Taming Thor: A New Approach to Modeling Weather Effects
38 Pages Posted: 7 Oct 2019
Date Written: January 30, 2018
Weather effects are large, non-linear, and heterogeneous. Standard storm and element-based analyses are not designed to study weather effects on the whole. I use the LASSO machine learning method to develop a one-dimensional weather index reflecting how favorable weather conditions on the whole are for specific outcomes and contexts, capturing complex effects without over-fitting data. I illustrate the index with national daily store-level apparel and sporting goods sales. The worst five percent of weather days decrease outdoor store sales 22 percent, compared to 12 percent reductions at indoor malls on their bottom fifth percentile days. A bootstrap with historical weather data indicates that up to one-third of variability in same-store sales growth may be due to weather.
Keywords: LASSO, Machine Learning, Weather, Retail, Variation
JEL Classification: C4, Q54, L8
Suggested Citation: Suggested Citation