Dynamic Covariance Forecasts for Risk Budgeting and Portfolio Optimization

Summary

This report compares methods for forecasting asset return covariance, a key input to minimum-variance and risk-budget portfolios. It evaluates historical and shrinkage covariance estimates, exponentially weighted moving averages, and multivariate GARCH approaches, including DCC-GARCH and GO-GARCH. The motivation is that asset volatility and correlations change over time, so a fixed historical estimate can be inaccurate.

Across Chinese and global asset and equity-sector portfolios, the report finds that EWMA and GARCH methods generally reduce covariance forecast errors and improve volatility tracking versus historical estimates. In a monthly rebalanced stock, bond, and gold risk-budget backtest, DCC-GARCH improves reported return, return-to-risk ratio, and drawdown relative to the historical covariance method. These results come from historical samples and depend on the tested assets, periods, and model setup; the report cautions that models may fail when market conditions shift. Better covariance estimates do not guarantee higher returns in every portfolio.

Key ideas

• Historical covariance estimates assume volatility and correlations remain stable, an assumption that may not hold.
• EWMA gives more weight to recent observations to reflect changing covariance conditions.
• DCC-GARCH and GO-GARCH generally reduce forecast errors in the tested asset groups.
• More accurate covariance forecasts can improve risk-budget weights and backtest outcomes, but do not ensure higher returns.
• The reported findings are historical and may not persist under changed market conditions.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.