Reading 12 says: Durbin Watson is used to detect autocorrelation. Also in reading 12, it says: Durbin Watson is appropriate for trend models, but not autoregressive models. To determine whether error terms are serially correlated in an AR model, significance of autocorrelation should be tested using t-stat. The top is saying we should use DW to detect autocorrelation, while the 2nd is saying we shouldn’t use it. Can someone please explain this?
DW is used to detect autocorrecelation,
However, you cannot use it for autoregressive models
It’s partly because the AR model always exhibits some form of autocorrelation. So it would be inappropriate to test for autocorrelation using DW test. Instead of doing so, we use the t-test on the error term.