Gravity Model of Trade

Gravity model of trade is the foundational empirical and theoretical framework in international economics within business and economics theory for explaining the volume and direction of bilateral trade flows across countries and regions. Introduced by Dutch economist and first Nobel laureate in Economics Jan Tinbergen in his 1962 treatise Shaping the World Economy, the model draws a formal structural analogy to Isaac Newton's law of universal gravitation: just as gravitational attraction between two celestial bodies is proportional to the product of their masses and inversely proportional to the squared distance separating them, bilateral merchandise or service trade between origin economy i and destination economy j scales proportionally with their economic sizes (measured by Gross Domestic Product or aggregate sectoral expenditure) and decays with geographic distance and policy frictions. Although initially viewed as an orphaned empirical regularity lacking neoclassical microfoundations alongside classical comparative advantage and the Stolper–Samuelson theorem, subsequent breakthroughs demonstrated that structural gravity equations arise organically from virtually every major class of general equilibrium theory trade models.

The Naive Multiplicative Gravity Equation

In Tinbergen's (1962) original econometric specification, the nominal value of exports X_ij shipped from exporter i to importer j is written in multiplicative power-law form:

where Y_i denotes total nominal output (GDP) of exporting country i, Y_j denotes total nominal expenditure (GDP) of importing country j, D_ij measures the bilateral great-circle distance (or composite trade friction) between the economic centers of i and j, G is a global proportionality constant, and α, β, and θ > 0 are constant elasticities. Across more than half a century of empirical trade studies—synthesized in meta-analyses by Anne-Célia Disdier and Keith Head (2008) and Keith Head and Thierry Mayer (2014)—the estimated output elasticities α and β cluster tightly near unity (0.8 to 1.1), while the distance elasticity θ averages approximately 0.9 to 1.1, implying that doubling bilateral distance reduces bilateral trade by roughly half.

Taking natural logarithms of both sides and appending observable bilateral trade cost shifters alongside a multiplicative stochastic error term η_ij yields the traditional log-linearized regression equation estimated historically by ordinary least squares (OLS):

Here Contig_ij, Lang_ij, and RTA_ij are binary indicator variables equal to 1 when trading partners i and j share a contiguous land border, an official common language, or membership in a regional trade agreement (such as the European Union or USMCA), respectively.

The Anderson and van Wincoop Expenditure System

While Tinbergen's intuitive specification achieved exceptional statistical fit (R² values routinely between 0.70 and 0.85), it suffered from an omitted-variable bias that James E. Anderson and Eric van Wincoop (2003) famously termed the "gold-medal mistake" of empirical trade economics. Naive gravity models assume that trade between i and j depends only on the absolute bilateral barrier t_ij between those two countries, ignoring the relative trade barriers that each partner faces with the rest of the world.

Building on Anderson's (1979) Armington differentiation framework—in which goods are differentiated by country of origin and consumers maximize constant-elasticity-of-substitution (CES) utility with elasticity of substitution σ > 1—Anderson and van Wincoop (2003) derived the microfounded structural gravity system. Let p_i denote exporter i's factory-gate price and let t_ij ≥ 1 denote the iceberg bilateral trade cost factor (so that t_ij units must be shipped from i for 1 unit to arrive in j, yielding destination price p_ij = p_i t_ij). Imposing global market clearing (Y_i = ∑_j X_ij) where Y_W ≡ ∑_i Y_i is world GDP yields:

where E_j is total expenditure in importer j, and the price indices Π_i and P_j are solved simultaneously by the Multilateral Resistance Terms (MRTs):

The inward multilateral resistance P_j captures importer j's ease of market access to all global suppliers, whereas the outward multilateral resistance Π_i measures exporter i's aggregate incidence of trade costs across all global destination markets. Two geographically remote economies such as Australia and New Zealand exhibit high multilateral resistance (Π_i and P_j are large), so for any given bilateral distance t_ij, the relative trade cost ratio t_ij / (Π_i P_j) is small—explaining why Australia and New Zealand trade far more intensively with one another than two comparably sized European economies separated by the same physical distance. Anderson and van Wincoop demonstrated that accounting for multilateral resistance resolved John McCallum's (1995) famous border puzzle, reducing the estimated tariff-equivalent reduction of the U.S.–Canada national border on inter-provincial vs. state–province trade from a factor of 22 down to a modest, theoretically plausible friction.

Equivalence Across Theoretical Microfoundations

A landmark insight of modern international trade theory, unified by Costas Arkolakis, Arnaud Costinot, and Andrés Rodríguez-Clare (2012), is that isomorphic structural gravity equations emerge from radically different supply-side microfoundations:

In Jonathan Eaton and Samuel Kortum's (2002) Ricardian model, countries draw continuum-good productivities from a Fréchet distribution F_i(z) = exp(−T_i z^−θ), where T_i governs absolute advantage and θ governs comparative-advantage dispersion; lowest-cost sourcing across countries yields a structural gravity equation in which the trade elasticity is the Fréchet shape parameter θ rather than a taste parameter. In Thomas Chaney's (2008) synthesis of Marc Melitz's (2003) heterogeneous-firm model under Pareto productivity draws G(φ) = 1 − φ^−γ, lower bilateral variable trade costs t_ij not only expand exports of incumbent exporters (the intensive margin) but also lower the productivity cutoff φ*_ij required to cover bilateral fixed export costs f_ij, inducing new firms to enter the export market (the extensive margin) so that the aggregate trade elasticity equals the Pareto tail index γ.

In applied econometrics, estimating the log-linearized gravity equation by ordinary least squares introduces two major statistical biases identified by J. M. C. Santos Silva and Silvana Tenreyro (2006) in their classic paper "The Log of Gravity":

1. Jensen's Inequality and Heteroskedastic Errors: Because the expected value of the logarithm of a random variable does not equal the logarithm of its expected value (E[ln η_ij] ≠ ln E[η_ij]), whenever the variance of the multiplicative error term η_ij depends on the regressors (Y_i, Y_j, D_ij)—as is universally true in skewed trade data—the conditional expectation of ln η_ij is correlated with the covariates, rendering OLS point estimates of trade elasticities inconsistent.
2. Sample Selection from Zero Bilateral Trade Flows: At the bilateral country-pair or disaggregated Harmonized System (HS) product level, roughly 30% to 50% of potential bilateral trade flows are exact zeros (X_ij = 0) because no firm in i is productive enough to overcome the fixed export threshold f_ij. Because ln(0) is undefined, log-linear OLS either drops all zero-trade pairs (inducing severe sample-selection bias) or adds an arbitrary constant such as ln(X_ij + 1), which yields unit-dependent, inconsistent parameter estimates.

Santos Silva and Tenreyro (2006) resolved both pathologies by estimating the structural gravity equation directly in multiplicative levels via Poisson Pseudo-Maximum Likelihood (PPML) with exporter-time and importer-time fixed effects:

where μ_i and χ_j are exporter and importer fixed effects that absorb the structural Multilateral Resistance Terms Π_i and P_j exactly, and Z_ij is the vector of bilateral trade cost variables. Crucially, as Johannes Fally (2015) proved, estimating structural gravity via PPML with exporter and importer fixed effects automatically satisfies the general-equilibrium adding-up constraints (∑_j X̂_ij = Y_i and ∑_i X̂_ij = E_j), making three-way fixed-effects PPML (incorporating exporter-time, importer-time, and country-pair fixed effects alongside intra-national trade flows X_ii) the modern gold standard for ex-post policy evaluation of free trade agreements and WTO accession.

Why is distance still so powerful in the digital age ("the distance puzzle")?

Despite steep declines in ocean container shipping rates, air freight costs, and telecommunications prices since 1950, empirical estimates of the distance coefficient θ ≈ 0.9 have remained remarkably stable or even risen slightly—a phenomenon known as the missing globalization puzzle or distance puzzle. Economists explain this persistence through compositional shifts toward time-sensitive intermediate inputs in just-in-town supply chains, informational and contracting frictions that decay with physical and time-zone separation, cultural familiarity gradients, and the inclusion of domestic trade flows (X_ii), which reveal that border and distance barriers have declined relative to internal trade even when cross-border distance elasticities appear constant.

How does the Arkolakis–Costinot–Rodríguez-Clare (ACR) formula use gravity estimates to measure welfare gains from trade?

Arkolakis, Costinot, and Rodríguez-Clare (2012) proved that across Armington, Krugman, Eaton–Kortum, and Melitz–Pareto gravity models, the percentage change in real per-capita welfare (Ŵ_j) from any foreign trade shock depends on just two sufficient statistics: the observed share of expenditure devoted to domestically produced goods (λ_jj = X_jj / E_j) and the structural gravity trade elasticity ε > 0, via the compact welfare formula Ŵ_j = λ̂_jj^−1/ε.

1. Tinbergen, J. (1962). Shaping the World Economy: Suggestions for an International Economic Policy. New York: The Twentieth Century Fund. https://doi.org/10.1002/tie.5060050113
2. Anderson, J. E., & van Wincoop, E. (2003). "Gravity with Gravitas: A Solution to the Border Puzzle." American Economic Review, 93(1), 170–192. https://doi.org/10.1257/000282803321455214
3. Eaton, J., & Kortum, S. (2002). "Technology, Geography, and Trade." Econometrica, 70(5), 1741–1779. https://doi.org/10.1111/1468-0262.00352
4. Santos Silva, J. M. C., & Tenreyro, S. (2006). "The Log of Gravity." The Review of Economics and Statistics, 88(4), 641–658. https://doi.org/10.1162/rest.88.4.641
5. Head, K., & Mayer, T. (2014). "Gravity Equations: Workhorse,Toolkit, and Cookbook." In Handbook of International Economics (Vol. 4, pp. 131–195). Elsevier. https://doi.org/10.1016/B978-0-444-54314-1.00003-3