Quant-Trading Primer

From discetionary traders to equities research analysts – good traders are relying on strict rules and processes to determine what they will buy, how much of it, when they will sell and what the risks are expected to be. In some cases more than others, some of these rules are too hard to quantify and code. Perhaps the most insightful example of such decision-making would be the ‘gut feel’ that a private equity investor applies when talking to management of a firm they are interested in. A difficult process to code to be sure, yet even in this case, the investor is sub-consciously crunching data, and experience has defined the parameters.

In the exchange traded markets, decisions can be made much quicker than when one is weighing up whether to buy or sell a particular private company. A greater number of decisions can also be made, given all the available options, and as the number of transactions increases – the law of large numbers begins to play a role and efficient processes become all the more important. The law of large numbers simply tells us that, for a particular series of events, as the number of observations increases, the average outcome of those events gets closer to the “expected value”. In the case of a coin flip, the more you throw it, the number of heads you see will approach ever closer to 50%, which is the expected value. If an investor expects to make 7 great investments out of 10, on average, and he or she only makes 10 investments in a year, there is little confidence we can have that in any given year 7 of those investments will be positive. Perhaps it will be 8, perhaps it will be 3. On the other hand, if they make 1,000 investments a year, and the process is strict, then we can be much more confident that about 700 of those “trades” will be good. Working with large numbers has its advantages; giving us more confidence in our expected outcomes.

The markets give us an absolutely enormous amount of data – data that has been meticulously collected back through history. With all of this at our disposal, how can we not be expected to poke around in it and ask some obvious questions: “how often does such and such an event occur?” and “what happens immediately after said event?”. When we can comb through data in seconds, and get detailed answers to all of our questions, we can look for anomolies in a way we could never have dreamed of in the past. Further, we can test our assumptions, our views of the world. The moment I personally discovered this capability, there was no turning back. As a discretionary trader I had continually sought out patterns that repeated themsevles, but I knew I was reacting to each one differently, or failing to see things in the moment, and my trading was perfect in hindsight only. By asking the computer to do the searching and testing, I was forced to become extremely prescriptive about how I would act in any given situation, and then let the computer tell me the honest truth about “what would have been” if I applied those trading rules over the last 20 years of history. “Oh, I’d have lost a lot of money? That’s good to know!”

By trading systematically, not only are we exploiting the nature of probabilities, but we are also enforcing discipline. Emotions are not part of the trade. Sick-days don’t factor in either. Let’s face it, we live in an age of automation, so anything we can automate, we should. If the computer can do it, it frees us up to do more research, or other activities that the computer can’t do.

There is also a great deal of transpency with algorithmic trading, despite the media hype and talk of ‘black boxes’ and Artificial Intelligence taking over. When an algorithm is assigned to the trading task, we know it will do the same thing, each and every day. When an individual is behind the wheel, we actually don’t know that they will follow the trading/investing process they promised us they would on any given day. With an algorithm we can ask the question “so show me how you would have performed over the last 20 years” and it will answer, it will answer truthfully too.

An entirely new suite of risk management tools open up to us in the algorithmic trading world. We can monitor a number of different portfolio metrics and continuously compare them to expectations. When then overlay the human element when we make judgement calls about how to interpret this data, so the human element is never lost.

The truth is, that despite the fact that humans come first (with the grand ideas, or with their hand on the kill switch), we consistently get beaten by machines. That’s why we invented them in the first place! Cognitative biases are plentiful (we go into that on the About page) and issues like over-confidence through to confirmation-bias are notoriously hard to shake. Take this wonderful little example from the medical industry:

A group of doctors gave some researchers in Oregon a simple list of seven factors which could be used to determine whether an ulcer was malignant or not. They created a very simple agorithm, equally weighting these factors. The researchers then tested the doctors by asking them to judge the probability of cancer in ninety six different individual stomach ulcers, on a scale from “definitely malignant” to “definitely benign”. Unbeknown to the doctors, the researches actually showed them each ulcer twice, mixing up the duplicates randomly through the samples so the doctors wouldn’t know they were looking at the same ulcer as they had before. The researcher’s goal was to see if they could create an algorithm that would mimic the decision-making of doctors.

The UCLA analysed the data of the doctors and the algorithm and the story got interesting. Firstly, even the first attempt at a simple model proved to be extremely good at predicting the doctors’ diagnoses. The doctors might want to believe that their thought processes were subtle and complicated, but a simple model captured these perfectly well. That did not mean that their thinking was necessarily simple, only that it could be captured by a simple model. Surprisingly, the doctors’ diagnoses were all over the map: the experts didn’t agree with each other. Even more surprising, when presented with duplicates of the same ulcer, every doctor had contradicted himself and rended more than one diagnosis: these doctors apparantly could not even agree with themselves.

So basically, if you wanted to know whether you had cancer or not, you were better off using the algorithm than asking the radiologist to study the x-ray. The simple algorithm has outperformed not merely the group of doctors; it had outperformed even the single best doctor.

There’s no one best way to trade or invest, but there are plenty of advantages to algorithmic trading. Explore the blog posts on the Resources page for more in-depth articles. Whilst there is plenty of money to be made by discretionary traders, or fundamental investors, we believe that properly harnessing all the tools at our disposal today gives us a serious competitive advantage, and allows us to produce better risk-adjusted returns than we have seen anywhere else. The human element is still essential: we build the models right! We are always seeking that continual improvement and refinement that the systems just can’t do on their own (yet!).