In the previous two chapters, we looked at univariate, or time series, forecasting methods that only used data from the sales history to produce their forecasts. These methods therefore don't make use of information on factors that might be influencing or driving sales, such as expenditure on marketing activities, pricing, or the weather. By modeling the effect of these factors, we might obtain more accurate forecasts. But this is not guaranteed; the simpler univariate methods often do better.
In this chapter, we will look at how to use your software to create models that attempt to explain variations in sales by measuring the influence of potential drivers. If you have information on the future values of these drivers, you can then use the models to produce forecasts. The models we will look at are called regression models. The process of obtaining these models is underpinned by a number of technical assumptions that you will find in the appendix to the chapter. We start by looking at the simplest form of regression (so-called bivariate regression) where only one factor is used to predict sales. Towards the end of the chapter we will compare the advantages and disadvantages of univariate methods and regression.
6.2 BIVARIATE REGRESSION
Consider the following example. A supermarket suspects that sales of 10.5-ounce cans of chunky chicken soup are related to the weather, with more soup being purchased in colder weather. It gathers ...