Probabilistic forecasts

Forecasts that show their uncertainty, and their track record

Price ranges for every major coin from 4 hours to 1 year, with bull/base/bear targets, a prediction cone and every forecast scored in public.

Try asking

How likely is a 10% drop in bitcoin over the next 7 days, according to the model?

Set a price alert near the edge of the 80% range.

Compare forecast ranges for two assets with similar Research Scores.

Check whether the 80% intervals have actually contained 80% of outcomes.

Median plus 50%, 80% and 90% price intervals show what is plausible, not just one guess.

See the model's probability of a move beyond ±5%, ±10% and ±20% by the end of each horizon.

Every forecast is written to an append-only ledger when it is made and evaluated automatically when it matures.

No directional call is shown unless a model has beaten a 50/50 baseline in walk-forward testing. Otherwise we say so.

How it works

Model DK-EWMA-LN v1 estimates volatility from daily log returns with exponential weighting (λ = 0.94).

Volatility is scaled to each horizon and turned into log-normal price intervals and threshold probabilities.

Each forecast enters the ledger at creation and is scored for coverage, Brier score, log loss and calibration when it matures.

Knowing what a tool can't do is part of using it well.

DK-EWMA-LN v1 is a volatility model: it does not claim to know which way price will move.
Crypto has fatter tails than a log-normal model assumes, so extreme moves can exceed the 90% range more often than expected.
Performance statistics need at least 30 matured forecasts per horizon before they are shown as reliable.
Forecasts are statistical estimates, not advice or promises.