Loading...
Loading...
Explainer · empirical
Traditional finance underwrites two independent questions and needs a yes to both. Recoverable grants need a yes to neither. That single asymmetry is why the market serves one quadrant and exiles the rest — and, classified against two entire national charity registers, the exiled region turns out to be where most organisations actually live.
The two axes are bankability (can a venture service a fixed, scheduled, priced claim?) and extractability (is there a financial surplus a financier can capture?). Debt needs the first; equity needs the second. So the market fully serves only the top-right, and equity reaches partway into a second quadrant for the exitable few. Everything else is starved. A recoverable grant — philanthropic capital deployed with the expectation but not the obligation of return — is the one instrument indifferent to both axes: patient and contingent (so bankability is moot), and return of capital not on it (so extractability is moot). Its cumulative deployment is the geometric series T(R) = 1 / (1−R) in the recovery rate R — formally the same object as the Keynesian multiplier, with R in the role of the marginal propensity to consume. R = 0.90 gives 10×, 0.95 gives 20×, 0.98 gives 50×.
⚠️ T(R) is not the System Value Multiplier, and the difference is large. T(R) counts dollars put to work over an unbounded horizon. PSC’s SVM counts value created over N cycles, scaling turnover by a benefit coefficient γ: SVMPSC = γ·TN(R). At R = 0.98 that is 50× here, 38.6× there, and 22.7× for this paper’s own quantity truncated to 30 cycles. Three defensible numbers, none interchangeable — and the gap widens as R → 1. Quoting either without saying which quantity it measures is the kind of claim a diligence process breaks.
bankable · non-extractable
Steady earned revenue (a café or cleaning business employing people with barriers) but a mission-capped margin. It can repay — it just can't afford priced debt or dilute to equity.
A dollar at this recovery does
10.0×
T(R) = 1 / (1 − R) = 1 / (1 − 0.90)
…of cumulative work as it recycles — and it comes back to work at 26%/yr (a full recycle roughly every 3.9 yr).
Convex: the last points of recovery dominate. 90→95% doubles it; 95→98% more than doubles it again.
Empirical anchor: diversified early-stage impact recovers ~91c/$ (Acumen); the most concessional capital targets capital preservation only (Omidyar B2). No systematic recoverable-grant benchmark exists in either country — these R’s are illustrative.
T(R) counts dollars deployed, not value created. It is not PSC’s System Value Multiplier, which truncates at N cycles and scales by a benefit coefficient γ: at R = 0.98 this is 50× and that is 38.6×. The two are not interchangeable, and the gap widens as R → 1.
Blend a pool across the map
The pool is two things: a high-R engine that regenerates the corpus, and a low-R consumption tranche it subsidises.
9.7×
pool multiplier
60% in the engine (R ≥ 80%)
Even a consumption-tilted pool beats a traditional grant’s 1×. First-order estimate (Σ w·1/(1−R)); a full model credits dynamic redeployment into the engine.
The model: plot four archetype ventures, dial the recovery rate, and read the multiplier. Only the top-right corner is served in volume; the archetypes span the exiled region.
The empirical result
The map is not only a diagram. We classified every financially-reporting charity in two countries — 44,196 in Australia (ACNC, 2021–24 AIS) and 283,771 in the United States (full-990 501(c)(3)s via IRS 990 / NCCS Core, 2021–23) — on identical axes, weights, cutoffs and sensitivity, with no re-tuning. In both, the bankable, non-commercial cell is the single largest population by entity count, and the share is stable across 40/50/60 cutoffs. The finding survives a sixfold change in population and a change of jurisdiction, regulator and accounting form: the structure lives in the axes, not one country’s funding mix.
⚠️ What this census does and does not establish. Extractability is near zero across a legally non-distributing population, so it does not vary here and these data cannot test whether the two axes are independent — that argument is made structurally, and stands or falls there. What the data establish is the sizing: how many organisations can carry a scheduled claim, and how many of those are funded by one-and-done gifts. Both are the right measurements for that question, and neither is a proxy for extractability. The map is the argument; the census is the target.
🇦🇺 Australia
52.6%
bankable · non-commercial (23,247 of 44,196)
Addressable target: 10,551 orgs · $10.1B/yr
donation-dominant, less endowed grant-makers and one-off capital transfers.
reconciled within 2–4% of ACNC aggregates.
🇺🇸 United States
48.6%
bankable · non-commercial (137,913 of 283,771)
Contributions-dominant subset: 127,907 orgs · $600.4B/yr
not the same quantity as Australia's — US contributions bundle government grants, so a donation-only target cannot be isolated.
reconciled within 3.7% of IRS SOI.
In Australia, 52.6% of the 44,196 classified ACNC charities are bankable but non-commercial — the exiled region, and the largest cell by entity count. Sector revenue reconciles within 2–4% of the ACNC's published figures in all four years.
All 44,196 classified ACNC charities (2021–2024), binned by bankability × commerciality. Darker = denser; the mass sits bottom-right.
| Cell (bankability × commerciality) | 🇦🇺 % entities | 🇺🇸 % entities |
|---|---|---|
| Bankable · non-commercial | 52.6% | 48.6% |
| Bankable · self-funding | 19.6% | 23.6% |
| Fragile · non-commercial | 19.9% | 17.8% |
| Fragile · trading | 7.9% | 10% |
Two honest caveats. (i) US contributions bundle government grants (not separable in NCCS Core), so US “donation-only” cannot be isolated at the org level; the like-for-like cross-country quantity is the entity share, not the dollar split.(ii) Both registers include grant-making intermediaries (DAF sponsors, community foundations) — 4.2% of US entities but ~13% of the exiled cell’s dollars; excluding them by NTEE classification leaves the exiled entity share essentially unchanged (48.6→47.9%), so the entity-count headline is robust while the dollar total is not. ⭐ Australia has no equivalent NTEE category, but it carries the money flows, so the same population is reachable by behaviour instead: grants made at or above half of expenditure. That screen is much broader — 18.4% of entities, because it catches anything that regrants half its spending and not only what an administrator has coded as a foundation — and takes the cell from 52.6% to 50.0%. Two jurisdictions, two independent constructions of the same idea, and neither moves the entity-count headline by more than three points. ⚠️ They are not the same instrument and should not be quoted as one: an NTEE code says what an organisation is, an expenditure ratio what it did in a window. Method & per-year reconciliation: companion working paper, Bankable and Extractable (Ghadamian, IRSA Institute).
The entity count is the paper’s claim; the aggregate dollar figures corroborate it from the other direction — conventional finance dominates by volume (it funds the safe, collateralised top-right) yet serves a minority of entities.
How much is exiled — the evidence
~95%
of AU SME lending is collateralised — unsecured SME credit sits below 5%.
RBA, 2018–2025
85–92%
of the ~$157B AU impact market is market-rate GSS bonds; concessional capital is a small minority.
Impact Investing Australia / RIAA, 2025
~7%
of the ~$222B charity sector's revenue is donations/bequests; repayable capital is negligible.
ACNC, 2023
~91c/$
is the recovery ceiling even for diversified early-stage impact; no AU recoverable-grant benchmark exists.
Acumen / Omidyar
Verified via adversarial multi-source review. The market serves the top-right by dollar volume; by number of ventures and share of the mission economy, the exiled region is the majority.
The single parameter that governs the multiplier — the recovery rate R— is the least measured. The best anchors are global and ex-ante (Acumen ~91¢/$; Omidyar’s most concessional tier targets capital preservation), and no systematic recoverable-grant recovery benchmark exists in either country. Individual programmes report figures — Ford (2013) documents a PRI-capitalised revolving loan fund repaying at 97% — but a single well-run instance is not a base rate for a population selected on these axes, and it is that base rate a pool at scale would be first to generate. The binding uncertainty is empirical, not conceptual: the map says the target is large, real and countable; what remains is to deploy, recover, and measure — which a pool at scale would be the first to do.
Entity-level classification of both registers on bankability (surplus consistency, margin, asset backing, revenue stability) and commerciality (market-facing earned income, a proxy for extractability, structurally near zero across a non-distributing sector). Reconciled to each regulator’s published aggregates per year. Recovery rates in the model are illustrative. Both classifications, their per-year reconciliations and the pipeline are deposited: Australia at doi.org/10.5281/zenodo.21817460, the United States at doi.org/10.5281/zenodo.21817482 — separate records, not interchangeable. Full method, code and reconciliation in the companion working paper and analysis/capital-map.