Machine Learning in a Dynamic Limit Order Market

Abstract

We use a novel machine learning approach to tackle the problem of limit order management. Applying our framework to data, we show that the most important variable for a trader to consider is the price level of their order, followed by the queue sizes of the order book, volatility and finally queue position. Further, we show the option to cancel a limit order is valuable and contributes approximately 15% of a limit order's total expected value. This paper takes an important step towards describing pervasive features and dynamics that exist in financial markets.