Correlation Shrinkage: Implications For Risk Forecasting

Vol. 18, No. 3, 2020

Jose Menchero and Peng Li

In this article, we study the impact of shrinking sample correlations toward zero. We find that while such shrinkage may be beneficial from a portfolio-construction perspective, there is virtually no benefit in terms of the accuracy of risk forecasts. In fact, we show that correlation shrinkage typically increases the errors in risk forecasts, sometimes by a large margin. Hence, we conclude that for purposes of estimating portfolio risk, the estimated correlations should not deviate significantly from the sample correlation.