Regresar

Analysis of Chaos and Pbkp_redicting the Price of Crude Oil in Ecuador Using Deep Learning Models

Abstract:

This paper studied deterministic chaotic behaviour and pbkp_rediction of WTI crude oil daily price time series from 2015 to 2020 in Ecuador. To understand the price of crude oil, the dynamics and time delay of the system were reconstructed through the Average Mutual Information and False Nearest Neighbours methods, the chaotic characteristics was determined using the Lyapunov exponent, and finally a BDS test was applied to determine the nonlinearity of the series. Then, three neural networks are used to pbkp_redict oil prices, which were validated by estimating four goodness-of-fit measures. The results show that the neural network models produce a good pbkp_rediction rate, confirmed by a maximum error of 0.0058927 % from the Radial Basis Function, which indicates a significant similarity between the pbkp_rediction and the real data. Pbkp_redictions made with monthly values perform better than those made with daily values, possibly due to the level of noise in the daily time series. Among the three models, NARX performs best, with a percentage error of 2.6213 · 10- 13%.