Forecasting South Africa’s Consumer Price Index Using Seasonal ARIMA, LSTM, Support Vector Regression, and Gradient Boosting Models

Abstract

Accurate forecasting of the Consumer Price Index (CPI) is essential for effective monetary policy, inflation targeting, and macroeconomic planning. This study evaluates the forecasting performance of a traditional statistical model, Seasonal Autoregressive Integrated Moving Average with exogenous variables (SARIMAX), alongside three machine learning (ML) models: Long Short-Term Memory (LSTM), Support Vector Regression (SVR), and Gradient Boosting Trees (GBT) for forecasting South Africa’s CPI. Monthly data spanning February 2012 to January 2026 are used, incorporating CPI and key macroeconomic predictors, namely the Producer Price Index (PPI), Gross Loan Debt, Foreign Debt, Total Exports, Total Imports, and the Foreign Exchange Rate. Forecasting accuracy is assessed using MSE, RMSE, and MAPE. The empirical results show that the LSTM model achieves the lowest forecast errors, with an out-of-sample MAPE of 1.192% and an RMSE of 1.4177, followed by the GBT model with a MAPE of 1.320%. The SARIMAX and SVR models exhibit comparatively higher forecast errors. The Diebold–Mariano test indicates that the difference in predictive accuracy between the LSTM and GBT models is not statistically significant, suggesting statistically comparable forecasting performance. Overall, the results demonstrate the potential of ML-based approaches, particularly LSTM and GBT, for capturing the complex dynamics of South African CPI. The study contributes to the economic forecasting literature by providing empirical evidence on the application and comparative performance of statistical and ML models for CPI forecasting in an emerging economy. The findings provide useful insights for central banks, policymakers, economists, and financial institutions seeking improved CPI forecasts to support monetary policy formulation, economic planning, and financial decision-making.