Do regional trade agreements really boost trade?

In 1995 the GATT became the WTO and the floodgates opened. Where there had been a few dozen preferential agreements there are now over 350. Baier & Bergstrand (2007, JIE) showed that once the endogeneity of RTA membership is addressed with country-pair fixed effects and lagged RTA indicators, the average agreement approximately doubles bilateral trade between members within a decade. The figures below do not replicate that identification strategy, we report pooled OLS partial correlations and a descriptive deep-vs-shallow premium on the CEPII Gravity data, so the estimates here read as conditional associations, not Baier-Bergstrand-style causal effects. Four questions across the data: how much of world trade moves under an RTA, what the gravity coefficient on an RTA dummy looks like year by year, which blocs matter most, and whether deep agreements (EU, USMCA, CPTPP) deliver larger conditional trade premia than shallow ones.

RTA coverage of world trade

The first question is volume: how much of merchandise trade actually crosses a border that is covered by an active preferential agreement? We take every country-pair flow in CEPII's Gravity V202411 file (BACI-merged bilateral tradeflow 1996-2020), flag whether CEPII's fta_wto dummy is 1 for that pair and year, and sum by year. The fta_wtovariable is a binary notification dummy ('in force' = 1) and does not distinguish between PTAs that have been signed-but-not-ratified and those that are fully implemented in the given year; CEPII Gravity V202501 extends coverage through 2022 and the qualitative pattern below is unchanged.

RTA coefficient from a gravity regression

Coverage is necessary but not sufficient: two rich neighbours may trade heavily whether or not they have a treaty. The gravity equation gives us a conditional premium. For each year we estimate ln Xᵢⱼ = α + β₁ ln Dᵢⱼ + β₂ ln(YᵢYⱼ) + β₃ RTAᵢⱼ + εᵢⱼ by OLS (closed-form 3×3 normal equations) on all pairs with positive trade, non-zero distance and GDP on both sides, and a non-missing fta_wto. We are after β₃, the log-point uplift on trade from having an agreement, holding distance and economic mass fixed. Baier & Bergstrand (2007) found this coefficient is biased down in naïve OLS; here we report the OLS slope without country fixed effects, so the reader can see the raw conditional correlation.

Which blocs actually trade inside themselves

The RTA dummy treats every agreement alike. In reality blocs differ by an order of magnitude in how much of their member exports stay inside the bloc. For each of fifteen hand-curated blocs (EU-27 post-Brexit, USMCA, ASEAN, CPTPP, RCEP, Mercosur, AfCFTA, GCC, SAFTA, EAC, EFTA, CETA, EU-Singapore, Australia-NZ CER, COMESA) we compute intra-bloc exports / total exports of member countries from BACI 2020.

Deep versus shallow

Mattoo, Rocha & Ruta (2020) score modern RTAs along a depth index running from pure tariff cuts to behind-the-border commitments on services, investment, competition, IPR, labour and environment. We take their binary shorthand: deep for EU-27, USMCA, CPTPP, EFTA, CETA, EU-Singapore, Australia-NZ CER; shallow for ASEAN, RCEP, Mercosur, AfCFTA, GCC, SAFTA, EAC, COMESA. For each CEPII pair in 2020 we flag the pair as living inside a deep bloc, a shallow bloc, or neither, then compare mean ln(trade) after netting out mean ln(GDPₒ·GDP_d) using a pooled slope. The exponentiated gap is the trade-volume premium relative to non-RTA pairs.

Are RTAs bilateral or plurilateral?

CEPII's pair-level dummy hides the architecture. Some pairs are covered by a one-off bilateral deal (Japan-Mongolia, US-Chile before CPTPP); others sit inside dense networks (any EU-EU pair shares 25 common RTA partners). For each pair with fta_wto = 1 we count the number of third countries k such that both endpoints also have an RTA with k. Zero common partners means a pure bilateral; dozens of common partners means plurilateral embedment.

Post-CPTPP trade diversion

The Comprehensive and Progressive Agreement for Trans-Pacific Partnership was signed 8 March 2018 in Santiago and entered into force 30 December 2018 for the first six ratifiers (Australia, Canada, Japan, Mexico, New Zealand, Singapore), with Vietnam following 14 January 2019 and Peru, Malaysia, Chile, Brunei staggered thereafter. A clean pre/post window is 2016-2017 vs 2019-2020, dropping 2018 as the transition year. Diversion, in the Viner (1950) sense, means trade shifts toward bloc members and away from excluded third parties. Below we compare raw growth of four bilateral-flow categories: CPTPP intra-bloc, CPTPP-to-non-members, non-members-to-CPTPP, and non-member-to-non-member (the control). The intra-bloc growth minus the non-member control growth is the Baier & Bergstrand (2007) difference-in-differences diversion proxy before adding pair fixed effects.

Intensive vs extensive margin of post-CPTPP export growth

Figure 6 measures aggregate trade-volume change. Melitz (2003, Econometrica 71(6): 1695-1725) and Chaney (2008, American Economic Review 98(4): 1707-1721) showed that the aggregate mass that a trade agreement moves decomposes into an extensive margin (number of product lines in the export basket) and an intensive margin (average value shipped per active line). For every CPTPP signatory we pool country_year_product, count distinct HS6 lines with positive exports per year (extensive), and divide total exports by that count (intensive). Averages are taken over pre = 2015-2017 and post = 2019-2023, so the pre/post deltas below net out single-year noise. 2018 is dropped as a transition year.

Integration-depth proxy: RTA network degree per country

Dür, Baccini & Elsig (2014, Review of International Organizations 9(3): 353-375) code every preferential agreement on seven provision dimensions (tariff, services, investment, competition, IPR, public procurement, standards-discipline) to produce the DESTA depth index. The DESTA panel is not ingested in the workbench, so this panel uses a transparent width-proxy: for each country, the number of distinct RTA partners carrying fta_wto = 1 in CEPII Gravity 2020. Mattoo, Rocha & Ruta (2020) Handbook of Deep Trade Agreements Table 2.1 shows that partner-count and DESTA depth correlate at roughly 0.7 across the 279 agreements in force by 2020; countries embedded in the EU+FTA network (partner count 30+) systematically carry higher DESTA depth scores than pure-bilateral signatories (partner count 1-5).

The income-tier architecture of the RTA system

Mansfield & Reinhardt (2003, International Organization57(4): 829-862) and Baccini, Dür & Elsig (2015, Review of International Organizations10(1): 113-138) document that PTA growth has been concentrated in the cross-income-tier and high-income clubs, with the South-South dimension thinner than the rhetoric of regional integration suggests. We classify each ISO3 country into a 2015 GDP-per-capita tier (high ≥ USD 12,000, middle USD 4,000-12,000, low < USD 4,000), then cut the 2020 active-RTA pair set into the six unordered tier-tier combinations.

Method note: from gravity β̂ to welfare

Baier & Bergstrand (2007, Journal of International Economics71(1): 72-95) show that naïve pooled OLS of the RTA dummy is biased by self-selection into treaties; once country-pair and time fixed effects are included and the RTA dummy is lagged, the average agreement raises bilateral trade roughly 100% over a decade (log point ≈ 0.7-1.0). Head & Mayer (2014, Handbook of International Economics vol. 4, ch. 3) meta-summarise modern gravity estimates with a modal RTA coefficient near 0.5. Limão (2016, Handbook of Commercial Policy vol. 1B, ch. 5) synthesises the preferential-trade-agreements literature and argues that deep PTAs deliver most of their trade-creation effect through behind-the-border provisions (services, investment, IPR), not tariff cuts alone. Conversion to welfare requires the structural gravity 'trade elasticity' σ (typically 4-8 in the Arkolakis-Costinot-Rodríguez-Clare 2012 ACR-class) and counterfactual simulation: a 0.5-log uplift at σ = 5 translates to roughly a 10% real-income gain for small open economies, with heterogeneity across sectors.

Policy read

The paper-over-paper multiplication of RTAs is real (Figure 1), but the average bilateral uplift is smaller than casual coverage numbers suggest (Figure 2). Depth matters more than count (Figure 4), and the world of RTAs is structurally plurilateral (Figure 5), which means the marginal treaty is mostly about bloc integration rather than bilateral trade creation. For a ministry: join the deep bloc you are geographically closest to, negotiate behind-the-border provisions rather than shallow tariff cuts, and do not over-estimate the trade-creation effect of a one-off bilateral deal. For a firm planning a sourcing shift: the relevant question is which bloc the supplier sits inside, not whether a bilateral exists.