↪ [[default rate - Snapshot]] # default rate - Reading *2026-06-30 · mode: light · type: scientific concept (credit risk) · 7 sources · 8 dimensions covered* ## Definitions & Glossary **Primary source:** Federal Reserve / BIS research notes ✅ Default rate = realized (ex-post) measure of how many borrowers/issuers actually defaulted over a period, vs. **probability of default (PD)** = forward-looking (ex-ante) estimate of likelihood of default. The two are easily conflated but temporally opposite: "the number of defaulted borrowers divided by the number of total borrowers within a certain portfolio is known as the observed default rate, while the predicted one is called the expected default rate." (corroborated across multiple independent sources; primary Open Risk Manual page is login-gated — ⚠️ secondary corroboration only) > "Default Rate is calculated as the amount in default over the past 12 months divided by the total outstanding volume of loans that are not in default at the beginning of the 12-month period." Source: BIS / Federal Reserve research ✅ ### Glossary | Term | Definition | Source | |------|------------|--------| | Default rate | Realized % of portfolio that defaulted in period | BIS/Fed ✅ | | Probability of Default (PD) | Forward-looking estimate of default likelihood | Wikipedia (IRB) ◽ | | Default (Basel) | Borrower 90 days past due OR unlikely to pay in full | BIS Basel Framework CRE36 ✅ | ## Domain & How It Works Basic formula: defaulted loans ÷ total loans outstanding at period start (e.g. 20/1000 = 2%). Two methods used in practice: **cohort method** (track fixed borrower group over time) and **portfolio-based** (rolling 12-month window). Threshold for "default" varies by convention — commonly 90 or 270 days past due, not universally fixed. Source: BIS, Wall Street Oasis ✅/⚠️ ## Alternatives / Before | Alternative | What it is | Key difference | Source | |-------------|------------|----------------|--------| | Delinquency rate | % of loans 30+ days past due, still accruing interest | Earlier-stage signal, loan still "alive" | Federal Reserve ✅ | | Charge-off rate | Loans written off as loss, net of recoveries, annualized | Realized loss, later-stage than default | Federal Reserve ✅ | | Probability of Default (PD) | Forward-looking model estimate | Ex-ante vs default rate's ex-post | Multiple ◽ | Progression: delinquency → default → charge-off. Default sits between the two. ## Market & Players ### Categories | Category | Role in the stack | |----------|-------------------| | Rating agencies | Publish issuer/portfolio default-rate forecasts (NRSRO status) | | Central bank / regulator data | Official charge-off & delinquency series (macro-level) | ### Feature map | Player | Category | Key features | Differentiator | Positioning | Source | |--------|----------|--------------|----------------|-------------|--------| | Moody's | Rating agency | Issuer-level forecasts, private credit coverage | Baseline + scenario range (1.7–8.3% spec-grade 2026) | Macro + private credit lens | Moody's ⚠️ | | S&P Global | Rating agency | 12-month trailing speculative-grade rate | US 4.4% / Europe 3.5% (Oct 2025) | Trailing realized-rate focus | search snippet ⚠️ | | Fitch | Rating agency | Leveraged loan + HY forecast ranges | Splits US (4.5–5%) vs Europe (3–4.25%) for 2026 | Regional granularity | search snippet ⚠️ | | Federal Reserve / FRED | Official data | Charge-off & delinquency series, all US commercial banks | Free, official, time-series, no model bias | System-wide ground truth | Fed ✅ | Moody's, S&P, Fitch (NRSROs) control ~95% of global rating market; figures diverge by methodology and cutoff — treat any single agency number as one view, not consensus. ## Regulation & Standards | Body | Document | Year | What it requires | Link | |------|----------|------|-----------------|------| | BIS / Basel Committee | Basel Framework CRE36 (IRB approach) | Basel II, updated Basel III | PD for a grade must be "long-run average of one-year default rates"; default = 90 days past due or unlikely to pay; banks need ≥7 borrower grades, annual reviews, 3-yr use test | bis.org/basel_framework/chapter/CRE/36.htm ✅ | ## Adjacent Concepts | Concept | Distinction from default rate | Source | |---------|------------------------|--------| | Loss Given Default (LGD) | % of exposure actually lost after recovery, not whether default happened | Wall Street Prep ◽ | | Exposure at Default (EAD) | Outstanding balance + undrawn commitments at moment of default | numberanalytics ⚠️ (link 403, corroborated elsewhere) | | Expected Loss (EL) | EL = PD × LGD × EAD — default rate feeds PD, only one of three inputs | Multiple ◽ | Common error: treating PD like a credit spread — spreads price in liquidity premium and risk aversion too, not just default risk. ## Artefacts | Type | Name/Link | Notes | |------|-----------|-------| | Live data series | [FRED DRALACBN](https://fred.stlouisfed.org/series/DRALACBN) | Delinquency Rate on All Loans, All Commercial Banks — official, free, downloadable ✅ | | Dataset | Lending Club loan dataset (Kaggle) | ~30k samples, 22% default rate, used for ML default-prediction benchmarks | | Dataset | "Give Me Some Credit" (Kaggle) | DTI, delinquency count, income — classic PD-modeling dataset | ## Research Snapshot | Paper | Authors | Year | Key finding | Epistemic class | Link | |-------|---------|------|-------------|-----------------|------| | Advancing financial resilience: systematic review of default prediction models | Alvi, Arif, Nizam | 2024 | Deep learning (LightGBM, XGBoost) trending toward higher accuracy/citations vs classic statistical models (logistic regression still 70–85% accurate) | ⚠️ empirical finding, not regulatory mandate | [PMC11564005](https://pmc.ncbi.nlm.nih.gov/articles/PMC11564005/) ✅ open access | **Consensus:** emerging — ML/DL methods show measurable edge in research literature, but logistic regression remains respectable and is still what most regulatory IRB models use in practice (interpretability requirement). ## Next: Explore - **Loss Given Default (LGD)** — the other half of Expected Loss; default rate alone says nothing about how much is actually lost - **Basel III IRB calibration** — how "long-run average" default rate gets turned into regulatory capital - **Credit scoring / PD modeling (logistic regression vs ML)** — the techniques that turn historical default rates into forward PD estimates *(Run `/penguru:snapshot <concept>` to go deeper)* --- ## Sources | # | Title | URL | Type | Status | |---|-------|-----|------|--------| | 1 | Fed Notes — Why is the Default Rate So Low | https://www.federalreserve.gov/econres/notes/feds-notes/why-is-the-default-rate-so-low-20210304.html | ✅ | live | | 2 | Moody's — US Corporate Default Risk in 2026 | https://www.moodys.com/web/en/us/insights/credit-risk/private-credit/us-corporate-default-risk-in-2026.html | ⚠️ | live | | 3 | Federal Reserve — Charge-Off and Delinquency Rates | https://www.federalreserve.gov/releases/chargeoff/delallsa.htm | ✅ | live | | 4 | Wikipedia — Internal Ratings-Based Approach | https://en.wikipedia.org/wiki/Internal_ratings-based_approach_(credit_risk) | ◽ | live | | 5 | BIS Basel Framework CRE36 (IRB) | https://www.bis.org/basel_framework/chapter/CRE/36.htm | ✅ | live (text extraction blocked, link verified) | | 6 | PMC — Systematic review of default prediction models (2024) | https://pmc.ncbi.nlm.nih.gov/articles/PMC11564005/ | ◽ | live | | 7 | FRED — Delinquency Rate Series DRALACBN | https://fred.stlouisfed.org/series/DRALACBN | ✅ | live | *Dimensions with "not found": none* *Dead links dropped: wallstreetoasis.com (403), numberanalytics.com (403), openriskmanual.org x2 (login-gated, not cited as primary)*