Enhancing Malaysia’s gross domestic product estimation: advanced Gaussian processes in investigative and comparative analyses

Abstract

Understanding gross domestic product (GDP) is crucial for effective planning and anticipating economic developments, including in Malaysia. This study aimed to enhance Malaysia’s GDP estimation using advanced Gaussian processes (GPs). Three single kernel GPs models and three advanced GPs models with new kernel compositions were developed for GDP modeling, and four traditional regression models were employed for comparative analysis. The top-performing GPs and regression models were subsequently applied for GDP forecasting in three different cases. This study evaluated several metrics to measure the predictive accuracy in GDP modeling and forecasting. The metrics revealed that squared exponential-Matérn 3/2 (SE-Matérn) GPs model produced the smallest deviations with more stable error trend among the GPs models and better confidence interval coverage, while quadratic model obtained the highest coefficient of determination. For GDP forecasting, the analysis indicated that the SE-Matérn GPs model performed strongly in Case 1. However, in Cases 2 and 3, it showed larger deviations compared to quadratic model, though the absolute percentage error difference remained within 1.5%, showcasing the model’s dependable forecasting ability. These findings indicated that the advanced GPs model can effectively estimate Malaysia’s GDP during period of decline due to its flexibility and non-parametric nature.