Contribution
It replaces bilateral trade, the usual measure of economic dependence in the trade-and-conflict literature, with a general-equilibrium elasticity: how much country n’s real income moves when country i’s productivity rises. It computes this for 143 countries over 1970–2012, shows that about 30 percent of pairs are “enemies” (negative elasticity), and then asks whether the elasticity predicts political alignment. With two instruments it finds that it does, and that the coefficient on bilateral trade, controlled alongside it, is small and changes sign from one specification to the next while the elasticity’s does not.
Key insight
Dependence is not trade. A country can depend on a partner it barely trades with, through common markets and the cost of living, and two exporters of the same raw material can be harmed by each other’s growth. Measured this way, a country whose income comes to rise with a partner’s productivity comes to vote with it, rivals it less and allies with it more.
Assumptions
- The political model (1)–(3). Each country’s representative agent chooses costly political actions ξ_ni toward each partner i that raise i’s productivity, z_i = z̄_i f_i(ξ_Ni). In the Nash equilibrium the first-order condition is U_ni · (∂ln z_i/∂ln ξ_ni) = (∂ln v_n/∂ln ξ_ni) (v_n/u_n): the larger n’s real-income exposure U_ni, the more n does for i. The political part is non-parametric; the paper tests only this monotone prediction.
- The trade model. Constant trade elasticity θ (Armington in the exposition; Eaton–Kortum, Krugman, multi-sector and Caliendo–Parro input-output versions in the appendix). The empirical work uses the multi-sector input-output version with one median input-output matrix for all countries.
- The exposure matrix. U ≡ −(θ/(θ+1))(I − S)(I − V)⁻¹M + S, from linearizing market clearing and indirect utility, with S the expenditure shares, T the income shares, M = TS − I and world income the numeraire (12). Arc elasticities from exact-hat algebra give the same results, the paper says, for shocks up to the sample’s cumulative productivity change.
- Identification, China. China’s growth after 1978 is taken as supply-driven; the instrument is the counterfactual change in exposure from a 100 percent rise in Chinese productivity, computed from the 1980 equilibrium.
- Identification, air travel (Feyrer 2019). Year-varying elasticities of sectoral expenditure shares to air and sea distance, with pair, importer-year and exporter-year fixed effects, predict shares, from which predicted exposure is computed. The exclusion restriction is that the changing relative cost of air and sea routes affects political alignment only through economic exposure.
Key results
- Descriptive. Mean exposure is small and rises over time; about 30 percent of pairs are enemies, typically raw-material exporters competing for markets (Chile and South Africa; Saudi Arabia and Niger). China moves from outside the top five authority scores in 1980 to first in 2010. PTA formation is followed by a rise in exposure, with no pre-trend (two-way fixed effects and de Chaisemartin–D’Haultfœuille).
- China, long differences 1980–2010 (Tables 2–3, 119 countries). OLS coefficient on Δexposure 44.12 (SE 15.10) for the κ-score of UN voting similarity, “a one standard deviation in real income exposure … leads to a 0.26 standard deviation increase in political alignment”; similar for the S- and π-scores. IV 101.3 (27.67), first-stage Kleibergen–Paap F 21.82. With Δlog trade added (instrumented by 1980 trade), exposure stays at 81.23 (22.68) and trade is negative (−0.0173, n.s.), F 10.66.
- Air travel, pair panel (Tables 4–5). OLS 24.20 (3.824) on the κ-score, 653,214 pair-years. IV 60.66 (15.55), F 33.69, 480,452 pair-years; “a one standard deviation increase in real income exposure leads to a 0.094 standard deviation increase in bilateral political alignment”. With log trade instrumented as well, 77.09 (18.23), and trade enters positive (0.0255).
- Rivalry and ideal points (Table 6). IV coefficients on any, positional and spatial rivalry −21.41, −10.35, −15.55, all significant; ideological −3.819 (2.701), not significant, as the text later concedes. Ideal-point distance −88.06 (26.88).
- Alliances (Table 7). IV coefficients positive and significant for any alliance (17.79), defence pacts (15.02), non-aggression (21.90) and ententes (15.94); neutrality 10.75 (5.368), with an Anderson–Rubin p of 0.1.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Countries that became more exposed to Chinese productivity growth moved toward China in UN voting, 1980–2010 | moderate | one cross-section of 119; IV F 21.82, 10.66 with trade |
| C2 | Within country pairs, exposure predicted by air-versus-sea distance raises UN voting similarity, lowers rivalry and raises alliances | moderate | Tables 5–7; the exclusion restriction is argued, not tested |
| C3 | The exposure measure is not proxied by bilateral trade | moderate | trade’s sign and significance change across tables, exposure’s do not |
| C4 | Economic dependence causes political alignment | moderate | two instruments agree in sign; the effect in the panel is 0.094 SD |
| C5 | The mechanism is strategic: costly political actions to raise a partner one depends on | weak | the model motivates the regression; no test separates it from other mechanisms |
Concepts
- Real income exposure U_ni: d ln u_n / d ln z_i, the elasticity of n’s real income per capita to i’s productivity, all general-equilibrium effects included.
- Economic friend / enemy: U_ni > 0 / U_ni < 0, after the “friends and enemies” of neoclassical trade theory. U need not be symmetric.
- Authority and hub scores (Kleinberg 1999): the leading eigenvectors of U′U and UU′; an authority’s growth moves others, a hub is moved by others’.
- Market size, cross-substitution and cost-of-living effects: the three channels through which a partner’s productivity reaches n’s real income in (10)–(11).
Connections
It enters the trade-and-conflict literature (Polachek 1980; Mansfield 1995; Barbieri 2002, the book after LIT-1422’s article; Martin, Mayer and Thoenig 2008) by changing the outcome from war to softer alignment, which it argues is where most political influence lies (Nye 1990). Its second instrument is Feyrer’s air-travel design; by its abstract, the later NBER paper of Feng and others on militarized conflict (LIT-1415) uses air-against-maritime transport technology in the same way. Its measures of alignment come from Signorino and Ritter, Bailey, Strezhnev and Voeten, Thompson and Colaresi, and the Correlates of War alliances.
Bearing on the record
- CASE-150. The flag on the case can be lifted: the outcomes include UN voting (S-, π- and κ-scores and ideal-point distance) and strategic rivalries (any, positional, spatial, ideological), and also formal alliances, which the case does not mention. The case’s description is otherwise right. Three things the reading adds. The effect is small in the panel, 0.094 SD per SD of exposure. Ideological rivalry, the rivalry nearest to a change of view, is the one outcome the instrument does not move. And the case’s “what it does not show” is sharper than it knew: the paper’s own model is the strategic explanation the case sets against changed interpretive organization, since alignment there is an action chosen for its payoff, with preferences held fixed.
- CLAIM-428. The reading supports the claim’s first half, that material interdependence couples outcomes, with a measure that can be negative: 30 percent of pairs are enemies. That a pair’s dependence is organized as opportunity or as threat is, in the paper, decided by the sign of U, not by perspective; the paper offers nothing on the perspective that organizes the coupling.
- CLAIM-404. Not more than the case already says: alignment of states through what their incomes depend on is influence through payoffs, not through evidence or criteria.
No instruction for machine-learning practice; nothing for the anthology.
Limitations
- Alignment is not attitude. UN votes, rivalry codings and alliances are actions and classifications of states. Nothing in the paper speaks to opinion within a country.
- The China cross-section is small (119 observations), the R² of its OLS regressions under 0.1, and its IV with trade has a first-stage F of 10.66.
- The air-travel exclusion restriction is argued from the fixed effects and not tested. Pairs whose air routes shortened relative to sea may have changed in other ways that matter politically (travel, migration, communication).
- Errors and slips. Footnote 4 says that by Perron–Frobenius U′U and UU′ “each have a unique eigenvector with all positive entries”; the theorem applies to positive or irreducible non-negative matrices, and U has negative entries for the 30 percent of enemy pairs, so the claim needs an argument the paper does not give. Table 2’s column (1) has one regressor and R² 0.0484, so the standardized effect of exposure should be about √0.0484 = 0.22 SD, not the 0.26 stated, unless the regression carries unreported controls. The text first says that for rivalries “In all cases, we find a negative and statistically significant relationship” and a page later excepts ideological rivalry. “Inverse hyperbolic sign” (fn. 17) is sine; Table 5’s third F is printed 33.629 beside 33.69; “Kleinbergen-Paap” in Table 3’s note.
Open questions
- Does exposure move public opinion toward the partner, or only the state’s votes and alignments?
- Can the strategic mechanism be told from a change in how a country’s public or elite regards the partner, for instance by whether alignment persists after exposure falls?
- Why does ideological rivalry not respond to the instrument?