Skip to content This library is free, offered as dana. If it helps your work, donate here →
penguin tree ai
← back to libraryDonate here

This is one of 3,622 free prompts in the penguin tree ai library. Sign up once and the whole library opens, with new prompts by email.

open the library →

Credit risk trend summarizer

Synthesizes raw portfolio credit metrics into a plain-language trend summary across seven risk dimensions, flags anomalies against thresholds and prior periods, and delivers ranked recommended actions with an escalation…

download .md

Purpose

Convert raw portfolio credit metrics into a concise, plain-language trend narrative that identifies where risk is moving, flags statistically or directionally notable anomalies, and recommends specific actions ranked by urgency — so analysts stop hand-writing commentary and risk committees get a decision-ready summary.

When to Use

Invoke this skill when the user asks to:

  • Summarize month-over-month or quarter-over-quarter credit performance
  • Write portfolio commentary or narrative for a risk committee or leadership deck
  • Explain why delinquency, roll rates, or losses moved in a given period
  • Flag anomalies or deteriorating segments in a credit portfolio
  • Turn a metrics table or dashboard export into a written trend summary
  • Prepare talking points for a portfolio review or credit review meeting

Inputs Required

Before starting, collect:

  • Portfolio/segment: What book is being summarized (e.g., unsecured consumer installment, SMB revolving, auto)
  • Reporting period: Current period plus the prior period(s) being compared
  • Current-period metrics: Values for available dimensions — DQ buckets (30/60/90+ DPD), roll rates, vintage curves, net charge-off rate, recovery rate, approval/originations volume, average FICO or score band mix
  • Prior-period metrics: The same metrics for at least one comparison period (ideally 3+ for trend direction)
  • Thresholds/appetite: Any risk appetite limits, early-warning triggers, or target ranges the portfolio is measured against
  • Known drivers: Any policy changes, seasonality, marketing shifts, or macro events that could explain movement
  • Audience: Risk committee / senior leadership / working team (sets depth and tone)

If thresholds are missing, benchmark against the portfolio's own trailing 6-period average and note the assumption. If only two periods are provided, flag that trend confidence is limited.

Instructions

Step 1: Establish the Baseline and Direction

Before summarizing anything, orient the reader:

  • State the portfolio, period, and total exposure/balance if provided.
  • For each metric, compute the period-over-period change in both absolute terms and basis points/percentage points — never report a raw value without its delta.
  • Determine trend direction over the available periods (improving / stable / deteriorating) — a single-period move is a change; three periods in one direction is a trend. Label each accordingly.

Step 2: Assess Across Seven Risk Dimensions

Walk each dimension the data supports. Skip dimensions with no data rather than inventing figures.

  1. Delinquency migration — Movement across 30/60/90+ DPD buckets. Rising early buckets (30 DPD) are a leading indicator; rising late buckets (90+) confirm realized deterioration.
  2. Roll rates — Rate at which accounts progress from one delinquency bucket to the next (e.g., 30→60). Roll-rate acceleration is often the earliest reliable warning before losses appear.
  3. Vintage performance — How recent origination cohorts perform versus older ones at the same months-on-book. A newer vintage curving above an older one signals underwriting or mix drift.
  4. Charge-offs and losses — Net charge-off rate and gross loss trend. Distinguish flow (new losses this period) from stock (cumulative).
  5. Recovery/collections — Recovery rate and post-charge-off cash. Deterioration here compounds net loss even when gross defaults are flat.
  6. Originations mix and quality — Volume, approval rate, and score-band or DTI mix of new business. Front-book shifts explain much of the back-book trend 6–12 months later.
  7. Concentration and macro sensitivity — Exposure clustering by geography, product, score band, or channel, and sensitivity to any macro drivers supplied.

Step 3: Flag Anomalies

For every dimension, apply this flag logic and label each finding:

  • 🔴 Breach: Metric crosses a stated threshold or risk appetite limit.
  • 🟠 Early warning: Metric moved materially (default rule: >15% relative change or >2 std dev from trailing average) but hasn't breached, or a leading indicator is turning.
  • 🟡 Watch: Directional move worth noting but within normal variance. For each flag, state the metric, the magnitude, and the most likely explanation drawing on the known drivers provided (separate correlation from confirmed cause — say which it is).

Step 4: Recommend Ranked Actions

Produce 4–7 actions, ranked by urgency, each with: the trigger it addresses, the specific action, and the owner function (Underwriting / Collections / Policy / Analytics / Committee). Order: breaches first, then early warnings, then structural improvements. Every action must tie back to a flagged finding — no generic best practices.

Step 5: Make the Escalation Call

State one clear judgment: does this period warrant no escalation, a watch item, or immediate committee attention? Be decisive and give the single reason.

Output Format

  • Structure: Executive summary paragraph → dimension-by-dimension trend read → flagged anomalies list → ranked action table → escalation call
  • Length: 600–900 words
  • Format: Plain-language prose for the summary and dimension reads; bulleted flags; a numbered or table format for ranked actions
  • Sections: Executive Summary → Trend by Dimension → Flagged Anomalies → Recommended Actions (ranked) → Escalation Call
  • Tone: Credit risk analyst briefing a committee — precise, quantified, plain-spoken. No jargon without the number behind it; no hedging where the data supports a call.

Quality Checklist

Before delivering output, verify:

  • [ ] Every metric cited includes its period-over-period delta (absolute and bps/pp) — no bare values
  • [ ] Each dimension is labeled improving / stable / deteriorating based on the number of periods available
  • [ ] Every anomaly carries a flag level (🔴/🟠/🟡), a magnitude, and a stated likely driver
  • [ ] Correlation and confirmed causation are explicitly distinguished for each driver claim
  • [ ] Every recommended action maps to a specific flagged finding and names an owner function
  • [ ] A single, decisive escalation call is made — not left open

Notes

  • Leading indicators (30 DPD entry, roll-rate acceleration, front-book mix shift) deserve more weight in the narrative than lagging ones (net charge-offs) — losses confirm what earlier signals already predicted.
  • Watch for Simpson's paradox: a stable blended metric can hide a deteriorating segment offset by a growing healthy one. Always check whether mix shift is masking underlying trend.
  • When only two periods are available, frame findings as changes, not trends, and say so — false trend confidence is a common failure in short-window commentary.

How to use this skill

This file is a Claude Code skill in Anthropic's standard format — markdown with YAML frontmatter at the top.

Installing in Claude Code:

  • ~/.claude/skills/<folder-name>/SKILL.md — auto-loads in every session anywhere on your machine (recommended for skills you'll use across multiple projects)
  • <your-project>/.claude/skills/<folder-name>/SKILL.md — auto-loads only when working in that project (recommended for project-specific skills)

Restart Claude Code or open a new session to register.

Using in Claude.ai, ChatGPT, Gemini, or other AI tools:

  1. Drag this file into a new chat, or copy and paste its contents
  2. Send your task or request as your next message — the AI will apply the skill where it fits

For Claude Code skill details: https://docs.anthropic.com/en/docs/claude-code/skills