On inferring the direction of option trades

To sign option trades as buys and sells, researchers often employ stock trade classification rules including the quote, the Lee and Ready (1991), the Ellis, Michaely, and O'Hara (2000), and the tick methods. Using a proprietary CBOE dataset that reports trade direction, we find that these four rules sign correctly 83%, 80%, 77%, and 59% of all classifiable trades, respectively. These rates are based on separate classifiable samples because each of the four rules fails to classify some trades (e.g., the quote rule cannot classify midspread trades). Outside-quote and reversed-quote trades are highly misclassified by all four rules. The probability of such trades is related to trading frequency, trade size, moneyness, and maturity. Underlying asset price changes around the time of the trade improve classification precision. We find that the components of index option complex trades not executed on the Retail Automated Execution System are misclassified almost 50% of the time by any method. The elimination of these trades (15% of the sample) results in a success rate of over 87% for the quote rule.