Convolutional Attention in Betting Exchange Markets

TL;DR

This work tackles short-term price forecasting in betting exchange markets (Betfair) by leveraging market-depth data within an end-to-end automated trading framework. It introduces convolutional attention mechanisms and a novel roll padding strategy for multivariate time series, evaluating multiple architectures (CNN-LeNet, LSTM, ConvLSTM2D, WaveNet) in a production-like setting. The study shows that convolutional attention, particularly with multi-head Conv1D processing, improves forecasting and trading decisions in the UK To-Win Horse Racing pre-live window, achieving a best-case accuracy of $30.92\%$ and profitable simulated trading under liquidity constraints. The findings demonstrate the practical potential of advanced neural architectures in real-time exchange markets, while noting limitations from data size and market dynamics, and suggesting future work on stake optimization and model generalization for broader deployment.

Abstract

This study presents the implementation of a short-term forecasting system for price movements in exchange markets, using market depth data and a systematic procedure to enable a fully automated trading system. The case study focuses on the UK to Win Horse Racing market during the pre-live stage on the world's leading betting exchange, Betfair. Innovative convolutional attention mechanisms are introduced and applied to multiple recurrent neural networks and bi-dimensional convolutional recurrent neural network layers. Additionally, a novel padding method for convolutional layers is proposed, specifically designed for multivariate time series processing. These innovations are thoroughly detailed, along with their execution process. The proposed architectures follow a standard supervised learning approach, involving model training and subsequent testing on new data, which requires extensive pre-processing and data analysis. The study also presents a complete end-to-end framework for automated feature engineering and market interactions using the developed models in production. The key finding of this research is that all proposed innovations positively impact the performance metrics of the classification task under examination, thereby advancing the current state-of-the-art in convolutional attention mechanisms and padding methods applied to multivariate time series problems.

Figures (24)

• Figure 1: Average trading volume, average liquidity at the bid and ask price, and average number of ticks variation in absolute value per minute.
• Figure 2: High level architecture for automated betting exchange.
• Figure 3: State machine for a given order.
• Figure 4: Simplified graph scheme for a Back$\Rightarrow$Lay scalp implementation.
• Figure 5: Simplified graph scheme for a Back$\Rightarrow$Lay Trailing-Stop implementation.