A1734
Title: Reassessing price discovery measures in high-frequency data: Evidence from the Japanese stock market
Authors: Makoto Takahashi - Hosei University (Japan) [presenting]
Masahiro Yamada - Tokyo University of Science (Japan)
Abstract: Price discovery across investor types is studied using high-frequency data from the Tokyo Stock Exchange, showing that two widely used measures, Weighted Price Contribution (WPC) and Hasbrouck's VAR model, suffer from Structural biases that can lead to misleading conclusions about which traders incorporate information into prices. Through Monte Carlo simulations with a calibrated data generating process, it is shown that (i) WPC overestimates the Contribution of noise traders by capturing transitory Price fluctuations and volume heterogeneity as Price discovery, and (ii) VAR-based measures produce different results depending on the Cholesky ordering of variables. A seasonality-adjusted long-horizon WPC (Adj LH-WPC) is proposed that simultaneously addresses three sources of bias: transitory noise via long-horizon Price changes, volume heterogeneity via regression coefficients, and intraday seasonality via Weighted least squares. The method recovers the true permanent Price impact shares in simulations and provides stable, economically interpretable estimates of each investor type's Contribution in empirical analysis. Building on these findings, a Structural VAR framework is explored that exploits the U-shaped intraday volatility pattern as an identification condition, offering a path toward resolving the simultaneous equation bias between order flow and Price changes that remains in reduced-form methods.
