Statistical analysis of correlated fossil fuel securities

Abstract

Forecasting the future prices or returns of a security is extraordinarily difficult if not impossible. However, statistical analysis of a basket of highly correlated securities offering a cross-sectional representation of a particular sector can yield information that is potentially tradable. Securities related to the fossil fuels industry are used as the basis of a practical application to two distinct forecasting techniques. The first method, forecasting using conditional multivariate Gaussian statistics, was shown to yield, in a relative sense, the best results for those securities which exhibited a high correlation with the rest of the basket. For the second method, principal component analysis was done on a basket of commodity futures to reveal a small number of dominant factors governing the movements of the portfolio. Autoregressive models were then applied to both the factors and futures, but results showed both to be essentially Markov processes.