Research

Video recording from the online Chamberlain seminar | Twitter thread | Replication package

Key takeaways: 

• • Read the paper if your treatment or instrument is computed by a formula from multiple sources of variations, only some of which (the "shocks") are as-good-as-random

  • We show how to leverage the underlying true or natural experiment, purging OVB from non-random exposure to the shocks, while not imposing further assumptions (like parallel trends) that conventional approaches often require

  • The key requirement is to be explicit about what the natural experiment is: if some shocks exogenously happened, which alternative vectors of shocks could have happened instead, equally likely?

  • We give many suggestions how shock counterfactuals may be specified, and illustrate the ideas in applications.

  • The procedure yields corrected standard errors (via randomization inference)

Stata commands available via ssc install 

• did_imputation (implements the robust and efficient imputation estimator)

• event_plot (makes event study plots for our estimator, alternative robust estimators, and conventional OLS)

An R command is available from Kyle Butts here (only basic functions!).

Twitter thread | Presentation video | Replication package

The paper supersedes the 2017 version with Xavier Jaravel.

Key takeaways: 

• • When using shift-share IVs, one has to take an a priori stand whether shares or shocks are exogenous. 

  • We propose a formal framework for the exogenous shocks view.

  • This approach has practical implications for choosing controls (especially when shares do not add up to one), computing standard errors, running balance tests, etc.

  • Our Stata command recasts the shift-share IV regression as an equivalent IV regression at the level of exogenous shocks, helping to get correct SE, F-stats, and tests.

Twitter thread  |  Slides on shift-share IV methods broadly

Stata command: ssaggregate, available from ssc
R command: ssaggregate (by Kyle F. Butts)

Replication archive: https://github.com/borusyak/shift-share

Key takeaways:

• • Welfare effects of trade shocks are linked to the labor market and consumption exposures of agents

  • On the labor market  side, exposure to trade and therefore the welfare effects of trade shocks vary within, but not across, groups of U.S. workers with similar earnings.

  • Trade can generate winners and losers but not have substantial effects on the shape of the income distribution

This paper supersedes one part of our draft "The Distributional Effects of Trade: Theory and Evidence from the United States", which concerns the labor market effects of trade in general equilibrium.

Slides | Replication package

We review the econometric insights from the recent literature on design-based identification with formula instruments, such as linear and nonlinear shift-share instruments.

• Intuition and checklists for the "many exogenous shifts" and "exogenous shares" approaches to shift-share IV validity

• Answers to many common questions (Appendix A)

• Shift-based standard errors for regressions with multiple shift-share instruments or treatments (Appendix A.9)

Shows the key role of design-based specifications (which control for the expectation of the treatment or instrument) not only to avoid OVB but also to avoid sign reversals due to negative weights

CReAM Discussion Paper 22/20 | cemmap working paper 15/21

VoxEU article

A short video for general audiences: on Youtube

Media coverage: Frankfurter Allgemeine Zeitung (FAZ), Berliner Zeitung, The European, Morgenbladet

New version (now with Josh Feng) coming soon.

Key takeaways:  

• • All income and education groups in the U.S. have very similar % of expenditure on imports, overall or with specific trade partners.

  • Therefore they gain equal purchasing power from small trade shocks , in contrast to a common view that the poor benefit more.

  • Note: The labor market effects of trade shocks are now in a separate manuscript, "Are Trade Wars Class Wars? The Importance of Trade-Induced Horizontal Inequality")

NBER Working Paper 28957: https://www.nber.org/papers/w28957

Media coverage: New York Times (2018), New York Times (2019), Deutsche Welle, Les Echos

Twitter thread | Our older 2018 draft

[Presentation video from the NBER ITI meeting (60 min)]

Regressions of local population changes on exogenous local labor demand shocks may reveal little about the local responsiveness of internal migration to those shocks.

We study regression discontinuity settings where the outcome is defined at a more aggregated level than the RD shocks, e.g. when studying the effects of the fraction of women in the state legislature on state-level outcomes, where some seats are determined by a close election between a woman and a man. Our approach also works to study spillovers from RD events. We propose two new estimators and apply them to estimate the effects of unionization on wage inequality at the state-by-industry level.

(Based on the material previously included in our working paper "Non-Random Exposure to Exogenous Shocks: Theory and Applications")