Even as algo wheels are used by more traders, the challenges of accurately identifying the better performing algorithm remain. With limited data available to most traders, it is often impossible to differentiate between random noise and actual, better or worse performance. A series of simulated experiments conducted by Pragma’s research team show that even months of trading data are often insufficient for accurate transaction cost analysis and can lead to incorrect conclusions.
In this paper, we illustrate how easy it can be to get the wrong answer from an algo wheel experiment. To combat this challenge, we provide readers with a list of best practices, as well as a new execution benchmark we call Trajectory shortfall.