Job market paper
Abstract We introduce a new correlation estimator that combines the Quadrant correlation estimator with subsampling. Unlike the well-known Pearson and Kendall correlation estimators, the subsampled Quadrant estimator is consistent under time-varying volatilities and more precise than the Quadrant without subsampling. When we estimate correlations of high-frequency financial data, the subsampled Quadrant compares favorably to alternative estimators. The subsampled Quadrant estimator is robust to the microstructure noise, rounding error, asynchronous trading, and jumps found in financial data. On implementing the new estimator with actual high-frequency stock transaction data, we illustrate distinct patterns of correlation estimates along different sampling frequencies. Also, we propose a new approach that embraces the subsampled Quadrant to estimate intra-day market betas and find new insights about the intra-day variance in market betas for 22 stocks.