eagL / crypto markets

Pricing crypto prediction markets from first principles

Crypto strikes and brackets are the one corner of prediction markets where you can price the contract properly, because the underlying trades continuously and the settlement rule is arithmetic. This is where eagL runs a real model, across 11 assets and 14 Kalshi series.

Why crypto contracts are priceable

A contract that pays out if Bitcoin closes above a strike is a digital option. Its fair probability follows from the distribution of the underlying at expiry, and for short horizons that distribution can be estimated from recent price behavior. This is the same machinery an options desk runs, pointed at event contracts. Political and sports markets have no such structure, which is why eagL scans them but does not pretend to model them.

The model

eagL fits geometric Brownian motion to stored minute candles per asset. Drift is estimated per minute. Variance uses an exponentially weighted moving average with a twenty-period half-life, so the estimate reacts to a volatility regime change in tens of minutes rather than days. Strike and bracket probabilities are read off the implied lognormal. Past a two-hour horizon the model drops drift entirely: at that distance the drift estimate is noise, and noise with a sign is worse than no opinion.

The grade

Every interval the model prices gets scored once it resolves: Brier score, log loss and a ten-bin calibration curve. The scores live in the app and accumulate, so the question "does this model earn its keep" has a running answer instead of a marketing one. The validation regime around it is on the backtesting page.

Model to market

A model probability only matters against an executable price. eagL reads the top eight book levels, computes microprice and depth imbalance, charges Kalshi's fee formula at your projected fill VWAP, and sizes the position by walking the ladder until the next contract stops paying. Where the same event trades on another venue, the cross-venue accounting applies.

Honest limits

GBM assumes lognormal moves and crypto has fat tails; that is precisely why the scoring exists. The model is a disciplined baseline you can audit, not an oracle. Its full specification, including everything it refuses to publish on thin samples, is on the methodology page.