Risk Models

Risk models adjust target weights. They do not inspect pending orders and they do not block order-mode submissions. In the current engine they are a weight-mode overlay between raw weights and rebalancing.

Engine Contract

Risk models answer: "Should these target weights be modified before they become realized exposure?" They are not stop-losses and they are not broker order checks.

Level 1: Position Limit

from backtester.risk import PositionLimitModel

strategy = MyStrategy(
    ...,
    execution_mode="weights",
    risk_model=PositionLimitModel(
        max_weight=0.25,
        max_total_leverage=1.0,
    ),
)

Level 2: Volatility Target

from backtester.risk import VolTargetModel

strategy = MyStrategy(
    ...,
    risk_model=VolTargetModel(
        target_vol=0.15,
        lookback=63,
        max_leverage=1.5,
        rebalance_freq="BMS",
    ),
)

Level 3: Risk Chain

Apply multiple models in sequence.

from backtester.risk import RiskModelChain, VolTargetModel, PositionLimitModel

risk_model = RiskModelChain([
    VolTargetModel(target_vol=0.12, lookback=63, max_leverage=1.5),
    PositionLimitModel(max_weight=0.20, max_total_leverage=1.0),
])

strategy = MyStrategy(..., risk_model=risk_model)

Data Contract

Failure Modes

• Expecting a risk model to reject raw order submissions in order mode.
• Applying leverage caps without checking row gross exposure.
• Forgetting that risk model output still goes through rebalancing and drift.
• Treating a volatility target as a guarantee of realized future volatility.

Related Guides

• Weights
• Rebalancing
• Order Lifecycle