Forecasting the crude oil prices based on Econophysics and Bayesian approach

Abstract

In view of the importance and complexity of crude oil price, we investigate the dynamic forecasting of crude oil prices via Bayesian and Econophysics approaches. We propose information entropy to measure the predictability of crude oil prices and employ the rolling window approach to forecast the dynamic price of crude oil. In this article, Bayesian approach is applied to estimate the parameters, at the same time, the classic estimate approach is also adopted. Finally, we compare the forecasting results of the two methods. The results of this paper indicates that (1) both estimate approaches can effectively estimate the parameters of Heston model; (2) under a specific loss function, Bayesian approach has advantage over classical estimation approach when estimated sample size is small, while the estimate sample size increases, classical method slightly outperforms Bayesian method; (3) the predicted entropy of the estimated interval of 200 days under the Bayesian method are closest to the actual information entropy, while the estimated interval of 600 days under the maximum likelihood method is closest to the information entropy; (4) except for the estimation interval of 200 days, the predicted entropy under the maximum likelihood estimation is more representative of the predictability of the crude oil price. we also consider the more general case, we test this two methods of different steps ahead and compare the stability of the two methods.