lending-risk-brief
GitHub生成组合级信贷风险简报，涵盖集中度、账龄表现、评级迁移叙事、宏观压力测试及观察名单。适用于撰写风险报告、信审会材料或季度资产质量回顾。
触发场景
安装
npx skills add mohitagw15856/pm-claude-skills --skill lending-risk-brief -g -y
SKILL.md
Frontmatter
{
"name": "lending-risk-brief",
"homepage": "https:\/\/mohitagw15856.github.io\/pm-claude-skills\/skill\/lending-risk-brief.html",
"metadata": {
"openclaw": {
"emoji": "🏦"
}
},
"description": "Write a portfolio-level lending risk brief: concentration analysis by sector, geography and single name, vintage performance, migration matrix narrative, macro-sensitivity scenarios, top watch names, and actions. Use when asked to write a portfolio risk report, credit risk committee brief, loan book review, or quarterly portfolio quality update. Produces a structured risk brief with concentration tables, migration narrative, scenario read, watch list, and recommended actions."
}
Lending Risk Brief Skill
A loan book fails in patterns before it fails in names. This skill writes the portfolio-level brief that makes the patterns visible: where the book is concentrated, which vintages are misbehaving, which way the grades are migrating, what the macro could do to it, and which names need decisions now.
What This Skill Produces
- Concentration analysis: sector, geography, single-name — each against limits
- Vintage performance comparison
- A migration-matrix narrative (not just the matrix)
- Macro-sensitivity scenarios (base / adverse / severe) with named transmission channels
- A top-10 watch-names table
- Recommended actions with owners
Required Inputs
Ask for what's available; compute what the data supports and mark the rest [data gap — request from portfolio systems]:
- Portfolio snapshot — exposures by borrower, sector, geography, grade, origination vintage
- Concentration limits from the risk appetite statement, if set
- Grade migrations this period (upgrades/downgrades by exposure)
- Delinquency/NPL and provision figures, current and prior periods
- Watch-list candidates already known to the team
Portfolio Framework
1. Concentration. Three cuts, each vs its limit (or vs a stated reference norm if no limit exists — and flag the missing limit as a finding): top sector and top-3 sector share; geographic share; single-name — top-10 and top-20 obligor share, largest single exposure vs capital. Concentration that grew via passive drift (runoff elsewhere) deserves the same flag as active growth — say which it was. Correlated concentrations count together (e.g. construction lending + commercial-real-estate collateral is one bet, not two).
2. Vintage performance. Compare cohorts at the same age on-book (delinquency/default at month 12, 24…), not calendar snapshots — a young book always looks clean. A vintage underperforming its age-matched predecessors signals an underwriting-standards question for that origination period; name the period and what changed in criteria then, if known.
3. Migration narrative. Report net migration by exposure, not count. The narrative must answer: is movement drift (broad one-notch slippage → macro/sector pressure) or jumps (multi-notch falls → underwriting or monitoring misses)? Which sectors drive the downgrades? Are downgrades arriving before delinquency (grading works) or after (grading lags — a finding in itself)?
4. Macro scenarios. Base / adverse / severe. For each: the named driver (rates, unemployment, property values, sector shock) and its transmission channel into this specific book ("+200bps hits the 34% of book on floating rate at refinance; DSCR<1.2x share rises from X to Y [compute from data]"). Severity framing over precision — label all scenario numbers as estimates.
5. Watch names. Top 10 by exposure-weighted concern: name/ref, exposure, grade and recent movement, the concern in one sentence, the action and its owner and date.
6. Actions. Each tied to a finding: limit proposals, sector pause/tighten, deep-dive reviews, provision considerations, data fixes. An observation without an action is a gap — either act or state why watching is the action.
Output Format
Portfolio risk brief: [portfolio / as-at date]
1. Headline read — 3–4 sentences: direction of book quality and the one thing committee must decide. 2. Concentration — table per cut: segment | exposure | share % | limit | headroom | trend. 3. Vintage performance — cohorts at matched age, worst vintage named. 4. Migration — net migration by exposure + the drift-vs-jumps narrative. 5. Scenarios — base/adverse/severe: driver | transmission channel | estimated impact. 6. Top-10 watch names — ref | exposure | grade Δ | concern | action | owner | date. 7. Actions — numbered, each tied to its finding, with owner.
End with: "This brief is analytical support, not a credit, provisioning, or capital determination. Decisions follow your institution's risk policy and applicable regulation."
Quality Checks
- Every concentration cut is compared to a limit, or the absent limit is flagged as a finding
- Correlated concentrations are counted together, not reported as separate comfort
- Vintages compared at matched age on-book, not calendar date
- Migration reported by exposure with a drift-vs-jumps interpretation
- Each scenario names its transmission channel into this book, not a generic macro headline
- Every watch name and every finding has an action with an owner
- Estimated figures labelled as estimates; missing data marked
[data gap]
Anti-Patterns
- Do not let a young book's low arrears pass as quality — age-match or say you can't
- Do not present the migration matrix without the narrative — the matrix is data, the drift-vs-jumps read is the analysis
- Do not report single-name and sector concentration as independent when they overlap in the same names
- Do not write a scenario without its transmission channel into this specific book
- Do not list an observation without an action or an explicit "monitor, because…"
- Do not invent portfolio statistics — compute from provided data or mark the gap
版本历史
- 54fad50 当前 2026-07-19 12:24
