A nonlinear Markov switching approach in demand forecasting for aggregate production planning problem

The aggregate production planning has been constantly among the inseparable elements of the production. Regarding the complexity of production conditions, this planning plays a critical role in the success of large manufacturing companies. Demand forecasting is among the effective factors that lead to cost reduction in aggregate production planning. Since the forecast does not exactly match reality, it is necessary to minimize the prediction error as much as possible. A wrong forecast of demand leads to a production stock-out or backlog. On the other hand, the correct forecasting in the production planning leads to risk reduction and performance improvement in the company's business. Due to the fluctuating and nonlinear trend of demand and the variables affecting it in the different periods, the linear models have low efficiency in achieving asymmetric changes. Therefore, in this study, for the first time, the Markov switching model is applied to predict demand in the problem of aggregate production planning. In this regard, after demand predicting, the decision variables and total costs are estimated, and the results are compared with the actual total costs. The obtained results showed that the Markov switching model has better performance compared to the autoregressive moving average (ARMA) and vector autoregressive (VAR) models based on the cumulative absolute forecast error (CAFE) criteria and the aggregate production planning costs.