Looking at the formula for Durbin Watson, is the statistic biased against negative correlation? It seems like negative covariance will increase the DW statistic, making it more likely to be above the dl-du parameters whereas positive covariance has the opposite effect.
The tests seems symmetrical to me. quote: To test for positive autocorrelation at significance á, the test statistic d is compared to lower and upper critical values (dL,á and dU,á): If d < dL,á, there is statistical evidence that the error terms are positively autocorrelated. If d > dU,á, there is statistical evidence that the error terms are not positively autocorrelated. If dL,á < d < dU,á, the test is inconclusive. To test for negative autocorrelation at significance á, the test statistic (4 - d) is compared to lower and upper critical values (dL,á and dU,á): If (4 − d) < dL,á, there is statistical evidence that the error terms are negatively autocorrelated. If (4 - d) > dU,á, there is statistical evidence that the error terms are not negatively autocorrelated. If dL,á < (4 − d) < dU,á, the test is inconclusive.
I have not seen that formula, 4 - d, for Durbin-Watson. Is that from Schweser?
Yes. And from CFAI. And from wikipedia. (http://en.wikipedia.org/wiki/Durbin-Watson_statistic) and from every stats book ever. Why? Because that is the Durbin-Watson test statistic.
Chrismaths, would you mind guiding me to the page in the CFAI text deals with Durbin-Watson? I am only going by pgs. 301-302, which seem to provide a short summary. Thanks for the wikipedia tip.
pg 303. note 50
Thanks very much; I completely missed that note.