Do Algorithmic Executions Leak Information?

George Sofianos and JuanJuan Xiang

Asset managers are concerned that the algorithms they use to execute their orders may leak information to predators. Predators are traders who use this information to trade in the same direction as the asset managers, increasing the asset managers’ trading costs. Asset managers are particularly concerned about leaking information to high-frequency trading (HFT) predators.

In this chapter, we develop and empirically test a framework for evaluating whether algorithmic (“algo”) executions leak information. Based on the Goldman Sachs Electronic Trading (GSET) algo executions we tested, our main finding is that these algo executions do not leak information that predators, including HFT predators, can profitably exploit. The algos we tested, by slicing up large high-Alpha orders into smaller orders and executing them over time, make it expensive for predators to identify and profitably trade along these orders.

In the next section, we define information leakage and distinguish between good and bad information leakage. We next show how bad information leakage increases execution shortfall and introduce the BadMax approach for testing whether algos leak information to predators. In