Trade classification accuracy for the BIST

• Much higher than for other markets, the trade initiator is identified for 94.2% of all trades on this developing market.
• Highest accuracy rate (above 95%) in the literature for the LR algorithm are obtained for period including Lehman collapse.
• LR classifier accuracy is highest (lowest) for trades for mixed agency/principal (pure principal) client/broker trades.
• Accuracies of trade classifiers are generally higher for long versus (mostly buyer-initiated) short trades.
• Trade-initiator misclassifications tend to be higher in the first versus last 30 min of both daily trading sessions.

The accuracy of five algorithms for classifying trades as buyer- or seller-initiated is assessed for BIST-30 index constituents over a period including the Lehman collapse. The highest classification accuracy rate (over 95%) is for the one-second lagged Lee & Ready (LR) algorithm. The LR's classification accuracy is highest (lowest) for trades representing mixed agency and principal (pure principal) relations between clients and executing brokers. Unlike for U.S. markets, almost all trades are classifiable with accuracy rates of 90-plus percent for both long and short trades. As for U.S. markets, higher misclassification rates occur for trades in the first versus last 30 min of the trading day, as the time between consecutive trades decreases, and for decreasing trade sizes.