What the options market expects, worked out live from free data: Deribit's bitcoin and ether options and futures, the US Treasury's yields and the New York Fed's overnight dollar rate.
Prices and rates are real; the portfolios are made up. Each panel says what it shows. New to options? The short primer on the About page explains calls, puts, strikes and implied volatility in a few paragraphs. The same workers would run on a customer's own share or currency options.
Each dot is one option expiring on the chosen date: its strike across, and the price swing it implies (its implied volatility, as a yearly percentage) up. The grey bars span the swings implied by the best bid and ask, what one could actually sell and buy at; the line is a smooth curve fitted through the dots; the dashed line is the forward, the market's expected price on that date. The curve is usually lowest near the forward and rises on both sides, because options that only pay off after a large move cost extra. One worker follows each date, and recalculates the whole curve whenever any of these prices changes.
The swing the market expects at the forward, for each expiry date, against the time to that date. A rising line means traders expect rougher markets further out; a falling one, that they expect calm to return. The dashed line is DVOL, Deribit's index of the swing expected over the next 30 days, which this worker checks against its own 30-day figure. Dates less than two days away are left out: they have too few options and too little time left to measure reliably.
A future is a contract to buy or sell the coin at a price fixed today for delivery on a later date. When it costs more than the coin does today, someone who buys the coin and sells the future locks in the difference: shown here as a yearly rate, so dates far apart compare fairly. Next to it is what US dollars earn over the same time, from the US Treasury's yields; the spread is the difference, in basis points (hundredths of a percent). The perpetual never expires; its buyers and sellers pay each other a funding rate every 8 hours instead, compared here with SOFR, the overnight dollar rate.
| Future | Days to delivery | Price | Premium, a year | Dollars earn, a year | Difference |
|---|
How the value of the portfolio's crypto and options reacts to change, in the client's currency. Delta: the gain or loss if the coin's price moves 1%. Gamma: how much delta itself changes on a further 1% move; options make exposure grow or shrink as prices move. Vega: the gain or loss if expected swings rise by one percentage point. Theta: the value lost in a day as the options' remaining time runs down. Below them, the table shows the gain or loss if bitcoin and ether moved by up to 20% either way while expected swings changed by up to 10 points; each cell recalculates every option from scratch, because the four measures above only hold for small moves.
| Bitcoin and ether move | Swings 10 points lower | 5 lower | Unchanged | 5 higher | 10 higher |
|---|
Two workers answer the same question in very different ways: how much could this portfolio lose in a day when the coin's price and its expected swings move together? Neither reads the other's answer; a third worker compares them.
The last ten minutes. The network's figure moves with every tick; the Monte Carlo's follows when its run finishes, so the dotted line moves first and the solid line steps after it. Below, how far apart the two were each time the check compared them, against its bands, and how far apart the two fitted surfaces were.
| Percentile of the day's gain or loss | Exact | Network |
|---|
How the network sees the book. The network sees the book as these sensitivities, per coin and per expiry, in the same units as the greeks above (delta and vanna per 1% move, gamma per 1% move squared, vega per vol point, volga per vol point squared). Its delta, gamma and vanna follow each expiry's smile as the price moves, as the model's smile does, while the greeks above hold each option's vol fixed, so the two differ for the same book. The Monte Carlo revalues every position on every path.
Heston has no jumps, so one-day tails on a jumpy asset are too thin. This is a demonstration of a neural-network approximation monitored against the model it stands in for, not a crypto risk model.