The Research

→ To what extent do machine-learning–based models or algorithms influence the trading decisions and average execution quality of institutional firms?

Although, through the form of predictive algorithms, machine learning is widely used to generate predictive signals and inform decisions while trading, the execution quality of any given order relies on how it affects and is affected by the liquidity (the availability of buyers and sellers at different prices), volatility (rapid fluctuations in price), and timing when placing the order, causing human judgement to have a larger impact on the quality rather than the algorithm.

• 01While ML models are widespread in use to generate predictive signals, they do not directly influence execution quality.
• 02Execution quality is bound by constraints such as liquidity, volatility (rapid changes in price), and market structure at the moment any given trade is executed.
• 03If the order size is greater than the depth available at the best bid or ask price, the trade must “walk the book,” which means that it fills progressively at worse price levels.
• 04Execution quality not only depends on whether the direction of the trade is correct, but also on how the trade changes the behavior of the market and reshapes liquidity. This means that even a correct prediction generated by a model can result in poor execution.
• 05While models can easily process data and identify patterns, they actually lack the understanding required to adapt to rapidly changing market conditions. As a result, traders on their own must interpret model outputs and make adjustments to their decisions.