Impact of Weather on Retail Sales - Causal Analysis
If there is a suspicion that sales are affected by the weather it is a good idea to carry out a causal analysis to evaluate the impact of weather on retail sales or demand for other products and services. At Forecast Solutions we use specialist statistical software to analyse the impact of weather on demand, to determine if a causal relationship is present.
Historical data on sales and weather is analysed, including measures of temperature, rainfall and sunshine hours in order to identify the key factors. Reliable historical data on UK weather can be obtained from the Met Office or other suppliers. One good alternative is WeatherNet, a non-governmental provider of weather history and on-line weather applications.
If there proves to be a causal effect we can quantify the impact of weather on retail sales or demand for other products and services that may be further down the supply chain. After carrying out the causal analysis the findings can be used to improve the forecasting process. There will be a better understanding of the historical effects of weather and a more accurate future forecast can be made, even when a reliable weather forecast is not available (see below).
Weather Related Forecasting
To make the fullest use of any causal factor in forecasting it is desirable to make use of an accurate forecast of that causal factor over the forecast horizon. With weather data this is sometimes possible when making a very short term forecast directly of retail sales, but for companies further down the supply chain it can be more difficult. For these companies there is often a need to make a somewhat longer term forecast in larger time buckets e.g. weeks rather than days and a reliable weather forecast can be hard to find.
However, a causal analysis on weather can still improve the forecast through a better interpretation of historical effects. By way of example, if there has just been an extended period of unseasonally high rainfall and we are in the business of selling horticultural products, sales will undoubtedly have dipped during the wet weather. If the effect of that extreme weather is understood from a causal analysis, the sales history can be adjusted for the effects of weather prior to running a statistical forecast. Then the future forecast can be made on a weather-neutral basis and will not be pushed down erroneously due to the previous weather event. So there is still a significant benefit to be derived.
See some case studies on the impact of weather on sales.
Need for expert help
Although a basic causal analysis using the regression tool in Excel's data analysis tool set can give an initial indication of weather effects, specialist statistical software is usually necessary for a thorough analysis. One reason for this is that the various measures of weather such as mean, minimum and maximum temperature, sunshine hours and rainfall tend to be closely correlated with each other and are very easy to misinterpret.
The likelihood is It will be necessary to include the possibility of time lags due to supply chain effects. For example, for some businesses the effect of high rainfall this week might have a distributed effect across this week, next week, following week, etc. This process requires specialist software and, preferably, an analyst with a good body of previous experience .
Of course the weather is an integral part of market seasonality for many products. Therefore great care is needed to avoid confusion of the results with natural seasonality or inherent trends in market size or share. It is often necessary to seasonally adjust historical weather data and/or historical sales to obtain the best understanding from the causal analysis.
Forecast Solutions Planning has the necessary software, skills and experience for this type of analysis and can help integrate the results into an improved sales forecasting process. We can also analyse the impact of other drive factors such as price sensitivity or the effect of economic indices to build causal models incorporating several variables.