Reform of a multilateral institution changes behaviour only when it changes what the institution must count

Every few years the multilateral institutions are told to reform, and the proposals arrive in a familiar order: shares, chairs, mandates, capital. The debate is conducted almost entirely in the language of governance, which is the language of who decides. My position is that this is the wrong end of the problem. What an institution does is set by three things no vote reallocates: the balance sheet it must protect, the appropriation it must survive, and the measure it is obliged to report against. A reform that leaves all three intact changes the authorship of a decision rather than the decision.

Where this is written from

The vantage here is narrow and worth naming before anything rests on it. I am an analyst writing inside The Cefa Brief, working from published statistics and public documents and from nothing else. The figures below are annual series redistributed by FRED and retrieved for this site on 19 September 2026, read alongside the publishing agencies' own descriptions of what each series counts. I hold no seat in any institution, no brief from any government, and no sight of a board paper, a term sheet or a staff appraisal.

That position carries a specific blindness, and it happens to be the same blindness the reform debate carries. This site holds national series, published by national statistical agencies, at annual frequency. It holds nothing about the institutions this piece is about: no lending book, no staffing, no project record, no disbursement schedule. The asymmetry deserves a moment. A sovereign economy is documented to a decimal place by agencies obliged to publish on a timetable. The bodies lending into those economies are documented by themselves, at their own discretion, in formats of their own choosing. An argument about multilateral reform built from published statistics is an argument made from outside the institution, and I will hold the claims to what the outside supports.

What the debate is usually about

Reform proposals fall into three families. The first concerns voice: how shares are allocated, how chairs are filled, which capitals hold a veto in practice. The second concerns the balance sheet: how much risk the capital base can carry, how much of it is callable, what a rating agency will tolerate. The third concerns instrument and mandate, meaning what may be lent, on what terms and to whom. Communiqués bundle the three together, which obscures that they are not the same kind of proposal.

Voice reform is the loudest of the three and the least binding. Reallocating shares changes who authorises a decision and who answers for it afterwards. It does not change the constraint under which the decision is taken. A board with a different distribution of votes, facing the same capital position, the same rating and the same reporting obligation, approves broadly the same book, and nothing available from this vantage would distinguish the two. That is not an argument against reallocation, whose case rests on legitimacy and requires no claim about behaviour at all. It is an argument against the expectation that reallocation is what will change what these institutions do.

The balance sheet the capital is asked of

Capital adequacy reform asks shareholders to stand behind more risk, and that request lands on national balance sheets. One such balance sheet is documented here. Federal debt as a share of American GDP reads 122.3 percent in 2020, 120.9 in 2021, 118.4 in 2022, 117.7 in 2023, 120.4 in 2024 and 120.7 in 2025. The whole range is 4.6 points, from a low of 117.7 in 2023 to a high of 122.3 in 2020, and the ratio does not fall below 117.7 anywhere in the window. Concessional money sits worse still, because it is appropriated rather than guaranteed. The American federal surplus or deficit, in billions of dollars, reads minus 3,096 in 2020, minus 2,774 in 2021, minus 1,374 in 2022, minus 1,687 in 2023, minus 1,815 in 2024 and minus 1,775 in 2025. Every observation is a deficit, the narrowest is 1,374 billion, and the most recent is 401 billion wider than the narrowest. A shareholder guarantee is a contingent claim on the first of those positions. A replenishment is a line item inside the second. Neither is altered by the distribution of votes.

The measure is the instrument

The third constraint is the one the debate rarely reaches. These institutions sort countries using published statistics, and the statistic does the sorting whether or not anyone defends it. Income per head is the usual instrument. Real GDP per person in the United States, in chained 2017 dollars, reads 61,127 in 2020, 64,758 in 2021, 66,026 in 2022, 67,411 in 2023, 68,681 in 2024 and 69,749 in 2025, a rise of 8,622 dollars across five steps. The first step is 3,631 dollars and the last is 1,068, with 1,268, 1,385 and 1,270 in between. That is one country, and the only one this site holds such a series for. The shape is the point rather than the level. A measure of this kind is an average over a population, it carries no distribution inside it, and it moves smoothly enough that a threshold drawn across it will be crossed by arithmetic rather than by anything a ministry would recognise as a change.

An eligibility threshold is a policy instrument nobody votes on. So is a disclosure requirement, the definition of a concessional loan, the point at which an exposure is marked impaired, and the rule fixing when a project counts as complete. Each decides what staff are rewarded for producing, and that is the only mechanism by which an institution's behaviour reliably changes. An organisation manages the thing it is required to count. It does not manage the thing a communiqué says it values.

Which reforms would change behaviour

Three, on this reading. Change what must be published, at what frequency, and in a form that permits comparison across institutions, which is the discipline national statistical agencies already work under and lending institutions do not. Change the definitions that decide eligibility and classification, because that is where money is allocated in practice. Change the rule determining when a failure is recognised and who books it, because an institution that never records a loss has no measurable appetite for risk and cannot be argued with about whether it is taking enough. None of the three requires a reallocation of shares. All three are duller than a reallocation of shares, which is part of why the debate prefers the shares.

What this cannot establish should be stated as plainly as the rest. Six annual observations of three American series say nothing directly about any multilateral institution. They do not measure a lending book, they cannot date a decision, and they cannot show that a disclosure rule has ever changed a staff incentive anywhere. The claim is narrower, and holding it exactly matters: the constraints that bind these institutions are of the kind recorded in accounts and statistics, the reform debate is conducted in the kind recorded in communiqués, and the absence of any comparable published series about the institutions themselves is why an outsider cannot settle the question. That absence is itself one of the reforms.