Estimating the Correlation of Non-Contemporaneous Time-Series

Abstract

Daily financial time-series are often observed with different closing times: for example the FTSE stock index closes 11am NY time while the S&P500 stock index closes 4pm NY time. The non-contemporaneous observations mean the na¿ve correlation estimator is biased. This paper reviews the problem, discusses some simple estimators previously used in the literature, develops a maximum likelihood estimator, and compares estimators via simulations. We find two important results. First, a pseudo-ML estimator performs best, while simple method-of-moment estimators perform well in certain cases. Second, the standard error of the correlation estimator can be surprisingly large relative to the contemporaneous observations case.

Keywords: Covariance, Correlation, Asynchronous Trading, Non-synchronous Trading, Closing-Time Problem

JEL Classification: C13, G10