Our Methodology

Multi-Model Composite Rating System

Our star ratings are powered by an ensemble of proprietary quantitative models, validated across 15 years of market data (2010–2025). Each model captures a different dimension of stock quality, and the composite signal is significantly more robust than any single factor alone.

+4.54%

12M Long/Short Spread

Top vs. Bottom Decile

0.988

Decile Monotonicity

Spearman ρ on Deciles

p < 10⁻¹²

Statistical Significance

Spearman p-value

80K+

Observations Tested

Across All Horizons

15 Years

Backtesting Period

2010 – 2025

How Our Ratings Work

A three-stage pipeline from raw data to actionable star ratings.

Step 01

Multi-Factor Signal Generation

Multiple proprietary models independently evaluate each stock across fundamental quality, momentum, valuation, earnings stability, and macro sensitivity. Each model contributes a distinct, uncorrelated perspective.

Step 02

Ensemble Aggregation

Individual model outputs are combined into a unified signal using an optimized weighting scheme. The ensemble approach dramatically reduces noise and model-specific biases, producing a more reliable assessment than any single model.

Step 03

Star Rating Assignment

The aggregated signal is translated into intuitive 1-5 star ratings using calibrated thresholds. Stars make it easy to instantly identify high-conviction opportunities without interpreting complex data.

Decile Performance Analysis

Higher-rated stocks consistently deliver higher excess returns across all time horizons.

Decile
Excess 1M
Excess 3M
Excess 6M
Excess 12M
D1 (Lowest)
-0.27%
-0.97%
-2.25%
-4.38%
D2
-0.33%
-0.94%
-1.80%
-3.27%
D3
-0.13%
-0.45%
-1.22%
-2.56%
D4
-0.13%
-0.68%
-1.40%
-2.09%
D5
-0.02%
-0.35%
-0.89%
-1.78%
D6
-0.09%
-0.27%
-0.63%
-1.14%
D7
+0.08%
-0.03%
+0.03%
-0.65%
D8
+0.01%
-0.10%
-0.31%
-0.42%
D9
+0.14%
+0.09%
+0.47%
+0.23%
D10 (Highest)
+0.13%
+0.33%
+0.25%
+0.17%
Spread (D10 – D1)
+0.41%
+1.29%
+2.50%
+4.54%

Why Our Approach Works

The composite methodology offers significant advantages over single-factor or discretionary approaches.

ρ = 0.988

Near-Perfect Monotonicity

Decile monotonicity of 0.988 means higher-rated stocks almost always outperform lower-rated ones. There are no "gaps" or inversions — the signal is remarkably clean across the entire spectrum.

+4.54% at 12M

Strengthens Over Time

The long/short spread grows from +0.41% at 1 month to +4.54% at 12 months. This suggests the composite captures durable stock quality rather than short-term noise — ideal for buy-and-hold investors.

p < 10⁻¹²

Statistically Robust

All correlation metrics are highly significant (p < 10⁻¹²) across 80,000+ observations. This isn't curve-fitting — the signal holds up across diverse market conditions and time periods.

Multi-Model

Ensemble Diversification

By combining multiple independent models, the ensemble signal is far more stable than any individual factor. Model-specific weaknesses cancel out, resulting in smoother, more reliable ratings over time.

Correlation Statistics

Measuring the relationship between our ratings and future excess returns.

Metric
Type
1M
3M
6M
12M
Pearson r
Linear
-0.0074
-0.0054
-0.0063
-0.0051
Spearman ρ
Rank
0.0248
0.0336
0.0513
0.0579
Spearman p-value
Significance
9.95E-13
7.20E-22
9.50E-48
3.53E-58
Decile Monotonicity
Spearman ρ on deciles
0.9390
0.9760
0.9640
0.9880
Observations
Sample size
82,531
81,534
80,041
77,073

Key Findings

Higher Ratings = Higher Excess Returns

The decile analysis shows a clear, nearly monotonic relationship. Stocks in the top decile (D10) outperform the bottom decile (D1) by +0.41 pp/month, +1.29 pp/quarter, +2.50 pp/half-year, and +4.54 pp/year.

The Effect Strengthens Over Longer Horizons

The long/short spread grows from +0.41% at 1M to +4.54% at 12M, suggesting our ratings are more predictive at longer time frames — making them ideal for medium to long-term investors.

Statistically Significant but Economically Measured

Spearman correlations (0.02-0.06) are highly significant (p < 1e-12) due to the large sample size (~80K+ observations), but the effect sizes are modest. Decile monotonicity is strong (ρ = 0.94-0.99), confirming systematic predictive power.

Consistent Across Market Conditions

Our ratings are tested across 15 years of market data (2010–2025), spanning bull markets, corrections, and bear cycles. The winsorized analysis controls for outliers, ensuring the results reflect persistent stock selection alpha rather than extreme return events.

See the Ratings in Action

Explore our full stock coverage with live market data, star ratings, and historical analysis.

Explore Dashboard