EBA欧洲银行-C1-Garcia-Cespedes2C20Moreno-Estimating-the-distribution-of-total-defaultlosses_35页_806kb

Summary of "Estimating the distribution of total default losses on the Spanish Financial system"

Core Content

This document presents a study on estimating the distribution of total default losses in the Spanish financial system, using the Vasicek (1987) credit risk model and importance sampling methods. The research is motivated by the need for accurate risk measurement and allocation tools, especially in light of the financial crisis and the significant impact of defaults on the financial system.

Main Points

1. Credit Risk Overview

• Credit risk is the risk that a borrower will default on debt, and it is the most significant risk for financial institutions.
• During the financial crisis, credit losses were the main source of P&L losses.
• The Vasicek model is widely used for credit risk measurement, assuming constant recoveries. However, under general conditions, it is time-consuming to estimate and allocate risk.
• The model assumes that a borrower's assets value $ V_j $ is influenced by macroeconomic factors and idiosyncratic risk.

2. Spanish Portfolio Analysis

• The study analyzes the portfolio of Spanish financial institutions covered by the deposit guarantee fund at the end of 2010.
• Key parameters such as Probability of Default (PD), Loss Given Default (LGD), and Exposure at Default (EAD) are estimated.
• PD is based on external ratings from S&P, Moody's, and Fitch. Institutions without ratings are assigned a notch lower than the average.
• LGD is estimated using historical data and updated to reflect losses on total assets rather than just deposits.
• The study includes a list of merged or affiliated institutions and their respective PD, LGD, and EAD.

3. Importance Sampling Methods

• Importance sampling is a technique used to improve the efficiency of Monte Carlo simulations in estimating rare events.
• The method involves modifying the simulation distribution and adjusting results with weights.
• Two main changes are proposed:
  • Adjusting the default probability based on macroeconomic scenarios.
  • Changing the distribution of macroeconomic factors to focus on high-loss scenarios.
• The weight for each simulation is calculated based on the ratio of the original and modified probabilities.

4. Portfolio Results

• The study compares the results of different models (constant vs. random LGD, default vs. market mode) and risk allocation methods (VaR vs. ES).
• The 99.9% probability loss under the ASRF model is 13,733 MM €, but under importance sampling, it increases to 32,102 MM €.
• The confidence intervals with 10,000 simulations are very narrow, indicating high precision.
• VaR-based risk allocation has a sharp profile, while ES-based allocation is smoother.
• With random LGD, the 99.9% probability loss increases further to 37,934 MM €, and confidence intervals widen.

5. Model Extensions

• The study extends the importance sampling framework to include random LGD and market mode valuation.
• A macroeconomic factor $ z_{LGD} $ is introduced to affect LGD behavior.
• The LGD can be modeled as constant or random, with parameters inferred from historical data.
• The migration matrix and accumulated migration probabilities are used to estimate the probability of default and rating transitions.

6. Further Research

• The study suggests future research directions, including the use of point-in-time PDs, structural approaches for LGD, and improving the measurement of interrelations and contagion effects.
• References are provided for the methodologies and data sources used in the study.

Key Information

• Vasicek Model: A flexible model for credit risk, with the assumption of constant recoveries. It is extended to handle random recoveries and market mode valuation.
• Importance Sampling: Used to estimate the loss distribution and risk allocation more efficiently by focusing on high-loss scenarios.
• Spanish Financial Institutions: The study includes a detailed list of institutions, their PD, LGD, EAD, and factor sensitivities.
• Risk Allocation: Based on VaR and ES, with different profiles depending on the model assumptions.
• Loss Distribution: The loss distribution varies significantly with different models and assumptions, with the highest losses observed under the random LGD and market mode scenarios.
• Data Sources: PD is based on external ratings, LGD is updated using FDIC data, and EAD is derived from balance sheet information.

Conclusion

• The importance sampling framework has been successfully extended to handle random recoveries and market mode valuation.
• The study highlights the importance of considering different models and assumptions in estimating credit risk and loss distribution.
• These methods are valuable for regulators to identify and measure the impact of systemically important financial institutions and to perform stress tests on the financial system.