AR models use a lagged dependent variable as independent rght ?? isnt that a misspecification of the model

there’s no difference b/n independent variable and dependent variabe in an AR model… model correctly specified if autocorrelations of errors is not significant

In AR models the dependent variable is regressed against previous values of itself. No distinction of independent or dependent. You can’t us the DW test for these models. You must look at the autocorrelation of the residuals and make sure they are not significant. Use the T test with standard error of 1/root N. null is no serial correlation, alternative is serial correlation. That is for this test make sure that the t stat is below t critical!!! In your other regressions you wanted to reject the null of = 0. so you wanted t tests above t critical. In AR the null is no serial corr, you don’t want to reject that! If it is rejected, add another variable to the regression and test again!

What is this “Use the T test with standard error of 1/root N.” What is standard error of 1?

standard error = 1/ root of N To get the t-stat the formula is = Correlation / (1/root of N)

You mean the estimated Coefficient/ (1/root of N)

^^ no…i mean the residual autocorrelations…which are calculated using lags you’re testing if your autocorrelations are significant , so in a sense that is your coefficient in this case…