Forecasting Short-Term Oil Price with a Generalised Pattern Matching Model Based on Empirical Genetic Algorithm

Abstract

Price is an important guideline for measuring the changes in the oil market. Therefore, the forecasting of oil prices has become an important issue in oil market research. One of the problems, however, is that oil price is a non-linear or chaotic time-series, leading to difficulties in such research. In the forecasting methods commonly used, pattern matching model is a good method because of its simplicity, non-linearity, and accuracy, but when calculating its important input parameters, pattern matching model encounters certain problems in terms of accuracy and stability. In this case, the accuracy of the model prediction results will be affected. In this paper, the loss function is used to detect the source of the complexity of oil price forecast. On the basis of generalised pattern matching model based on genetic algorithm (GPGA), we introduce empirical distribution into genetic algorithm, which can dynamically compare the fitness among populations and tracks changes in individual evolutionary fitness to improve multiple modules. By using these information, directional evolution and full search elements are ensured. Finally, a generalised pattern matching model based on empirical genetic algorithm (GPEGA) is proposed. Empirical studies show that the accuracy and stability of GPEGA are 59.0% and 0.8% higher than that of GPGA. Moreover, the performance is 71.2% and 72.2% better than that of BPNN and ARIMA on mean square error. This study can help decision makers quickly and accurately grasp market information and provide support and reference for decision making on stabilizing economic markets and people’s lives.