AI Techniques for Price Prediction in Commodity Markets
In this tutorial, we wish to cover the foundational, methodological, and system development aspects of Price Prediction of certain raw materials in spot markets (such as Ethylene, Hydrocorbons and Methyl Methacrylate) that are volatile as well as not traded through online Exchanges. Efficient price predictions of such raw materials are needed on daily basis and it is an important problem for industries as they spend several billion dollars in a year to procure such commodities for their business needs. There exists diverse and evolving information sources which can potentially influence the prices of raw materials in the markets. Multiple machine learning based methodologies – such as non-linear regression, random forest, and expert-based learning – are present in the literature to predict the price of raw materials. However, recent research showed that artificial prediction markets can aggregate such a diverse and evolving data more effectively than the standard machine learning models (Jahedpari et al. 2017). The reason behind the success of artificial prediction market based models is that the participating agents evolve through market intelligence with time in terms of their knowledge and this leads to better price predictions. Also, our recent experiments at IBM Research confirm the effectiveness of artificial prediction markets (Jahedpari et al. 2017) in order to derive efficient price predictions using actual (diverse and evolving) data sources. Motivated by this, this tutorial provides the conceptual underpinnings of the use of artificial prediction markets for predicting raw material’s price in spot markets. Broadly the contents of this tutorial belong to the topic of collaborative decision making in the area of multi-agent systems in Artificial Intelligence.