Using the theory that prices go up around pay-day more than other days on average due to capital flowing into the market on those days.
This backtest assumes some leverage (5x 100% of cash is exposed to the market on each trade). This could probably be done IRL with CFD's.
Algorithm goes long around 3 days before the end of the month (invests 5x portfolio into benchmark) and closes all portfolio positions around 8 days after the start of month. That's all it does. No TA, no fundamentals, no machine learning.