Machine learning and trade direction classification: insights from the corporate bond market

Abstract

Leveraging the availability of a large panel of signed trade data in the corporate bond market, we explore how machine learning methods can be used to improve upon standard trade direction classification methods in markets in general, both with and without pre-trade transparency. Using the signed data set allows us to show how both the trading and information environment at the time of the trade critically affect the accuracy of existing trade classification rules in general and also illustrate the importance of optimizing the feature set in correctly classifying trade direction. These insights extend to the equity market.