Risk Parity Stock Optimization Using Principal Component Quantile Simulation

Abstract

Today's portfolio optimization models are often too sensitive to stochastic input parameters and the use of outdated risk measures, resulting in poor risk adjusted return. This thesis presents a solution to avoid these diculties by introducing a new simulation framework to obtain multivariate return distributions for correlated assets. Next a model to obtain a risk parity portfolio using Conditional Value at Risk (CVaR) is oered making the model able to capture asset specic risk characteristics present in the tails of the marginal distributions. Quantile regression and principal component analysis (PCA) are combined to form a factor model able to capture the entire return distribution and maintain dependencies between correlated assets. A new method to simulate future principal components is presented making the simulation algorithm quick and eective. The resulting marginal distributions show asset spesic risk characteristics and tail behaviour. This is in turn re ected in the risk parity portfolio weights, conrming CVaR as a risk measure oering better intelligence to investors.