Evolution of limit order book dynamics: One machine learning high frequency trading model
Stevens Institute of Technology
Over the last few years, information technology, especially in the electronic trading system area, developed dramatically. Hence more and more financial products have traded electronically. This trend also provides our financial engineer researchers the possibility to obtain high frequency trading data. Based on the past work of our peers, we propose a machine learning model, trees method oriented, to predict the mid-price movement and cross over opportunities for limit order book data. L1 penalty Least Absolute Shrinkage and Selection Operator(Lasso) regression was firstly used to conduct features selection. Comparison among the empirical results of multi-class support vector machines, random forest and decision trees model was presented after. Additionally, some statistical properties of limit order book in US market, such as the power law of the arrival rate and self-excited process, were verified and used to calibrate the parameters. The accuracy rate of our models demonstrates the feasibility to capture the short dynamics of the limit order book. Our limit order book data were obtained from NASDAQs. Python and Matlab were used to realize our models.
Keywords: electronic trading, limit order book, Lasso, support vector machines, decision tree, random forest