Autocorrelation between error variables can be detected using DW statistic. If the model has autocorrelation, use AR(1). If the residual from AR(1) still exhibit autocorrelation, use AR(2) and so on. However it seems that DW test cannot be used to test autocorrelation where one of the explanatory variables used is lagged value of dependent variable. In that case then how can we test for autocorrelation ? A bit confused here. Thanks for any clarifications.
DW test cannot be used for AR models. for AR models, autocorrelation is detected by plootting the pests, ar model, or the dick test
There is a diff between Serial Correlation (Multiple regression) and AutoCorrelation (Time series). SC = DW AC = NO DW. AC is tested for by looking at the significance of the t-stat = 1/sqrt(n) and checking its distance from the t-crit on the t-tables. If AC is significant - start to add the lags. If lags do not work - move up to AR(2).
Thanks CP. I was incorrectly thinking that Autocorrelation is same as Serial Correlation. For some reason, Schweser uses the terms interchageably. I will appreciate if someone can shed more light on the difference between the two.
cpk123 Wrote: ------------------------------------------------------- > There is a diff between Serial Correlation > (Multiple regression) and AutoCorrelation (Time > series). > > SC = DW > > AC = NO DW. > > AC is tested for by looking at the significance of > the t-stat = 1/sqrt(n) and checking its distance > from the t-crit on the t-tables. If AC is > significant - start to add the lags. If lags do > not work - move up to AR(2). Autocorrelation is in fact tested with DW, just not when the model includes a lagged value of the dependent variable (i.e. AR model). Autocorrelation and serial correlation are essentially the same, and most texts treat them as synonymously. You can make an argument for the difference between the two, but any problems associated with one, and methods to overcome those problems, are the same.
autocorrelation only happens when the variances of the residuals are related to a “self” (auto) dependent variable. Like time e.g. in a time series.
cpk you are prob technically right but page 200 book 1 schweser says “serial correlation, also know as autocorrelation” so for purposes of the exam i will treat them as the same
Ok man…take up the argument with Gujarati and chapter 12 of his book.
we just got schweezed!