Angeloni et al. (2015) replication issues


#1

Dear MMB team,

I’ve been looking at your mod file for the paper Monetary policy and risk taking by Angeloni, Faia, and Lo Duca from the JEDC (NK_AFL15), but I’ve found some issues with it. I noticed this since I’ve also been busy replicating the paper from scratch and have been struggling getting it to work properly. These are some of the issues I’ve found with the MMB mod file:

  • The IRFs of the replication are qualitatively similar, but not quantitatively identical to the original paper (Fig. 3, pp 300). Since the calibrated parameters are identical to the original paper, the IRFs should also be identical.

  • Some of the steady state values in the MMB mod file are strange. For example, bank capital (bk) is negative in the steady state (approximately -0.007), implying that banks are operating with negative net worth. Furthermore, the gross return on capital/assets in the steady state is approximately equal to 1.40. This would imply net quarterly returns of 40%, which is also rather extreme. I also think the steady state value for the nominal interest rate is a bit strange: the mod file says this is 1 divided by the subjective discount factor, but equation (4) (pp. 292) shows that the variables g and phi should also feature in the steady state condition.

Sorry if this sounds a bit harsh, because I do appreciate the effort everyone’s putting in this wonderful project!


#2

Dear macro_econ,

many thanks for your mail and the interest in the MMB.

Regarding your first point, one should mention that both the consumption and the investment response are scaled by their inverse steady-state ratio, in line with the original paper IRFs and exactly replicating them. However, you are right to mention that the IRFs for the deposit ratio and bank riskiness are slightly off in quantitative terms.

We also apparently missed the issues you mentioned with respect to the steady-state during our replication procedure. Following your message, we are currently in contact with the original authors to answer your questions appropriately. Thank you for raising these issues, we will come back to you as soon as possible.

Best regards,
Philipp (from the MMB team)


#3

Thank you so much Philipp!


#4

Dear macro_econ,

after correspoding with the authors, we could verify that the code in our replication package of the MMB matches the original code.

Regarding the IRFs, my guess is that the plots of the IRFs for the deposit ratio and bank riskiness were slightly rescaled in the case of the monetary policy shock. As the other responses match for the monetary policy shock and all responses match for the technology shock, I see no other possibility.

Regarding the steady state values (and here I summarize the answer from the authors): this paper is a spin off of the JME paper by Angeloni and Faia (2013). The JME model was written in Matlab and in that version they checked the empirical plausibility of the steady states. The model in the JEDC paper (the one in our database) was written in Dynare. Here, the authors adapted the model framework, but focussed primarily on the analysis of the dynamics. As the steady state values were using different solvers that could have been a reason for why some steady states differ from the original, conventional levels.

Thanks again for raising the question and your patience in awaiting the answer,

Best regard,
Felix (from the MMB team)


#5

@fstrobel If I understand you correctly, we have to take the authors at their word. Or did you verify that the steady state of the 2013 JME paper are OK? I am asking because some time ago I came across this post:

I don’t see how the resource constraint of the paper adds up, because the transfers to new bankers seem to be missing.
Footnote 16 of the working paper (e.g. http://www.norges-bank.no/Upload/Konferanser/2011-06-09/Faia1.pdf) states that
“Notice that the bank capital accumulation equation includes the money that households transfer in every period to new bankers, given by a fraction of the value of the project: \phi Q_t K_t.The steady state value that helps to pin down the return on asset, R_A is 0.075. Since such term is negligible we have omitted that in the dynamic.”
That implies that the equation (13) it refers to in the working paper is not the actual equation used. Rather, the equation needs to be amended by an additional term -0.075QK to make the resource constraint of the economy hold in steady state (as it is only omitted in the dynamics).
However, the same equation reappears in the published version as equation (8), but the corresponding footnote (16) does not mention this “trick”, which still seems to be necessary.


#6

Hi JohnDoe,

we have only verified that the replicated code of the JEDC model that features in our MMB matches the original code, and that the problem in the steady state of that paper, which macro_econ spotted, already exists in the original code.

The statement, that I passed on from the authors, is meant to provide an explanation for why they set up the steady state that way and how the problematic values entered the steady state of the JEDC paper.

However, we have not attempted a replication of the JME model, so we haven’t put the statement of the authors to a test. Nonetheless, I hope that information helps you and macro_econ in your own replication efforts.

You point out that the problem in the steady state carries over to the JME paper. Have to tried to set up the code for this model? If so, and if you can’t replicate the results of the JME paper without a clean steady state, I suggest you contact the authors about it. When we find inconsistencies in a code or are not sure, whether and/or where we might have made a mistake in a replication, we contact the respective authors as well. Usually, they are helpful, and ideally they send you the code.

Best,
Felix