Analyzing One-Step and Multi-Step Forecasting to Mitigate the Bullwhip Effect and Improve Supply Chain Performance

This study presents the results of an investigation into effectiveness of one-step and multi-step (h-step-ahead) forecasting methods in mitigating the Bullwhip Effect and improving supply chain performance within an order-up-to-level inventory control system. the Bullwhip Effect, a phenomenon in which small variations in consumer demand cause increasingly larger fluctuations upstream in the supply chain, presents significant challenges for inventory management and cost control. Traditional forecasting methods, such as Moving Average and Exponential Smoothing, have been extensively studied for their impact on supply chain performance. This study is among the first to introduce machine learning forecasting models, specifically Long Short-Term Memory and LightGBM, within this research domain, comparing their performance under various demand conditions, including autoregressive processes with and without seasonality, as well as the well-known M5 forecasting competition dataset. The results reveal that the multi-step-ahead forecasting capability of Long Short-Term Memory and LightGBM significantly outperforms one-step-ahead forecasting in reducing demand amplification, leading to improvements in key supply chain metrics such as order fulfillment rate, variance of inventory, and average end inventory across all autoregressive and M5 series. The findings demonstrate the superior accuracy and stability of machine-learning methods, particularly in scenarios with high demand autocorrelation, seasonality, and variability. These results provide new insights into the potential of advanced forecasting techniques to better manage supply chain variability and reduce the Bullwhip Effect, thereby offering valuable guidance for optimizing inventory control strategies.