PRAM: A Novel Approach for Predicting Riskless State of Commodity Future Arbitrages With Machine Learning Techniques

Abstract

Arbitrage risk management is a very hot and challengeable topic in the commodity future market. To resist the possible risk of an arbitrage, exchanges have to withdraw margin from clients referring to the case of maximum risk. However, if this arbitrage is in the riskless state actually, the capital of clients will be inefficient. Therefore, by investigating the applications of machine learning techniques, we here propose a novel algorithm named PRAM to predict the riskless state of arbitrage, by integrating multi-scale data ranging from contract quotation to contract parameters. Unlike the traditional models, PRAM explores the arbitrage risk management from the view of minimum risk, which can form a powerful supplement with the available risk management systems. Benchmark results based on DCE database implicate that PRAM outperforms existing methods. Then, we discover that features of different arbitrage types depended by PRAM are odds with being identical. In addition, we identify some trade situations, such as delivery and near-delivery months, which seriously impact the effectiveness of PRAM. Furthermore, considering different varieties involved in intra-commodity arbitrages, we create personalized PRAMs, which can deeply improve the accuracy of prediction.

Keywords

• Arbitrage risk management
• riskless state
• machine learning
• trade situations
• personalized models