Assessing accuracy of trade side classification rules. Methods, data, and problems

Trade side classification algorithms enable us to assign the side that initiates a transaction and distinguish between the so-called buyer- and seller-initiated trades. According to the literature, such classification is essential to assess both market liquidity and dimensions of market liquidity based on high frequency intraday data. The main problem is that trade and quote data is not publicly available for many stock markets and researchers have to utilize indirect methods to infer trade side. The aim of this paper is to investigate major problems in assessing accuracy of trade side classification algorithms. We evaluate and compare four most frequently utilized procedures using intraday data for 105 companies from the Warsaw Stock Exchange (WSE). Moreover, an analysis of the robustness of the results is provided over the whole sample period from January 2, 2005 to December 30, 2016, and three consecutive sub-periods of equal size, covering the pre-crisis, crisis, and post-crisis periods. The empirical experiment shows that the Lee-Ready (1991) algorithm and tick rule perform better than other methods on the WSE, regardless of the choice of the sample.