Bootstrap refinements in tests of microstructure frictions

Bootstrapping is often used as a substitute for asymptotic distributions when the latter are not available. Recent developments in the theory of the bootstrap show that combining the bootstrap with a known asymptotic distribution yields inferences that improve on those drawn from asymptotic distribution theory or bootstrapping alone. We review the key to obtaining the improvement and compare asymptotic and bootstrap inferences of three variance ratio tests used in microstructure research. The more precise bootstrap inferences lead to conclusions that differ from those found in extant research on transitory volatility. Asymptotic tests are biased toward rejection, and bootstrap and asymptotic critical values are not generally close to each other. These findings suggest that the more precise bootstrap inferences should be used in future applications of these tests, as well as in various other empirical applications where intradaily or other high frequency data are modeled using vector autoregressions