Lasse Heje Petersen:
Machine Learning about Implementable Portfolios
Time and place
Friday, 25.11.22, 12:00-13:00, online: Zoom
Abstract
We develop a framework that integrates trading-cost-aware portfolio optimization with machine learning (ML). While numerous studies use ML return forecasts to generate portfolios, their agnosticism toward trading costs leads to excessive reliance on eeting small-scale characteristics, resulting in poor net returns. We propose that investment strategies should be evaluated based on their implementable ecient frontier, and show that our method produces a superior frontier. The superior net-of-cost performance is achieved by integrating ML into the portfolio problem, learning directly about portfolio weights (rather than returns). Lastly, our model gives rise to a new measure of "economic feature importance".