Option Trade Classification

We evaluate the performance of common stock trade classification algorithms including the quote, tick, Lee and Ready (1991), and Ellis, Michaely, and O'Hara (2000) rule to infer the trade direction of option trades. Using a large sample of matched intraday transactions and Open/Close data, we show that the algorithms' success rate to correctly classify option trades is considerably lower than for stocks. In particular, the prevailing Lee and Ready algorithm is only able to correctly sign between 60% to 64% of option trades, which is a similar magnitude as using the quote rule alone. We find that the overall weak performance is due to sophisticated customers who often use limit orders instead of market orders to implement their trading strategies. We develop additional rules that can be used together with existing classification algorithms, improving correct classification by more than 10%.