Exclusive Interview | Expert Quantitative Strategist and Researcher

Artur works as a Quantitative Strategist at the Swiss wealth management company Julius Baer in Zurich. His focus is on quantitative models for systematic trading strategies, risk-based asset allocation, and volatility trading. Prior to that, he worked as a front office quant in equity and credit at Bank of America, Merrill Lynch and Bear Stearns in New York and London with emphasis on volatility modelling and multi-asset derivatives valuation, trading and risk-managing. His research area and expertise are on econometric data analysis, machine learning, and computational methods with their applications for quantitative trading strategies, asset allocation and wealth management.

He has a PhD in Statistics focused on stopping time problems of jump-diffusion processes, an MSc in Industrial Engineering from Northwestern University in Chicago, and a BA in Mathematical Economics. He has published several research articles on quantitative finance in leading journals and is known for his contributions to stochastic volatility and credit risk modelling. He is a member of the editorial board of the Journal of Computational Finance and keeps a regular blog on quant finance and trading at artursepp.com.

Grace: Could you just start off by telling us a bit about your educational background?

Artur: I have a PhD in Statistics and three masters’ degrees – one in Statistics, one in Industrial Engineering and one in Financial Mathematics. I have been exploring different fields and have taken any opportunity to expand my knowledge base. As for my professional career, I started at Merrill Lynch in New York back in 2007 and I started doing quantitative modelling for structured credit derivatives. Sure it was right at the tip of the bubble, but being on the trading floor during the financial crisis I learned to understand the human aspect of trading, the limitations of quantitative models, and the importance of risk management. Later I spent six years at Bank of America Merrill Lynch in New York and London working on quantitative models for volatility trading and option book management. The last three years I have been at Julius Baer in Zurich focusing on systemic strategies and client portfolio advisory.

“…being on the trading floor during the financial crisis I learned to understand the human aspect of trading, the limitations of quantitative models, and the importance of risk management.”

Grace: Your bio highlighted your contribution to stochastic volatility in credit risk modelling. Can you tell us a bit about that?

Artur: When I was doing my PhD in Statistics, I wrote a few research papers that had good traction. I found that the experience of writing a research paper taught me to be punctual about details and explanations. Actually, the necessity to explain to outsiders my ideas made me look at the topics and models with more scrutiny and it improved my outlook as a result. My recent goal is to fill the gap in the research on stochastic volatility because there is very little knowledge about how to apply models in practice for out of sample modelling. On credit risk also, I think a lot of research was done prior to the financial crisis but it was simplistic and after the financial crisis I think all the research dried up.  Nevertheless, it’s an interesting idea and it’s obviously very mixed in the past.

“My recent goal is to fill the gap in the research on stochastic volatility …”

Grace: I know that you did a presentation about three quant strategies. Those are equity bond and credit carry, volatility carry, and trend following. Can you discuss those briefly and how you use them?

Artur: Bonds, credit, and commodity futures carry are classic strategies based on the expected cash flow. For bonds the cash flow arises from coupons, for forex it derives for forward rates, and for commodity futures it comes from the futures term structure. The nature of these strategies is that, if nothing happens, we get the implied premia. Therefore, the carry strategies perform well in the low volatility regime. On the other hand, for the volatility carry strategies we should look at markets where realized volatility is expected to mean revert, while the implied volatility is rich enough to compensate for risk taking. In my experience and my implementation, volatility carry must be considered a mean-reversion strategy. Trading implied volatility from both the short and long side should be focused on the volatility mean reversion not price mean-reversion. Finally, trend following is a very well-known strategy that has worked for a long time. The driver behind the performance of the trend-following strategy is autocorrelation in prices; when asset price changes today are highly correlated with recent changes. It is based on the statistical significance of how price changes are correlated over short periods of time. If the autocorrelation is positive, then asset prices are conditionally predictable, and the trend-following strategy is expected to perform well.

Grace:  Can you tell us about how you go about developing your trading system?

Artur: From my perspective, it is most important to look at strategies on the overall portfolio level and it is important to understand the regimes when strategies are expected to outperform and underperform. I have already mentioned carry, volatility carry, and trend following. So, for an example, if an investor owns equity portfolio with a strong momentum then obviously you look for something to mitigate that risk. The strategy must be customizable and the important thing is to try to understand exactly what you are trying to achieve. For an example, is it an alternative beta, hedging strategy or absolute alpha? For me, the most important thing is to understand the cyclicality of the strategy. Also make sure that, if you include the bad periods, the strategy performance is robust overall; because some systems fare very well in good periods but once a bad period occurs, they lose everything.

Grace: Can you talk more about the research you’ve done on using machine learning for models for volatility prediction?

Artur: The first stage in systematic trading of volatility comes from the ability to forecast volatility. There are a lot of different models for volatility predictions like simple ones, such as close to close estimator, then more advanced GARCH model, and Markov-Chain regime-switching models with volatility regimes. I have created a machine learning tool that analyses these models based on various tests. One of the most important tests is to measure the volatility forecast with a one or two step forward.  And then it is also important to model volatility using the same model in each asset class. Therefore, it is a big data mining process that should look at different models, at different assets, and different periods and make recommendations for a volatility model. And then that model is applied in the decision making for selecting optimal trades and for the risk management of options book. That is not only applicable for options trading but for all quantitative strategies that use volatility for some sort of decision making process, either for risk minimization or portfolio construction. I construct my portfolio based on the signals that I receive.  For the risk metrics, I typically use different volatility models that is not related to my signal generation. What is also important is that the system is learning over time. There are different cycles, some models perform differently so within each cycle I am sure I am getting the best volatility forecast that I can use in other models and other applications.

Grace: In regards to machine learning and quantitative finance, where do you see that in the next decade?

Artur: Currently there are several directions where machine learning is applied; in finance, banking and asset management. The first is automatization of back office functions where it is not so much learning from data as it is just streamlining processes and making systems adaptive. The next application is the reduction of discretionary decision-making for research, investment advisory, and asset allocations because right now a lot of these things are still done manually.  Many of related actions are repetitive any way. Therefore, there is a lot of flexibility for optimization that can be done. I think what is important is that most investment banks are now working on the optimization of trading. For example, with central risk books, the process of inventory management or processing exchange orders is being automated. The biggest area for this optimization is cash bonds, and OTC markets like FX Swaps. The final application is fitted to the investment strategies, in particular, factor-based strategies. For an example, for the equity value factor, there are a lot of choices to be made about what kind of balance sheet data to apply. Because for any factor there are lot of possible attributes, machine learning helps to select features with the most predicative power and make decision rules adaptive.

 “machine learning helps to decide the best course of action and make your rule adaptive.”

Grace: You had mentioned factor-based investing. Can you share more of your thoughts on factor based investing and where you see it in about ten years?

Artur: I think there is a lot of potential behind the factor-investing. I see a strong demand from the institutional side. For an example, we currently see a push towards market-neutral factor strategies, where you hedge the market impact. If you want a momentum strategy, you want only pure momentum without the market impact. I think a lot of these ideas are now coming into the fixed income, credit and other markets. As the liquidity improves, the more we can apply it. I also I think another big topic is application of these factor-based rules for less liquid strategies like private equity or real estate. I think there is a lot of potential is applying the factor-based approach to client portfolios – whether it be high net worth individuals or smaller institutions where you make factors part of your rule-based investment and retirement planning. For example, dividends are a quality factor where the investor may not rely on market gains but on a steady income from the portfolio. As a result, the investor should screen for companies with sustainable dividends and sustainable earnings with relatively small market beta.

Grace: I’m curious, do you think traditional discretionary investing is outpaced?

Artur: Yes, it’s very difficult especially for people who are just starting in this space. It is probably much more difficult to become a discretionary portfolio manager. On the other hand, it is somewhat easier to become a quantitative portfolio manager if you have some proper skills. In my opinion, whenever you have good ideas, maybe on the discretionary side, there are almost always ways you can quantify it and make it robust. Once you quantify, you can explain it and you can optimize your decision-making process.

“In my opinion, whenever you have good ideas, maybe on the discretionary side, there are almost always ways you can quantify it and make it robust.”

Grace: In regards to your trading, do you happen to have a preferred coding language or platform?

Artur: I think there are two layers; one layer is research and the other layer must perform fast, have a lot of libraries and have a good connection to data sources. For the first layer I use Matlab and Python. For the next layer, regarding the execution where the speed is more important, I use Java.

View Artur Sepp’s website here.

Ready to start coding? Access a free API-enabled demo to automate your trading strategy.


Risk Warning: The FXCM Group does not guarantee accuracy and will not accept liability for any loss or damage which arise directly or indirectly from use of or reliance on information contained within the webinars. The FXCM Group may provide general commentary which is not intended as investment advice and must not be construed as such. FX/CFD trading carries a risk of losses in excess of your deposited funds and may not be suitable for all investors. Please ensure that you fully understand the risks involved.