstein 23:06
So you ended up spending a few years a tutor, and then another seven, sort of, for lack of a better word as a hired gun, working at different firms. As you got to see the inner workings of all these different firms on the buy side, what were some of your biggest lessons that you learned,
Angus Cameron 23:25
so I didn’t make the cut it tutor, kind of bad and volatile start really trashed my Sharpe ratio. So I left tutor and then I traded futures out of Singapore for about six months. And then I got a job working for a German bank in Tokyo, I kind of became co manager of Asian prop trading for them did really well for about three years. And then I went to London to really kind of move away from Asian markets and focus on G 10, where I worked as an FX and reach trader for a while, I spent some time at a pretty well known hedge fund as well. Once I kind of cracked the trading approach that suited my personality, it was really a question of where am I going to have capital and where am I gonna get paid to be somewhat mercantilist about the whole thing, but that was really what I was looking for? I think so there’s kind of two ways I can answer that question. One is, from an individual strategy or individual traders perspective. And then from a firm or portfolio level, I think the lesson that I took away at an individual level is that there’s different ways to generate alpha. There’s no single solution. There’s different edges in the market. There are different trading models that you can deploy. And you’ve got to be very cognizant that the type of trading that you employ matches your personality to some degree, right. There’s risk associated with trading and some people react differently to different trading environments. At a firm level, it seemed to me that the firm’s that were profitable in the longer term, were really kind of focused predominantly on risk management, the risk of the portfolios and the trades was the most important variable. And once they get to a certain point, the idea would be to have diversification across multiple strategies. But the core competency of those firms that last a long period of time, is risk management and diversification. Now, occasionally, you’ll see situations where there is liquidity risk that rears its ugly head across markets. But if you can focus on risk and diversification, you should be able to survive that. The other feature, I think, which is important, and I didn’t really see this in many firms, but over time, I think it’s probably one of the most important factors is this idea of alpha decay, and continuous evolution. And the sad reality is that once you find a stream of alpha, and you start making money from a stream of alpha, it becomes more popular. And as it becomes more popular becomes more crowded. And so you have this kind of either structural decay or cyclical decay in the performance of a strategy. So I think going forward, the companies that do well are those that are really focused on continuous evolution and improvement through technology, people and processes, and identifying new series streams of alpha, and integrating that into their portfolios.
Corey Hoffstein 26:42
So I want to jump ahead a bit to 2006, when you founded Comodo capital, which from talking to you sort of seems like a bit of a precursor, philosophically, to the firm that you run. Now, my understanding of Komodo is it was sort of a macro multi Strategy Fund, but I was hoping maybe you could walk through some of the core investment philosophies that you were really looking to build the firm on, as it was really the first time you had a chance to build it from the ground up? Yeah, I should say I
Angus Cameron 27:13
was a pretty reluctant hedge fund manager, I was very reluctant and hesitant to start a fund, I kind of felt that unless you’re generating a Sharpe ratio is greater than one and a half, you really have no right, risking other people’s money. So my Sharpe ratio was good, up until that point of time, in the long term, but I would go through patches where I have flat returns. So I had this idea that if I kind of take that old BT model of a diversified Alpha stream, and take the macro returns that I had, and add different kind of more procyclical strategies to it, I can get to a Sharpe ratio, that would be consistently at a high level. And so the objective of Comodo was to see if we could generate alpha over the entire cycle. And the principles by which we operate it are really threefold. The first one was preserving capital at a trade level. So every trade and every strategy have a stop. The second principle was to identify strategies that had attractive return profiles. So we would have a series of trades that had the same entry logic structure and accepts that over time generated positive expected returns. And then the third principle was to see if we could combine those strategies into a diversified portfolio of alpha streams. And ideally, what you want to find is strategies with complementary risk reward characteristics, such that when the cycles come and they do come, you could ride that with a degree of smoothness. So that, you know, the types of strategies that we have were directional strategies, relative value and volatility. And the challenge was assembling them into a portfolio that could do well over a multi cycle period. And so for us, it was, we will I was lucky, I found some investors that like the idea, they gave us seed capital. And I guess I could break down Komodo into two phases. The first was kind of oh six to 2009, where we generated solid, absolute returns of the entire crisis, predominantly kind of macro directional and long volatility positions were driving that. And over that period of time out, Aum went up significantly. Proud to say that we navigated or never suspended any redemptions over the whole period of time, provided monthly liquidity, including during the Lehman and Madoff crisis. And thanks to a really kind of smart team of people on the operation side, we sidestepped A lot of the counterparty disasters that affected hedge funds in that period, there were kind of paradoxically some funds that were perfectly positioned for the sell off in, oh 708, that ended up losing money because they chose the wrong prime broker or they didn’t manage the cash. So that was a facet of the whole episode, that doesn’t really get much discussion. So we did very well from kind of Oh, six to 2009. And then in 2010 2013, the markets moved into really a growth, recovery phase of low inflation. And then we pivoted to more relative value Long, short equity strategies, we had a long bias on our macro books, and generated returns, which on an absolute basis was still good. But relative to our peers, who were employing a slightly different strategy, more of a hedged long strategy, we didn’t do as well. So it was, the whole idea of a diversified portfolio of alpha strategies wasn’t theory, a great idea. In practice, it was very much harder, as I discovered, to remain at the efficiency frontier of each of these strategies, we had to execute and kind of retrain these strategies on a continuous basis. And then the real challenge was in the post financial crisis world, we’ve gone from an environment where correlations between assets geographically spiked into the financial crisis. And then the correlation never declined. So from an operating perspective, we went from trading eight hours a day to trading 24 hours a day, in a way we ran out risk meant that this was going to be a lot harder than it had done in the past. So over that period of time, we did well, investors were happy, we were happy, everything was great. From the outside, you know, the am I’m grew, the returns was solid, we had a multi strap fund, we had a macro fund, and we had a systematic fund. And we won a bunch of awards, which is kind of nice when it occurs. But that 24 hour trading cycle, started to really create friction, and started to wear people down. We weren’t generating quite the same returns that were done in the past. But more importantly, the whole crisis itself had this enormous wear and tear on people. And I think pretty much everyone I knew who sailed through that storm was to some degree falling apart, health issues, exhaustion, just the stress of the whole thing was kind of diabolical. And I had made a commitment to our investors at the very beginning that we needed to generate, we promised to generate superior returns over all business cycles. So in 2012, to the surprise of many, and the delight of some, I returned or client money, I made sure my staff had positions to go to, and we closed Comodo to outside investors. And that was the end of that chapter.
Corey Hoffstein 33:23
And from there, we’re gonna fast forward half a decade and halfway around the world to a scene in San Francisco 2017, where you launched liminal capital, that from your description to me, in many ways seems to be tied to a lot of the same sort of philosophies and intentions as Komodo, arguably a lot of the same markets and instruments. What are the biggest differences between what you were doing at Komodo and what you’re hoping to achieve with lemon oil today?
Angus Cameron 33:57
So I went back to Australia in 2012. I did some policy advisory work for the government. I was as a separate issue. I was very confused at the policy reaction to the financial crisis. I’m not sure if I’ve got answers that make a lot more sense. But I needed to decompress. So I spent some time trading futures from my own account. I didn’t want to get a cold hand. And I went back to school, I went to what is equivalent to a community college and studied software engineering, I’m proud of that have been much more of a hacker than having any formal training in software engineering. And I remember the time I kept on reading about these organizations like deep mind and open AI, and I had this idea ticking over my head that I wonder if it would be possible to build an AI agent trade. And what was interesting to me was that a lot of The developments were really focusing on game playing. And what I’d found anecdotally was that there’s a lot of similarities between playing games and trading. In fact, some of the best traders I knew were also gaming themes. They also kind of find it interesting that there’s a high correlation with people who played Dungeons and Dragons, and those who end up on the prop desk, I thought you’d find that particularly interesting. But we realized that I kind of came to the conclusion that the way that the whole market making industry had been transformed by technology, and you’d had this emergence of high frequency traders, the same forces were at work and ultimately trading investing, were going to be automated to a high degree. So with that, just hunch in mind, I packed up my family, and we moved to Palo Alto. And my wife at that point was used to kind of big family changes based on hunches. But it was really based on this vague, nebulous idea that we could use machine learning to automate a trading process. So when we arrived, we really needed to use the Silicon Valley binocular proof of concept, can we build an agent trade a market, and we chose Mini s&p futures. And we chose a directional trading strategy. And the timeframe was short to medium term. And so that original model took us about 18 months to build. Now, a lot of that was just me coming to terms with the technology, and what was available. But ultimately, we built an agent that could receive information, process information, make decisions, real time and adapt to changing market conditions. So for us, it was really looking at the different components of what a trading processor is constructed from. And you know, in a human based system, you have analysts, portfolio managers, and traders. And then we basically looked at what those functional responsibilities were, and could we codify those functional responsibilities. So interestingly, where we are now, from an external perspective, we don’t look anything differently from a typical fund, we have a product, which has a universe of instruments, a target, a methodology, a risk budget, but under the hood, it doesn’t look anything like a hedge fund. We don’t have analysts and traders and portfolio managers, we have a group of software engineers, data scientists, DevOps people, quantitative researchers. So we look a lot more like a software company than we do look like a hedge fund, even though the problem that we’re solving is the same problem that hedge funds look to solve. You
Corey Hoffstein 37:56
explain to me that one of your key insights during the development process of the technology stack actually came from talking to folks at Tesla, about their autonomous driving process? What was that breakthrough for you?
Angus Cameron 38:10
I think there’s so many issues associated with Tesla from Wall Street that you probably need it issue a trigger warning to your podcast listeners that you are going to mention the T word. But I think, to some degree, I’ve never been afraid of asking stupid questions. And that’s possibly one of my greatest edges. So when I arrived in the Bay Area, I met a lot of people who were solving kind of non trivial problems that involved complex decision making in different areas, from autonomy to medicine, to robotics. And so the takeaway from our conversations with the guys in that space was less towards less the specifics of any solution. But how do you frame the problem. So if I can give you an example, if you look at something like driving as a human, we just get into a current drive. There’s a lot of things going on, that we’re not necessarily conscious of. But if you give that problem to a software engineer, to accomplish that task, you’re going to need a fundamentally different approach. So if you’re looking at autonomy, and what’s going on behind the scenes and autonomy, you’re really solving a lot of problems simultaneously, you’re solving navigation, object avoidance, driving the car, staying in your lane controlling the car. So you’d have a combination of Sub procedures with effectively a controller mechanism. And the Sub procedures are all independently generating predictions and recommendations. And the controller mechanism is one which is kind of waiting those predictions and those recommendations in order to decide what is the action path. So from Tesla, it was really understanding their decision making architecture that we really kind of feel thought long and hard about, we met people from Facebook as well. That’s also an very interesting company. Because although the core product is not as complicated as kind of a self driving car, they do have unique challenges. And their challenges really are associated with operating at scale. And this is really, I think, where, but potentially, Zuckerberg has not really appreciated that he solved a lot of scale issues in the early days of Facebook, which probably would have crushed other companies. So we looked at how they manage data at scale. And we also have a really good relationship with Amazon. They’re a big player here through AWS. And those guys have been super supportive in terms of making our engines or trading engines and our information engines continuously available. And they’re very open with helping us solve infrastructure problems that kind of occur. So we did this almost like full immersion, not into Silicon Valley, but also into the manner in which they solve problems, which is very different from how you do it on Wall Street.
Corey Hoffstein 41:13
When we think about the traditional Wall Street process, we might sort of think about breaking down a trade into these individual components that might be their own part of the tech stack. So for example, idea generation entry or market timing, trade, structuring how that position gets put on, and then the ongoing trade and risk management of that position. At what part of the stack, do you think machine learning provides the greatest edge? And why?
Angus Cameron 41:41
That’s a really interesting question. I think machine learning is not necessarily the edge, I think it really enhances whatever edge that you’ve developed. So in terms of if you can frame it in terms of those different edges that we’re looking at machine learning through natural language processing allows us to read news and extract signal from the noise a lot faster than somebody just sitting at the desk staring at a news stream. It also kind of allows you to process complexity in a different way to how a traditional analysts would be doing it, you’d naturally get a speed edge, because you can process information a lot faster. And in terms of idea generation, and trading, I do think that the ability to analyze a lot of information in real time gives you a higher hit rate, and ultimately, expectancy on a strategy by using machine learning. But just to be clear, we don’t really know where the most effective use of machine learning is going to be. It’s now permeated so deeply into the fabric of what we’re doing that it’d be very difficult for us to go back to a traditional approach. It is, however, recommended that you take it very gently, because the use of machine learning creates its own set of problems that don’t exist. In a human based process. You have issues with data quality, you have issues, particularly in our area of finance, where data is incredibly sparse, overfitting is a huge issue. And to get the whole machine to run, you’ve got infrastructure issues that are kind of unheard of in a traditional asset management company. So I wouldn’t say machine learning creates an edge in itself. It kind of enhances whatever existing edges you have. But it is no panacea for the hedge fund industry and for traders going forward. It creates as many issues that you have to resolve them actually does resolve itself.
Corey Hoffstein 44:01
So Angus have just about anyone I’ve spoken to in the three and a half seasons I’ve done with this podcast now you are probably the person who has put the most emphasis on the importance of actual trade structuring and ongoing management of those trades. And in fact, in a prior conversation, I think you sort of alluded to, or may even call it sort of a hidden source of alpha in the markets that a lot of people overlook. I’m curious as to where this view came from for you and why you think it’s so important.
Angus Cameron 44:32
In a way it’s a difference between whether or not you’re a trader or what not you consider yourself an investor, if your main functionality is trading, and what I mean by trading is if you have leverage and you have a stop, whereas investing is where it’s kind of the absence of of leverage in the average absence of the stock. Trading and managing the position size over the life of the trade becomes more and more of an important kind of feature of outperformance the kind of longer you trade. So I wouldn’t necessarily say it’s a hidden source of alpha. But I guess it’s a rarely discussed source of alpha, outside of trading circles. People generally like high hit rate strategies, they like that there’s this intuitive sense to be in desire to be right on a trade. And the problem was trading strategies that have high hit rates, is that they don’t always translate into positive expectancy of the longer term. So if you look at it as a kind of a single trading strategy, through the lens of a payoff formula, you have, basically three input variables, you’ve got your hit rate, which is how often are you right, you’ve got your risk reward, which is how much you make when you’re right, versus how much you lose when you’re wrong. And the other variable is the number of trades that you put on in a given period. And ideally, as a trader, you’re looking for a strategy that has a positive expectancy, which basically implies that your hit rate multiplied by your winners, minus one minus your hit rate multiplied by your losers is going to be a positive number. And you can either focus on improving the hit rate of a strategy, we can focus on improving the risk reward of the strategy. And over the longer term, I found that strategies with higher risk reward or what you would call convexity, or asymmetry, in payoff, generate a more sustainable long term return. It’s the difference between buying volatility and selling volatility. Selling volatility is a strategy which has a high hit rate, but a really awful risk reward. Buying volatility has a low hit rate and a positive risk reward. So you’re constantly dealing with those features. So really, kind of the issue of traders and what they talk about is, what is my trade expression? What is my trade structuring that generates that high risk reward? Rather than do I think Microsoft is going up? Or do I think Microsoft is going down? I guess that’s the point that I was making about a hidden source of alpha. But to kind of put it into a framework that’s a little bit easier to understand, there’s really only two functional tasks in both the kind of trading and investment world, there’s really two domains of activity. The first domain is kind of idea, generation or research and prediction. And that impact entails new idea generation and the monitoring of existing ideas. And then the second domain is really the action space, which is trading and investment strategies. And both of those are skills that require time to develop as a discretionary trader and time to codify as a systematic trader. And some people are more predisposed to one and not the other. And so just kind of going back to our Dungeons and Dragons reference, I forgot to ask you, what was your favorite class and Dungeons and Dragons? This was always a good question,
Corey Hoffstein 48:19
is a favorite trading question on the desk,
Angus Cameron 48:21
it was actually,
Corey Hoffstein 48:23
I actually am a big believer that your class sort of matches your personality. But I always feel like I have trouble guessing people’s class. My default has sort of always been a half elven Ranger, but they get very weak at the high levels. I think anything but a wizard at the high levels is just pathetic.
Angus Cameron 48:41
I tend to agree with kind of general classification is that if you’re a cleric or wizard, you should probably go into the reset side. If you’re a fighter, Ranger or Paladin, you should probably get into trading. And the joke was, If your answer was beef, you should probably go into high frequency trading. So we could spend a lot, we could almost spend the entire podcast talking about this. But if you think about this two dimensions of idea generation and trading, you can take an idea from really the world of statistics and think about I’ve got a matrix similar to a statistical hypothesis testing matrix where I have good predictions and bad predictions, and I’ve got good trades and bad trades. And if you think about that, in a two dimensional matrix, there’s really four outcomes. The obvious one is that if I have a good prediction, and I can treat it well, I’ve generally got a good outcome. And if I have a bad prediction, and I treat it badly, I’ll generally get a negative outcome. Where it becomes interesting is that the better you become or the more adept that you become a trading, the more you start to skew those outcomes on a particular trading strategy. Generally speaking, people’s starting point is pretty bad. I think the harsh reality is that 80 to 90% of People are predisposed to be bad traders. It’s not because they have any deficiency, they’re biologically predisposed to make bad decisions. And that’s the cause that’s driven predominantly by our limbic system, and how we react to stress and uncertainty. And unless you train yourself to do the right thing, or you’ve got someone to teach you to do the right thing, you’re going to make a lot of mistakes. So as a discretionary trader, I always found it was the takes a lot of time and effort to train oneself, or to trade a individual to be a good trader. So kind of going back to that idea of their matrix. The good prediction with a good trade generates a good outcome. And you really have these two other edge cases. One is that you have a good prediction. And it’s badly traded, which is what we call a type two error. And that’s really common with a lot of people when they’re moving into the trading space. There’ll be right on their prediction. But they were tying the trading correctly, those sizes trading correctly, they will manage the trading correctly to generate a bad outcome for them. And you can see this with a lot of data that comes out of retail trading platforms, that on average, 80 to 90% of people will lose all their money in three to six months. And it could be a situation like we’re in now, where it’s just we’re going to clear boom, off in equities, it should be relatively easy. The other edge case is a really interesting one where you have a bad prediction and a good outcome. And from experience, I’ve found that investors often have great difficulty in processing, that there is this assumption that if I’m right on the idea, I should make money. And if I’m wrong on idea, I should lose money. But sometimes you can be right and lose money. And sometimes you can be wrong and make money. So type one error is not quite as bad as a type two error. But as you can understand, if you focus on those particular domains of a degeneration and research and trading, the better you become a trading, the better than the outcome becomes, the better you become an idea generation, the better than the outcome, the top. So if you look at how this problem is addressed, or the whole kind of landscape is addressed within the hedge fund community, the default configuration for a trading team is where you take the trading problem and the idea generation problem. And you have an analyst, a trader and a PM, the analysts job is to find good trade ideas, and revise existing trade ideas or biases that they have. The traders job is to execute those trade ideas and the feedback information to the pm and an analyst and the portfolio managers effectively. In America, you’d like to use the term quarterbacking they’re effectively quarterbacking the whole operation, the sizing, the structuring them managing the trade through the life of its trade. And if you can find a team of people that do this together, it’s a lot of fun. And it’s very rewarding. There’s a lot of feedback between different roles and responsibilities. But if you look at most hedge funds, this is how they can figure it they have this kind of three roles, the analyst, the trader and the PM. And this basic model comes in different variations. You could have like the example of the lodge fund where you have a rockstar PM, and he’s got 10 analysts working for him and 10 traders working for him. That’s really the tutor model and the point 72 model where you could do Alternatively, you could have like a big multi strap like a millennium, which has three 400 different pods with those functional roles, again, performed in diversify strategies across different asset classes. But the problem that you have with that kind of human centric model of animals trader and portfolio manager, and this is what we learned over time, is that obviously you’ve got continuing problems, right? Firstly, you’re dealing with a non deterministic process and the outcome can’t be guaranteed. The second issue is really a bigger issue which is the drivers of price on the idea side are continuously changing. The markets will focus on particular variables or features, and then they’ll get bored of them and they will move on to something else. So the drivers of asset classes from an idea generation process are changing. And unless you have adaptation built into your process, you’re going to have a decay in the efficacy of your idea generation over time. And the other problem is to trade efficiently to train that position or that idea efficiently. It really needs to be managed real time on a 24 hour basis, because you have predicted events coming through kind of scheduled events, like data releases and earning statements coming through the process, and you have unscheduled stuff coming through as well. But you have to process and that was the big change that we saw post financial crisis is that correlations between asset classes between the US, Europe, US and Asia increased dramatically, almost to one during the financial crisis, and then they never really went down. So if you’re trading Euro stocks, if you’re trading Nikkei, your future training Chinese equities, a lot of the prices are driven by what the s&p did overnight. And unless you’re monitoring both the s&p 500 and the Hang Seng or the Crossfield and AKA, you can’t optimize the trading of that. I guess the other problem, conceptually, and this is things that you discover when you’re trying to take a human process and convert it into code is that trading is really more of a synchronous process rather than a synchronous process. It’s really the idea generation and the decisions that are being made. From an idea front from a trading front Oh, operating a synchronously in the old days, I would sit there coming on Monday morning, and start going through all my asset classes and decide whether or not I want to be long or short, the more asset classes that I was trading, the longer that process was going to last. And then once I’ve done that process, I can then look at which ones I like, in which trades I could put on. But that’s a synchronous process for the problem is actually an asynchronous problem. So we had to rethink considerably how we manage that, as we evolve. And the biggest problem that was kind of a very human centric processes, this is limited throughput. A single analyst or researcher can probably monitor, maybe five to 10 different assets, and a trader PM, can probably monitor about five to 10 trades. And to a great degree, that was the real challenge that we had, trying to move from what was effectively a systematic macro process. And to scale a systematic macro process, we had always been very quantitative, in our approach AI as a kind of rule of thumb was uncomfortable trading on any untested assumptions or heuristics. So I was very good at kind of deep research, analysis and trends in terms of understanding how assets or trades were affected by different variables. So at commodo, we continue to the quantitative work in the systematic macro and systematic relative value, we built a bunch of fundamental models to predict the direction of assets, albeit not as sophisticated as what we have now. And my research team was very good. We got to mid 60s In terms of predictive accuracy for a lot of the instruments that we were trading. But as we systematize, the idea generation process, and you can think of this as a factory, as we automate on one side of the factory, we have the benefit of the number of ideas were increasing, and the quality and efficacy of those ideas were increasing. But at the end of the factory, those trades were passed to a bunch of monkeys of which I was a monkey, we had to take that idea. And we had to convert it into a trade, manage the trade, size, time structure, the trade. And so in this attempt to be more robust and systematic about our process, we’ve kind of created a problem for ourselves where a systematized degeneration process feeding into a manual trading process was going to cause something to break. And we broke my, my trade is good, not like a new world of systematized trading. So the solution that we had, when this was a problem was either to hire more traders. And that was difficult because it takes a long time to train them to get on the right side of that trading dimension, or we had to look for a better solution. So I had a bit of a hiatus for a period of time. I got involved in politics, which we can talk about a bit later. But I really started looking to how technology could solve both the idea generation problem and the trading process. And there’s a concept in computer science called pattern called a finite state machine. The idea of a finite state machine is that you have a process that goes through multiple states. And so that was really a big clue for us about how to approach the problem. And so we just started playing around. And we spent probably 24 months, just trying out different trading approaches of different trading algorithms, which is what we call agents for just the s&p 500. With our prototype, we call the T 100. Don’t know, if you kind of get the late 80s pop culture reference. But that was the Arnold Schwarzenegger version of the Terminator. But we ran that t 100. prototype for 18 to 24 months just trying to find which is the model that was going to work, how we could run that at scale, kind of the Silicon Valley version of trashing blowing up rockets, which is what they do at SpaceX or, but for us, it was just kind of, can we get something to work, only trading Mini s&p futures. And we finally found, after a lot of trial about an error, we found that approach that work. And the interesting thing was is that we took that kind of t 800. And we trained it on Mini s&p futures Bucha. Okay, that works reasonably well for millions simply future’s not as good as a guy who’s sitting there full time is 20 years experience trading Mini s&p futures but not far off. So we’re not beyond human level, but we’re not far away from the level. And then we took that and we said, okay, is that going to work trading currencies, we found the slight changes at work trading currency, does that work trading single stocks? Yes, it works with trading single stock. So when I use the term agents were referred to as this idea of a trading algorithm that does the kind of the period, the resizing and retreating on sensitivity to external events. And each one is somewhat unique, that each market is somewhat unique. There’s certain peculiarities certain a cadence or tempo to prices in each market. And there’s obviously a different sensitivity to external events. But you know, that basic model is really what we’ve spent a lot of time doing. Once you kind of receive some kind of success in that space, then we then had another challenge, which was to see if we can operate more than one of them simultaneously. And that kind of led us into a whole different world of DevOps, and building infrastructure. And building a system where we could not only run, but we could also manage the risk of 50 100 1000 different agents running simultaneously.
Corey Hoffstein 1:03:07
I want to talk a little bit more about the risk management side sort of that final piece and the ongoing trade, painting with really broad strokes, there’s sort of two camps when it comes to risk management, one that tends to focus on risk management through the lens of portfolio construction, looking at all the risks in aggregate, and then the other camp tends to focus on risk management more on a trade by trade basis, not really caring about how the aggregate picture looks just really making sure risk is done for a particular trade. Which camp Do you fall into? And why?
Angus Cameron 1:03:41
It’s kind of a false dichotomy, right? Because there’s both, there’s definitely multiple approaches to risk management. And unless you have your head around risk management, and the potential downside of an investment portfolio or a trading portfolio, you’re going to be in serious strife. And the core mathematics behind it is really based on probability. So I don’t see them as necessarily incompatible or different kind of sects within a religious world. And to some degree, if you look at Ben Graham’s concept of margin of error, that’s effectively a stop mechanism that is formalized in how he approaches markets. So I don’t see them as being necessarily contradictory. And I would say, if every strategy or every program has a stop, it’s whether or not you’re working at someone else’s work. Yeah. That’s an important point to remember. But coming from the trading space where we use leverage a lot. It’s necessary to have stops and it’s necessary to have an understanding of the risk of it, individual trade and how that individual trade relates to other trades within your portfolio. So it does create a degree of tension and has created a degree of frustration because people like to see the world from the perspective of am I right? Am I wrong? Did I make money did I lose money. And so if you can think about it from a mate, that matrix, we all like to be right and make money. And if we’re wrong, there’s an expectation that we lose money, which is okay as well. But if you trade with stops, you’re dealing with situations where you’re right on the idea, but you lose money on the implementation. And that doesn’t sit well with a lot of people. And Alternatively, you could be wrong on the strategy, but you could make money on trading it, which is puzzling to a lot of people from the outside. But I’ve been in this game long enough to see everything happen, or a lot of things happen that I did not anticipate. So from our perspective, we just, we integrate stops and trading risk management into the entire process of what we do. If we’re running a portfolio, and we’re hoping that our loans will balance our shorts, we’re still running a stop, we’re just not running a stop and the individual positions, we’re running a stop on the portfolio itself, and doesn’t necessarily need to be hard stop. It could be the scale stop, or it could be a reduction stop. But I just think markets have a tendency to blow up things beyond your imagination has a tendency to happen. And you’ve got to have a mechanism to deal with that. The downside is that sometimes you get it right, you don’t make any money. And I think a lot of people don’t like that about the whole stop mentality, that I don’t know what the alternative is.
Corey Hoffstein 1:06:52
One of the original reasons we got connected was because you had read my paper liquidity cascades and wanted to reach out and share a lot of the experiences you had over time and thoughts about how market microstructure had really changed. I thought it was really fascinating discussion that probably deserves its entire own hour. But I was hoping maybe to sort of end this podcast, you’d be willing to discuss some of those changes that you’ve seen throughout your career, and maybe what you think the most important ones are that are affecting market microstructure today.
Angus Cameron 1:07:23
Yeah, I did really enjoy your paper, I thought, wow, here’s somebody that’s really understanding things beyond the superficial analysis that we get in most of the media. So I certainly applaud you for having the courage to go out and put it down on paper and then publishing it to the world. My experience in the, I guess, in the market making space to some degree is that, you know, when I was in London, my partner and I, who’s also at liminal, and was at Komodo, built some of the original market making models for the FX business, that turned out to be wildly profitable. And we also ran a Comodo some trading strategies that were somewhat dependent on ultra low latency in foreign exchange markets that, again, were quite successful. The point that I wanted to make not to correct you because I think the general thrust of your arguments were correct was that the market making ecosystem has changed quite dramatically over the past 10 years. And this is an important factor because market makers are effectively the pipes that connect participants in the market. And the two changes really have been in the number of participants and the diversity of participants. So on the number of participants, we’ve seen the number of people in the market actively making markets declined precipitously, the investment banks used to be big players, thanks to the Volcker rule and the Dodd Frank Act. They can’t hold a lot of risks. So they’ve stepped away to some degree from market making. And the proprietary market makers, what we found was to maintain a speed and access edge, it became almost insanely expensive. We went from the infrastructure and the lines costing 10 to $20,000 a month to half a million to a million dollars per month. And so what this has meant is, from an economic perspective, increasing returns to scale and a consolidation where you end up with to some degree a Pareto distribution where you have five to 10 dominant players in that market making environment that really have been driven by regulatory changes and technology. On the issue of diversity. There’s been a trend To within these small number of players to very, very similar models and very, very similar risk appetites. So the market then kind of reaches a state where it’s either kind of providing liquidity or there’s no liquidity, there’s no Shades of Grey. In this new world, it’s either on or off. And when something unusual happens, it’s very quick for these market makers to just hit kill, and go to passive market making and all of a sudden, liquidity evaporates. So that was really what I wanted to kind of share with you. My suggestion is that you’ve kind of renamed the paper liquidity cascades 2021. Because these cascades have different origins. But they seem to occur on a more frequent basis, your require, almost like a bubble in assets with a confluence of factors that create the risk of low liquidity. So 98, we went through the GK to default, which resulted in LTCM. And all the banks pulling liquidity, which was quite dramatic if you’ve gone through it, in Oh, 708, you had a property subprime bubble, and you had all the investment banks who are market makers all of a sudden gets smashed on their balance sheets, so they pulled the Quiddity. And we’re kind of in we’re in this new regime of activist central banks, passive flow dynamics, and this hyper concentration of market makers, which, in the medium term, creates the preconditions for, I think, some pretty exciting volatility markets going forward.
Corey Hoffstein 1:11:49
Last question for you. We’ve got vaccine rollouts going on around the world. I’m trying to stay as optimistic as I can here. Assuming the world starts opening up in the summer, this fall with mass vaccinations, what are you most looking forward to doing again,
Angus Cameron 1:12:06
surfing. I’ve kind of fallen for the Northern California. It’s a great surfing culture, hook in so much as I thought, okay, Northern California is going to be great, I can go swimming and I can go surfing. The problem I found is that the water temperatures insanely cold at all times of year. And so my usual kind of relaxation, which is to go to the beach and go swimming and body surfing is no longer available to me. So as soon as the pandemic is over, I’m off to a beach somewhere with my family. And that’s how I kind of recreation at this age, my life.
Corey Hoffstein 1:12:46
Angus, this has been fantastic. Thank you so much for joining me. You’re
Angus Cameron 1:12:50
more than welcome.
Corey Hoffstein 1:12:55
If you’re enjoying the season, please consider heading over to your favorite podcast platform and leaving us a rating or review and sharing us with friends or on social media. It helps new people find us and helps us grow. Finally, if you’d like to learn more about newfound research, our investment mandates mutual funds or associated ETFs. Please visit think newfound.com. And now welcome back to my ongoing conversation with Harley Bassman quantitative easing from 2008 to 2020 really failed to generate any measurable inflation, at least through traditional measures like CPI. I’m curious as to your thoughts as to whether we should expect anything different from the monetary and fiscal responses that we saw from the COVID-19 crisis.
Harley Bassman 1:13:45
Is it rude to push back on your host after he’s been so nice to me? Of course, you’ve had inflation, I guess you put in the caveat of CPI but of course we’ve had inflation. Look what financial assets have done, look at Gold, look at housing, look at everything, everything went up except for union wages. By union, I mean, basically hourly workers, people making 30 to 70k. That was the original plan of quantitive. Easing, zero interest rates was to go into push money into the system to create moderate inflation that would then reduce the debt we had because we built up way too much debt prior to the housing crisis. The way you get rid of debt is your default, or you inflate, and inflation is a slow motion to fault. There is a third scenario where you have significant real GDP growth, which is how we solved the debt crisis in the 40s and 50s. After the war, I do not think we’re going to have four or five 6% real growth in the economy. So therefore I’m left with default or inflate, and we have had inflation. What’s coming up in the future is we’re going to go and try to use fiscal policy in addition to our current monetary policy to come Create velocity MV equals PQ equals GDP, money times velocity, price times quantity GDP, the M went up, the V is gone down, which we need to deal with the Fed needs to do with the government is to do is to somehow stop philosophy from declining. It doesn’t even need to go up. If it just stops going down, then the increase money will increase GDP and likely that will be happen via inflation, which at a moderate level three 4% would be a good public policy outcome.