How Would You Like Your Model-Based Financial Disclosures?

Model-based risk assessments are at the heart of several current industry initiatives that have been adopted, ostensibly, to improve the quality of financial disclosures.

CECL and IFRS 9 – which use projections of expected loss to determine loan-loss reserves – are two of the most obvious examples. But what do investors really want from such model-driven disclosures?

Tony Hughes

They don’t want all the trade secrets, because that would aid competitors and erode returns. They do want to know if there are skeletons in the closet that could haunt the company’s future performance.

Models can be a way to provide useful summaries of portfolio-level information. Unfortunately, they also bring several unwanted side effects.

Disclosures may be skewed or manipulated by clever modelers, so there is a strategic element that may be difficult for auditors to detect or control. Models, moreover, feature a randomness element that make disclosures more volatile and a technical element that make disclosures more complex – and potentially harder for investors to comprehend.

So, what would constitute an optimal model-based disclosure? Let’s now address this question in the context of expected loss accounting.

From Incurred Loss to CECL and IFRS 9

Before breaking down the modeling options for financial disclosures, we need to understand how we arrived at this stage.

Prior to the global financial crisis (GFC), distressed loans were only recorded in a company’s financial statements when some sort of adverse event had been observed. If a borrower missed a payment or declared bankruptcy, for example, the impairment would be recognized by a firm’s accountants. Conversely, if nothing bad happened, nothing would be recorded. This approach, known as the “incurred loss” (IL) methodology, did not rely on credit-loss models.

In the lead-up to the GFC, banks were originating a lot of really bad loans – mainly mortgages. By the second half of 2006, these loans they were sitting on bank balance sheets.

Since the economy was growing and house prices were rising, the bad loans were technically still performing. Under the IL method, however, investors were largely oblivious to their presence, and banks still appeared to be highly profitable.

We all know what happened next.

In response to the GFC, international accounting bodies proposed an overhaul of the way loans are treated by accountants. The push for restructuring began in 2008, and CECL – the protocol that replaced IL in the U.S. – was formally adopted in 2016.

Under CECL, at the point of loan origination, a bank must model and project its expected lifetime losses and report this figure, in present value form, in its financial statements. The projections must consider tail risk, which means that a handful of stress scenarios must also be factored into the loss valuation. The international standard, IFRS 9, is more complicated, but follows the same broad brushstrokes.

Exploring Disclosure Options

Imagine you are an investor who wants to gauge a financial institution’s performance and to be told of any skeletons that may be present. What you’d need to do first is to deconstruct the expected loss calculation into its constituent parts. To determine the type of disclosure that best meets your needs, consider the following five information packages:

Package A: Tells you only when loans suffer an impairment – the same as that offered by IL methodology.

Package B: Tells you information about the observable credit characteristics of the loans in the portfolio – i.e., whatever is available, given a particular loan product type. This may include average credit scores, debt service burden, loan-to-value, leverage ratios and/or ratings.

Package C: Tells you through-the-cycle estimates of expected losses for the loans in the portfolio. If calculated correctly, these numbers will not change at different points in the economic cycle. They assume steady state economic conditions and are agnostic to any particular economic forecast. Moreover, they capture the core creditworthiness of the borrowers and summarize it in the form of expected loss estimates.

Package D: Tells you point-in-time estimates of expected losses. These can be thought of as unconditional baseline forecasts that incorporate – either implicitly or explicitly – a prediction of future macroeconomic conditions. These estimates will vary across the economic cycle, but otherwise convey the same information as through-the-cycle estimates.

Package E: Tells you everything you would expect from a CECL disclosure. This is basically Package D, combined with a range of different scenario-based predictions of tail events.

You can see that the information being offered starts out in its rawest, purest form, and, with the help of a bank’s internal staff, becomes more and more processed as we move to the later packages. A and B are raw; C adds in models to determine overall creditworthiness; D layers on specific baseline economic projections; and E adds scenarios, including assumptions about the probability that they will actually occur.

Under CECL, you are getting Package E and nothing else. With IFRS 9, you are basically getting Package E combined with a little bit of Package A; the international standard has what’s called “staging,” which classifies loans based on actual impairments.

Do-It-Yourself vs. Ready-Made

Would you like to be given  the raw ingredients that will allow you to make your own sausage, or do you want the fully processed, ready-made version? That’s the key question investors must consider when pondering the type of financial disclosure that will best suit them.

Seeing the raw ingredients (perhaps akin to the IL methodology) will expose all the bad stuff, no doubt, but it may take some effort to actually find it. With the ready-made sausage (CECL), you don’t have to do any work yourself – but you risk allowing the bank to hide some nasty, cheap off-cuts of meat with fancy seasoning provided by clever in-house modelers.

Let’s leave the sausage-making metaphor for now – it’s giving me indigestion.

As a garden-variety investor, you may not be able to do your own credit modeling, especially since the bank will only disclose aggregate data in the packages offered. You will, though, have in your mind your own forecast for the broader economy and be just as capable of imagining your own downside scenarios.

What do investors gain by outsourcing the construction of these elements to the banks? It's not like the insights of bankers into the prospects of the economy or the nature of possible downside scenarios are really more incisive than those of the investor.

I suspect that every investor would like access to a bit more information from packages A and B. If then given a choice between C, D and E, I would choose C – the least processed of the available packages. I wouldn’t want to entrust the bank with any of the stuff I can accomplish perfectly well on my own, thank you very much.

Parting Thoughts

Primarily what investors should be asking for is raw, unprocessed data – and agnostic, uncontroversial, empirical modeling that would be difficult for them to do on their own. The IL method had few advantages, but a big one was that the disclosure was based on raw data. There was, in short, power in its simplicity.

We should be very careful what we wish for with banks. Elaborate Rube Goldberg-style disclosure systems don’t help investors very much. We’d be better off following a raw food diet and requiring the banks to stick to the unvarnished facts.

A heavily processed diet is quite unhealthy for investors, so I fail to see why global central banks are so keen to impose it.

Tony Hughes is an expert risk modeler. He has more than 20 years of experience as a senior risk professional in North America, Europe and Australia, specializing in model risk management, model build/validation and quantitative climate risk solutions. He writes regularly on climate-related risk management issues at UnpackingClimateRisk.com.