Tuning Market Manipulation Model Sensitivity

The Market Manipulation model uses advanced AI and behavioral analytics to flag potential compliance risks, such as spoofing, layering, front-running, or insider trading across your communication channels.

To maintain the optimal balance between mitigating regulatory risk and managing team alert fatigue, compliance administrators can adjust the sensitivity thresholds of this model. This guide explains how to safely tune the model's sensitivity based on your firm’s risk appetite.

Prerequisites

• Permissions: You must have System Administrator or Compliance Risk Manager privileges.

• Environment: It is highly recommended to test adjustments in a UAT/Sandbox environment before deploying changes to Production.

Step-by-Step: Adjusting the Sensitivity Threshold

1. Navigate to the Policy Management Panel Log into the Shield platform, click on the Settings gear icon in the bottom-left navigation bar, and select Policies & AI Models.

2. Locate the Market Manipulation Model Under the AI Classifiers tab, locate the Market Manipulation model card. Click Edit Settings.

3. Modify the Sensitivity Slider You will see a slider indicating the current sensitivity score (typically from 1 to 5, or 10% to 100%).

   • Increasing Sensitivity (Higher Value): Captures more subtle behavioral anomalies and keyword correlations. Note: This will increase the volume of alerts and may generate more false positives.

   • Decreasing Sensitivity (Lower Value): Restricts flags to only high-confidence matches. Note: This reduces alert volume but increases the risk of missing marginal violations.

4. Refine Lexicon Weights (Optional) If your model configuration includes sub-triggers (e.g., specific terms associated with pumping or collusive behavior), you can manually adjust individual weights to further fine-tune what triggers an alert.

5. Run a Policy Simulation (Dry Run) Before saving, click Run Simulation. Select a historical 7-day data sample to see how the new sensitivity threshold would have impacted your alert volume.

6. Save and Deploy If the simulated alert volume looks manageable, click Apply Changes. Select Deploy to Production.

Best Practices for Sensitivity Tuning

Warning: Dropping model sensitivity too low can result in regulatory non-compliance. Always document the rationale behind sensitivity adjustments for audit purposes.

• Incremental Adjustments: Do not make drastic changes all at once. Adjust the sensitivity score by no more than 10-15% at a time, then monitor the alert volume for 48 hours.

• Desk-Specific Policies: Instead of applying a blanket global policy, consider creating scoped policies. For example, you might want higher sensitivity for Equities trading desks compared to back-office support functions.

• Quarterly Reviews: Market jargon and regulatory focus areas shift. Schedule a quarterly review of your false-positive rates to ensure the model remains aligned with current market conditions.