A Benchmark for Limit Order Book (LOB) Forecasting Deep Learning Models
Published in Submitted to NeurIPS, 2024
We propose a comprehensive benchmark to evaluate the performance of deep learning models on LOB data. Our work makes three significant contributions: (i) we construct a futures LOB dataset and evaluate state-of-the-art LOB models on this dataset; (ii) we present the first benchmark study to evaluate prediction models on both primary LOB prediction tasks: mid-price trend prediction and mid-price return forecasting; and (iii) we assess the performance of general-purpose time series forecasting models on LOB data. Our empirical results highlight the value of our constructed futures LOB dataset, demonstrating a performance gap between the commonly used open-source stock LOB dataset and our futures dataset. Most importantly, the results unequivocally demonstrate that LOB-aware model design is essential for achieving optimal prediction performance on LOB datasets.
Recommended citation: Please wait for publication to cite!
Download Paper