Hi XXX,

I apologize for the somewhat canned email. I have large email flows and dont have bandwidth to answer all questions in full.  The answer below is somewhat standardized as the question you ask does comes up from time to time.  Also I wanted to note that we will likely add two more years to the database at some point in the fall, perhaps around October.

*** Note

There is a richer version of the data on the website that has granularity mapped to SIC-2, which is richer than SIC-3 (in the baseline database).  This fatter network is close to full information as pairwise similarities beyond SIC-2 granularity are getting closer to zero and dont matter much economically in tests we performed over the years.

The main idea to get close to full information in the network is to use that “larger” database in full, and fill in permutations that are not in that database with zeroes.

Steps
1) You should get the "larger version" of the TNIC database:
	http://hobergphillips.tuck.dartmouth.edu/industryclass.htm
2) First extract all unique gvkeys in that database.
3) Take all pairwise permutations.
4) Merge with the larger database you downloaded into this permtuations database by gvkey pair.  If the given pair has an observation in the downloaded database, use the pairwise score from that database as the score.  If it is not in that database, set its score to zero as it means the pair is quite distantly related.

I hope this is useful and I appreciate your interest in this work.  Best wishes with your research!

Sincerely,
Jerry