How do you apply a linear regression built in SPSS to new data and generate prediction intervals I am trying to use SPSS to build a linear regression on historical data (dependent and independent variables) and then apply this to new data (independent variables only) to generate predicted values and associated prediction intervals.
I've looked in detail at the documentation on the REGRESSION procedure within SPSS, and while it is obvious how I would get the prediction and interval for the data used to build the regression (using a /SAVE subcommand to save the temporary variables PRED, LICIN and UICIN) I'm not seeing any functionality that would allow me to apply this to new data.
Essentially I'm looking for the equivalent of PROC SCORE in SAS, or predict.lm in R.
 A: If you have SPSS Version 19, I believe they introduced "Scoring Wizard" under Utilities that apparently can accomplish this sort of task.  That said, I have tried to get it to work and do not have the desire to debug the errors I am getting since it is very easy to do in R.  
I echo @Jeromy's response; if you need to stay within SPSS, I would use the R plugin and the ?predict function.  
A: I believe the ability to save the parameter file and score new data is in quite older versions than 19, but a general solution is to stack the datasets and run the regression with the original data, then save the predicted values. As long as the hold out data has all of the independent variables SPSS will still provide predictions, even if the dependent variable is missing. Below is an example:
*Old data.
SET SEED 10.
MATRIX.
SAVE {UNIFORM(100,2)} /OUTFILE = * /VARIABLES = X1 X2.
END MATRIX.
DATASET NAME Old.
COMPUTE Y = 0.5*X1 + -0.2*X2 + RV.NORMAL(0,0.05).
EXECUTE.

*New data without Y.
MATRIX.
SAVE {UNIFORM(100,2)} /OUTFILE = * /VARIABLES = X1 X2.
END MATRIX.
DATASET NAME New.

*stack the files on top of one another.
DATASET ACTIVATE Old.
ADD FILES FILE = *
  /FILE = 'New'
  /IN = HoldOut.

*Now run the regression and save the predicted values.
REGRESSION
 /DEPENDENT Y
 /METHOD=ENTER X1 X2
 /SAVE=PRED(PredVal) ICIN(Int).

The ICIN command saves the upper and lower limits for the 95% prediction intervals to the new data.
A: Why would you use linear regression on time series in the first place ? If you have time series data there may be lags required for all series and adjustments for Pulses , Level Shifts , Seasonal Pulses and /or Local Time Trends. Additionally you might have parameters that change over time (N.B. this is not rectified by Arima structure) and/or error variance that may change over time (N.B. Not necessarily rectified by Power Transforms such as reciprocal square roots, logs et. al.).You might need to update your tool set as you are abusing the methodology of linear regression by using it incorrectly on time series data.
You should be using Transfer Function Models (Chapter 10) in the seminal BoX-Jenkins text on time series analysis. Routine implementation of these procedure facilitate re-use of models, re-estimating parameters and even augmenting the older model with newly identified structure from the "new data". Try Googling Transfer Functions or AUTOMATIC Transfer Functions
