Comparative Forecasting Models for Optimizing MSME Production: A Time Series Analysis

Abstract

Accurate short-horizon forecasting is essential for Indonesian food-service MSMEs that plan production with perishable inputs and holiday-driven demand swings. Using monthly sales from Martabak Tip Top, Tarakan (December 2023–November 2024), this study compares a three-period moving average with single exponential smoothing under a one-step-ahead out-of-sample evaluation on a common test window. Accuracy is assessed with mean absolute percentage error (primary), mean absolute error, and root mean squared error. Single exponential smoothing delivers lower error than the moving average during the test period (MAPE 8.0 per cent versus 9.2 per cent) and projects a December requirement of about 1,710 units (moving average: about 1,720). The head-to-head evidence in an emerging-market MSME setting shows that giving greater weight to recent observations provides a more reliable operational signal than equal-weight averaging when modest level shifts occur around public holidays. Practically, using single exponential smoothing as the default planning input supports tighter bills-of-materials conversion, leaner safety-stock and reorder-point settings derived from observed forecast errors, and steadier labour scheduling, thereby reducing stockouts and waste while improving working-capital efficiency. The approach is transparent and spreadsheet-ready, offering actionable guidance for operations, finance, and policy audiences concerned with MSME performance in developing-region contexts.