Title: A noise-robust trade classification algorithm
Authors: Simon Jurkatis - Freie Universitaet Berlin (Germany) [presenting]
Abstract: A new trade classification algorithm is proposed that provides accurate estimation of the trade initiator in data environments that have challenged the literature on trade classification. These environments are imprecise timestamps relative to the frequency of quote changes and misalignments of trade times and their corresponding quote changes. Using data from Nasdaq's limit order book, which provides information of the trade initiator, the method is compared against the common alternatives under various data environments by artificially decreasing timestamp precision and adding noise to recorded transaction times for more than 130m trades. The results show that the new algorithm is the dominant choice with improvements of up to reducing misclassification by half. The empirical relevance of these differences is demonstrated by estimating various measures of liquidity.