Cholesky identification

[BenjaminBorn]

Hi Luca,

Sorry to bother you again but I tried to use your package to replicate some Cholesky identified IRFs generated using old Matlab codes of mine and I didn't succeed. I then noticed that in the irf_chol() function you have the following line:
https://github.com/lucabrugnolini/VAR.jl/blob/2f0f8e51e32c92494df1598b1095b5718811cf2b/src/VAR.jl#L548
This is puzzling as full(cholfact(A)) just reconstructs the matrix A but I think what you need is the lower diagonal matrix cholfact(A)[:L]. See the following example

A = [2.0 2.0; 2.0 2.0]
Base.LinAlg.Cholesky{Float64,Array{Float64,2}} with factor:
[1.41421 0.0; 1.41421 2.10734e-8]

cholfact(A)[:L]
2×2 LowerTriangular{Float64,Array{Float64,2}}:
 1.41421   ⋅        
 1.41421  2.10734e-8

full(cholfact(A))
2×2 Array{Float64,2}:
 2.0  2.0
 2.0  2.0

If I replace line 548 by mSigma = cholfact(V.Σ)[:L], I can replicate my results. I'm probably missing something and it would be great if you could help me out.

Best, Benjamin

[lucabrugnolini]

Hi Benj, Thanks, I will look into that. My prior is that some changes in the Julia function performing Cholesky could have created some bug in the code. It would be not the first time given that, as far as I remember, the built-in function was just modified in the previous version and I had some headaches. I used the code to replicate Kilian and Kim (2011) paper in the past, thus I am pretty confident with results. However, I will tell you my posterior as soon as I check it. Meanwhile, I am adding you to the project Luca 2018-07-03 22:48 GMT+02:00 Benjamin Born <notifications@github.com>:

…

Hi Luca, Sorry to bother you again but I tried to use your package to replicate some Cholesky identified IRFs generated using old Matlab codes of mine and I didn't succeed. I then noticed that in the irf_chol() function you have the following line: https://github.com/lucabrugnolini/VAR.jl/blob/ 2f0f8e5/src/VAR.jl#L548 This is puzzling as full(cholfact(A)) just reconstructs the matrix A but I think what you need is the lower diagonal matrix cholfact(A)[:L]. See the following example A = [2.0 2.0; 2.0 2.0] Base.LinAlg.Cholesky{Float64,Array{Float64,2}} with factor: [1.41421 0.0; 1.41421 2.10734e-8] cholfact(A)[:L] 2×2 LowerTriangular{Float64,Array{Float64,2}}: 1.41421 ⋅ 1.41421 2.10734e-8 full(cholfact(A)) 2×2 Array{Float64,2}: 2.0 2.0 2.0 2.0 If I replace line 548 by mSigma = cholfact(V.Σ)[:L], I can replicate my results. I'm probably missing something and it would be great if you could help me out. Best, Benjamin — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#2>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AKLhwIrn3teAyG-FNEL64nFAcHRvHEgCks5uC9iMgaJpZM4VBmm8> .

-- *Luca Brugnolini* *Central Bank of Malta, Research Department* *University of Rome Tor Vergata, Department of Economics and Finance* *Personal website <http://lucabrugnolini.github.io>*

[BenjaminBorn]

Thanks for looking into this and for adding me to the project.

I guess it would be worthwhile to set up some tests so that if Julia syntax breaks we are alerted. Maybe we could test against the codes supplied with the Kilian/Luetkepohl book at: http://www-personal.umich.edu/~lkilian/book.html

[lucabrugnolini]

Agree, this was exactly the book I had in mind for the implementation of the various identification strategies. Using it also for testing it's actually a great idea. 2018-07-04 13:34 GMT+02:00 Benjamin Born <notifications@github.com>:

…

Thanks for looking into this and for adding me to the project. I guess it would be worthwhile to set up some tests so that if Julia syntax breaks we are alerted. Maybe we could test against the codes supplied with the Kilian/Luetkepohl book at: http://www-personal.umich.edu/~lkilian/book.html — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AKLhwF__Awb7uQh35N7pPdM1eFDRTaXhks5uDKg4gaJpZM4VBmm8> .

-- *Luca Brugnolini* *Central Bank of Malta, Research Department* *University of Rome Tor Vergata, Department of Economics and Finance* *Personal website <http://lucabrugnolini.github.io>*

[lucabrugnolini]

I did some fix and added an example in test/runtest.jl with known DGP. Have a check if you still get the same problem, and in case let me know.

Thanks

[BenjaminBorn]

Yep, that fixes the issue. Thanks a lot.