Can Two Banks Finance the Same Commodity?

Cross-lender collateral visibility is the missing layer of digital trade finance

In 2014, the same stacks of copper and aluminium at one Chinese port were allegedly pledged to bank after bank. In 2020, a Singapore oil trader that owed about US$3.5 billion to 23 banks turned out to have far less inventory than the loans assumed. Different continents, different commodities, same structural question: what stops two banks from financing the same barrel, tonne or container?

The uncomfortable answer is that, for most of the history of commodity finance, nothing structural did. Each bank checked its own file. Nobody could check the other banks’ files. This article takes that question apart, shows where the gap sits in the technology stack, and sets out what the digitalisation of trade can and cannot fix.

Short answer

Yes. Today, two or more banks can finance the same economic commodity without either knowing, and not only through fraud. The commodity is a physical, often fungible lot; each bank holds a claim on a document or a data record about it; and nothing links those claims across lenders.

Digitising documents is necessary but not sufficient. A durable fix needs five things together: lot identity, controlled (singular) rights records, a cross-lender claim registry, lifecycle events, and governance that settles priority, liability and data protection.

Each bank’s controls are locally sound. The failure sits in the space between the banks.

1. What does “the same economic commodity” actually mean?

The phrase hides a lot of structure. A commodity financing touches five different layers, and the “same” commodity can be the same at one layer and different at another. Most failures come from confusing them.

Two banks can therefore sit on the “same” commodity in several ways. Some are fraudulent; many are legitimate. A registry that cannot tell them apart will either miss the fraud or drown banks in false alarms.

A taxonomy of multi-bank exposure to one commodity

2. The evidence: this is not a theoretical risk

Qingdao, 2014: one lot, many receipts

In mid-2014 Chinese police investigated Decheng Mining and its parent, Dezheng Resources, suspected of using duplicated warehouse receipts to raise several loans against the same metal at Qingdao and nearby Penglai. Reuters reported that by late May banks and traders checking their Decheng-linked stock found single cargoes that appeared to have been used for multiple financings. Accounts of the scale differ: state media cited by AFP reported more than US$2.5 billion borrowed from 18 Chinese banks, while a later court ruling reported by GTR put the proven fraud at RMB 3.6 billion (about US$520 million) across 13 banks. Foreign lenders, including Standard Chartered and Citi, were also drawn in.

The technical lesson sits in a warehouse source’s comment to Reuters: a receipt is a document of entitlement, yet the market treated it as proof of ownership. The market behaved as if the document were the asset.

Singapore, 2019–2020: a cluster of failures

Hin Leong collapsed in April 2020 owing around US$3.5 billion to 23 banks, with about 273 outstanding LCs. Its interim managers found inventory of US$212 million against US$646 million owed to banks under inventory facilities, and about 60 LCs worth US$1.5 billion tied to buy-and-sell transactions in the same cargo. Its founder admitted concealing roughly US$800 million of trading losses and selling oil pledged as collateral. HSBC separately alleged that ZenRock used the same cargo for several credit facilities, and ING told a court that Agritrade obtained multiple financing on the same cargoes with duplicate documents. Bloomberg tallied more than US$9 billion of potential losses for lenders across the cluster.

Sources: Bloomberg tally of 2019–2020 Singapore cases; Trafigura statements and press coverage of the 2023 nickel claim. Figures are reported estimates on different bases and are not additive.

Trafigura, 2022–2023: the other half of the problem

Trafigura’s nickel case is a different species. Trafigura says it paid for containerised nickel that turned out to hold little or no nickel, and estimated losses of about US$577 million, later described as above US$600 million. Of the 1,104 containers involved, it had inspected at least 156 by February 2023 without finding nickel. The alleged scheme unravelled when Citi stopped supporting the trades on 27 October 2022 and inspections followed. Gupta’s side contests Trafigura’s account; the trial is a dispute about knowledge and complicity, not about whether the cargoes were nickel.

Why include it?

Because it separates two risks that get mixed up. Double collateral (one asset, many claims) is a visibility problem. Ghost collateral (no asset, or the wrong asset) is a verification problem. A cross-lender registry addresses the first; only physical truth addresses the second. A serious architecture needs both.

3. Why it is structurally possible: six gaps

The arithmetic of multiple financing

Take an illustrative lot: 10,000 tonnes of aluminium at US$2,600 per tonne, worth US$26.0 million. A lender advancing 85 percent lends US$22.1 million against it. One lender is covered 1.18 times. Add lenders and cover collapses, because each assumes it holds the whole lot.

The formal condition is simple. For a lot l with value V(l,t) at time t, lender i advances A(i) against it. Safe financing needs the sum of A(i) to stay at or below V(l,t) after the haircut. No lender can test that inequality today, because it can observe only its own A(i). The registry’s job is to make the sum observable without revealing each term.

4. Where the windows open: the collateral lifecycle

A lot is not exposed to duplicate claims at one moment. It is exposed at specific transitions. Mapping them shows what a digital control must do at each point.

5. Does digitalisation fix it? Only if you digitise the right layer

The industry’s instinct is that electronic documents will end duplication. That is only partly true, and in one respect the opposite can be true.

1. A copy is free. A scanned or PDF receipt multiplies at zero cost. Digitising the picture of a document makes duplicate pledging easier, not harder.
2. Singularity comes from control, not format. UNCITRAL’s Model Law on Electronic Transferable Records (MLETR, 2017) gives an electronic record functional equivalence with a paper transferable document, provided a reliable method identifies the record, keeps it under exclusive control and protects its integrity. The UK Electronic Trade Documents Act 2023, in force since September 2023, lets an electronic trade document such as a bill of lading be possessed in law. Both work at the level of the single document.
3. Many collateral records are not title documents. If the underlying receipt only evidences entitlement, a singular electronic record preserves the ambiguity rather than resolving it. The rights layer has to be defined, not just digitised.
4. Fungible goods need lot identity. Document singularity says nothing about whether two different documents describe the same tonnes.
5. Cross-lender visibility is a network property. Even perfectly singular documents leave open whether the same cargo is also backing an LC at another bank through a different document. Only a shared view across lenders answers that.

A reference model, not a product. Layers L0 to L2 make claims real and singular; L3 makes them visible across lenders; governance makes the result legally usable.

Identity is commodity-specific

6. Designing the cross-lender registry

The design problem is to answer one question, “is this already financed elsewhere?”, without letting banks read each other’s books. The workflow below is the pattern behind industry registries such as Singapore’s.

Three technical details that decide whether it works

1. Keyed fingerprints, not plain hashes. Trade fields such as quantity, dates and commodity have low entropy; a plain hash can be guessed by brute force. Matching should use keyed hashing (HMAC) under a registry-controlled key, or a private set intersection protocol, so a bank learns only that a match exists.

Get Andrea Frosinini’s stories in your inbox

Join Medium for free to get updates from this writer.

Remember me for faster sign in

2. Several keys, with tolerance. An exact hash on one composite key misses duplicates when one lender records 9,998 tonnes and another 10,000 or writes the vessel name differently. Practical registries use multiple fingerprints over different field subsets (for example, document reference; counterparties plus commodity plus date window; vessel plus quantity band) and score the overlap.

3. Atomic register or conflict. If a bank queries first and registers afterwards, two banks acting at the same moment both see “no match” and both lend. The registry must evaluate and record in a single serialised operation.

Minimum data model for a cross-lender claim record

7. What has actually been built: Singapore’s Trade Finance Registry

The clearest live test of the concept is the Trade Finance Registry (TFR) in Singapore, created in direct response to the 2020 failures.

Sources: ABS media release (23 June 2023), ABS TFR page, MonetaGo and Finextra announcements, DBS and Standard Chartered 2020 release coverage, southeastasiainfra.com on Bahrain.

What the TFR does and does not close

8. Architecture options compared

The patterns are complements, not rivals. A credible stack uses E or D at the physical edge, B or C for rights, and A across lenders.

9. What changes inside a bank

Metrics that show the system is working

10. Open research questions

1. Lot identity after blending. How should a canonical identifier survive mixing, refining and re-packing, with quantity conservation enforced?
2. Chain-aware matching. How can a registry distinguish a legitimate sequence of sales of the same cargo from a duplicate claim, without drowning banks in false positives?
3. Legal effect of a registry hit. Is a registry entry evidence of priority, notice, or merely a warning? That varies by jurisdiction and needs harmonisation.
4. The waterbed effect. If regulated banks close the gap, does duplicate financing migrate to non-participants such as funds, private credit and trading houses’ own lines?
5. Liability. Who bears loss when a registry misses a duplicate, or a bank files a false registration?
6. Privacy-enhancing technology at scale. Can private set intersection or secure multiparty computation meet latency and competition-law constraints in production?
7. Registry and control records. When an electronic bill of lading or receipt is already a controlled record, which system is the source of truth for encumbrance?
8. Incentives. Who pays, who benefits, and what stops banks free-riding on a network that only works with near-universal participation?

Conclusion: from “do I trust this document?” to “who else can see this asset?”

Trade finance has spent decades improving the answer to a bilateral question: is this document genuine, and is this client reliable? Duplicate financing is a multilateral question: what else is attached to the thing I am lending against? No amount of bilateral diligence answers it.

That is why the digitalisation of trade matters most at the seams between institutions. Electronic records that cannot be copied, lot identities that survive the supply chain, event streams that tell lenders when collateral changes, and registries that match claims without exposing books are the plumbing that lets one bank see what another has done.

The Singapore experience shows the pattern is buildable: about three years separate the 2020 failures from the 2023 launch. It also shows the remaining work: coverage beyond banks, legal effect across borders, and physical verification that no registry can replace.

A question for the community

If you work in trade finance, collateral management, legal or standards: which layer would you fix first, lot identity, controlled rights records or cross-lender matching? And who should own the registry: banks, regulators or neutral operators?

Sources and notes

Case figures are reported estimates from press and court-related reporting; they use different bases and should not be added together. Illustrative numbers in the arithmetic section are a model, not market data. Technical design patterns in sections 6 to 9 are the author’s analysis. This article is not legal advice.

1. Reuters via IOL/Business Report, China’s vanishing metals hit confidence (June 2014). https://iol.co.za/business-report/companies/2014-06-12-chinas-vanishing-metals-hit-confidence/
2. Reuters via AJOT, China port fraud probe prompted by corruption inquiry (June 2014). https://www.ajot.com/news/china-port-fraud-probe-prompted-by-corruption-inquiry
3. AFP via Business Standard, China group used same metal stocks to borrow USD 2.5 bn (June 2014). https://www.business-standard.com/amp/article/pti-stories/china-group-used-same-metal-stocks-to-borrow-usd-2-5-bn-114061801242_1.html
4. GTR, Qingdao fraud probe ends with jail term. https://www.gtreview.com/news/asia/qingdao-fraud-probe-ends-with-jail-term/
5. Bloomberg via Daily Maverick, After $9 billion credit hit, banks seek trade finance overhaul (July 2020). https://www.dailymaverick.co.za/article/2020-07-09-after-9-billion-credit-hit-banks-seek-trade-finance-overhaul/
6. Bloomberg via The Edge Singapore, HSBC and other banks unlikely to recoup Hin Leong loss. https://testing.dev.theedgesingapore.com/news/oil-gas/hsbc-and-other-banks-unlikely-recoup-hin-leong-loss
7. Reuters via Gulf Today, Banks to tighten norms for Singapore commodity sector (July 2020). https://gulftoday.ae/business/2020/07/05/banks-to-tighten-norms-for-singapore-commodity-sector
8. Bloomberg via BloombergQuint, HSBC targets Singapore oil trader ZenRock. https://www.bloombergquint.com/business/hsbc-targets-singapore-oil-trader-as-credit-concerns-escalate
9. American Banker, How Citigroup helped lift the lid on Trafigura’s nickel nightmare. https://www.americanbanker.com/articles/how-citigroup-helped-lift-the-lid-on-trafiguras-nickel-nightmare
10. GTR, Trafigura’s US$600mn nickel fraud trial against Prateek Gupta begins. https://www.gtreview.com/news/global/trafiguras-us600mn-nickel-fraud-trial-against-prateek-gupta-begins/
11. ABS, Trade Finance Registry launch release (23 June 2023). https://www.abs.org.sg/docs/library/abs-launches-trade-finance-registry-to-counter-trade-finance-fraud_v3.pdf
12. ABS, Trade Finance Registry page. https://abs.org.sg/industry-guidelines/compliance-and-risk-management/trade-finance-registry
13. MonetaGo selected by ABS to deliver the Trade Finance Registry (Business Wire, June 2022). https://www.businesswire.com/news/home/20220615005146/en/MonetaGo-Selected-by-the-Association-of-Banks-in-Singapore-to-Deliver-Trade-Finance-Registry
14. Finextra, Standard Chartered and DBS to set up blockchain-based trade finance registry (2020). https://finextra.com/newsarticle/36694/standard-chartered-and-dbs-to-set-up-blockchain-based-trade-finance-registry
15. Southeast Asia Infra, Bahrain and Singapore advance deployment of cross-border digital public infrastructure. https://southeastasiainfra.com/bahrain-and-singapore-advance-deployment-of-cross-border-digital-public-infrastructure/