Modeling and Forecasting Nepal’s Nominal GDP: An Updated Box Jenkins ARIMA Analysis

Abstract

Background: Gross domestic product (GDP) is one of the principal indicators used to summarize the scale and evolution of economic activity and provides an important information base for macroeconomic planning, fiscal management, invest- ment decisions, and the allocation of public resources. Reliable GDP forecasts are particularly valuable for economies experiencing structural transformation and recurrent external shocks. This study focuses the ARIMA analysis of the annual nominal GDP from 1990/91 to 2024/25. The study aims to model and forecast Nepal’s annual nominal GDP using the univariate Box–Jenkins time-series methodology. Method: The study uses 35 annual observations of Nepal’s nominal GDP at current prices, expressed in million Nepalese rupees. The univariate Box Jenkins approach was applied in R to identify and estimate appropriate autoregressive integrated moving average (ARIMA) models. The GDP series was logarithmically transformed to reduce scale effects and stabilize variation. Stationarity was examined using graphical analysis, autocorrelation and partial autocorrelation correlograms, the Augmented Dickey–Fuller (ADF) test, and the Kwiatkowski–Phillips–Schmidt–Shin (KPSS) test. Six pre-specified ARIMA models were estimated using maximum likelihood and compared using Akaike information criterion (AIC), corrected Akaike information criterion (AICc), Bayesian information criterion (BIC), log-likelihood, root mean square error (RMSE), mean absolute deviation (MAD), and residual Ljung–Box diagnostics. An automated model- selection procedure was also used as a robustness check. Results: The GDP series exhibited a strong non-stationary trend in its original form. After logarithmic transformation and first differencing, the KPSS test supported stationarity, although the ADF test continued to provide mixed evidence regarding the presence of a unit root. Among the six pre specified candidate models, ARIMA(1, 1, 2) produced the lowest AIC (-98.7618) and AICc (-97.3825). The residual Ljung Box test was statistically insignificant at the 5% level, indicating no strong evidence of remaining autocorrelation in the residuals. However, the autoregressive coefficient was estimated at the boundary of the parameter space, resulting in an unavailable standard error and suggesting a potential near unit root estimation issue. The automated model selection procedure identified the more parsimonious ARIMA(0, 1, 0) model with drift and reported lower information criteria than the restricted candidate model. Forecasts based on ARIMA(1, 1, 2) indicate a continued increase in Nepal’s nominal GDP from NPR 6,353,083 million in 2025/26 to NPR 7,907,673 million in 2029/30, although the prediction intervals widen substantially as the forecast horizon increases. Conclusion: The findings demonstrate both the usefulness and limitations of univariate ARIMA modeling for forecasting Nepal’s aggregate GDP. Within the restricted set of pre-specified models, ARIMA(1, 1, 2) was identified as the preferred model according to AIC and AICc, while the automated analysis suggested that the simpler ARIMA(0, 1, 0) model with drift provides an important parsimonious benchmark. The results indicate that Nepal’s nominal GDP is expected to continue increasing over the forecast period, although forecast uncertainty increases over time. The study highlights the importance of comparing alternative specifications and suggests that future research should evaluate multivariate, hybrid, and other forecasting approaches alongside univariate ARIMA models to improve the robustness and accuracy of GDP forecasts for Nepal.