This blog is dedicated to exploring the concept and ideas of graybox trading. Some traders are excellent at reading and predicting markets. These traders make use of primarily two specific skills that help explain much of their abilities: implicit learning or pattern recognition and market cognition which can be summarized as either (1) understanding how the other traders will act and/or (2) the factors that drive their markets.
Quantitative or systematic traders, can, also do quite well. Traders can use their market cognition, thinking about markets, and test those ideas on historical data. The systems that test out in the past and pass the developer’s tests can then be executed with consistency. System testing cannot ascertain which systems will be profitable in the future but they can eliminate extrapolating the wrong expectations from historical data.
Traditional systems consist of algorithms. Algorithms cannot think but can encapsulate specific lines or forms of thought. I think cognitive systems could be the next generation of systems development. Cognitive systems would be able to make use of several qualities that discretionary traders have, and those qualities are (1) human-like pattern recognition, (2) ability to incorporate non-price data into decision making, and (3) ability to infer future price development in ways that cannot be inferred from historical price data alone.
Addressing point three for a moment, for example, traditional technical analysis suggest that when there are more puts in the market that the market is more likely to reverse. The puts in this case can be seen as hinting at another trader’s intentions. This is well-known and unlikely enough to build a strong system on. However, if a system were able to combine such knowledge and use that knowledge to classify price patterns then it starts to become cognitive in nature. The system is starting to think. As an aside, a difficulty is that most of this value-added data is not easily available to test and incorporate into a system. Mike Harris describes the concept of “availability bias” which suggest traders gravitate toward the available methods, and obviously the methods or tools that are not easily available are more likely to infer an edge.
But, how can a discretionary trader take a basic system and without advanced techniques, use their discretion to improve the results? Well, as in a recent post, I suggested that the first step is to decrease the selectivity of the system. The objective is to get in more “reps” to train the pattern recognition parts of the brain. It sounds counter-intuitive because you must basically lower the bar. But, this can be thought of as the cost for the possible reward for being able to improve the results. Another way to think about it if you have a system that was most selective and only took winning trades then how would could you improve it?
It is possible to backtest the above idea, albeit in a limited form. The first step is to take your system that is the candidate for discretionary improvement, and you decrease the profit selectivity such that it generates as many trades as possible with a break even profit. Next, you would need to bin your market data into training, test, and validation sets. Finally you would need to use either (1) market replay or (2) print out the price patterns for the training set. You would study the patterns and the resultant activity. Next, you would test yourself on the unseen test set. You could re-use this set if needed. Finally, you verify your results on the validation set. You compare your results on the validation set against the properly optimized system variant.
Trade costs are obviously a very relevant factor, as well. If costs are high then it is better to trade a selective system which is normally going to be quantitative system. On the other hand, the ability to improvise or use discretion becomes both more possible and more valuable when trading costs are minimal.
Curtis is passionate about markets. He has developed top ranked futures strategies. His core focus is (1) applying machine learning and developing systematic strategies, and (2) solving the toughest problems of discretionary trading by applying quantitative tools, machine learning, and performance discipline.
Please log in again. The login page will open in a new tab. After logging in you can close it and return to this page.