I was searching for approximation around risk steady state/stochastic SS when I came across this article titled “New Approaches for Modeling Risk in Macroeconomics The Risky Steady State” by Nicolas Coeurdacier et al (2011, AER). They showed that non-stationarity in SOE RBC models (Refering to Schmitt-Grohé and Uribe 2003 closing devices, which provides solution to non-stationarity) is because of approximation around a deterministic steady state instead of the risky one. If we use approximate around SSS then we will not have non non-stationarity.
I have seen old discussion on the forum regarding approximation around the stochastic steady state. But, I could not find any concrete methods or codes to use this methods. Would you please let me know if it is possible to approx. around SSS in Dynare. Or, there any codes. I have seen Michel Juillard but it did not work which posted a question in another post. Risky steady state file given by MichelJuillard, error message
As far as I know, several people failed in replicating that paper. See e.g. the old paper by Giovanni Lombardo.
Okey Thanks a lot ! This is enough for me to know that. Because I was wondering why this their method did not get popular.
Referring to one of your old notes as above. I tried Nonlinear moving average method of Alexander Meyer-Gohde. So, there is method is approximation around the risky steady state right?
Using their add in “Nonlinear_MA 1.0.8.” for dynare at order 3 I found that It produced very similar second moments as when I do it with dynare with built in pruning. I am wondering why it is? Why there is no difference between these.
I dig a bit deeper I found that pruning technique of Andreasen (2013,2017) and LAN, H., AND A. MEYER-GOHDE produced very similar second moments at the 3rd order. I also checked with ’fernandez-villaverde_et_al (2011) pruning technique as you mentioned too in your dynare version of risk comments paper. However, it produced very large second for some model I am using. It also produced very higher persistence in autocorrelation of consumption, output etc. (It means there is unit root).
Moreover, I found that DEN HAAN, W. J., AND J. DE WIND
(2012) is best in the sense that second moments matched with the data very well. I found comments of Stéphane as mentioned at forum "I also Other alternatives, which I tend to prefer but are not yet implemented in Dynare, exist. For instance, see den Haan et alii"
I think, these are only creditable pruning techniques?
My understanding is that the pruning scheme by Andreasen et al and Lan/Meyer-Gohde are virtually identical, with the former being the more prominent source. The one by Fernandez-Villaverde et al. always struck me as wrong an I am not sure that it really prevents explosiveness.
The Den Haan/De Wind paper is very different and harder to implement, but conceptually preferable.
Thanks for your input. I utilized codes for pruning codes using pruning_abounds.m given by Lan/Meyer-Gohde. First, I compare, moments by using dynare built in pruning option with their codes for Andreasen et al. I wanted to make sure these codes are fine. so Lan/Meyer-Gohde Andreasen and dynare pruning give same moments. So codes are perfect. I did not use NLMA rather simulated moments with half million simulation in dynare.
In the same way, I utilized their codes for Den Haan/De and Fernandez-Villaverde et al. This is how I found that Den Haan/De is much better than any others.