Robust portfolio optimization with covariates

Abstract

In this project, we propose ARIMA regression as a methodology for the inclusion of covariate information into a robust CVaR minimization portfolio as a method to improve the performance of the portfolio optimization model. This methodology is compared with a robust CVaR minimization portfolio and an equal weights portfolio and is found to have poor performance in terms of Sharpe ratio and certainty-equivalent return but exhibits better performance when it comes to maximum drawdown. This suggests that while the methodology is flawed, it still holds promise in certain niche applications.