Trading Model Regime Fidelity Index (TMRFI)

GMI-EU-027Energy & UtilitiesIntegration-Required

Definition

Measures whether the AI bid optimization strategy is generating positive alpha (value above a naive benchmark) in the current market regime, or whether regime change has eroded the model's edge.

The Measurement Gap

Traditional energy trading and risk management compares actual P&L to budget or plan. Budget reflects expectations set at the beginning of the period; it does not reflect what a passive, uninformed strategy would have earned given the exact market conditions that occurred. This means all of the variance between actual results and budget reflects market conditions, weather, and AI performance combined: with no way to isolate the AI's contribution.

The specific fall-short: when trading results are poor, the standard narrative is "the market was unfavorable." When results are strong, the narrative is "the AI added value." Neither is necessarily true. A brutal winter may produce strong P&L even with a value-destructive AI strategy. A mild year with stable prices may produce weak P&L even with a high-quality model. Without a market-condition-controlled benchmark, trading AI performance is fundamentally unauditable.

The exact operational decision TMRFI enables:

whether to continue running the AI bid strategy, retrain it with current-regime data, or revert to a simpler benchmark while the model is recalibrated.

What goes wrong without this information:

a trading AI that is destroying value relative to a naive strategy continues operating because absolute P&L is positive in a favorable market. The model's regime misalignment only becomes visible during an adverse period. When the cost of the misalignment is highest and the ability to act on it is most constrained.

How AI/ML makes this measurement possible:

AI trading platforms log each recommendation separately from the execution record. This recommendation log is what enables the benchmark comparison: without it, there is no way to know what the AI would have done differently from what was executed.

Formula

TMRFI = (P&L_AI_recommended − P&L_benchmark) / |P&L_benchmark|

Variables:

P&L_AI_recommended: profit/loss of trades recommended by the AI optimization model in the measurement period ($). Source: trading optimization platform recommendation log cross-referenced to ETRM settlement records

P&L_benchmark: profit/loss of the naive benchmark strategy (day-ahead price-taker or historical average hedge ratio) over the identical period, computed using the same market data ($). Source: derived calculation applied to market data feeds

|P&L_benchmark|: absolute value used as denominator; when benchmark P&L approaches zero (stable low-volatility periods), use absolute dollar difference rather than ratio to avoid instability

Calculation Frequency: Monthly, with a rolling 3-month trend computed alongside the current-month value.

Edge Cases:

Separate weather-driven P&L variance from model-driven variance where feasible. Do not credit the AI for a price spike driven by an extreme cold snap if the model's recommendation did not specifically anticipate it. Minimum 30 trading days per measurement period.

Output:

Fractional ratio. Positive values indicate the AI generates alpha above the benchmark. Negative values indicate the model is value-destructive relative to a naive strategy. Benchmark threshold: TMRFI above 0.05 (5% alpha) justifies continued model use. Values negative for two consecutive months indicate value destruction. Values negative for three consecutive months trigger reversion to benchmark strategy and model retraining review.

Proposed formula, pending validation pilots.

At a Glance

Category
Grid Operations, Dispatch, and Edge Autonomy

Decision Supported

Is the trading AI still adding value, or has the market regime shifted enough that the model's strategy is now value-destructive?

← Hedge Concentration Risk Index (HCRI)Trim Deferral Risk-Adjusted Savings (TDRAS) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment