Hello Quantopians,

This post is related to my second task of investment strategy course, where I was asked to study the relationship between market sentiments and different kinds of returns. For this purpose, I was using Intraday, Overnight, First 30 mins and last 30 minutes of returns. For the purpose of capturing the notion of market sentiments, I am using VIX close price and to complete the picture, I took Market Beta as my second factor. It is pertinent to mention here that Market Beta was calculated for the period of 252 days (a year) and to eliminate the look ahead bias, both the factors were shifted. Another caveat, that I removed the data for those dates when earnings reports were announced for a particular security, in order to ensure robustness.

Eyeballing the heatmaps, I found some arbitrage opportunities in case of First 30 minutes and Last 30 minutes. They are convincing in the sense that the number of observations are high in corresponding intersection of deciles and standard deviation is low. Looking at the first 30 minutes in conjunction with the no of observations indicates that when the VIX implied volatility for S&P500 is high for the past business day, the first 30 mins returns are strongly negative regardless of the market beta of securities. This has a simple behavioral explanation i.e. when the implied volatility is higher, traders sell more actively than buying thereby increasing the supply and decreasing the returns.

Similarly, the mean returns for the last 30 mins are high and positive when the VIX close was high yesterday. This is simply because of the fact that due to flooding of securities in the market for the first 30 mins, the arbitrageurs find it lucrative to purchase the assets at the end of the day when the prices hit to its lowest, thereby increasing the demand and consequently the returns. Hence we observe positive returns for the highest VIX decile regardless of Market Beta.

It is important to point out, that the standard deviation of the corresponding decile intersections is relatively lower in case of last 30 minutes in comparison to first 30 minutes. This shows that the above described market behavior is more consistent in the second case than the first one. The robustness of these results were verified by plotting similar heatmaps for Geometric mean and the above results were consistent with them too.

In the end, the success rate was also calculated and plotted as heatmaps to reiterate the findings that were discussed above.