A Hybrid Forecasting Technique to Deal with Heteroskedastic Demand in a Supply Chain

Abstract

Under demand uncertain environment, maintaining a proper safety stock is very important to cope with the stock-out situation. Improper estimation of safety stock quantity leads to an improper estimation of the order and further causes bullwhip effect and net-stock amplification. In practice, demand is heteroskedastic in nature i.e. the variance of the demand varies with time. Therefore, it is important to predict the changing demand variance to update safety stock level in each replenishment cycle. The Autoregressive Integrated Moving Average (ARIMA) model applied to predict the mean demand assuming it is homoscedastic in nature. Whereas, the Generalized Autoregressive Conditional Heteroskedasticity (GARCH) model deal with heteroskedastic demand and help in projecting the changing demand variance. Hence, a combined approach of ARIMA and GARCH (ARIMA-GARCH) model has been proposed to evaluate the safety stock level and order quantity. The performance of ARIMA and ARIMA-GARCH has been evaluated considering the demand from a cement manufacturing company. The cement demand is seasonal in pattern and highly fluctuate. Using cement demand data, ARIMA (2, 1, 1) (0, 1, 1)12 and GARCH (2, 1) model is identified to forecast 12-months ahead mean and variance of demand to determine the safety stock and order quantity in each replenishment cycle applying the equations proposed by Zhang (2004) and Luong & Phien (2007). Further, bullwhip effect and net-stock amplification ratio are estimated to evaluate the performance of ARIMA-GARCH model against the ARIMA model. From the study, it has found that ARIMA-GARCH model outperforms the ARIMA as it updates the safety stock to calculate order quantity in each replenishment cycle.

Applications of Artificial Intelligence for Demand Forecasting

“…Also, (Nguyen et al, 2021) discovered that LSTM outperformed ARIMAX regarding operating and financial metrics such as bullwhip effect, net stock amplification, transportation cost, and inventory turn. Similarly, (Jaipuria & Mahapatra, 2021) developed a hybrid ARIMA and GARCH (ARIMA-GARCH) approach to evaluate the safety stock level and order quantity. Its performance was better than ARIMA in terms of BWE and net-stock amplification.…”

“…Predictions help balance demand and supply at the customer level and utility planning (Bot et al, 2020). Demand variability is one of the main variables utilized for calculating the safety stock held by a firm for handling stock-out circumstances caused by fluctuations in supply and demand (Jaipuria & Mahapatra, 2021). Similarly, (Syahrir et al, 2022) developed an effective drug order system to estimate optimal order quantities, frequencies, and safety.…”

Research on the Application of Multi-Objective Algorithm Based on Tag Eigenvalues in e-Commerce Supply Chain Forecasting

“…Dong performed multi-objective optimization of hybrid composites based on multi-objective *Corresponding Author www.ijacsa.thesai.org regression to verify the effectiveness of the positive hybrid effect in improving the flexural strength of materials [7]. Irodov et al solved the multi-objective optimization problem of pellet burners by performing a multi-objective regression optimization analysis on the work of tubular gas burners [8]. Wang et al solved the multi-objective constrained optimization problem of variables by introducing random forests and other methods to continuously approach the objective and constraint functions [9].…”

Non-stationary time series data for natural rubber inventory forecasting: A case study

“…ARIMA models have shown high accuracy and precision in predicting time series data at the nearest lag. However, the ARIMA-GARCH model may be superior to ARIMA since it updates safety stocks and calculates order quantities at each replenishment cycle [26]. LSTM models have also demonstrated superiority over ARIMA and have been found to reduce error rates and improve long-term projections [27]- [30].…”