The performance of PSO-SVM in inflation forecasting

Abstract

Analyzing inflation forecast problem, this paper proposes a SVM-based approach. Firstly, the paper reviews some former studies about inflation forecasting and predicting methodology, finding that SVM is a nonlinear adaptive data-driven model with strong approximation and generalization ability, which can be applied to complex forecasting tasks. Secondly, the paper establishes a SVM model and discusses the selection of kernel functions. Thirdly, the Particle Swarm Optimization (PSO) and the Genetic Algorithm (GA) are introduced to optimize the models. Then the SVM-based models (Fixed-SVM, PSO-SVM, GA-SVM) together with a BP neural network are employed to forecast Chinese inflation rate. The results show that the PSO-SVM performs better than BP and any other SVM-based model since its MSE of testing group is 0.006 and its absolute errors between predictions and real values are all below 0.02. It reveals that the final PSO-SVM model is promising in short-term inflation forecast.