Demand Volume Forecasting for Wood Products Using ARIMA and Holt’s Double Exponential Smoothing
Abstract
This study compares forecasting methods for predicting wood product demand volume measured in cubic feet in an export-oriented make-to-order furniture manufacturing company. Demand fluctuates because production depends heavily on customer orders, making early planning for capacity, storage, container allocation, and workload preparation more difficult. Monthly demand data from January 2021 to December 2025, consisting of 60 observations, were aggregated into two-month periods by summing every two consecutive monthly observations, resulting in 30 observations. A logarithmic transformation was applied to reduce the influence of extreme values. A quantitative time-series approach was used to compare Autoregressive Integrated Moving Average (ARIMA) and Holt’s Double Exponential Smoothing. The evaluation used a fixed holdout test set consisting of 24 training observations and 6 testing observations, representing a 12-month forecast horizon. Forecasts generated on the transformed scale were converted back to the original cubic-feet scale before calculating MAE, MSE, and MAPE. The results show that ARIMA(0,1,1) achieved slightly better predictive performance, with an MAE of 26,177.90, an MSE of 1,159,058,148.96, and a MAPE of 25.08%, while Holt’s Double Exponential Smoothing produced an MAE of 27,739.79, an MSE of 1,177,626,282.12, and a MAPE of 27.45%. However, the MAPE indicates moderate forecasting accuracy; therefore, the model should be used as a preliminary planning aid combined with managerial judgment rather than as the sole basis for operational decisions.