Walt Pohl (Norwegian School of Economics) 

Finance research has identified accounting data as an important source of explanatory variables in the cross-section of stock returns. In this paper, I investigate whether machine-learning algorithms can use the same information to generate excess returns. I find that they can, but that the algorithm may be picking up on systematic risk that is not well-explained by traditional risk factors. Despite that caveat, machine-learning algorithms can exploit the information in the cross section to produce Sharpe ratios above 1.