TCN-LOB: Temporal Convolutional Network for Limit Order Books

Abstract

Deep convolutional neural networks (CNNs) are commonly used for predicting limit order book (LOB) data. However, using only convolutional neural networks may lead to neglecting long-distance dependencies. Recent research has addressed this issue by incorporating attention mechanisms and adding Long Short-Term Memory Network(LSTM) layers. In contrast to previous methods, we combine Temporal Convolutional Network(TCN) and Squeeze-and-Excitation(SE) to enhance the network's ability to capture long-distance dependencies and relationships between feature channels. Based on the layers mentioned above, our proposed multi-horizon forecasting model has been validated on a publicly available benchmark dataset containing millions of high-frequency trading events. The results demonstrate the effectiveness of our proposed model, with comparable performance to state-of-the-art algorithms in short-term prediction horizons and outperforming other methods in long-term prediction horizons.