Accurate forecasting of daily consumer spending is crucial for strategic decision-making in the retail sector, yet the dynamic influence of weather is often underestimated or insufficiently integrated into predictive models. This study presents a comprehensive evaluation of incorporating both historical and 7-day weather forecast data on predicting consumer spending amounts across three diverse sub-industries: Grocers, Home Improvement, and Casual Dining. We employed a robust methodology involving a comparative analysis of eight distinct machine learning (ML) models, from linear regression to ensemble methods, each trained both with and without weather data to isolate meteorological contributions independent of algorithmic choice. Our experimental framework encompasses 500k+ individual model training runs across all 50 US states, multiple sub-industries, and weather configurations, representing one of the most comprehensive evaluations of weather-informed consumer spending prediction to date. Our experiments demonstrate that incorporating weather data provides broad improvements across most model-sector combinations, with models utilizing weather data typically exhibiting substantial reductions in Root Mean Squared Error (RMSE) of predicted consumer spending amounts in some cases exceeding 60\%. Post hoc analysis confirms that these improvements are statistically significant across nearly all configurations. Overall, these findings establish weather data as a broadly applicable enhancement for consumer spending forecasts regardless of the underlying ML approach, providing actionable insights for inventory optimization, resource allocation, and targeted marketing strategies in the retail sector.
Gerg et al. (Thu,) studied this question.