Training the model

We are going to use sales data of quarters for the years 2012 to 2015 to design and train the forecast time-series model. With this design, we will test the predictions with a group of known sales for 2016 to 2017. Finally, we will make a forecast for 2018 to 2019.

The model has to take both of these components (known sales and forecast) to make a good prediction. The steps to develop a forecast, as we will see in this chapter, are as follows:

1. Look at the data chart to decide whether it has autocorrelation or not.
2. Test the autocorrelation with the Durbin-Watson test.
3. Calculate the moving average (explained in Chapter 12, Working with Time Series Using the Centered Moving Average and a Trending Component) to smooth the peaks of the data.
4. Design the model, calculating the seasonal trends.
5. Test the forecast by multiplying the seasonal trend by the regression line.
6. Use the model to make forecasts.

In this chapter, we will design...