Daniel J. Henderson and Daniel L. Millimet, "Is Gravity Linear?", Journal of
Applied Econometrics, Vol. 23, No. 2, 2008, pp. 137-172.

There are 12 separate text files, all of them ASCII files in DOS format.
They are all zipped in the file hm-data.zip. Unix users should use
"unzip -a",

Naming convention:

  uscfs in title implies U.S. data from the Commodity Flow Survey (CFS)
  intl in title implies cross-country data on aggregate trade flows
  con/lib in title denotes conservative or liberal definition of
    disaggregated good types
  diff/homo in title denotes differentiated or homogeneous goods
  fs in title denotes "Full Sample"
  forecast in title denotes data set used for estimation in-sample and
    forecasting out-of-sample

uscfsfs.txt
  N = 4228
  Variables (8)
    trade (in levels)
    origin state ID 
    destination state ID
    home dummy
    border dummy
    distance
    origin state remoteness
    destination state remoteness

uscfsforecast.txt
  N = 4228
  Variables (9)
    trade (in levels)
    origin state ID 
    destination state ID
    home dummy
    border dummy
    distance
    origin state remoteness
    destination state remoteness
    sample dummy (1 = in-sample, 0 = hold-out-sample)

intlfs.txt
  N = 4315
  Variables (10)
    trade (in levels)
    country 1 ID
    country 2 ID
    currency union dummy
    common language dummy
    border dummy
    regional trade agreement dummy
    # landlocked
    # islands
    distance

intlforecast.txt
  N = 4308
  Variables (11)
    trade (in levels)
    country 1 ID
    country 2 ID
    currency union dummy
    common language dummy
    border dummy
    regional trade agreement dummy
    # landlocked
    # islands
    distance
    sample dummy (1 = in-sample, 0 = hold-out-sample)

The remaining data files are:

  condifffs.txt (N = 1901)
  condiffforecast.txt (N = 1883)
  conhomofs.txt (N = 1492)
  conhomoforecast.txt (N = 1470)
  libdifffs.txt (N = 1880)
  libdiffforecast.txt (N = 1863)
  libhomofs.txt (N = 1678)
  libhomoforecast.txt (N = 1667)

The structure of each of these data sets, including the number and order of
the variables, mirrors intlfs.txt or intlforecast.txt.  The number of
observations is in parentheses.  

Finally, Appendix B and Appendix C of the paper are made available here as
PDF files.