Latency and the Look-Ahead Bias in Trade and Quote Data

Abstract

The NYSE Trade and Quote (TAQ) dataset reports events sequentially out of order. Trades and quote changes are clustered but recorded with differential latency, partly due to geography, effectively mixing up the order of events. Consequently, quote changes after, and in response to, a trade are regularly reported before it, creating a look-ahead bias. This bias creates errors in signing and spread measurement when using the TAQ NBBO. Errors intensify with latency: around 20% of high-latency trades are signed incorrectly and effective spreads are understated by over 40%. We propose a more accurate latency-free signing methodology, grounded by exchange rules.

Keywords: TAQ, SIP, Latency, Trade Signing, Effective Spread, Price Impact JEL Classification: C81, G12, G14, G20

JEL Classification: C81, G12, G14, G20