Evolving the Vertex Quality Score: How Genetic Optimization Uncovers Global Compounders

At Ynves, our mission is simple: eliminate black-box guessing and help investors build long-term wealth by discovering true, high-quality compounder stocks.

Central to our platform is Y-VQS (Vertex Quality Score)—our proprietary scoring engine built on four core financial pillars: Solvency & Safety, Profitability & Efficiency, Growth Compounding, and Capital Discipline.

For a long time, quantitative quality scores in Wall Street and fintech relied on human intuition. Financial analysts would sit in a room and hardcode static rules:

“Award 10 points if Return on Invested Capital (ROIC) is above 15%. Deduct 5 points if Net Debt/EBITDA exceeds 3.0x.”

While these rules sound sensible, they raise fundamental questions:

Why 15% ROIC and not 16.2%?
How should debt thresholds adapt across different industries, like Banks, REITs, or Tech?
How do we know these static human rules won’t fail when market regimes shift?

To solve this, we built Y-Evolve—our in-house quantitative genetic optimization framework. Here is how we used evolutionary algorithms to refine Y-VQS and uncover superior global compounders.

What is Y-Evolve?

Think of Y-Evolve as Darwinian natural selection for financial scoring models.

Instead of manually guessing parameter weights, Y-Evolve tests candidate parameter formulas across a vast dataset of historical stock fundamentals. In each generation:

It takes our baseline scoring model.
It introduces controlled parameter mutations across 200+ financial variables (adjusting targets for ROIC, cash conversion, debt caps, and category weights).
It evaluates how effectively each mutated candidate selects winning compounders.
It crowns the winning candidate, which then becomes the parent seed for the next generation—compounding improvements over time.

The Benchmark Universe: Going Truly Global

A scoring model trained only on US Large-Cap tech stocks will fail when applied to European industrials or Asian small-caps.

To eliminate regional and home-country bias, Y-Evolve is universe-agnostic and built to train against broad global benchmarks. As a prime real-world example, we ran our latest evolutionary optimization across AVGV (Avantis All Equity Markets Value ETF).

AVGV provides a compelling multi-region testbed because it operates as a Fund-of-Funds containing over 2,000+ stocks across six distinct market segments:

US Equity: Large-Cap (AVLV), Mid-Cap (AVMV), and Small-Cap (AVUV) Value
International Developed: Large-Cap (AVIV) and Small-Cap (AVDV) Value
Emerging Markets: Value (AVES)

By evaluating candidate formulas across broad global ETF universes like AVGV, Y-VQS learns to identify universal quality traits that transcend geographic borders and market-cap tiers.

Measuring Model Edge: Validating Statistical Alpha

A common misconception in quantitative finance is confusing quality with performance.

Not every stock that delivered high past returns was a pristine business—many were speculative turnaround bets or temporary momentum spikes. Conversely, with Y-VQS, our goal is not simply to evaluate fundamental business health in a vacuum, but to measure how reliably those quality traits translate into superior, risk-adjusted returns over time.

Y-Evolve evaluates how effectively that quality signal generates real-world statistical outperformance. In every generation, Y-Evolve ranks the entire universe by Y-VQS score and isolates the Top 20% highest-rated quality tier. It then measures whether holding that top-scoring portfolio achieves superior risk-adjusted returns and active alpha over the passive market benchmark.

To ensure parameter formulas are trained to identify true long-term compounders rather than short-term market noise, Y-Evolve evaluates stock performance across a hierarchical multi-year CAGR waterfall (10-Year $\rightarrow$ 5-Year $\rightarrow$ 3-Year $\rightarrow$ 1-Year annualized compound returns). This forces the model to reward companies with durable multi-year economic moats while eliminating single-quarter price spikes.

To verify whether a mutated parameter formula increases our predictive edge, Y-Evolve evaluates candidate models using a composite Fitness Score built on three risk-return principles:

$$\text{Fitness Score} = \text{Sharpe Ratio} + 0.80 \times \text{Information Ratio} - 1.50 \times \text{Underperformance Penalty}$$

Here is what these three components measure in plain English:

1. The Sharpe Ratio (Risk-Adjusted Return Efficiency)

High returns mean little if achieved through extreme, nerve-wracking volatility. The Sharpe Ratio measures excess return divided by portfolio risk, forcing Y-Evolve to favor steady, resilient compounders over erratic high-beta stocks.

2. The Information Ratio (Predictive Alpha Strength)

How reliably does a high Y-VQS score translate into beating the market? The Information Ratio measures active outperformance relative to our global benchmark (AVGV), proving that selecting top-rated Y-VQS stocks generates consistent, repeatable stock-picking alpha.

3. The Downside Penalty (Capital Preservation)

Protecting downside capital is mathematically more important than chasing upside—a 50% loss requires a 100% gain just to break even. We apply an aggressive $1.5\times$ penalty if a candidate formula exhibits excessive drawdown risk ($1.85 \times \text{volatility}$) or trails the benchmark, eliminating parameter formulas that accept uncompensated risk.

Out-of-Sample Validation: Guarding Against Overfitting

The single biggest failure mode in quantitative modeling is overfitting.

Think of a student who memorizes last year’s exam answers word-for-word. They might score 100% on practice tests, but fail when faced with new questions. An overfitted stock model does the exact same thing: it looks genius on past data, but fails in live market regimes.

To ensure parameter adjustments reflect genuine statistical edge rather than historical coincidence, Y-Evolve enforces a strict 70/30 train/test split:

 ┌──────────────────────────────────────────────────────────────────────────┐
 │                     GLOBAL STOCK UNIVERSE (AVGV)                         │
 └────────────────────────────────────┬─────────────────────────────────────┘
                                      │ (Deterministic Shuffle)
                   ┌──────────────────┴──────────────────┐
                   ▼                                     ▼
     ┌───────────────────────────┐         ┌───────────────────────────┐
     │   70% IN-SAMPLE DISCOVERY │         │  30% OUT-OF-SAMPLE VAULT  │
     │   (Training Mutations)    │         │  (Generalization Gate)    │
     └─────────────┬─────────────┘         └─────────────┬─────────────┘
                   │                                     │
                   ▼                                     ▼
      Discovers Candidate Formulas            Evaluates Real-World Edge

70% In-Sample Dataset: Used by the evolution engine to discover candidate parameter mutations.
30% Out-of-Sample Dataset: Held out completely to verify real-world generalization.

A candidate parameter set is crowned a winner only if its fitness score improves on the 30% unseen dataset. If a mutation improves in-sample metrics but degrades out-of-sample performance, Y-Evolve discards it immediately.

The Results: Compounding Superior Alpha

By combining Cumulative Elitist Evolution, Global Universe Benchmarking (AVGV), and Strict Out-of-Sample Validation, Y-Evolve systematically compounded performance across successive evolutionary runs:

Optimization Run	Out-of-Sample Fitness	Sharpe Ratio	Information Ratio (Alpha)	Max Drawdown Est.	Relative Fitness Gain
Original Baseline	`1.1070`	`0.68`	`0.53`	`19.99%`	Baseline
Generation Run #1	`1.4145`	`0.85`	`0.70`	`19.29%`	+27.8% (+0.3075)
Generation Run #2	`1.4480`	`0.87`	`0.72`	`19.07%`	+30.8% (+0.3410)
Generation Run #3	`1.4794`	`0.89`	`0.74`	`19.07%`	+33.6% (+0.3724)
Generation Run #4	`1.4831`	`0.89`	`0.74`	`19.22%`	+34.0% (+0.3761)
Generation Run #5	`1.5260`	`0.92`	`0.76`	`18.86%`	+37.8% (+0.4190)
Generation Run #6 (Current)	`1.5385`	`0.92`	`0.77`	`18.87%`	+39.0% (+0.4315)

Over successive evolutionary passes, Y-Evolve achieved a +39.0% total out-of-sample fitness gain:

0.92 Sharpe Ratio: To put this in perspective, broad global equity market indices historically operate at a Sharpe ratio around 0.45 – 0.65. Pushing Y-VQS to 0.92 over 1.5x increases the risk-adjusted return efficiency of holding a passive market index.
0.77 Information Ratio: In institutional portfolio management, an Information Ratio above 0.50 is considered strong, and above 0.70 is top-decile. At 0.77, Y-Evolve proves that selecting top-rated Y-VQS quality stocks can generate consistent, repeatable outperformance over a benchmark index rather than relying on fluke luck.

Every winning parameter discovered by Y-Evolve is automatically validated and promoted into our live Y-Vertex Quality Score pipeline.

Continuous Model Evolution: These benchmark results represent just the beginning of our optimization framework. Because global market dynamics and macro regimes continuously shift, Y-Evolve operates as a living, automated pipeline. We constantly run new evolutionary passes in the background to refine our financial model and promote newly discovered parameter edge into our live engine in real time.

Transparent, Auditable Wealth Engine

At Ynves, we believe you deserve full visibility into how your money is analyzed. We don’t hide behind Wall Street jargon or secret black boxes.

By combining cutting-edge genetic algorithms with rigorous financial principles, we provide institutional-grade stock selection tools directly inside your Ynves Portfolio Cockpit.

Want to see how your portfolio holdings score under the newly evolved Y-VQS engine? Log into your dashboard today and explore your Y-Vertex Profile!