CORRELATION VS. CAUSATION
Theo Wyld, Research Analyst
Correlation vs. causation; what is actually the difference? To illustrate this, consider the below example.
Suppose you are consistent in the times that you eat your three wholesome meals: at 8am, 12:30pm, and 6pm. Twice a year, the setting sun will correlate with you tucking into your dinner. Were an independent observer to take those periods in isolation he might claim that the setting sun causes your hunger. Clearly, common sense tells us that this is not the case; here we have correlation but no causation.
To mix the two terms can be easy to do but can prove costly in the context of investing, based on statistical inference.
Many quantitative strategies are based on time series analysis; that is manipulating one or more sets of data, looking for patterns and making subsequent predictions. There are endless statistical techniques that are designed to filter out the ‘meaningless correlations’ but no system is perfect. Therefore all strategies are exposed in some way to this misinterpretation of correlation.
There are some systematic hedge funds out there with exemplary track records, Renaissance Technologies being the golden boy. However, there are an awful lot that have gone bust betting big on what was seen as a bullet-proof quantitative strategy which proved to be spurious correlation.
‘Sell in May and go away’ is a simple example of this in my mind; looking purely at the statistics, this looks to be a good bet but it only takes a year like this one to remind you of the dangers of the ‘well, it’s happened before’ attitude.
Undoubtedly there are some things that move in tandem 99.9% of the time, but even then always make use of the extra information you have gleaned since the last data point before betting it all on red.