Commenter Tom came over for a visit to this blog through a link at Marginal Revolution. Among his objections to the information transfer model included the scenario where Russia experienced inflation with the expansion of its monetary base while the US did not experience inflation with its expansion. However, this is exactly the kind of thing the information transfer model explains. Russia appears on the left side of this graph, where the slope of the curves P(M0) and NGDP(M0) is much greater:
Russia is still on the "quantity theory" side of the graph, not the "liquidity trap" side (the right hand side). Here are the fit to the price level and the information transfer index, for reference (the lack of NGDP data for Russia before 2003 limits the range over which I could test the model):
I show the GDP deflator, CPI and CPI less food in the graph above.
Jason, kappa = log(M/c) / log(NGP/c) with M = currency component of base ("M0"), correct? Did I read on one of your blog posts somewhere that you calculated a best fit for c for one country, and then have used that same value for other countries? So for the case of Russia here, did you use a c that had already been fixed elsewhere? Thanks.
ReplyDeleteActually the piece that was fixed was that c = γ*MB0 with a fixed γ across countries:
Deletehttp://informationtransfereconomics.blogspot.com/2013/07/universalizing-model-kappa-sigma-space.html
Where MB0 is the constant setting the scale for the currency component of the base (actually, I should probably say M00, but that gets a bit confusing).
And yes, I kept γ fixed in this particular fit. However the results don't change very much if I let the fit be a 3-parameter fit, letting it choose γ as well.
Jason, thanks. I'm not quite getting this:
Delete"The former parameter is only an overall normalization that depends on the reference year for the price level, so once γ is fixed, all subsequent fits are effectively one-parameter fits."
You're referring to alpha there ("the former parameter"), which looks to be NGDP0/M0. So you're down to two parameters at that point: M0 and alpha, and alpha has M0 in the denominator. So you're saying essentially that NGDP0 shouldn't really count as a parameter?
The price level normalization is completely arbitrary (e.g. scaled to 100 in 1982), so the overall normalization (NGDP0/M0) does not add information. M0 is the only parameter once γ is fixed.
DeleteOr another way, I use the freedom to pick the year P is scaled to in order to eliminate NGDP0/M0.
Ah, OK, that makes sense. Thanks.
Delete