This study investigates the monthly air conditioner (AC) production trends in Bangladesh over a 12-year period from January 2013 to December 2024. The primary objective is to identify production patterns, seasonal behavior, and forecast future production to support strategic decision-making in inventory management, marketing, and resource allocation within the Bangladeshi AC market.The dataset exhibits an additive trend with multiplicative seasonality, characterized by higher production in March–June and lower output in October–January. To model and forecast these dynamics, four classical time series approaches were employed: ARIMA, Exponential Smoothing State Space Model (ETS), STL+ARIMA, and STL+ETS. Each model was trained and tested using an 84:16 data split, and their performances were evaluated using key error metrics such as MSE, RMSE, MAE, and MAPE.Among the models, the ETS (M,A,M) model—representing the Multiplicative Holt-Winters approach—showed the best forecasting performance, achieving the lowest error values (MSE = 13,860,263; RMSE = 3,722.94; MAE = 2,846.54; MAPE = 33.86%). This model effectively captured the additive trend and multiplicative seasonality present in the data. In contrast, the STL-based hybrid models (STL+ARIMA and STL+ETS) exhibited higher error values, indicating relatively weaker adaptability to the dataset.The five-year forecast (2025–2029) indicates a continuing upward trend in production, with peak levels expected during the summer months, potentially reaching up to 40,000 units in the highest demand periods. These findings demonstrate that time series forecasting can play a crucial role in enhancing production planning, optimizing inventory management, and aligning marketing strategies with seasonal demand fluctuations in the Bangladeshi air conditioner market.

Keywords

Hybrid time series models, Classical time series models, Time Series Forecasting, ARIMA, ETS, STL Decomposition, Seasonality, Air Conditioner Production.