The Gateway to Algorithmic and Automated Trading

Rage against the machine? What experts say about modelling for emotional markets

First Published 12th September 2012

Are human beings poor decision-makers? When is it better to let a machine take over? Find out what some of the experts say when it comes to building models that grapple with irrational behaviour and extreme events.

Panel members (from left) are Scott Reamer, Tigrane Kibarian, Paul Domjan, Felix Gasser, Rafael Molinero, Bradford Paskewitz and John Moody. Photo by Battle of the Quants photographer D-Mo Zajac.

London - When it comes to trading financial markets, there's little doubt that machines have earned the respect and admiration of their carbon-based rivals. But not every problem is best solved with models and raw computational power.

At this week's Battle of the Quants in London, a large group of executives and experts gathered to offer their thoughts on when it makes sense to adopt systematic approaches to dealing with some of the thornier issues that come up in financial markets.

Below are some of the highlights of a roundtable discussion on the virtues and occasional vices of automation.

John Moody, CEO of J E Moody & Company: Our price strategy is actually systematic but based upon commodity fundamentals. The key insight there is that in the commodity markets supply and demand affect price in ways that you don't see in the financial futures. So there's actually a big distinction between trading commodities and financial futures.

We find that by having a model structure that's based upon market fundamentals, that we end up with much more robust performance in the end. In terms of whether the rationale changes over time, there are elements of the markets -- essentially the global macroeconomic background, the demographics of investors participating in commodities, exogenous factors, general investor sentiment and expectations - that do change over time. But certain fundamental truths don't change, and those are the laws of supply and demand. So our models seek to try and capture, in a way that constrains the range of possible trades, those things that are constant over time. But in order to keep up with the markets and changing market dynamics, we obviously have to have a re-estimation of parameters.

Tigrane Kibarian, COO of Odin Capital Management: Actually, the most important thing is to understand the market environment you are working in, because that is actually the most important driver of your performance. So, at the end of the day, you need to have a system, (and) understand the limitations, the failings of it. Where we basically differ from amongst others, is that we will try and consider this market environment as well as the changing market relationships in a very active manner. Because some correlations change, the inter-relationships within asset classes and across asset classes would change. And trying to work through this to look for trade ideas, why this is happening, I think works pretty well for us. Some things might be obvious for a human mind but aren't at all for a machine. I mean, how can you describe Lehman before it even happens? And before that, how would you describe LTCM before LTCM happens? For a human mind this is incredibly obvious and the implications of it are then where you make your money.

Felix Gasser, head of managed futures and portfolio manager at FRM: It's important to see that it's not intelligence that you give to the machine, but you really just give it the framework and the heavy load of computation within the framework that you figure out, to make sure that if you get into price moves that go against you, you get out of them on time. And also, if you get into price moves that go in your favour, you have to understand you cannot forecast how long the move will actually endure. Therefore you have to give this job to a machine to make sure you go in on time and get out again in a systematic way.

Paul Domjan, managing director of Roubini Country Insights: If you think about our space, which is understanding fundamental country economics, the machine is very good at - our systematic models are very good at - reflecting balance sheets.

We probably understand them better than the analyst because they (machines) can capture a much wider range of data to model that balance sheet. On the other hand, they're very bad at forecasting policy change, they're very bad at making judgements about politics. We can try to do that systematically, but you begin to bring that in as a component to understanding, for example, macro strategies.

We would argue for a combination of a systematic approach and also the expert judgement.

If you were to ask a question like, where do we draw a cut line to get out of the European periphery, it's very very difficult to do that based on news flow. News flow is going to be changing so quickly. Part of that will become policy, part of it won't. I think fundamentals provide a much stronger basis.

Bradford Paskewitz, CEO of Paskewitz Asset Management (when asked about when to change models): Basically, we monitor our strategies on a monthly basis. That said, one of our core strategies on S&P I've traded for something like 21 years, without changing the models at all. So it's not necessarily the case that every model has to be tweaked or fine-tuned. I prefer to keep them stable and steady, barring some indication of failure.

Scott Reamer, CIO of Rotella Chora: Markets are made of human beings and interacting human beings … are irrational, and persistently so. When you aggregate a lot of irrational agents that then produces a type of behaviour that is quite anomalous. It happens to be unusual but also it happens to be, in some cases, somewhat predictable. That's what we focus on. Emotion, underlying those price changes, is the absolute core of markets. It's the reason, for me, that markets are complex, social, adaptive systems, prices simply an emergent factor of all those local interactions. They're driven by emotion in an information-infection sort of way. One person's risk preference is transferred to another until everybody's buying or selling the euro, or whatever.

If you're a discretionary global macro investor, you might wake up one day and have a bad day and make a bad call and that can wipe out 10 years of your performance. To me systematic investing is the answer to the fact that humans are persistently irrational and in groups manifest in certain patterns. Those patterns are what we analyse and try to predict.

Tigrane Kibarian of Odin Capital: You (Scott Reamer) said one very interesting thing, which is that emotions actually drive discretionary investors and I think this is when it becomes a problem. If, instead of that, you consider emotion an input, then you might actually take advantage of it. So how would that translate? It's basically when you realise that something is wrong. Something has gone wrong - you just cannot put your finger on it, but something is wrong. And this is usually where the opportunities are.

Felix Gasser of FRM: Humans are not bad decision makers, but we are probably not conditioned to make good decisions in an algorithmic or numeric framework. We are evolutionarily conditioned actually to respond to physics. It takes us back to the emotions. There's actually a reason we respond emotionally, because, in previous lives we had to respond in a very violent way. There's a logic behind this. If you're being attacked by something that you don't know how fast it moves - so you have an x there - actually the more that we respond the better, so we produce adrenaline. It's the other way round as well. If you attack...the stronger you hit or attack, you have a higher chance to win. So we have this exaggeration of response that's built in. Now in trading, this doesn't work. That's why, when you get emotional, you will make poor decisions.

We are not bad decision-makers in life outside, but I think in a new world, which is numeric, we need new guard rails and a new framework.

Rafael Molinero, CEO of Molinero Capital Management: (In some cases during a crisis) people will stick to their decisions, and will have a hard time going from long to short. Models don't have this issue. When you have a crisis, and the crisis is long enough, you can benefit from that on the systematic side. Having said that, you do have extreme events where, let's say, a politician will speak or something like that that's unpredicted. That's definitely much harder to capture unless you're extremely short term.

Scott Reamer of Rotella Chora: We differentiate between extreme events that are of an endogenous nature and those that are exogenous. No one can predict a geopolitical event, no one can predict sunami earthquakes. Our hypothesis is that markets are complex and social systems, that people produce the endogenous events, the black swans, the tail events, the multi-sigma events. Personally, and professionally, we don't care what happens in the middle of the distribution. We only care about what happens on the edge of the distribution.

Our hypothesis is that most of those ends of the distribution are a function of that endogenous (origin). Someone is following somebody else, someone's risk preference is affecting 10 other people. When hamburger triples in price you don't buy less of it, but when Apple triples in price people buy more of it.

How you take advantage of those (endogenous extreme events) to me, is entirely a systematic strategy. But they're sourced from an underlying emotional irrationality that I think is central to understand so you can find the right mathematics. So extreme events are all we care about, and whatever scale that you're looking at, whether it's a 10-minute scale or a 10-year scale, for us they come from that endogenous herding process. That's what we try to model.

Paul Domjan of Roubini Country Insights: If you think about extreme events over the last couple of years, or the last decades - Asia, Russia, Argentina, Europe - most of these are driven by economic issues in the countries.

If you're going to try to reflect them by picking up the fundamental drivers, some of those fundamental drivers are just representing the data. And either - because the pattern series is too short or the lead-lag is too unstable or there are inter-relationships that are difficult for a system or systems to capture, or the data is too noisy - you can't do (it), from an optimisation statistical perspective. And I think that's part of what drives you into a certain quant approach, that involves algorithms that reflect executing decision-making.

John Moody of J E Moody & Company: I think it's not so much a market environment issue as a trading style issue. If you have a model that can capture a lot of repeatable events, a systematic approach works well. Whereas human traders are more likely to perform better on one-off trading situations.

Another general comment is that systematic modelling is much better at risk management than discretionary traders tend to be. If you look at the commodity markets, all the major blow-ups in the last 30 years that have been headline events have been due to, or accomplished by, discretionary managers. So that includes the Hunt brothers in silver, Metallgesellschaft in crude oil, Sumitomo in copper, Amaranth's 2005 disaster in natural gas, Ospraie's flagship fund's 2008 collapse with the commodity bubble, and most recently in 2011, Blue Gold's huge loss in energies.

These are all discretionary managers that blew up because of poor risk management. And I believe a systematic approach can capture alpha and at the same time bring disciplined risk management. It seems to fly out the window more often than it should in the discretionary space.