FINS 5550 / FINS 3650 International Banking

Topic 4 Recap and Post-Lecture Readings

Premise The lectures are intended to provide structure, on the basis of which you will be able to understand and interpret primary sources. Thus each lecture is the preparation for the suggested readings, not the other way around. These recaps are intended to reinforce the structure of your understanding, after you have finished the readings. The following recap is offered as another tool to provide structure. As will be evident, you need to put in some additional effort to turn this recap into a proper discussion of the topic. Please feel to post your recap on the appropriate discussion thread. Credit Risk Eighty percent of the average banks capital is held against credit risk. If credit risk accounts for 80% of the banks inventory cost, its a fair bet that credit transformation accounts for a similarly large portion of bank profits. Credit risk arises whenever the bank has an exposure which requires a counterparty to remit funds. The exposure can arise from a loan or loan-type product derived from a given origination channel (direct solicitation, agent solicitation, brokered, or reverse inquiry). [What is reverse inquiry?] The exposure also can arise from a contingency such as a line of credit [Whats the difference between a line of credit and a revolving line of credit?], letter of credit [Whats the difference between a letter of credit and a line of credit?], or performance bond. Alternatively, the exposure can arise from a swap exposure. [Describe an example of when a bank has an exposure due to an interest-rate swap. How does one quantify the exposure?] While this recap generally refers to the exposure as a loan, the form of the exposure doesnt really matter. What does matter is the likelihood that the exposure will be repaid. Traditional Underwriting Traditional banking calls for each exposure actual or potential to be individually underwritten a process which demands time, effort, and expertise. Many discussions of credit underwriting begin with the so-called Four Cs of underwriting (see Dun & Bradstreets take at http://smallbusiness.dnb.com/business-finance/business-loansbusiness-credit/12154-1.html, or a longer laundry list at http://www.creditguru.com). Not everyone agrees what constitutes the Four Cs, and what is indisputable is that more than four aspects should be taken into account: The Five Original Four Cs Character D&B lists a number of factors, mostly speaking to the business history of the counterparty. IMHO, the real meaning of Character is

Topic 4 Recap and Readings

integrity. There is an old adage that, if youve picked the wrong party to lend to, you cant make it better by negotiating tougher terms. Capacity the ability of the counterparty to pay debt service (principal and interest) on the exposure, with a focus on free cash flow. Implicit in this C is the notion that the proceeds of a loan increases the capacity of the counterparty to generate cash, for example, where the counterparty builds a new factory or adds new equipment. The late economist Hyman Minksy has an interesting spin on this. Read Paul McCulleys brief discussion at http://www.pimco.com/EN/Insights/Pages/Global%20Central%20Ban k%20Focus%20May%202009%20Shadow%20Banking%20and%20 Minsky%20McCulley.aspx, [Find your own examples of speculative or Ponzi borrowing. Where does Australian housing fit on the spectrum? Chinese housing? Owning a professional sports team?] Capital the financial resources of the counterparty. IMHO, the only way this makes sense in conjunction with Capacity is if Capital refers to financial resources OTHER THAN those arising from the loan itself. Conditions exogenous factors (i.e., factors generally applicable to the economy, industry, or region, rather than factors which are specific to the counterparty). Collateral the right to seize property or other assets if the counterparty defaults. IMHO, collateral should be viewed as an exit strategy, rather than an excuse for entering into a loan. The only time one needs to exercise ones collateral rights is if the counterparty doesnt have the capacity to make debt service. In other words, dont be willing to lend on the strength of the collateral unless you are prepared to own the collateral. [How does my comment here relate to Minskys view of the world?] Other Factors Conveniently Beginning with the Letter C Covenants binding promises made by the counterparty to constrain its future behaviour, such as keeping total leverage (the ratio of assets to equity) within a prescribed maximum. Compliance the willingness and ability of the counterparty to comply with all regulatory requirements. IMHO, this is only on the list because of rules unique to the United States where a lender which forecloses on a mortgage can be held liable for environmental cleanup costs. Competition the counterpartys competitive advantages and disadvantages. IMHO, this is a bit of a reach, as it probably falls within most peoples understanding of Conditions. [Can you think of any additional factors an underwriter should consider?] The Modern Approach to Credit While nothing quite compares to old-fashioned underwriting, the realities of modern banking are that the average bank wants to create more loan exposures than it

FINS XX50 International Banking

Topic 4 Recap and Readings

has the personnel and time to examine on a case-by-case basis. The modern approach to credit automates the process. The general approach is probabilistic in nature, whereby losses are considered to be stochastic: The expected amount of loss on a given exposure equals the probability of loss (PD, or, old school, frequency) multiplied by the loss given default (LGD, or, old school, severity) multiplied by the amount of exposure at default (EAD). Thus actual losses are expected to follow a probability distribution. The margin priced into the loan product contributes to loan-loss provisions, which are intended to cover expected losses, or the mean of the loss-probability distribution. Capital is required to cover unexpected losses, which corresponds to that part of the loss-probability distribution greater than the mean. Some unexpected losses are SO unlikely that it is uneconomic to hold capital against them. The probability of unexpected losses greater than the banks capital corresponds to the probability of bank failure.

[What does the area in red represent?] In general, the banks board, through its risk appetite, sets the risk of failure which it is willing to tolerate. (Of course, this risk of failure cannot be greater than the risk of failure which the regulator is willing to tolerate; that risk is reflected in the minimum capital requirement.) In practice, bank boards often express their tolerable level of failure in the form of a rating. For example, a banks board may express their intention to maintain a rating of double-A. Based on historical defaults by double-A rated institutions, this corresponds to a one-year chance of failure of between 0.03% and 0.07%. APRA generally sets the tolerable risk of failure at 0.10%. As an overview of the credit process, read the BCBSs Principles for the Management of Credit Risk, or at the very least, read the introduction to this document on the webpage. The shorter Sound risk assessment and valuation for loans is worth opening up. Australian Prudential Standard 220 has these principles in mind when it discusses credit quality, and sets the general requirements for credit risk management, monitoring and provisions. [What is the difference between specific provisions and general provisions? What is the impact of provisioning on capital adequacy?]) APRA has also

FINS XX50 International Banking

Topic 4 Recap and Readings

published four sets of Guidance Notes: Impaired Facilities, General Provisioning, Specific Provisioning, and a shorter one on Ratings Systems. (For anyone still seeking an assignment topic, consider writing up a recap of this standard and these guidance notes.) Since the dawn of banking, there have been regulatory and practical constraints on betting the bank exposing too great a percentage of the banks capital to a single counterparty. Australian Prudential Standard 221 is the current version of this rule. [What are the prudential limits set out in APS 221? The standard sets out a prior consultation limit. Do the requirement that an ADI consult with the regulator give APRA a right to prohibit an ADI from increasing its exposure?] Even if a banks exposures are within the prudential limits, however, recall that credit concentration risk is a Pillar 2 risk, and a bank is expected to budget more capital the lumpier its exposures are. As to calculating the basic capital requirements, Australian Prudential Standard 112 sets out the standardised approach to credit risk. In an approach highly similar to Basel I: Multiply each exposure by a credit conversion factor, and then by a risk-weighting, where the capital required to support credit risk equals the sum of the products. [What are the major categories of exposures? What are some key CCFs?] The big difference from Basel I, though, is that the risk weights are based on rating rather than just the type of counterparty (e.g., government, bank, or corporate). The regulatory capital requirement equals 8% of total risk-weighted assets. APS 112 in effect establishes 8% of risk-weighted assets (subject to adjustment in the form of the Prudential Capital Requirement) as the amount of capital which produces a tolerable level of unexpected losses for standardised banks. Check out the APRA discussion paper on implementing the standardised approach to credit risk. Australian Prudential Standard 113 allows banks to be authorised to use their internal ratings models to determine their regulatory capital (also subject to a PCR adjustment). [What is the difference between Foundation IRB and Advanced IRB banks?] The four biggest Australian banks are all authorised as AIRB banks. The only Australian FIRB bank is Macquarie, and there is the expectation that Macquarie qualify for AIRB status presently. There is a discussion paper on implementation for this approach as well. Estimating Losses Under the modern framework for credit-risk management, losses are stochastic, which means that the range of possible losses follows a probability distribution of some sort. The term expected loss refers to the expectation of credit losses, or the mean of the distribution. Estimating expected loss (EL) is critical for two reasons. From the perspective of optimising bank profitability, EL should be reflected in the pricing of credit products, as the margin charged on any credit exposure should be large enough at the very least to cover credit losses. From a prudential perspective, EL serves as the basis for general provisions. The standard deviation of the probability distribution is a measure of the volatility of expected credit losses. The term unexpected loss (UL) projects losses corresponding to a given tail value of the probability distribution. For example, assume a tolerable risk of failure of 0.10% (which is the standard APRA and other regulators use as a default setting). Assume further that the loss probability distribution is normally distributed (not a very good assumption, by the way). The term UL would refer to the losses at 3.09 standard deviations above the mean of the loss probability distribution, MINUS losses at

FINS XX50 International Banking

Topic 4 Recap and Readings

the mean. Thus the sum of EL and UL equals losses @ 99.9% on the cumulative probability distribution, or EL + UL = L99.9%. Estimating Losses Probability of Default Again, Loss L = PD LGD EAD. How does one estimate L? One approach might be to use the rating of the counterparty (for after all, thats the principal difference between Basel I and Basel II, in that Basel II uses ratings to determine risk weights. (In Basel speak, a rating agency is known as an external credit assessment institution, or ECAI.) But is estimating loss what the rating agencies claim to be doing? (Please examine the websites of both Standard & Poors and Moodys. After having taken much criticism during and after the GFC, the rating agencies are trying to be more cuddly, and have produced a number of explanatory publications for lay-readers. In particular, check out S&Ps website www.aboutcreditratings.com, which has links to a number of publications, including guides on ratings essentials, criteria, and performance. The charts shown in the lecture theatre were drawn from these.) The rating agencies have differing approaches. Standard & Poors and Fitch state that their ratings represent an opinion as to the likelihood of full and timely repayment of principal and interest in other words, these two agencies consider the probability of default. Moodys, on the other hand, considers not only the likelihood of default, but also the financial loss suffered in the event of default in other words, the product of PD LGD. So at first glance, the rating agencies seem to be estimating loss, either directly (in Moodys case) or indirectly (in S&Ps and Fitchs). However, on closer examination, the rating agencies dont claim to be estimating the likelihood of default or loss. Instead, they claim to be assessing the RELATIVE likelihood of default or loss. In other words, they state that a counterparty with a triple-A rating is less likely to default than a counterparty with a double-A rating, which is less likely to default than a counterparty with a single-A rating, and so on. For this reason, the actual default levels for a given rating fluctuate throughout the business cycle. That said, the rating agencies are very protective of their through the cycle cumulative default rates, i.e., the default rates over any given ten- or fifteen-year cycle. Thus it should come as no surprise that both the implementation guidelines for Basel II and APS 113 require a through-the-cycle probability of default. But SHOULD banks be using a through-the-cycle probability of default, when the banks capital is intended to cover unexpected losses which could be incurred over the next year? [This is the issue of pro-cyclicality. What is pro-cyclicality, what is the Basel III approach to pro-cyclicality, and what is your opinion as to pro-cyclicality?] Alternatives to Ratings But remember, the whole point of APS 113 is that an authorised bank can substitute its own ratings system for the agencies system. Many banks approach credit risk from a completely different angle: Rather than model through-the-cycle default probabilities, these banks calculate the point-in-time probability of default. [How does this fit in with the concept of pro-cyclicality?]

FINS XX50 International Banking

Topic 4 Recap and Readings

One approach is to estimate the probability of default from accounting data. One of the earliest models was Altmans Z-Score (see the slides for the exact formula; the original paper "Financial Ratios, Discriminant Analysis and the Prediction of Corporate Bankruptcy was published in the Journal of Finance in 1968). In general, there is very little evidence that accounting models have a high predictive power more than one or two years into the future. Then again, if one is calculating the one-year PD, maybe an accounting model is sufficient. A more popular approach is to use a structural model of default, often called a Merton-style model after Robert Merton. Merton worked very closely with Myron Scholes and Fisher Black, after whom a certain options-pricing model was named. Merton considers the equity of a firm to be an option over the firms enterprise value, with a strike price equal to the amount of the debt outstanding. In other words, the shareholders can claim the company if they pay off the bondholders. In some cases, the option expires worthless. In other words, the enterprise value is less than the amount of the debt. In other words, the company defaults. The distance to default is the market capitalisation of the firm as a multiple of the standard deviation of the enterprise value. Assuming, as Merton does, that the enterprise value follows a normal distribution (again, not a very good assumption), if the distance to default is 3.09 standard deviations, the probability of default is 0.10%. The most popular Merton-style model is KMV, which was purchased by Moodys in 2002. Instead of mapping the distance to default onto a normal probability distribution, KMV has developed an empirical map that converts distance to default into what they call an Expected Default Frequency. Perhaps the most compelling feature of KMV and similar models is that time is an input. So, for example, we know that the cumulative default rate on single-A bonds increases with time. Yet the rating of a one-year bond is the same as the rating of a fiveyear bond from the same obligor. However, the five-year EDF is higher than the one-year EDF. APRA made a submission to the BCBS when Basel II was still in the formulative stage, discussing the use of credit models by Australias largest banks. Read the report at http://www.apra.gov.au%2FADI%2Fupload%2FCredit-Risk-Modelling-Current-Practicesand-Applications-October-1999.pdf. Estimating Losses Loss Given Default Basle II and APS 113 set out two key principles relating to loss given default. First, a bank must use a downturn LGD. If the exposure is collateralised, this means that any projected recovery reflects prices in a down market. Second, the LGD should reflect the economic loss, which in turn reflects both enforcement costs and delay. An interesting question is what discount rate a bank should use when calculating the net present value of projected recoveries. [First, which discount rate produces the most conservative results? the least conservative results? What do you think is an appropriate discount rate? APRA thinks that the discount assumption should be no less conservative than the contract rate.]

FINS XX50 International Banking

Topic 4 Recap and Readings

Estimating Losses Exposure at Default Basel II says very little about calculating exposure at default. At the very least, the EAD is not less than any amounts actually drawn on any facility. Some commentators suggest that the credit conversion factor should be used to calculate the EAD with respect to undrawn lines, but a more conservative approach is to assume that all lines will be drawn before a counterparty goes into bankruptcy. [What do you think? APRA generally is in the latter camp, except in the case of credit cards, where actual patterns of usage are more relevant to estimating EAD than the formal credit line. Do you think APRA has it right. Or, in other words, if YOU were going to go bust, would you max out on your credit cards?] Finally, review what the BCBS has to say about validating internal ratings systems. Credit Scoring When discussing the IRB approach to credit risk, it is too easy to get caught up in the ratings produced by the agencies. While lots of companies are rated, vastly more counterparties are not rated, especially small companies and individuals. To assess credit in these sectors, it is essential that a bank use credit-scoring models. By far the most popular credit-scoring model is the one produced by Fair Isaac & Company, commonly known as FICO score used by 90 f the top 100 banks in the world, and every single one of the U.S. top 100 banks. The Fair Isaacs website has a series of white papers on credit scoring (registration required), and the company history is interesting in its own right. Most important, though, is their description of the composition of the FICO score. [How does the FICO approach to credit compare to the Four Cs?] A credit score is a binomial classification system: Based on a score, the bank decides either to extend or deny credit. The modern history of binomial classification systems commences with the Second World War, and stems from the use of radar to detect enemy aircraft. Indeed, the receiver operating characteristic, a key tool used in evaluating binomial classification systems, takes its name from the radar receivers, and the ability of the operators to distinguish between flocks of seagulls and formations of bombers. Much of the literature on evaluating binomial classification systems is in the field of medical research, and the introductory article on a website dedicated to anaesthesiology is strongly recommended. In short, the receiver operating characteristic (or ROC) is an ordered pair: the x value is the percentage of false positives returned by a given test and a given discrimination threshold, and the y value is the percentage of true positives by the same test at the same discrimination threshold. The receiver operating characteristics across all discrimination thresholds traces the ROC curve, and the area under the ROC curve (or AUROC, also called the AUC) is a commonly used measure of the effectiveness of a test. [How does one plot a ROC curve? What does it mean to say that there is a tradeoff between sensitivity and specificity? What is the confusion matrix? What are the costs of Type I and Type II errors? If one has plotted a ROC curve for each of several competing tests, how does one choose the best test? The AUROC is commonly used, but is it a good indicator of the best test?]

FINS XX50 International Banking

Topic 4 Recap and Readings

In the context of a credit-scoring model, the critical factor is the banks scorecard. The key to formulating a scorecard is to determine an acceptable level of risk for a given product, which determines the tolerable level of false positives (i.e., applicants who score above the discrimination threshold, but which prove to be defaulters). The optimal scorecard is the one which generates the greatest amount of business (true positives, or applicants who score above the discrimination threshold and in fact repay their exposures). Its actually more subtle than that. Remember that price is part of the equation, in that the credit spread on a product must be large enough to cover the expected cost of credit losses. If the credit spread is high enough to cover the marginal applicant, it might well be too large for the best applicant, and the best applicant might take up a product from a competitor which is geared towards better credits, and which is therefore more finely priced. [Coming up with an example of this is a perfect opportunity to earn extra credit / class-participation points.] See you in the lecture theatre.

FINS XX50 International Banking