Gaussian Copula: Modeling Default Correlation

In previous articles, I discussed the reduced-form approach to default correlation, which explains correlation through hazard rates but does not fully capture the relationships between individual companies.

As I mentioned then, there is another family of models known as structural models, which explain default correlation through the value of a company’s assets and its financial leverage. If asset values themselves are affected by economic conditions, it is natural to expect defaults to become more frequent during periods of economic stress. (For more details, see: What Equity Prices Tell Us About Default Risk.)

In this article, however, I would like to focus on a different question:

How can we estimate the probability that several companies default at the same time?

The difficulty is that default is a binary event — a company either defaults or it does not.

While it is relatively easy to measure correlations between stock returns or asset values, measuring correlation between default events is much more challenging.

To address this problem, the Gaussian Copula model was developed.

This mathematically elegant model does not link defaults directly. Instead, it introduces a Hidden Credit Score and defines correlation between these hidden scores.

How does this work?

Suppose a company has a cumulative five-year default probability of 15%, based on its credit rating.

The Gaussian Copula transforms this probability into an equivalent position on the standard normal distribution:

15% cumulative default probability

↓

15th percentile (the company belongs to the bottom 15% of firms that are expected to default)

↓

$N^{-1}(15\%) = -1.04$

(On the familiar bell-shaped normal distribution, 15% of the area lies to the left of a value that is 1.04 standard deviations below the mean.)

Many readers become confused at this point. In reality, nothing fundamental has changed. The model is simply replacing one scale with another.

A 15% cumulative default probability and a threshold of -1.04 represent exactly the same percentile.

In simple terms:

If a company’s hidden credit score falls below -1.04, it is expected to default within five years.

What is the point of doing this?

Real-world default probabilities are not normally distributed. Many companies never default at all, and the distribution of corporate defaults certainly does not resemble a bell curve. As a result, it is difficult to model correlations directly.

The solution is a mathematical trick: transform default probabilities into variables that follow a normal distribution. That variable is the Hidden Credit Score.

What is the benefit of the Hidden Credit Score?

Once all companies are expressed on the same standardized scale, we can introduce a common economic factor that affects them all. In other words, the hidden credit score can be divided into two components: a factor that affects everyone and a factor specific to each individual company.

This is usually written as:

$x_i = aF + \sqrt{1-a^2}\,Z_i$

where:

• (x_i) = Hidden Credit Score of company (i)
• (F) = Common economic factor
• (Z_i) = Company-specific factor
• (a) = Sensitivity to the common economic factor

The intuition is straightforward.

Every company is influenced by two forces: the overall economy and events specific to that company.

The common economic factor (F) affects many companies simultaneously.

The company-specific factor (Z_i) captures events such as:

• Management mistakes
• Competitive pressures
• Product failures
• Operational disruptions
• Other company-specific developments

When economic conditions deteriorate, (F) becomes negative and pushes the hidden credit scores of many companies downward at the same time.

As a result, many firms move closer to their default thresholds simultaneously.

This is precisely where default correlation comes from. It emerges directly from the sensitivity of companies to the common economic factor (F) (which is assumed to be constant in the model).

When a crisis occurs and (F) becomes more negative, the entire distribution of hidden credit scores shifts to the left. Consequently, a larger number of companies end up below the default threshold.

Suppose we describe economic conditions using the following values.

If the economy moves to:

F = -0.8

we obtain the following picture:

Notice that the area to the left of the green curve becomes larger, meaning that more companies are expected to default.

The key point is that the default threshold itself does not change.

What changes is the distribution of hidden credit scores.

During periods of economic stress, that distribution shifts to the left, causing a greater proportion of companies to fall below the default threshold.

Adapted from:
Options, Futures, and Other Derivatives — John C. Hull