Stochastic Demand Forecast and Inventory Management of a Seasonal Product in a Supply Chain System

Estimation of seasonal demand prior to an active demand season is essential in supply chain management. The business cycle of the seasonal demand is divided into two stages: stage-1, the slow-demand period, and stage-2, the peak-demand period. The focus here is to determine an appropriate demand forecast for the peak-demand period. In the first set of forecasting model, a standard gamma and an inverse gamma prior distribution are used to forecast demand. The parameters of the prior model are estimated and updated based on current observation using Bayesian technique. The forecasts are derived for both complete and incomplete datasets. The second set of forecast is derived by ARIMA method using Box-Jenkins approaches. A Bayesian ARIMA is proposed to forecast demand from incomplete dataset. A partial dataset of a seasonal product, collected from the US census bureau, is used in the models.Missing values in the dataset often arise in various situations. The models are extended to forecast demand from an incomplete dataset by the assumption that the original dataset contains missing values. The forecast by a multiplicative exponential smoothing model is used to compare all the forecast. The performances are tested by several error measures such as relative errors, mean absolute deviation, and tracking signals. A newsvendor inventory model with emergency procurement options and a periodic review model are studied to determine the procurement quantity and inventory costs. The inventory cost of each demand forecast relative to the cost of actual demand is used as the basis to choose an appropriate forecast for the dataset.This study improves the quality of demand forecasts and determines the best forecast. The result reveals that forecasting models using Bayesian ARIMA model and Bayesian probability models perform better. The flexibility in the Bayesian approaches allows wider variability in the model parameters helps to improve demand forecasts. These models are particularly useful when past demand information is incomplete or limited to few periods. Furthermore, it was found that improvements in demand forecasting can provide better cost reductions than relying on inventory models.