Stochastic Simulation of the Energy Price Shocks and the Macroeconomy-Model

Dynare contributions and examples
1

Hello all,

i tried to replicate the Dynare simulation of the Energy Price Shocks and the Macroeconomy: The Role of Consumer Durables-Model of Dhawan and Jeske and so far i finished the code and the model is running in Dynare. As i’m trying to replicate the different simulation results in Table 4 of the paper, i’m struggling to replicate (or to be precise get values close to their simulation) their percent standard deviation. I used their parameterization and tried to change the equations to the logarithm’s of the variables as i have contacted one of the authors and he said they used the standard deviation of the log-levels. Maybe i did a mistake writing the code or just interpret the results incorrect.
The mod-file corresponds to the DEA model in the table and for example i get 0.23 for the Hours variable instead of 0.72 with botch shocks present. I know that they don’t have to be totally equal as it’s stochastic, but it feels like my results are too different to be correct.
I have attached my mod-file as well as the paper.
Maybe one of you sees a mistake in my code.
Kind regards
Nico
dhawan_jeske08_Energy_Price_Shocks_and_the_Macroeconomy_The_Role_of_Consumer_Durables.pdf (868.8 KB)
Test3.mod (8.8 KB)

2

Yes, that difference is a bit puzzling. From what I can see, you are correctly using HP-filtered standard deviations of logged variables. That should not explain the difference. I would recommend using theoretical moments at order=1. That should roughly correspond to many short simulations.

3

Thanks for the reply,
as mentioned by you i removed the periods option to get theoretical moments and set the order to 1. I run 100 simulations, and nevertheless the results didn’t change a lot. Here is a list of computed moments compared to Table 4 of the paper:
Variable: Computed Moment / Table 4 Value
Output: 1.73 / 1.41
HH Energy Use: 3.05 / 2.10
Hours: 0.23 / 0.72
The rest of the moments are off too, this is just an example.
Any further recommendations for adjustments are welcome as i wanted to replicate the simulation and then extend it with my own parameterization.
But if it’s not possible to solve this issue, i’ll probably just accept that my results are different and carry on as i’m running out of ideas how to solve it.
Kind regards
Nico

4

The generic advice in such cases is

In such cases, it’s hard to find out who made a mistake when you do not have access to the original codes.

5

I reread the part of the parameterization again and saw that they set some parameters different for the same model specification in the chapter and the appendix. Nevertheless both parameter settings can’t reproduce the values in their tables, especially i am wondering about the volatilities that should be equal by construction ,as the authors claim, to be different.
The different parameters for the same named specification DEA are:
\rho, \psi, \omega_{1d}, \omega_{1k}
values in chapter text: -2.8748, -0.7, 0.8165, 15.9726
values in the appendix: -2.7870, -3.2149, 0.5600, 17.4095
I ran 10, 100 and 1000 simulations with the simulation settings mentioned in the paper for both parameter settings and the one from the chapter text is far better, but as mentioned can’t hit the values that should be equal by construction. Here is a list with their values and mine from 10, 100 and 100 simulations respectively:
Output: 1.41, 1.39, 1.35, 1.31
Consumption: 0.80, 0.84, 0.76, 0.75
NDS ex energy: 0.43, 0.42, 0.41, 0.40
HH Energy Use: 2.10, 2.27, 1.89, 1.87
Durables: 4.55, 4.95, 4.44, 4.41
Fixed Investment: 5.37, 5.40, 5.12, 4.99
Durables + Fixed Investment: 4.26, 4.39, 4.06, 3.97
Hours: 0.72, 0.71, 0.68, 0.66

As you can see they still don’t replicate the values from the paper, but are quite near in my opinion. If no one has another suggestion, i’ll close the topic and just take this parameterization to continue working.
Thanks for the reply @jpfeifer!
Kind regards
Nico

6

From what I can see, your new numbers look reasonably close. I would proceed to work with this setup.