Application of Time Series Analysis Methods in Forecasting Demand and Production Planning for Ready-Mixed Concrete: A Case Study at Vina 21 Concrete Company

Abstract

This paper presented a comparative study of four time series forecasting methods—Seasonal Naive, SARIMA, Holt-Winters, and Prophet—applied to real-world data from Vina 21, an SME producing ready-mixed concrete in Vietnam. By testing two different approaches to historical data (including or excluding the Covid-affected year 2021), the study provides insights into how atypical periods may influence model performance. Prophet and Holt-Winters show clear advantages when the full dataset (2021–2023) is used, displaying reduced RMSE and more robust handling of extended historical variability. In contrast, SARIMA performed poorly or did not converge under the heavier data load, likely due to the highly irregular demand patterns introduced by pandemic-related disruptions. However, SARIMA was more feasible when the 2021 data was omitted, albeit with a higher absolute error and only modestly improved MAPE relative to other approaches. From the perspective of Vina 21’s management team, adopting a forecast model that can handle significant swings in demand and maintain reasonable accuracy across diverse conditions is critical. Prophet emerged as a consistently strong candidate, especially for minimizing absolute forecast errors (RMSE), which directly supports production planning and raw material procurement decisions. Moreover, the findings indicate that integrating all available data generally benefits methods like Prophet and Holt-Winters, enabling them to capture broader patterns in seasonal and trend components. Future work should delve deeper into integrating external factors such as weather data, government construction policies, and project-specific timelines to further refine forecasting accuracy. Additionally, exploring ensemble approaches and advanced machine learning methods could yield further improvements. For SMEs like Vina 21, these results underscore the value of data-informed decision-making, validating that the choice of forecasting model should be aligned with the enterprise’s goals—whether it prioritizes minimizing absolute risk of under/overproduction or focuses on cost control and error ratios. The research results can be expanded to benefit the construction material manufacturer sector in Vietnam.