Risk Aversion

Uses mean-variance utility from financial portfolio theory. Each offer’s score is penalized by its variance — offers with uncertain outcomes are scored lower than offers with consistent (even if slightly lower) acceptance rates.

Algorithm

Config value: "approach": "behaviorAlgos", "sub_approach": "riskAversion"

For each offer, the algorithm computes a risk-adjusted utility score:

\(p = \frac{\text{response\_count}}{\text{count}}\)

\(\text{variance} = p \times (1 - p)\)

\(\text{utility} = p - \frac{\gamma}{2} \times \text{variance}\)

Where \(\gamma\) is the risk aversion coefficient. Scores are aggregated per offer and normalized to sum to 1.0.

An offer with 50% acceptance and low variance can beat an offer with 60% acceptance and high variance when the risk aversion coefficient is sufficiently high.

Parameters

• riskAversionCoefficient (\(\gamma\)): Higher values penalize variance more heavily. Default: 1.0. Range: 0 (risk-neutral, equivalent to raw acceptance rate) to >2 (highly risk-averse).
• Processing Window: Time window in milliseconds for historical data.
• Historical Count: Max records to process per update cycle.

Cold Start

Recommendations are always returned. The real-time training path in RollingBehavior always produces a scored options array, regardless of whether the algorithm has interaction history:

• No history: Every offer in the options store receives a uniform random score. All offers are ranked and passed to the post-score class.
• Partial history: Offers scored by Risk Aversion receive their risk-adjusted utility; offers the algorithm has not seen receive a random score. Both are included in the result.
• Established history: Risk Aversion scores drive the ranking for all known offers.

The scored options are then sorted by arm_reward and handed to the configured dynamic post-score class (e.g. PlatformDynamicEngagement), which controls the final offer selection, eligibility filtering, and response formatting.

When To Use

• When consistency matters more than maximum expected return
• When you want to avoid “volatile” offers with unpredictable acceptance rates
• Financial services, insurance, compliance-driven contexts where predictability is valued
• After sufficient interaction history has accumulated (the platform handles cold start with random scores until then)

When NOT To Use

• When you want to maximize expected acceptance regardless of variance
• When all offers have similar variance (the algorithm adds no value)
• When you need built-in exploration (combine with epsilon at the deployment level)

Example

from prediction.apis import deployment_management as dm
from prediction.apis import online_learning_management as ol
from prediction import jwt_access
 
auth = jwt_access.Authenticate("http://localhost:3001/api", ecosystem_username, ecosystem_password)
 
deployment_id = "demo-risk-aversion"
 
online_learning_uuid = ol.create_online_learning(
        auth,
        algorithm="ecosystem_rewards",
        name=deployment_id,
        description="Risk Aversion configuration",
        feature_store_collection="set_up_features",
        feature_store_database="my_mongo_database",
        options_store_database="my_mongo_database",
        options_store_collection="demo-deployment_options",
        randomisation_processing_count=5000,
        randomisation_processing_window=2592000000,
        contextual_variables_offer_key="offer",
        create_options_index=True,
        create_covering_index=True
)
 
online_learning = dm.define_deployment_multi_armed_bandit(epsilon=0, dynamic_interaction_uuid=online_learning_uuid)
 
parameter_access = dm.define_deployment_parameter_access(
    auth,
    lookup_key="customer_id",
    lookup_type="string",
    database="my_mongo_database",
    table_collection="customer_feature_store",
    datasource="mongodb"
)
 
deployment_step = dm.create_deployment(
    auth,
    project_id="demo-project",
    deployment_id=deployment_id,
    description="Risk Aversion demo deployment",
    version="001",
    plugin_post_score_class="PlatformDynamicEngagement.java",
    plugin_pre_score_class="PreScoreDynamic.java",
    scoring_engine_path_dev="http://localhost:8091",
    mongo_connect=f"mongodb://{mongo_user}:{mongo_password}@localhost:54445/?authSource=admin",
    parameter_access=parameter_access,
    multi_armed_bandit=online_learning
)