Skip to content
Code to reproduce aspects of "The Consumption Response to Trade Shocks: Evidence from the US-China Trade War"
Jupyter Notebook
Branch: master
Clone or download
Latest commit 82c38f5 Dec 27, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
data president vote, county data Oct 11, 2019
docs added auto change Dec 27, 2019
.gitignore
LICENSE create license Oct 4, 2019
README.md update readme Dec 17, 2019
alt_hs_naics_mapping.ipynb a fix to concordance issue, does not matter Oct 4, 2019
bls_quarterly_county.ipynb
bls_single_file.ipynb dirver updates Dec 17, 2019
countylevel_tariffs_and_exports.ipynb should all run Dec 17, 2019
employment_analysis.ipynb
main_driver_file.ipynb test, it all ran Dec 17, 2019
political_aspects_of_retaliation.ipynb added correlation with trump vote notebook Oct 11, 2019
pretrend_notebook.ipynb dirver updates Dec 17, 2019
simple_county_by_quantile.png major updates Dec 5, 2019
tariff_map.ipynb added auto change Dec 27, 2019
trade_analysis.ipynb added main driver Dec 16, 2019
updated_tariff_data.ipynb

README.md

"The Consumption Response to Trade Shocks: Evidence from the US-China Trade War"

This repository contains code to reproduce aspects of the paper "The Consumption Response to Trade Shocks: Evidence from the US-China Trade War." Only results associated with non-proprietary data are available, i.e. Chinese retaliatory tariffs and it's projection to the county level and then their correlation with trade and employment. The auto analysis notebook is not included here as the data used is proprietary.


Update (12/2019) The paper and aspects of the code have been revised relative to the September 2019 NBER working paper version. Below are some key updates:

  • An alternative concordance is used. To reproduce the NBER version results, read the notes around code cell 15 in countylevel_tariffs_and_exports.ipynb and uncomment the appropriate parts.

  • I now use the BLS single files rather than the county high-level files. I will keep the bls_quarterly_county.ipynb file in the repository, but now to generate the files you must use bls_single_file.ipynb. The BLS data now goes to June 2019. In this notebook, I also layer in population, income, rural share data from the Census.

  • Pretrends. The notebook pretrend_notebook.ipynb now provides an exploration of pretrend issues in employment data. This notebook essentially mimics what is done for autos.

  • In the paper and the notebooks, population is now used as the main weighting variable rather than employment. To revert to the previous results, simply change the weighting variable to total_employment.

  • A driver file that runs everything and reports results discussed in the paper is posted as main_driver_file.ipynb. Also you can use this file to inspect the output, etc.


There are several files associated with this repository. Almost all of the notebooks directly pull data from the original source (trade data using the Census International Trade API, concordance from the US Census, employment data from the BLS, shapefiles from the US Census). After running the files, most of the output is saved as a .parquet files and stored in the data folder.

Below is a description of each jupyter notebook in the repository. If you don't know what a "jupyter notebook" is or what to do with them, this website is a great place to start: https://datascience.quantecon.org/.

The files below are the individual components and are organized in the sequence for which they must be run if you do not have the intermediate files.

  • updated_tariff_data.ipynb which uses and shapes the tariff data from Bown, Jung, and Zhang. I'm hosting their datafile as I had problems with a direct link.

  • alt_hs_naics_mapping.ipynb starts from the Census concordance which provides a mapping from hs10 to naics codes and then creates a mapping from hs6 to naics. In the data folder there is alt_concordance.parquet which you can download and directly run to proceed to the next step below.

  • countylevel_tariffs_and_exports.ipynb takes the tariff data above and then projects down to the county level, in addition to merging it with US export data and US employment data (just for the year 2017). The projection method simply takes a employment weighted averages of tariffs at the NAICS 3 digit level. Waugh (2019) explains more in detail, in addition to the text in the notebook.

  • tariff_map.ipynb using the tariff data above, it creates a map of tariff exposure. It downloads the required shapefiles and then plots the change in a county's tariff at the county level. It assumes that you have a folder below the repository called shapefiles and then county and state for which the code will download and unzip the shapefiles into that folder.

  • bls_single_file.ipynb grabs the Quarterly Census of Employment and Wages quarterly, single files from the BLS and then creates employment measures at the county-level, monthly frequency. It is then merged with the trade data from above. Note the single files are very large and take time to download and unzip. This notebook constructs several measures of employment, total, goods producing, retail employment, non-goods producing. The advantage of the single files is that many different measures of employment can be constructed.

  • employment_analysis.ipynb performs the employment analysis in Section 5 and has the unweighted regression results as well. The analysis also mimics most aspects of the auto analysis in the paper (e.g., visualizations, tabular analysis, regression results).

  • pretrend_notebook.ipynb performs the pretrend analysis, Section 6, and generates figures 5 in paper. Here you must specify the employment measure emp_all is total employment, emp_gds is goods employment, emp_rtl is retail employment.

  • trade_analysis.ipynb performs the trade analysis in Section 5 (us exports to china, exports in total) and it contains the unweighted regression results as well. This notebook trade_elasticity.ipynb estimates a trade elasticity from the product-level trade data...it's 4!

Finally, I must highlight that there is a license regarding this work, it's use, and so forth.

Old files
  • bls_quarterly_county.ipynb grabs the Quarterly Census of Employment and Wages files from the BLS and then creates employment measures at the county-level, monthly frequency. It is then merged with the trade data from above. Note there is a place where, depending on if commented, creates a dataset with goods producing employment or total private employment.
Other Stuff
  • political_aspects_of_retaliation.ipynb Explores the issue about where the tariffs were directed and how that correlates with the propensity of a county to vote for President Trump. Short answer is yes, but most the variation is about places that export the most to china. The interesting observation is that there is a strong correlation between those that exported a lot to china and the propensity to vote for President Trump.
You can’t perform that action at this time.