Nice article! Just one remark: i think that there still might be predictability in the entire joint dataset by taking into account the correlation between the individual stocks. For instance, if some stocks are highly correlated, then its likely that their stock prices result from some underlying hidden process, e.g. “the market" process. Even if this market process is a random walk, temporary deviations of individual stocks from the market will tell us whether a stock is currently mispriced and we can buy and sell accordingly.