Could be interesting to see performance distribution for random strategies on that stock universe as a comparison. The reverse could also be interesting: how do the models perform on data that is random?