Costinot–Donaldson–Komunjer Ricardian Trade Model

Costinot–Donaldson–Komunjer Ricardian trade model (often cited alongside the Costinot–Rodríguez-Clare multi-sector quantitative gravity synthesis) is the benchmark multi-industry structural model of comparative advantage in business and economics theory, developed by economists Arnaud Costinot, Dave Donaldson, and Ivana Komunjer in their April 2012 Review of Economic Studies article "What Goods Do Countries Trade? A Quantitative Exploration of Ricardo's Ideas". For nearly two centuries after David Ricardo introduced the law of comparative advantage in 1817, empirical international economics faced a disconnect between theory and data: deterministic two-country textbook Ricardian models predicted complete corner specialization that vanished in multi-country, multi-industry data, while empirical researchers relied on Béla Balassa's (1965) ad-hoc Revealed Comparative Advantage (RCA) index without a general-equilibrium microfoundation. By embedding a continuum of stochastic intra-industry product varieties drawn from a Fréchet distribution (as in the aggregate Eaton–Kortum model of 2002) inside an economy with I countries and K disaggregated manufacturing industries, Costinot, Donaldson, and Komunjer (CDK) derive a log-linear structural multi-industry gravity equation that links bilateral sectoral trade flows x_ij^k to fundamental cross-industry productivity differences z_i^k with a single structural Ricardian elasticity θ̂ ≈ 6.53, enabling counterfactual quantification of the welfare gains from comparative advantage via the multi-sector Arkolakis–Costinot–Rodríguez-Clare welfare formula.

In David Ricardo's (1817) numerical example—and in Rudiger Dornbusch, Stanley Fischer, and Paul Samuelson's (1977) continuum-of-goods formulation—a country i has a comparative advantage in industry k1 relative to industry k2 compared to country i′ if the ratio of its labor productivity in k1 to k2 exceeds that of country i′:

Despite the simplicity of this inequality, empirical trade economists from G. D. A. MacDougall (1951, 1952) and Bela Balassa (1963, 1965) to James Harrigan (1997) and Stephen Golub and Chang-Tai Hsieh (2000) encountered three roadblocks when attempting to test Ricardo's theory across many countries (I > 2) and many industries (K > 2):

1. The Multi-Country, Multi-Good Assignment Problem: In a deterministic Ricardian model with I countries, K goods, and bilateral iceberg trade costs d_ij^k ≥ 1, bilateral trade flows are governed by high-dimensional corner solutions (zero trade vs. positive trade). There is no general monotonic linear relationship between bilateral productivity ratios and bilateral export shares once I > 2 and K > 2.
2. Unobservability of Productivity in Non-Produced Goods (Selection Bias): In a deterministic Ricardian world where countries specialize in the goods they export and cease domestic production of the goods they import, an econometrician cannot observe a country's domestic labor productivity z_i^k in the industries where its comparative advantage is weakest. Observing only realized factory productivity in surviving domestic firms confounds fundamental country-industry technology with endogenous selection into production.
3. Lack of Microfoundations for Balassa's (1965) Revealed Comparative Advantage (RCA): To infer comparative advantage from trade data, Balassa (1965) defined the empirical index RCA_i^k = (X_i^k / ∑_k′ X_i^k′) ÷ (∑_i′ X_i′^k / ∑_i′,k′ X_i′^k′), comparing country i's share of exports in industry k to the world average. However, raw export totals X_i^k conflate supply-side productivity z_i^k with proximity to large importing markets, bilateral tariff preferences, and destination demand bias.

Costinot, Donaldson, and Komunjer (2012) solved all three obstacles by introducing intra-industry stochastic variety heterogeneity from Eaton and Kortum (2002) into each of the K industries, transforming deterministic corner assignments into smooth, differentiable interior trade shares across every country pair and industry.

Two-Tier Preferences and Labor Endowments

Consider a world economy consisting of I countries indexed by i, j ∈ {1, …, I} and K disaggregated industries (goods) indexed by k ∈ {1, …, K}. Labor is the primary factor of production, mobile across industries within each country i at competitive national wage rate w_i > 0, and immobile across international borders, with inelastic national labor supply L_i.

Within each industry k ∈ {1, …, K}, there exists a continuum of product varieties indexed by ω ∈ [0, 1]. A representative household in importing country j maximizes a two-tier utility function:

• Upper Tier (Across Industries k = 1, …, K): Cobb–Douglas preferences assigning fixed expenditure share α_j^k ∈ (0, 1) (with ∑_k=1^K α_j^k = 1) of country j's total expenditure X_j to industry k, so total spending in country j on industry k is X_j^k = α_j^k · X_j.
• Lower Tier (Across Varieties ω ∈ [0, 1] Within Industry k): Constant Elasticity of Substitution (CES) aggregation over varieties ω ∈ [0, 1] with elasticity of substitution σ^k > 0:

Within-Industry Fréchet Productivity Draws and Fundamental Productivity z_i^k

In every country i and industry k, constant-returns-to-scale production of variety ω requires 1 / z_i^k(ω) units of labor per unit of physical output. Rather than assuming that all varieties ω within industry k have identical productivity, Costinot, Donaldson, and Komunjer assume that each variety productivity z_i^k(ω) is drawn independently across i, k, and ω from a Fréchet cumulative distribution function:

This two-parameter distribution separates comparative advantage across industries from comparative advantage within industries:

1. Fundamental Country–Industry Productivity (z_i^k > 0): The scale parameter z_i^k governs the central tendency of country i's productivity across all varieties ω in industry k. The geometric mean (and up to a Gamma-function constant Γ(1 − 1/θ), the unconditional arithmetic mean) of variety productivity in country i and industry k is proportional to z_i^k. Thus, Ricardo's concept of comparative advantage is captured by variations in the I × K matrix [z_i^k].
2. Intra-Industry Dispersion / Ricardian Trade Elasticity (θ > max{1, σ^k − 1}): The shape parameter θ > 1 governs the degree of idiosyncratic variety-level productivity dispersion within each industry:
   • When θ is low, varieties ω within industry k differ widely in productivity across countries, generating strong intra-industry comparative advantage and making aggregate industry trade shares less sensitive to changes in z_i^k or trade costs.
   • When θ → ∞, intra-industry heterogeneity vanishes (z_i^k(ω) → z_i^k for all ω), recovering the deterministic multi-industry Ricardian model.

Perfect Competition, Iceberg Trade Costs, and Bilateral Sectoral Trade Shares

International trade from exporting country i to importing country j in industry k incurs iceberg trade costs d_ij^k ≥ 1 (with d_ii^k = 1), meaning that d_ij^k units of variety ω must be shipped from country i for 1 unit to arrive in destination j. Under perfect competition, the unit cost at which producers in country i can deliver variety ω of industry k to buyers in country j is c_ij^k(ω) = (d_ij^k · w_i) / z_i^k(ω), and buyers in country j purchase each variety ω ∈ [0, 1] from the lowest-cost supplier around the world:

By the min-stability property of the Fréchet distribution (the property derived at the aggregate economy level in the Eaton–Kortum model), the probability π_ij^k that exporter i is the lowest-cost supplier of a variety ω in industry k to importer j—which also equals the share of country j's expenditure on industry k devoted to imports from country i—is given by the multinomial logit gravity formula:

Correspondingly, the CES consumer price index p_j^k of industry k in importing country j is:

Solution to the Endogenous Selection Problem: Observed vs. Fundamental Productivity

In the data, country i does not produce all varieties ω ∈ [0, 1] of industry k; rather, domestic producers in country i survive and produce the subset of varieties Ω_ii^k where country i beats foreign import competition in its own domestic market (a fraction π_ii^k of varieties). Using the conditional distribution of Fréchet maxima, Costinot, Donaldson, and Komunjer derive the selection-correction formula relating observed average productivity z̃_i^k across active domestic firms to fundamental productivity z_i^k:

Whenever a country is more open to imports in industry k (so its domestic market share π_ii^k < 1 is smaller), foreign competition weeds out the least productive domestic varieties, pushing observed average productivity z̃_i^k above fundamental productivity z_i^k by the factor (π_ii^k)^−1/θ. Dividing observed productivity by (π_ii^k)^−1/θ (or using industry producer output prices) recovers fundamental technology z_i^k.

The Log-Linear Structural Gravity Equation and Econometric Replacement for Balassa RCA

Suppose bilateral iceberg trade costs decompose multiplicatively into a country-pair component and an importer-industry component: d_ij^k = d_ij · d_j^k · exp(−u_ij^k). Taking the natural logarithm of the bilateral sectoral trade flow x_ij^k = π_ij^k · X_j^k yields the Costinot–Donaldson–Komunjer Structural Gravity Equation:

where:

• δ_ij ≡ −θ · ln(d_ij · w_i) is an exporter × importer country-pair fixed effect, absorbing macroeconomic wage differences w_i, bilateral geographic distance, common language, border adjacency, and aggregate trade agreements;
• δ_j^k ≡ ln X_j^k − θ · ln(p_j^k · d_j^k / γ^k) is an importer × industry fixed effect, absorbing destination-specific demand tastes α_j^k, multilateral price competition p_j^k, and destination industry tariffs;
• θ · ln z_i^k is the Ricardian comparative advantage term.

This equation delivers three major results:

1. Cross-Ratio Test of Ricardo: Taking the double difference (cross-ratio) across two exporters (i, i′) and two industries (k1, k2) shipped to the same destination market j cancels out both δ_ij and δ_j^k:

In a stochastic Fréchet world, Ricardo's two-country, two-good comparative advantage prediction holds log-linearly for every pair of exporters (i, i′) and every pair of industries (k1, k2) inside every destination market j.

1. Econometric Revealed Comparative Advantage (Replacing Balassa, 1965): When direct productivity data z_i^k are unavailable, replacing θ ln z_i^k with an exporter × industry fixed effect δ_i^k in the gravity regression ln x_ij^k = δ_ij + δ_j^k + δ_i^k + ε_ij^k recovers exp(δ_i^k/θ) = z_i^k—providing a theory-consistent, geography-purged measure of Revealed Comparative Advantage that strips out bilateral distance and destination demand size.
2. Empirical Estimate of the Ricardian Trade Elasticity (θ̂ ≈ 6.53): Using 1997 data across 21 OECD countries and 13 disaggregated manufacturing industries from the Groningen Growth and Development Centre (GGDC) productivity database (instrumenting inverse producer prices with R&D expenditure), Costinot, Donaldson, and Komunjer estimated a structural Ricardian elasticity of θ̂ = 6.53.

Having estimated θ̂ = 6.53 and recovered the I × K matrix of revealed fundamental productivities z_i^k, Costinot, Donaldson, and Komunjer (and Costinot and Rodríguez-Clare in their 2014 Handbook of International Economics synthesis) use Dekle–Eaton–Kortum (2007) "hat algebra" (x̂ ≡ x′ / x) and the multi-sector Arkolakis–Costinot–Rodríguez-Clare welfare formula to measure the welfare gains from cross-industry Ricardian comparative advantage:

In the multi-sector ACR formula, the total welfare gains from trade (moving from autarky π_jj^k = 1 to observed trade shares π_jj^k < 1) arise from two sources:

1. Within-Industry Gains from Trade: Driven by variety-level Fréchet heterogeneity within each industry k, which would exist even if all countries had identical relative productivities across industries (z_i^k = z_i · z^k).
2. Across-Industry Ricardian Comparative Advantage Gains: Driven by cross-industry differences in relative fundamental productivity (z_i^k1 / z_i^k2 ≠ z_i′^k1 / z_i′^k2).

Running a general-equilibrium counterfactual in which cross-industry differences in relative fundamental productivity are eliminated (z′_i^k = z_i · z_ref^k) while adjusting aggregate country productivity z_i so that autarky welfare is unchanged, CDK find that across the 21 OECD economies and 13 broad manufacturing sectors, eliminating cross-industry Ricardian comparative advantage inflicts an average real income loss of roughly 0.3% to 1.5% of GDP (about 5% to 20% of the total manufacturing gains from trade), while growing substantially at finer sectoral disaggregation or when incorporating input–output intermediate linkages (as in Costinot and Rodríguez-Clare, 2014).

How does the Costinot–Donaldson–Komunjer (2012) model extend the Eaton–Kortum (2002) model?

The original Eaton–Kortum model (2002) treats the entire manufacturing sector as a single aggregate industry (K = 1) with a single national technology parameter T_i per country, so it cannot study why a country exports aircraft while importing apparel. Costinot, Donaldson, and Komunjer (2012) generalize Eaton–Kortum to K distinct disaggregated industries, assuming that within each industry k, varieties ω ∈ [0, 1] are drawn from a Fréchet distribution centered on an industry- and country-specific fundamental productivity level z_i^k. This preserves the tractable gravity model of trade within every sector while allowing classical Ricardian comparative advantage to operate across the K industries.

Why is observed factory productivity z̃_i^k different from fundamental Ricardian productivity z_i^k?

Because of international competition, domestic firms in country i do not produce all varieties ω ∈ [0, 1] of industry k; they shut down in the varieties where foreign exporters are cheaper, surviving only in the fraction π_ii^k of varieties where country i has its highest idiosyncratic productivity draws. Consequently, observed average productivity across surviving domestic producers, z̃_i^k = Γ(1 − 1/θ) · z_i^k · (π_ii^k)^−1/θ, is biased upward in import-competing industries (where π_ii^k is small). CDK show how to correct for this endogenous selection bias using either domestic producer price indices or the domestic expenditure share π_ii^k.

Why is the CDK structural gravity fixed effect superior to Balassa's (1965) Revealed Comparative Advantage (RCA) index?

Balassa's (1965) traditional RCA index divides a country's raw export share in industry k by the world export share in industry k. As a result, Balassa's index is contaminated by bilateral trade costs (such as a country sitting next door to a neighbor with high demand for industry k) and destination market size. By estimating the structural gravity equation ln x_ij^k = δ_ij + δ_j^k + δ_i^k + ε_ij^k, the exporter-by-industry fixed effect δ_i^k = θ ln z_i^k purges all bilateral geographic barriers (δ_ij) and importer demand effects (δ_j^k), isolating supply-side comparative advantage z_i^k.

What is the empirical value of the Ricardian trade elasticity θ estimated by Costinot, Donaldson, and Komunjer?

Regressing bilateral OECD manufacturing trade flows on fundamental industry productivity (measured via inverse producer price indices and instrumented with industry R&D expenditures to correct for measurement error and reverse causality) alongside exporter–importer and importer–industry fixed effects, Costinot, Donaldson, and Komunjer (2012) estimate a preferred structural Ricardian trade elasticity of θ̂ = 6.53.

1. Costinot, A., Donaldson, D., & Komunjer, I. (2012). "What Goods Do Countries Trade? A Quantitative Exploration of Ricardo's Ideas." Review of Economic Studies, 79(2), 581–608. https://doi.org/10.1093/restud/rdr033
2. Eaton, J., & Kortum, S. (2002). "Technology, Geography, and Trade." Econometrica, 70(5), 1741–1779. https://doi.org/10.1111/1468-0262.00352
3. Costinot, A., & Rodríguez-Clare, A. (2014). "Trade Theory with Numbers: Quantifying the Consequences of Globalization." In G. Gopinath, E. Helpman, & K. Rogoff (Eds.), Handbook of International Economics (Vol. 4, pp. 197–261). Elsevier. https://doi.org/10.1016/B978-0-444-54314-1.00004-5
4. Dornbusch, R., Fischer, S., & Samuelson, P. A. (1977). "Comparative Advantage, Trade, and Payments in a Ricardian Model with a Continuum of Goods." American Economic Review, 67(5), 823–839. https://www.jstor.org/stable/1828066
5. Balassa, B. (1965). "Trade Liberalisation and 'Revealed' Comparative Advantage." The Manchester School, 33(2), 99–123. https://doi.org/10.1111/j.1467-9957.1965.tb00050.x