단기 운영계획을 위한 데이터 길이와 수요 특성 기반 수요예측 정확도 분석

초록

Purpose – This study examines the impact of data length and demand characteristics on forecast accuracy in short-term operational planning. Design/Methodology/Approach – Using daily demand data from a manufacturing company, we compare the performance of exponential smoothing (ETS) and the Hidden Markov Model (HMM) across different data lengths ranging from two to 24 months. Forecast accuracy is evaluated over a 21-day horizon, focusing on the second and third weeks. In addition, a rolling forecasting approach is employed to validate out-of-sample performance. Findings – The results indicate that the optimal data length varies depending on demand characteristics. For stable demand, ETS shows superior performance, particularly with 12 months of data. In contrast, for highly variable demand, HMM demonstrates improved performance, especially with shorter data lengths, such as six months. However, its performance is not consistent across all conditions. Furthermore, forecast accuracy decreases as the forecast horizon increases, with significantly higher errors observed in the third week. Research Implications – These findings highlight the importance of selecting appropriate data length and forecasting models based on demand characteristics and planning horizons. The study contributes to forecasting research by incorporating data length as a key analytical factor and provides practical insights to improve demand forecasting strategies in short-term operational settings.

키워드