risk-credit
Calculate portfolio credit VaR from obligor exposure, PD and LGD, either with the Vasicek asymptotic single risk factor quantile (the model behind the Basel IRB formula, CRE31) or with a seeded default-mode one-factor Gaussian copula Monte Carlo for a finite portfolio, giving expected loss, unexpected loss, credit VaR and Expected Shortfall. Use for default loss on a loan or bond book. Not for counterparty exposure (use risk-counterparty), not for market VaR (use risk-tail), not for IRB capital or risk weights, and not for rating migration.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-14
- Stars
- ★ 1.6k
- Forks
- 491
01 前言 Front Matter
- name
- risk-credit
- description
- Calculate portfolio credit VaR from obligor exposure, PD and LGD, either with the Vasicek asymptotic single risk factor quantile (the model behind the Basel IRB formula, CRE31) or with a seeded default-mode one-factor Gaussian copula Monte Carlo for a finite portfolio, giving expected loss, unexpected loss, credit VaR and Expected Shortfall. Use for default loss on a loan or bond book. Not for counterparty exposure (use risk-counterparty), not for market VaR (use risk-tail), not for IRB capital or risk weights, and not for rating migration.
02 资源文件 3
- 参考文档reference.md4.2 KB
- 脚本scripts/check_artifact.py2.6 KB
- 参考文档workflows/credit-var.md1.2 KB
03 SKILL.md 正文
Risk Credit
Two tools, both read-only.
risk_credit_asrf: the Vasicek asymptotic single risk factor loss quantile. Fast and exact for an infinitely granular portfolio; it ignores name concentration. CLI:riskdesk credit-asrf --input PATH.risk_credit_copula: a seeded Monte Carlo of the same one-factor model for the actual, finite list of obligors. It captures concentration and gives Expected Shortfall. CLI:riskdesk credit-copula --input PATH.
Examples: examples/credit_asrf.json and examples/credit_copula.json (one six-obligor book).
The contract, in short
Both take as_of, currency, source, data_mode, confidence (strictly between 0 and 1),
horizon_days (the horizon over which every PD applies) and obligors, each with a unique
id, positive exposure (EAD in the request currency), pd (strictly between 0 and 1) and
lgd (0 to 1).
- ASRF obligors add
correlationR (strictly between 0 and 1), the asset correlation with the single factor. - Copula obligors add
factor_loadinga (strictly between -1 and 1). The pairwise asset correlation is a_i * a_j, so R = a^2 for a homogeneous pool. The request addsscenarios(1,000 to 1,000,000) andseed(non-negative integer). Scenarios times obligors may not exceed 200,000,000.
PD, LGD, correlation and loadings are inputs. The tools do not estimate them, and you must not invent them. A supervisory correlation (the CRE31.5 corporate formula, for example) may be used only if the user asks for it; say so when you do.
What to report first
degradedanddegraded_reason. For the copula, degraded means fewer than 20 scenarios of tail probability mass: raisescenarios.credit_varatconfidenceoverhorizon_days, with currency.expected_loss,unexpected_loss(credit VaR minus expected loss), and for the copulaexpected_shortfall,scenariosandseed.- For the ASRF, the largest
contributions. They add up to the credit VaR.
Sign conventions
Loss is positive. Credit VaR and Expected Shortfall are losses at and beyond the quantile.
Unexpected loss here is credit VaR minus expected loss, the quantity the IRB formula capitalises
before its maturity adjustment. It is not the standard deviation of loss, which the copula
reports separately as loss_standard_deviation.
What not to claim
- Neither tool gives IRB capital or a risk weight. They apply no maturity adjustment, no CRE31.7 correlation multiplier and no PD or LGD floor.
- The ASRF result assumes infinite granularity. For a concentrated book it understates the loss quantile; the copula shows by how much.
- The copula result carries Monte Carlo error. Quote the seed and scenario count with it.
- Default mode only. Rating migration and spread loss are not modelled. LGD is fixed, not random. There is one systematic factor, so sector concentration is not captured beyond what the loadings express.
- A loss quantile is a model output, not a forecast that has been backtested.
Going deeper
reference.md: the formulas, the derivation of quantile additivity, the CRE31.5 source, and the published and exact checks the tests use.workflows/credit-var.md: which tool to run and in what order.scripts/check_artifact.py RESULT.json: one-line verdict on a saved result.
04 同仓库更多 Skills
worktable-review-with-annotations
Create, inspect, reply to, update, or resolve situated Worktable annotations. Use when feedback belongs to a Doc section, block, text range, or HTML Doc, or when the user asks to address existing review feedback and coordinate a requested artifact edit.
worktable-find-and-synthesize-context
Find, assess, and synthesize context from Worktable Docs and Records. Use when the user asks what a workspace contains, wants workspace-backed research, needs current facts reconciled across artifacts, or requests a cited synthesis without changing workspace state.
worktable-create-or-update-docs
Create, edit, rename, or maintain narrative Worktable Docs. Use for plans, decisions, briefs, notes, research syntheses, and other durable prose when the user requests a new artifact or a safe update to an existing one.
worktable-create-or-manage-records
Organize repeated structured information as durable Worktable Records. Use for research datasets, inventories, directories, pipelines, trackers, and other collections whose items need identity, fields, filtering, links, validation, or individual updates.
worktable-create-or-manage-interactive-html
Create or improve durable visual and interactive artifacts in Worktable. Use for dashboards, trackers, calculators, mini apps, slide presentations, and other requests that benefit from a polished HTML Doc.
worktable-collaborate-threads
Read, answer, and hand off messages through durable Worktable threads with users or other authenticated participants. Use when the user asks to check addressed messages, reply to a participant, continue a thread, request another agent's help, wait briefly for a response, or watch for comments and messages sent to the agent.