Feature request: Seasonal dummies
Posted: Sat Mar 19, 2011 5:00 pmHi,
It would be great if you would add seasonal shocks to Dynare to assist with estimating models on non seasonally adjusted data. (Our models really ought to be able to match seasonal movements given e.g. seasonal changes in demand shocks. My strong hunch is that estimating off non-SA data would greatly help with identification too.)
There are two ways that this could be done. The easy way would be to perturb in the magnitude of the seasonal movements, meaning to a first order seasonal fluctuations wouldn't change the response to non-seasonal shocks. To solve you'd then just follow the appendix to Chatterjee and Ravikumar (1992). The slightly more correct way would be to solve for the steady state in each quarter, then to stack the resulting 4 sets of dynamic equations into one (the state vector would be expanded to include the state in each quarter, plus possibly some lagged expectation blocks to ensure that shocks from the 4th quarter aren't known in the first), see Braun and Evans (1995) and the references contained within. You could do the latter entirely in the preprocessor even if you didn't want to touch your existing solution algorithm (if I understood correctly).
Interface would be something like:
Regards,
Tom
It would be great if you would add seasonal shocks to Dynare to assist with estimating models on non seasonally adjusted data. (Our models really ought to be able to match seasonal movements given e.g. seasonal changes in demand shocks. My strong hunch is that estimating off non-SA data would greatly help with identification too.)
There are two ways that this could be done. The easy way would be to perturb in the magnitude of the seasonal movements, meaning to a first order seasonal fluctuations wouldn't change the response to non-seasonal shocks. To solve you'd then just follow the appendix to Chatterjee and Ravikumar (1992). The slightly more correct way would be to solve for the steady state in each quarter, then to stack the resulting 4 sets of dynamic equations into one (the state vector would be expanded to include the state in each quarter, plus possibly some lagged expectation blocks to ensure that shocks from the 4th quarter aren't known in the first), see Braun and Evans (1995) and the references contained within. You could do the latter entirely in the preprocessor even if you didn't want to touch your existing solution algorithm (if I understood correctly).
Interface would be something like:
- Code: Select all
set_seasonality(4);
var theta ...;
varexo epsilon ...;
parameters s1 s2 s3 s4 ...;
model;
theta = s1 * seasonal(1) + s2 * seasonal(2) + s3 * seasonal(3) + s4 * seasonal(4);
...
end;
Regards,
Tom