Option Trade Classification

Abstract

We evaluate the performance of common stock trade classification algorithms to infer the trade direction of option trades. Using a large sample of matched intraday transactions, we show that the algorithms' success is considerably lower than for stocks. The reason are sophisticated customers who often use limit orders to implement their trading strategies. These traders' activities vary over time and across exchanges with different pricing models. We design new rules that improve classification accuracy by 6% to 47%. In a long-short trading strategy for stocks based on option order imbalance, the new rules increase Sharpe ratios from 2.22 to 4.25.

Keywords: buyer/seller initiated trades, trade direction, limit order, Lee and Ready algorithm, quote rule

JEL Classification: C10, G12, G13, G14