Page 1 of 1
Working day adjusting variable and Autocorrelation
Posted: Sat Sep 22, 2012 12:13 pm
by relaxdave
I am estimating monthly GDP, which works pretty good until I control for the number of working days per month.
When I include this variable into the models (AR, DL and ARMA), I get serious problems with autocorrelation in the residuals, the R2 is higher though.
Does anybody have an idea why this is the case and how else I could control for working days?
Calendar adjustment would be very costly since I want to pursue forecasts.
Thank you in advance!
Dave
Re: Working day adjusting variable and Autocorrelation
Posted: Mon Sep 24, 2012 2:46 am
by trubador
Autoregressive structure of these models may create such problems. If you really "do not" want to pre-adjust your variable, then you can use working days series as a factor for the dependent variable instead of an exogenous variable. You may either use (Y/wdays) or (Y*@max(wdays)/wdays) as your dependent variable...
Re: Working day adjusting variable and Autocorrelation
Posted: Mon Sep 24, 2012 7:42 am
by relaxdave
Thank you, Trubador!
I tried transforming the variable the way you described it and in the end I could get rid of the serial correlation by adding more lagged endogenous variables.
I would like to estimate seasonally adjusted data as well - the only problem I have is, that I have to forecast the seasonally UNadjusted values later and I don't know how to unadjust the forecasted data. Do you have any suggestions?
Thanks, Dave
Re: Working day adjusting variable and Autocorrelation
Posted: Tue Sep 25, 2012 4:03 am
by trubador
You can directly model the seasonality along with your model via using additional AR/SAR lags or creating seperate dummy variables. Something along the following lines:
Code: Select all
Y*wdays/@max(wdays) c ar(1) ar(12)
or
Code: Select all
Y*wdays/@max(wdays) ar1 @expand(@month)
If you want to drop insignificant seasonal dummies, then you can use @seas() command afterwards.