Forecasting Price Movements in Betting Exchanges Using Cartesian Genetic Programming and ANN☆
Introduction
People have loved to bet on various events since the beginning of written history [1]. In particular, many people are gambling on sporting events. In past few years gambling has grown into multi billion industry. In the UK alone, remote (online) gambling sector generated £4.47bn (April 2015 – March 2016). Out of that, online betting and betting exchanges generated total of £1.72bn Gross Gambling Yield (GGY) [2]. This has led to increased interest in forecasting sporting event outcomes using mathematics and statistics for years.
With the rise of betting exchanges in 2000, punters can not only take bets, but also offer their own, in peer to peer fashion. Following the concept of more familiar stock exchange markets. This has attracted a new sort of customer, the full time, high-volume trader who buy and sell odds just like financial traders buy and sell stock or trade on foreign currency exchange. Although, the value of betting exchanges cannot be compared to financial markets, Betfair – the leading betting exchange – processes more than seven million transactions each day – more than all European stock exchanges combined [3].
This research investigates possible advantages of using machine learning with Feed forward multilayer perceptron (MLP) Artificial Neural Networks (ANN) and Cartesian Genetic Programming (CGP) to predict price movements on pre-race GB horse racing markets. The paper has been structured as follows: Section 2 explores current state of the art approaches, Section 3.1 outlines the machine learning and back testing platform and Section 3.2. describes the machine learning models explored. Furthermore, Section 4 reports on the results achieved.
Section snippets
Related work
There have been various attempts to predict the outcome of various sporting events using machine learning – dog racing (greyhound) [4] [5], tennis [6] [7], soccer [8] [9] [10], cricket [11] and horse racing [12] [13] [14].
However, little work has been done on forecasting the price movements inside the betting exchanges. Because betting exchange is very similar to traditional stock or currency exchange markets, there have been couple of attempts using stock market strategies and analysis on
Framework
Fig. 1 demonstrates the overall framework diagram. At first, technical market data, such as price and volume, is extracted from Betfair API. Data is then transferred to a local server on a weekly basis where it is formatted in open-source Protobuf-net for fast deserialization speeds and processing. Additionally, fundamental data such as horse form and running history is imported into the MySQL database. On the local server, each Strategy gathers the necessary data (if any) and develops a model
Results
This section describes results for multiple strategy models. All models were trained on training set for 24 hours using Intel Core i7 3930 K @ 4.2 GHz CPU (100% of CPU utilisation with parallel processing), then evaluated based on overall profit during training set (Table 4) and for unseen data – testing set (Table 5). Furthermore, all models used a limited liability staking plan of 100 units. Which means that with any single bet maximum potential loss is 100 units. Default Back/Lay bet stake
Conclusion
This research has addressed multiple challenges designing and implementing a back-testing and machine learning framework for betting exchange markets. Furthermore, developed entirely new trading strategies using machine learning algorithms.
The proposed framework is generic enough that it can be adopted to various financial markets, however, the only difference is that betting markets are time constrained, i.e., market is finalized as the winner is determined, while most other financial markets
References (30)
- et al.
Forecasting outcomes in tennis matches using within-match betting markets
Int. J. Forecast.
(2010) - et al.
Designing a neural network for forecasting financial and economic time series
Neurocomputing
(1996) - et al.
Forecasting the winner of a tennis match
Eur. J. Oper. Res.
(2003) An Economic and Social History of Gambling in Britain and the USA
(1996)A short history of the betting exchange industry
- et al.
Expert prediction, symbolic learning, and neural networks. An experiment on Greyhound racing
IEEE Expert
(1994) - et al.
Greyhound racing using support vector machines: a case study
Machine Learning for the Prediction of Professional Tennis Matches
(2015)- et al.
Predicting Soccer Match Results in the English Premier League
(2014)
Game ON! Predicting English Premier League Match Outcomes
Predicting Sports Events from Past Results. Towards Effective Betting on Football Matches
Using Machine Learning to Predict the Outcome of English County twenty over Cricket Matches
How to Win at the Track
Horse racing prediction at the Champ de Mars using a weighted probabilistic approach
Int. J. Comput. Appl.
Cited by (6)
Drivers analysis and empirical mode decomposition based forecasting of energy consumption structure
2020, Journal of Cleaner ProductionCitation Excerpt :In order to predict groundwater levels, Moghaddam et al. (2019) served evaporation, temperature, aquifer recharge and other factors as inputs of the conceptual model, the Bayesian network and ANN models. Dzalbs and Kalganova (2018) introduced ANN to predict the price movements. Amber et al. (2018) considered solar radiation, temperature, wind speed and others as driving factors of electricity consumption, which was predicted by multiple regression, genetic programming, ANN, deep neural network and SVM.
Implementing the BBE Agent-Based Model of a Sports-Betting Exchange
2021, European Modeling and Simulation Symposium, EMSSEffects of three drying methods on kinetics and energy consumption of carrot drying process and modeling with artificial neural networks
2021, Energy Sources, Part A: Recovery, Utilization and Environmental EffectsComparing State-of-the-Art Neural Network Ensemble Methods in Soccer Predictions
2020, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
- ☆
This article belongs to Special Issue: HEST4BDAA.