Risk-grading philosophy: through the cycle versus point in time

RMA Journal, The, Nov, 2003 by Jeremy Taylor

Three closely related questions are examined in this article: 1) How cyclically sensitive should a risk-grading system be? 2) How should risk grades be converted into default probabilities? and 3) How does all this interact with facility coding and recovery rates?

The issue of cyclicality in an institution's risk-grading system is integral to the system's design and critical to how it gets used. It's a question that peeks out from various points of discussion in the latest Basel documentation but without ever getting posed--or answered--directly. This is probably because of the innate difficulties in defining and measuring it, as well as the fact that there are both advantages and disadvantages to having risk grades that are stable through the cycle.

Basel's CP3 document implies that risk grade (RG) systems should reflect longer-term considerations:

   Although the time horizon used in
   PD estimation is one year ...
   banks must use a longer time
   horizon in assigning ratings. (1)

But it also emphasizes the importance of risk grades that are sensitive to underlying changes in credit quality as they occur and, through appropriate stress-testing, sensitive to possible downside scenarios.

RG systems are typically ordinal--for example, we know that a RG 5 borrower has a higher likelihood of defaulting than a RG 4 borrower, but we don't know how much higher. It is not until each RG is assigned a probability of default (PD) that we can say how much more risky a RG 5 borrower is, thus making the system cardinal. Because cardinality is becoming increasingly necessary--for pricing, profitability measurement, capital allocations, limits-setting, and so forth--it's hard to separate how cyclically sensitive a bank's risk grade should tar from translating risk grades to default probabilities. We'll begin discussing the first question by assuming static factors--i.e., a fixed translation of each RG into a PD (or PD range). Then in the following section, we'll discuss how best to come up with that translation. To understand how we go from default to loss estimation, we then need to bring in cyclicality, in recovery rates.

How Cyclically Sensitive Should a Bank's Risk Grades Be?

We should start by asking what a RG system needs to do:

1. Credit approval--to guide loan structuring, conditions, pricing, profitability.

2. Credit monitoring at the borrower level--to guide whether additional scrutiny is required, and whether the terms or pricing of a transaction need adjusting/renegotiating as circumstances (including renewal) allow.

3. Credit monitoring at the segment/line of business/portfolio level--to guide the bank's marketing and exit activities and resource allocations, which includes limits-setting and approval authorities.

4. Loss reserving and capital planning--as input into the determination of the loan loss allowance, expected loss (EL) and economic capital (EC) requirements.

For the first two roles, it's advantageous to have RGs that reflect the current point in the cycle and thus translate into PDs that are conditioned on how the industry and economy are currently performing. In deciding whether to approve a new transaction, it makes a difference whether the industry in question is nearer a peak or a trough of its cycle. This suggests a point-in-time (PIT) approach, whereby borrowers are regraded immediately as their fortunes change, whatever the cause.

For the third role, PIT grading has advantages (such as where to look for vulnerable borrowers). At the same time, the strategic issues involved make a longer-term viewpoint more appropriate (to determine resource allocation based on how volatile an industry tends to be over time rather than on where it happens to be today).

The case for cycle-neutral, through-the-cycle (TTC) grading is clearer when we turn to the fourth role. As long as we're facing recurring cyclicality (or mean-reversion, as we'll explore later) in the portfolio, then EL and EC should be relatively stable with any short-term ups and downs in actual losses around that mean value absorbed by capital. It's because of variability and covariability in losses that capital is required. EL is intended to represent what the bank might lose on a given customer in a long-run average sense, irrespective of the point in the cycle at which a loan is booked. The provision for credit losses is a cost of doing business and is covered by spread income, leaving a return on the capital employed as a residual.

We'll confine our discussion to changes in credit quality driven by changes in the macro environment. This corresponds to the systematic risk component of the capital asset pricing model. Whatever ratings philosophy a bank adopts, whether PIT or TTC, it's clearly appropriate for any credit quality change that's due to company-specific (idiosyncratic) factors to trigger a grade change.

Where are banks in their ratings philosophy? A 1998 study by William Treacy and Mark Carey, both from the Federal Reserve Board, found that "... [E]very bank we interviewed bases risk ratings on the borrower's current conditions." (2) But because the categorizations are so nebulous, their conclusion regarding the prevalence of PIT grading has to be viewed cautiously. How did the responding bankers interpret the question? Was it just the RGs or also the PDs into which the grades map?