Machine learning and speed in high-frequency trading

Abstract

The creative destruction wrought by high-frequency algorithmic trading has raised increasing concerns about the effect of machine learning behaviors and ultra high-frequency trading on financial markets. By employing a genetic algorithm with a classifier system as an adaptive learning tool, we address some of these concerns by studying a dynamic limit order market model with asymmetric information and varying speeds of high-frequency trading (HFT). We show that HFT benefits uninformed traders, improves information efficiency but reduces market liquidity. We find that there is a trade-off where a competition effect erodes the information and speed advantages of high-frequency traders, increasing trading speeds of HF traders reduces market liquidity but generates a hump-shaped relationship to the profitability of high-frequency traders and information efficiency. This research finds there may be potential benefits to throttling the trading speed arms race to improve market efficiency. We also find that strategic algorithmic trading compensates for diminishments in speed advantages, providing an insight on machine behavior in the FinTech age.