Portfolio-Size Constraints on Default Correlation: Exact Coherence Tests and Sharp Asymptotic Bounds
Palaisti
Because defaults are absorbing, default correlations at different horizons must satisfy joint restrictions; in the one-factor Gaussian model they define an exact interval of admissible later correlations.
Source published · In the Daily Brief of 18 November 2026
Filed under Credit risk, Probability
A default, once it happens, cannot be undone, so the distributions of cumulative defaults at successive horizons constrain one another. The paper characterises these constraints for finite homogeneous portfolios. For binomial mixtures whose mixing distribution functions cross at most once, stochastic ordering of the whole default count reduces to two inequalities: the probability of no defaults must not increase and the probability of all defaults must not decrease.
In the one-factor Gaussian model this gives an exact interval of admissible later correlations from two scalar equations, and the intervals are nested as the portfolio size \(n\) grows. With earlier and later default probabilities \(p_0 < p_1\) and earlier correlation \(\rho_0\), both ends approach \(\rho_0\) at rate \((\log n)^{-1/2}\), with leading magnitude \(\rho_0\sqrt{2(1-\rho_0)}\,[\Phi^{-1}(p_1)-\Phi^{-1}(p_0)]\). The paper also gives a finite-size certificate that a prescribed set of horizon laws is incoherent.