How Flagium Calculates Structural Risk

Structural Risk Calculation Framework

Flagium AI's scoring methodology departs from traditional linear models to reflect the non-linear realities of corporate distress. The calculation passes through seven distinct mathematical operations to arrive at the final risk score:

---

The 7-Step Scoring Pipeline

1. Base Stress Mapping

Each input metric (e.g. Interest Coverage, Debtor Days, Promoter Pledge) is mapped to a raw stress value between 0 and 100 based on sector-specific triggers.

2. Bounded Pillar Caps

To prevent a single metric from dominating the entire score, metrics are grouped into pillars with strict capacity caps:

• Governance: Max 25 points
• Solvency: Max 20 points
• Earnings Quality: Max 15 points
• Balance Sheet Stress: Max 15 points
• Competitive Position: Max 15 points
• Growth Sustainability: Max 10 points

3. Non-Linear Saturation Curves

Distress scaling is non-linear. Minor issues are smoothed out, while severe breaches are amplified. The mapping follows a sigmoidal saturation function:

$\text{Stress}_{\text{saturated}} = \frac{100}{1 + e^{-k(\text{Stress} - x_0)}}$

This ensures that once a company crosses critical thresholds, its risk score escalates rapidly, mirroring institutional liquidation behaviors.

4. Duration-Based Persistence Logic

Risk is sticky. A flag active for 4 consecutive quarters carries more structural weight than a temporary spike:

• 1 Quarter: 0.8x multiplier
• 2 Quarters: 1.0x baseline
• 4 Quarters: 1.1x boost
• 8+ Quarters: 1.2x penalty

5. Correlation Suppression (Clustered Discounting)

When multiple overlapping signals fire (e.g. gross margin, operating margin, and NIM all compressing), the engine suppresses subsequent signals to avoid double-penalizing the company:

• 1st Signal: 100% impact
• 2nd Signal: 70% impact
• 3rd Signal: 40% impact
• 4th+ Signal: 20% impact

6. Compounding Interaction Overlays

When distress channels cross-infect (such as high leverage combining with a sudden governance qualification), the overall system fragility spikes. An interaction multiplier of 5% to 15% is dynamically applied.

7. Temporal Stability Constraints (Drift Control)

To filter out noise from quarterly reporting variations, the engine restricts score volatility to $\pm 10$ points per quarter, unless a major regulatory event or default triggers an immediate bypass.