Working Paper: CEPR ID: DP13459
Authors: Stephen J. Redding; David E. Weinstein
Abstract: One of the most successful empirical relationships in international trade is the gravity equation, which relates bilateral trade flows between an origin and destination to bilateral trade frictions, origin characteristics, and destination characteristics. A key decision for researchers in estimating this relationship is the level of aggregation, since the gravity equation is log linear, whereas aggregation involves summing the level rather than the log level of trade flows. In this paper, we derive an exact Jensen's inequality correction term for the gravity equation in a nested constant elasticity of substitution (CES) import demand system, such that a log-linear gravity equation holds exactly for each nest of this demand system. We use this result to decompose the effect of distance on bilateral trade in the aggregate gravity equation into the contribution of a number of different terms from sectoral gravity equations. We show that changes in sectoral composition make a quantitatively relevant contribution towards the aggregate effect of distance, particularly for more disaggregated definitions of sectors.
Keywords: gravity equation; trade; aggregation
JEL Codes: F10; F14; C43
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| gravity equation (D50) | accurate modeling at aggregate and sectoral levels (E10) |
| log-linear form of gravity equation (C29) | holds at each level of aggregation (E10) |
| distance (R12) | contributions to bilateral trade (F10) |
| origin fixed effects (C23) | contributions to bilateral trade (F10) |
| destination fixed effects (C23) | contributions to bilateral trade (F10) |
| sectoral variations (R11) | impact on aggregate effect of distance (R11) |
| sector definitions (R38) | contributions to aggregate effect of distance (R12) |
| estimated coefficients for aggregate trade elasticity with respect to distance (F12) | statistically significant and robust (C12) |