Layout of stage1_estimates file
(sas version and csv
format)
stage1_estimates_naics1997 (for 466 6-digit NAICS industries for 1997)
stage1_estiamtes_sic1992 (for 457 NAICS industries for 1992)
stage1_estimatesNAICS97_for_gauss.asc
(this is an ascii version of the above for input into
Gauss)
August 1, 2012
Files used in paper "An
Alternative Theory of the Plant Size Distribution, with Geography and Intra-
and International Trade"
by
Thomas J. Holmes (University of
Minnesota, Federal Reserve Bank of Minneapolis, and NBER)
and
John J. Stevens (Board of
Governors of Federal Reserve System)
Note: The statistics
reported in this paper derived from Census micro data were screened to ensure
that they do not disclose confidential information.
Description of File:
These files contain the
first stage estimates of the parameters of the distance adjustment function for
each industry. The cost efficiency
parameter γi for each location i, i = 1,...,177,
is not reported here. (In other files,
we report the γi that arise in the
second stage iterative procedure.)
For the estimate, we define distbar and then used shipment observations with distance
(in miles) greater than or equal to distbar. For most cases we use distbar=100. As explained in the paper, if, after exlcuding shipments below 100 miles, the implied value of
a(100) in the semilog case turned out to satisfy
a(100)≤.2, we reestimated the model with all
the shipments and used this estimate instead.
For the five industries impacted this way, we constrained η₂=0 and just allowed for
the linear term η₁.
The scaling of eta1 and eta2
below take distance in 1,000s of miles.
If distance is in miles, the formula for the adjustment is
a(distance)= exp(-.001*eta1*distance -
.000001*eta2*distance2)
Layout of both files (see
below for layout of stage1_estimatesNAICS97_for_gauss.asc)
Variable |
Description |
naicstext (or sictext) |
text description of
industry |
naics (or sic) |
industry code |
obs_shipments_total |
number of shipment observations
for that industry, before truncating such that distance of the shipment
exceed distbar. |
obs_shipments_used |
Number of shipments
observations from the commodity flow survey used in the estimation. (The number with distance>=distbar). |
flag_spec2 |
= . means standard
estimation worked fine with two parameters, eta1 and etat2. = 1 means 2 parameter trans cost function
gave nonsense results so opted for 1 parameter function (i.e., constrained
eta2=0). = 2 means standard error on eta2 was
problematic so used 1 parameter function. = 3 we imposed eta2=0 after first running
the estimates and finding a(100)<=.2, as discussed above. We then set distbar=0
and estimated the model with eta2=0 and eta1 free. This happens for 5 cases for both NAICS1997
and SIC1992 |
eta1 (η₁) |
semilog coefficient on miles (in 1,000 mile units) |
eta2 (η₂) |
semilog coefficient on miles2 (in 1,000 mile
units) |
se_eta1 |
standard error of eta1 (semilog) |
se_eta2 |
standard error of eta2 (semilog) |
LogLike |
log likelihood (semilog) |
distbar |
dist cutoff used for this
analysis |
eta_lnln |
coefficient on lnln specification |
se_eta_lnln |
standard error of eta_lnln |
LogLike_lnln |
log likelihood |
Layout of
stage1_estimatesNAICS97_for_gauss.asc
Has no header row.
Column 1 |
naicsindex (this is an index from 1 to 473 of the 1997 NAICS manufacturing
industries, with industries sorted by NAICS.
See file mandat_naics2 for a link from naicsindex
to NAICS. |
Column 2 |
eta1 |
Column 3 |
eta2 |
Column 4 |
Loglike (semilog) |
Column 5 |
eta_lnln |
Column 6 |
Loglike (lnln) |