These are personal opinions....not the opinions of BMO Financial Group
Nick Le Pan
A bank’s IRB estimates are intended to be predictive.It is hard to understate the importance of this statement by the head of the "Accord Implementation Group". There is a lot of mumbo jumbo in Basel II about adding "conservatism" to a bank's internal risk measures. It is important to keep in mind that the good risk management is based on good risk measurement. Risk measures which are consistently "conservative" (or "optimistic") will lead to poor decisions. Nice to have a record of a regulator placing this in black and white.
Saturday 4 September 2004
Til Schuermann and Samuel Hanson
....it is impossible to distinguish notch-level PDs for investment grade ratings—for example, a PDAA- from a PDA+.Agreed. That's why I believe that the rating agencies, and many banks, have too many ratings in the high quality end of the spectrum.
Mark Illing and Graydon Paulin
....if historical relationships are a good indicator of the future, changes in required capital and provisions for commercial and industrial, interbank, and sovereign exposures will likely be countercyclical under Basel II (i.e., capital requirements will increase during recessions).
More fuel for the fire. One of the interesting comments in Illing and Paulin's paper is the suggestion that, if regulatory capital is cyclical, banks will be induced to change their risk appetite over the cycle. As conditions improve, and regulatory capital requirements decline, they fear that a bank will be tempted to utilize its excess capital by adding new assets, and this may induce the bank to reduce its credit standards to achieve growth. Conversely, as conditions deteriorate, regulatory capital requirements will increase and so the bank may not have sufficient capital to suppport new assets. Under these conditions the bank would select only the highest quality loans, thereby starving otherwise acceptable borrowers of needed financing.
I think this is, at least in part, looking through the wrong end of the telescope. One of the more difficult things for a bank to achieve among all its lending staff, is a common understanding of the bank's risk appetite . So changing lending standards can be a dangerous undertaking, both because it is very difficult to be certain how much more or less risk the bank is willing to incur, but also because it destablizes the bank's risk management discipline. A temporary loosening of credit standards in order to add assets can be very difficult to turn around when the "excess" capital has been used up.
Rather, I think that banks need to be concerned about the cyclicality of regulatory capital because a bank may be under pressure to issue or redeem capital at the nadir or apex of the cycle, when both would be inappropriate.
BBA, LIBA and ISDA
The joint industry-working group believes that the substantial assets in Low Default Portfolios should not be excluded from the IRB approach due to the absence of statistical data to establish and validate PD, LGD and EAD estimates.
Thursday, 20 May 2004
Daniel Rösch
Point in Time Ratings may be penalized under Basel II in economic downturns when borrowers’ short term default probabilities rise. This drawback stems from the presumed fixed asset correlation under Basel II which does not take into account the lower correlation and the lower economic risk which the Point in Time Rating produces.Also worth a look is A traffic lights approach to PD validation by Dirk Tasche
Tasche's paper illustrates something that has been troubling us. Any reasonable confidence interval describing the possible default rates of a given risk grade is likely to include default rates that are also consistent with the immediately adjacent, and possibly other neighbouring, risk grades.
Wednesday, 12 May 2004
Michael Gordy and Bradley Howells:
In he absence of some consensus on policy options [for dampening the cyclicality of Basel II capital], national regulators may pursue their own informal measures (such as encouragement of through-the-cycle rating methodologies). This may have competitive implications, as well as implications for cross-jurisdictional comparison of disclosed regulatory capital ratios.Gordy and Howells continue to advance the argument that "through-the-cycle" ratings are essentially a very bad idea. It is a rating philosophy which ensures that you never really know what risks you are facing.
They propose that banks measure their capital on a "point-in-time" basis, but for purposes of determining compliance with the regulatory minimum, capital be smoothed over time (either by the bank itself, or using a "counter-cyclical" scaling factor provided by national regulators). I don't know about you, but I think it would require a major change in mindset on the part of regulators to expect them to publish a scaling factor which would moderate bank capital levels as the economy deteriorates. I cant envision this actually happening. Despite its other shortcomings, the time weighted averaging of each individual bank's capital requirement seems to me to be much more desirable/practical. It is actually quite similar, in its effect, to the scaling of PDs to long term average levels that I suggested here.
Tuesday, 11 May 2004
Consensus achieved on Basel II proposals
I'm glad the committee members are all on the same page. It's not quite clear to me what has been agreed to. The paragraph on LGD estimation is particularly opaque.
Notes from IACPM meetings
I had a number of interesting discussions with various folks at the IACPM meeting in Paris last week. There seems to be concurrent realization by many credit risk modelers that allocating capital to individual assets on the basis of "unexpected loss contribution" just ain't right. We have found that, in the KMV implementation, UL contribution can result in bizarre results such as capital for an asset being close to the estimated LGD for the asset. In a presentation to the Data and Models Committee, Deutsche Bank said that they have even seen capital being greater than the nominal amount of the assets. Exceeding LGD is strange, exceeding the principal amount is nonsense.
Some banks are thinking about implementing an allocation routine which is based on contribution to the "tail" of the distribution, but there are widely divergent views on what this actually means. There are also concerns with the behaviours that it will reward. Allocating capital based on tail-risk contribution will reduce the capital allocated to higher risk assets. This might not be consistent with the portfolio manager's objectives.
Wednesday, 28 April 2004
Moody's acquisition of KMV a few years back seems to be paying off. Went to see their presentation on the new super-duper private firm default risk model yesterday. They claim that it works substantially better than the old Moody's RiskCalc product, which itself was, surprisingly, found to work better than KMV's Private Firm Model.
The new model produces PD estimates which are tuned to current economic conditions. This was, to my mind, a huge shortcoming of the previous RiskCalc. The presentation showed some quite impressive accuracy ratios for the new model, but they didn't flash anything on the screen to address the issue of whether the PD levels were correct. Oh well, at least I can read the technical document on the GO train.
More from the folks at Kamakura:
Major financial institutions seeking compliance with the New Capital Accords from the Basel Committee on Banking Supervision are no longer comfortable assuming all pairs of companies in a first to default swap basket have the same correlation in their probabilities of defaultMark Levonian:
One can’t answer the question “Does the rating system work well?” without answering the question “What is the rating system supposed to do?”Saturday, 24 April, 2004
Alan Greenspan:
Operating with better information does not mean that banks will necessarily reduce credit availability for riskier borrowers. It does mean that banks can more knowingly choose their risk profiles and price risk accordingly.Thursday, April 22, 2004
Andrew Willis:
...an intensely competitive dealer community is taking on more risk for less reward. This is a trend that can only end in tears.Monday, April 19, 2004
Nassim Nicholas Taleb (edge.org)
We have a bad habit of finding 'laws' in history (by fitting stories to events and detecting false patterns); we are drivers looking through the rear view mirror while convinced we are looking ahead.I agree with Taleb on many fronts. His book, Fooled By Randomness is worth a read. I'd lend you my copy, but someone else took it and didn't give it back....now I've forgotten who it was.
Sunday, April 18, 2004
The Economist has published a good overview of the current status of Basel II. (Paid subscription required)
Saturday, April 17, 2004
Robert A. Jarrow and Donald R. van Deventer
The pursuit of perfection in credit risk modeling will continue for decades, and practical bankers should plan for and implement smooth transitions from one model to the next as the state of the art improves.Jarrow and van Deventer have put together a very useful paper which provides in depth discussion of a variety of credit model validation techniques. Their concluding sentence, quoted above, is wise counsel. No doubt they hope that bankers will be planning for transition from current models to the models that Jarrow and van Deventer sell through Kamakura.
Long Bets update
My "mark-to-market" prediction has been accepted and published as the 149th long term prediction on longbets.org. Apparently each prediction has to be personally approved by Stewart Brand, which is kinda cool. I remember being intrigued, but mostly puzzled, by Brand's Whole Earth Catalog in the early 70's.
Wednesday, April 14, 2004
Peter Kepler, senior analyst at Financial Insights:
In fact, after the market turmoil during the past couple of years and discussions of balance sheet transparency, Kepler eventually expects to see regulations forcing this issue, although he acknowledges that is probably years down the road. Experts say banks for now are interested in the issue of mark-to-market largely for internal reporting, as opposed to trying to sell more loans into the aftermarket.I have submitted a prediction to longbets.org that marking to market will not be required by US bank regulators or accountants before 2024, that's 20 years, by which time I expect to have retired. If accepted, you'll be able to bet fer me, or agin me.
Tuesday, April 13, 2004
The question of what, precisely, the LGD used in portfolio modeling is supposed to represent is turning out to be very hard to pin down. Mich Araten at JPMorgan Chase sent me this Basel discussion document which suggests that:
Banks would be required to report expected LGDs that would not need to explicitly focus on loss severities during economic downturns. A supervisory mapping function would then be used to extrapolate appropriate “economic-downturn” LGDs from bank-reported expected LGDs.Monday, April 5, 2004
I keep wrestling with the issue of deciding on the discount rate to use when estimating LGD. I find it hard to avoid the conclusion that the rate of return required by investors to hold distressed assets includes a premium for the uncertain economic value of the asset, and as such is not an appropriate discount rate to determine the expected economic value of those assets.
Saturday, March 27, 2004
Report of the CIA Subcommittee on Credit Risk: Actuaries have much to learn from these developments about the practice of credit risk analysis.
That's the Canadian Institute of Actuaries.....not the CIA you were thinking of.
Friday, March 26, 2004
Kenneth Emery, Richard Cantor and Roger Arner: In conclusion, there remains much to be explained: even our best regression model explains only about 23% of the variation in loan recovery rates across issuers.
The folks at Moody's identify a number of things which appear to be linked to recovery rates. Unfortunately many of them are not characteristics that can be identified at origination (e.g. nature of the default), which means that this isn't all that helpful in desiging a process to estimate LGD for assets before they default.
Things that do matter: relative seniority and collateral (probably the same thing). Something that doesn't seem to matter: industry (at least there isn't enough data to show that it does matter).
One really interesting observation: recovery rates on loans don't seem to fluctuate with macro factors to the same extent as recovery rates on bonds. This might make it easier to argue that an LGD estimate derived from default weighted average LGD observations would be sufficient to meet the "stressed LGD" requirement proposed for Basel 2.
Lisa Washburn et al: If municipalities were rated on the corporate scale, Moody's would likely assign Aaa ratings to the vast majority of general obligation debt issued by fiscally sound, large municipal issuers...[and] to the bulk of the senior obligations issued by large, fiscally sound municipal providers of essential services.
Alfred Hamerle, Thilo Liebig and Daniel Rosch: We show how the Basel II one factor model which is used to calibrate risk weights can be extended to a model for estimating PDs and correlations.
Alfred Hamerle, Thilo Liebig and Harald Scheule: We find that the inclusion of variables which are correlated with the business cycle improves the forecasts of default probabilities.
Bernd Engelmann, Evelyn Hayden and Dirk Tasche: We show in detail how to identify accounting ratios with high discriminative power, how to calculate confidence intervals for the area below the ROC curve, and how to test if two rating models validated on the same data set are different.
Monday, March 22, 2004
Erik Heitfield is a researcher with the Fed, who published a great paper on double default/double recovery effects last year. He was in Toronto last week going over some new findings relating to the relationship between a bank's rating philosophy and the capital it will be required to hold in the Basel II framework.
Some of his conclusions are that a) banks that have a "point-in-time" ("PIT") orientation in their rating systems will be more able to backtest their PD estimates than banks which take a "through-the-cycle" ("TTC") rating approach, and b) the Basel II rules will lead to different regulatory capital requirements for banks which assume exactly the same risks in their portfolio, but differ in their rating philosophies.
This started me thinking about a couple of things:
Should banks be held to a constant solvency standard throughout the "cycle"?
I think the objective of a stable level of solvency for regulated banks throughout the “cycle” is ill-conceived. It doesn’t reflect the way that the risk profile of, even highly and successfully regulated, banks evolve over the “cycle”. The 99.9% solvency threshold in Basel 2 was, I believe, established with reference to the average default rate observed for rated companies with a particular agency rating (probably around “A”). But the default risk of “A”-rated companies is not stable over the course of a cycle. A-rated companies might have 10 bps of default risk during an average year, but in any particular year, their default risk will be higher or lower than this. The magnitude of the shifts in default risk have been quantified (with an unknown degree of precision) by KMV in their database of EDFs for financial companies.
The same logic must apply to regulatory capital. A bank must capitalize itself properly so that, at the nadir of a normal cycle they are able to maintain their access to markets and, under extreme shocks, still have sufficient buffer to protect their creditors/depositors from loss. However it isn’t realistic to expect a bank to constantly adjust its capital so that, regardless the macro environment, it is always 99.9% likely to avoid failure. If banks were able to do this with any consistency, none would ever fail.
How can regulators standardize capital requirements among PIT and TTC rating philosophies?
I suggest that the way to deal with the potential differences between regulatory capital for TTC banks and PIT banks would be to allow both banks to scale their PDs such that the EL for their portfolio matches the average losses they have incurred in the past. The amount of scaling that banks would need to do would depend on how close they are to either the PIT or TTC end of the spectrum. A PIT bank would scale its PDs down during a recession, and up during a boom. A TTC bank might not need to scale at all.
This would normalize the regulatory capital requirements for both banks against their own past performance. A bank with higher average losses would need to maintain regulatory capital that recognizes this, regardless of whether they are PIT or TTC. This also avoids the problem of having a bank, or (worse) a regulator, trying to adjust levers on the capital model for each bank based on some subjective assessment of where the current state of the economy lies within the cycle (not something that can be determined with any degree of precision).
This would also allow practitioners to use PIT measures to manage their portfolio without worrying that this will create a pro-cyclical capital requirement, placing them at a competitive disadvantage. Most practitioners I come across want to use PIT risk measures for portfolio management, but feel that Basel 2 is pushing them in the direction of TTC.
I know that this is not a panacea. In essence this approach assumes that the risk taking activities of a bank remain constant over long periods of time. Regulators would need to be alert to secular changes in the risk profile in a bank’s portfolio, which might cause the long term loss rate to increase or decrease. If a bank enters a new market, it would need to make some reasonable assessment of what the long term loss rate from the new market is likely to be, and what the impact on the overall portfolio loss rate would be. This would not be easy, but I think it is a manageable problem. To some extent these estimates are selfcorrecting because the new business will generate losses which will then become part of the long-term average that the bank must scale to in future. Gaming the system would be a temporary benefit only.
Til Schuermann has issued a working paper dealing with "What do we know about Loss Given Default". He makes the (correct) statment that the proper discount rate to use for estimating LGD is far from obvious, but some inappropriate choices include the pre-default interest rate of a loan and the risk-free rate. I'm not so sure....
What is the "correct" discount rate to estimate LGD?
It is common to calculate the observed LGD experienced from actual defaults, and then use the average of such LGDs as a factor to be applied in the capital model to predict the loss that could be experienced from new defaults. When quantifying historical LGD experience, it appears logical to discount cash received on account of the defaulted asset at a rate which approximates the return that an investor in distressed assets would have required in order to invest in the defaulted asset. The return required by such investors is invariably quoted to be quite a “high” rate (often 15-25% depending on the prevailing risk free rate), and at least part of the rationale for earning such a high return is the fact that the distressed investor assumes a considerable degree of uncertainty regarding the timing and amount of cash that he will receive from the defaulted asset. In effect, he is being compensated for recovery uncertainty.
In the commercially available capital models (KMV, CreditMetrics), the LGD applicable to a an asset which defaults during a particular Monte Carlo draw is a stochastic variable. In KMV, which I am most familiar with, the LGD is modeled as a Beta distribution, centered on the LGD % input by the user. Assume that the LGD % entered for each asset is the average LGD that we expect to incur if the asset defaults during an economic downturn, this means that recovery uncertainty is explicitly modeled within the capital model. This implies that it is not necessary or appropriate to also use a discount rate that incorporates recovery uncertainty when estimating the LGD % to be entered into the model. Maybe all that the historical LGD analysis needs to consider is the pure time value of money...maybe the risk free rate is the "correct" discount rate to use after all.
I am not sure what the implications of this might be for the ASRF model used for Basel 2. Clearly, it was calibrated against a model which included recovery uncertainty, but the ASRF model itself does not incorporate recovery uncertainty. What are the implications for the discount rate used for LGD % estimation for the Basel 2 LGD estimates?
Tuesday, January 27, 2004
Kamakura:The KRIS Private Firm model incorporates financial ratios, macro-economic factors and public company default probabilities from the KRIS-cr default probability service to correctly model the cyclicality and correlation in small business defaults.
Tor Jacobson, Jesper Lindé, and Kasper Roszbach: [B]anks do not implement counterparty risk rating systems as envisioned by the new Basel II Accord, in such as way that they result in consistent estimates of portfolio credit losses.
This paper reinforces the findings of the recent IACPM Middle Market survey - that bank's credit risk estimates for the middle market are, ahem, somewhat inconsistent.
Brooks Brady, Standard & Poor's: Three investment-grade companies defaulted in 2003
Friday, January 23, 2004
FitchRisk: [P]ublic equity market information, while highly predictive over the short-term, diminishes in importance when analyzing longer term credit quality.
The Economist: [B]ankers would have to be more than usually stupid not to thrive.
Moody's: [N]o defaulting issuers began the year with an investment grade credit rating, resulting in a 2003 investment-grade default rate of zero.
That's the first time that's happened since 1997.
Sunday, January 18, 2004
Daniel M. Covitz and Paul Harrison: [R]ating changes do not appear to be importantly influenced by rating agency conflicts of interest but, rather, suggest that rating agencies are motivated primarily by reputation-related incentives.
A really worthwhile paper, and I would not have predicted the conclusions. In particular, I would have expected Covitz and Harrison to find some evidence that the rating agencies were slower to downgrade companies from investment to speculative grade. But they didn't...and the rating agencies deserve a pat on the back for this.
Of course, I can't help mentioning that the whole analysis is predicated on the notion that the credit markets already know about, and have reacted to, changes in credit risk well before the agencies change a rating. So, overall, a change in rating simply confirms what the market already knows (or suspects). Covitz and Harrison demonstrate that the agencies are equally slow to react no matter how significant the downgrade, or important, the issuer.
Fitch: Self-Referenced CLNs Raise Questions & Concerns (Free registration required)
Thursday, January 15, 2004
Basel Committee on Banking Supervision: Continued progress toward Basel II
These press releases are 99% spin and 1% content, IMHO.
On a different topic.....it seems that my "Test Deck Exercise" might see light of day. I have been advocating a project where banks are supplied a list of actual transactions and they submit their credit risk measures (for the deals that they are in) on a confidential basis to an impartial, trusted third party. That way banks can see whether their risk assessment process are consistent with the market.
This project will build on previous, similar, exercices conducted through RMA and, in Canada, by Mercer Oliver Wyman. I should know in early February whether we will be able to proceed.
Wednesday, January 14, 2004
Associated Press: J.P. Morgan Chase, Bank One Plan Merger
Holy shit!
Tuesday, January 13, 2004
Nick Le Pan: In the area of national discretion, the AIG is cataloguing approaches by member
jurisdictions. That will be an important way to gauge consistency. No decision
has yet been made by the AIG or the Committee to provide this catalogue
publicly...
I guess OSFI doesn't believe that the benefits of full disclosure that they, rightly, want banks to enjoy would be good for regulators. Strange.
Aslo from Nick Le Pan: The Committee expects internationally active banks to operate above
the Pillar 1 minimum requirement.
Then why, oh why, does the Committee believe that the new Accord must be calibrated to maintain the level of capital in the banking system at current levels? If regulators expect some banks to choose to hold capital in excess of minimum requirements, then surely the minimum requirement should be below the level of capital that banks currently hold.
Tuesday, January 6, 2004
NERA: Credit Ratings for Structured Products: A Review of Analytical Methodologies, Credit Assessment Accuracy and Issuer Selectivity Among the Credit Rating Agencies
NERA says that "[t]he impetus for the study is the market debate over the treatment by individual agencies of securities included in a rated structured transaction – in particular, the technique known as "notching." Notching refers to the industry practice whereby one agency adjusts ratings of bonds within a structured finance instrument that may have been initially rated by a different agency." [emphasis added]. Curiously, I've scanned the report and can't find any section that actually deals with the differences in ratings applied to securities included in structured transactions. Unless I'm missing something, the study looks at the differences between ratings assigned to the structured transaction themselves. The ratings assigned to the securities underlying the structured deal aren't actually compared.
Mark Adelson at Nomura thinks that NERA also missed the boat because they focus on rating transitions and ignore the fact that Moody's ratings are intended to reflect expected loss, but S+P and Fitch claim to rate likelihood of default.
In an article that is suspiciously similar to Adelson's report, Saskia Scholtes also finds the NERA report underwhelming.
Kamakura: ....the number of troubled companies in the United States continued to decline in December.
This work is licensed under a Creative Commons License.