REGIMES.md

REGIMES.md

Quick reference for regime signals and suggested strategy behavior.

Core Idea

Use recent-session features as priors to select mode first, then run a mode-specific strategy.

---

Strong Regime Signals Found

1) Low-move regime (next session likely quiet)

• W=1: vol_mean <= 1.82697e-05 → lower chance of high abs move next session
• W=5: range_max <= 0.00375927 → lower chance of high abs move next session

Mode mapping: QUIET

• Use fade/mean-reversion behavior
• Smaller sizing
• Tighter stops

---

2) Vol-persistence regime (next session likely to move)

• W=3: vol_max >= 3.28199e-05 → much higher chance of high abs move
• W=5: range_min >= 0.00248672 → much higher chance of high abs move
• W=6: range_mean >= 0.00392654 → much higher chance of high abs move

Mode mapping: VOL_PERSIST

• Use breakout/trend-follow behavior
• Allow larger sizing
• Wider stop/hold logic

---

3) Directional up bias

• W=1: crosses >= 16 → higher P(next_up)
• W=2/3: crosses_trend high (e.g., >=15/16) → higher P(next_up)

Mode mapping: BIAS_UP

• Prefer long-up entries
• Raise threshold for down entries or disable down entries

---

4) Directional down / anti-up bias

• W=6: crosses_max <= 15 → lower P(next_up)

Mode mapping: BIAS_DOWN

• Prefer down entries
• Require stronger confirmation for up entries

---

5) Low-chop vs high-chop proxy

• W=5: above_max >= 0.997 → lower chance of next session being cross-heavy

Interpretation: very one-sided recent midpoint behavior often precedes less crossing noise.

Mode effect: can favor single-entry conviction behavior over multi-reentry scalping.

---

Suggested Priority Stack for Mode Selection

1. If strong vol-persistence signals hit → VOL_PERSIST
2. Else if strong low-move signals hit → QUIET
3. Else apply directional bias (BIAS_UP / BIAS_DOWN)
4. Else NEUTRAL

---

Notes

• These are exploratory empirical mappings, not guaranteed alpha.
• Use minimum sample-size guards and periodic revalidation.
• Optimize net PnL after friction, not win rate alone.