Durbin-Watson-Test with ARIMA for single time series
Dbscheuf
Diagnosing an ARIMA model for a given time series requires a closer look at the
residuals (ERR_1). But how - if at all - do you use the Durbin Watson-Test with
ARIMA?
With normal regression analysis you need an independent variable (or more) and
a dependent variable and run SPSS selecting the Durbin-Watson-Test-Option. Is
it the same with ARIMA? Is the following procedure for Durbin Watson and ARIMA
correct:
Open SPSS Menues "statistics" -> "regression" -> "linear". In the Dialog Box
include the original time series as dependent variable and the FIT_1-Series
(from arima-modelling the series) as independent. Select "Durbin Watson" under
the Statistics-Option and click OK.
Bertram Scheufele
Instute for Communication Research
University of Munich / Germany
da...@autobox.com
The Durbin-Watson test (1950) was developed to test whether of not the
residuals from an OLS model were free of autocorrelated disturbances of order
1 . Modern time series analysis is concerned with more than that.
The residuals from a model should be tested for
1. autocorrelation of all lags 2. dependence on lags of input series (if any)
3. constancy of the mean of the residuals (Intervention Detection) 4.
constancy of the variance of the residuals (regime change , GARCH ) 5.
constancy of the distribution via tests for constnacy of model parameters and
model form over time.
For more on time series , ARIMA , regression (transfer functions) , outlier
detection (Intervention Detection) please visit
http://www.autobox.com the home page for AUTOBOX
Dave Reilly
Automatic Forecasting Systems (AFS)
(devbelopers of AUTOBOX celebrating 23 years )
P.S. The quick answer is you should not use DW as it is not robust (correct)
when other gaussian violations are present.
-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own