Machine Learning, Quantitative Portfolio Choice, and Mispricing

What happens to mispricing when quantitative learners---asset managers who use quantitative methods to make portfolio choice decisions---enter the market? Mispricing can actually increase when these learners enter and trade against historical mispricing because estimation error and model error limit their ability to properly adapt to changing prices caused by their own asset demand. This causes some asset prices to be corrected relatively little, while other assets that are initially underpriced (overpriced) become overpriced (underpriced). In a model with an estimated dividend process and a Koijen and Yogo (2019) style demand system, learner entrants who invest with some canonical quantitative methods---such as Brandtet al. (2009), Kozak et al. (2020), and DeMiguel et al. (2009) methods---tend to increase mispricing. When mispricing does not increase, a substantial amount of mispricing remains even when the learners have access to a long time series of data.