Scribd
Upload
117 views

Forecasting - Guide To Forecasting in SPSS

This document provides guidance on various forecasting techniques in SPSS, including multiple regression, seasonal decomposition, exponential smoothing, ARIMA, and combination forecasting. Multiple regression uses correlated independent variables to forecast a dependent variable, while seasonal decomposition extracts trend, seasonal, and irregular components from time series data. Exponential smoothing and ARIMA (autoregressive integrated moving average) models are also described for generating forecasts from time series data, as well as methods for combining different forecasting techniques into a single combination model.

Uploaded by

xueligao1001
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
117 views

Forecasting - Guide To Forecasting in SPSS

This document provides guidance on various forecasting techniques in SPSS, including multiple regression, seasonal decomposition, exponential smoothing, ARIMA, and combination forecasting. Multiple regression uses correlated independent variables to forecast a dependent variable, while seasonal decomposition extracts trend, seasonal, and irregular components from time series data. Exponential smoothing and ARIMA (autoregressive integrated moving average) models are also described for generating forecasts from time series data, as well as methods for combining different forecasting techniques into a single combination model.

Uploaded by

xueligao1001
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 3

1

Guide to Forecasting in SPSS


Fr. McGowan, S.J.
I. Multiple Regression
a. Preparing the Data
i. Plotting the Time Series
1. Analyze → Forecasting → Sequence Charts
2. Place desired variable(s) under Variables and Date variable under Time Axis Labels
3. Note that you can transform the data as well as have one chart per variable
4. Click OK
ii. Entering a Trend Component
1. Transform → Compute Variable
2. In Target Variable, enter “Trend” label
3. Select “Miscellaneous” under Function group and then select “$Casenum” from the selection below.
Use the nearby arrow to add “$Casenum”
4. Click OK. The number of cases should now be counted in the “Trend” column
iii. Entering Date Columns
1. Data → Define Dates
2. Select the appropriate date format and enter the first date for the data
3. Click OK. It will automatically create several date columns
iv. Entering a Seasonal Dummy
1. Transform → Count Values within Cases
2. Enter the appropriate title in the Target Variable and Target Label fields (e.g. Q1Dummy)
3. Select the date variables (quarters, months, etc.) and place it under Numeric Variables
4. Click Define Values, enter the desired value (e.g. 1 for Q1), click Add, then Continue
5. Click OK. There will now be “1” for you dummy and “0” in all other cases.
b. Building a Multiple Regression
i. Evaluating Variables
1. Analyze → Correlate → Bivariate When selecting variables from a list:
2. Select all independent and the dependent variable you
want to correlate; click OK  Hold Ctrl to select multiple values
3. Looking at the readout, you want to use the independent  Hold Shift and press up or down
variables that are most correlated to the dependent but arrow to select a group
not correlated to each other. Significant correlations will  Hit Ctrl + A to select all values
2
be flagged so look for the highest R among those
ii. Creating a Multiple Regression Model
1. Analyze → Regression → Linear
2. Place the desired variables in their respective Dependent and Independent fields
3. Click Statistics; check the Durbin-Watson box, then Continue
4. Click Plots; place ÐEPENDNT under X, *SRESID under Y, then Continue
a. The other way to check the error terms is to go to Save and save Standardized Residuals; then go
to Graphs → Chart Builder and graph using a scatter plot
AMDG
2

5. Finally click OK and check the Durbin-Watson and errors terms in the readout. The model and
individual variables should all be significant

II. Seasonal Decomposition


a. Creating a Seasonal Decomposition Model
i. Creating the Model and Forecasts
1. Analyze → Forecasting → Seasonal Decomposition
2. Place desired variable(s) under Variables, select Model Type, and click OK
3. Run a linear regression using the new SAS_1 variable as the dependent variable and the trend as the
independent variable. Note the y-intercept (constant) and slope.
4. You need to seasonalize your forecast so use the factors under the SAF_1 column or the SPSS
readout and add or multiply as necessary for each period

III.Exponential Smoothing
a. Creating an Exponential Smoothing Model
i. Creating the Model and Forecasts
1. Analyze → Forecasting → Create Models
2. Place desired variable under Dependent Variables, select Exponential Smoothing in the Method
drop down, and click Criteria to select model type
3. Under the Statistics tab, check any desired accuracy measures, at least Normalized BIC
4. Under Plots, select Fit Values to graph your model estimates with the data
5. If you have narrowed down your most accurate model types, go to the Save tab and save Predicted
Values. Keep in mind that you data can get clustered if you save all of the model types you test so it
may be best to only save when you have a few best model types.
6. To generate forecasts, go to Options, select “through a specified date” in Forecast Period, and enter
the latest data point you want forecast

IV. ARIMA
a. Creating an Exponential Smoothing Model
i. Viewing the ACF and PACF
1. Analyze → Forecasting → Autocorrelations
2. Place the dependent variable under Variables and click OK
3. Analyze the spikes and shapes in each and repeat with Difference or Seasonal Differencing if needed
ii. Creating the Model and Forecasts
1. Analyze → Forecasting → Create Models
2. Place desired variable under Dependent Variables, select ARIMA in the Method drop down, and
click Criteria to select model type
3. Under the Statistics tab, check any desired accuracy measures, at least Normalized BIC
4. Under Plots, select Fit Values to graph your model estimates with the data
5. If you have narrowed down your most accurate model types, go to the Save tab and save Predicted
Values. Keep in mind that you data can get clustered if you save all of the model types you test so it
may be best to only save when you have a few best model types.

AMDG
3

6. For ARIMA models, the goal is to have white noise in the residuals so also save Noise Residuals
7. To generate forecasts, go to Options, select “through a specified date” in Forecast Period, and enter
the latest data point you want forecast
8. Click OK
9. In the readout, remember that the model should have a significance over .05
iii. Testing for White Noise
1. Analyze → Forecasting → Autocorrelations
2. Place saved noise residuals from your ARIMA model under Variables and click OK
3. There should be no spikes in the graph and all points should be within the range of the black lines

V. Combination
a. Creating an Exponential Smoothing Model
i. Creating a Combination Model
1. For a combination model, you will use a multiple regression with exponential smoothing or ARIMA
2. Create your most accurate multiple regression (see above), click Save while in the window, and save
the Unstandardized Predicted Values
3. Create your most accurate exponential or ARIMA model (see above), and save Predicted Values
under the Save tab
4. Once you have both of these values, create another multiple regression but use these two columns as
the independent variables. Also, click on Options and uncheck Constant
5. Click OK. In the resulting readout, find the “B” value for each of the two variables. The two “B”
values represent the weight that each forecast will have in the weighted average of the combination
model. Multiply the forecasted values for each model by their respective weights and add the two for
the combination model forecast. The weights (“B” values) should add up to 1 and you may need to
round to make sure they do.

AMDG

You might also like