Author: Anja Helk Value at risk (VAR), as an established measurement of market risk, is now revealing its virtues to credit risk management. VAR indicates the maximum likely loss over a specifIc period within a given confidence level, and is increasingly used as a tool for credit risk monitoring, to perform risk/return analysis of credit portfolios and for capital allocation decisions. Importantly, VAR measures economic capital, which is thought to replace the less flexible regulatory capital requirements that are currently in place.
Euromoney takes a look at three publicly available models and the way they produce a measure of credit risk: PortfolioManager (published by KMV, 1993), CreditMetrics (published by RiskMetrics, 1997), and CreditRisk+ (published by CSFB, 1997).
CreditMetrics assesses individual and portfolio credit VAR on a mark-to-market basis. The credit VAR stems from changes in credit quality caused by upgrades, downgrades, defaults, and portfolio concentration.
Methodology: CreditMetrics establishes an exposure profile of each instrument. It then computes the volatility in value for each instrument by rating migration and default risk. The likelihood of default and rating changes is specified in a transition matrix.
The individual volatilities and the correlation between default events, which are derived from industry and country allocations, give an aggregate portfolio volatility. The loss probability distribution is generated by a Monte Carlo simulation.
Users: CreditMetrics is aimed at investment portfolios that are actively traded. The model can evaluate the impact of specific exposures on their portfolio, measure risk adjusted capital requirements for a specific investment strategy and can be used to make asset allocation decisions. “Our model appeals to fund managers because it allows for issuer-specific decisions, and it will be increasingly used for optimization decisions,” says Ethan Berman, the CEO of RiskMetrics.
Components: CreditMetrics Methodology, historical data sets, software package (CreditManager).
Advanced features: Risk due to recovery rate uncertainty can be accounted for. Stress testing of inputs possible. Customizable reports on different types of risk.
Input data Loss exposure of individual assets, default rates of a rating class, recovery rates of a rating class (optional), transition matrix, industry/ sector correlations, spread curves by sector and rating.
Strengths: “Flexibility and openness is an important distinction of our model,” says Berman. “Our clients can use their own data sets, or use the data available from CreditMetrics. The CreditMetrics package offers a variety of data available over the web – default rates, transition data, recovery rates and pricing information, as well as underlying data needed for measuring correlations between industries and between countries.”
Weaknesses: Credit spread curves for each rating class are difficult to validate in the European market. Spread changes for bonds that do not change ratings, and changes in the base of interest rates are not incorporated. Company debt has to be rated.
KMV’s PortfolioManager measures a portfolio’s credit risk and the contributions of individual assets to portfolio risk. What makes KMVs approach distinct in that it is based on an option theoretic valuation of debt, after Robert Merton. Debt can be valued as if it were an option on the value of the underlying asset of the firm. The probability of default is the same as the probability of the option being exercised.
Methodology: PortfolioManager uses the EDF credit measure (expected default frequency) for individual default risk. The EDF is calculated by KMV’s CreditMonitor, which uses the equity market capitalization to deduce the value and volatility of the entire value of the business – the asset value. When the market asset value falls below the face value of obligations, the firm reaches the default point – the firm will default. The EDF is the ex-ante probability of this occurring. The credit risk of the whole portfolio is quantified by the relationship between the expected default risk of each pair of assets in a portfolio. The correlation of defaults is inferred from correlation of the asset values and the respective EDFs. A loss probability distribution can be obtained both analytically and from a Monte Carlo simulation.
Users: “When we started in 1989, our focus was on the banking world, but demand for our model has picked up strongly over the last two years among the non-bank sector. Fund managers now increasingly appreciate our model for diversification, individual buy/sell decisions, and to minimize concentrations,” says Stephen Kealhofer, KMV’s inventor.
Data inputs For CreditMonitor: no required inputs from user. For PortfolioManager: Exposure characteristics. Risk information is supplied with the model, but users can put in own data.
Components: CreditMonitor and Private Firm Model for individual credit risk, PortfolioManager and Gcorr for portfolio level.
Advanced features: Private firm models provides EDF measures for non-listed firms. A database for default probabilities and history for specific names covers 10,000 US companies, 6,000 public companies in Europe, and 24,000 globally.
Strengths: “A credit risk measure from the equity world has a continuous range, so that we have a timely and finely graded picture of individual firm risk,” says Kealhofer. “And default correlation can be measured by using two firms’ asset correlation, which are in turn derived from equity volatility, meaning there is good quality and ample data. The model is also open to the users’ own data. It is even possible to put changes in the capital structure of a firm into the model.”
Weaknesses: The correlation between asset and equity value is not yet accepted industry wide. Users must supply expected recovery rates. Much of the empirical work is based on US experience.
CSFB’s CreditRisk+ is a framework for calculating the portfolio loss distribution, which gives information on the average loss, the worst case loss of the portfolio at the 95th or 99th percentile and the economic capital. CreditRisk+ is an event model, i.e. it does not account for changes in credit quality other than default. A mark-to-market approach to default is unobtainable within this framework.
Methodology: This can be described as a reduced form approach, because it treats default as a random event. The structural approach, as employed by CreditMetrics and PortfolioManager, models default risk as a function of the behaviour of the issuer. The default rate and default rate volatility – the standard deviation – are the inputs needed to generate the full distribution of losses.
Unusually, CreditRisk+ approximates this function analytically by using a mathematical technique that originated in the insurance industry. Information on default correlations is taken from sector analysis, and is contained in the default rate volatilities.
Users: CreditRisk+ tends to be applied to large loan portfolios, where light data requirements of the model are perceived to be an advantage and value changes of the portfolio are not seen as relevant.
Data inputs: Exposures (net of any losses in the event of default), default rates, default rate volatility.
Components: CreditRisk+ stands apart from the other two models, in that CSFB does not provide software or consultancy services. An example spreadsheet is freely available.
Strengths: “The CreditRisk+ model is easy to use, because it is elegantly constructed, and requires little data. The analytical approach makes the model easier to understand and more attractive to some people. However, the most important contribution made by CreditRisk+ is possibly that it has helped to build the discussion of how credit portfolio models should work. The fundamental concepts common to other models are particularly explicit in CreditRisk+,” says Tom Wilde, director in risk management at CSFB.
Weaknesses: The standard deviation (volatility) of default rates is not readily available from historical observation, so a structural model is needed to infer this data from other available market data. The mathematical structure of the model is in some respects less flexible.
McKinsey’s CreditPortfolioView (1998), another popular model, doesn’t use historical data, but its default probabilities depend on the current state of the economy.
Moody’s Risk Management Services entered the arena in April 2000 with RiskCalc. This model combines the structural approach based on Merton’s options theory and the statistical approach that analyzes historical data.