I'm not sure where my confusion is stemming from, but it seems that equivalent tests (Durbin Watson, and a simple significance test) for serial correlation in the errors (of lag 1) sometimes yield different results. Consider the series,

$$Y=\beta_0+ \beta_1x_t + \epsilon_t$$

As I understand it, the Durbin watson test, is performed as follows (with hypothesis spelled out)

$$\epsilon_t= \phi \epsilon_{t-1}+v_t, \hspace{0.2cm} v_t \sim \text{iid}$$ $$H_0: \phi=0 \hspace{0.2cm} \textrm{(no serial correlation)},\hspace{0.2cm} H_1: \phi \neq 0 \hspace{0.2cm} \textrm{(serial correlation)}$$ $$DW=\frac{\sum_{t=2}^T(\hat{\epsilon}_t-\hat{\epsilon}_{t-1})^2}{\sum_{t=2}^T\hat{\epsilon}_t^2}$$

Moreover, it seems to me that the Durbin Watson test test is equivalent to testing the hypothesis that $\phi=0$, which is done by first estimating $\phi$ by OLS in the following model,

$$\hat{\epsilon_t}= \phi \hat{\epsilon_{t-1}}$$

and subsequently testing whether $\phi =0$. However, I've seen these tests yield different conclusions (i.e. failure to reject with DW and a very significant coefficient for the autoregressive model of predicted residuals). What gives?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.