Credit Risk Analysis
MANAGING CREDIT RISK: THE CHALLENGE FOR THE NEW MILLENNIUM Dr. Edward I. Altman
From 2004, I've been working on a Project "Credit Risk Analysis using Advanced Data mining Tools". More info on aires.dei.uc.pt
Some of the techniques that we have been using, include:
- Manifold Learning is a technique for dimensionality reduction. It unfolds high dimensionality data and project it on a lower dimension manifold where it becomes more meaningful and easier to interpret. This technique is widely used since we are considering tenths of financial ratios to get a sharper perspective of the company.
- Neural Networks: Neural Networks are connectionistic machines capable to learn automatically in most the same way as human brains. They are very good in finding higher order correlations and discover hidden patterns in the data.
- Support Vector Machines SVM are one of the most advanced algorithms for data classification. They perform much better than linear regression methods like Logistic and discriminant analysis.
- Genetic Algorithms are algorithms inspired by nature. They use much the same approach as natural selection: evolvable genes that are mixed, mutated and the best individuals selected for the next generation.
Our team have been working on these topics for some time and you can click here for a list of publications on this project.
ABOUT CREDIT RISK
An Overview of Credit Risk Management practices – A Banker's perspective Sumant A. Palwankar Credit Risk Analyst 8 th
Companies extend credit to each other in two fundamental ways: trade credit and leverage. Each of these modes expect a certain return and each encompass a certain risk. You cannot avoid risk. You can only ensure that you are sufficiently compensated for the risk taken. When defaults occur, credit investors suffer losses. Accordingly, issuers should pay investors a spread over the default-free rate of interest proportional to their default probability. Leverage, in the form of short and long-term liabilities, is usually a more permanent, and extensive, feature of most companies' capital structures.
Here, the turnover, or amortization rate, is in the vicinity of 5% to 25% annually, sharply lower than trade credit. Another feature of leverage is an explicitly applied rate of interest that, presumably, reflects the riskiness of the obligor. If the providing investors are not satisfied with their portfolios' return expectations, the remedy may be to obtain credit of riskier obligors bearing higher rates of interest. Here, the individual obligors' return-to-risk characteristics are enhanced as investors exercise control over their portfolios' diversification. Diversification, a portfolio effect, diminishes risk without disturbing the return expectation. In fact, diversification lowers, by 90% or more, the risk-to-return characteristic of the typical "stand-alone" exposure. Trade creditors typically exercise less control over the diversification of their accounts receivable books than leverage-investors over their portfolios.
However, trade creditors do worry when their accounts receivable book is over-concentrated in too few buyers (obligors). Credit losses are a cost of doing business that cannot be ignored. In the instance of banks, credit losses can be sufficiently severe so as to threaten their existence. Accordingly, in both forms of credit extension, lenders need to understand the default rates of obligors in their portfolios. Even when a rigorous collection procedure is instigated upon default, both trade and leverage lenders usually suffer losses. Thus, interest income, or the extent of the receivable owing, must be decreased by an amount of this "loss expectation" to understand the economics of the extension of credit. Ninety percent, or more, of the uncertainty embedded in the economics of credit extension traces to the probability of default of the particular obligor. Thus, getting to the heart of the economics of credit means obtaining timely and accurate estimates of individual obligor default probabilities. Before the fact, there is no method to discriminate unambiguously between issuers who will default and those who will not. Ex ante, that is, we can only make probability assessments of the prospects of default. Default is a deceptively rare event. The typical borrowing firm has a default probability of around 2% in any year. Thus, there is a 98% complementary probability of that firm not defaulting.
However, there is considerable variation in default probabilities across firms. For example, the chances of a triple-A rated firm defaulting is only about 0.02% per annum. A single-A rated firm has a chance of around 0.10% per annum, five times higher than a triple-A. At the bottom of the rating scale, a triple-C rated firm's chances are about 15%, 750 times higher than a triple-A. A naive approach to investing in credit would be, simply, to assume firms will not default. After all, 98% of the time they don't. However, the consequent losses would destroy investment performance or if the investor is a highly leveraged institution, such as a bank, would bankrupt the lender in short order. Strictly subjective judgements are no longer an adequate basis for discriminating among firms' default prospects. The measurement of default probabilities has evolved into a science. There are two critical ingredients to competitive default probability measurement: data and models. Models are the means by which data are transformed into default probabilities. Data pertinent to estimating default probability arise from two sources: financial statements and market prices of firms' debt and equity.
Historically, far more use was made of financial statements than market prices in estimation of a firm's default probability. Statements are, inherently, reflections of what happened in the past. Prices, by contrast, are forward looking. Prices are formed by capital providers as they anticipate the future prospects of the firm. Prices contain, thereby, ex ante information. The most accurate default measurement derives from models employing both sources. There is a limit, of course, to the information that can be extracted from statements or prices. The most functional models are grounded in theory that works. Unfortunately, there is not much theory in economics, macro or micro, which works. Most models are ad hoc, i.e. they lack structure that reflects the causative linkage among the included variables. Even so, ad hoc models have considerable predictive power. But, alas, we cannot determine why they work. Ad hoc models are destined to remain undecipherable "black boxes", even to their designers. When estimating the default probability on private firms, we only have statement data.
We must do the best we can with statement data. The model needs to deal with firms whose capital structure is at variance with the norm. For example, when examining the possible impact of a recapitalization, we will want to obtain an accurate estimate of the new default probability. These requirements should lead us to a model based on causation rather than correlation. The Problem Investing in credit assets requires resolution of two fundamental questions: (1) what is the likelihood of default and (2) what will be lost if default occurs? The probability of default derives from the dynamic fortunes of the issuer. Default occurs when the issuer's resources are depleted to such an extent that a promise to pay cannot be met. As far as what will be lost if default occurs, cross-default clauses in debt contracts usually ensure that the default probabilities for each class of debt for a firm are the same. That is, the default probability of the firm determines the default probability for all of the firm's debt or counterparty obligations. However, the loss in the event of default for each of the classes of obligations can vary widely depending upon on their nature (security, collateral, seniority, etc.). Therefore, the facility agreement bears significantly on the prospects of loss should default occur. For example, typical loss rates in the event of default for senior secured bonds, subordinated bonds and zero coupon bonds are 49%, 68%, and 81% respectively. Although loss given default is an important source of uncertainty for the credit investor, the dominant source of uncertainty, and thereby risk, is the default probability itself The loss given default and probability of default determine the expected loss of the facility. The principal challenge of credit managers is to characterize the default probability of each issuer and then follow the dynamic evolution of the default distribution to monitor for quality degradation. There are three basic types of information available that are relevant to the default probability of a firm: financial statements, market prices of the firm's debt and equity, and subjective appraisals of the firm's prospects and risk. Financial statements, by their nature, are inherently backward looking. They are reports of the past.
Prices, by their nature, are inherently forward looking. Investors form debt and equity prices as they anticipate the firm's future cash flows. In determining the market prices, investors use, amongst many other things, subjective appraisals of the firm's prospects and risk, financial statements, and other market prices. This information is combined using their own analysis and synthesis and results in their willingness to buy and sell the debt and equity securities of the firm. Market prices are the result of the combined willingness of many investors to buy and sell and thus prices embody the synthesized views and forecasts of many investors. Thus, market prices have a measure of uncertainty, or discount rate, embedded in them. By contrast, financial statements are an agglomeration of past transactions; they contain no embedded discount rate. However, the discount rate embedded in an issuer's stock price is not directly applicable to the same issuer's bonds. Debt is a prior claim, relative to equity, on the cash flows generated by the firm's assets. Accordingly, the appropriate discount rate is systematically lower for debt. Creating a model to take into account all these factors and uncertainties is in itself a challenge and a big risk. Risk can never be eliminated. However, we can, and should, use all our ingenuity to predict it and bound it.