Let's talk ↗
Spec sheet 01 / 06 — Research paperAll research ↗

The Redemption Gap: Why Full Reserves Are Not the Whole Story

A Hong Kong-focused study separates reserve backing from cash available at the redemption deadline. A transparent stress model tests the gap.

Originally published by Panasia Research.

Topic
Stablecoin reserves
Published
2026-09-12
Reading
15 min
Sections
10
References
14
Cover
The Redemption Gap cover: a sandstone portico against grayscale architectureIUM LABSResearch
Cover artwork and image sources
  • Cover: AI-generated editorial illustration · ium Labs Research.
01 Three balances, three different questions — 02 What the public rules establish — 03 Distribution changes the observed route — 04 A fully backed balance sheet can have a cash queue — 05 The price of replenishing cash — 06 The economics of a larger cash sleeve — 07 What an actual test would need to record — 08 What This Changes for a Korea Go-to-Market — 09 Conclusion — 10 Method and Limits —
Spec sheet 02 / 06 — Key takeaways

Key takeaways

  1. 01

    Reserve coverage and timely cash delivery answer different questions. The HKMA guideline distinguishes the reserve pool from the redemption mechanism. [1]

  2. 02

    Hong Kong's guideline already asks licensees to track how much of the reserve can become cash within set timeframes. That is the deadline view modeled here. [1]

  3. 03

    In a hypothetical HKD 100m arrangement with HKD 102m of reserves but HKD 10m accessible cash, a 20% redemption wave leaves a HKD 10m immediate cash gap.

  4. 04

    March 2023 showed the gap in public: about 8% of USDC's reserves were stuck at a failed bank over a weekend, and USDC traded as low as 86 cents. [10]

  5. 05

    Issuer plans describe distribution designs, not measured redemption outcomes. A useful test records the full interval from request to usable bank funds.

Spec sheet 03 / 06 — The paperReading 15 min
10Sections

An issuer can hold assets worth more than all outstanding tokens and still need time to deliver bank money. For a company using stablecoins to meet an invoice, the relevant balance is the cash accessible before the deadline. This paper separates three questions that are often collapsed into a single claim of backing: whether assets cover liabilities, whether cash can be mobilized on time, and whether this holder has an operational path to receive it.

Our contribution is a reproducible liquidity model, informed by Hong Kong's supervisory framework and two dated issuer announcements. In a hypothetical HKD 100 million arrangement with HKD 102 million of reserves, a 20% redemption wave can leave a HKD 10 million immediate cash gap when only HKD 10 million is accessible. The arrangement is overcollateralized in the example; the payment queue still matters. This is a conditional result, not an assessment of a named issuer's reserves or a prediction of a run.

01

Three balances, three different questions

Define outstanding tokens at their reference value as L, total reserve market value as R, and cash deliverable through the redemption route before a chosen deadline as C. Reserve coverage is R/L. It compares asset value with a liability. Immediate cash coverage is C/Q, where Q is the value of valid requests due by that deadline. These ratios have different denominators because they answer different questions.

The distinction is useful even without a crisis. A business that buys a token on a secondary venue may not use the same withdrawal process as a distributor with a direct issuer account. A token credited to a wallet and money credited to a bank account are different endpoints. A research report that records only reserve composition leaves the second endpoint unmeasured.

Access is part of the route. A 2024 Federal Reserve note found that only Circle's direct customers, cleared through an application process, can use the primary market for USDC; most retail users buy from intermediaries. [9]

We define the redemption gap as the amount of a specified request cohort that cannot be funded before its specified deadline using accessible cash and eligible incoming liquidity. The term is analytical. It is not a statutory ratio or a claim that a product has breached an obligation. A deadline must be chosen first: an internal treasury deadline, a contractual promise and a supervisory processing expectation are not interchangeable.

Hong Kong's guideline draws the same line. It asks licensees to monitor liquidity indicators "such as cash ratio and proportion of reserve assets that could be converted to cash for meeting redemption demand within certain timeframes." [1] In this paper's notation, that is C measured against a deadline.

02

What the public rules establish

Hong Kong's August 2025 supervisory guideline requires reserve market value at least equal to outstanding par value and expects an appropriate buffer. It distinguishes reserve management from redemption arrangements. Paragraph 3.3.3 generally expects valid requests to be processed within one business day after the day of receipt, unless otherwise approved. The qualifications concerning validity, reasonable fees and prior regulatory consent matter; the wording is not an unconditional promise of instant bank credit. Paragraph 3.3.1 gives holders a right to redeem at par, a right to direct the disposal of the reserve pool pro rata and a right to claim any shortfall; paragraph 3.3.2 expects an independent legal opinion that those rights exist. [1]

The guideline also limits reserves to cash, deposits of up to three months, government-type debt securities with up to one year to maturity, overnight reverse repos and dedicated funds, and asks for stress tests at least quarterly. [1]

The FSB's July 2023 recommendations similarly treat governance, risk management, recovery and redemption rights as separate components of a stablecoin arrangement. They call for robust legal claims and timely redemption, with par redemption into fiat for single-currency arrangements. These are recommendations to authorities, not a substitute for a product contract or a global license. [2]

Singapore provides a useful historical comparator. MAS's August 2023 framework announcement specifies a five-business-day par-redemption requirement for the single-currency stablecoins within its defined scope. We use that announcement to show why a numerical processing window must be read with its jurisdiction, scope and date. This paper does not treat the announcement as verification of the current licensing or operating status of any Singapore product. [7]

The US GENIUS Act of July 18, 2025 requires reserves on an at least 1 to 1 basis, admits Treasury bills with a remaining maturity of 93 days or less, and requires a public redemption policy with procedures for "timely redemption." [11] In the passages we read, it sets no number of days.

Our inference is narrower than a jurisdictional ranking: comparable evidence needs the same start and end events. A request submitted, a request accepted as valid, a token returned to the issuer and bank funds becoming usable can occur at different times. Ranking products by one headline number before aligning those events would measure documentation choices as much as payment performance.

03

Distribution changes the observed route

On 10 April 2026, HSBC announced its Hong Kong stablecoin issuer license and a planned launch in the second half of 2026. Its selected release describes intended integration into PayMe and the HSBC HK App, with payments and tokenized investments among the planned uses, and says each coin will be fully backed by high-quality, liquid assets held in segregated accounts. The statement is evidence of the license and product plan at that date. It does not supply a series of completed redemptions or demonstrate the final launch status as of this paper. [4]

Standard Chartered's announcement on the same date describes Anchorpoint's license and a planned B2B2C distribution model for HKDAP, using selected authorized distributors. Anchorpoint, a joint venture of SCBHK, HKT and Animoca Brands, targets a phased approach from the second quarter of 2026. This is a different distribution design from an application-centered proposition. The release does not disclose each distributor's redemption contract, cash inventory or service-level results. [5] The HKMA's release says both licenses took effect that day. [8]

A June 2026 government reply provides an important intermediate observation. It states that the two licensed issuers needed to complete technology testing, risk arrangements and other preparations before official business launch. Its list of preparations includes redemption arrangements. The reply also counts 36 applications in the initial period and expects launch "as early as the middle of this year." This prevents a license date from being silently recoded as the date of full service availability. We retain the dated observations rather than infer what happened afterwards. [3]

Dated observationWhat it supportsWhat it does not establish
HSBC, 10 April 2026 [4]License announcement and intended application channelsCompleted customer redemptions or measured cash-arrival times
Anchorpoint, 10 April 2026 [5]License announcement and intended distributor modelUniform redemption terms across future distributors
HKMA, 10 April 2026 [8]Licenses effective that day; launch intended "in the coming few months"A launch date for either product
Government reply, 10 June 2026 [3]Pre-launch preparations described by the authoritiesA later launch date or subsequent transaction outcomes
Dated observations on Hong Kong's first two licensed issuers (Sources: HSBC, Standard Chartered and HKMA, April 10, 2026; HKSAR Government, June 10, 2026)

For an empirical study, distribution is a variable to measure rather than a reason to declare one design superior. An integrated bank application might simplify account matching. A distributor network might widen access. Either could also change where operational work sits. The question is whether the end customer can identify the entity accepting the request, the asset owed, the fees, the exceptions and the completion timestamp.

04

A fully backed balance sheet can have a cash queue

Consider a deliberately simplified arrangement. Tokens outstanding are HKD 100 million. Reserves have a market value of HKD 102 million: HKD 10 million is immediately accessible cash and HKD 92 million is securities that cannot produce settled cash before the chosen deadline. No new issuance receipts, credit lines or securities-sale proceeds arrive inside that window. The example assumes no default and initially no price change.

For a redemption fraction q of outstanding tokens, requests equal Q = q × L. Immediate funding equals the lesser of Q and C. The uncovered amount is max(Q − C, 0). Opening reserve coverage is 102%. Request coverage below is the share paid immediately, min(C, Q) / Q, capped at 100%.

Redemption requestsRequest valueImmediately paidCash gap at deadlineRequest coverage
5% of tokensHKD 5mHKD 5mHKD 0m100%
10% of tokensHKD 10mHKD 10mHKD 0m100%
20% of tokensHKD 20mHKD 10mHKD 10m50%
40% of tokensHKD 40mHKD 10mHKD 30m25%
Immediate cash coverage by redemption wave (ium Labs scenario, HKD)

These rows are scenario arithmetic, not observed issuer data. The securities remain assets even when their proceeds arrive later. Calling the cash gap an asset shortfall would therefore misstate the example. Equally, reporting only the 102% reserve ratio would omit the deadline. Both statements can be true: the balance sheet has a positive buffer, and some requests cannot be funded inside the selected window without another source of liquidity.

The model also exposes what would falsify the apparent bottleneck. If HKD 15 million of reliable incoming cash arrives before the deadline, the 20% request cohort is fully funded. If a committed facility can be drawn in time, accessible liquidity rises. If the contractual deadline is later than securities settlement, the same opening balance sheet can meet that later deadline. Those are changes to assumptions, not contradictions of the calculation.

Timing should be modeled as a schedule

For each time interval t, use opening cash C(t), eligible inflows I(t), and requests due Q(t). The paid amount is limited by available cash; unfilled requests carry forward. Once a security is sold, record the proceeds at settlement rather than at trade execution. Once a facility is drawn, record both the cash and the new repayment obligation. A reconciliation that increases available cash without recognizing the corresponding liability overstates the improvement.

Do not count new minting receipts as permanent rescue capacity. If new tokens are issued against those receipts, liabilities rise as well. Net inflows can help a short window, but their absence should be tested independently. A robust exercise should also distinguish committed facilities from indicative quotations and legally segregated cash from cash held elsewhere in an organization.

A public example: March 2023

The distinction is not only theoretical. When Silicon Valley Bank failed on Friday, March 10, 2023, Circle could not withdraw $3.3 billion of USDC reserves, around 8% of the total. It shut its primary market over the weekend. On Sunday, regulators said all SVB depositors would be fully protected. [10]

86 cents: USDC's lowest secondary-market price in March 2023 (Source: Federal Reserve FEDS Notes, December 17, 2025)

An earlier Fed note had observed that USDC issuance and redemption "is constrained by the working hours of the U.S. banking system." [9] The episode touched both axes of this paper: part of the cash could not move, and until regulators protected the uninsured deposits, Circle "faced the possibility of a significant shortfall." [10]

05

The price of replenishing cash

The next question is how much reserve value is consumed when the cash gap is filled. Suppose the arrangement must raise HKD 10 million by selling securities, and net proceeds are 99.5% of their pre-sale reference value. Securities worth 10 / 0.995 = HKD 10.050251 million must be sold. The incremental loss is approximately HKD 50,251. This is an assumed execution discount, not a market quote.

After paying HKD 20 million in total redemptions, reserves are about HKD 81.949749 million and tokens outstanding are HKD 80 million. The remaining buffer is about HKD 1.949749 million. The accounting identity is important: paying a redemption reduces both assets and token liabilities, while the execution loss reduces the buffer. Subtracting the entire redemption from the buffer would be incorrect.

Assumed execution discountSecurities sold to raise HKD 10mLoss against reference valueRemaining absolute buffer
0.0%HKD 10.000000mHKD 0HKD 2.000000m
0.5%HKD 10.050251mHKD 50,251HKD 1.949749m
1.0%HKD 10.101010mHKD 101,010HKD 1.898990m
2.0%HKD 10.204082mHKD 204,082HKD 1.795918m
Cost of raising HKD 10m by selling securities at an assumed discount (ium Labs scenario, HKD)

The starting value must be defined consistently. If reserve market value already incorporates the same execution discount, applying it again would double-count the loss. The Hong Kong guideline, for one, expects valuation at the bid price where applicable. [1] Here the discount represents an additional cost relative to the opening reference valuation. The calculation assumes the remaining securities retain their reference value; a portfolio-wide price shock would require revaluing them too.

This creates two independent stress axes: time to accessible cash and deterioration in realizable value. A fast sale at a material discount may solve the first while worsening the second. A high-quality portfolio with slow operational release may show the reverse. Combining both axes produces a more informative test than a single reserve-percentage headline. The BIS sees in stablecoins' growth "a tail risk of fire sales." [6]

06

The economics of a larger cash sleeve

Holding more immediately accessible cash reduces the modeled gap but can carry an opportunity cost. If an extra HKD 10 million is held in cash earning 1% rather than assets earning 3%, the assumed annual income difference is HKD 200,000. Relative to HKD 100 million of outstanding tokens, that is 20 basis points per year. It is a cost of these selected assumptions, not evidence that any issuer earns those rates or should choose that allocation.

A useful decision model compares that annual cost with expected benefits: fewer delayed requests, lower emergency funding costs, less forced selling and operational resilience. These benefits cannot be assigned arbitrary monetary values and then presented as measured savings. If event frequencies and loss severity are unknown, report a break-even condition instead. For example, HKD 200,000 of annual avoided costs would offset the assumed income difference; the paper does not estimate the probability of achieving it.

The wider monetary context also matters. The BIS's 2025 framework places tokenized central bank reserves, tokenized commercial bank money and tokenized government bonds on a unified ledger, and calls asset-backed stablecoins "akin to digital bearer instruments." Moving a record onto a programmable platform does not make these different liabilities economically identical. We use that distinction to motivate identifying the final asset owed to the holder, not to rank the resilience of the two issuer plans above. [6] Our companion paper on tokenized bonds applies the same funding-time arithmetic to settlement.

07

What an actual test would need to record

A useful dataset would begin with every request in a defined observation period, including cancelled, rejected and delayed requests. Record submission, validation, token receipt, payout instruction and usable-bank-credit times, along with reason codes and fees. A median computed only from successful automated requests would understate the experience of users whose cases enter manual review.

Pair that request log with a time-stamped liquidity schedule. It should identify cash by custodian and currency, security trades and their settlement dates, committed facilities and draw conditions, and operational outages. The two records allow an analyst to distinguish a cash shortage from an identity-check delay, incorrect bank details or a payment-system interruption. Each has a different remedy and should not be compressed into a single blockchain latency number.

Compare like with like. A customer who is already onboarded should be separated from a first-time holder. Weekday requests should not be compared unadjusted with weekend requests. Large institutional requests may use different paths from small app withdrawals. Publish cohort sizes, percentile completion times and the share still unresolved at the observation cutoff. This is a proposed study design; no such issuer-level dataset was obtained for this paper.

The central hypothesis would weaken if deadline-specific accessible liquidity added no explanatory value after controlling for request validity and operating interruptions. It would strengthen if cash mobilization consistently explained the tail of completion times. Either outcome would be more useful than treating asset backing as a complete measure of customer liquidity.

08

What This Changes for a Korea Go-to-Market

Korea was still drafting its rules in mid-2026: the July 14, 2026 growth strategy put a stablecoin regime, including won coins, into second-half legislation. [14] The Bank of Korea's June 2025 Financial Stability Report warned that lost trust in reserves could turn de-pegging and mass redemption requests into a "coin run," with no deposit insurance behind the coins. [12] In February 2026 the Democratic Party's task force was reported to plan reserves of "100%+α" in deposits or government bonds held with outside institutions, under a headline arguing that 100% is not enough without continuous verification. [13] Our August paper, Korea's Framework Act Is Still in Subcommittee, tracks the bills.

  • Pitching Korean partners. Show the cash route as well as the reserve report: who accepts redemption, what is paid, the fees, the hours and the completion time.
  • Assessing a coin for Korean users. Ask what share of the reserve turns into cash within one business day. [1]
  • Campaign copy. Do not call a coin instantly redeemable. Stated windows run from one business day in Hong Kong to five in Singapore. [1] [7] The payout side of the same cash gate is in Stablecoin Payments in Asia.

None of this is legal advice; a Korean lawyer should review any redemption promise made to Korean users.

09

Conclusion

The defensible finding is conditional and operational: full asset backing and timely cash delivery are distinct tests. A buyer should be able to follow both the claim and the cash route. A future empirical paper should measure the entire redemption interval, rather than use reserve coverage or token-transfer speed as a proxy for the endpoint that the holder actually needs.

10

Method and Limits

This is a public-source analytical paper with an explicit hypothetical balance sheet. Source selection is purposive: supervisory texts, a US statute, dated issuer plans, Federal Reserve and BIS analysis and Korean reporting. The fourteen references are not fourteen independent empirical observations. The two firms are research subjects; the paper implies no relationship with or endorsement of ium Labs.

The calculations use unrounded values. They exclude interest accrual during the short stress window, taxes, bank failure, FX mismatch, operational capacity limits and feedback from token-market prices. Adding those effects could materially change results. No customer transactions, internal reserve records, legal opinions or confidential contracts were reviewed. Regulatory passages provide dated context rather than a product-specific legal conclusion.

The paper is written as of September 12, 2026 and uses only material published on or before that date. The Bank of Korea report and the Korean task force plan are cited through press reports, and the USDC figures through Federal Reserve staff notes. Still uncertain: whether either Hong Kong issuer had launched by the cut-off, and the size of the redemption gap in any real product, since no issuer-level redemption log was available.

Spec sheet 04 / 06 — References

References

14Sources

[1]Hong Kong Monetary Authority. 2025-08. Guideline on Supervision of Licensed Stablecoin Issuers. hkma.gov.hk.

[2]Financial Stability Board. 2023-07-17. High-level Recommendations for the Regulation, Supervision and Oversight of Global Stablecoin Arrangements: Final report. fsb.org.

[3]HKSAR Government. 2026-06-10. LCQ6: Development and regulation of stablecoins. info.gov.hk.

[4]HSBC. 2026-04-10. HSBC welcomes HKMA's grant of a Hong Kong stablecoin issuer licence. about.hsbc.com.hk.

[5]Standard Chartered. 2026-04-10. Standard Chartered-backed Anchorpoint granted Stablecoin Issuer Licence by the Hong Kong Monetary Authority. sc.com.

[6]Bank for International Settlements. 2025-06. The next-generation monetary and financial system. bis.org.

[7]Monetary Authority of Singapore. 2023-08-15. MAS Finalises Stablecoin Regulatory Framework. sgpc.gov.sg.

[8]Hong Kong Monetary Authority. 2026-04-10. Granting of stablecoin issuer licences. hkma.gov.hk.

[9]Board of Governors of the Federal Reserve System (FEDS Notes). 2024-02-23. Primary and Secondary Markets for Stablecoins. federalreserve.gov.

[10]Board of Governors of the Federal Reserve System (FEDS Notes). 2025-12-17. In the Shadow of Bank Runs: Lessons from the Silicon Valley Bank Failure and Its Impact on Stablecoins. federalreserve.gov.

[11]U.S. Government Publishing Office (govinfo). 2025-07-18. Guiding and Establishing National Innovation for U.S. Stablecoins Act (GENIUS Act), Public Law 119-27. govinfo.gov.

[12]Blockmedia (Yonhap). 2025-06-25. "스테이블코인 확산, 금융시장에 위험 요인"…한은·BIS 경고(종합2보) (Bank of Korea Financial Stability Report, first half of 2025). blockmedia.co.kr.

[13]Asia Business Daily (아시아경제). 2026-02-22. [스테이블코인 감독]②준비자산 100% 적립으론 부족, 핵심은 상시 검증…지갑 분리로 '제2 빗썸' 막아야. view.asiae.co.kr.

[14]ET News (전자신문). 2026-07-14. 비트코인 현물 ETF, 국내 상장 길 열린다…정부 하반기 도입 추진 (the 2026 second-half economic growth strategy). etnews.com.

Spec sheet 05 / 06 — Continue readingAll research ↗

Continue reading