Using
EViews to Decompose Time Series Data
1. Open EViews
EViews is a econometric forecasting package for micro computers. It is available on business school computers on the h drive.
2. Begin a Workfile that contains data base. In the File menu click on New and then Workfile. Designate the workfile frequency and beginning and ending dates. The ending date includes the periods in the future (ex ante) that you wish to forecast.
Observations are expressed in columns in EViews. Two columns are reserved for the constant c (intercept) and residuals (resid) and are already expressed in the workfile.
3. Enter your quarterly data into your workfile.
Data may also be imported from Excel file using the import command under the file menu. (Note the Excel file must be closed in order to import. New data series can be inserted under the Object menu as New Object which when named is included in the workfile. Data can be inserted or deleted by double clicking on the series and clicking on the Edit +/- command. (Data may also be Fetched from your data disk if it has been previously Stored)
After data is included in your workfile, then save the workfile by clicking on Save in the File menu. Once saved it may be opened by using the Open command on the File menu.
4.
Use the seas command to create a new series that is seasonally adjusted. Seasonally
adjust the data as follows: Highlight the Sales series.
5. View your quarterly data by double clicking on the series and clicking on the View command. Observe the line graph for trend and seasonal changes. Click on correlogram and observe the level series first to determine if a trend exist. Repeat the correlogram with 1 difference to see if a linear trend exists and the data is now stationary. Repeat with 2 difference to see if a quadratic trend exists and the data is now stationary. Observe the autocorrelation coefficients for patterns in lags of the stationary data to determine if seasonal variation exists.
6. Calculate a least squares trend line by first entering a new series, X, that starts with 1 and increases by one unit for each time period. Use the Genr command to generate values of X^2 by using the following equation: X2 = X^2. You may repeat for X3=X^3. Estimate a linear trend equation by typing in the command LS and following directions. Alternatively, you may type LS c X for a linear trend line and LS c X X2 for a quadratic trend line. (Log values could also be generated for sales that may be regressed against X for a nonlinear trend equation.) The "best fit" has the highest R-squared.
7. Determine a forecast based upon trend and seasonal factors by multiplying your trend forecast by its seasonal factor. For example, LS c X X2 is a model that includes a quadratic trend. To determine a quadratic forecast series click on the command Forecast when the regression output is shown. Forecast error analysis and graphs may be requested. The forecast series will be named (ex. salesf) and entered into your workfile. To determine a forecast based upon trend and seasonal factors genr a new forecast series by multiplying salesf by it factor. (Example: Genr salestf = salesf *factor)
8. Determine the CI relative by dividing the original data by its forecast value based upon trend and seasonal influences. Use the Genr command to determine a new CI series. (Example: Genr CI = (sales/salestf)). The CI series will vary around one or it could be multiplied by 100 to vary around a 100 base. A value of 95 means that CI influences resulted in sales being 5 percent below the amount predicted by trend and seasonal influences.