Does Conditional Covariance Outperform Empirical Covariance? A Study of Portfolio Optimization Techniques

Akshay Damani, Anish Nair, Urmi Mangal, Radhika Bhutra

Abstract

This study analyses two academically utilised approaches of dynamic conditional correlation GARCH (DCC-GARCH) and historical/empirical covariance for covariance matrix estimation and their impact on portfolio optimization is studied during crisis periods. The researchers have collected data for a period of 9 months (i.e. the crisis period) of a multi-asset portfolio that is diversified across countries. This paper also compares different optimisation techniques to not only have a more robust comparison between results from the estimation of covariance matrix, but also to study which optimisation technique benefits the most from the change in covariance matrix estimation method and performs the best during the studied crisis period. It evaluates four strategies: mean-variance optimization (MVO), cluster risk parity (CRP), naïve risk parity (NRP), and equal weighting. The results indicate that DCC-GARCH based covariance matrix estimation provides a much better result for portfolio optimisation during the crisis period studied, and that max Sharpe portfolio optimisation technique benefits the most from said estimation method while CRP and NRP witness no benefit from the same. These results add to the existing literature showing the outperformance of DCC-GARCH in covariance estimation and highlights an additional nuance of variance in results based on which optimisation technique is used.

Source: semanticscholar · PDF

Read the AI summary and key takeaways for traders on WOBR Quant Research.


Open the interactive page on WOBR AI → · WOBR.AI home