Working Paper: NBER ID: w22368
Authors: Antonio Ciccone; Elias Papaioannou
Abstract: Empirical cross-industry cross-country models are applied widely in economics, for example to investigate the determinants of economic growth or international trade. Estimation generally relies on US proxies for unobservable technological industry characteristics, for example industries' dependence on external finance or relationship-specific inputs. We examine the properties of the estimator and find that estimates can be biased towards zero (attenuated) or away from zero (amplified), depending on how technological similarity with the US covaries with other country characteristics. We also develop an alternative estimator that yields a lower bound on the true effect in cross-industry cross-country models of comparative advantage.
Keywords: No keywords provided
JEL Codes: F10; G30; O40
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Benchmarking estimator leads to attenuation bias (C51) | Estimated effects biased towards zero (C51) |
| Technological similarity with the U.S. is uniform across countries (O57) | Estimates reflect downward bias (C51) |
| Benchmarking estimator leads to amplification bias (C51) | Estimated effects biased away from zero (C51) |
| Countries grouped based on comparative advantages (F12) | Benchmarking estimator yields a lower bound on true strength of comparative advantage (C51) |
| Technological characteristics vary across countries and industries (O30) | Implications for using benchmark data in empirical research (C80) |