In the
case study, we used actual weather data in conjunction with a prior sales model to assess the impact of advertising. However, we also use weather forecasts to generate
estimates of sales
day-by-day, for the upcoming week.
As part of our services, we provide graphs of
how well our predictions matched actual
sales. The image to the right shows the predicted sales generated by our mathematical
model, represented by the curve. Bar heights indicates the actual number of sales
made each day.
|
We frequently use weather data to predict future sales. However, weather forecast
accuracy decreases as forecasters look further into the future. The yellow bands
on the graph indicate the range of sales predictions this model might produce, given
the region's past weather forecast accuracy. The bands widen as we estimate sales
further into the future because of the corresponding decrease in accuracy of weather
forecasts. View a report with more detailed
information on this important issue. |
|
|
Blue bar height indicates sales numbers we knew about when we developed the model.
Gray bars indicate the actual number of sales made, collected a week after the predictions
were made. The difference between the purple line and the height of the gray bars
indicates how closely our predictions matched actual sales.
|
Our
online regression tool lets you compare your sales to any combination of weather
variables. How would you describe the relationship between sales and high
temperature displayed in the image
below?
Some of the weather variables we examine include temperature, precipitation,
wind speed and gusts, adverse events, hours of sun, temperature range,
and many interactions between these variables and time-and-day-of-week-related
variables.