Identifying Default Correlation via a Mix of Correlated Bernoulli Distributions

Default correlation presence imply higher credit losses. Its proper identification from empirical data is challenging. It turns out that the real-world data is a mixture of distributions, rather than a homogenous one. We show how to model it using a mixture of correlated Bernoulli ones. We show that the United States total loan loss statistics for the last 30 years is best proxied by a mix of 87% for a risk-free distribution and a 13% for the one with the default probability of 22 % and default correlation of +26%.