The Counterparty Problem

Executive Summary

The single most important finding is that “U.S. reliability” is not one question but two, and conflating them produces bad risk analysis. The United States has a real, documented and recently accelerating record of formally exiting binding treaties — approximately fifteen significant withdrawals or repudiations since 2001, clustered in two windows (2017–2020 and 2025–2026). But there is no honest way to compute a “failure-to-convert” base rate for non-binding instruments like Memoranda of Understanding, because no registry of U.S. MoUs exists.

For a sovereign allocator, the operationally correct conclusion is that the 24 July 2026 PIF–EXIM MoU should be underwritten as a political signal whose value depends entirely on a Congressional reauthorization that has not yet happened — not as a $15 billion guarantee.

The fair reading is not “the U.S. always defects.” Democracies changing policy across administrations is normal; the United States has honoured core commitments — NATO Article 5, the Bretton Woods institutions, the treaty alliances with Japan and Korea — over decades, and empirical scholarship finds states generally comply with ratified treaties. Binding-treaty exits are real but procedurally constrained and often reversible: Paris and the WHO were both rejoined. MoUs carry near-zero legal weight and are best read as signals. The risk that matters sits between those poles, and §4 states it precisely.

1. The Immediate Context

Independently verified, current to 26 July 2026.

The PIF–EXIM MoU (24 July 2026). Saudi Arabia’s Public Investment Fund and the Export-Import Bank of the United States signed a Memorandum of Understanding creating a framework of up to $15 billion in long-term export-credit financing for eligible PIF portfolio companies procuring U.S.-origin goods and services — spanning advanced technology, aerospace, infrastructure, future mobility, water security and critical minerals. PIF stated that procurement from the U.S. by the fund and its portfolio companies since 2017 totals $65 billion, contributing an estimated $35 billion to U.S. GDP (PIF release). On the same day PIF signed related MoUs with the World Bank Group’s IFC and MIGA arms totalling up to $9.5 billion, for a combined announced figure of up to $24.5 billion. The instrument is explicitly an MoU establishing a “framework” — a non-binding statement of intent, not an authorized transaction.

The charter sunset. EXIM’s statutory charter expires 31 December 2026. Congress has not enacted reauthorization.

The coalition letter (21 July 2026). The U.S. Chamber of Commerce published a coalition letter to Congress signed by roughly 200 state and national organizations urging a “robust, long-term reauthorization” — signatories include the National Association of Manufacturers, the Aerospace Industries Association, Boeing, Lockheed Martin, RTX, Ford, and the Business Roundtable. The letter states that “more than 115 foreign export credit agencies actively support their national exporters” (U.S. Chamber), and the Chamber’s companion advocacy notes China “provided an estimated $39 billion in export credit support in 2022 — fourteen times the volume officially supported by the United States that same year” (U.S. Chamber).

2. Universe One — The Export-Credit Record

EXIM is the sharpest illustration of the failure mode that matters: an institution neutralized not by crisis or default, but by legislative inaction.

The lapse

EXIM was created by executive order in 1934 and chartered as an independent agency under the Export-Import Bank Act of 1945 (12 U.S.C. §635 et seq.), subject to periodic Congressional reauthorization. Two overlapping failures define the modern record. The charter itself lapsed for five months — 1 July to 4 December 2015, the longest gap in the agency’s history. And on 20 July 2015 the bank lost its board quorum when two members’ terms expired, leaving two of the required three seats. The quorum was not restored until 9 May 2019 (CSIS). For nearly four years the board could not approve any transaction above $10 million.

The quantified impact

A 2024 National Bureau of Economic Research study (NBER Working Paper w32019; Kabir, Beaumont, Matray & Xu, “EXIM’s Exit: The Real Effects of Trade Financing by Export Credit Agencies”) finds that the value of EXIM’s financial support to firms declined by almost 85% between 2014 and 2019, and estimates that a $1 reduction in EXIM trade financing reduces exports by approximately $4.50. In FY2014 — the last full year with a quorum — EXIM authorized roughly $20.5 billion, supporting about $26.5 billion of exports and about 164,000 jobs. Medium- and long-term new authorizations fell to near zero during the lapse. The Mercatus Center documents average approval size falling from $6.7 million (3,157 transactions, FY2014) to $1.9 million (1,968 transactions, FY2018), with 1,189 fewer companies served.

EXIM’s own Office of Inspector General (OIG-EV-23-04, 29 September 2023) concluded the lapse “caused substantial damage to its institutional reputation and medium- and long-term portfolio,” from which EXIM “has not yet fully recovered.”

The stranded counterparties have names. The roughly 76-year EXIM–Pemex relationship was suspended, re-engaging only with a proposed $400 million facility after March 2020 applications. Boeing forfeited a potential Ethiopian Airlines sale. GE relocated a plant and 350 jobs to Canada to access reliable export-credit financing (PIIE). A $3.5 billion Egyptian petrochemical facility went to competitors financed by Germany, China, the UK and Canada. As of end-FY2018, roughly $40 billion in transactions sat stranded in EXIM’s pipeline (CNBC).

The 2019 reauthorization and the partial fix

The Further Consolidated Appropriations Act, 2020 (P.L. 116-94, Div. I, Title IV, signed 20 December 2019) extended EXIM’s authority through 31 December 2026 — the longest reauthorization in its history — and added a mitigation mechanism: under 12 U.S.C. §635a(c)(6), if a board quorum lapses for 120 consecutive days within a President’s term, a “temporary board” of specified officials acts in the board’s stead, though it cannot change policies or procedures (CRS IF10017). The delegated-approval threshold was raised from $10 million to $25 million.

Reauthorization status, late July 2026

The Senate bill — S. 3772, the Export-Import Bank Reauthorization Act of 2026, introduced in February 2026 by Senators Mark Warner (D-VA) and Kevin Cramer (R-ND), both Banking Committee members — would extend EXIM’s charter and the China and Transformational Exports Program by ten years, to 2036. Per the Congressional Research Service (R43581, updated 7 May 2026), the bill as introduced retains the existing $135 billion lending cap. The widely reported $205 billion figure is a sponsor’s stated negotiating goal, not enacted text. On the House side, no numbered companion has been introduced; a discussion draft dated 11 March 2026 at House Financial Services proposes a more modest five-year extension. Committee action to date consists of hearings only — no markup, committee vote, or floor vote has occurred.

No source located predicts confident on-time passage. Positives: bipartisan Senate sponsorship and reported administration support. Risks: the chambers diverge on length and cap, and nothing has advanced out of committee by mid-2026. Analysts point to a year-end vehicle — the FY2027 NDAA or an appropriations package — as the probable path (CSIS). The 2015 precedent is the cautionary base case.

The comparative record

The global count of official export-credit providers grew from 85 to 115 — a 35% increase — between 2015 and 2019, precisely the window in which EXIM was constrained (EXIM Competitiveness Report). No peer G7 export-credit agency suffered a comparable multi-year authorization interruption: JBIC/NEXI, KEXIM/K-Sure, UKEF, Euler Hermes, Bpifrance, EDC and SACE all expanded product offerings while EXIM was idle. The top medium/long-term providers in 2019 were China ($33.5 billion), Italy ($11.1 billion) and Germany ($10.5 billion); the United States ranked eighth at $5.3 billion. From 2015 to 2019, China’s official MLT export-credit activity alone was at least equal to 90% of that provided by all G7 countries combined. The strategic backdrop is stark: a self-imposed four-year outage during the very period the principal competitor was scaling most aggressively.

EXIM’s MoU roster — a bounded, honest sample

EXIM has signed numerous MoUs with foreign export-credit agencies, sovereign funds, chambers and governments over the last decade: Hungary’s HEXIM; Nigeria’s NEXIM; Saudi EXIM (January 2024); the Abu Dhabi Investment Office; a 13-agency multilateral MoU on Ukraine reconstruction; a transatlantic infrastructure MoU; and now PIF (July 2026). The analytically correct measure is the conversion rate — did each MoU produce an authorized facility, guarantee or co-financing? That rate cannot be computed honestly, because public disclosure of MoU-to-transaction conversion is fragmentary. What is documentable is the direction of the risk: EXIM’s own MoU partners were exposed during 2015–2019, when the bank could not authorize the large transactions its cooperation frameworks contemplated. The Pemex suspension is the clearest example of a decades-long partner stranded by an authority lapse. An MoU “framework of up to $15 billion” is contingent on EXIM retaining the authority to authorize the underlying transactions — an authority that expires in five months absent Congressional action.

3. Universe Two — The Treaty and Major-Agreement Record

Formal withdrawals, repudiations, and non-implementations since 2001 — with mechanism and consequence preserved, because the distinctions matter.

The following are documented, significant instances. Several are non-ratifications or “unsignings” rather than withdrawals from binding treaties; the categorization is preserved throughout.

• Kyoto Protocol (2001): ratification abandoned; the Senate had signalled non-ratification 95–0 (Byrd–Hagel). Non-implementation of a signed protocol.
• ABM Treaty (2002): formal withdrawal from the 1972 U.S.–Russia treaty, exercised via the treaty’s own withdrawal clause.
• Rome Statute / ICC (2002): “unsigned” — the model for later unsignings.
• JCPOA (2018): withdrawal from a multilateral arrangement the prior administration itself treated as a non-binding political commitment; sanctions snapback and de facto collapse followed.
• 1955 Treaty of Amity with Iran (2018): withdrawal after an adverse ICJ provisional-measures ruling — a binding bilateral treaty exit.
• INF Treaty (2019): withdrawal citing Russian non-compliance; eliminated a treaty with robust on-site verification.
• Arms Trade Treaty (2019): “unsigned” — signature returned from the Senate (AJIL).
• Open Skies Treaty (2020): withdrawal effective 22 November 2020, over allied objections.
• Paris Agreement — two full cycles: withdrawal effective 4 November 2020; rejoined 19 February 2021; second withdrawal ordered 20 January 2025, effective 27 January 2026.
• WHO: withdrawal announced 2020; reversed January 2021; withdrawal ordered again 20 January 2025.
• UNESCO: withdrawal announced three times (1984, 2017, 2025 — effective 31 December 2026), rejoined twice (2003, 2023). The archetype of oscillating commitment.
• UN Human Rights Council (2018; 2025): two exits.
• TPP (2017): signature withdrawn before ratification; the remaining eleven proceeded as CPTPP.
• Broader 2025 disengagement: moves to exit or defund numerous bodies, and the closure of USAID — a systemic pattern in that window, not an isolated one.

The base rate in the other direction

NATO Article 5 has held for over 75 years and has been invoked exactly once — by the allies, on behalf of the United States, after 9/11. The Bretton Woods institutions have had continuous U.S. membership and leadership for eight decades. NAFTA was renegotiated into USMCA, not repudiated — renegotiation is a distinct and less severe phenomenon than withdrawal. The bilateral mutual-defence treaties with Japan and the Republic of Korea have been sustained for decades and carry categorically more weight than Budapest-style assurances.

4. The Two Base Rates, With Explicit Sample Frames

Definition A — formal withdrawal or repudiation

Sample frame: significant U.S. treaty and major-agreement exits, unsignings, and non-implementations, 2001–2026, identifiable from primary sources (State Department notices, Federal Register and executive orders, UN depositary notifications) and authoritative secondary sources (CRS reports R44761 and R48504; the compliance literature). Count: approximately fifteen significant instances, plus the broader 2025 wave of organizational exits. Distribution: heavily clustered in 2017–2020 and 2025–2026; predominantly executive-driven; several reversible and in fact reversed. The clustering is itself the key analytic point — the phenomenon is strongly correlated with specific administrations, which is both evidence of unreliability to a counterparty needing multi-decade certainty and evidence of normal democratic policy alternation.

Definition B — failure to convert

This paper declines to state a base rate for stated intents, MoUs and pledges that never produced their signalled action, because no honest denominator exists. There is no central registry of U.S. MoUs; they are signed by State, Commerce, DoD, DOE, Treasury, USTR, DFC, NASA, HHS and others; many are unpublished; and Case-Zablocki Act reporting is demonstrably late and incomplete (Just Security). GAO report GAO-26-108186 found the State Department reported nearly one-third of covered international commitments to Congress late between October 2023 and March 2025, and consistently missed the website-publication deadline. Scholarship by Hathaway, Bradley and Goldsmith documents decades of late and incomplete reporting, and the rising use of non-binding instruments precisely to avoid scrutiny. The bounded substitute offered here is EXIM’s own MoU roster (§2) — and even there, deal-by-deal conversion is not publicly tracked. That absence is itself the finding. Constructing a “last 100 MoUs” denominator would require inventing a sample frame the underlying records cannot support.

5. What a Fair Reading Concedes

A hostile Atlanticist analyst, a compliance officer and a sophisticated sovereign allocator would each press the following — and each has force.

“Democracies changing policy is a feature, not a defect.” Conceded. A newly elected government reversing its predecessor’s discretionary commitments is legitimate democratic practice, not bad faith. The Paris and WHO reversals were policy oscillation, not repudiation of binding obligation. This is why the paper separates policy change across administrations, legislative-authorization failure (the EXIM case), and bad-faith repudiation — three phenomena with different risk implications.

“The U.S. has honoured its most important commitments for generations.” Conceded and emphasized. NATO Article 5, Bretton Woods, and the treaty alliances with Japan and Korea have held for 70–80 years. The withdrawals cluster in discretionary, often executive-agreement or non-binding categories — not in the ratified, deeply institutionalized treaties that anchor the security and financial order.

“Other major states have worse records.” Conceded. Russia’s violation of the Budapest Memorandum and Chinese opacity in export finance are more serious breaches of specific commitments than most U.S. exits. The empirical compliance literature finds that states, including the United States, comply with ratified treaty obligations most of the time; the U.S. is not a systematic outlier on formal treaty compliance. What distinguishes the U.S. record is volatility of discretionary commitments, not breach of binding ones.

What a fair reading nonetheless concludes for a counterparty: the relevant risk is not that the United States defaults on binding treaties — it rarely does. It is that (a) discretionary and non-binding commitments are exposed to sharp reversal on a four-to-eight-year electoral cycle, and (b) even statutorily-created institutions can be neutralized by Congressional inaction, as 2015–2019 proved. The MoU’s value is real but conditional; the binding question is the 31 December 2026 reauthorization.

6. What Could Not Be Determined

• A true MoU conversion rate. No denominator exists; U.S. MoUs are decentralized, frequently unpublished, and inconsistently reported. Any specific percentage would be fabricated.
• Deal-by-deal outcomes of most EXIM MoUs. Public sources announce signings but rarely track whether a given MoU produced authorized facilities.
• The final terms and passage odds of S. 3772. In committee with no markup; House and Senate versions diverge; the likely year-end vehicle and its timing are not settled. The $205 billion cap figure is a sponsor’s goal, not enacted text.
• A precise count under Definition A. “Approximately fifteen significant” is deliberate; the exact figure depends on whether unsignings, non-ratifications, re-withdrawals of the same agreement, and the 2025 organizational wave are counted as discrete events. Reasonable analysts will bracket differently.

7. Contested Figures and Source Notes

• China’s export-credit volume is cited variously as ~$39 billion (2022, Chamber letter), ~$33.5 billion MLT (2019, EXIM), ~$24 billion MLT (2024, per the July 2026 coalition letter), and ~$76 billion total (2019, EXIM conservative estimate). The figures differ by year and by whether the measure is MLT-only or total official financing; Chinese figures are inherently opaque and should be read as estimates.
• EXIM “lost opportunity” figures (~$20 billion lost; ~$40 billion stranded pipeline) originate largely from EXIM proponents and officials. The ~85% financing decline is the more rigorous, independent figure (NBER w32019).
• The Senate Banking hearing date shows an internal inconsistency across official sources (19 vs. 26 March 2026); unresolved.
• MoU legal status. Per U.S. State Department guidance, calling a document an “MoU” does not automatically render it non-binding for the United States — intent and wording (“will/should” vs. “shall/agree”) govern. Most MoUs are political commitments whose breach triggers reputational, not judicial, consequences. This is why conversion of stated intent, not breach, is the analytically correct lens.

Engage with this Working Paper

Substantive comment, technical critique, and named response are welcomed. The Working Paper is maintained as a public version-controlled document on GitHub; the version above is the editorially-controlled v1.0. Indicative figures are flagged as such throughout; contested figures are reported rather than resolved. It is analysis, not intelligence, and carries no classification.