Contrary to conventional wisdom in nance, return prediction R2 and optimal portfolio Sharpe ratio generally increase with model parameterization, even when minimal regularization is used. We theoretically characterize the behavior of return prediction models in the high complexity regime, i.e....