A fantastic new paper that investigates the efficacy of machine learning in stock selection. Cliff Notes as follows – look at the machine learning and artificial intelligence models that you would find in a graduate level quantitative finance textbook. Run them on tons of data and see if they are helpful in stock selection. They conclude that the models are marginally beneficial (not a surprise since some of the authors do quant finance for a living). Really astonishing were the variable results. They threw a ton of stock information into the hopper, fundamental, price, industry, you name it. One variable dominated all the others in every model – Price Momentum. If a stock had gone up recently, the models wanted to bet that it would continue.
Momentum is the least thoughtful, least rational of indicators. It looks at the recent past and simply extrapolates. There is no notion of value, no grounding in fundamentals – just the recent past. What these models are picking up is the dominance of short term thinking on our lives and discourse generally, not just in finance.