Credit VaR for Fixed Income Portfolios: Spread VaR, Jump-to-Default Risk, and Credit Portfolio Models

Defining Credit VaR: Two Sources of Credit Loss

Credit risk in a bond portfolio generates losses through two distinct channels, and rigorous credit VaR must address both:

• Spread VaR: loss from OAS (Option-Adjusted Spread) widening without default. The bond's price falls because the market demands a higher risk premium — even if the issuer ultimately pays every coupon and principal. Spread VaR is continuous, can be observed daily, and is captured by spread DV01 × spread volatility models.
• Jump-to-Default (JTD): the binary, catastrophic loss if an issuer defaults. This is not well captured by spread vol models because spreads often move only modestly before a sudden default event. The 2001 Enron and WorldCom defaults, 2008 Lehman Brothers default, and 2023 Silicon Valley Bank convertible bond losses all involved sudden, large JTD exposures.

The distinction matters for risk management: spread VaR is managed through spread duration limits and OAS hedging (CDS). JTD is managed through single-name concentration limits, sector diversification, and credit quality thresholds. The Compliance and Risk templates include both components in a unified credit risk framework.

Spread VaR: Calculation Framework

Spread DV01 (also called "spread sensitivity" or OASD × notional × 0.0001) is the dollar change in portfolio value for a 1bp widening in OAS across all credits in the portfolio. Spread VaR then applies a statistical model of OAS volatility:

• Spread VaR = SpreadDV01 × σ_OAS (daily, in bps) × Z_α × √(holding period)
• Where σ_OAS is the daily standard deviation of the relevant OAS index (IG, HY, or sector-specific)

• "Spread VaR for an IG corporate bond portfolio: $50M portfolio in investment-grade corporates. Average OAS = 118bps (Bloomberg US Corporate IG index). Average OAS duration (OASD) = 4.8 years. Spread DV01 = 4.8 × $50M × 0.0001 = $24,000/bp. Historical OAS volatility (Bloomberg US Corporate IG OAS daily changes, 252-day lookback): σ_OAS = 3.8bps/day. Calculate: (1) 1-day 99% spread VaR = SpreadDV01 × σ × Z_99 = $24,000 × 3.8 × 2.326, (2) 10-day 99% spread VaR (scale by √10), (3) 1-year spread VaR (scale by √252 from daily), (4) what OAS level would this portfolio reach at the 99th percentile 1-year OAS shock? (current OAS + 1Y 99% OAS change). Compare to 2020 COVID peak IG OAS of ~370bps (spread from 118bps widened approximately +252bps at peak). Was the COVID event a 99th percentile OAS move or worse?"
• "Multi-sector spread VaR with correlation: Portfolio breakdown: (A) Financials IG $20M, OASD 3.9Y, OAS vol 4.8bps/day. (B) Industrials IG $18M, OASD 5.2Y, OAS vol 3.5bps/day. (C) Utilities IG $12M, OASD 6.8Y, OAS vol 3.1bps/day. OAS correlation matrix: Financials-Industrials ρ = 0.78, Financials-Utilities ρ = 0.65, Industrials-Utilities ρ = 0.72. For each sector: SpreadDV01_i = OASD_i × Notional_i × 0.0001. Compute: (1) sector spread VaR in isolation (1-day 99%), (2) portfolio spread VaR using the correlation matrix: VaR_portfolio = Z_99 × √(w^T Σ w) where w_i = SpreadDV01_i × σ_i and Σ is the correlation matrix, (3) diversification benefit = sum(standalone VaRs) − portfolio VaR, (4) what happens to the diversification benefit if correlations rise to 0.95 in a stress event (flight to quality)?"

Jump-to-Default Risk: Single Name Exposure

JTD risk for a single bond position is straightforward in concept but complex in aggregation across a portfolio. The key inputs are: (1) default probability by rating from a long-run transition matrix, (2) loss given default (LGD) by seniority and industry, (3) correlation between default events (copula model).

• "Jump-to-default analysis — single BBB-rated issuer: $15M position in a BBB-rated senior unsecured bond. Parameters: (1) Moody's 1-year historical average default rate for Baa-rated issuers: 0.20%; (2) LGD for senior unsecured bonds: 60% (recovery rate 40%). Calculations: (A) Expected 1-year loss = $15M × 0.60 × 0.0020 = $18,000 (the expected credit cost); (B) Stressed JTD (full default today): gross loss = $15M × 0.60 = $9M; (C) 1-year JTD VaR at 99.9% confidence: for a single Bernoulli default event, the 99.9th percentile outcome is 0 defaults if PD = 0.20% (since 0.1% > 0.20%, the 99.9th percentile loss is zero — the single-name JTD only matters in portfolio context through tail probability modeling). Explain why JTD is primarily a portfolio aggregation problem, not a single-position problem. (D) Now assume the bond's CDS spread widens from 120bps to 450bps one week before default — compute the spread loss from the pre-default spread widening: $15M × 0.60 × (0.045 − 0.012) / (0.45 × (1+0.045)^5) ≈ estimate the mark-to-market loss on the bond from the CDS-implied spread move."
• "Portfolio JTD — concentration risk analysis: I have a bond portfolio with the following 5 largest single-name exposures: (1) $25M BBB tech issuer, LGD 55%; (2) $20M BB energy issuer, LGD 65%; (3) $18M BBB financial, LGD 50%; (4) $15M B-rated retail, LGD 70%; (5) $12M BBB utility, LGD 45%. Default probabilities (1-year, from S&P transition matrix): BBB 0.23%, BB 0.80%, B 4.15%. Calculate for each position: (A) Expected loss, (B) JTD gross loss. For the portfolio: (C) total expected loss, (D) total gross JTD if all 5 default simultaneously (worst case, zero correlation benefit). For the Basel Large Exposures Rule (CRR Article 395): maximum single-name exposure = 25% of Tier 1 capital. If Tier 1 = $180M, what is the maximum permitted single-name exposure? Do any positions breach the limit? (E) Compute a simple portfolio 1-year credit VaR using the independent default assumption (Poisson approximation): probability of at least one default across the 5 names = 1 − Π(1 − PD_i). What is this probability? What is the expected loss conditional on at least one default occurring?"

CreditMetrics Migration Model

CreditMetrics models credit losses from both defaults and rating migrations — a BBB bond that migrates to BB loses value because the market now demands a higher yield (BB OAS is wider than BBB OAS). The model uses an annual rating transition matrix (probability of each rating moving to each other rating over 1 year) and reprices each bond at the new rating's OAS level.

• "CreditMetrics 1-year migration analysis for a single BBB bond: $20M face, 5-year maturity, 4.80% coupon, current OAS 120bps (priced at par since coupon matches yield). S&P 1-year transition matrix for BBB: to AAA 0.02%, to AA 0.23%, to A 4.56%, to BBB 88.48%, to BB 4.44%, to B 0.84%, to CCC 0.19%, to Default 0.24%. OAS levels by rating at 5-year maturity (current market levels): AAA 30bps, AA 45bps, A 70bps, BBB 120bps, BB 220bps, B 430bps, CCC 900bps. Recovery in default: 40% of face ($8M recovery, $12M loss). For each end-state rating: (1) compute the new bond price at the new OAS level (approximate: ΔPrice ≈ −OASD × Face × ΔOAS, where ΔOAS = new_OAS − 120bps, OASD ≈ 4.5 years for a 5Y bond priced near par), (2) probability-weight the price change by the migration probability, (3) construct the 1-year P&L distribution for this position, (4) compute expected loss, (5) compute 99th percentile loss (at 99th percentile, the bond has migrated to B or worse — what is the cumulative probability of migrating to B or worse?)."
• "Portfolio CreditMetrics — Monte Carlo with default correlation: Portfolio: 50 IG corporate bonds, $4M average position size, ratings mix: 30% A, 50% BBB, 20% BB. Default correlations (Gaussian copula, asset correlation): same-sector pairs ρ_asset = 0.35, cross-sector pairs ρ_asset = 0.18. Generate 50,000 Monte Carlo trials: (A) for each trial, draw a correlated normal vector Z = (z_1, ..., z_50) using the Cholesky decomposition of the 50×50 correlation matrix, (B) for each name i, default occurs if z_i < Φ^(-1)(PD_i) where Φ^(-1) is the inverse normal CDF and PD_i is the 1-year default probability, (C) rating migration occurs if z_i falls in the migration threshold range from the transition matrix, (D) for each trial, compute total portfolio loss from defaults (LGD × notional) and migration mark-to-market, (E) from the 50,000 trial loss distribution: expected loss, 99th percentile credit VaR, 99.9th percentile credit VaR, and Expected Shortfall at 99%. How does increasing default correlation from 0.18 to 0.45 affect the 99.9th percentile credit VaR?"

Stressed Credit VaR: Historical Crisis Scenarios

The two critical historical stress scenarios for IG credit portfolios are the 2008 Global Financial Crisis and the 2020 COVID shock. Both produced OAS widening events that far exceeded any parametric spread VaR prediction.

• "2008 credit crisis stress scenario: Bloomberg US Corporate IG OAS rose from approximately 150bps in June 2008 to a peak of 650bps in November 2008 — a widening of 500bps in 5 months. IG OAS daily volatility during the crisis period: approximately 18bps/day (vs. 3-4bps/day in normal markets). Apply to a $200M IG portfolio (OASD 4.8Y, SpreadDV01 $96,000/bp): (1) P&L loss from the full +500bp widening: SpreadDV01 × 500bps, (2) loss as percentage of AUM, (3) peak daily loss (use the worst single day in Oct 2008 — approximately +38bps in 1 day): SpreadDV01 × 38, (4) how does the 2008 peak OAS move compare to a 3-sigma OAS event under normal-period volatility (3 × 3.8bps × √252 daily days)? This shows why stressed VaR is required: the 2008 event was a 30+ sigma event under normal-period vol assumptions, (5) had you held a 5% notional CDS hedge (bought protection on $10M of the portfolio at 150bps CDS spread): CDS hedge P&L at peak = notional × (ΔOAS / 10000) × duration / (1+CDS_spread). How much did the hedge offset?"
• "COVID 2020 — 'dash for cash' credit stress: From February 20 to March 23, 2020 (22 trading days), IG OAS widened from approximately 100bps to 373bps (+273bps). HY OAS went from 340bps to 1087bps (+747bps). Even AAA-rated structured products saw spread widening. For a mixed credit portfolio: IG tranche $180M (OASD 4.5, SpreadDV01 $81,000/bp), HY tranche $40M (OASD 3.8, SpreadDV01 $15,200/bp): (1) IG spread loss: $81,000 × 273bps, (2) HY spread loss: $15,200 × 747bps, (3) total loss and as % of $220M AUM, (4) daily OAS moves during this period averaged approximately 12.4bps/day for IG — compare to the normal-period VaR (1-day 99% = $81,000 × 3.8 × 2.326). How many 'normal VaR' days does the COVID event equal?, (5) the Fed's Primary Market Corporate Credit Facility (PMCCF) and Secondary Market Corporate Credit Facility (SMCCF) announcements on March 23 reversed IG OAS by ~80bps in 1 day. Compute the P&L on day of the Fed announcement."

Credit Concentration Risk and Regulatory Limits

• "Basel Large Exposures Rule — compliance check: Under CRR Article 395, a bank's exposure to a single counterparty or connected group cannot exceed 25% of Tier 1 capital. For a bank with Tier 1 capital of €240M and a corporate bond portfolio with 40 names: 3 issuers have exposure exceeding €55M: Issuer A €62M (tech sector), Issuer B €58M (energy), Issuer C €51M (financials). (1) Compute the breach for Issuers A and B: excess = exposure − (25% × €240M), (2) what is the minimum reduction in each position required to comply?, (3) assume Issuer A and B are in connected groups (Issuer B is a subsidiary of a holding company also related to Issuer A): what is the group exposure? Apply the connected counterparty definition from CRR Article 4(1)(39), (4) how does the EBA GL/2021/15 on large exposures affect exposure calculation for collateralized bond positions? (Exposure = gross exposure − eligible collateral under the Financial Collateral Comprehensive Method)"

Where to Start

Credit VaR analysis for bond portfolios spans spread risk (daily observable) and default risk (tail event modeling). The Compliance and Risk templates include a Credit Risk Module covering spread VaR calculation, CreditMetrics migration model setup, and Basel Large Exposures compliance checking. For the rate VaR component that combines with credit VaR to give total VaR, see Fixed Income VaR. For yield curve level analysis that underlies rate sensitivity of credit portfolios, see Yield Curve Analysis with AI. The Quant Finance category includes CDS pricing tools that can be used to extract market-implied default probabilities for credit VaR calibration.

Frequently Asked Questions

What is the difference between spread VaR and credit VaR?

Spread VaR is the narrower concept: the loss from OAS widening in normal trading conditions, captured by a statistical model of daily OAS changes. It is analogous to rate VaR — a daily-frequency risk measure based on observed price changes. Credit VaR is the broader concept encompassing all credit loss outcomes: spread VaR, rating migration losses, and jump-to-default losses at a given confidence level over a defined horizon (typically 1 year for capital purposes). Credit VaR is measured above expected loss — the capital needed to absorb unexpected credit losses beyond what has already been priced into the portfolio's credit spreads. A portfolio earning 120bps OAS on $200M ($2.4M/year) already "earns" the expected loss through carry; credit VaR measures the additional capital needed to survive a 99th or 99.9th percentile loss event.

How is default correlation modeled in a bond portfolio?

Default correlation is the key driver of portfolio credit VaR tails. Low correlation = losses are diversified and portfolio VaR is low. High correlation = losses concentrate and portfolio VaR spikes. The industry standard is a Gaussian copula (or t-copula for fatter tails): each issuer's default is governed by a latent variable (asset value) drawn from a correlated multivariate normal distribution. Default occurs when the asset value falls below a threshold calibrated to the issuer's PD. The correlation between assets (ρ_asset) is the key parameter — typically proxied by equity return correlations between sectors, estimated at 0.15-0.35 for cross-sector IG pairs and 0.30-0.50 for same-sector pairs. The t-copula captures the empirical observation that defaults are more correlated in crisis periods than implied by the Gaussian copula — it was widely adopted after the 2008 crisis exposed the Gaussian copula's inability to capture extreme joint default events.

What is the FRTB treatment of credit spread risk?

Under the FRTB Sensitivities-Based Method (SBM), credit spread risk for non-securitization positions is separated into CSRN (Credit Spread Risk Non-Securitisation) buckets: IG buckets by sector and HY buckets by sector (8 IG sectors × 2 HY buckets = 10 buckets, plus a miscellaneous bucket). Delta sensitivities to OAS are computed at each tenor point (0.5Y, 1Y, 3Y, 5Y, 10Y) for each issuer. Risk weights are prescribed by rating and tenor: for BBB-rated financial corporates at the 5Y point, the SBM risk weight is 2.0% of sensitivity. Intra-bucket correlations within IG sectors are 0.65 for same-name, 0.35 for different names. Inter-bucket IG/HY correlations are 0.20. CSRN capital = √(Σ_b K_b² + Σ_{b,c} 2ρ_{bc} K_b K_c). Importantly, FRTB SBM disallows correlation between CSRN and GIRR (General Interest Rate Risk) — so the negative rate/spread correlation that creates diversification in economic models is not recognized in regulatory capital under the standardized approach.

How do high-yield bonds change the credit VaR calculation?

High-yield bonds have substantially higher OAS volatility (daily σ_OAS typically 15-25bps vs. 3-5bps for IG), higher default probability (B-rated 1Y PD approximately 4%, CCC approximately 15% vs. BBB 0.23%), and higher default correlation (HY issuers tend to be smaller and more macro-cyclical). The spread DV01 is typically lower for HY (shorter effective duration due to higher yields and call features), but the spread volatility more than compensates. A common rule of thumb: 1-day 99% spread VaR as a percentage of portfolio value is approximately 0.5-0.8% for IG and 1.5-3.0% for HY. For a mixed IG/HY portfolio, the total credit VaR depends critically on the IG-HY correlation — historically around 0.65 in normal markets, rising to 0.85-0.90 in crisis periods. The JTD risk in HY is meaningful even on a 1-year horizon and must be modeled explicitly, while in IG portfolios it is often treated as a tail Pillar 2 consideration rather than a Pillar 1 VaR driver.