Exit Criteria That Actually Work in Small-Caps
We tested multiple exit models in a small-cap strategy and found that binary exits destroy value. A score-driven decay framework with hysteresis improved returns by 7.4pp, raised Sharpe, and cut turnover by 46%.
At a Glance
Using historical data from Nov 2022 – Nov 2025, we redesigned and tested our exit logic for small-caps.
What changed — and what it delivered:
- +7.4 percentage points higher total return
- Sharpe +0.05 (1.29 → 1.34)
- Turnover down 46% (1028% → 550%)
- Whipsaws: 27 → 0
- Max drawdown improved by 0.4pp
The core insight was counter-intuitive:
Score drops predict positive forward returns.
Binary exits were systematically selling before mean reversion.
The Hidden Cost of “Clean” Exit Rules
Most systematic equity strategies rely on exits like:
- Score < X → sell
- Price down 20–25% → review
- Earnings miss → exit
They look disciplined.
In small-caps, they are often the largest source of silent alpha leakage.
Why?
Because small-caps are structurally different:
- Information arrives in jumps, not smoothly
- Prices overshoot fundamentals
- Mean reversion after drawdowns is strong
Binary exits assume precision in signals.
Our data shows that precision does not exist.
What Our V3 Exit Logic Looked Like
V3 (Production until Jan 2026):
| Trigger | Action |
|---|---|
| Score < 3.0 | Exit within 5 days |
| Price ↓ >25% | Review; exit if score <5 |
| Governance breach | Cap score |
Observed problems:
- Forced selling during temporary dislocations
- High churn in noisy quarters
- Loss crystallization just before recovery
The Discovery That Changed Everything
We ran a simple analysis across 12 quarters (Nov 2022 – Nov 2025):
Forward Returns After Score Drops
| Score Change | Observations (N) | Avg Forward Return |
|---|---|---|
| 6 → 5 | 93 | +9.8% |
| 5 → 4 | 159 | +4.0% |
| 4 → 3 | 176 | +5.2% |
| < 3 | 140 | +3.0% |
| Stable ≥6 | 275 | +2.1% |
Interpretation:
- Score drops often coincide with price drawdowns
- Selling pressure has already occurred
- Forward returns are higher, not lower
Binary exits were selling into mean reversion.
The Principle We Adopted
When ranking quality is weak (IC ≈ 0.02), exits must be smooth — not precise.
This led to a complete redesign.
The V4 Exit Framework (What We Use Now)
Layer 1: Score-Driven Position Decay
Instead of hold vs exit, every stock flows through decay buckets, applied only at rebalance:
| Conviction Score | State | Multiplier |
|---|---|---|
| ≥ 6.0 | FULL | 1.00 |
| 5.0 – 5.9 | REDUCED | 0.70 |
| 4.0 – 4.9 | MINIMAL | 0.40 |
| 3.5 – 3.9 | WATCH | 0.15 |
| < 3.5 | EXIT | 0.00 |
Effective weight
Effective Weight = Kelly Base Weight × Decay Multiplier
This preserves conviction while shedding risk gradually.
Layer 2: Hysteresis
- Downgrades require two consecutive quarters
- Upgrades happen immediately
Why this matters:
- Score volatility is ±0.5 quarter-to-quarter
- Without hysteresis, boundary noise causes churn
- With hysteresis, temporary weakness is ignored
Result: whipsaws reduced from 27 → 0.
Layer 3: Rebalance-Only Execution
- No intra-quarter exits
- No daily reactions
- No price-based triggers
All decay happens at scheduled rebalances, reducing turnover materially.
Layer 4: Hard Exits (Structural Only)
Binary exits still exist — but only for irreversible failures:
| Hard Exit Trigger |
|---|
| ROE ≤ 0 |
| Net worth ≤ 0 |
| Promoter pledge > 80% |
| Promoter holding ↓ >15pp (12M) |
| Missing financials > 2 quarters |
Everything else is signal degradation, not invalidation.
What Changed in the Numbers
Exit Logic Backtest Comparison
| Configuration | Return | Sharpe | Turnover | Whipsaws |
|---|---|---|---|---|
| V3 Binary | 123.4% | 1.29 | 1028% | 27 |
| Decay (no hysteresis) | 128.9% | 1.34 | 831% | 21 |
| Decay + Hysteresis | 130.8% | 1.34 | 550% | 0 |
| Decay + Regime Adjustment | 131.5% | 1.34 | 570% | 2 |
Final choice:
👉 Decay + Hysteresis (simplest, most robust, lowest churn)
Why We Rejected Regime-Aware Exit Thresholds
We tested regime-specific exit cutoffs.
Result:
- +0.7pp incremental return
- Higher turnover
- Reintroduced whipsaws
Given:
- Weak IC
- Strong mean reversion after score drops
- Added complexity
We removed it entirely.
Exit logic is regime-invariant by design.
Regimes already control exposure and factor gating, not exits.
How a Stock Lives (and Exits) in Our System
Weekly – Universe Refresh
- Market cap ₹500–15,000 Cr
- Liquidity filter
- Promoter holding ≥25%
- ROE > 0
- Net worth > 0
- Data freshness ≤ 2 quarters
Quarterly – Portfolio Decisions
- Governance caps
- Conviction scoring (0–10)
- Kelly-based position sizing
- Regime-driven exposure
Ongoing – Exit Logic
- Score-driven decay
- Mandatory hysteresis
- Structural hard exits only
Exits are the final probabilistic layer, not a blunt instrument.
The Big Takeaway
Most exit rules fail because they assume:
- Scores are precise
- Fundamentals degrade linearly
- Early selling is always safer
In small-caps, the opposite is often true.
Our conclusion:
- Be slow to punish
- Be fast to reward
- Let conviction decay, not snap
Exit logic is not about being right.
It’s about not being forced wrong at the worst possible time.
This framework is now live in production (V4).
We do not expect to revisit exit logic again until signal quality materially improves.