Forecasting Beta Using Ultra High Frequency Data

Abstract

This paper examines if using ultra high frequency (UHF, e.g., tick-by-tick) data could improve the accuracy of beta forecasts compared with using only moderately high frequency (MHF, minute-level) data. We propose a novel two-step paired t-test for performance evaluation. Our test exploits the cross-sectional variations in the beta forecasts and avoids the issues associated with the traditional approach which requires choosing a proxy for the true beta. Our tests provide strong evidence that using UHF data generally yields more accurate beta forecasts than using MHF data. Furthermore, we show that the UHF estimator consistently belongs to the group of best risk-hedging performers for portfolios constructed based on both industrial classifications and size and book-to-market ratios. However, we also find that using UHF data of a coarser scale (e.g., 5 or 15 s) leads to reduced benefits compared with using tick-by-tick data. Our conclusions hold when different UHF estimators and sample periods are used.