1. Probability
A calibrated probability of the specified downside event occurring during the forecast horizon.
Frozen model · live collectionObjective forecasts.
Objective forecasts.
Transparent by design.
Probabilistic crypto risk forecasts built to be checked, not hyped. Every live forecast is timestamped before the outcome, tracked through its full horizon, and never rewritten afterward.
Frozen modelNo retraining during forward validation.
Live validationReal-world forecasts, resolved automatically.
Ξ
ETHUSDTEthereum / Tether
LIVE
Downside 24h probability
—%
Probability of a ≥4% downside move within the next 24 hours.
TargetETH falls 4%+ within 24hDownside event from the next-hour anchor
Probability—
Risk bandLOADING
Status—
Generated—
Horizonnext 24 hours
ETHUSDT price · 1h closeLoading…Waiting for market data
How to read the forecast
This page predicts a defined event, not an exact ETH price. A 20% reading means the model estimates roughly a one-in-five chance that ETH trades at least 4% below the next-hour anchor within 24 hours.
2. Risk band
A simple context label derived from the frozen model's historical probability distribution.
3. Status
PENDING forecasts are still active. RESOLVED forecasts are locked after the 24-hour horizon.
Forward validation
Loading live validation…The frozen model is now being judged on forecasts published before the outcome is known. Live forward metrics are kept separate from the historical backtest.
Forward validation is collecting. No model retraining or recalibration is allowed during this period.
Historical validation v1.10 OOS backtest
These are historical out-of-sample research metrics, shown for context only. They are not presented as live forward results.
Historical out-of-sample validation. v1.10 was informed by earlier research and is not an independent forward confirmation.
Recent forecasts
Every recorded forecast keeps its original timestamp, probability, risk band and outcome status.
| Asset | Target | Probability | Risk | Status | Generated UTC |
|---|---|---|---|---|---|
| Loading forecast history… | |||||
Why this is different
The validation process is designed to make hindsight harder.
Frozen model
Model weights, features and threshold stay fixed during forward collection.
↗
Live tracking
Forecasts are generated from the latest closed hourly candle and stored immediately.
No retroactive backfill
If the valid publishing window is missed, the system records a skip instead of inventing a forecast later.
Locked outcomes
After the horizon completes, the actual event is resolved and kept with the original prediction.
Methodology
The current research target is intentionally narrow: estimate the probability that ETH trades at least 4% below the forecast anchor during the following 24 hours.
59 price-invariant features
Returns, realized volatility, candle structure, drawdown, breakout distance, normalized volume behavior and ETH/BTC relative features. Raw absolute ETH/BTC prices are excluded from the model feature vector.
Frozen probability calibration
The model uses the calibration frozen at the start of forward validation. Live outcomes do not alter the model during the validation period.
Strict forecast timing
The collector waits for the exact latest closed signal candle and will not publish retroactively after the 15-minute anchor-lag limit.
Forward pass criteria
The forward gate cannot become eligible until at least 60 days and 1,200 resolved frozen forecasts are collected, alongside discrimination and calibration checks.
Crypto coverageEthereum · ETHLIVEFrozen 24-hour downside probability model in forward validation. Bitcoin · BTCRESEARCHPublic BTC forecasts will launch only after a dedicated backtest and frozen forward model are approved. Solana · SOLRESEARCHPublic SOL forecasts will follow the same no-hindsight validation standard.
One validation standard for every cryptocurrency
Crypto Prediction Today is built for multi-asset forecasting, but public predictions are added only after each model passes the same research process.
No unvalidated Bitcoin or Solana prediction is published just to fill a page. New assets are added only after their own model passes validation.
Ethereum forecast research
How this live ETH forecast is designed to be used
A probability forecast instead of a price target
Crypto Prediction Today does not try to publish a single exact Ethereum price for tomorrow. The current model answers a narrower question: what is the probability that ETH trades at least 4% below the forecast anchor during the next 24 hours? Framing the problem as a defined event makes the forecast easier to evaluate. Each prediction has a timestamp, a probability, a fixed 24-hour horizon and an outcome that can later be checked without changing the original record.
This is different from a conventional Ethereum price prediction that may say ETH will reach a specific dollar value. A probability can still be wrong, but it communicates uncertainty directly. A 20% downside probability does not mean ETH will fall 20%, and it does not mean a fall is guaranteed. It means the frozen model currently assigns roughly a one-in-five chance to the predefined 4% downside event.
What information the model uses
The frozen research model uses 59 price-invariant features derived from ETH and BTC market behavior. These include returns over several windows, realized volatility, candle range and wick structure, drawdown, distance from rolling highs and lows, breakout behavior, normalized volume changes and relative ETH/BTC movement. Raw absolute ETH and BTC prices are intentionally excluded from the feature vector so the model is not allowed to rely on a particular historical price level.
The site also displays the latest ETH market price for context, but that displayed price is not the same thing as an absolute-price model feature. The public forecast is generated from the completed hourly market data available to the collector, then stored before the future 24-hour outcome is known.
Historical validation versus live forward validation
The historical metrics shown above come from the v1.10 out-of-sample backtest. They describe how the research model behaved on historical periods that were separated chronologically during testing. Those numbers are useful context, but they are not presented as proof that the model will work in the future. The live forward section is intentionally separate.
During forward validation the model, its feature definition and probability calibration remain frozen. New forecasts are collected in real time and resolved only after their horizon completes. The system does not create a forecast retroactively if the valid publishing window was missed. This design is intended to make hindsight and silent history editing harder.
How to interpret the live Ethereum risk level
The probability is the primary model output. The risk band is only a simpler label that helps readers scan the current reading relative to the frozen model's historical probability distribution. A NORMAL reading is not a guarantee of market stability, and an ELEVATED or HIGH reading is not a trading instruction. Market events, liquidity shocks and information not represented by the model can produce outcomes that differ materially from the forecast.
Crypto Prediction Today is therefore best treated as an auditable forecasting experiment: a continuously updated ETH risk estimate with a visible methodology, a locked forecast history and a growing forward-validation record. It is not personalized financial advice, a promise of profitability or a substitute for independent risk management.
FAQ
Ethereum forecast FAQ
Short answers to the most common questions about the live ETH probability model.
What exactly does the ETH forecast predict?
It estimates the probability that Ethereum trades at least 4% below the forecast anchor at some point during the following 24 hours.
Is 19.8% a prediction that ETH will fall 19.8%?
No. The percentage is a probability. The defined event itself is a downside move of 4% or more within the 24-hour horizon.
How often is the forecast updated?
The collector works from completed hourly market candles and attempts to create a new forecast inside a strict publishing window after the relevant candle closes.
What happens after 24 hours?
A pending forecast is resolved against the actual market path and kept with its original probability. Resolved records are not rewritten retroactively.
Why are historical and forward metrics separated?
Historical backtests can influence research decisions. Forward validation tests the frozen version on predictions published before their outcomes are known, so the two types of evidence are shown separately.
Does the model provide BUY or SELL signals?
No. The public output is a probabilistic research forecast for a defined ETH downside event. It is not personalized investment advice or a guaranteed trading signal.
Public read-only API
The same data shown on this page is available as JSON. These endpoints do not modify the model or database.
GET /api/latest
GET /api/history?limit=168
GET /api/stats
GET /api/chart?hours=168
GET /health
GET /api/history?limit=168
GET /api/stats
GET /api/chart?hours=168
GET /health
Research product only. Forecasts are probabilistic and may be wrong. This site does not provide personalized investment advice or promise trading profitability.