Readme for industry crosswalks
Brendan Price
July 29, 2014

This folder contains numerous industry crosswalks, needed to convert CBP and NBER-CES data into a consistent set of 4-digit sic87dd and 3-digit sic3 industry codes.

(1) The folder "naics" contains crosswalks mapping from 2007 to 2002 NAICS codes (naics07_naics02.dta) and from 2002 to 1997 NAICS codes (naics02_naics97.dta). The 2007-2002 crosswalk is adapted from American FactFinder table EC0700CBDG1. The 2002-1997 crosswalk is adapted from http://www.census.gov/econ/census02/data/bridge.

(2) The crosswalk "naics97_sic87.dta" maps 1997 NAICS codes into 1987 SIC codes. It is adapted from David Dorn's crosswalk "cw_n97_s87.dta" (available at http://www.cemfi.es/~dorn/data/cw_n97_s87.zip). Many of the SIC codes used in Dorn's crosswalk do not appear in the CBP data. For instance, Dorn has separate codes 722, 723, and 724, whereas CBP uses a single code 720. I reassign codes accordingly.

(3) The crosswalk script "sic87_sic87dd.do" maps 1987 SIC codes into our preferred set of consistent industry codes, which we call sic87dd. The sic87dd scheme used in this paper is closely akin to, but slightly modified from, the schemes of the same name used by the 2013 Autor, Dorn, and Hanson "China Syndrome" paper and the 2014 Acemoglu, Autor, Dorn, Hanson, and Price "Solow Paradox" paper. The script is modified from David Dorn's script subfile_sic87dd.do (available at http://www.cemfi.es/~dorn/data/subfile_sic87dd.zip). The file "sic87dd_key.dta" is closely adapted from www.census.gov/econ/cbp/download/sic88_97.txt, with industry descriptions for non-manufacturing industries heavily modified to reflect the aggregations we performed.

(4) The crosswalk "sic87dd_sic3.dta" maps sic87dd codes into 3-digit sic3 codes. For manufacturing industries, sic3 is just the first three digits of the sic87dd code (augmented with 0). For non-manufacturing industries, we simply set sic3 = sic87dd (since our non-manufacturing sic87dd codes already reflect substantial aggregation of non-manufacturing industries). The industry descriptions stored in "sic3_key.dta" are based on CBP industry names listed at www.census.gov/econ/cbp/download/sic88_97.txt. We create new industry names for many of the non-manufacturing industries, in keeping with the aggregations we performed.

(5) The crosswalk "sic87dd_to_sic87dd_pooled.dta" maps non-manufacturing SIC codes (here labeled sic87dd) into the 87 aggregated codes we use in our analysis (here labeled sic87dd_pooled). Note that in other files the aggregated non-manufacturing codes are simply referred to as sic87dd codes.

(6) The crosswalk "sic2io.dta" maps 1987 SIC codes into the industry codes used in the 1992 BEA benchmark input-output tables. This crosswalk is derived primarily from http://www.bea.gov/scb/account_articles/national/1197io/appxA.htm.

(7) The folder "nesting" contains .dta files that characterize the hierarchy of SIC and NAICS codes used in the County Business Patterns. These hierarchies facilitate cleaning of the CBP files. An especially subtle issue is the treatment of auxiliary industries (present during SIC-coded years), which really live under the 1-digit level; I create fictitious 2-digit and 3-digit codes to act as "parents" to these auxiliary industries. In the nesting files, the variable "agg_level" indicates the level of aggregation of a given industry code.