On the asymptotic structure of Brownian motions with a small lead-lag effect

TL;DR

This paper analyzes the asymptotic behavior of likelihood ratios for two correlated Brownian motions with a small delay, considering measurement errors, and explores the implications for estimating the lag parameter.

Contribution

It characterizes the asymptotic structure of the likelihood ratio process for Brownian motions with a tiny lag under measurement errors, revealing dependence on error levels.

Findings

The limit experiment's structure varies with measurement error dominance.

The model is asymptotically equivalent to discrete observations with endogenous errors.

Efficient estimation of the lag parameter is discussed.

Abstract

This paper considers two Brownian motions in a situation where one is correlated to the other with a slight delay. We study the problem of estimating the time lag parameter between these Brownian motions from their high-frequency observations, which are possibly subject to measurement errors. The measurement errors are assumed to be i.i.d., centered Gaussian and independent of the latent processes. We investigate the asymptotic structure of the likelihood ratio process for this model when the lag parameter is asymptotically infinitesimal. We show that the structure of the limit experiment depends on the level of the measurement errors: If the measurement errors locally dominate the latent Brownian motions, the model enjoys the LAN property. Otherwise, the limit experiment does not result in typical ones appearing in the literature. The proof is based on the fact that the model is asymptotically equivalent to discrete observations of two Brownian motions under endogenous measurement errors. We also discuss the efficient estimation of the lag parameter to highlight the statistical implications.