Accurate forecasting of daily consumer spending is crucial for strategic decision-making in the retail sector, yet the dynamic influence of weather remains underutilized in predictive models. Grounded in the Stimulus-Organism-Response framework and demand theory, this study examines how weather acts as an environmental stimulus triggering behavioral responses that differentially affect spending across sectors of varying demand elasticity. We present a comprehensive evaluation of weather data integration for consumer spending prediction across three retail sectors: grocers, home improvement, casual dining. We employ a robust methodology involving eight distinct machine learning models, from linear regression to ensemble methods, each of which is trained with and without weather features to isolate meteorological contributions independent of algorithmic choice. Our experimental framework encompasses 1.2 million individual model training runs across all 50 US states over 10 years, evaluating multiple scenarios ranging from operational forecasting to theoretical performance bounds. Models incorporating weather data achieve a mean symmetric Mean Absolute Percentage Error (sMAPE) improvement of 11.5 % compared to baselines using only economic features, with some methods exhibiting statistically significant gains in 74 % of combinations across states and industries. Performance gains vary systematically by sector, with grocers achieving 20.2 % improvement, casual dining 12.2 %, and home improvement 3.3 %, reflecting differential weather sensitivity across necessity versus discretionary goods, consistent with demand theory predictions. These findings demonstrate weather data’s substantial predictive value for consumer spending forecasting across diverse machine learning approaches and geographic contexts, with sector-specific performance differences reflecting underlying demand elasticity and weather-driven behavioral mechanisms predicted by economic theory.
扫码关注我们
求助内容:
应助结果提醒方式:
