Multi-Factor Composite Scoring: Integration Across Correlated Signals
Discover how Parallax normalizes individual factor percentiles into an institutional 0-10 composite score using dynamic ML weighting and correlation boundaries.
The Mathematics of Multi-Factor Score Normalization
Constructing a reliable multi-factor equity model requires synthesizing heterogeneous variables—from historical EBITDA multiples and return-on-equity to twelve-month momentum trends and order imbalance metrics. Raw financial metrics cannot be directly combined due to differing measurement scales, heavy-tailed distributions, and extreme industry-specific outliers.
To create an investable multi-factor composite, raw metrics are first normalized within peer investment universes into standardized percentile ranks, scaled from 0 to 10:
- Scores 8–10: Top quartile, representing strong to exceptional factor strength.
- Scores 6–7: Above median, reflecting moderately attractive characteristics.
- Scores 4–5: Below median, representing neutral to weak signal exposure.
- Scores 0–3: Bottom quartile, representing poor to very poor performance.
Resolving Factor Correlation and Signal Cancellation
A primary obstacle in multi-factor design is cross-factor correlation. Core equity factors frequently exhibit pronounced inverse correlations—most notably Value versus Momentum. During sharp market trends, cheap value stocks frequently display deteriorating momentum profiles, while high-momentum leaders trade at elevated valuation multiples.
Naive equal-weighted aggregation causes severe signal cancellation: combining a top-decile Value score with a bottom-decile Momentum score collapses the composite rank to an uninformative median score of 5.0, stripping the portfolio of active exposure to both factors.
Dynamic ML-Optimized Weighting and Boundary Enforcement
Parallax resolves signal cancellation by employing an adaptive machine learning model that dynamically calibrates factor weights according to prevailing market regimes:
- Regime-Aware Covariance Estimation: The ML model forecasts near-term factor expected returns and inter-factor correlation structures, adapting allocations across changing macroeconomic regimes.
- Explicit Factor Weight Boundaries: To prevent unintended factor concentration or excessive turnover, the composite optimization enforces strict structural constraints:
- Value: 10% minimum to 40% maximum.
- Quality: 10% minimum to 40% maximum.
- Momentum: 10% minimum to 40% maximum.
- Defensive: 10% minimum to 40% maximum.
- Tactical: 10% minimum to 40% maximum.
- Isolation of Size: As documented in the methodology, Size is evaluated independently for portfolio screening but is excluded from the Overall Score calculation to safeguard composite quality.
Portfolio managers review the complete scoring framework at docs.chicago.global/methodology/scoring and examine individual signal architecture at docs.chicago.global/methodology/factors.