Robust Estimation of Realized Correlation: New Insights about Intraday Fluctuations in Market Betas
Time-varying volatility is an inherent feature of economic time series and complicates correlation measurement at high frequencies. While the quadrant correlation estimator is robust to volatility dynamics, it has low efficiency under Gaussian sampling. We introduce a subsampled quadrant correlation estimator that improves efficiency while retaining robustness and computational simplicity. The estimator is based solely on sign information and involves no smoothing parameter beyond the sampling horizon, which makes it attractive for high-frequency financial data and large cross sections. An empirical application to a broad panel of U.S. stocks reveals substantial intraday variation in market betas. Decomposing beta dynamics shows that relative volatility declines over the day by a similar amount for all stocks, so that the differences across stocks in how their betas move within the day are predominantly driven by differences in intraday correlation changes.