The Next Trillion Dollars of Stablecoins Will Be Built on Options

On June 1, Vitalik Buterin posted a proposal to rebuild DeFi's synthetic dollars on options instead of debt. As the leading options protocol on RWAs we were naturally excited, and skeptical.

Every synthetic dollar in production today is a loan. You deposit ETH, you borrow dollars against it, and the protocol keeps itself solvent by selling your collateral the moment a price feed says your cushion is too thin. Most of the worst days in DeFi's history trace back to one of those two parts, the forced selling or the price feed.

Vitalik's construction doesn't borrow anything, so there is nothing to liquidate and no feed to defend. What nobody had done was put numbers on it. We pulled the live ETH options surface and ran the strategy through nine years of crashes to find out what holding the thing actually costs.

How a synthetic dollar works today

The dominant design is the collateralized debt position (CDP), and it has barely changed since 2017. Deposit $200 of ETH into MakerDAO, mint $100 of DAI against it. The system stays solvent through one enforcement mechanism: if your collateral ratio falls below a threshold, your ETH is seized and sold, with a liquidation penalty on top, before the debt can go underwater. Aave, Liquity, and every perp venue's margin engine run the same logic with different parameters.

Notice what you paid for that $100. Your $200 of ETH is locked, earning nothing, backing half its value in usable dollars. Liquity v1 pushed the minimum to 110 percent; most systems sit at 150 percent or higher because thinner cushions liquidate constantly. Overcollateralization is the capital cost of the debt design, and the liquidation engine is its operational cost.

And the liquidation engine has a dependency that cannot be relaxed: it must know the price right now, every block, because solvency is enforced in real time. Vitalik's argument is that this dependency, not leverage itself, is the original sin:

Real-time oracles are very hard to make safe. You can only rely on a limited number of actors, who are watching real-time signals in an automated way. You cannot use mechanisms that incorporate any notion of recourse.

The history is on his side. Below are four of the worst blowups in DeFi, and only one of them was really a market crash. The rest were the oracle or the liquidation engine failing on its own.

Black Thursday is the one everyone remembers. ETH fell by half in a day, the network clogged, Maker's oracle stalled, and the keepers who were supposed to buy the seized collateral couldn't get their transactions through, partly because someone stuffed the mempool to crowd them out. The auctions cleared at zero, and Maker handed out ETH for free. The other three didn't even need a crash. Compound liquidated $89 million of loans because DAI printed $1.30 on a single exchange for a few minutes. A trader drained $116 million from Mango by pumping its own mark price. And last October, a tariff headline hit record leverage in a thin weekend book, and the auto-deleveraging spirals turned it into the largest liquidation event in crypto's history.

The industry has spent years patching this instead of replacing it. Maker delays its own price feed by an hour (the OSM) so it has time to catch a bad print. Liquity wired in an oracle fallback. Chainlink built SVR to claw back the MEV that liquidations leak. RAI dropped the dollar peg to take pressure off the mechanism. crvUSD turned the liquidation cliff into a gradual slide, though it still reads a live oracle and still sells collateral into the fall. Terra tried to skip collateral altogether and lost $40 billion in a week.

All of it still leans on the same two things: forced selling, and a price feed that has to be right every second. Vitalik wants to get rid of both.

Vitalik's idea: split the coin instead of borrowing against it

Start with the intuition before any formulas. You and a friend jointly buy 1 ETH for $1,660 and lock it in a box, to be opened on a fixed date. You agree: when the box opens, you get the first $830 of whatever the ETH is worth, your friend gets everything above that.

Think about what each of you now holds. If ETH ends at $2,500, you get your $830 and your friend gets $1,670. If ETH ends at $1,000, you still get your $830 and your friend gets $170. Your claim is worth $830 in every world except one: ETH finishing below $830, where you simply get the whole ETH. Your friend, meanwhile, holds something that moves harder than ETH itself: they paid roughly half the price of the coin for all of its upside.

You are holding a synthetic dollar. Your friend is holding leverage. Nobody borrowed anything, and there is no scenario in which either of you owes more than what is in the box, so there is nothing to liquidate, ever.

That is the entire proposal. Three names before the formulas, all three from our example: the cutoff price ($830) is called the strike, written S. Your claim, the first S dollars of the coin, is the dollar token, which Vitalik calls P. Your friend's claim, everything above the strike, is the leverage token, N. Anyone can split 1 ETH into the pair, and anyone can recombine the pair into 1 ETH, any time before expiry. At expiry an oracle reports the ETH price x, and:

• the dollar token P receives min(1, S/x) ETH, which is worth min(x, S) dollars
• the leverage token N receives max(0, 1 - S/x) ETH, which is worth max(0, x - S) dollars

Options traders will recognize both halves. The leverage token is a call on ETH struck at S. The dollar token is a bond minus an out-of-the-money put, the payoff every covered call writer knows. This is put-call parity minted into two ERC-20s: one share of anything equals a bond plus a call, so splitting the asset manufactures both. Vitalik's own words:

we remove the entire concept of liquidations by making the "base building block" of the system options rather than debt.

The whole thing rests on P + N = 1: the two tokens always add back up to the coin, by construction. That removes liquidations, and it also means the system never needs a price until maturity. Splitting and recombining tokens is just bookkeeping, and anyone who rolls before expiry never touches settlement at all.

That unlocks the oracle designs debt can never use: optimistic oracles with dispute windows, Kleros-arbitrated settlement, prediction markets in front of an expensive recourse oracle. Vitalik notes the construction "literally is 'just' a prediction market", a scalar market of the kind Seer trades today, so synthetics and prediction markets can share one slow, hardened oracle. It is the asset-layer version of his info finance and low-risk DeFi essays.

One more piece of intuition: how do you actually hold a dollar this way? You do not sit at the strike. You hold the dollar token struck far below spot, say half, where its delta is near zero and the token barely notices ETH moving. If ETH grinds down toward your strike, you roll to a lower one, on your own schedule, before it gets close. Compare that to the debt holder's experience of the same crash:

The debt claim is all or nothing: worth par until it breaches the liquidation line, then worth whatever a forced auction returns mid-crash. The dollar token has no line to breach. It reprices smoothly, and even with ETH sitting at the strike it is still a 91-cent claim. How much it loses depends on how far below the strike ETH settles, which is why Vitalik calls the failure mode drift rather than default.

None of this is new. TradFi has run the same construction for decades. Buffer ETFs hold over $70 billion of index exposure built entirely from options, with no margin and no liquidation machinery anywhere in the stack. JPMorgan rolls a $22.7 billion collar on the S&P every quarter (we covered the March roll). Box spread traders build synthetic Treasury bills out of options, billions of notional a month. On-chain, Lien Protocol shipped almost exactly this split in 2020 and still died. Why it died is the last section of this article.

What it costs, on live data

So does it work? It hangs on two things: what the embedded option costs to carry, and what the dollar does in an actual crash. We pulled the Deribit ETH surface on June 11, with spot at $1,660, for the first, and built a nine-year backtest for the second.

The dollar token's only market exposure is the put embedded in it, so the carry question reduces to: what does a far out-of-the-money ETH put cost, by tenor, by distance?

Start with distance, because it is almost the whole cost. At 90 percent moneyness, rolling protection costs 1,900 to 3,700 bps a year, which is hopeless. At 50 percent it is roughly 260 bps a year at the 6-month tenor. At 25 percent it rounds to single digits. The cost falls about an order of magnitude for every 25 points of extra distance, which is exactly why you want to sit deep and roll early. Short-dated and deep is basically free: a 7-day put at half of spot prices at about a tenth of a basis point, below Deribit's minimum tick.

The premium that does exist comes off skew, not off some fat volatility premium. On pull day, 30-day at-the-money IV was 55.6 percent against 58.5 realized, while the 10-delta put wing ran 68 to 75. Whoever ends up short these wings is selling crash insurance, and pricing that well is a real business that has to exist for any of this to scale.

Then there is the decay everyone worries about. The obvious objection to "hold options as your savings account" is theta, and it came up within hours of the post. It has the sign backwards:

The dollar holder did not buy a put. They hold a bond minus a put: they are the seller, and time decay accrues to them, the way it does for every covered call writer. The one paying that decay is the leverage holder, in exchange for the leverage. That side of the split usually gets ignored, so look at it:

This is the cleanest part of the whole design. You get real leverage with none of the machinery that blows perp traders up, no funding to bleed and no level where you get force-closed, and the premium its buyers pay is exactly what funds the dollar side. The two legs subsidize each other, or neither exists, the same matching problem every DeFi option vault was built to solve.

Nine years, five crashes, zero liquidations

Pricing a snapshot is not holding the thing. So we ran the proposal's own strategy, mechanically, through every day from August 2017 to this week: hold one synthetic dollar in the dollar token, strike at half of spot, roll the 30-day maturity at expiry, roll the strike down whenever ETH falls within 1.5x of it. Black-Scholes at mid, r = 0, real Deribit DVOL where it exists, flat bps per roll as the cost knob:

Run the full window at 75 percent vol with free rolls and the synthetic dollar drifts minus 0.7 percent a year. Almost none of that comes from the scheduled rolls: the embedded short put, three standard deviations out, expires worthless nearly every month. It comes from the 20 early roll-downs, each of which sells the old token below par after ETH has fallen toward the strike. On real DVOL (2021 onward), where the elevated 2021-22 vol meant real premium to collect, the dollar carried positive 0.18 percent a year. Costs dominate from there. At 10 bps per roll the drift becomes minus 2.0 percent a year (minus 1.1 on DVOL); at 25 bps, minus 4.0 (minus 3.0); at 50 bps, minus 7.1 (minus 6.1). That is thirteen rolls a year times whatever each roll costs. What you pay to run this is mostly the cost of trading it, not the cost of the options.

Now the crashes, which are the reason anyone would build this in the first place:

Through a 62 percent COVID crash, a 71 percent Luna/3AC unwind, FTX, and two crashes this cycle, the synthetic dollar was never liquidated, because there is nothing to liquidate. It did take real hits. The worst was COVID: on March 12, 2020 ETH closed about 7 percent above the strike, the token marked down 5.4 percent, and the roll-down rule sold it there. On real DVOL, Luna cost 5.7 percent peak to trough and FTX 2.7 percent, because crash vol ran far above 75 percent. The strike-down roll fired 20 times in 8.8 years. The comparison vault, rebalanced with perfect daily discipline, still got liquidated five to eight times, because daily discipline does nothing against a price that gaps down between rebalances. The debt vault's worst case is losing the position to an auction. The option dollar's worst case so far is about six cents, taken on a roll.

The parameters matter more than the proposal suggests. Sitting at 40 percent of spot instead of half cuts early roll-downs from 20 to 6 and the worst dip to 2.4 percent, for slightly less premium. Rolling weekly cuts the worst dip to 1.1 percent, at four times the number of trips through the book. Either way you trade tracking for execution, which is the same problem again.

The missing piece is the venue

So the construction holds up and the options are cheap. The real cost is getting filled, thirteen times a year, in deep out-of-the-money strikes that have no resting bids today, including a handful of times in the middle of a crash. Vitalik flags this himself:

Rebalancing would be more like one-sided market making than like making an instant sell.

On the scheduled roll, a dollar holder is non-toxic, schedule-insensitive flow: they do not care if the fill happens today or Thursday, they just cannot pay 50 bps for it (the backtest prices that path at minus 6 to 7 percent a year). Flow like that is exactly what price-improvement auctions exist to monetize: makers quote uninformed, patient flow far inside the screen because it cannot run them over. The roll itself, sell one strike, buy another, is a two-leg package, which is RFQ territory; resting it on a book leg-by-leg is how you donate the spread twice. We wrote up both mechanisms, and why options need them more than perps, in our RPI piece.

The early roll-down is the opposite kind of flow. It happens in a crash, it is forced by a rule anyone can read, and the dollar holder is buying back a put at the moment puts are most expensive. Nine years of data say that trade, not the monthly roll, is where the losses come from. A venue that can fill it near fair value in a crash is the difference between a six-cent hit and something worse.

History has already run this experiment in both directions. Lien built this exact split in 2020 with no options market structure around it; the legs sat illiquid and the protocol died. The buffer-ETF complex built the same payoff on top of deep, mature options markets and grew to $70 billion. Same payoff, opposite outcomes, and the only thing that differed was the market underneath it. Builders are already moving on the idea: four days after the post, someone had a physically-settled version live on Base, and the problems they ran into were liquidity for the two tokens and slippage on the roll. Not the math. The trading.

That is why we think this proposal matters more for venues than for protocols. DeFi's first generation of stablecoins was built on money markets: Maker, Aave, Compound, lending books with liquidation engines bolted on. If Vitalik is right, and our numbers say the economics check out, the next generation gets built on options exchanges: deep books in the wings, auctions for patient one-sided flow, atomic multi-leg rolls, and portfolio margin that understands a covered position has no liquidation price.