I have witnessed a disturbing trend in the industry of late: the over-optimisation or curve fitting of trading systems. This practice involves tweaking the parameters of a system to fit the historical data, often resulting in unrealistic and inflated performance metrics. Optimisation is a dangerous and misguided approach that can lead to disastrous outcomes for traders and investors.
Over-optimisation is based on the assumption that the past performance of a system is indicative of its future performance. However, this assumption is flawed for several reasons. First, market conditions are constantly changing and evolving, and a system that works well in one period may not work well in another. You may have tested your system in a bull market and when the market swings bearish your system flounders. Second, the historical data is often noisy and contains random fluctuations that are not representative of the true underlying patterns. Third, the optimisation process itself introduces a bias in the system, as it selects the parameters that maximise the fit to the data, but not necessarily the robustness or stability of the system.
The pitfalls of over-optimisation are manifold. First, it can lead to false confidence and over-trading, as traders may believe that their system is infallible and can generate consistent profits. However, when the system encounters new or unexpected market situations, it may fail miserably and incur large losses. Second, it can lead to curve-fitting risk, which is the risk that the system performs poorly out-of-sample or on unseen data. This risk is especially high when the system is optimized on a small or limited data set, or when it uses too many parameters or indicators. Third, it can lead to over-complexity and fragility, as traders may add more and more features or rules to their system to improve its fit to the data. However, this can make the system more difficult to understand, maintain and execute, and more prone to errors or breakdowns.
The failings of over-optimisation are inherent in its nature. It is a form of data mining or cherry-picking that ignores the underlying logic or rationale of the system. It is a form of hindsight bias that assumes that what worked in the past will work in the future. It is a form of over-fitting that sacrifices generality and adaptability for specificity and accuracy. It is a form of illusion that creates a false sense of security and success.
Optimisation is retrospective – it is not prospective, and it is this feature that traders fail to understand. Instead of looking for the single magic variable or tool that will cause them to print money, they should focus on developing simple, logical, and robust ideas.
