The Gateway to Algorithmic and Automated Trading

Perfect Asymmetry?

Published in Automated Trader Magazine Issue 21 Q2 2011

Dalton H Mota runs Asymmetrica, a research consultancy in Brazil. He’s a trader as well as a researcher, and finds time to hold down a professorship too. An engineer by (early) training, he’s consulted by global investors and corporations, and he got his start trading popcorn futures. David Dungay went to meet him.

Dalton H Mota

David Dungay: Tell us about yourself and how you got into quantitative research.

Dalton H Mota: I studied engineering at first, then switched to business school. I did business analysis for a global consulting firm for a few years, then I was approached by a big financial-risk consulting shop to create and develop custom quantitative tools for large asset managers. After four years of this, and completing a Masters in Finance, I planned to apply for a quant PHD in the US or UK, but began trading my own money instead. I already had quant experience, my trading went well, and some consulting projects were happening, so I just kept going. My research shop, Asymmetrica, brings together the things I love doing, financial research, statistics and computers, and trading. By trading real money I can become a better financial researcher. By doing high level applied research I become a better trader. And by interacting with global clients I share my experience and learn too.

I use my quant models and research to trade my own and close friends' money. My trading activity currently generates the bulk of my income. I am not a hedge-fund or institutional asset manager, and right now, I am not registered to manage client´s money. I am also a finance professor at FGV Mba (Getulio Vargas Foundation) in Rio and Sao Paulo, teaching executive finance classes since 2005.

David Dungay: You started trading very early, I believe?

Dalton H Mota: Growing up in the eighties, in a country with tremendous economic volatility, had advantages. I had a premature financial education.

When I was around 11 years old, Brazil was experiencing hyperinflation, like 15% a month. Amazed by the price increases, I began tracking in my old Lotus 1-2-3 the prices of the goods I consumed every week. This was my first quantitative analysis. I realized popcorn prices were going up more than the soft drink. In my childhood naivety, I had the idea of negotiating with the street seller to buy popcorn for three-month delivery in the future, locking in the price, just to sell at the future date and be able to buy more drinks. That was my first futures trading. I actually did this , but the price momentum switched, and I ended up being able to buy fewer drinks than I had expected. That was my first encounter with financial risk.

David Dungay: How do you distinguish between what you do and algorithmic trading?

Dalton H Mota: We do quantitative financial research. In lower frequencies, we use models to identify the market regime for an asset: is it a bull, bear or sideways market? Is volatility more likely to go higher or lower from current levels? Conditioned on these variables, the system chooses an appropriate higher-frequency model. So, conditioned on a lower-frequency model, we trade the signals generated in higher frequencies. What we do is algorithmic in the sense that we strictly follow our own rules, implemented in software.

David Dungay: How do you generate your ideas and do you have a formal pipeline process for taking those ideas into production? Tell us about your back-testing methods also.

Dalton H Mota: As a small research house, we have a lot of flexibility to do new things, but a really robust trading process is not easily built. I read a lot and observe a lot. You can´t forget the fact that a market is essentially a human expression. Context is important. When I have an idea, it usually takes some time to mature. I do all the backtesting in MatLab. The backtesting process is done using commonsense goals, profit factor, risk of ruin, et cetera. We also walk forward every system, and paper-trade them in real-time. So, it's a very intensive process before we start trading with real money.

David Dungay: Which markets and products do you trade? Why these markets and do you have plans to branch out into other areas?

Dalton H Mota: We trade index futures and strategies with vanilla options. We have found these instruments to be suitable for the type of trading we do, essentially volatility trading and trend-following variations. We currently have plans to extend trading to other areas, but that's always a possibility.

David Dungay: Do you buy or build your technology? Why?

Dalton H Mota: We code every piece of software we use, from collection of financial data, to analysis and generation of trading signals. As for hardware, we simply don´t need a very expensive hardware infrastructure for what we do. A few fast computers will do the trick.

Dalton H Mota

David Dungay: What is your preferred programming language? Why?

Dalton H Mota: I have been using MatLab since college, and I am very impressed with the evolution of the product over the years. Since we don´t do ultra-high frequency, it is suitable to our research needs and trading systems.

David Dungay: How do you quantify risk into your models. Also, how have recent global events affected their performance? Do you have metrics to adjust to events like Libya and Japan or do you just stop the model trading?

Dalton H Mota: We use some simple metrics, like average true range, and we also do scenario simulations. We have a few other proprietary market-wide risk models. Our signal-processing model has been useful in anticipating changes in short-term volatility behaviour. We don´t use parametric Gaussian methods.

In volatility spikes that we can attribute to well-defined global events, such as Libya/Japan, the whole process switches to a different type of trading where economic analysis plays a greater role; that is, we see if we have any edge based on information we have, and look at how we could exploit statistical opportunities opened by the event.

What is short-term price volatility from a fundamental point of view? For me, it´s the degree of disagreement among agents in a given time window, adjusted by liquidity. A higher volume of transactions happens until short-term prices clear the change in risk aversion of one group of agents ( the sellers ), with the end vector being another group of agents ( the buyers ) holding the assets, the event risk, and a higher expected return; that is, a premium for bearing the additional perceived risk. The risk-neutral probability may become very divergent from the true real probability.

So, it opens room for you to make a good return, if you have a different opinion, based on a different information set or different quantitative insights, and some kind of quantifiable edge.

David Dungay: What is the average life expectancy of your models?

Dalton H Mota: A few models died very early. As a basic rule, the more parameters one model has, the shorter the expected life, because the probability you are overfitting data increases a lot. So if you identified essential themes, deeply rooted in observed market characteristics that persist over time, and adapted a robust model to exploit this behaviour, there's a higher probability it can survive.

David Dungay: What time frames do you trade in, and do you have any plans to delve into HFT?

Dalton H Mota: Models generate signals at an hourly frequency, but the holding or flat period can be a few hours or days. Volatility trading with options can have a longer time frame. I would have to think a lot about doing HFT, because it's a whole different game from what I do, and I would probably need much more capital, so it would be a very different business from the one I have now, which is essentially an applied research business.

David Dungay: How instrumental is agent-based modelling in your quantitative systems and can you tell us how it has helped you pioneer new strategies?

Dalton H Mota: For an applied researcher who also trades, agent-based modelling is an essential simulation tool to change perspectives. It is a more realistic map of reality than conventional economic theory.

Trading and real business investing is very difficult. ABM helps on the soft side, in terms of a better understanding of the underlying forces that shape the world we see now. And when you think outside the box, you increase your odds of coming up with new ideas, or connecting seemingly unrelated dots. So, my experience shows me that real decisions and agent simulations feed on each other.

David Dungay: Please tell us more about how you use low-frequency fundamental models in calculating economic risk in real time?

Dalton H Mota: To my approach, a fundamental economic model is an essential component. I think it is important to understand why people make money through trading, what kind of risks they are being
paid to hold, how their strategies fit into the great scheme of things. Everything has an economic purpose, even if one´s not aware. As I see it, there´s very little true alpha in the global market, only multiple betas, most we don´t even understand and can't estimate.

And I try to understand the beta I want to be exposed to. For example, when you provide liquidity - buying stocks for low prices at the depths of recessions when fear is tremendous, banks are failing, people being laid-off, cutting expenses, et cetera - you are performing the legitimate real economic function of being an insurer; that is, holding risks other people are happy (or need ) to transfer to you. For that task you receive a premium in the form of a huge expected return.

As agents see the situation stabilising and risk fading, stock prices go up, because agents see premiums implied into stocks as attractive, buying the assets and lowering expected returns in the process. If you bought at the lows and held, you pocketed the change in prices that resulted from the change in expected returns and perceived risk. If you traded in the months between the low and the recovery, economic risk awareness would have kept you on the right side of the trend, and your quant model would have behaved according to this market regime.

Dalton H Mota

So, we map a few economic and market indicators to estimate the broad economic expectations more likely being priced into asset risk premiums at a given point in time, and this model helps us assess the more probable market regime. The pricing of economic risk is mostly a low-frequency process, and it is reflected in practice in what traders call a bull or bear market. From a theoretical point of view, it is all a repricing of economic risk, even if risk is myopically priced, like it is in bubbles.

David Dungay: How does your approach to quantitative trading systems give you that edge over the field? Is there more opportunity in the way you analyse data than perhaps a typical quant would find?

Dalton H Mota: I don't know, what is a typical quant? I am only concerned with doing the best I can. Trading is so difficult that every year you make money is also a year you learned something about survival and adaptation. Our research has a lot of number crunching, but with a lot of understanding too.
The quantitative framework structures the thinking, the information, the features of behaviour, and generates the signals. But the most important thing is a scientific mind, to look at the world in a scientific way, and be always open to the realities of the market. Human intuition is essential, but betrays you a lot, so quantitative trading keeps noise and emotional bias out.

David Dungay: What is a typical consulting project for you?

Dalton H Mota: We are approached by global investors and corporations. To the first group, we extend some of the proprietary quantitative research and economic analysis we do for ourselves. Also, we can develop custom-made quantitative tools if needed. Non-financial businesses look for broad financial and economic perspectives, and specific high-tech decision-making tools.

David Dungay: How automated is your end product? Can you explain the differences between automating the process but not necessarily the order placement?

Dalton H Mota: We are a quantitative financial research shop, that develops, tests and uses strategies, and shares knowledge and analysis with other agents through consulting. We are more on the intelligence side of things.

A bit of terminology here. I understand 'automating the process' as feeding all the information to the computers and generating a signal, a position-sizing, a stop-loss, a probability distribution of profits, et cetera. As we do not trade very high frequencies, nor trade a large number of assets, we don´t have the need directly to automate the order placement, although we could do so, and probably will do so to some degree in the future, if the number of instruments we trade grows.

I am a small prop trader, financial researcher and consultant, and so far things are under control this way.

David Dungay: Do your models have any self-adaptive elements to them?

Dalton H Mota: They all change with the environment, so in a sense, you can say they self-adapt.

David Dungay: Do you deliberately avoid or minimise the use of fixed parameters in your models?

Dalton H Mota: Yes. We try to use variable parameters or different sets of fixed parameters when possible. Economic risk modelling uses time-varying parameters. Short-term trading with futures uses fixed parameters, but they are only fixed for a given identified market and volatility regime. When these change, the model uses a different set of fixed parameters.

David Dungay: Dalton, thank you very much.

Dalton H Mota