The International Accounting Standards Board (IASB) and the US Financial Accounting Standard Board (FASB) are updating the approach for setting Allowances for Loan and Lease Losses (ALLL). In International Financial Reporting Standard 9 (IFRS-9) the IASB require allowances to be set according to the lifetime credit losses weighted by the estimated probability of default 12-months from the reporting date1. A modified version of this is FASB’s Current Expected Credit Loss Model (CECL). With CECL, FASB requires financial institutions to set their ALLL according to each transaction’s Expected Loss (EL)2. One of the effects of IASB-9 and CECL is that ALLL for unimpaired loans will increase during an economic recession. Depending on how the financial institution implements the PD or EL grading, from top to bottom of an economic cycle the ALLL may increase significantly, e.g., by a factor of 10. However, with more careful choices in the ratings framework, the increase in ALLL may be much smaller, e.g., less than a factor of 2. This has a very significant effect on the reported solvency of the bank. Critically, the degree of this pro-cyclical swing depends on the choices made when the rating framework is first implemented by the institution. This paper outlines why the swing occurs and recommends how rating frameworks should be designed to minimize pro-cyclicality.

The Inherent Pro-cyclical Nature of IFRS-9 and CECL

Currently banks assess their Allowance for Loan and Lease losses (ALLL) only for loans classified as impaired. With IFRS-9 and CECL, ALLL will be based on the Expected Loss (EL) for all loans, including those which are not currently impaired. One of the concerns of the American Bankers Association is that the pro-cyclicality of CECL should be better understood3. For IFRS-9 and CECL, pro-cyclicality is the phenomenon whereby as the economy moves into recession, the required ALLL for unimpaired loans increases just at the same time as earnings are down and impairments and write-offs are increasing. A large increase in the ALLL pushes the reported balance sheet towards insolvency and increases the amount of capital to be raised. The choice in the ratings framework can therefore have a significant effect on the reported balance sheet and the survivability of the institution.

As explained below, depending on the EL estimation approach implemented by the bank, the degree of the pro-cyclicality may be moderate or severe. Importantly the degree of this swing will be determined by choices made at the time that the rating approach is adopted by the bank. This is because accountants, regulators and investors will take a dim view of a bank trying to change its rating methodology in the middle of a crisis to make its ALLL lower.

How Choices for Rating Frameworks affect the degree of Pro-cyclicality

To illustrate the impact of the rating choices, consider three different approaches to rating loans. For this illustration assume we are dealing with commercial real estate loans, and the only known risk factor is the loan to value (LTV). As markets improve, values increase and LTV falls, thereby affecting the rating. However, the extent of the change in rating depends on the type of PD or EL model chosen and the philosophy as to what the rating is designed to measure.

Rating Approach 1: Dynamic inputs and a static model

In this example the LTV varies with economic conditions, and at each reporting period the EL grade is assessed by looking at the average loss rate over the economic cycle for loans that started with the given LTV at some point in the cycle, i.e., the model is built by looking at historical data, and, for each year, bucketing the loans according to their LTV at the start of the year, then calculating the average loss rate over the next few years4. So for example if a 65% LTV property fell in value to a 90% LTV, you would look at the historical data set, identify all cases with 90% LTV, calculate how many of those defaulted over the following year, and that would give the estimated probability of default for this fallen loan. The result can be thought of as a look-up table showing the EL for any given starting LTV.

This is a static or “stiff” model in that it does not change with the state of the market, i.e., it does not recognize that when the market is high there is more chance of a fall in the next few years, and when the market is low, there is a higher probability of a rise. Consequently the resulting grades swing with the input LTV but without compensation from the model to take into account the fact that the state of the economy is affecting the inputs. Another way to look at this is that although the inputs vary, the model implicitly assumes that current conditions will persist forever, be they good or bad.

Rating Approach 2: Dynamic inputs and a dynamic model

In this approach the LTV varies with economic conditions, and at each reporting period the EL grade is assessed by using a model which includes the current state of the market and the best forecast of the range of movement over the next few years. The model recognizes that when the market is high there is more of a tendency to fall in the next few years, and when the market is low, there is more of tendency to rise. This consideration of the market returning towards the long-term average acts to counter-balance the change in grade caused by the changing LTV and the result is that the estimation of EL is more stable.

Rating Approach 3: Dynamic inputs and a dynamic model but looking out one year

This is the same as Approach 2, but only looking out over one year, effectively it is a one-year point-in-time model. Given the current LTV and current economic conditions, this approach estimates the EL for the following year. In this case the LTV is changing, and the model is changing because it includes the forecast, but only for a short distance out. In good times the LTV is high, and it is likely that the market will be high the following year. In bad times the LTV is poor and the subsequent year is likely to be poor. This short-term view produces great swings in the one-year EL.


The actual degree of pro-cyclical swing will depend very much on the assets, the details of the model chosen and the historical economic and loss data. However, to demonstrate the potential degree of swing, a simple5 illustrative example6 was set up and produced the results in the table below.

Lowest EL Highest EL Ratio
Approach 1 1.2% 11.1% 8.9
Approach 2 2.9% 4.5% 1.5
Approach 3 0.7% 8.7% 12.7

In this table, the Lowest EL is the lowest estimate of EL during the cycle according to the given rating approach. The Ratio is the EL assigned during the recession divided by the EL assigned during the boom. These numbers are illustrative, but the main message is that the choice of rating methodology can have profound consequences for the stability of the results, and therefore for the stability of the bank.


These results of course point out that that it is important for banks to test what will happen to their rating system as the inputs change during a downturn. The results will vary from bank to bank, but we can already draw some conclusions:

  1. The choice of ratings approach is highly important.
  2. A key factor is that the model should be anti-cyclical: when the market is high, it should be assessing the average future as being worse, and when the market is low, it should assess the average future to be better.
  3. To be able to make that argument to accountants, regulators and investors in the midst of a downturn, there needs to be a track record of using the model in the good times as it will be very difficult to justify switching approaches to give lower ALLL estimates mid-way through a crisis.
  4. To support the modelling approach credibly, there needs to be objective rigor behind the model. Rigor requires clear thinking, and it also requires the collection of as much historical data as possible.
  5. The other key is that the rating should look out over several years, and not be a short-term rating. There is the valid argument that it is more difficult to forecast further into the future, but implicitly we already forecast into the middle future, for example we do not close all banks as soon as the market drops, because our implicit forecast is that the market will recover at some point in the middle future. This forecast can be made explicit, and given some rigor to allow us to make explicit expected loss projections over several years.

One of the consequences of using a counter-cyclical model is that ALLL will be higher in the good times than for a cycle-neutral model. That may have the consequence of constraining business in the good times, especially as the market gets to its peak. This will be difficult to justify to the loan producers, especially as the boom goes on, but it is one of the key building blocks in making banks more robust.

CECL is a good step towards improving the stability of the financial system, but it allows for complexity, and in preparing for it, banks need to think about the consequences of how their ratings systems will treat ALLL in a downturn.

Dr. Chris Marrison
CEO, Risk Integrated

Download a PDF copy of this article:

Your Name (required)

Your Email (required)

E-mail me when Risk Integrated publishes a new article

Input text below


= Required Fields
Contact Us

Contact Risk Integrated today

Want to learn more about this article? Speak to our experts today.

Contact Us
  1. IFRS 9 Financial Instruments
  2. FASB Summary of Board Decisions
  3. Impairment ABA Workshop letter to FASB - Jan 2015
  4. "The next few years" may mean either ladder of years, a fixed number of years, e.g., 5, or the remaining contractual life of the loan. For the purposes of pricing, life of the loan is probably best, but for reserving, a fixed period is likely best, given that in good times the bank will seek to renew the loan, and in bad times the bank will be forced to renew the loan.
  5. In this illustration the PD was assumed to be equal to the probability of the LTV going above 100%, where the average LTV varies with the cycle and then there is a Normal distribution of value around that mean. The economic cycle was a climb in values from an index of 100%, up to 140%, then down to 90%. LGG was assumed to be a fixe 35%. Varying these assumptions causes significantly different results, but the order of magnitude in the differences and the rank ordering of the different ratings approaches remains the same.
  6. Note that this example did not use Risk Integrated’s Specialized Finance System. For a discussion of pro-cyclicality in a different context, see Cyclicality: Good Times get Worse, Bad Times get Better