Market Microstructure Noise, Sparsely Sampled Realized Variance, and Optimal Futures Hedging

[[abstract]]It is documented that realized variance (RV) sampled at ultra-high frequency is unreliable when observed prices are contaminated by market microstructure noise. Accordingly, in practice, it is common to choose a moderate frequency ranges from 5 to 30 minutes instead of at every tick for balancing a bias/variance trade-off. This study aims to investigate the efficient frequency for daily RV of spot and futures assets, and compare the performance of forecasting spot-futures distribution using the sparsely sampled RV estimates. Obtaining tick-by-tick records from NASDAQ 100 markets, both unconditional and the conditional sampling rules are investigated. It is found that the optimal frequency for sampling the assets is more likely to be in the neighborhood of fifteen minutes, and RV estimates reveal substantial downward-(upward-) bias for the spot (futures) at ultra-high frequency. Out-of-sample forecasting results show that, using a fixed 10-minute frequency on hedging practice seems to work well, but performs poorly as tick-by-tick frequency is employed. The hedging performance is judged by both criteria of variance reduction size and expected utility growth within an augmented GARCH framework