About & method

BTC Event Oracle is an independent, non-commercial, educational project. It receives no compensation, promotes no products, and gives no personalized advice — it publishes one general, systematic forecast on a fixed hourly schedule. It is not financial advice.

How the forecast is made

Each hour a quantitative baseline estimates Bitcoin's volatility with a GJR-GARCH model (an asymmetric volatility model that out-performs simpler methods for crypto) and builds a central estimate, a confidence range, and a probability of being higher, for 1-week, 1-month, and 1-year horizons. A volatility-regime check widens the published range during turbulent periods, when any model is less reliable. Then Claude (an LLM) reads condensed world-event signals — the Fear & Greed index, perpetual funding, open interest, Deribit implied volatility, and news — and applies a small, hard-capped adjustment. If the model is unavailable, the site falls back to the untouched baseline.

How it's graded

Every forecast is scored when it comes due against a random-walk benchmark ("tomorrow = today") using proper scoring rules — Brier for direction, CRPS for the full price distribution, and interval coverage for the ranges. We publish the calibration visually (a reliability diagram and a PIT histogram), test whether the method significantly beats the random walk with a Diebold-Mariano test (Newey-West HAC, which corrects for the overlap between hourly forecasts) that defaults to "no significant difference," and we line our odds up against a real-money market and keep score of who's been closer. Skill near zero means the method roughly ties a coin flip — the honest, expected result at short horizons.

Honesty & limits

Nobody can reliably predict Bitcoin's price. This site exists to show a transparent method and hold it accountable — not to promise returns. Read the plain-English guide and the full disclaimer.