Credit Risk: Probability of Default & Loss Given Default

Credit risk is one of the most fundamental risks that bond investors face. When you buy a bond, you’re lending money — and there’s always a chance the borrower won’t pay you back. Understanding how to measure that risk is essential for pricing bonds, managing portfolios, and making informed investment decisions. The three pillars of credit risk measurement are probability of default (PD), loss given default (LGD), and expected loss (EL). This guide covers all three — what they mean, how they’re estimated, and how they work together to quantify the credit risk embedded in bond prices.

What is Credit Risk?

In credit analysis, a default is formally defined as any of the following events: a missed or delayed payment of interest or principal, a bankruptcy filing, or a distressed exchange where creditors accept less favorable terms to avoid a formal bankruptcy. The specific definition matters because probability of default estimates depend on how “default” is defined.

Credit risk takes several forms: default risk (failure to make payments), downgrade risk (a rating cut that reduces bond value), and credit spread risk (the market demanding wider credit spreads, reducing prices even without a default). All three directly affect bond prices — investors demand higher yields to compensate for bearing credit risk.

Probability of Default (PD)

Probability of default (PD) is the likelihood that a borrower will fail to meet its required debt payments within a specified time horizon. It is typically expressed as a percentage and measured over a one-year horizon, though multi-year cumulative PDs are also widely used.

PD estimates come in two flavors: point-in-time (PIT) estimates reflect current conditions and fluctuate with the business cycle, while through-the-cycle (TTC) estimates average default risk across economic cycles and are commonly used by rating agencies.

When working with multi-year horizons, it’s important to distinguish marginal PD (the probability of defaulting in a specific single period, conditional on surviving to that point) from cumulative PD (the total probability of defaulting at any point through a given horizon). Cumulative PD always increases with time — even highly rated issuers have a higher probability of defaulting over 10 years than over 1 year.

Key drivers of PD include financial leverage (debt-to-EBITDA), cash flow stability, industry conditions (cyclical industries face higher default rates in downturns), and the broader macroeconomic environment (recessions and rising rates raise defaults across the board).

Loss Given Default (LGD)

Loss given default (LGD) measures the percentage of exposure that is lost when a default actually occurs. It is the complement of the recovery rate — the fraction of the debt that creditors ultimately recover through bankruptcy proceedings, asset liquidation, or restructuring.

For example, if creditors recover 40 cents on the dollar after a default, the recovery rate is 40% and the LGD is 60% — meaning 60% of the exposure is lost.

Recovery Rates by Seniority

The single biggest factor affecting recovery rates is the seniority of the debt in the issuer’s capital structure. Senior secured creditors are paid first from the proceeds of asset sales, while subordinated creditors receive whatever remains.

Note: These are illustrative ranges based on long-run averages. Actual recovery rates vary significantly by industry, economic conditions at the time of default, and the specific terms of the debt instrument.

The Expected Loss Formula

Expected loss (EL) combines all three components of credit risk into a single measure — the average loss you should anticipate from credit events over a given time horizon.

Where:

• PD — probability of default (as a decimal)
• LGD — loss given default (as a decimal)
• EAD — exposure at default, the total value at risk. For bonds, this is typically the outstanding principal plus any accrued interest.

It’s important to understand that expected loss represents the mean of the loss distribution — the long-run average. In any given year, the actual loss could be zero (no default occurs) or could be the full exposure (default with zero recovery). Banks and institutional investors also calculate unexpected loss — the potential for losses to exceed the expected level — to set capital reserves for tail risk scenarios.

Credit Risk Example

Now compare this to a B-rated bond from a more speculative issuer like Carvana, with the same face value and recovery assumptions:

Credit Ratings and Probability of Default

Credit rating agencies like S&P Global, Moody’s, and Fitch assign letter grades that reflect an issuer’s creditworthiness. Each rating corresponds to an approximate range of historical default probabilities. The table below maps S&P ratings to long-run average 1-year default rates:

Source: S&P Global Ratings, Annual Global Corporate Default and Rating Transition Study. Values represent long-run weighted averages (1981–2023, published 2024) for global corporate issuers. Actual annual default rates vary significantly by year and economic conditions.

The dividing line between investment grade (BBB− and above) and high yield (BB+ and below) is one of the most important thresholds in fixed income markets. Many institutional investors — pension funds, insurance companies, and bank portfolios — are restricted to holding only investment-grade bonds, which creates a sharp difference in demand, liquidity, and pricing across this boundary.

PD vs Credit Spread

Probability of default and credit spreads both relate to credit risk, but they measure fundamentally different things. Understanding the distinction is critical for proper credit analysis.

A crucial distinction is between historical (real-world) PD and risk-neutral (market-implied) PD. Credit spreads embed a risk-neutral PD that is typically higher than the statistical PD derived from historical default data. This is because spreads compensate investors not only for expected defaults but also for liquidity risk, loss severity uncertainty, and general investor risk aversion. For a deeper analysis of how credit spreads are measured and compared, see our guide on Z-Spread vs G-Spread.

How to Estimate Credit Risk

Professional credit analysts use multiple approaches to estimate an issuer’s creditworthiness. No single method is sufficient on its own — each provides a different lens on the same underlying risk.

1. Credit Rating Approach

The most straightforward method is to use agency credit ratings (S&P, Moody’s, Fitch) as proxies for PD. Rating agencies publish extensive historical default studies that map each rating to observed default frequencies. This approach is simple and widely accepted, but ratings can lag behind rapidly changing fundamentals.

2. Market-Based Approach

You can derive an implied PD from credit spreads using a simplified relationship:

For example, if a bond’s credit spread is 150 basis points (1.50%) and the assumed LGD is 60%, the implied PD ≈ 1.50% / 0.60 = 2.5%.

3. Fundamental Analysis Approach

Analyzing an issuer’s financial statements provides direct insight into default risk. Key ratios include Debt / EBITDA (leverage), interest coverage (EBIT / Interest Expense), current ratio (short-term liquidity), and free cash flow / total debt (repayment capacity). Deterioration in these metrics often precedes rating downgrades.

To learn more about credit risk and other fixed income concepts in depth, explore our Fixed Income Investing course.

Common Mistakes

Credit risk analysis involves several concepts that are easy to confuse. Avoid these common errors:

1. Confusing PD with credit spread. PD is a statistical probability of default; credit spreads are market prices that include compensation for default risk plus liquidity premiums, risk aversion, and other factors. A wider spread does not necessarily mean a proportionally higher PD.

2. Ignoring recovery rates. Two bonds from different issuers can have identical PDs but very different expected losses if one is senior secured (high recovery) and the other is subordinated (low recovery). Always consider LGD alongside PD.

3. Treating PD as static. Default probabilities are not fixed — they evolve as a company’s financial health, industry conditions, and the macroeconomic environment change. A BBB-rated issuer today can deteriorate to B-rated within a few years.

4. Using a 1-year PD for multi-year exposures. A one-year PD of 0.15% does not mean the five-year default probability is 0.75% (simply multiplied). Cumulative PD must account for conditional survival probabilities and the compounding nature of multi-period default risk.

5. Mixing horizons, units, or levels of analysis. Comparing a 1-year PD (a percentage) directly to a credit spread (in basis points) without conversion is a common error. Similarly, applying an issuer-level PD to instrument-level loss without adjusting for seniority produces misleading results — a senior secured bond from a defaulted issuer will recover far more than a subordinated bond from the same issuer.

Limitations of Probability of Default

1. PDs are estimates, not certainties. Historical PDs represent long-run averages, not guarantees. Even AAA-rated issuers can default, and PDs derived from historical data often differ significantly from market-implied PDs — the gap reflects risk premiums, liquidity, and supply/demand dynamics.

2. Model risk. Different credit risk models — structural (Merton), reduced-form, and statistical scoring — can produce materially different PD estimates for the same issuer. The choice of model and its assumptions matters.

3. PD doesn’t capture tail risk. Expected loss represents the average outcome, but the actual loss distribution has fat tails — extreme losses occur more often than normal distributions predict. This is why banks hold capital reserves well above expected loss levels.

4. Ratings lag. Agency credit ratings — the most common source of PD proxies — are slow to update. A company’s financial condition can deteriorate significantly before a rating downgrade occurs, making rating-based PD estimates stale during periods of rapid change.