This directory contains the July 1998 and the July 17 1995 version of
"Error Bands for Impulse Responses," by Chris Sims and Tao Zha as a zipped
Microsoft Word binary files and as pdf files \. The files'
MathType preferences are in factory.eqp.

The 1995 version of the paper is in the "m2" files.  It includes an
exposition of why the bootstrap does not produce "small sample" accuracy
in confidence intervals and of the difference between confidence intervals
and posterior probabiity intervals.  The 1998 version in the "m4" files
eliminates much of this discussion, but argues more pointedly against
trying to use confidence intervals rather than likelihood-characterizing
intervals (like Bayesian posterior probability intervals).

The directory also contains RATS input files that implement the Bayesian
Monte Carlo method for over-identified models described in the 1995
version of the paper, and a data file to go with the RATS files. We now
realize this code did not work quite as well as we thought it did.  We may
put up our current, better code, but have not yet (7/27/98).

A still earlier version of the paper was available as Cowles Foundation
discussion paper number number 1085, and a version close to that one,
without the figures, is available in this directory as m.pdf.  (The Cowles
web site lists CFDP 1085 as available in pdf form, but what is actually there
is the later version available here has m2.pdf.)  The early
version contains some mistakes, but may be of interest because it contains
a detailed explanation, cut from the later version, of how Blanchard and
Quah computed their error bands and the nature of the error they made.