Extracting Knowledge from Technical Reports for the Valuation of West Texas Intermediate Crude Oil Futures

Abstract

We propose and demonstrate an approach for the often attempted problem of market prediction. We restrict our study to a widely purchased and well recognized commodity, crude oil, which experiences significant volatility. Robust debate exists over the applicability of the weak and semi-strong versions of the Efficient Market Hypothesis (EMH) to financial markets. In this paper we train nine learners using features extracted from monthly International Energy Agency (IEA) reports to predict undervalued, overvalued, and accurate valuation of the oil futures. Investor decisions are driven by both market and external forces such as geopolitical or natural events, which are discussed in IEA reports. Our results show, when considering F-measure and G-measure, the addition of text features significantly improves performance compared to only using price history from the oil futures data. Top learners using both feature spaces performed statistically better than random at the 95% confidence level, challenging the validity of the weak and semi-strong versions of the EMH in the crude oil market.