Short-Term Inflation Forecasting in Pakistan Using ARIMA Models: A High-Frequency Analysis from Pre- to Post-COVID Era

Abstract

This article develops and estimates univariate ARIMA models to forecast Pakistan's monthly inflation using data from July 1993 to June 2024, covering both pre- and post-COVID economic periods. Unlike studies that rely on annual data or focus solely on pre-pandemic trends, we emphasize high frequency, short-term accuracy essential for real-time policymaking. While Seasonal ARIMA (SARIMA) and machine learning techniques (e.g., LSTM) are viable alternatives, ARIMA is chosen for its interpretability and established effectiveness in short-term macroeconomic forecasting (Romanian Journal of Economic Forecasting, 2021). Data from the World Bank, State Bank of Pakistan, and IMF were thoroughly preprocessed using stationarity tests (ADF) and residual diagnostics (Ljung-Box Q-test). The best-performing model, ARIMA (1,1,3) (AIC = 332.14), was evaluated using RMSE, MAE, and Theil’s U (0.4361). Projections to 2025 yielded a forecasted inflation rate of 1.04% compared to the actual rate of 3.50%, demonstrating the model’s value despite exogenous shocks (e.g., commodity price volatility). This empirically grounded approach equips policymakers with a transparent tool for addressing inflation amid the complexities of emerging economies. Future research may incorporate multivariate models to adjust for global confounding variables.