Thalesians Ltd Director (and Founder and Director of Thalesians Marine Ltd), Oleksandr “Alex” Bilokon, has just interviewed Dr Joseph Simonian, Senior Investment Strategist at Scientific Beta. Dr Simonian is a noted contributor to leading finance journals and is also a prominent speaker at investment events worldwide. Simonian is also currently the co-editor of the Journal of Financial Data Science and on the editorial board of The Journal of Portfolio Management. He holds a PhD from University of California, Santa Barbara, an MA from Columbia University; and a BA from the University of California, Los Angeles. He resides in Newton, Massachusetts, United States.
OB: You have just published a book called “Quantitative Global Bond Portfolio Management.” The book is co-authored with Frank Fabozzi and Gueorgui Konstantinov. What are some of the main highlights of the book?
JS: Well, the book is really the first book on global bond investing in quite a while. And the fact that it looks at global bond investing through a quantitative lens makes it even more unique. More specifically, the book offers the most extensive discussion of currency risk in global bond management in at least a decade. It also presents an in-depth discussion of inter-curve yield curve strategies and also discusses how data science approaches can be useful to bond portfolio management. We believe to book does a good job of combining academic rigor with practical utility and thus serves as a perfect reference text for investors, academics, and students alike.
OB: Speaking of data science, you have been involved in machine learning and data science for many years. You are in fact one of the founding co-editors of The Journal of Financial Data Science. One of your primary areas of focus has been on the importance of bridging the gap between traditional econometrics and machine learning. Please explain why you think this is important?
JS: I think this is important because we still have quite a bit of what I call “statistical tribalism” in the investment business. On the one hand, you see some quants express skepticism regarding machine learning and on the other we see another subset of quants, often coming from computer science backgrounds, who have no knowledge of and see no value in econometrics. And I believe that both camps are mistaken as each paradigm is important for producing high quality investment research. Machine learning algorithms are generally better at prediction, while econometrics is often better at explaining what has driven economic events in the past. In my view, having some understanding of both paradigms and how they can complement each other is important to having a comprehensive understanding of economic and financial data and in being able to build viable investment strategies.
OB: You have also been heavily involved in factor investing, most recently with your work with Scientific Beta. In your view, what are the biggest risks in factor investing?
JS: For me, poor model validation is a major risk in any type of quantitative investing, including factor investing. Our business is still unfortunately plagued by many overfitted, insufficiently tested, false positive models masquerading as viable investment strategies. In addition, I believe it is important to select a set of factors that have some economic basis and intuition. This is especially important in equity factor investing. With fixed income you can often effectively create so-called statistical factors; a well-known example is the decomposition of the yield curve by means of principal component analysis. But with equities, there is typically a lot of interpretation involved with these types of synthetic factors, so I tend to avoid them when building models for equity investing. I think the one exception to what I have said is high frequency trading, where statistical factors as I have described them can often be useful.
OB: Another area you have worked on is the topic of causality in economics and finance. Can you speak to some of the unique issues that researchers in finance may encounter that sets them apart from their counterparts in the natural sciences?
JS: Yes, well it starts with the fact that we cannot conduct closed experiments in finance in the way that we can in natural science. Given this, it often becomes difficult if not impossible in some contexts to isolate causes, confounders, and effects with the level of precision we can in natural science. The other way in which causal analysis in finance is different from that in natural science is the fact that we often observe significant lags between causes and effects, in contrast to the natural world where the cause-and-effect relationship is often nearly instantaneous. Now this does not mean that we are at a total loss, because with the proper care we can still apply different methods of causal inference to investment questions. In fact, I have published several papers over the last few years which do this. In one of them we apply a technique called double machine learning to the causal analysis of contagion in financial markets. Double machine learning attempts to isolate the treatment effect of one variable on another in the presence of confounders. However, it is important to keep in mind that when applied to investment questions, a framework like double machine learning must be applied with an awareness of the unique characteristics of investment phenomena. So, I am optimistic about causal research in finance, if we are thoughtful about how we approach the problem.
OB: You have been dedicated to quantitative investing for your entire career. What benefits do you think quantitative investing presents over fundamental investing?
JS: I have always thought that the rigor and discipline that quantitative methods bring to the investment process present distinct advantages to portfolio managers. Quantitative methods give us the ability to structure our investment ideas in a very thorough and systematic way. They also allow us to take emotion out of the implementation of the investment process. And emotion, as you know is one of the biggest failings of investors, a failing that often leads them to poor decision making. Now that is not to say that there are no good fundamental mangers. I used to work for perhaps the greatest bond investor ever, Bill Gross. And he clearly had the “secret sauce” if you will. But I think the vast majority of fundamental managers do not have that level market insight. I should also say, that as time has gone on and markets have become even more complicated and sophisticated, quantitative methods have become even more indispensable. So, while I think that having market intuition is crucial to being a successful investor, sophisticated quantitative methods have really become indispensable to the practice of investment management.
OB: Some quants are enamored of natural science and think that quantitative investing has much to learn from it e.g., physics. What is your view on this matter?
JS: I don’t agree with it for the most part. Finance has spent much of its history trying to mimic the natural sciences, with mechanistic models and simple causal explanations of market behavior. However, as I mentioned in my answer to a previous question, unlike research in the natural sciences, finance does not have the benefit of closed experiments and repeated trials. It is furthermore burdened by a much more complex phenomenon: human behavior. Nevertheless, much of the profession has, over time, pursued a path of creating ever more elegant mathematics without the same level of commitment to producing tools that can actually be put to practical use. I believe this is a mistake. In contrast, the development of behavioral finance has been a more positive development in my view. Now saying all this, I would add that what natural science can teach finance is how to rigorously test ideas. For all its purported rigor, quantitative finance often proceeds in a decidedly unscientific manner when building investment products. So how does science operate? Well, if a scientist has a theory, they will typically only accept that theory if they have made attempts to falsify it through robust empirical and conceptual analysis, including the presentation of counterexamples. If a theory passes experimental and logical scrutiny, then it will be accepted. Unfortunately, in finance we typically see investment strategies, which are our analogs to theories, developed in the opposite manner. Investment teams will often test various combinations of signals, constraints, and asset weightings, choose one that performs well during a backtest, scream “eureka” and try to play off their “discovery” as a viable investment strategy. And then when it comes time to actually perform, these types of strategies will typically fail as they are the product of overfitted, insufficiently tested, false positive investment ideas.
OB: You have been recently been writing and speaking about emerging market investing and how a quantitative approach can be helpful here. Can you share some of your thoughts on this topic?
JS: Yes, as many of your readers know actively managed emerging market equity strategies, especially fundamentally driven ones, have experienced a bout of extremely poor performance over the last several years. And my basic view is that many managers have not been able to keep pace with the changing emerging market landscape. We have seen that some emerging markets have become more-or-less uninvestable – Turkey and Russia, for example. These countries once featured prominently in many emerging market equity indexes but have since seen their weights decrease radically or disappear altogether. On the other hand, India, Saudi Arabia, and Thailand, among other countries, have greatly increased their weights in the major emerging market indexes. And what we have seen is that the investment processes of many fundamental managers have broken down as these changes have occurred. At the same time, we have other types of emerging market equity strategies, quantitative factor strategies for example, that have fared much better. Scientific Beta has such a strategy, and what has driven it outperformance against the standard emerging market indexes is a laser focus on extracting risk premia and an emphasis on risk control and management, which are of course extremely important in the often volatile markets of developing countries. So here again, we see that the advantages of quantitative methods can often be significant if applied correctly.
OB: Finally, you are currently working on a book called where you apply game theory and machine learning to global macro investing. How does it approach the use of geopolitical information in investment decision-making?
JS: There have really been two approaches to using geopolitical information in finance. One is what I call the “soft” approach where there is no formal rigor to geopolitical analysis, just individuals editorializing on current events. This is the most useless type of geopolitical analysis, especially when it is done by investment professionals with little knowledge of geopolitics. The second approach is what I call the “hard” approach where geopolitical analysis is forced into the mold of finance, with time series models etc. I also believe this approach is misguided, given the unstructured nature of political information and the lack of adequate data. Analyzing financial time series is hard enough without pretending that we have the means to competently apply time series analysis to geopolitics. So, in the book we propose a third way, what I call a “structural analysis” of geopolitical events, which is driven by computational game theory and reinforcement learning, among other algorithmic approaches. The idea is that by applying the foregoing methods to geopolitical questions within computer simulations, we can start to get an idea of the likely decisions of geopolitical actors over various time horizons, information which can then be used to inform our investment decisions. So, the aim of the book is to try and inject some rigor into the analysis of geopolitics without pretending that we can analyze geopolitics with the tools of statistical inference. Now because the book is addressed to global macro investors, we also show how all of our ideas can be expressed in a portfolio context. So, while the book has some technical content, it is also a very practical book.