Forecasting Annual Electricity Consumption in Myanmar: A Comparative Time Series Analysis for Strategic Energy Planning
Ni Ni Win Naing *
International Leadership University, Meikhtila, Yan Myo Aung Quarter, Mandalay, Myanmar.
Richard Dare
International Leadership University, Zabuthiri Township, Nay Pyi Taw, Myammar.
Naing Naing Htun
International Leadership College, Zabuthiri Township, Nay Pyi Taw, Myammar.
May Phu Pwint Soe
International Leadership University, Zabuthiri Township, Nay Pyi Taw, Myammar.
*Author to whom correspondence should be addressed.
Abstract
Electricity demand forecasting is essential for effective energy planning resource allocation, and policy formulation. This study analyzed comprehensive historical time series data spanning 63 years, from the 1961–1962 to the 2023–2024 fiscal years. The research extensively utilizes secondary time series data on annual electricity consumption sourced from the Myanmar Statistical Yearbooks, published by the Central Statistical Organization (CSO) under the Ministry of Planning and Finance. The Augmented Dickey-Fuller (ADF) unit root test was applied to determine the stationary of the data series. Three distinct univariate forecasting techniques the Box-Jenkins Autoregressive Integrated Moving Average (ARIMA) model, Brown’s Double Exponential Smoothing model, and Holt’s Double Exponential Smoothing mode were implemented to capture modeling trends. The best-performing forecasting model was determined through a comparative evaluation of standard residual fit statistics, specifically the Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and the Bayesian Information Criterion (BIC). The initial ADF test for stationary yielded a value of 3.68 with a p-value of 1.00, confirming that the original series was non-stationary and required a first-differencing transformation. Ljung-Box Q-statistic of 18.74 (df = 18, sig = 0.82) indicated no significant residual autocorrelation. In comparison, Holt’s model produced error accuracy values of 1285.32 (RMSE), 7.03% (MAPE), and 14.45 (BIC). Brown’s Double Exponential Smoothing model achieved an RMSE of 1277.84, a MAPE of 7.53%, and a BIC of 14.37. Based on the superior fitness parameters of the ARIMA model, the 95% confidence interval forecast estimates for electricity consumption (in kWh) are 33,977.3 for 2024–2025, 36,809.3 for 2025–2026, 39,877.3 for 2026–2027, and 43,201.0 for 2027–2028. Identifying precise time-series frameworks such as the ARIMA model provides highly accurate forecasts capable of capturing short-term fluctuations and data shifts within the temporal series. Regular implementation and monitoring of these refined statistical models, using continuous secondary updates, allow for the stable identification of long-term electricity demands. Consequently, these quantitative forecasting insights strategically empower energy planners and government policymakers to balance industrial expansion and economic growth with the long-term sustainability of the national grid infrastructure.
Keywords: Annual electricity consumption, Myanmar, time-series forecasting, ARIMA model, Brown’s double exponential smoothing, holt’s double exponential smoothing, augmented dickey–fuller test, energy planning, forecasting accuracy, national grid planning