Trade Classification Algorithms : A Horse Race between the Bulk-based and the Tick-based Rules *

We compare bulk-volume classification (BVC) proposed by Easley, Lopez de Prado, and O’Hara (2012b) to the traditional tick rule (TR) for a sample of equity trades executed on NASDAQ’s INET platform. Applying BVC leads to substantial time savings when a researcher uses pre-compressed data like Bloomberg and to smaller time savings when a researcher uses TAQ. Notably, this efficiency comes at a significant loss of accuracy. Specifically, misclassification increases by 7.4 to 16.3 percentage points (or 46% to 291%) when switching from TR to BVC. Additionally, TR produces more accurate estimates of order imbalances and of order flow toxicity (VPIN).