Transparency · Deterministic Engine
Methodology
All diagnostics produced by Quant Research Critic are fully deterministic and reproducible.
The same inputs will always produce the same outputs. AI is used only to interpret and explain
results — it does not alter, score, or influence any calculation.
Input Format
Strategy CSV Input
Each uploaded CSV file represents a single strategy return series. The engine expects a date column and a return column.
date,return
2023-01-01,0.0021
2023-01-02,-0.0014
2023-01-03,0.0008
Multiple CSV files can be uploaded to construct a portfolio. Each file is treated as one strategy sleeve.
Core Calculations
Correlation Matrix
Pairwise Pearson correlation is computed across all strategy return series. The result is a symmetric N×N matrix where each cell represents the linear relationship between two strategies.
corr(i, j) = cov(rᵢ, rⱼ) / (σᵢ · σⱼ)
Values range from −1 (perfect inverse) to +1 (perfect co-movement). Diagonal values are always 1.00.
Diversification Score
The diversification score measures how independent the strategies in a portfolio are from one another. A score near 1.0 indicates a highly diversified portfolio; a score near 0.0 indicates high overlap.
Diversification Score = 1 − mean(|corr(i, j)|) for all i ≠ j
Only off-diagonal correlations are used. The absolute value ensures that both positive and negative correlations are treated as overlap.
Overlap Risk Classification
Portfolio overlap risk is derived from the average pairwise correlation across all strategy pairs. The threshold boundaries are fixed and deterministic.
High Overlap
avg correlation > 0.70
HIGH
Moderate Overlap
0.40 – 0.70
MEDIUM
Low Overlap
avg correlation < 0.40
LOW
Cluster Analysis
Strategies are grouped into clusters based on return correlation. Strategies within the same cluster behave similarly and represent the same underlying exposure. The number of clusters indicates how many truly independent sleeves exist in the portfolio.
A portfolio with 4 strategies but 2 clusters is effectively a 2-sleeve portfolio — capital is concentrated even if it appears diversified by strategy count.
Allocation Weights
Capital weights are computed using a correlation-penalized allocation model. Strategies with high overlap receive reduced combined weight. Independent strategies receive a premium allocation proportional to their diversification contribution.
weight(i) ∝ diversification_contribution(i) / Σ diversification_contribution
Weights sum to 100%. The allocation engine does not use expected returns or volatility forecasts — only realized correlation structure.
AI Interpretation Layer
Deterministic vs AI
Every number, score, classification, and allocation weight produced by this tool is fully deterministic. The same inputs will always produce the same outputs regardless of when the analysis is run.
AI is used exclusively as an interpretation layer. It receives the deterministic outputs and generates plain-language explanations — portfolio commentary, allocation rationale, and risk narratives. AI does not score strategies, set thresholds, or influence any calculation.
Math decides. AI explains.