Analysis of Trend and Seasonality in Live Streaming Sales and SARIMA Prediction Research
Sales forecasting serves as a cornerstone of modern e-commerce supply chain management, playing a pivotal role in inventory optimization and logistics planning. Diverging from traditional shelf-based models, livestreaming e-commerce is characterized by prominent “pulse-style” consumption patterns, featuring complex data dynamics and inherent seasonal fluctuations. Conventional simple moving average methods fail to capture such non-stationary volatility, often resulting in significant forecasting lag. Based on daily toy sales data from an e-commerce platform during 2022–2024, this study employs STL decomposition to extract long-term trends and weekly seasonality. It constructs a SARIMA (1,1,1)×(1,1,1)₇ model for empirical prediction. The model achieved a Root Mean Square Error (RMSE) of 599.30 and a Mean Absolute Percentage Error (MAPE) of 21.46%. The results indicate that while SARIMA effectively captures the weekly periodicity in livestreaming sales, it exhibits limited capacity to predict “pulse effects” triggered by marketing campaigns. The research further investigates the impact of livestreaming interactions on model residuals, validating both the strengths and weaknesses of statistical models in identifying periodic regularities. Potential directions for future enhancements are also discussed.