Florian Badorf

Who they are

This page records the public professional record only. Nothing adverse is recorded and no motive is attributed.

His affiliation and wider career were not verified for this page, so none is asserted here. The record used is one paper. It is the study this site relies on for two points about retail: that the response of sales to weather differs by store and by product, and that a weather forecast improves a sales forecast only over a limited horizon.

Key papers

Badorf, F., & Hoberg, K. (2020), 'The impact of daily weather on retail sales: An empirical study in brick-and-mortar stores', Journal of Retailing and Consumer Services, 52, 101921, doi:10.1016/j.jretconser.2019.101921. Read at the level of the published abstract.

His wider publication record was not surveyed for this page.

Key ideas

The study uses data from 673 stores and a random coefficient model, which lets each store have its own response to the weather. The model captures non-linear effects and seasonal differences across several weather measures, and the authors use it to measure how much weather information explains daily sales, to find store-specific effects and to see how different sales themes respond.

The main finding is that the effect is not one number. Its size and direction depend on the location of the store and on the sales theme. It can reach 23.1 per cent depending on location and 40.7 per cent depending on sales theme. A model that ignores non-linear effects can misestimate the impact of extreme good or bad weather. This is the evidence behind weather-sensitive demand.

The second finding concerns forecasting. Adding weather forecasts improved the accuracy of daily sales forecasts up to seven days ahead, though the benefit shrank at longer horizons. The practical rule drawn from it in What a Weather Forecast Does for Sales is to use a weather forecast only inside the window where it has been shown to help.

Two limits apply. The study was read here at the level of its abstract, so the retailer, the country and the product groups behind the sales themes are not described on this page. And one study of one set of stores is a pattern to test in other shops, not a rule.

How to read the result

The 673 stores are the strength of the study and also the reason to read it carefully. A random coefficient model gives each store its own response to the weather instead of imposing a single average, so the same warm day can be seen lifting one store's sales and not another's. That is the method behind the finding that the effect reaches 23.1 per cent across locations and 40.7 per cent across sales themes. The two figures are ceilings on how much the response varied, not averages, and they describe different cuts of the data: one by place, one by sales theme.

Direction matters as much as size. The abstract's point is that the effect depends on the store's location and on the sales theme, so a warm day need not lift every range of goods. A planner who applies one national temperature rule to every store is making the mistake the paper warns against. The paper also reports that models which ignore non-linear effects can misjudge extreme weather, which is when a forecast is most wanted.

The forecasting result has a boundary built into it. Adding weather forecasts improved daily sales forecasts up to seven days ahead, and the gain shrank beyond that. The sensible reading is operational: use the weather forecast inside the week, and do not build next month's orders on it. The related finding about displacement, in Martha Starr-McCluer's and Brigitte Roth Tran's work, covers what happens to sales after the weather has passed. See demand displacement.

What the record does not show

Read at the level of the abstract, the paper does not name the retailer, the country or the product groups behind the sales themes on this page, so this site does not either. It does not test whether a buyer who sees the forecast does better than a model, and the record gives no figure for the average gain from adding weather. The co-author, Kai Hoberg, is named in the citation and shares the credit for every result described here.

---