|
To: tenreyro@fas.harvard.edu Subject: Issues with MA data Cc: arose@haas.berkeley.edu |
For various reasons, I decided today to pursue the issue of
data-averaging. In particular, I want to figure out whether and why
averaging trade over time seems to give you lower estimates of the currency
union effect on trade (your Table 3). You may recall that when I read
your paper for the first time yesterday I couldn't really understand why
averaging the data corrected any bias (what's the bias?), and now I'm even more
skeptical.
The first step is obviously to replicate your finding.
It seems like it should be easy to do, since my data set with Glick covers all
IFS codes from '48 through '97. I've simply inserted zero values where the
trade data was missing, and proceeded (you can see this in the descriptive
statistics on "trade" which are mostly zero). I've created
(logs of) 3-, 5-, and 7-year moving averages of trade alone (and then estimate
the equation on data sampled every 3/5/7 years as appropriate). Nothing
else is changed.
Unfortunately, I've been unable to get anything close to
your results. My estimate for the effect of currency union on trade (in
pst1.log which I attach), shows an estimate of 1.19 (se .11) when I use annual
data without any correction. When I use 5-year moving averages and sample
the data every 5 years, I get 1.18 (.11). Results are comparable for 3-
and 7-yr MAs. Also, FYI, when I use unadjusted annual data every 5 years,
I get 1.27 (.11), though obviously on a smaller sample size. Results rise
somewhat if I use only the last 20 years of data.
I'll try to make my data set available on the internet soon,
so that you can play with it if you like. In the meantime, I'll just post
the output.
Over to you.
PS: I know this is going to be a pain for you (and maybe me
too). But since you only sent me your paper yesterday, there's literally
no way I could have been much faster.
********
Andrew K. Rose, Rocca Professor Tel: (510) 642-6609
Haas School of
Business Fax: (510) 642-4700
University of
California
E-mail: arose@haas.berkeley.edu
