Mid-Price Forecasting in Limit Order Books with Gaussian Process Models

Abstract

The limit order book (LOB) is a key component of modern financial markets, representing a dominant trading mechanism. Accurate and efficient price prediction for LOBs has long been a significant challenge due to their inherent high noise and stochasticity. In this study, we explore the application of Gaussian Processes (GPs), a Bayesian probabilistic model, to predict mid-price of LOBs. By leveraging parallel computing techniques, we address the computational challenges associated with GPs, significantly reducing runtime. Our experiments utilize a large-scale LOB dataset, revealing that the GP model outperforms most baseline models. While it does not quite match the performance of the state-of-the-art model in their accuracy evaluation metrics, the GP model surpasses in uncertainty estimation, providing valuable insights into prediction reliability. These findings highlight the potential of GPs in LOB price forecasting and provide insights for deeper exploration in this field.