Current and prospective estimate of counterparty risk through dynamic neural networks

Abstract

The estimate of the probability of default plays a central role for any financial entity that wants to have an overview of the risks of insolvency it may incur by having economic relations with counterparties. This study aims to analyze the calculation of such measure in the context of counterparty risk from a current and prospective standpoint, by using dynamic neural networks. The forecasting aspect in the calculation of such risk measure is becoming more and more important over time as current regulation is increasingly based on a "Through the Cycle" and not a "Point in Time" assessment, consequently giving fundamental importance to such estimate. To this end, three different models aimed at calculating the Probability of Default have been investigated: the CDS method, the Z-Spread method, and the KMV method (Kealhofer, Merton and Vasicek). First, the different techniques have been applied to one of the main suppliers of gas and energy in Italy as a reference company. Then, they have been applied to calculate the same risk measure on the 50 companies included in one of the most important European indices, the Euro Stoxx 50.

Keywords

• default probability
• counterparty risk
• credit default swap
• corporate bond
• kmv model
• nonlinear auto-regressive (nar) network
• forecasting