A Comparative Analysis of Traditional SARIMA and Machine Learning Models for CPI Data Modelling in Pakistan

Abstract

Background. In economic theory, a steady consumer price index (CPI) and its associated low inflation rate (IR) are very much preferred to a volatile one. CPI is considered a major variable in measuring the IR of a country. These indices are those of price changes and have major significance in monetary policy decisions. In this study, different conventional and machine learning methodologies have been applied to model and forecast the CPI of Pakistan. Methods. Pakistan's yearly CPI data from 1960 to 2021 were modelled using seasonal autoregressive moving average (SARIMA), neural network autoregressive (NNAR), and multilayer perceptron (MLP) models. Several forms of the models were compared by employing the root mean square error (RMSE), mean square error (MSE), and mean absolute percentage error (MAPE) as the key performance indicators (KPIs). Results. The 20-hidden-layered MLP model appeared as the best-performing model for CPI forecasting based on the KPIs. Forecasted values of Pakistan's CPI from 2022 to 2031 showed an astronomical increase in value which is unpleasant to consumers and economic management. Conclusion. The increasing CPI trend observed if not addressed will trigger a rising purchasing power, thereby causing higher commodity prices. It is recommended that the government put vibrant policies in place to address this alarming situation.