Equity Decimation: The 43-to-18 Concentration Model

The Concentration Thesis

In Q4 2025, the Refine Engine identified a critical inefficiency: 43 equity positions were generating net drag on portfolio alpha. The signal-to-noise ratio had deteriorated below our acceptable threshold of 0.72.

This dispatch documents the complete methodology behind our concentration protocol — a systematic reduction from 43 legacy holdings to 18 high-conviction core positions.

Methodology

The reduction was executed through a three-phase protocol:

1. Phase I — Signal Decomposition: Each position was scored against our proprietary alpha-factor model, isolating idiosyncratic returns from beta exposure.
2. Phase II — Correlation Pruning: Positions exhibiting >0.85 rolling 60-day correlation with existing core holdings were flagged for removal.
3. Phase III — Liquidity Audit: Remaining candidates were stress-tested against a 2-sigma liquidity shock scenario.

Key Performance Metrics

Metric	Before	After	Delta
Total Positions	43	18	-58.1%
Portfolio Beta	1.12	0.94	-16.1%
Sharpe Ratio	1.34	1.89	+41.0%
Weekly Alpha	+4.1 bps	+12.4 bps	+202.4%
Tracking Error	6.2%	3.8%	-38.7%

The Noise Elimination Framework

“The essence of investment is not diversification — it is the disciplined elimination of noise.” — Refine Internal Memo, January 2026

Our framework operates on a simple axiom: every position must independently justify its inclusion. There is no room for “portfolio filler” or positions held purely for diversification theater.

The 25 eliminated positions fell into three categories:

• Correlation Redundancies (14 positions): Exposures adequately captured by remaining core holdings
• Alpha Decay (7 positions): Once-productive positions whose alpha signals had mean-reverted to zero
• Liquidity Traps (4 positions): Positions with execution cost exceeding expected alpha contribution

Implementation Notes

The trim was executed over 12 trading sessions using our proprietary VWAP-Adaptive algorithm to minimize market impact:

# Refine Engine v5 — Trim Protocol
def execute_trim(positions: list, target_count: int) -> TrimResult:
    scored = alpha_factor_model.score(positions)
    pruned = correlation_filter(scored, threshold=0.85)
    validated = liquidity_audit(pruned, shock_sigma=2.0)
    return validated[:target_count]

Total slippage during execution: 0.8 bps — well within our 2.0 bps tolerance.

Conclusion

The 43-to-18 concentration model represents a paradigm shift in how we construct portfolios. By treating restraint as an alpha source, we achieved a 3x improvement in risk-adjusted returns while dramatically simplifying operational overhead.

This protocol is now embedded in the Refine Engine as Trim_v2 and runs automatically at each quarterly rebalance window.

---

Dispatch Classification: OPTIMIZED — All parameters verified and production-deployed.