Large-scale portfolios using realized covariance matrix: evidence from the Japanese stock market

This paper examines effects of realized covariance matrix estimators based on high-frequency data on large-scale minimum-variance equity portfolio optimization. The main results are: (i) the realized covariance matrix estimators yield a lower standard deviation of large-scale portfolio returns than Bayesian shrinkage estimators based on monthly and daily historical returns; (ii) gains to switching to strategies using the realized covariance matrix estimators are higher for an investor with higher relative risk aversion; and (iii) the better portfolio performance of the realized covariance approach implied by ex-post return per unit of risk and switching fees seems to be robust to the level of transaction costs.