Helpman–Melitz–Rubinstein Trade Model

The Helpman–Melitz–Rubinstein (HMR) Trade Model is the landmark theoretical and econometric framework in international trade and business and economics theory developed by economists Elhanan Helpman, Marc J. Melitz, and Yona Rubinstein in their 2008 Quarterly Journal of Economics article "Estimating Trade Flows: Trading Partners and Trading Volumes". Unifying the heterogeneous-firm Melitz trade model (2003) with the empirical gravity model of trade (Tinbergen, 1962; Anderson and van Wincoop, 2003) and James Heckman's (1979) sample-selection econometrics, the HMR model resolved two major empirical anomalies that traditional trade models could not explain: first, why more than 50% of all country pairs in global trade data exhibit zero bilateral trade flows (T_ij = 0 in one or both directions); and second, why bilateral trade flows between trading partners are systematically asymmetric (M_ij ≠ M_ji). By replacing the unbounded Pareto productivity distribution of standard Melitz–Chaney models with a bounded truncated Pareto distribution G(a) on unit-input support [a_L, a_H], Helpman, Melitz, and Rubinstein showed that whenever bilateral fixed export costs f_ij and iceberg transport costs τ_ij push the zero-profit export cutoff a_ij below the most productive firm's unit input requirement a_L, zero firms export from country j to country i (W_ij = 0). Furthermore, they proved that conventional log-linear gravity regressions estimated only on positive trade pairs suffer from two distinct econometric biases—Heckman sample-selection bias from dropping zero-trade pairs and an omitted-variable bias from unobserved firm-level extensive-margin heterogeneity W_ij—which induces standard gravity estimates to overstate the direct intensive-margin elasticity of trade with respect to distance by roughly 35% to 50%.

Two Major Anomalies in Bilateral World Trade Data

Since Jan Tinbergen (1962), the empirical gravity model of trade—which posits that bilateral trade volume M_ij imported by country i from exporter j is proportional to the product of their economic sizes Y_i Y_j and inversely proportional to bilateral distance D_ij^−γ—has been one of the most empirically successful equations in economics. Theoretical microfoundations from Anderson (1979), the Krugman new trade model (1980) with Dixit–Stiglitz monopolistic competition, the Eaton–Kortum model (2002), and Anderson and van Wincoop's (2003) multilateral resistance equation all yielded symmetric log-linear gravity equations:

Yet Helpman, Melitz, and Rubinstein (2008) documented that when this symmetric gravity equation is confronted with comprehensive cross-country bilateral trade matrices (158 countries from 1970 to 1997, comprising 24,806 ordered country pairs per year), two empirical facts violate every symmetric representative-firm model:

1. Pervasive Zero Bilateral Trade Flows (T_ij = 0): In any given year, roughly half of all country pairs do not trade with each other at all, or trade in only one direction. Moreover, the growth of world trade since 1970 occurred overwhelmingly among country pairs that were already trading in 1970, while entry of new country-pair trading relationships is governed by fixed regulatory, linguistic, and institutional barriers. Because ln(0) is undefined, standard log-linear OLS gravity estimations discard the 50% of country pairs with M_ij = 0.
2. Directional Asymmetry of Bilateral Trade (M_ij ≠ M_ji): In symmetric gravity models where bilateral trade costs are symmetric (τ_ij = τ_ji), the predicted ratio M_ij / M_ji depends only on fixed exporter and importer country effects. In reality, bilateral trade exhibits pronounced directional imbalances driven by asymmetric fractions of exporting firms.

Why Standard Melitz Models Cannot Generate Zero Bilateral Trade

In the original Melitz trade model (2003) as parameterized by Thomas Chaney (2008) with an unbounded Pareto distribution of firm productivity φ = 1/a ∈ [φ_min, ∞) (equivalent to unit labor requirements a ∈ (0, a_H] reaching down to a_L = 0), every exporting country j possesses a positive density of arbitrarily hyper-productive firms with unit cost a → 0. Consequently, no matter how high the bilateral fixed export cost f_ij or variable iceberg trade cost τ_ij is, a positive mass of firms in country j always finds it profitable to export to country i, making zero bilateral trade impossible (M_ij > 0 for all i, j).

Preferences, Monopolistic Competition, and Firm Unit Input Costs a

Consider a world economy consisting of J countries indexed by i, j ∈ {1, 2, ..., J}. Consumers in importing country i maximize a Constant Elasticity of Substitution (Dixit–Stiglitz) utility function over a continuum of differentiated varieties l ∈ B_i with elasticity of substitution ε = 1 / (1 − α) > 1 (where 0 < α < 1), yielding country i's demand for variety l at price p_i(l) as x_i(l) = [p_i(l) / P_i]^−ε (Y_i / P_i), where P_i is country i's ideal CES price index and Y_i is total expenditure in country i.

In exporting country j, a mass N_j of monopolistically competitive firms each produces a distinct variety. Each firm in country j draws a firm-specific unit input requirement a (so 1/a is firm productivity), meaning that producing 1 unit of output requires a bundles of country j's factor inputs at unit factor cost c_j (yielding marginal production cost c_j a). To sell its variety in destination market i, a firm from country j with unit input coefficient a faces two bilateral trade costs:

1. Variable Iceberg Trade Cost τ_ij ≥ 1 (with τ_jj = 1): Delivering 1 unit of the good to market i requires shipping τ_ij units from country j, so marginal delivered cost in market i is τ_ij c_j a, yielding constant-markup delivered price p_ij(a) = τ_ij c_j a / α.
2. Bilateral Fixed Export Cost c_j f_ij > 0 (with f_jj = 0 for surviving domestic producers): A firm in j must pay a fixed overhead cost c_j f_ij (covering destination-specific regulatory compliance, distribution networks, and customs documentation) to enter market i.

Zero-Profit Export Cutoff a_ij and Truncated Pareto Distribution G(a)

A firm in country j with unit input requirement a earns bilateral net operating profits π_ij(a) = (1 − α)[(τ_ij c_j a) / (α P_i)]^1−ε Y_i − c_j f_ij from exporting to country i. Setting π_ij(a_ij) = 0 defines the bilateral zero-profit export cutoff unit input requirement a_ij:

Only firms in country j with unit input cost a ≤ a_ij find it profitable to export to country i. To explain zero bilateral trade flows, Helpman, Melitz, and Rubinstein (2008) assumed that unit input requirements a across all countries follow a Truncated Pareto Distribution G(a) on the bounded support [ a_L, a_H ] with 0 < a_L < a_H and shape parameter k > ε − 1:

Here a_L > 0 is the unit input requirement of the most productive technological frontier firm (with maximum productivity φ_max = 1/a_L < ∞). Consequently, two regimes arise for every ordered country pair (i, j):

• Zero Bilateral Trade Regime (a_ij ≤ a_L): If bilateral fixed costs f_ij or iceberg costs τ_ij are high enough that a_ij ≤ a_L, then not even the most productive firm a_L in country j can cover the fixed export cost c_j f_ij. Zero firms export from j to i, and aggregate bilateral trade is zero: M_ij = 0.
• Positive Bilateral Trade Regime (a_ij > a_L): All firms in country j with a ∈ [a_L, a_ij] profitably export to country i, generating positive aggregate bilateral trade M_ij > 0.

Closed-Form Expression for Aggregate Bilateral Imports M_ij

Integrating firm-level export revenues over all exporting firms a ∈ [a_L, a_ij] in country j yields the HMR structural gravity equation:

where V_ij ≡ ∫_aL^{max{a_L, a_ij}} a^1−ε dG(a) is proportional to the Extensive-Margin Exporting-Firm Multiplicative Factor W_ij:

Log-Linear Gravity Equation With the Extensive Margin Term ln w_ij

Whenever bilateral trade is positive (T_ij = 1 ⇔ a_ij > a_L ⇔ W_ij > 0), taking natural logarithms of M_ij and adding an i.i.d. shock u_ij (where τ_ij^ε−1 = D_ij^γ exp(−u_ij)) yields the HMR Log-Gravity Equation for Positive Trade Flows:

where m_ij = ln M_ij is log bilateral imports of i from j; λ_j and χ_i are exporter and importer fixed effects; −γ d_ij is the Intensive Margin effect of log bilateral distance d_ij = ln D_ij (how distance reduces the export volume of each individual firm that is already exporting); and w_ij = ln W_ij = ln[(a_ij / a_L)^k−ε+1 − 1] is the Extensive Margin term capturing the endogenous fraction of firms in country j that export to country i. Because the zero-profit cutoff ratio a_ij / a_L is itself a decreasing function of bilateral distance d_ij and fixed export costs f_ij, omitting w_ij from a traditional gravity regression forces the distance coefficient γ̂_OLS to absorb both the intensive margin and the extensive margin!

Stage 1: Probit Equation for Selection Into Trading Partners (z_ij and ρ_ij)

How can an econometrician estimate w_ij = ln[ (a_ij / a_L)^k−ε+1 − 1 ] when firm-level export counts are unobserved in aggregate country-level trade matrices? Helpman, Melitz, and Rubinstein defined the latent export profitability ratio Z_ij ≡ (a_ij / a_L)^ε−1 of the most productive firm a_L. Taking logs yields the linear latent variable z_ij ≡ ln Z_ij:

Positive bilateral trade occurs (T_ij = 1) if and only if Z_ij > 1 ⇔ z_ij > 0. Therefore, in Stage 1, HMR estimate a Probit model for the probability ρ_ij = Pr(T_ij = 1) = Φ(z_ij*) that country j exports to country i. Using the predicted probability ρ̂_ij from Stage 1, HMR construct two structural estimators for every positive-trade observation (T_ij = 1):

1. Predicted Latent Cutoff Index: ẑ_ij* = Φ⁻¹( ρ̂_ij ).
2. Inverse Mills Ratio (Heckman Selection Correction): η̂_ij* = E[ η_ij | T_ij = 1 ] = φ( ẑ_ij* ) / Φ( ẑ_ij* ).

For econometric identification independent of functional form, Stage 1 includes exclusion restrictions φ_ij that shift fixed export entry costs f_ij without directly altering variable shipping costs τ_ij—specifically, bilateral regulatory entry costs (number of days, legal procedures, and startup costs required to register a business in the exporter and importer, or common religion indicators).

Stage 2: Nonlinear / Polynomial Gravity Estimation Controlling for Both Biases

Given z_ij = z_ij* + η_ij, the conditional expectation of the extensive-margin term w_ij = ln{ exp[ δ (z_ij* + η_ij) ] − 1 } (where δ ≡ (k − ε + 1) / (ε − 1) > 0) for positive-trade pairs (T_ij = 1) is a function of the single sufficient statistic ẑ_ij ≡ ẑ_ij* + η̂_ij*: ŵ_ij* = ln{ exp(δ ẑ_ij) − 1 }. Substituting ŵ_ij* and the Heckman selection term β_uη η̂_ij* into the log-gravity equation yields the HMR Stage 2 Nonlinear Gravity Specification (which can also be estimated via OLS using a cubic polynomial in ẑ_ij alongside η̂_ij*):

Quantitative Findings: Extensive Margin vs. Heckman Selection Bias

Applying this two-stage estimator to the 1986 cross-section of 158 countries (24,806 country pairs, of which 11,146 had positive trade), Helpman, Melitz, and Rubinstein (2008) established three empirical findings:

1. Upward Bias in Traditional Gravity Distance Elasticities: In a benchmark OLS gravity regression on positive flows, the estimated distance coefficient is γ̂_OLS = −1.167. Controlling for both selection and the extensive margin in HMR Stage 2 reduces the intensive-margin distance coefficient to γ̂_HMR = −0.750 to −0.813—proving that roughly 35% of the apparent distance effect in standard gravity regressions is actually the extensive margin (fewer firms exporting to distant destinations rather than smaller shipments per firm).
2. Firm Heterogeneity Bias Dominates Heckman Sample-Selection Bias: When HMR included only the Heckman selection term η̂_ij* (omitting the firm-heterogeneity term ŵ_ij*), the distance coefficient barely moved; by contrast, including the extensive-margin firm-heterogeneity correction ŵ_ij* accounted for nearly the entire bias reduction.
3. Counterfactual Trade Volume Predictions: Because the extensive-margin elasticity ∂w_ij/∂z_ij varies nonlinearly with the baseline fraction of exporting firms z_ij, reducing trade barriers generates a larger proportional surge in trade between small, distant country pairs (where z_ij is near zero and many marginal firms cross the export threshold) than between large, proximate trading partners.

Why does an unbounded Pareto productivity distribution fail to explain zero bilateral trade?

Under an unbounded Pareto distribution of firm productivity φ ∈ [φ_min, ∞), there is no upper bound on firm productivity (equivalently, the lower bound on unit input requirements is a_L = 0). Therefore, for any finite bilateral fixed export cost f_ij > 0 and any finite iceberg transport cost τ_ij, there always exists a non-empty tail of hyper-productive firms with unit cost a ∈ (0, a_ij] that earn positive export profits in destination i, predicting positive trade M_ij > 0 for 100% of country pairs. By bounding the technological frontier at a_L > 0 via a truncated Pareto distribution, the HMR model ensures that whenever a_ij ≤ a_L, zero firms export (M_ij = 0).

What is the difference between the Heckman sample-selection bias and the HMR extensive-margin bias in gravity regressions?

Heckman sample-selection bias (η̂_ij*) is an econometric disturbance-term bias arising because unobserved shocks u_ij to trade volume are correlated with unobserved shocks η_ij to the probability of positive trade (E[u_ij | T_ij = 1] ≠ 0). By contrast, the HMR extensive-margin bias (ŵ_ij*) is a structural omitted-variable bias that persists even if all error terms are zero: as bilateral distance d_ij falls, not only does each existing exporter ship more output (intensive margin), but additional lower-productivity firms enter the export market (extensive margin W_ij), causing naive OLS regressions that omit w_ij to conflate firm entry with firm-level price elasticity.

What exclusion variables identify the First Stage Probit in the HMR procedure?

Rigorous identification requires at least one variable φ_ij that affects fixed export entry costs f_ij (Stage 1 selection into trading) without affecting variable per-unit shipping costs τ_ij (Stage 2 intensive-margin volume). Helpman, Melitz, and Rubinstein (2008) used cross-country data on regulatory costs of firm entry (the number of legal procedures and days required to start an enterprise in both the importing and exporting country) and shared religion, verifying that these fixed-cost barriers strongly predict whether two countries trade at all while having no direct effect on per-firm shipment volume once ŵ_ij* is controlled for.

1. Helpman, E., Melitz, M., & Rubinstein, Y. (2008). "Estimating Trade Flows: Trading Partners and Trading Volumes." Quarterly Journal of Economics, 123(2), 441–487. https://doi.org/10.1162/qjec.2008.123.2.441
2. Melitz, M. J. (2003). "The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity." Econometrica, 71(6), 1695–1725. https://doi.org/10.1111/1468-0262.00467
3. Chaney, T. (2008). "Distorted Gravity: The Intensive and Extensive Margins of International Trade." American Economic Review, 98(4), 1707–1721. https://doi.org/10.1257/aer.98.4.1707
4. 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
5. Silva, J. M. C. S., & Tenreyro, S. (2006). "The Log of Gravity." Review of Economics and Statistics, 88(4), 641–658. https://doi.org/10.1162/rest.88.4.641