Distributionally Robust Portfolio Optimization

This lecture covers the concept of distributionally robust portfolio optimization, where an investor aims to maximize the expected value of a concave piecewise linear utility function under uncertain rates of return. Different approaches using training samples to estimate the expected utility are compared, including sample average approximation and partial information assumption. The lecture also delves into the evaluation and comparison of different methods, such as SDP relaxation, SOCP relaxation, and dualization, in the context of portfolio optimization. The instructor provides hints on implementing these methods and emphasizes the importance of finding optimal objective values and runtimes for different systems.