Stepwise methodology and choice

[tkon]

Hello, this is the first time i post here. Im still a student, and I have to prepare a project with eviews on a simple OLS regression with multiple variables. Im using the 7.0 version.

So i choosed my variables (11), generated the linear equation and then started to use different stepwise methods to get the best possible model. I selected between 3 models the one with the higher R² (0.97) and the lowest Akaike/Schwarz criterion (with all my betas significantly different from 0). Then i tested the normality of resids with the JB stats (the distribution is normal), the heteroskedasticity with the White ones (its homoskedastic so its fine) and the autocorrelation with the Ljung-Box test : sadly i have a serial correlation problem at all lagss and a positive one at order 1 with a low DW stats (this is also the case for the other models).

My teacher told me to use an AR model to cure the serial correlation problem so i added AR(1) in the model i selected before and generated it again => all went fine, the serial correlation disappeared at all lags (Ljung-Box Q-stats low and p>0.2 everywhere) but now half of my betas arent significantly different from 0 anymore.

So my question is : since i added an AR(1) term, do i have to restart the stepwise method from the beginning with the autoregressive term ? I mean do i have to restart the selection with all my variables instead of just adding the ar(1) term in the model selected before ?

Thanks