Market regime

The four regimes

MarketRegime has four variants:

Regime	Meaning
BullishExpansion	A strong trend, pointing up, with price above its average
BearishExpansion	A strong trend, pointing down, with price below its average
Consolidation	No strong trend and low volatility
Transition	Everything in between, and the warmup

The rules

pub fn classify_regime(bars: &[Bar], adx_val: f64, atr_val: f64) -> MarketRegime

bars is the history up to and including the current bar. adx_val is the current ADX value, atr_val the current ATR as a fraction of price (ATR / close).

1. Fewer than 21 bars: Transition.
2. The function computes the 20-bar simple moving average of the close, for the current and the previous bar, and its slope: (sma_20 - prev_sma_20) / prev_sma_20.
3. Trending means adx_val > 20.
   • Slope above 0.001 and close above the average: BullishExpansion.
   • Slope below -0.001 and close below the average: BearishExpansion.
   • Otherwise: Transition.
4. Not trending: atr_val > 0.02 gives Transition, anything else Consolidation.

ATR units

The atr indicator reports 100 * ATR / close, a percentage. classify_regime expects the fraction, so divide by 100 before passing it on. The crate’s own pipeline does the same.

use kestrel_chartkit::{build_checked, classify_regime, Bar};
use std::collections::HashMap;

fn regimes(bars: &[Bar]) -> Result<(), Box<dyn std::error::Error>> {
    let mut adx = build_checked("adx", &HashMap::new())?;
    let mut atr = build_checked("atr", &HashMap::new())?;

    for (i, bar) in bars.iter().enumerate() {
        let (Some(a), Some(t)) = (adx.on_bar(bar), atr.on_bar(bar)) else {
            continue; // still warming up
        };
        let regime = classify_regime(&bars[..=i], a.value, t.value / 100.0);
        println!("{} {}", bar.timestamp, regime);
    }
    Ok(())
}

The market regime chart in the showcase runs exactly this loop over synthetic bars and shades every run of equal regimes.

Building on the regime

regime_advanced holds small, independent helpers for consumers that track the regime over time:

• RegimeMarkovModel counts observed transitions and returns transition_probability(from, to) and next_state_distribution(from).
• RegimePersistenceTracker reports how long the current regime has lasted.
• HysteresisBand switches levels only after a value has crossed separate enter and exit thresholds, so a reading that hovers at a boundary does not flip back and forth.

The composite scoring uses the regime to grade a signal: see Composite scoring.

The analytics module has its own on-demand regime vote with different inputs. It is a separate function, not a wrapper around classify_regime.