Credit Risk Calculator

Credit Risk Calculator

When to Activate

Trigger when user asks: "what's the default probability for [borrower]?", "what if revenue drops 20%?", "recalculate PD with adjusted inputs", "explain the credit risk drivers", "SHAP waterfall", "risk score", "PD sensitivity", "which borrowers are highest risk", "what drives the risk rating"

Workflow

Step 1: Get Current Features

Tool: credit_risk_analyzer or credit_portfolio_analyzer

Query: "Current credit features for borrower [name]"

Extract: All model input features (financial ratios, covenant metrics, ESG scores, macro indicators)

Step 2: Present Baseline Profile

Present:

• Headline: "[Borrower] PD: [X.X]% ([RISK_RATING]), [trend] from prior quarter"
• Feature summary table:

STOPPING POINT

Present the baseline, then offer: "I've pulled [Borrower]'s current risk profile. I can:

• Run baseline PD calculation (recalculate probability of default from current features)
• Model a stress scenario (specify revenue drop, leverage increase, rate rise, or covenant breach)
• Compare multiple scenarios side by side (e.g., +0.5x / +1.0x / +1.5x leverage)
• Explain key risk drivers (SHAP waterfall showing exactly what drives the PD score)

Which would be most useful?"

Step 3a: Baseline PD Calculation (if user chooses)

Tool: code_execution

You MUST use ONLY the code from pd_model.py. Copy the relevant function(s) into the code_execution tool and call them with the data from Step 1. Do NOT write custom scoring code, do NOT import libraries other than numpy and pandas, and do NOT attempt alternative approaches.

Convert borrower features from Step 1 into a pandas DataFrame, then call predict_pd(model, features).

Present:

• PD Score: X.X% (Risk Rating: [LOW_RISK / MODERATE / ELEVATED / HIGH_RISK])
• Comparison to prior quarter PD

Step 3b: Stress Scenario (if user specifies a what-if)

Apply user scenario adjustments:

• "Revenue drops 20%" → adjust REVENUE and derived ratios
• "Interest rates rise 200bps" → adjust INTEREST_COVERAGE, DSCR
• "Leverage increases 1x" → adjust LEVERAGE_RATIO
• "Covenant breach" → set BREACH_COUNT += 1, adjust headroom

Tool: code_execution with apply_scenario(features, adjustments) then predict_pd

Present:

Step 3c: Multi-Scenario Comparison (if user chooses)

Run multiple scenarios side by side using code_execution:

Include: "Threshold analysis: [Borrower] crosses into HIGH_RISK at [X.X]x leverage"

Step 3d: SHAP Waterfall (if user chooses risk drivers)

Tool: code_execution with explain_with_shap(model, features, feature_names)

Present:

• Base PD: [X]% → Final PD: [Y]% after all feature contributions
• Key insight: "[Feature] is the single largest risk driver, contributing [X]% to PD"

Rating Boundaries

Audience-Specific Presentation

• Risk Committee: Headline PD + rating + top 3 SHAP drivers (concise)
• PM/Analyst: Full SHAP waterfall + multi-scenario comparison + threshold analysis
• Client/Investor: Plain language — "This borrower's credit quality is [strong/adequate/under pressure] with key risks from [factor]"

Output Template

## Credit Risk Assessment: [Borrower]

**PD Score**: [X.X]% | **Risk Rating**: [RATING] | **Trend**: [Direction]

### Key Risk Drivers (SHAP)
| Feature | Value | Impact | Direction |
|---------|-------|--------|-----------|
| [Top driver] | [Value] | [Impact] | [Direction] |

### Scenario Sensitivity
| Scenario | PD | Rating | Change |
|----------|-----|--------|--------|
| Current | [X]% | [Rating] | — |
| [Stress] | [Y]% | [Rating] | [Change] |

Cross-Skill References

• For covenant status of the borrower → covenant-monitoring skill
• For rate sensitivity impact on coverage → rate-sensitivity-analysis skill
• For portfolio-wide risk overview → credit-portfolio-review skill

Stopping Points

• After Step 2 (baseline profile): Offer 4 branching options
• After any Step 3 branch: "Would you like to run another scenario, or shall I compile the full risk assessment?"