How to Read Model Predictions
Every row of data includes forecast fields from two model families. None of them call price direction - they measure risk and volatility. Here's what each one actually means, with worked examples. See published accuracy for how reliable each one is.
The Four Horsemen - per-bar risk & volatility classifiers
Each outputs a confidence field, a probability from 0.0 to 1.0 that a specific condition is true for the next single candle. Values near 1.0 mean high confidence "yes", near 0.0 mean high confidence "no", and near 0.5 mean a coin-flip - treat it as no edge.
tail_risk_conf
Predicts: probability the next candle's move is larger than the 90th-percentile absolute move of the prior 20 candles. Direction is not part of the call - it could be up or down.
Low (≤0.1): expect a normal-sized move. Near 0.5: weak signal, low trust.
high_vol_conf
Predicts: probability the next candle's 20-bar rolling volatility sits above its 50-bar median - i.e. a choppier-than-usual bar. This is the Regime State signal.
Low (≤0.1): expect a calmer-than-usual bar relative to the benchmark.
regime_change_conf
Predicts: probability the next candle is the first high-vol bar after a calm stretch - a low-to-high volatility transition, not an ongoing high-vol state.
Low (≤0.1): strong belief the calm stretch continues.
regime_emergence_conf
Predicts: probability that short-horizon (10-bar) volatility is about to spike above its 30-bar median - an earlier, faster-twitch cousin of regime change.
Reads the same scale as the others: near 1.0 is high confidence, near 0.0 is low confidence, near 0.5 is weak.
Core Alpha - volatility magnitude models
A separate model family that forecasts how much the market is likely to move, not a yes/no regime call and not a direction call.
future_vol
Predicts: expected volatility (as a decimal, e.g. 0.018 ≈ 1.8% typical move) over the next few bars. A single continuous number, not a probability.
| Value | Interpretation |
|---|---|
| 0.005-0.010 | Low vol - calm, range-bound |
| 0.010-0.020 | Normal vol - typical daily moves |
| 0.020-0.040 | Elevated vol - active, news-sensitive |
| 0.040+ | High vol - large gaps, macro events |
Example: future_vol: 0.032 with recent realized vol of 0.018 → sharp moves (~3.2%/day) are expected; often paired with a widening stop-loss, not a long/short call.
expansion_probability
Predicts: probability (0.0-1.0) that volatility expands beyond its 20-day realized baseline. Distinct from high_vol_conf above - this is its own model and its own window.
| Probability | Interpretation |
|---|---|
| 0.0-0.25 | Very unlikely to expand |
| 0.25-0.50 | Slight contraction bias |
| 0.50-0.75 | Slight expansion bias |
| 0.75-1.0 | Very likely to expand |
Example: expansion_probability: 0.87 alongside already-elevated realized vol → high chance of a further volatility breakout.
None of these fields predict price direction. They describe risk and volatility regime only - pair them with your own directional thesis or entry signal before sizing a trade. Full field-by-field accuracy is published on the methodology page.
API Documentation
Interactive OpenAPI Specification for all data endpoints.