spain_shock_decomposition_y.pdf (209.2 KB) Hi there,
I am puzzled by the results of my shock decomposition. Can anyone give me so insight relative to how it looks? I have understood that the shock decomp is the joined contribution of the shocks to the variable. Since my variables are in log, it is the percentatge deviation.
I am not sure if I am interpretting good my shock decomp however. Am I right thinking that ea shock is lasting more and pushing the growth rate down?
From what I can see your data still has a trend. That is a problem. Please refer to Pfeifer(2013): “A Guide to Specifying Observation Equations for the Estimation of DSGE Models”
I have followed the dataset of Smets and Wouters 2007, and done the same operations applying logs to the data and detrending the hours, because ‘constelab’ is added and centered around a 0 value in the model.
Find the code used and the datset in the following file. Can you provide me with some guidance?
Moreover albeit I have done the same transformations as Smets and Wouters did in his dataset (I am using spanish data instead) as I understood the model features a deterministic growth rate driven by labor-augmenting technologiccal progress so the data does not need to be detrended before estimation.
Thank your dor your help.
code.zip (1.1 MB)
I am not entirely sure what you did. But some data like
labobs is close to mean 0, while you have a steady state of
1.32. If I were you, I would work with demeaned growth rates and forget about imposing cointegration for now.
Also, are you sure your data is properly seasonally adjusted?
Hereby I attach a description of the dataseries I am using, all of them has been downloaded already seasonally adjusted, except the interest rate and theaverage number of actual weekly hours in main job, however I am just using this serie for computing an index used to calculate the hours. Can the problem be related to one of these series?
Regarding the demeaning, I will look carefully at your guide and reconsider the way I have calculated the observation equations, even though I have followed the procedure that Smets and Wouters explain in his file . Note that that my GDP is not deflated since it is the real. Do you see any mistake?
Thank you for your help
data.zip (16.5 KB)
I have re-done the dataset with a series take it fro BDREMS database, used for the REMS model of the Spanish Economy, all the series are sesonally adjusted and I have used the same methodology as Smets and Wouters. However, I have some questions.
First, what do you suggest me to estimate my own constant for labor (That’s ‘constelab’) Also, should I estimate my own constants for growth rate (ctrend) inflation and interest rates (constepinf, conster)?.
Second, on your GitHub code for Smets_Wouters_2007 for dynare 4.5 the constelab parameter is also set to 0 whereas in the bayesian estimation is set with a prior mode of 1.2918. Why? The posterior mode is also found to be 1.32 as in my steady state.
Finally, when I run the estimation with my new data I got an error message because the ‘hessian matrix at the mode is not positive definite’. I guess is because I should chage the values of the parameters. Is this true? What should be my approach to this?
Thank you again for your time.
I am attaching the new data in case you want to take a look. I have much more observations and is quarterly data. The series are seasonally adjusted and at constant prices. See sheet 3. BDREMS ALTERATION.xlsx (300.8 KB)
I would need to see the final estimation and data files
I attach a folder with:
1st) dataset (sheet 2 contains the estimated series while sheet 3 the observation equations)
2nd) the matlab matrix with the observed variables
3rd) results of the code and error mesage
4th) the folder with the stored results of dynare
Last, the US_shock_decomp_mode that I am using as mode_file.
Some posterior modes as you will see are really deviated from the estimated prior mode. (csadjcost and constelab: labor constant)
How should I proceed?
results.zip (419.4 KB)
mode_compute=5 and everything runs smooth. The mode it get is
Spain_mode.mat (8.6 KB)