Developing market sentiment indicators for commodity price forecasting using machine learning

• Main Author: Sohail, Tariq
• Other Authors: Boyd, Milton (Agribusiness and Agricultural Economics)
• Published: 2017
• Subjects: artificial intelligence; market sentiment indicator; machine learning; autoregressive model; crude oil; forecasting; commodity prices; Python; Amazon Mechanical Turks; Beautiful Soup; Natural Language Toolkit; Mechanize
• Online Access: http://hdl.handle.net/1993/32038

The objective of this study is to develop a market sentiment model for financial markets using machine learning, and to illustrate these methods using commodity price data. A market sentiment model may capture the fundamental and crowd psychology of the market, through a variable that uses positive and negative words and phrases. The commodity price used is the daily price of the spot crude oil exchange-traded fund (ETF), United States Oil Fund (USO). The forecasting power of the market sentiment model is compared with a traditional autoregressive model. The results showed that the autoregressive models did not have significant forecasting power for the oil data over the time period examined and the addition of the sentiment model did not improve the forecasting power. Machine learning is a relatively new forecasting method. Therefore, further research on this topic is needed before any firm conclusions can be drawn regarding the effectiveness of this approach. === February 2017