Deep Learning Models of High-Frequency Financial Data
We develop and evaluate deep learning models for predicting price movements in high-frequency data. Deep recurrent networks are trained on a large limit order book dataset from hundreds of stocks across multiple years. Several data augmentation methods to reduce overfitting are analyzed. We also develop and evaluate deep reinforcement learning models for optimal execution problems with limit order book data. "Optimal execution" is the problem of formulating, given an a priori determined order direction (buy or sell) and order size, the optimal adaptive submission strategy to complete the order at the best possible price(s).The performance of deep recurrent models is compared against other types of models trained with reinforcement learning, such as linear VAR models and feedforward neural networks.
Bio: Justin Sirignano is an Associate Professor at the Mathematical Institute at the University of Oxford, where he is a member of the Mathematical & Computational Finance and Data Science groups. He received his PhD from Stanford University and was a Chapman Fellow at the Department of Mathematics at Imperial College London. His research interests are in the areas of applied mathematics, machine learning, and computational methods.