Inferring Price Discovery from Frictional Markets

Abstract

This paper identifies measurement error when using high-frequency transaction price data in the estimation of error correction models and cointegration analysis. Within the framework of price discovery analysis, we show that high-frequency transaction price variables are not relevant to the conceptual variables of interest, that discreteness and bid-ask bounce are the sources of the measurement error, and that this measurement error results in biased estimates. To deal with this bias, we endogenize a vector of bid-ask components, re-identify the error correction model, and propose a modified algorithm for the calculation of Johansen's (1988) full-information maximum likelihood estimator. Numerical examples show that our procedure essentially eliminates the bias and that the procedure is more robust to data error.

JEL Classification: G0, G1, G2