What is G-Score? The Edge Detection Metric for Systematic Trading
April 4, 2026 · By Ashim Nandi
G-Score is a proprietary metric developed by System R that measures whether a trading strategy's edge is real or illusory by applying the geometric growth criterion. Unlike arithmetic averages that can show positive returns while capital actually shrinks, G-Score reveals the true compounding potential of a strategy after accounting for the destructive effect of volatility. It answers the question every systematic trader should ask before deploying capital: will this edge actually grow my account over time?
The Problem G-Score Solves
Most traders evaluate performance using arithmetic averages. A strategy that gains 50% one year and loses 50% the next shows a 0% average return. That looks like breakeven.
It is not breakeven.
Start with $100,000. Gain 50% to reach $150,000. Lose 50% and you are at $75,000. A 25% loss disguised as a 0% average. The arithmetic lied.
This is not an edge case. This is the fundamental mathematics of compounding, and it affects every equity curve in existence. The gap between what traders think they are earning and what they are actually earning is driven entirely by volatility.
Mark Spitznagel of Universa Investments coined the term volatility tax to describe this effect. The mathematical relationship is precise:
Geometric Return ≈ Arithmetic Return - σ²/2
Your real return is always less than your apparent return. The gap is a direct function of variance.
G-Score takes this relationship and turns it into a decision tool.
The G-Score Formula
G-Score is built on the geometric growth criterion:
G = E[R] - σ²/2
Where:
| Variable | Meaning |
|---|---|
| G | G-Score, the geometric growth rate |
| E[R] | Expected return (arithmetic mean of the strategy's returns) |
| σ² | Variance of returns (standard deviation squared) |
| σ²/2 | The volatility tax, the compounding cost of variance |
A positive G-Score means the strategy's expected return is large enough to overcome the volatility tax. A negative G-Score means volatility is silently destroying capital, even if the arithmetic average looks profitable.
Three Scenarios That Reveal the Power of G-Score
Scenario 1: Low volatility, positive edge
- Arithmetic return: 10%
- Volatility (σ): 15%
- Volatility tax: 0.15² / 2 = 1.125%
- G-Score: +8.875%
The edge is real. Capital compounds.
Scenario 2: Moderate volatility, same return
- Arithmetic return: 10%
- Volatility (σ): 30%
- Volatility tax: 0.30² / 2 = 4.5%
- G-Score: +5.5%
Still positive, but nearly half the growth is consumed by volatility. The strategy works, but the cost of variance is high.
Scenario 3: High volatility, same return
- Arithmetic return: 10%
- Volatility (σ): 50%
- Volatility tax: 0.50² / 2 = 12.5%
- G-Score: -2.5%
The arithmetic says 10% annual growth. The reality is capital destruction. The strategy feels profitable while slowly bleeding the account dry.
This is exactly the kind of edge G-Score was designed to expose.
How G-Score Is Calculated
Calculating G-Score requires three inputs from a strategy's track record:
Step 1: Compute the arithmetic mean return. Sum all period returns and divide by the number of periods. Daily, weekly, or monthly returns all work, but the period must be consistent.
Step 2: Compute the variance of returns. Calculate the standard deviation of the same return series, then square it.
Step 3: Apply the geometric growth criterion. Subtract half the variance from the mean return.
For a strategy with the following monthly returns over 12 months:
| Month | Return |
|---|---|
| 1 | +3.2% |
| 2 | -1.5% |
| 3 | +2.8% |
| 4 | +4.1% |
| 5 | -2.3% |
| 6 | +1.9% |
| 7 | -0.8% |
| 8 | +3.5% |
| 9 | -1.2% |
| 10 | +2.6% |
| 11 | +1.4% |
| 12 | -0.7% |
- E[R] = 1.08% per month
- σ = 2.18%
- σ²/2 = 0.024%
- G-Score = +1.06% per month (annualized: approximately 12.7%)
This is a healthy G-Score. The edge survives the volatility tax.
But here is the critical caveat: sample size matters enormously. Twenty trades showing positive G-Score carries almost no statistical significance. Two hundred trades across multiple market regimes starts to mean something. Three hundred or more gives real confidence. The law of large numbers applies directly to G-Score validation.
Why G-Score Matters More Than Sharpe or Sortino
The Sharpe ratio and Sortino ratio are the standard performance metrics in finance. Both have limitations that G-Score addresses directly.
| Metric | What It Measures | Key Limitation |
|---|---|---|
| Sharpe Ratio | Return per unit of total volatility | Penalizes upside volatility equally with downside |
| Sortino Ratio | Return per unit of downside volatility | Ignores the compounding cost of all variance |
| G-Score | Net geometric growth after volatility tax | Directly answers "does this edge compound?" |
The Sharpe ratio tells you whether returns justify the volatility. The Sortino ratio improves on this by only penalizing downside deviation. Neither directly answers the question that matters most to a trader deploying real capital: will this strategy actually grow my account over time?
G-Score answers that question. A positive G-Score means yes. A negative G-Score means no, regardless of what the arithmetic average, win rate, or Sharpe ratio say.
Consider a strategy with a Sharpe ratio of 1.2. That looks strong by conventional standards. But if the strategy operates with extreme leverage and high variance, the Sharpe can remain above 1.0 while the geometric return is negative. The strategy looks good on paper and destroys capital in practice.
G-Score catches this. Sharpe does not.
How ATOM Calculates G-Score Automatically
In ATOM, System R's trading platform, G-Score is computed automatically across every strategy, every backtest, and every live performance window. Traders do not need to run the calculations manually.
ATOM pulls the return series from a strategy's track record, computes the arithmetic mean and variance, and applies the geometric growth criterion. The result appears alongside other performance metrics, but with one key difference: G-Score is the metric ATOM uses internally to determine whether a trading edge is real.
When G-Score turns negative, ATOM flags the strategy. This is not a soft warning. A negative G-Score means the volatility tax exceeds the expected return. The strategy may win individual trades, it may even show a positive arithmetic average, but it is mathematically destroying capital through compounding.
ATOM also tracks G-Score across different market regimes. A strategy might show a strong G-Score during trending markets and a negative G-Score during range-bound conditions. This kind of regime-aware analysis is where G-Score becomes most powerful. It does not just tell you whether an edge exists. It tells you when the edge exists and when it does not.
G-Score and Position Sizing
G-Score connects directly to position sizing. The geometric growth criterion is mathematically related to the Kelly criterion.
Full Kelly sizing maximizes geometric growth. But full Kelly also produces drawdowns that most traders cannot survive. This is why professionals use fractional Kelly, typically quarter to half Kelly.
G-Score provides the reality check. If a strategy's G-Score is barely positive, even modest position sizing will push the effective G-Score negative once transaction costs, slippage, and real-world execution are factored in. The strategy's edge is too thin.
If a strategy's G-Score is strongly positive, it can absorb the friction of real trading and still compound. Position sizing can then be optimized within the bounds that G-Score defines as sustainable.
The relationship is simple: G-Score tells you whether an edge exists. Position sizing determines how aggressively you can exploit it without triggering the volatility tax that G-Score measures.
The Deeper Principle: Ergodicity
The reason G-Score matters goes beyond formulas. Physicist Ole Peters at the London Mathematical Laboratory published research challenging a foundational assumption in economics: ergodicity.
Imagine 100 traders each starting with $100,000 on the same system. After one year, you average results across all 100. Some are up, some are down, some are bankrupt. The average across the group might show a positive return. This is the ensemble average.
Now imagine one trader with $100,000 trading the same system for 100 years. The compound result of that single path through time is the time average.
In a multiplicative process like trading, these two averages are not the same. The ensemble average can be positive while the typical individual path trends toward zero.
G-Score measures the time average, the one that actually matters to a trader with one account and one equity curve. You do not get to average your results across parallel lives. You get one path through time, and G-Score tells you where that path leads.
FAQ
What is a good G-Score?
A G-Score above zero means the strategy compounds capital after accounting for the volatility tax. Above 5% annualized is strong. Above 10% is exceptional. The most important threshold is zero: positive means the edge is real, negative means it is not, regardless of what other metrics show.
Can G-Score be used for crypto or forex?
Yes. G-Score is asset-class agnostic. It measures the relationship between expected return and variance, which applies to any instrument with a return series. In highly volatile assets like crypto, G-Score is especially important because the volatility tax is larger and more likely to turn apparent profits into real losses.
How does G-Score differ from CAGR?
CAGR (Compound Annual Growth Rate) is a backward-looking measure of what actually happened. G-Score is a forward-looking estimate of what should happen based on the strategy's return distribution. CAGR is a single number from a single path. G-Score accounts for the statistical properties that drive future compounding, making it more useful for evaluating whether a strategy should continue to be deployed.
How many trades do I need for a reliable G-Score?
At minimum, 200 trades across at least two distinct market regimes. Below 50 trades, G-Score is dominated by noise. Between 50 and 200, it is directionally useful but not conclusive. Above 300 trades across varying conditions, confidence in the G-Score is high. Sample size discipline is non-negotiable in systematic trading.