AlphaPortfolio for Investment and Economically Interpretable AI
68 Pages Posted: 20 Apr 2020 Last revised: 15 Jul 2020
Date Written: May 21, 2020
We propose reinforcement-learning-based portfolio management, an alternative to the traditional two-step portfolio-construction paradigm (e.g., Markowitz, 1952), to directly optimize investors' objectives without relying on estimates of distributions of asset returns. Specifically, we extend cutting-edge AI tools such as Transformer to allow multi-asset sequence modeling, so as to effectively capture the high-dimensional, non-linear, noisy, interacting, and dynamic nature of economic data and market environments. The resulting AlphaPortfolio yields stellar out-of-sample performances even after imposing various economic and trading restrictions. Importantly, we use polynomial-feature-sensitivity and textual-factor analyses to project the model onto linear regression and natural language spaces for greater transparency and interpretation. Such ``economic distillations'' reveal key market signals, firms' financials, and disclosure topics, including their rotation and non-linearity, that drive investment performance. Overall, we highlight the utility of reinforcement deep learning and provide a general procedure for interpreting AI and big data analytics in finance and beyond.
Keywords: Artificial Intelligence, Distillation, LSTM, Machine Learning, Portfolio Theory, Reinforcement Learning, Textual Analysis.
Suggested Citation: Suggested Citation