Shilong Yang, JP Morgan Chase & Co
Deep Basket - An Application of Deep Learning in Portfolio Construction
Traditional portfolio construction is based on convex optimizations in Markowitz’s portfolio theory assuming underlying asset return’s mean and covariance are known. We use data driven approach, identify hidden deep factors (normally nonlinear) through neural networks (MLP, RNN, LSTM, etc), and further construct the tracking portfolios through hierarchical models. In particular, deep learning is demonstrated to detect and exploit interactions in the data that are invisible to any existing financial economic theory. The flexibility in neural networks is key to understand multi-horizon features hidden in the financial time series. The tracking baskets constructed using this novel technique are shown to outperform traditional approaches based on covariance structure in out-of-sample tests. Additional features or signals from different frequencies can be incorporated easily in the same deep neural networks, providing a consistent solution for portfolio construction problems.
Related Topics
Explore More Events
-
Conferences
AI in Fintech Forum: 2024
326 Galvez Street
Frances C. Arrillaga Alumni Center
Stanford, CA 94305
United States