DW test cannot be used for AR model?

[imthekingoftheworld]

Saw this from Wiley mock

[tickersu]

Correct. If you’re already accounting for at least some component of the autocorrelation, you shouldn’t use the DW test statistic. It essentially breaks down and suffers loss of power making it useless.

[takeo808]

DW can’t be used if the independent variable includes past values of the dependent variable (time-series model).

[S2000magician]

@takeo808

DW can’t be used if the independent variable includes past values of the dependent variable (_ time-series model _).

Specifically, an autoregressive model.

DW is fine for time-series models that aren’t autoregressive.

[tickersu]

@S2000magician

@takeo808

DW can’t be used if the independent variable includes past values of the dependent variable (_ time-series model _).

Specifically, an autoregressive model.

DW is fine for time-series models that aren’t autoregressive.

Right! I want to clarify what I meant by if you’re accounting for “at least some part of the autocorrelation”; this meant you could be using AR(1) for an AR(2) process, for example, but you still couldn’t use the DW test statistic (even a generalized form that’s designed for higher than order 1 autocorrelation) because you’re modeling the autocorrelation.