Quant Talk

A transcript of my interview with friend and former colleague Misha Boroditsky in New York

[11/17/2024]

[This interview has been edited for length and clarity.]

[Jérôme BUSCA] Hello everyone, and welcome to Quant Talk. I'm Jérôme BUSCA, your host, and today we'll be interviewing Misha BORODITSKY, who, until recently, was head of Fixed Income execution at Goldman Sachs Asset Management in New York City. I've known Misha for many years and I'm glad to call him a friend. We worked together at a firm called Knight Capital Group nearly 15 years ago, when he was leading the automated corporate bond trading team, alongside other roles in Equity market-making.

Knight Capital was at the time one of the top electronic trading firms in the world, and, for our listeners who want to get a sense of how things were in that period, you can for instance watch the 2010 Oliver Stone movie titled "Wall Street: Money Never Sleeps", some scenes of which were shot on the trading floor at Knight Capital Group, and for which some actual employees were hired as extras. The movie is also known as Wall Street 2, since it was intended as a sequel to the wildly popular 1987 film "Wall Street", with lead Michael Douglas starring in both. But back to Misha. His background is in physics, and after making the switch to finance, he has enjoyed a distinguished career in the high-stake, competitive field of high-frequency trading and market-making with different firms on Wall Street. He's an experienced quant, a successful trader, and a renowned expert on all aspects of quantitative trading and automating market-making. Without further ado, it is my pleasure to welcome Misha to this interview. So, hello Misha! I hope you don't resent my injecting a little Hollywood glamour into this introduction. We have to do our best to keep it entertaining. How are you?

[Misha BORODITSKY] Great. Wonderful to be here. Thank you for such a flattering introduction. And in return, I have to say that the listeners cannot expect a better instructor than you. I can also say that I read every MaverickQuant free article your bot alerts me to – love the elegance of exposition.

[JB] Well, thank you, Misha. I hope to live up to that standard. So, let's start with a little bit about you, your background and career. Can you give us a sense of how you got to this role at Goldman Sachs and what you studied, and how you got there basically?

[MB] Sure. Normally I would start with a disclaimer about not representing company view, but today, November 17 2024 I am done with my last gig at Goldman Sachs and starting at Bloomberg LLP tomorrow so my views can’t be construed as representing any other party.

Having said that, as Jérôme mentioned, I started in physics. I got my Ph.D. at UCLA, University of California, Los Angeles. It was fun, but very fast I realized that the product of my diligence and smartness is far from being enough to achieve anything good. So I moved to more practical things like telecommunications and engineering and I spent 6 great years at AT&T Labs, improving optical fiber communications, working with really smart people. But at some point it was time to move on and I switched to finance. My first financial job was indeed with Knight Capital markets, a great company to be at.

I actually spent almost 10 years there until, at some point, the trend was over, and I moved on to Cantor Fitzgerald. Then to another brokerage, and I eventually found myself at Goldman Sachs.

[JB] Excellent. And what a beautiful and inspiring career it is. Misha, I think, you've been one of the quite successful quant traders in very challenging times. For example, the 2008 crisis was certainly a breaking point for many people, although it must be said that probably market making wasn't a bad place to be. I don't know what you think about that.

[MB] My recollection was that, I was only two years into finance then, but it was one of the great years -- the greatest years in the history of electronic trading. In fact that was the year my first model went in production, it worked well, and I thought that will be like that for a while.

[JB] Exactly. And that's kind of the irony of the market, it is that some segments of the market can completely collapse: you probably all heard about Mortgage-backed Securities and the subprime loans, and banks being recapitalized and so on. But on the electronic market-making side, certainly those were transformational years, and good years in a broad sense. The crisis probably sped up some of the changes that were long overdue, and especially the electronification of the markets, because there always was some reluctance to moving to more electronic markets. And that certainly has sped up the thing, right? Was that something you’ve witnessed as well?

[MB] I hundred percent agree that electronification sped up, but I would also think that it also led to good changes in regulation and compliance. It's interesting because people often assume that market makers are printing money out of thin air, but the truth is, they actually do serve a very important function. Without educated and technologically advanced market-makers, the markets don't actually function well. Things become more expensive to buy, and more difficult to sell. So, in that sense, I think the changes of 2008 overall, long term, improved all capital markets.

[JB] Yes, absolutely. And we certainly saw that with the dramatic compression in spreads that happened, especially in equity markets, where we both worked at one point. And now, if you're a retail investor, you can basically enjoy a bid/offer spread that's close to zero, right, which was unthinkable 10 years ago. So, we certainly witnessed a lot of change during these 15 years. Can you describe in concrete detail, for our listeners, what your current role is? And can you give us a sense of what a typical day is for you?

[MB] It will be in the past tense now. (I will be happy to report what work at Blooomberg will be like in our next installment.) So let me just clarify that I worked in asset management. It's a one of the four divisions, the one responsible for watching -for a fee of course - after and growing other people's money, and by other people's money I mean individuals', insurance companies', pension funds' and so on.

So, as a fiduciary, the whole team was responsible for some decisions, which happen at multiple levels. At first, one group, portfolio managers, have to make decisions that affect long-term returns: which bonds to buy, which sectors, which countries, and so on.

Then, once the decision is made, a bunch of computations are made and people decide which exact instruments are going to be bought and sold. And at that point, the execution team steps in and actually implements all those things.

So, you can think of the portfolio management thinking on a longer term, and the execution team working on a shorter term: real time data, faster signals, faster decisions, more data. And, for me, that’s kind of more interesting, coming from the sell side, but obviously there is room for every interest and kind of modeling style in the business.

[JB] What is the difference between manual trading and systematic trading? Obviously, they're different styles.

[MB] Definitely, and it is fair to say that as you automate, more things can go wrong. So, on the one hand, you invest time into automating, and at the same time, you have to invest more effort into preventing any kind of error. So all these sanity checks are much more abundant when you trade automatically. And on the flip side, automated trading allows for better testing because everything you have done can be back-tested or replayed using the same software, which is a huge benefit of any algo trading.

[JB] How important is mathematical modeling to your job?

[MB] Modeling is extremely important. There are a bunch of different flavors of what people mean by modeling. And I think for a practitioner, the important pieces of modeling are simplicity, and to some extent, reproducibility. The markets are not exactly like a typical physical system, where you can run an experiment and get exactly the same results many times over.

So what we observe are not true facts, but rather "stylized facts", a term coined by a great French scientist and hedge fund manager, Jean-Philippe Bouchaud. So, the thing is, our intuition and our statistical analysis distill from tons and tons of dirty, noisy data what we come to believe as almost a fact, but which is still expected to change after the next market crash. But I think that, within the range of validity of any of those stylized facts, we can actually build models with a full understanding that we're betting the future will be somewhat similar to the past, but with the full understanding that it's not guaranteed.

[JB] I've always been puzzled at how much finance is really not a field per se, but a meeting point of a number of approaches. It seems that a lot of different people bring different approaches to the table, right? So, you know, economists will bring some econometric approach. You've got physicists, who have their own approach, mathematicians, more traditional finance guys and so on, and even behavioral approaches, which border on psychology, frankly. So, what's your thinking on that? Are these competing approaches or are they complementary?

[MB] Great question. I think those approaches are complementary and in, also in many cases, they actually are different ways of describing the same market behavior -- just observed from different angles. From my past experience, I could give a few instances where I would try to develop a great new model based on some new and exciting research paper, only to get a good working model, but strongly correlated with an existing model developed based on some absolutely different approach. Go figure! But that leads me to think that, indeed, people observe very similar behavior. They're looking at the same markets from different vantage points.

[JB] What do you think are the two or three great challenges in quantitative trading nowadays?

[MB] Too much data is both a blessing and a curse. On the one hand, you get your hands on a new data set from an alternative data vendor, and you spend tons of time sifting through it, and then often you realize that, again, it is pretty much the same thing just, as we discussed before, being looked at from a different angle. Another interesting thing is that we, as quants developing new models, are actually changing markets. So you can definitely say that, if I'm putting in a model, chances are, some other guy is putting in a similar model a few months after or a few months before me. So, the markets will have to have dynamic changes. The good thing is that this will keep us employed, but the bad thing is that our models most likely have a finite lifetime.

[JB] And on the latter point, we can reference the quant crash of 2007. That would be a good example of the market participants’ behavior changing the market. It's fascinating, because it's one of the more documented instances where everybody got into the same positions. Do you think these crashes could be explained by the phenomenon you describe: everybody changing the behavior of the market?

[MB] So, I think the core reason was indeed a strong correlation between everyone's models. It was my first year in the financial markets, so I was just happy that markets didn't go down forever, but it was indeed fascinating. I think the reader can also find lots of much more detailed analysis on the web, because there are tons of really fascinating studies looking at the microstructure, which firm caused which other firm to start selling or buying. But I think it also was a good warning to everyone in this space, that risk controls are extremely important.

[JB] I think that we're changing the regime of the markets with electronification, and we are creating these fat tails, and non-linear dependencies that probably become more prevalent. I'm also interested in hearing what you have to say about the future of data. There's a sense that data is more available than it used to be, and also that there are new ways to handle it, for example, artificial intelligence or big data-type techniques. Do you think there's something to that, or not really?

[MB] Something, yes. How much remains to be seen, because as our computers and networks become faster, it becomes easier to pump data at higher frequency, which doesn't mean that it has to be material or meaningful.

Often enough, we see data feeds where the same price is being printed every second, or every hundred milliseconds, which gives no value, and basically wastes electricity. So, every trade normally happens for a reason. I think every stylized fact are observed at a certain time scale. So, oversampling for the sake of providing higher frequency is more of a burden than a benefit. But, at the same time, of course more valid and reasonable data has contributed to pre- and post-trade transparency. Limits on how fast, say, a corporate bond trade needs to be reported and then disseminated to their market participants. These are all actual and valid and useful data.

[JB] A lot of buzzwords are being thrown around these days, like big data, AI -- everybody's a data scientist now. What do you think of data scientists?

[MB] So, in Equities , where data is more abundant, I do see a bunch of people I very much respect, getting very comfortable with models that are consuming lots of data, act like a black box, give them somehow a competitive advantage, without necessarily being interpretable.

I think it's totally fine in the setting of a hedge fund , where people are given money to multiply, assuming a risk of winning or losing.

In a more constrained and regulated environment, I think there is more need for interpretable models, and Machine Learning as a whole, I believe, has made good strides towards interpretability, but it's not always easy. Take the case of Fixed Income, which happens slowly while Fixed Income and Equity have similar market size; the rate of trading is different by a few orders of magnitude. The actual trades, which are the actual most important data, are coming in a thousand times slower. So, you can think of this as the process running at a much slower natural rate, meaning that your data set is smaller. So, you have to be much more mindful of any kind of overfitting, which comes, naturally, as a threat, with any artificial intelligence approach. Still advanced machine learning methods – not necessarily AI in today’s generative AI sense – are a new powerful tool for feature engineering, be it measuring traffic to warehouse or detecting CEO hesitation in the earning call’s Q&A session.

Over the last few years we saw another boom in models, that changes the way we learn and the way we work. It’s amazing that you can get a decent model prototype based on a research paper from the internet, that would use the data interface you described to the chatbot. To use the new capabilities well we need to ask the right questions is the right way, and be good in checking other’s – I almost said people - code.

[JB] What piece of advice would you give to students who want to launch into a quant trading career?

[MB] Well, number one, I really recommend trying to understand the taxonomy of quantitative finance better. Finance is not a monolith, there are tons of very different jobs, aligned with different personalities, skillsets, appetite for risk, and so on. Figuring out what you want, or what you absolutely do not want, before starting to have interviews, will be very helpful.

Once you've made that transition, to me – at that time - it actually very similar to academia. If you talk to people and you try to collaborate, chances are, whatever you want to build has possibly been built by someone in the group on another floor. So, talk to them, and give people credit when you use stuff that they have built. A piece of advice that I got from one of my managers at Knight Capital Markets, and which grew on me over time, was to start with the org chart. It is very important to understand who's who, not in terms of how to climb the ladder, but whom to ask when you have a question.

[JB] Well, that's certainly sound advice. And finally, if you were now to give students some advice on what they should study, do you think hard science is still the main pathway to a successful quant career, or do you think there are other options too?

[MB] Looking at people around me, I think pretty much anything goes, and I mean studying math, engineering or physics, definitely opens you a pathway into finance. In the same way, I see people coming into finance from, say, software engineering, which again -- maybe they are not as fluent in math, but they 100% make up for it, by developing faster, and in the current world, it is hard to be a modeler without having to code. So definitely good coding skills are a big plus.

[JB] And I think we probably will need a blend of skills. Well, that's fantastic. Misha, thank you so much for coming to our interview segment.