Determining optimal data length for short-term business forecasting: an empirical study in catering services

Purpose This study investigates the impact of data length on daily demand forecasting accuracy within a horizontal time series structure. While time series data can be structured hierarchically or as forecast intervals, this research focuses on optimizing data length to enhance forecasting performance. Design/methodology/approach Applying a time series model and the Hidden Markov Model (HMM), this study evaluates daily meal data from a large-scale catering service provider to analyze the impact of different data lengths on forecasting accuracy. Findings The results indicate that forecasts based on the 3-month data length achieve higher accuracy than those using the 24-month data length. This suggests that utilizing shorter, more recent data segments can enhance forecasting efficiency without compromising accuracy, challenging the traditional assumption that longer historical datasets always yield better predictions. Practical implications Forecasting models should align with business strategy and operations. Accurate data collection and validation are essential to prevent forecasting errors. Additionally, regular performance assessment using appropriate accuracy metrics is necessary for refining forecasting models. Originality/value This study empirically demonstrates that data length is a key structural factor in forecasting accuracy. Unlike previous research that primarily emphasizes model selection and parameter tuning, this study highlights how optimizing data length improves demand forecasting. By applying these findings to the catering service industry, this study provides actionable insights for businesses seeking to improve operational efficiency through optimized forecasting methodologies.