Deep Limit Order Book Forecasting (2403.09267v4)

Abstract: We exploit cutting-edge deep learning methodologies to explore the predictability of high-frequency Limit Order Book mid-price changes for a heterogeneous set of stocks traded on the NASDAQ exchange. In so doing, we release `LOBFrame', an open-source code base to efficiently process large-scale Limit Order Book data and quantitatively assess state-of-the-art deep learning models' forecasting capabilities. Our results are twofold. We demonstrate that the stocks' microstructural characteristics influence the efficacy of deep learning methods and that their high forecasting power does not necessarily correspond to actionable trading signals. We argue that traditional machine learning metrics fail to adequately assess the quality of forecasts in the Limit Order Book context. As an alternative, we propose an innovative operational framework that evaluates predictions' practicality by focusing on the probability of accurately forecasting complete transactions. This work offers academics and practitioners an avenue to make informed and robust decisions on the application of deep learning techniques, their scope and limitations, effectively exploiting emergent statistical properties of the Limit Order Book.

• The paper introduces DeepLOB, a hybrid CNN-LSTM framework, to forecast high-frequency mid-price changes in limit order book data.
• It employs a detailed microstructural analysis by classifying stocks based on tick size, linking liquidity and spread patterns with prediction performance.
• The research advances an evaluation method beyond traditional metrics, correlating forecasting success with practical trading execution probabilities.

Deep Limit Order Book Forecasting: A Microstructural Approach

Introduction

This paper investigates the predictability of high-frequency mid-price changes in the Limit Order Book (LOB) using deep learning models. A comprehensive approach is adopted, encompassing the analysis of LOB's microstructural properties and the application of state-of-the-art deep learning techniques. The paper introduces LOBFrame, an open-source software designed to efficiently process and model large-scale LOB data, facilitating the quantitative assessment of forecasting capabilities.

Microstructural Analysis

The examination of LOB data is grounded in its microstructural characteristics, focusing on the dynamics at play within a diverse pool of stocks on the NASDAQ exchange. This includes a quantitative method to classify stocks based on their tick size, segregating them into small-tick, medium-tick, and large-tick categories. Such classification yields insights into the liquidity, spread patterns, and informational richness of the LOB, correlating these properties with the subsequent forecasting performance.

Forecasting Methodology

The forecasting framework employed, DeepLOB, integrates Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) models to capture both spatial and temporal dependencies within LOB data. A detailed process outlines the data preparation, labeling, and training phases, emphasizing the model's adaptability to different prediction horizons and confidence levels. The results highlight the model's differential performance across stock categories, underscoring the influence of microstructural properties on predictability rates.

Evaluating Forecast Practicability

Beyond traditional metrics like the Matthews Correlation Coefficient (MCC), the paper advances a novel, strategy-oriented approach to assess the practicability of forecasts. This method, immune to class imbalances and assumptions-free, calculates the probability of executing a correct transaction based on the model's predictions. This evaluation offers a more nuanced understanding of model utility, differentiating between academically acceptable and practically viable forecasts.

Findings and Implications

The analysis reveals a stark contrast in forecasting performance and practicability across stock categories, with large-tick stocks showing higher predictability and practicability rates. The findings also illuminate the critical role of microstructural properties and class distributions in shaping forecasting outcomes. Additionally, the paper calls attention to the limitations of conventional machine learning metrics in capturing the efficacy of forecasts for trading strategies.

Future Directions

The paper concludes with recommendations for further research, including cross-exchange validations and explorations of different deep learning architectures. The potential of transformer models, diffusion models, and graph-based models in LOB forecasting is particularly highlighted, suggesting avenues for leveraging architectural nuances to address challenges inherent in modeling LOB dynamics.

Summary

This research contributes significantly to the field of high-frequency trading and market microstructure, offering both theoretical insights and practical tools for LOB forecasting. By bridging microstructural analysis with advanced machine learning techniques, the paper provides a foundation for developing more effective and practicable forecasting models, facilitating informed decision-making in financial markets.