Portfolio Optimization with Trend Following Strategies

Abstract

This thesis investigates how the mean-variance framework for portfolio optimization compares against that of risk-parity and the minimum conditional value-at-risk (CVaR) portfolio. Within the risk measure of portfolio variance, we find that the performance of the mean-variance portfolio is highly dependent on a well-conditioned sample covariance matrix while risk-parity appears to offer increased numerical stability. But with a regularized estimate, no method consistently outperforms the other. We suggest a minor extension to the risk-parity allocation objective with a resulting portfolio that exhibits superior properties in several central aspects. The minimum CVaR portfolio is built around the alternative risk measure conditional value-at-risk and we find that while the original problem formulation is prone to overfitting, a regularized version shows promising results worthy of further investigation.