Liquidity Biases in TRACE

Abstract

The transactions database TRACE is rapidly becoming the standard data source for empirical research on US corporate bonds. This paper is the first to thoroughly discuss the assumptions needed to clean the disseminated TRACE data and to suggest that different filters should be used depending upon the application. 7.7% of all reports in TRACE are errors and in some cases up to 18% of the reports should be deleted. Failing to correct for these errors will bias popular liquidity measures towards a more liquid market. The median bias for the daily turnover will be 7.4% and for a quarter of the bonds the Amihud price impact measure will be underestimated by at least 14.6%. Further, calculating these two measures on the same data sample would potentially bias one of them.