Performance Metrics for Algorithmic Traders

Abstract

Portfolio traders may split large orders into smaller orders scheduled over time to reduce price impact. Since handling many orders is cumbersome, these smaller orders are often traded in an automated ("algorithmic") manner. We propose metrics using these orders to help measure various trading-relatd skills with low noise. Managers may use these metrics to assess how separate parts of the trading process contribute execution, market timing, and order scheduling skills versus luck. These metrics could save 4 basis points in cost per trade yielding a 15% reduction in expenses and saving $7.3 billion annually for US-domiciled equity mutual funds alone. The metrics also allow recovery of parameters for a price impact model with lasting and ephemeral effects. Some metrics may help evaluate external intermediaries, test for possible front-running, and indicate sloppy or overly passive trading.

Keywords: trading performance, luck versus skill, order scheduling, price impact

JEL Classification: G12, G14, G23, G24