Ridge-LightGBM Ensemble with Covariance Shrinkage

Performance Ranking

0.15 78.8th percentile (Sharpe Ratio) 4.39

Better than 78.8% of the 80 other models

Performance Metrics

Primary Rankings

Sharpe Ratio Sharpe Ratio Risk-adjusted return: annualized return divided by annualized volatility. 3.5572

Rank Rank Position on the leaderboard, ordered by Sharpe ratio. #18

Returns

Annualized Return Annualized Return Average yearly return, after scaling the portfolio to the 10% volatility target. This model: Better than 79% of the 80 other models. 35.57%

Best 12M Return Best 12M Return The best compounded return over any 12 consecutive months of the test period. This model: Better than 93% of the 80 other models. 148.85% Oct 1999 – Sep 2000

Worst 12M Return Worst 12M Return The worst compounded return over any 12 consecutive months of the test period. This model: Better than 71% of the 80 other models. -6.16% Feb 2020 – Jan 2021

Risk

Max Drawdown Max Drawdown Largest peak-to-trough decline during the test period. This model: Better than 80% of the 80 other models. -9.49%

Downside Deviation Downside Deviation Standard deviation of negative monthly returns only. This model: Better than 53% of the 80 other models. 1.619%

Annualized Downside Dev Annualized Downside Deviation Downside deviation in annual terms. This model: Better than 53% of the 80 other models. 5.61%

Risk-Adjusted Performance

Sortino Ratio Sortino Ratio Return per unit of downside risk: like Sharpe, but only negative volatility counts. This model: Better than 69% of the 80 other models. 6.3433

Calmar Ratio Calmar Ratio Annualized return divided by the maximum drawdown. This model: Better than 83% of the 80 other models. 3.7501

Win Rate Win Rate Share of months with a positive return. This model: Better than 84% of the 80 other models. 87.7%

Distribution

Skewness Skewness Asymmetry of returns. Positive skew means more large gains than large losses. 0.370 (Near-symmetric)

Kurtosis Kurtosis Tail thickness. Higher values mean more extreme months, in either direction. 1.548 (Thin tails)

Volatility Targeting

Raw Volatility Raw Volatility Volatility of the portfolio before it was scaled to the 10% target. 4.45%

Scaling Factor Scaling Factor Multiplier applied to the portfolio to reach the 10% volatility target. 2.247

Observations Observations Number of monthly returns used for the evaluation. 408 months

Market Exposure (CAPM)

Beta Beta Sensitivity to market movements. 1.0 moves one-for-one with the market. -0.01

Alpha (ann.) Alpha (ann.) Average yearly return not explained by market exposure (the monthly regression alpha × 12). This model: Better than 81% of the 80 other models. 35.70%

Information Ratio Information Ratio Annualized alpha per unit of residual risk: alpha ÷ std(residuals) × √12. This model: Better than 81% of the 80 other models. 3.5662

R² R² Share of the return variance explained by market movements. 0.1%

Alpha t-stat Alpha t-stat Statistical significance of the alpha estimate. A |t| above about 2 suggests it is not chance. 20.54

Benchmark Comparison

vs. Benchmarks

Ridge-LightGBM Ensemble with Covariance Shrinkage

3.5572

Benchmark: Markowitz-ML

2.5212

Benchmark: Instrumented PCA factor portfolio

1.9481

Benchmark: Kozak-Nagel-Santosh rank-weighted factors portfolio

1.4195

Benchmark: Minimum Variance

0.9470

Benchmark: Factor-ML

0.8901

Benchmark: Equal-weight portfolio

0.4912

Performance Radar

Each axis shows the model's percentile rank within the leaderboard (outer edge = best). Customize the metrics using the dropdowns below.

vs. Leaderboard

Metric	This Model	Average	Top Model	vs. Avg
Sharpe Ratio Sharpe Ratio Risk-adjusted return: annualized return divided by annualized volatility.	3.5572	2.2472	4.3949	+58.3% (better)
Sortino Ratio Sortino Ratio Return per unit of downside risk: like Sharpe, but only negative volatility counts.	6.3433	4.8069	17.3078	+32.0% (better)
Calmar Ratio Calmar Ratio Annualized return divided by the maximum drawdown.	3.7501	2.0485	11.7867	+83.1% (better)
Information Ratio Information Ratio Annualized alpha per unit of residual risk: alpha ÷ std(residuals) × √12.	3.5662	2.2146	4.3562	+61.0% (better)
Win Rate Win Rate Share of months with a positive return.	87.7%	75.5%	91.2%	+16.2% (better)
Annualized Return Annualized Return Average yearly return, after scaling the portfolio to the 10% volatility target.	35.57%	22.47%	43.95%	+58.3% (better)
Max Drawdown Max Drawdown Largest peak-to-trough decline during the test period.	-9.49%	-18.68%	-3.51%	+49.2% (better)

Model Details

Language Python

Competition Year 2024

First Submitted Oct 7, 2026

Scored Oct 7, 2026

Downloads

Download Code (Python) 22 downloads Download Documentation 28 downloads

Have ideas for making this page more useful? Share your suggestions