Hi guys,
I have a general question first: Is it possible to take any model being estimated with proc model, and estimate it in proc autoreg (in other words is it always true that if a model can be estimated with proc model it can also be estimated with proc autoreg?
I ask because I have an autoregressive error correction model that has long since been estimated with proc model, but would like to transition to proc autoreg for it's capability for backward elimination of AR terms and ability to produce Durbin-H statistics. The only reason we've used proc model is because of the general way it allows you to specify the regression model '(rather than autoregs model statement where you just list the independent variables, in proc model you can more extensively express variable relationships, i.e. the "mean reversion" term of the error correction model (γ(α0 + α1*rt-1-Rt-1)).
For instance, here is my proc model code:
proc model data=for_procmodel outparms=parms_out;
dif_dep_variable = B1*dif_index_rate + B2*lag_dif_index_rate +
B3*lag_dif_dep_variable +
gamma*(A0 + A1*Index_Rate - lag_dep_variable);
%AR(dif_dep_variable,13);
fit dif_dep_variable / dw=1 godfrey=12;
run;
I know that the gamma can be distributed as such:
dif_dep_variable = B1*dif_index_rate + B2*lag_dif_index_rate +
B3*lag_dif_dep_variable +
(gamma*A0) + (gamma*A1)*Index_Rate -gamma* lag_dep_variable;
And then the parameter estimates can later be transformed to obtain the coefficients of the original model form, but I still can't get this exactly right in proc autoreg. May anyone possibly have any suggestions (or at least be able to reassure me that this is indeed entirely possible)?
Thank you so much in advance for any advice, and just for taking the time to read & consider my problem,
-Ryan
