Risk Management

Overview

Energy Copilot’s Risk Management module provides quantitative risk analytics for energy derivative portfolios. It combines standard financial risk methodologies (VaR, Greeks) with energy-specific risk factors (price spikes, weather volatility, renewable generation uncertainty).

Value at Risk (VaR)

Energy Copilot implements two VaR methodologies:

Historical Simulation VaR

Historical Simulation VaR uses the actual distribution of historical price returns, without assuming a normal distribution. This captures the heavy tails and skewness characteristic of electricity prices:

VaR(1-day, 95%) = -P5th_percentile(Σ daily_pnl_changes over lookback_window × current_position)

Parameters:

• Lookback window: 250 trading days (1 calendar year)
• Confidence level: 95% (standard) or 99% (regulatory reporting)
• Holding period: 1 day (operational) or 10 days (regulatory)

Monte Carlo VaR

Monte Carlo VaR simulates 10,000 price paths using a mean-reverting jump-diffusion model calibrated to NEM spot prices:

def simulate_nem_price(
    S0: float,         # Current spot price
    kappa: float,      # Mean reversion speed
    theta: float,      # Long-run mean
    sigma: float,      # Volatility
    lambda_j: float,   # Jump intensity (spikes/year)
    mu_j: float,       # Mean jump size
    sigma_j: float,    # Jump size std dev
    T: float,          # Time horizon (days)
    n_sims: int = 10000
) -> np.ndarray:
    """
    Merton jump-diffusion with mean reversion.
    Parameters calibrated separately per region from historical data.
    """
    ...

Note

The NEM spot price exhibits strong mean-reversion combined with infrequent large jumps (price spikes). Standard geometric Brownian motion significantly underestimates tail risk. The jump-diffusion model is calibrated quarterly.

Risk Dashboard

VaR Summary (/middle-office/risk/var)

• VaR at 95% and 99% confidence for each portfolio and the total book
• Comparison: Historical Simulation vs Monte Carlo VaR
• Back-testing: how many days in the past year did actual losses exceed VaR?
• Stressed VaR: VaR recalculated on the 2009 January heatwave scenario

Risk Factor Attribution

The risk is decomposed by driver:

Risk Factor	Description	Method
Region Price Risk	Exposure to spot price movements per region	Delta × price volatility
Tenor Risk	Exposure to curve shape changes	DV01 per tenor bucket
Volatility Risk	Options book exposure to implied vol changes	Vega × vol change
Correlation Risk	Regional price correlation breakdown	Off-diagonal covariance
Jump Risk	Spike event impact on unhedged positions	Monte Carlo jump paths

Greeks for Options and PPAs

Cap and Floor Greeks

For options positions (caps, floors, collars), Energy Copilot calculates Black-76 Greeks:

Greek	Symbol	Interpretation
Delta	Δ	Change in option value per $1 change in forward price
Gamma	Γ	Rate of change of delta per $1 change in forward price
Vega	ν	Change in option value per 1% change in implied volatility
Theta	θ	Daily time decay (option value lost per day)
Rho	ρ	Sensitivity to risk-free rate (small in energy markets)

PPA Greeks

Power Purchase Agreements (PPAs) have price and volume risk:

• Price Delta: sensitivity to change in wholesale forward prices
• Volume Uncertainty: P90/P50/P10 generation scenarios affect actual settlement
• Cannibalisation Risk: correlated generation across renewable fleet reduces captured prices

Credit Risk

Credit risk measures the exposure to counterparty default:

Credit_Exposure = max(0, MtM_value)
Credit_Limit_Utilisation = Credit_Exposure / Credit_Limit × 100

The platform generates alerts when any counterparty’s credit utilisation exceeds 80% of their approved limit.

Credit risk metrics:

• Current Exposure: sum of positive MtM values per counterparty
• Potential Future Exposure (PFE): maximum exposure at 97.5% confidence over deal life
• CVA (Credit Value Adjustment): discount for counterparty default probability

Position Limits Monitoring

Real-time limit monitoring against defined thresholds:

-- Query current limit utilisation
SELECT
    portfolio_name,
    limit_type,
    current_value,
    limit_value,
    ROUND(current_value / limit_value * 100, 1) AS utilisation_pct,
    CASE
        WHEN current_value / limit_value > 1.0 THEN 'BREACH'
        WHEN current_value / limit_value > 0.9 THEN 'WARNING'
        WHEN current_value / limit_value > 0.8 THEN 'CAUTION'
        ELSE 'OK'
    END AS status
FROM energy_copilot.gold.limit_monitoring
ORDER BY utilisation_pct DESC;

API Endpoints

# VaR summary
GET /api/risk/var?portfolio_id=1&confidence=0.95&method=historical

# Monte Carlo simulation results
GET /api/risk/montecarlo?portfolio_id=1&n_sims=10000

# Greeks for options book
GET /api/risk/greeks?portfolio_id=1

# Credit exposure by counterparty
GET /api/risk/credit-exposure

# Limit monitoring status
GET /api/risk/limits?portfolio_id=1