คู่มือโดยละเอียดเร็วๆ นี้
เรากำลังจัดทำคู่มือการศึกษาที่ครอบคลุมสำหรับ Credit Default Probability กลับมาเร็วๆ นี้เพื่อดูคำอธิบายทีละขั้นตอน สูตร ตัวอย่างจริง และเคล็ดลับจากผู้เชี่ยวชาญ
The Probability of Default (PD) is a core concept in credit risk measurement, representing the likelihood that a borrower — whether a corporation, financial institution, sovereign, or individual — will fail to make contractually required payments within a specified time horizon. PD is one of the three fundamental parameters in the Basel credit risk framework, alongside Loss Given Default (LGD) and Exposure at Default (EAD), and it drives both expected loss and regulatory capital calculations. There are three primary approaches to estimating PD. The structural approach, pioneered by Robert Merton in 1974, models a company's equity as a call option on its assets: default occurs when asset value falls below the debt face value (the 'default barrier') at maturity. The distance to default (DD) — measured in standard deviations of the asset value from the default barrier — is the key output, and PD is estimated from the normal distribution of this distance. Moody's KMV model is the leading commercial implementation of the Merton structural approach, using estimated asset value and volatility to compute the Expected Default Frequency (EDF). The reduced-form (intensity) approach treats default as a surprise event governed by a hazard rate λ (the instantaneous conditional probability of default), which can be calibrated to market prices of credit default swaps (CDS) or bond credit spreads. PD over horizon T = 1 − e^(−λT). This approach is more flexible than the structural model and can incorporate time-varying default intensities. Credit score-based approaches (including logistic regression models, rating agency models, and machine learning models) estimate PD from borrower financial characteristics: leverage ratios, profitability, coverage ratios, liquidity, size, and industry. Rating agency historical default rates provide an empirical mapping from credit rating to historical PD over various horizons. Accurate PD estimation is essential for loan pricing (the credit spread must compensate for expected loss = PD × LGD), regulatory capital calculation (Basel IRB approach allows banks to use internal PD models), portfolio credit risk management, and counterparty credit risk assessment in derivatives transactions.
Merton PD: PD = N(−d2) where d2 = [ln(V/D) + (μ − σ_V²/2)T] / (σ_V √T) Hazard Rate PD: PD(T) = 1 − e^(−λT) Credit Triangle: λ ≈ Credit Spread / LGD | CS ≈ PD × LGD / (1−PD)
- 1Choose the estimation approach: structural (Merton) for public companies with equity data; reduced-form for CDS-traded names; rating-based for companies with credit ratings.
- 2For Merton model: estimate asset value (V) and asset volatility (σ_V) by solving the equity-as-call-option system iteratively; identify the default barrier (typically current liabilities + 50% long-term debt).
- 3Calculate distance to default: DD = (V − D) / (σ_V × V), where D is the default barrier.
- 4Convert DD to PD using the normal distribution: PD = N(−DD) for the risk-neutral measure, or use empirical mapping (as Moody's KMV does) for real-world PD.
- 5For CDS/spread-based PD: obtain the market credit spread, assume LGD (typically 40–60%); calculate hazard rate λ = CS / LGD; compute PD(T) = 1 − e^(−λT).
- 6For rating-based PD: look up historical average cumulative default rate for the issuer's credit rating from S&P or Moody's annual default studies.
- 7Validate the PD estimate against market prices, rating agency ratings, and peer companies. Aggregate to portfolio expected loss: EL = Σ PD_i × LGD_i × EAD_i.
Company is 1.41 standard deviations above default — moderate credit risk
d2 = [ln(100/70) + (0.05 − 0.25²/2)×1] / (0.25×√1) = [0.357 + 0.019] / 0.25 = 1.505. PD = N(−1.505) ≈ 6.6%. Distance to default DD = (100 − 70) / (0.25 × 100) = 1.20 standard deviations. Moody's KMV maps this to an empirical EDF from their historical database. A 1.2 DD corresponds to roughly an 8–12% empirical 1-year default probability. If equity markets sell off and asset volatility rises to 35%, DD falls to 0.86 and PD jumps to 19% — demonstrating how equity market conditions directly impact credit risk.
CDS market implies 22% chance of default over 5 years
Hazard rate λ = CS / LGD = 300 bps / 60% = 500 bps = 5.0% per year. 1-year PD = 1 − e^(−0.05×1) = 4.88%. 5-year PD = 1 − e^(−0.05×5) = 1 − e^(−0.25) = 22.1%. This CDS-implied PD is a risk-neutral probability (includes risk premium), so real-world PD would be slightly lower. A 300 bps CDS spread is consistent with a BB credit rating, where historical 5-year cumulative default rates average approximately 10–15% — the CDS market implies a somewhat higher rate, reflecting market risk premium and liquidity premium.
Based on 40+ years of S&P historical default data by rating
Per S&P's 2023 Annual Global Corporate Default and Rating Transition Study, the average 3-year cumulative default rate for BB-rated issuers from 1981–2022 is approximately 8.2%. Annual default rates: BB Year 1≈0.84%, Year 2≈1.65% marginal, Year 3≈1.72% marginal. Investment-grade cumulative 3-year default rates are far lower: AAA≈0.01%, AA≈0.05%, A≈0.13%, BBB≈0.55%. These empirical rates are real-world PDs (not risk-neutral) and are used directly in Basel IRB expected loss calculations for loan portfolios.
Multiple financial ratios combined in logistic regression predict PD
In a logistic regression model, each financial ratio has a coefficient estimated from historical default data: Debt/EBITDA (5.0x — high leverage adds risk), interest coverage (1.8x — low coverage significantly increases PD), current ratio (0.9 — below 1.0 = negative working capital adds risk), and profit margin (4% — thin margin offers little buffer). The combined log-odds produce a PD of approximately 12.5% — elevated speculative-grade territory. This borrower would receive a B or B− internal rating, requiring pricing at approximately LIBOR/SOFR + 400–500 bps and tight covenant protections.
Bank loan origination credit approval and pricing (RAROC), enabling practitioners to make well-informed quantitative decisions based on validated computational methods and industry-standard approaches, which requires precise quantitative analysis to support evidence-based decisions, strategic resource allocation, and performance optimization across diverse organizational contexts and professional disciplines
Basel II/III IRB capital calculation for credit portfolios, helping analysts produce accurate results that support strategic planning, resource allocation, and performance benchmarking across organizations, where accurate numerical computation is essential for producing reliable outputs that inform planning, evaluation, and continuous improvement processes in both corporate and individual settings
IFRS 9 / CECL expected credit loss provisioning, allowing professionals to quantify outcomes systematically and compare scenarios using reliable mathematical frameworks and established formulas, demanding systematic calculation approaches that translate raw input data into actionable insights for stakeholders who depend on quantitative rigor in their daily professional activities
Credit default swap pricing and trading, supporting data-driven evaluation processes where numerical precision is essential for compliance, reporting, and optimization objectives, necessitating robust computational methods that deliver consistent and verifiable results suitable for reporting, auditing, and long-term trend analysis in professional environments
Bond portfolio credit risk monitoring and rating surveillance, which requires precise quantitative analysis to support evidence-based decisions, strategic resource allocation, and performance optimization across diverse organizational contexts and professional disciplines
In practice, this edge case requires careful consideration because standard assumptions may not hold. When encountering this scenario in credit default probability calculator calculations, practitioners should verify boundary conditions, check for division-by-zero risks, and consider whether the model's assumptions remain valid under these extreme conditions.
Professionals working with credit risk pd should be especially attentive to this scenario because it can lead to misleading results if not handled properly. Always verify boundary conditions and cross-check with independent methods when this case arises in practice.
In practice, this edge case requires careful consideration because standard assumptions may not hold. When encountering this scenario in credit default probability calculator calculations, practitioners should verify boundary conditions, check for division-by-zero risks, and consider whether the model's assumptions remain valid under these extreme conditions.
| Rating | 1-Year PD | 3-Year PD | 5-Year PD | Description |
|---|---|---|---|---|
| AAA | 0.00% | 0.01% | 0.04% | Highest investment grade |
| AA | 0.02% | 0.06% | 0.12% | Very high quality |
| A | 0.06% | 0.18% | 0.37% | High quality |
| BBB | 0.16% | 0.55% | 1.10% | Investment grade cutoff |
| BB | 0.84% | 3.41% | 7.18% | Highest speculative grade |
| B | 3.44% | 10.50% | 18.90% | Speculative |
| CCC/C | 26.78% | 42.60% | 52.20% | Highly speculative / distressed |
What is the difference between risk-neutral PD and real-world PD?
Risk-neutral PD is derived from market prices (CDS spreads, bond yields) and includes a risk premium on top of the pure default probability — investors demand extra compensation for holding default risk. Real-world (physical) PD is the actual historical frequency of default for a given risk category — typically estimated from rating agency data or internal bank loan performance data. Risk-neutral PD is used for derivatives pricing (CDSs, credit derivatives) and must match market prices. Real-world PD is used for Basel expected loss calculations, loan provisioning, and risk management. Risk-neutral PD is typically 1.5–3× higher than real-world PD for investment-grade issuers, reflecting the credit risk premium.
What are the Basel II/III approaches for PD estimation?
Under Basel II/III credit risk capital framework, banks can use: (1) Standardized Approach — risk weights prescribed by regulators based on external credit ratings, without internal PD estimates; (2) Foundation IRB (FIRB) — banks estimate PD internally but use supervisory LGD and EAD; (3) Advanced IRB (AIRB) — banks estimate PD, LGD, and EAD using their own internal models, subject to minimum data requirements and regulatory validation. AIRB allows the lowest capital charges for high-quality loan portfolios but requires years of default data, rigorous model validation, and ongoing model performance monitoring. Most large internationally active banks use AIRB for their major portfolios.
What financial ratios best predict corporate default?
Altman's Z-Score (1968) identified five key predictors of corporate bankruptcy from discriminant analysis: working capital/total assets, retained earnings/total assets, EBIT/total assets, market value of equity/book value of total liabilities, and sales/total assets. Research since then has confirmed that the most powerful individual predictors include: leverage (Debt/EBITDA, Debt/Assets), coverage (EBIT/Interest, EBITDA/Interest), liquidity (Current ratio, Quick ratio), profitability (ROA, profit margin), and size (larger firms have lower default rates for the same leverage level). Modern machine learning models use hundreds of financial and non-financial features.
How does recovery rate relate to PD?
Recovery rate (R) = 1 − LGD (Loss Given Default). Recovery rate is the fraction of exposure recovered after default through bankruptcy proceedings, asset sales, or restructuring. Historical recovery rates for unsecured corporate bonds average approximately 40% (LGD=60%), but vary enormously by seniority: senior secured bonds recover 60–80%; senior unsecured 35–55%; subordinated bonds 10–30%; equity near zero. Recovery rates are negatively correlated with PD — in systemic downturns when default rates spike, recovery rates simultaneously decline (procyclicality), amplifying credit losses beyond what independent PD and LGD estimates would suggest.
What is the 'through-the-cycle' vs. 'point-in-time' PD distinction?
Through-the-cycle (TTC) PD estimates the average default probability over a full economic cycle, smoothing out cyclical variation. Rating agencies use TTC ratings — a BBB rating should reflect similar default risk throughout the cycle. Point-in-time (PIT) PD reflects current economic conditions and moves with the business cycle: PIT PDs spike in recessions and compress in boom periods. Basel requires banks to use TTC PD for minimum capital calculations (to avoid procyclical capital amplification) but PIT PD for IFRS 9/CECL expected credit loss provisioning (which requires forward-looking, current-condition assessment). The PIT-to-TTC distinction is a source of significant practical complexity in credit risk management.
How is PD used in loan pricing?
Loan pricing uses the risk-adjusted return on capital (RAROC) framework: the credit spread charged must recover expected loss (EL = PD × LGD) plus generate an adequate return on regulatory capital. Minimum credit spread = (EL + Cost of Capital × Required Capital) / Loan Amount. For a 1-year term loan: if PD = 2%, LGD = 50%, EL = 1.0%. If required capital = 8% of loan, at 15% required return: capital cost = 8% × 15% = 1.2%. Minimum spread = 1.0% + 1.2% = 2.2% (220 bps). This RAROC framework ensures loans are only originated when the spread covers expected losses and provides adequate equity return.
Can machine learning improve PD prediction over traditional models?
Machine learning models — gradient boosting (XGBoost), random forests, neural networks — often achieve higher predictive accuracy (as measured by AUC/Gini coefficient) than logistic regression for PD prediction, particularly with large datasets containing many non-linear interactions between features. However, they face challenges in credit risk: regulatory explainability requirements (SR 11-7 guidance requires banks to understand and explain model outputs), model governance, and the risk of overfitting to historical patterns that may not persist. Hybrid approaches — using ML for feature engineering and variable selection, with logistic regression for the final model — are common in regulated banking environments.
เคล็ดลับโปร
Always validate your PD model with both Gini coefficient/AUC (discrimination — ranking borrowers from best to worst risk) and calibration tests (are predicted PDs close to observed default rates?). A model that ranks well but is poorly calibrated may still produce inadequate reserves.
คุณรู้ไหม?
The Merton model's key insight — treating a firm's equity as a call option on its assets — was published in 1974, just one year after the Black-Scholes option pricing formula. Robert Merton received the Nobel Prize in Economics in 1997 (along with Myron Scholes) largely for this body of work, which unified option pricing theory and credit risk in a single elegant framework.
เอกสารอ้างอิง