In this exercise, I was supposed to detect the dependencies between the Intraday returns and RSI of overnight/intraday/first 30m/last 30m returns. The goal was to find the best combinations of these factors such that the Intraday returns are maximized with minimal risks. To do that I compared average, geometric mean returns as well as the standard deviation of returns and percentage of cases when intraday returns were greater than zero across combinations of factors.

Time period investigated: 2006-2018
Investment universe: Q500US

Results:

Starting with the heatmaps of RSI factors with mean Intraday returns, I found strong arbitrage opportunities in every combination of factors, however, these results can be narrowed down using the corresponding heatmaps for number of observations in each deciles. In conjunction with the no. of observations, only RSI Last 30 mins and RSI Intraday exhibit potential arbitrage opportunities.

Starting off it is important to note that the entire diagonal going from left to right consists of large number of observations. The mean and geometric mean returns indicate opportunities for arbitrage i.e. 0.14% and 100.14%, respectively. However, the standard deviation is also high i.e. 0.026 for the relevant elements in heatmaps. For e.g. the geometric mean return for deciles 0.0/0.0 shows 0.14% returns, but the corresponding high standard deviation should not be neglected. Summing up the results based on the heatmaps of mean, geometric mean, standard deviation, no. of observations and success rates the combination of RSI Last 30 Minutes and RSI Intraday shows a visible pattern that could further be explored to come up with a proper strategy for statistical arbitrage. I also tried to plot the heatmaps for RSI with window length of 7, but since there were no significant differences in the results, I therefore made my analysis on the basis of conventional 14 days RSI.