How weather affects sales - analysis for demand forecasting

Analysis to measure how weather affects sales, weather sensitivity analysis, can help explain historical changes and can be an important component in forecasting.  Historical sales and weather data are used to develop a statistical model that explains the variations in demand that result from changes in one or more measures of weather.  Specialist statistical software is usually needed for weather sensitivity analysis, but when a suitable model has been found it can be implemented for ongoing use in Excel or alongside other forecasting software.

The resulting mathematical explanation of how weather affects sales can be used to remove the effects of weather variation from historical sales, so that a weather-neutral data stream can be used for demand forecasting.  This is important to help prevent the mis-interpretation of historical weather events as ongoing trend.  Medium-term weather forecasts are not reliable, but can be used in what-if analysis.  However, if daily forecasting is carried out, the short-term daily weather forecast can definitely be used in the sales forecasting process.

Seasonality and trends

Seasonality in demand can be caused by number of different factors as well as weather - school holidays for example, or seasonal events such as Xmas and Easter.  When forecasting, it is essential to test for the presence of seasonality and, if it exists, to deal with it effectively, but the problem is that short-term volatility in the weather often causes variations in sales well apart from seasonal averages.

When conducting analysis of how weather affects sales it is often necessary to seasonally adjust the historical sales and historical weather data, so that we are examining the effect of unseasonal weather.

Ongoing trends are often present too, in which case it is important to include this in the historical analysis and of course in future forecasting.

Finding the relevant measures of weather

A number of alternative weather measures are available and can be examined for their relevance regarding how weather affects sales:

• Temperature - minimum, mean or maximum (thresholds may apply)

• Rainfall (mm)

• Sunshine hours

• Wind speed - steady or gust (thresholds may apply)

These measures can correlate closely with each other, so there is a need for statistical software that can discriminate between the alternatives and can include more than one measure if necessary in a suitable statistical model.  For example it is sometimes helpful to include rainfall or sunshine hours as well as some form of temperature.

Thresholds and non-linear relationships

There are situations where sales response to weather only kicks in at certain thresholds.  For example, the use of fuel for domestic heating only starts significantly when temperatures decline to somewhere around 10 deg C, and sales of ice cream increase significantly at a certain temperature threshold. 

Also, the way that weather affects sales is not always in a straight-line fashion, therefore non-linear relationships are sometimes necessary for a good statistical model.  The use of specialist statistical software is essential for this type of analysis.

Testing for delays in sales response to weather

In some retail markets the sales response to variations in weather may take place very quickly, in which case a reaction in sales may take place in the same week or even the same day.  However, there are often delayed responses to the weather that are distributed over a longer period of time.  For example, if analysing weekly sales data, it may be necessary to include in a statistical model the weather data lagged by one or more weeks. 

Some example projects

A statistical analysis was carried out for Calor Gas, the major UK supplier of liquid petroleum gas for heating, temperature being of course a major factor.   Various measures of temperature, temperature thresholds, rainfall and sunshine hours were evaluated, alongside a number of economic indices.

Another project involved measurement of the effect of weather on insurance claims for household emergencies including boiler breakdowns.  Using the results of that analysis, a short-term forecasting system was created for the call centre, incorporating the use of daily weather forecasts.  A facility was also included to adapt the forecasts to gradual changes in the company's number of insurance clients.

A further study was for a major supplier of car breakdown insurance.  As well as the weather, a major factor in car breakdowns is the number of vehicles actually on the road, and there is no published data on that.  Therefore, the weather analysis was carried out in conjuction with a seasonal analysis.

Work was done for a major supplier of horticultural products to growers, diy retailers and garden centres.  This quantified the effect on sales of various weather measures, including temperature, sunshine hours and rainfall, together with related time lags.  The effects of weather were used to adjust historical sales in the forecasting system, prior to carrying out statistical forecasting on the resulting weather-neutral time series.

Need for expert help

Forecast Solutions Planning has the necessary software, skills and experience for the analysis of how weather affects sales and quantification of the effects of other causal factors, such as pricing or economic indicators.  We can help integrate the results into an improved sales forecasting process.