Official development assistance fell by 23% in 2025, the largest drop on record. Aid disbursements have tended to coincide with deposits in offshore banks. This column argues that once the Panama Papers raised the cost of offshore banking, part of that diversion migrated to cryptocurrency. Linking World Bank disbursements to blockchain forensics, it finds sharp, short-lived surges of anonymous crypto activity on tax-haven exchanges in the month aid arrives, implying leakage of two to six cents per aid dollar. The public ledgers that carry diverted funds also make them detectable, which argues for supervision timed to the moment the money lands.

In April 2026, the Organisation for Economic Co-operation and Development (OECD) reported that official development assistance from its donor members fell by 23% in real terms in 2025, to $174 billion, the largest annual contraction since records began (OECD 2026). The US accounted for three quarters of the decline, and Germany, France, and the UK cut as well. The immediate debate concerns the human cost: Sandefur (2026) traces the mortality consequences of the US withdrawal, and the CEPR assessment of the second Trump administration documents the wider damage to global health programmes (Gensler et al. 2025). Behind that debate sits a second one about accountability: donors under fiscal pressure want evidence that the money they still send reaches its purpose, and the case for aid is harder to make when part of each dollar is known to disappear. Where that part goes, and through which channel, is therefore a first-order question for the budgets that remain.

Where diverted aid used to go

Economists have measured aid leakage for two decades. Reinikka and Svensson (2004) found that only 13% of a Ugandan capitation grant reached the schools it was meant for, and Olken (2007) measured roughly a quarter of expenditures missing from Indonesian road projects. The most direct evidence on where diverted money ends up comes from Andersen et al. (2022), who show that World Bank disbursements to aid-dependent countries coincide with jumps in deposits in offshore financial centres known for bank secrecy, implying leakage of about 7.5 cents per aid dollar. Offshore wealth of this kind is highly concentrated at the top of the distribution (Alstadsæter et al. 2018), and for decades it relied on shell companies, nominee owners, and discreet banks in lightly regulated jurisdictions.

That infrastructure became more fragile after 2016. The Panama Papers exposed the owners behind hundreds of thousands of offshore vehicles (ICIJ 2016), and the enforcement that followed, from blacklisting to the automatic exchange of bank information, raised the cost of documentary concealment (Masciandaro 2016, Boas et al. 2024). Enforcement of this kind has a documented side effect: Johannesen and Zucman (2014) showed that the G20 crackdown on bank secrecy relocated hidden wealth between havens and left the total largely intact. Greater scrutiny of offshore banking need not have reduced the demand to hide diverted aid; it may have moved that demand elsewhere. The central question of our research is whether it moved to cryptocurrency.

Why cryptocurrency?

Cryptocurrency is the natural candidate for that migration. Pseudonymous wallets can be created in seconds, value crosses borders in minutes, and exchanges domiciled in lightly regulated jurisdictions provide entry and exit points with weak customer identification. Graf von Luckner et al. (2023) show that bitcoin is used to move capital across borders and around capital controls in emerging markets, and Di Casola et al. (2023) find that bitcoin trading in developing economies rises when the domestic currency is unstable. Foley et al. (2019) estimated that a large share of early bitcoin activity financed illegal trade, and Papsdorf et al. (2022) argued that the sector should face the same anti-money-laundering standards as conventional finance. For research, cryptocurrency differs from an offshore bank in one respect that matters: it conceals the owner while publishing every transfer.

Following the money in the month it arrives

In Agarwal et al. (2026) we use that ledger. We link project-level World Bank disbursements to 93 recipient countries between 2018 and 2024, $238 billion in total, to three sources of crypto data: on-chain bitcoin transactions, with wallets classified as identified when they belong to a known service and as anonymous otherwise; off-chain records of activity inside centralised exchanges; and IP-linked web traffic to exchange platforms, which attributes activity to countries. Identification comes from the timing of the money. World Bank tranches are released as projects clear procurement benchmarks, audit checkpoints, and board calendars, so the month in which a tranche lands is governed by administrative process (Kraay 2012, Andersen et al. 2022). We isolate 328 large tranche arrivals and compare crypto activity in the arrival month with the months around it and with countries receiving no disbursement.

The response is sharp, short-lived, and concentrated where concealment is easiest. Figure 1 plots the event study for transaction volume. Anonymous volume routed through tax-haven exchanges rises by roughly 137% in the arrival month, while identified volume rises by roughly 35%; neither series trends before arrival, and both return to baseline within one to two months. Entry into the crypto system runs almost entirely through newly created wallets that resolve to no known service. Exchange-reported activity in unregulated jurisdictions responds more than ten times as strongly as activity on regulated platforms, and mixers gain relative to payment services and ordinary exchanges. Read together, the patterns match the placement, layering, and integration sequence of money laundering, compressed into the weeks after the money arrives.

Figure 1 Crypto transaction volume around the arrival of a World Bank tranche

a) Transaction volume: Anonymous wallets

b) Transaction volume: Identified wallets

Notes: Event-time coefficients from a stacked event-study difference-in-differences on 328 tranche arrivals, 2018–2024. The outcome is geo-adjusted bitcoin transaction volume (inverse hyperbolic sine), attributed to countries by their share of web traffic to exchange platforms, for anonymous wallets (Panel A) and identified wallets (Panel B), on tax-haven and non-tax-haven exchanges. Month 0 is the month the tranche lands; the reference period is months −12 to −9. Shaded bands are 95% confidence intervals, clustered by recipient country.
Source: Agarwal et al. (2026), Figure 3.

Diversion of money in hand

The alternative reading is that economies under stress simply demand more crypto. Natural disasters that are not followed by a World Bank disbursement produce no crypto response at all, even though the countries in that placebo sample are richer and better connected than the aid recipients. The Pandora Papers leak of October 2021, a second shock to documentary concealment, attenuated the identified part of the post-disbursement response and left the anonymous part intact; that is the pattern substitution towards pseudonymity predicts, and a general contraction in crypto use would have attenuated both. Timing points the same way: a crisis that prompts aid could move crypto demand months before the money arrives, yet the response appears only in the arrival month. The channel also requires internet access, exchange accounts, and private-key custody, capabilities that remain scarce among intended beneficiaries; only 27% of people in low-income countries used the internet in 2023 (ITU 2023). The evidence points to actors close to the disbursement chain.

Two to six cents per dollar

Cumulating the rise in anonymous tax-haven volume over the months after arrival gives a gross figure of 11 cents per disbursed dollar. That is an upper bound, because the blockchain records gross flows and the same dollar is counted each time it moves between wallets. Figure 2 shows how net leakage falls as the number of hops rises. At the two to five hops that forensic practice treats as typical of layering (Chainalysis 2021), net leakage is two to six cents per aid dollar, or roughly $1.7 to $4.3 billion across the $78 billion that arrived in the 328 tranche months we study. Under plausible layering, the crypto channel is of the same order as the offshore-banking channel in Andersen et al. (2022). We observe bitcoin only, and stablecoins now carry most illicit on-chain activity (FATF 2025), so the total is likely larger.

Figure 2 Net leakage per aid dollar depends on how often each diverted dollar is observed

Notes: Net leakage equals the gross ratio divided by the turnover multiplier, the number of times a diverted dollar is observed as it passes through successive wallets and off-ramps. The solid line uses the cumulative post-arrival response (gross ratio 11.0%); the dashed line uses the arrival-month response alone (13.2%). The horizontal line marks the 7.5% offshore-banking estimate of Andersen et al. (2022).
Source: Agarwal et al. (2026), Figure 5.

Where it leaks

The response also varies systematically across sectors. Figure 3 ranks World Bank sectors by the arrival-month response. Transport, a category of large and discretionary construction contracts, leads, followed by Water and Sanitation, Social Protection, and Governance. In-kind Health aid, which creates no local liquid windfall, shows no response. Tranche size does not drive the ranking: Transport has the largest response and the smallest average tranche. The four sectors with measurable capture absorbed 48% of sector-classified World Bank disbursements and new commitments in the two years after our sample ends. We make no claim that the Bank rewards leaky sectors; the point is exposure, since close to half of continuing money flows to sectors where capture is measurable.

Figure 3 Which sectors capture more? Arrival-month response by World Bank sector

Notes: Each bar is the arrival-month coefficient for anonymous transaction volume on tax-haven exchanges, estimated on the tranche arrivals whose dominant sector is the one shown; whiskers are 95% confidence intervals and n is the number of arrivals. Filled bars are significant at the 10% level.
Source: Agarwal et al. (2026), Figure 6.

Supervision timed to the money

For donors trying to protect shrinking budgets, the first implication is timing. The signal appears in the month a tranche lands, a window that ex ante appraisal and ex post audit both miss, which argues for staged tranche release, transaction monitoring timed to arrival, and scrutiny of the exchange on- and off-ramps where laundered funds must touch the regulated financial system. The second is that procurement-heavy and distributive aid call for different instruments: beneficial-ownership disclosure for contractors and contract-level audits in Transport; payment integrity and biometric beneficiary reconciliation of the kind Muralidharan et al. (2016) show to be effective in Social Protection. The third is that enforcement aimed at one concealment technology should anticipate substitution towards another and be judged by what happens to hidden wealth in total. Piercing offshore banking took leaks of millions of documents. Blockchains publish their ledgers continuously, so the forensics we use can run in close to real time. The technology that moves aid out of sight now generates the public record by which it can be found.

Source : VOXeu