Recently I developed a peculiar interest in randomness and have been tweaking trading algos related to random numbers, inspired by the ideas presented in Getting Sid Programmatically by Ha C. I tried to pick random stocks and trade them - which I and apparently Ha and Fawce think are too risky. The I tried to pick multiple random stocks and trade the interesting stocks among them. The rationale is, assuming that each stock in the universe can be uniformly chosen, the expected return for the strategy that we buy the lowest price stock and sell the highest price stock would be positive - meaning that we can eventually make money. This is a right assumption but unfortunately it didn't work well since the volatility is too high. I realized that random numbers are innately unreliable, so I instead spread the risks into multiple drills (in this algo, 100 times per period) over multiple stocks (choosing the highest/lowest value stocks over 10 random stocks). It turns out that both strategies greatly reduced the volatility while most of the time yield positive gains.
From this simple experiment I believe that to some extent randomness can be of great help without making us suffer from potential huge losses.