Stablecoins: The Reserve Problem

Your intuition is correct, and it cuts straight to the heart of why stablecoins are structurally fragile. Let’s make the argument precise.

What a Stablecoin Promises

A stablecoin is a token that maintains a peg to a reference asset — almost always the US dollar. You give the issuer $1,youreceive1token.Later,youreturn1tokenandreceive$1 back. The issuer is running a narrow bank: take deposits, hold assets, redeem on demand.

The question is: what assets does the issuer hold?

---

The Three Regimes

Regime 1 — Full Reserve (100% cash)

The issuer holds exactly $1 in cash (or overnight Fed reserves) for every token outstanding. Reserve ratio:

$r=\frac{R}{L}=1$

where $R$ = total reserves (USD), $L$ = total liabilities (tokens outstanding × $1 face value).

This is safe. Any redemption can always be satisfied. The issuer makes money by charging fees on issuance/redemption, not by investing the float. This is the model of a pure currency board — Hong Kong’s HKD peg works exactly this way against USD.

The downside: it’s capital-inefficient and generates no spread income.

Regime 2 — Fractional Reserve Without a Central Bank Backstop

The issuer holds $r<1$ of assets per dollar of liabilities — perhaps 0.85 in US Treasuries, 0.10 in commercial paper, 0.05 in corporate bonds. Total reserve value ≈ $1, but only if all assets can be liquidated at par.

This is the model most large stablecoins actually use.

Regime 3 — Algorithmic (No Collateral)

No fiat reserves at all. The peg is maintained by a token-minting/burning mechanism. Pure confidence game.

This is Terra/Luna. We will watch it die in slow motion below.

---

Why Regime 2 Is Structurally Fragile: The Bank Run Model

Your intuition — a fractional reserve institution without a central bank backstop is weird — is precisely the insight that won Diamond and Dybvig the Nobel Prize in Economics in 2022.

Diamond-Dybvig (1983)

Consider a two-period economy:

• $t=0$: depositors place funds with the institution
• $t=1$: “impatient” depositors need liquidity; they can withdraw early
• $t=2$: assets mature; “patient” depositors collect returns

The institution invests in illiquid long-term assets that yield $R>1$ at $t=2$ if held to maturity, but only $\ell <1$ if liquidated early (fire sale). It promises to pay $c_1\ge 1$ to early withdrawers at $t=1$ and $c_2>c_1$ to late withdrawers at $t=2$.

This structure has two Nash equilibria:

Good equilibrium: Only impatient depositors withdraw at $t=1$. The institution honors all claims. Patient depositors wait and collect $c_2$. Everything works.

Bad equilibrium: Patient depositors, fearing others will run, also withdraw at $t=1$. The institution must liquidate everything at the fire sale price $\ell$. If $\ell \cdot A<c_1\cdot N$ (where $A$ is total assets, $N$ is total depositors), the institution fails before everyone gets paid. First-come, first-served turns rational fear into a self-fulfilling run.

The key result: both equilibria exist simultaneously. Which one materialises depends purely on coordination — on beliefs about what other depositors will do.

What Banks Have That Stablecoin Issuers Don’t

Banks escape the bad equilibrium through three interlocking mechanisms:

1. Deposit insurance (FDIC up to $250k in the US): Patient depositors no longer fear losing their money if they wait — the government will pay. The incentive to run evaporates.

2. Lender of last resort (the Fed’s discount window): Even if a solvent but temporarily illiquid bank faces a run, it can borrow reserves from the Fed against its long-term assets as collateral. No fire sale needed.

3. Regulatory supervision (stress tests, capital requirements, liquidity coverage ratios): Regulators verify that $r$ stays above minimum thresholds and that the asset side is liquid enough to survive a plausible run.

A stablecoin issuer like Tether has none of these. It is operating in the structurally unstable regime: fractional reserves, illiquid assets, no backstop. The good equilibrium holds only as long as everyone believes it holds.

---

Tether: Anatomy of a Fractional Reserve Stablecoin

Tether (USDT) is the largest stablecoin by market cap — over $100billionasofearly2026.Itclaimstomaintain$r = 1$ but the composition of those reserves is what matters.

The Reserve Quality Problem

Not all $1 of reserves is equal. Consider two portfolios:

Asset	Face Value	Liquidation Value Under Stress
Fed overnight reserves	$1.00	$1.00
3-month US T-bill	$1.00	~$0.999
Investment-grade commercial paper	$1.00	~$0.90–0.95
Corporate bonds	$1.00	~$0.70–0.85
Secured loans to crypto firms	$1.00	???

Tether has historically held a significant fraction of its reserves in commercial paper and other non-cash instruments. During normal conditions, $r\approx 1$. Under stress — when everyone tries to redeem simultaneously and Tether must fire-sell assets — the effective $r$ could fall below 1.

The solvency condition during a run is:

$r_{\text{stress}}=\frac{{\sum }_i{V}_i\cdot (1-h_i)}{L}\ge \frac{N_{\text{redeeming}}}{N_{\text{total}}}$

where $h_i$ is the haircut on asset $i$ under distress and $N_{\text{redeeming}}/N_{\text{total}}$ is the fraction of holders simultaneously redeeming. If enough people run at once, even a “fully reserved” stablecoin can break.

Tether has survived multiple stress events — the collapse of Terra/Luna in May 2022, FTX in November 2022 — but experienced brief depegs to ~$0.95 each time. It held. Whether it would hold under a more severe coordinated run is an open question.

The Opacity Problem

Unlike a bank, Tether is not required to publish audited financials in real time. For years it published only “attestations” from a small Cayman Islands accounting firm, not full audits. This opacity makes it impossible for outside observers to verify $r$, which makes the good-equilibrium fragile: rational depositors must rely on faith rather than verified balance sheets.

USDC (Circle) is more transparent — it publishes monthly reserve reports audited by Deloitte, holds primarily cash and short-term US Treasuries, and operates under US money transmitter licenses. Yet even USDC depegged briefly in March 2023 when $3.3 billion of its reserves were trapped in Silicon Valley Bank during its collapse — illustrating that even high quality reserves carry tail risks.

---

Terra/Luna: Regime 3 Collapses in Real Time

If Regime 2 is structurally fragile, Regime 3 (algorithmic, no collateral) is simply broken. The Terra/Luna collapse of May 2022 destroyed ~$40 billion of value in roughly 72 hours. Here is the mechanism.

The Mechanism

Terra (UST) was an algorithmic stablecoin. There was no fiat backing. The peg was maintained by an arbitrage mechanism involving a second token, Luna:

• If UST > $1:burn$1 of Luna → mint 1 UST, sell into market. UST supply rises → price falls back toward $1.
• If UST < $1:burn1UST\to mint$1 of Luna, sell. UST supply falls → price rises back toward $1.

This seems elegant. It works perfectly — as long as Luna has value.

The Reflexivity Death Spiral

The fatal flaw: Luna’s value depends entirely on confidence in UST. When UST began depegging:

Let $P_L$ = Luna price, $S_L$ = Luna supply, $P_{UST}$ = UST price.

When P_{UST} < \1$:

• The protocol burns UST and mints Luna to restore the peg
• More Luna supply → $P_L$ falls
• Falling $P_L$ means each dollar of Luna buys fewer UST → less stabilizing force per unit of minting
• UST confidence falls further → more redemptions → more Luna minting
• Luna hyperinflates, UST collapses to zero

Formally, define the stabilization capacity at Luna price $P_L$ and outstanding UST supply $S_{UST}$:

$\text{Capacity}=P_L\cdot S_L$

This is the total market cap of Luna available to absorb UST redemptions. As the spiral progresses, $P_L\to 0$ exponentially while $S_L\to \infty$, but the product collapses. There is no floor.

This is Soros’s reflexivity applied to monetary systems: the act of defending the peg destroys the instrument used to defend it.

---

The Stablecoin Trilemma

Your intuition captures a version of an impossible trinity specific to stablecoins. A stablecoin cannot simultaneously have all three of:

1. Price stability — reliably redeemable at $1 under all conditions
2. Capital efficiency — not requiring 100% fiat reserves locked up
3. Decentralisation / no central bank backstop — operating outside the regulatory and monetary infrastructure that makes fractional reserve banking safe

Type	Stable?	Capital-efficient?	Decentralised?
Full-reserve (narrow bank)	✓	✗	✗
Fractional-reserve (Tether)	Mostly	✓	✗
Overcollateralised crypto (DAI)	Mostly	✗	✓
Algorithmic (Terra)	✗	✓	✓

No entry gets all three checkmarks. This is not a coincidence — it is a structural result from monetary theory.

DAI: The Overcollateralised Compromise

MakerDAO’s DAI takes a different approach from Tether: instead of fiat collateral at $r=1$, it requires overcollateralisation in crypto assets, typically $r>1.5$ (150% collateral ratio). You lock $150 of ETH to borrow 100 DAI.

This preserves some decentralisation while buying a buffer against collateral price drops. If ETH falls more than 33%, positions get liquidated before DAI becomes undercollateralised. It works — until it doesn’t. In March 2020 (“Black Thursday”), ETH dropped 50% in hours, liquidation auctions broke down, and DAI briefly depegged. MakerDAO has since diversified into USDC and real-world assets as collateral, drifting toward Regime 2.

---

The Regulatory Endgame

Your insight — this is weird — is exactly what regulators have concluded. The US GENIUS Act (2025) and the EU’s MiCA regulation are converging on a simple principle: any stablecoin that looks like a deposit must be regulated like a deposit.

Concretely this means:

• 1:1 reserve requirements in cash or short-term government securities
• Regular third-party audits
• Capital buffers
• Resolution planning

In other words: turn them into narrow banks, or force them to get bank charters (and thus Fed access). The regulatory trajectory is toward collapsing the distinction between stablecoin issuers and regulated money market funds.

This is rational. The alternative — allowing large-scale fractional reserve stablecoin issuers to operate without deposit insurance or a lender of last resort — is recreating the pre-1933 era of private bank notes: periodically stable, catastrophically fragile, and prone to contagion runs.

---

Connection to Your Experience

At Credimi, you built credit scoring and underwriting systems. The credit risk machinery you built — PD estimation, LGD, portfolio concentration limits — was ultimately defending a similar solvency condition: that the asset side of the balance sheet could service the liability side under stress. The difference is that Credimi operated under Bank of Italy supervision with access to ECB liquidity facilities. Tether does not.

At Capital One, the card business is funded through a mix of deposits (FDIC-insured), ABS issuance, and equity. The funding structure is designed precisely so that a loss of confidence in one channel does not cascade into a run on others. Stablecoin issuers have none of that structural redundancy.

---

Questions to Test Understanding

1. Tether claims $r=1$ but holds 10% of reserves in secured loans to crypto firms. During a market stress event, those loans lose 40% of their value. What is the effective reserve ratio if 80% of USDT holders try to redeem simultaneously?

2. Why does deposit insurance eliminate the bad equilibrium in the Diamond-Dybvig model? What condition on the insurance premium is necessary for it to be credible?

3. The GENIUS Act requires 1:1 reserves in cash or T-bills. How does this affect stablecoin issuers’ economics? What business model do they use to generate returns if they cannot invest the float?

4. DAI’s 150% collateral ratio protects against a 33% ETH price drop. Why did this fail in March 2020? What does the failure reveal about the assumptions baked into the collateral ratio?

5. If Terra/Luna had held $10 billion of actual USD reserves (partial backing) alongside the algorithmic mechanism, would the May 2022 collapse have been prevented? Where does the death spiral start even with partial backing?