There has been a long debate in the risk management industry about whether traditional credit ratings are intended to be "point in time" ratings or whether they are intended to be "through the cycle" credit indicators. The reason that such a debate occurs is the vagueness of the maturity and default probabilities associated with a given rating for a specific company at any point in time. When viewed in the context of a modern quantitative default probability model like Kamakura's KRIS service, the debate about "point in time" versus "through the cycle" is a distinction without a difference. This post explains the reason why.
Even after more than 50 years of experience with traditional credit ratings, practical credit portfolio managers still don't have clear answers to a number of critical questions:
What is the default probability associated with a specific rating for a specific company over a given time horizon on a specific date?
What is the default probability associated with the same issuer over different time horizons?
How do the ratings and default probabilities associated with them change when macro-economic factors change?
Put differently, how does the default probability associated with a given rating change as the company moves through various points in the business cycle?
Because the rating agencies have been unable to articulate clear answers to these questions, two other questions are often asked and debated:
Is a given credit rating a "point in time rating," an index of risk that prevails today but (by implication) not necessarily tomorrow?
Is a given credit rating a "through the cycle rating" intended to measure the average level of credit risk for a given company over an unspecified but presumably long period of time that extends "though the cycle" both for the current business cycle and presumably many others?
A precise answer to these questions only becomes more muddled if one poses it to senior people at the rating agencies. In February 2003 at a well-lubricated dinner in mid-town Manhattan, I posed this question to the head of corporate credit ratings at one of the two leading rating agencies: "What would your agency do with respect to a company's rating if the company was a long run risk level of BBB/Baa but faced 60 days of CCC/Caa risk?" His answer, consistent with the benign credit conditions of the time, was this: "We can't afford to damage our firm's reputation by letting a BBB/Baa rated company default, so we'd set the rating at CCC/Caa." This comment contrasts with more recent statements by other senior rating executives who argue that they have a special obligation to avoid 'type 1' error that overly harshly rates companies. If the first quote is accurate in reflecting rating agency thinking, it means that a rating on any given date reflects the worse of (a) the short term risk of the company or (b) the long run risk of the company. From this view, the rating is neither "point in time" nor "through the cycle." A little reflection in a default modeling context makes the muddled semantics much more clear and transparent.
How does a modern quantitative default model like the Kamakura Risk Information Services suite of default models differ from ratings? The differences are very great and very attractive from a practice use point of view:
Each default probability has an explicit maturity
Each default probability has an obvious meaning. A default probability of 4.25% for a three year maturity means what it says: there is a 4.25% (annualized) probability of default by this company over the three year period starting today
Each company has a full term structure of default probabilities at maturities from 1 month to five years, updated daily. When version 5 of the KRIS default models is released, the longest maturity will be well beyond the five year point.
What does "point in time" mean in this context? All of the default probabilities for IBM, for example, prevailing on March 24, 2009 are default probabilities at that are "point in time" for IBM. Default probabilities at a different point in time, say March 25, will change if the inputs to the default models are different on March 25 than they were on March 24. What does "through the cycle" mean with respect to the default probabilities for IBM on March 24? "Through the cycle" implies the longest default probability available on the term structure of default probabilities because this maturity does the best job of extending through as much of the business cycle as possible. For KRIS version 4, the longest default probability available is the 5 year default probability. If the default probability for IBM on March 24 is 1% at five years, it means that the "through the cycle" default probability for IBM prevailing on March 24 is a 1% (annualized) default rate over the 5 years ending March 24, 2014. In KRIS version 5, the longest default probability maturity will be 10 years, so the "through the cycle" default probability from that point on will be the longest default probability maturity available, 10 years.
To summarize, all of the default probabilities prevailing for IBM on March 24, 2009 are the "point in time" default probabilities for IBM at all maturities from 1 month to 5 years. The "through the cycle" default probability for IBM on March 24 is the 5 year default probability, because this is the longest maturity available. The 5 year default probability is also obviously a point in time default probability because it prevails on March 24, the "point in time" we care about. On March 25, all of the point in time default probabilities for IBM will be updated, including the 5 year default probability, which has a dual role as the "through the cycle" default probability.
There is no uncertainty about these concepts: all default probabilities that exist today at all maturities are "point in time" default probabilities for IBM, and the longest maturity default probability is also the "through the cycle" default probability.
How can these default probabilities be mapped to ratings, and what rating would be "point in time" and what rating would be "through the cycle"? An experienced user of quantitative default probabilities would ask in return "Why would you want to go from an explicit credit risk assessment with a known maturity and 10,000 grades (from 0 basis points to 10,000 basis points) to a vague credit assessment with no known maturity and only 20 grades?" A common answer is that management is used to ratings and ratings have to be produced, even if they're much less accurate and much less useful than the default probabilities themselves.
The posts on March 23 and March 20 discuss in detail how Kamakura maps default probabilities to "implied ratings" which best mimic average rating agency behavior over the period from 1995 to 2004.
Donald R. van Deventer
Honolulu, March 24, 2009