Possible bug in qzdiv() in gensys.jl

[MattCocci]

Noticed that line 248 of gensys.jl has the following line

    if (root[j, 2] > stake) || (root[j, 1] < -.1)

whereas the qzdiv.m Matlab function has

      if (root(j,2) > stake | root(j,2) < -.1) 

We were getting errors until we changed the index in the second part of the Julia if-statement to match the Matlab version.

[sglyon]

Thank you for letting me know.

When you made the change, did it start working properly?

If so, would you (or whoever is working with this code) submit a pull request?

Thanks!

[sglyon]

Oh and I should also mention again that the qzdiv and qzswitch functions are both temporary.

Eventually we should rely on the ordschur function that was merged into Base.LinAlg (module built in to Julia). This will not be available on 0.3.x (latest official release), but it should be fairly easy to do a temporary back port for our purposes until 0.4 is officially released.

[sglyon]

@pearlzli @MattCocci

What do you think about the following plan:

• Push changes (bug fixes) to gensys that are sitting in DSGE.jl to this repo
• Find a permanent and more visible home for that code under the QuantEcon umbrella
• Have the NYFRB and others who might want this feature access the code via its new home under QuantEcon

It is not a super high priority, but I did want to initiate the conversation.

[pearlzli]

Hi @spencerlyon2, sorry for the delay! I spoke with @micahjsmith yesterday about this, and we're definitely on board. I can fork Hobo.jl, change the call to qzdiv to ordschur, and submit a pull request. Then once you move gensys to QuantEcon, we'll be happy to use it through the QuantEcon package.

[sglyon]

That's great, thank you very much.

Looking forward to that!