The global tariff landscape

Nominal tariff protection has fallen for three decades, but the decline is uneven across countries and sectors, and a dense lattice of preferential agreements now sits on top of the MFN schedule. This page traces the world average MFN tariff since 1996, the countries at either tail, the preferential margins granted inside free-trade agreements, and the HS sections where protection still binds.

The long descent

Applied MFN tariffs have fallen over three decades as successive WTO rounds, unilateral liberalisation in developing countries, and accession bargains locked in lower bindings. Anderson & van Wincoop (2004) documented that tariffs are only one slice of total trade costs, but they are the policy lever most directly observable and most frequently renegotiated. Bown (2014, Journal of Economic Perspectives 28(4): 177–202, “Trade policy instruments over time”) and the WTO’s annual World Tariff Profiles describe the post-Uruguay-Round plateau: average applied MFN rates stopped falling after roughly 2010 even as preferential and contingent protection continued to evolve.

The tails: most-protected and most-open

The cross-section at the latest common year shows dispersion the world average hides. Small island economies that rely on import duties for revenue sit near the top alongside large developing economies that retain protection for strategic sectors. Zero-tariff entrepôts and free-trade partners cluster near the bottom.

MFN by income quartile

Tariff levels co-move with income: lower-income economies retain higher applied tariffs on average, both because customs revenue is a larger share of fiscal receipts and because tariff protection is a tool for infant-industry and balance-of-payments management (Bastos & Silva 2010, Journal of International Economics). We bin reporters by GDP per capita (WDI NY.GDP.PCAP.CD, latest within 5 years of 2022) into quartiles and report the mean, median, and p10-p90 band of country-average MFN within each bin. Q1 = poorest quartile, Q4 = richest. The World Bank’s official income-group classification uses Atlas-method GNI per capita; a single-year GDP-PCAP NTILE is a coarser but closely correlated proxy.

Preferences on top of MFN

The MFN schedule is one tariff; the tariff actually charged on a shipment depends on whether the exporter qualifies for a preferential regime — an FTA, GSP, or regional customs union. The preferential margin m_ij = (MFN_i − pref_ij) / MFN_iis the share of MFN protection that the agreement strips away. Kee, Nicita & Olarreaga (2009) built their Overall Trade Restrictiveness Index on exactly this MFN-plus-preferences decomposition.

Which sectors still carry the most protection

Tariff peaks are not randomly distributed across product space. Agriculture (HS sections 1–4), textiles and apparel (section 11), and footwear (section 12) consistently top the list. Machinery and electronics (section 16) and chemicals (section 6) average well below the world mean — a pattern that reflects both WTO zero-for-zero agreements on IT products and the trade-negotiating muscle of manufactured-goods exporters.

Tariff-policy headroom within the applied schedule

The WTO distinguishes between the bound rate (the ceiling a member agrees not to exceed without renegotiation under GATT Article XXVIII) and the appliedrate actually charged. The gap between them — the binding overhang — is the policy space a country can reclaim unilaterally without WTO process. The canonical applied-vs-bound series from the WTO’s World Tariff Profiles is not ingested in this workbench, so we report a next-best proxy: the gap between each reporter’s HS4 line peak (and p95) and its country-average applied MFN. Where peaks sit well above the mean, the existing schedule already has sharp within-reporter dispersion — selective re-protection is policy-feasible without raising the average. Bouët & Laborde (2010, World Trade Review) document that the binding overhang is a better measure of available protection than the applied mean, particularly for developing economies.

Tariff escalation along the processing chain

Balassa (1965, Journal of Political Economy) showed that nominal MFN rates understate the effective protection downstream manufacturers receive when inputs enter duty-free and final goods face high tariffs. The pattern — low tariffs on raw inputs, higher on semi-processed intermediates, highest on finished goods — is tariff escalation, and Corden (1966, JPE) derived the effective-protection formula that makes the bias explicit: ERP = (t_f − a·t_m) / (1 − a), where a is the input-cost share. The WTO’s World Tariff Profiles, UNCTAD and the FAO’s Agricultural Tariff Escalation Monitoring all track this pattern explicitly for commodities-to-manufactures chains. We compute the simple-average applied MFN at each processing stage across five canonical chains.

Convergence vs. dispersion in the long descent

The world average in Figure 1 is one moment of the cross-section. The other moment is dispersion: how widely reporters disagree on the mean rate they apply. Subramanian & Wei (2007, Journal of International Economics 72(1): 151–175) argue WTO membership pushed countries closer to a common trade-policy stance, which should narrow the cross-reporter spread of applied MFN. Bown (2014, JEP 28(4): 177–202) emphasises that contingent protection migrated off the MFN schedule onto antidumping and safeguards, leaving the headline average flat while dispersion may evolve separately. The line below is the cross-reporter standard deviation of country-average MFN, year by year, on the same reporter-year cells used in Figure 1.

Notes & references

Method. All figures use applied MFN rates at the HS4 line level from WITS/TRAINS, aggregated via simple (unweighted) means. Weighted averages would downweight prohibitive tariffs (which by construction carry low import values), so the simple mean better captures stated policy. Preferential margins are built from reporter-partner-HS6 pairs in pref_tariff_hs6; only pairs with at least 500 HS6 lines on each side are kept to avoid artefacts from partial schedules.

CES tariff formula. Under a CES-Armington import demand with substitution elasticity sigma, an ad-valorem tariff t raises the landed price to P₁ = P₀(1 + t) and imports contract along Q₁ = Q₀((1 + t₁) / (1 + t₀))^−sigma. Revenue is t₁P₀Q₁; the Harberger (1964) deadweight loss is 0.5 (t₁ − t₀)² sigma P₀Q₀ / (1 + t₀). For welfare, we use Arkolakis, Costinot & Rodríguez-Clare (2012, QJE 127: 51–80): dW/W = (1 − lambda_ii)^{1/(1−sigma)}, which is the sufficient-statistic welfare formula for the Armington, Eaton-Kortum, Krugman, and Pareto-Melitz classes. A working bilateral calculator lives at /tariff-scenarios.

Data sources. Applied and bound rates are maintained by WITS (World Integrated Trade Solution, UNCTAD + World Bank) and TRAINS; cross-checked against WTO Tariff Analysis Online and USITC DataWeb. Tariff disputes and contingent protection (safeguards, antidumping, countervailing) are catalogued in Bown (2004) “Global Antidumping Database” (World Bank; later the Temporary Trade Barriers Database) and in the WTO’s Integrated Database.

References. Anderson, J. E. & van Wincoop, E. (2004) “Trade Costs,” Journal of Economic Literature 42(3): 691–751. Arkolakis, C., Costinot, A. & Rodríguez-Clare, A. (2012) “New Trade Models, Same Old Gains?” QJE 127(1): 51–80. Bown, C. P. (2004) “Trade Policy under the GATT/WTO: Empirical Evidence of the Equal Treatment Rule,” Canadian Journal of Economics37(3): 678–720, and “Global Antidumping Database” (World Bank, 2004). Bown, C. P. (2014) “Trade policy instruments over time,” Journal of Economic Perspectives28(4): 177–202. Hertel, T. W. (1997) Global Trade Analysis: Modeling and Applications(Cambridge). Kee, H. L., Nicita, A. & Olarreaga, M. (2009) “Estimating Trade Restrictiveness Indices,” Economic Journal 119(534): 172–199. WTO, World Tariff Profiles (annual); UNCTAD/World Bank WITS/TRAINS; USITC DataWeb.