Regime Transition Probabilities + Prediction Heads for Trading Bots

Beyond 'What Regime Are We In?'

Knowing the current regime is table stakes. The decisions that make or lose money are about what happens next and what that implies for direction, volatility and funding. The quant engine answers both with two extra layers on top of the regime label: transition probabilities and prediction heads.

Both live in the probabilities object of /api/v1/quant/market (and per coin). Transitions tell your agent where the regime is likely to go; the heads tell it what each outcome dimension looks like, conditioned on the current state.

Regime Transition Probabilities

regime_transitions is the forward distribution over the six regimes for your chosen horizon — the probability of staying put versus moving to each other state.

"regime_transitions": {
  "stays_same": 0.7326,
  "to_vol_spike": 0.1042,
  "to_strong_trend_bull": 0.0958,
  "to_squeeze": 0.0674,
  "to_strong_trend_bear": 0.0,
  "to_choppy_high_vol": 0.0
}

Here the market is 73% likely to stay in its current regime, but the largest migration risk is a 10% drift to vol_spike. That single number is a pre-emptive de-risking signal — the move hasn't happened yet, but the model is flagging rising odds of it.

The Prediction Heads: Six Conditional Forecasts

Each head is a probability distribution over outcome buckets, conditioned on the current regime. They turn one regime into six actionable forecasts:

Reading the Heads Together

curl -H "X-API-Key: cdk_live_your_key" \
  "https://cryptodataapi.com/api/v1/quant/market?horizon=24h"

"probabilities": {
  "directional": {"strong_down": 0.09, "mild_down": 0.21, "flat": 0.34,
                   "mild_up": 0.25, "strong_up": 0.11},
  "volatility":  {"low": 0.27, "medium": 0.34, "high": 0.38},
  "funding":     {"falls": 0.29, "stable": 0.39, "rises": 0.32},
  "liquidation_risk": {"low": 0.27, "medium": 0.34, "high": 0.38}
}

A flat-leaning direction, a slight skew to high volatility, and a one-third liquidation-risk read paints a coherent picture: no directional edge, but stay small — the book could get violent. Each head also carries a confidence so you know how much to trust it.

What Each Head Drives

• directional — bias and conviction for entries; the spread across the five buckets is your signal strength.
• volatility — straddle/strangle sizing and stop width; high vol means wider stops or no trade.
• liquidation_risk — cascade odds; pair with thin order-book depth to anticipate forced selling.
• funding — carry direction; informs basis and funding-harvest setups.
• open_interest — is the move backed by new leverage (continuation) or an unwind (exhaustion)?
• breadth — market-wide participation; a thinning breadth under a rising index is a classic divergence.

How AI Agents Use Transitions and Heads

• Pre-emptive de-risk. When to_vol_spike or to_strong_trend_bear rises above a threshold, cut leverage before the regime actually flips.
• Direction gating. Only take longs when directional skews up and the regime supports it — two independent confirmations.
• Carry & leverage reads. Combine funding + open_interest to spot crowded positioning ripe for a squeeze or flush.

Start from the regime probabilities for the state, then layer transitions and heads for the forward view. All of it is in a single /quant/market response.