A Hybrid Swarm-Based System for Commodity Price Forecasting During the Covid-19 Pandemic

Abstract

Commodity price forecasting is an essential task for the definition of economic strategies to reduce the impact of inflation in several countries. During the covid-19 pandemic, commodity prices have increased due to external factors such as the influence of brokers, traders, policymakers, and investors. Accurate forecasts are important to increase the quality of the decision making processes of governments in order to alleviate negative economic impacts such as inflation. The forecasting of commodity prices is a challenging task due to the dynamic characteristics of commodity prices and external influences. In this work, we presented a hybrid swarm-based system combining the Autoregressive Integrated Moving Average with exogenous variables (ARIMAX) model and a Support Vector Regression (SVR) model optimized by the Particle Swarm Optimization (PSO). In the proposed method the PSO algorithm selects the most relevant variables for the ARIMAX model and optimizes SVR parameters. The experiments were conducted on various time series of commodity prices and exogenous variables, with single models and hybrid methods in the literature. The proposed method significantly reduced MSE and MAPE metrics compared to other approaches. The most selected features were the new cases, cumulative deaths, and the dollar exchange rate.

Keywords

• Time series forecasting
• hybrid system
• particle swarm optimization
• commodity price
• support vector regression