Thread by @AndrewG_PhD

Lots on the effects of tariffs these days, but many public datasets or naiive approaches to measurement may set you up for endogeneity problems you probably haven't thought about: mismeasurement leads to biases and can even flip the sign of estimated coefficients. 🧵👇

https://twitter.com/nberpubs/status/1986539319444521257

Nearly all public datasets (eg
WITS) report tariffs as: Duties/Imports ( ie Ad Valorem Equivalent tariffs, AVE)
We usually think 👇AVE = liberalization,
but AVE depends not just on law, but also choices about aggregation and prices which are endogenous to most economic outcomes

Consider 1972-1988. AVE falls 1972-1979 when tariff law is largely fixed, and it goes UP when the GATT is implementing tariff cuts (1980-1988).
Is 1972-1979 welfare enhancing while, 1980-1988 is welfare diminishing?

Actually, it's exactly the opposite.
What's going on?

Decomposing AVE into components:

1.) Aggregation/composition issues would lead one to call the grey line the tariff level, but this includes a lot of endogenous variation which erroneously suggest the GATT increased tariffs -nope! Researchers only want PART of the blue line.

2.) Even statute-level AVE can include specific tariffs (per unit, f) and are thus a function of equilibrium prices (inflation & ex-rates). AVE can move WITH imports (opposite normal intuition) -> if not careful you can estimate biased, even POSITIVE import demand elasticites.

The rest of the paper shows:
1.) how to think causally in these settings
2.) how to do counterfactuals correctly in these settings
3.) these issues are still relevant in US Tariff data even up to the onset of the first US China trade war.