Model Behavior: Banks Must Get Data Right for IFRS 9

July 28, 2017

New regulations and accounting standards are the main drivers of changes to how banks worldwide must operate, maintain their books, and manage their data, with IFRS (International Financial Reporting Standard) 9 one of the most significant. As Wei-Shen Wong discovers, having the right data is imperative as it will affect the outcome of their models and their ability to comply with IFRS 9.
IFRS 9, which takes effect on Jan. 1, 2018, addresses recognition and measurement of financial instruments. This standard, which will replace the current recognition standard under International Accounting Standard (IAS) 39, is proving stressful for many banks, particularly because it changes the entire workflow of how firms account for a wide range of financial assets, and forces banks to be more forward-looking, such as making provisions for expected credit losses (ECLs) from the point of origination instead of recognizing these losses only as and when they occur.
The scope of the impairment requirements under the new standard are now much broader, says accounting and business advisory firm Baker Tilly Monteiro Heng in Malaysia.
The ECL model requires impairment loss to be recognized based on ECL even if a loss event has not occurred. “Entities are required to book in day-one credit losses under the ECL model and update the loss allowance for change in the ECL at each reporting date to reflect changes in credit risk since initial recognition,” says Peng Kuan Lock, partner and co-leader of audit and assurance at Baker Tilly (Malaysia).
A particular issue faced by banks in Asia is that some—especially in developing Asian countries, where financial markets are still nascent—lack the data required to build their ECL models, which affects their ability to calculate ECL.

“In general, banks in Asia lack data. Most of the banks only started on the IRB (internal ratings-based) framework recently. They might not have enough loss data, whereas Europe has been on the IRB approach framework for about 20 years. IFRS 9 is more complicated than IRB, which takes a snapshot, while IFRS 9 shows the change in credit quality since the inception of the exposure,” says Sky So, partner of financial services risk management at Ernst & Young.

The new impairment model would give rise to implementation challenges, particularly around getting the appropriate data and setting a consistent approach.

“Currently, most entities do not collect the amount of credit information required by the standard. Entities will need to significantly modify their current credit and information systems in order to gather the required information,” Lock says.

Historical data requirements in the region are highly dependent on the type of impairment models a bank intends to develop, says Adrian Lee, head of financial services at KPMG Malaysia. Conversely, the type of impairment models to be developed is also dependent on the data a bank has available.

“Banks with sufficient historical data, ranging a full business cycle for the particular product, or a full economic cycle (deemed to be seven to nine years) can consider more complex approaches to impairment models, should the quality of data be sufficient as well,” Lee says.

No Data? No Excuse

What happens when banks lack sufficient data to effectively and accurately build their ECL models? Not having the data required does not render a bank exempt from having to calculate ECL and make provisions for it: IFRS 9 is not optional, So says. “The challenges are much higher. It is an accounting requirement, and not a regulatory requirement. It is closer to a best-effort basis,” he adds.
If a bank lacks the data on, for example, the probability of default (PD), it can use proxy data or benchmark data. This is an acceptable practice, says Arnaud Picut, global head of risk practice at Finastra.
Banks affected by this would typically be those that do not currently operate within IRB standards. For example, banks in countries like Myanmar and Laos use the Philippines as a benchmark, finding their banks similar and comparable enough to do that. Then as a bank matures and starts collecting the data required to build these models and calculate ECL provisions accurately, it will slowly wean itself off the proxy benchmark.
“When they don’t have default data, they have a low benchmark. They have to find another country or another portfolio that can be used as a reference. It could be data from the central bank—or, for example, if they don’t have 90 days default data, they might move it down to 60 days, and then increase it artificially. It is accepted as long as it’s properly documented,” Picut says.
Although there are uncertainties and risks associated with benchmarking, it is an accepted practice in Europe, he adds. “Now there are so many data providers that provide benchmark data from different countries. So they aren’t too worried, because it is accepted. They are aware it is not totally correct, but they know in three years’ time, they have to get it correct. It’s a three-year journey.”
To meet the deadline, firms will deploy models that are relevant and proportionate to their current usable data. They can then consider refining and enhancing the models when their data quality and quantity improves.
KPMG Malaysia’s Lee says banks will apply a range of models to calculate the ECL components. ECL is a product of a firm’s forward-looking probability of default (PD), loss-given default (LGD) and exposure at draft (EAD).
“The type of models for each portfolio asset will depend on materiality, proportionality and relevance. Types of approaches range from the simple to the complex types, all dependent on data quality and availability, and the homogeneity of risk of the product segments. For example, for an immaterial portfolio where historical data is severely limited and the firm is applying judgmental credit rating on its customers, a ‘proxy approach’ can be considered in computing the ECL,” Lee says. “A proxy portfolio can be identified from the firm’s existing larger modeled portfolio where data availability is not an issue, or can be identified through external data released by Bank Negara Malaysia (Malaysia’s central bank).”

Calculation Complexity

One major Asia-based bank is working with solution provider SAS to deploy this project globally from Singapore. Bank officials, speaking on condition of anonymity, say the SAS solution allows the bank to use its data, link it with a set of models, and perform a fairly simple calculation of ECL discounted to today. The complexity is in getting the PD, LGD and EAD developed and implemented in an acceptable manner.
On the wholesale banking side, the bank has global models and systems in place. “For example, if you are in Bangladesh, China, Hong Kong or Singapore, when you meet a client and you onboard the client, we use the same systems, so I will have the same data across the bank,” says a risk executive at the unnamed bank.
On the retail side, however, there are differences in the data across countries. For example, some countries have credit bureau information, while others don’t. Thus, models will be impacted by the fact that there is less data available in a country where the credit bureau is non-existent or the quality of data is too weak or too new to be used.
The risk executive says the bank tries to have systems that are consistent across countries so when a deal is originated, it tends to have the same type of credit policies. And while the criteria for a personal loan might be different in Bangladesh, Sri Lanka and Thailand, the information a bank acquires about the borrower tends to be similar.
“We have to develop assessments based on the information that we have. If the information is different, we will have different models to try and compensate for that. For example, if I don’t have a credit bureau score, I’m going to rely more on the behavior of the client, how long they’ve been with us, and the information that I’ve been able to gather, such as level of indebtedness. In certain markets where we have a very small portfolio, we need to use proxies,” he says. “In some markets, I just don’t have enough defaults, so I can’t have a model that is built on the expectation of how much I’ll be losing or what the default probability would be. In those cases, we look at the legal structure, macroeconomic environment of different countries, and typically then use a proxy LGD model. We do this for small portfolios which are considered non-material for the operations of the bank.”
Over time, the bank will collect enough data to build a model. “Let’s say I need 40 losses and I’m at 20 today. As soon as I have 40, I will gather the data for the 40 and give them to modelers to then eliminate my need to have a proxy,” the risk executive adds.
Whether or not firms have to use benchmark data depends on the product type and geographical location, So says. Even under IRB, some banks lack the data and use data from local credit bureaus for retail portfolios, and other external data for wholesale or other low-default portfolios. Types of benchmark data relevant for IFRS 9 implementation includes external rating agency data, credit bureau data, macroeconomic data and forecast, and property price index, among others—all of which may come from different sources and have different definitions.
“They may not have the same transparency to users/modelers, and it could be difficult to assess the quality. When they are used on an ongoing basis, the change in the definition, calculation basis and quality may be difficult to be tracked for any change,” So says, adding that if banks use the same source for benchmark data, and that source is inaccurate or the relevance to their portfolios is not high, the accuracy of external data could affect the outcome of IFRS 9.

Different Degrees of Readiness

In such a diverse geographic region, banks are not uniformly prepared to meet the January 2018 deadline to start implementing ECL provisions in their books. Some countries in the region—such as Indonesia, Thailand, and India—have been given a year’s leeway by their domestic regulators, and must have their IFRS 9 systems up and running by January 2019 instead of January 2018.
“Whether the deadline is 2018 or 2019 (as in Thailand or Indonesia for example), readiness is not uniform within or across countries. Many banks that we’ve spoken with… will be ready in time, but with tactical solutions. The starting point for sophistication depends on which regulatory capital approach the bank has already implemented, and the level of advancement of any economic capital practices. Also, if a given country or bank hasn’t already implemented IAS 39, then significant measurement work is required,” says Abraham Teo, head of regulatory policy for APAC at AxiomSL.
Banks should be looking at a medium- to long-term target of having a comprehensive methodological and technology upgrade of risk and finance data management, credit modeling, and provisions calculation and auditability, Teo adds.

“This requires, on the one hand, to cater to the fundamental change that IFRS 9 will have on banks’ business models, risk appetite, portfolio strategy and commercial policy, and on the other hand, to address the necessary upgrade of banks’ IT technology,” he says. “After 15 years of regulatory pressure to calculate credit risk on a through-the-cycle basis and at a one-year time horizon, only the rare banks that have pursued improvements in their economic capital frameworks can today expect to have the point-in-time and lifetime data necessary to assess ECL. Those banks which have sufficient input data can drive ECL evolutions between reporting dates. Limited input data quality impacts the level and volatility of the ECL and P&L [profit and loss].”

For countries that will be meeting IFRS 9 requirements a year later, the Asian bank’s risk executive says he will have a central calculation that covers both group and specific country requirements. “So I’ll be able to calculate for Singapore, Malaysia, Hong Kong, China, Thailand, and Taiwan from the central point of view. For the countries implementing in 2019, these ones will continue to use IAS 39. In some cases, they have the option to migrate immediately, and in other cases, they will have to stay on IAS 39,” he says.
Although some countries have an extra year to start provisioning under IFRS 9, some private banks and privately owned banks are still going ahead with their own migration from IAS 39 to IFRS 9. “If they want to go international and they are not IFRS 9-compliant, it is as if they are not in good shape, especially if the bank is coming from an emerging country. So you have many private and private-owned banks that are still pushing for it because for them, they want to showcase that even if their country is late, they are up to date,” Picut says.
On the other hand, state-owned banks are less encouraged to comply, given the extended timeline by their local regulator.

Building Bridges

Market participants stress the importance of realizing that different business areas within banks can no longer work in silos.
“The reality… is that IFRS 9 requires input from many aspects of a firm—from IT (for data), risk (model development), finance (accounting and reporting), business (credit management, business practices, understanding of the portfolio, stage transfer requirements, customer behavior and trends), and economists and research (for forecasts). The scale of the project implies that each component will need to understand their role in each area and contribute effectively,” says KPMG’s Lee.
Being able to link risk data with finance and reporting information across the organization is one of the key requirements under IFRS 9.
“That requirement was somewhat there in the past, but now it’s going to be signed off in the financial statements from an auditing firm, rather than a regulator, so there is much more scrutiny in terms of reconciling the production of your ECL to your exposure level at the balance sheet,” says the Asian bank risk exec.
For most banks, this means both finance and risk data will have to be aggregated into a single dataset within an enterprise data mart. “It’s basically a central database that will have those finance/general ledger types of information with risk information like credit grades, country of operations, etc., linked to the client,” he says, adding that the cost of this process—just data acquisition, preparation, augmentation, and other functions—probably represents one-third of the cost of the entire project.
The other concept that IFRS 9 highlights is unbiasedness. Although it is not new, it means that banks cannot necessarily use their regulatory capital models because they tend to be conservative in nature. “So we need to adjust them, and we need to augment them for forward-looking views—for example, linking the macroeconomic variables forecasts to your forecasts of PD and LGD, and also developing a full-term structure to apply to your exposure at default models,” says the risk exec, whose firm has about 170 portfolios for which it has to create models. For example, for its Hong Kong credit card portfolio alone, the bank would need an LGD model, an EAD model, a PD model, a forward-looking PD model, and a term structure model.
It expects to review those models fairly regularly to be both point-in-time and forward-looking. For the retail models, the bank will implement an auto-recalibration calculation so that when the losses increase or decrease, the expectation of loss will also be impacted.
The bank is still in project mode, and has the equivalent of almost 200 full-time finance and risk staff working on it. Once the system is implemented, it will not need as many people, but the risk exec anticipates the bank will need an increase of between 80 and 120 staff.
“As a project, the bank will probably close this in the second quarter of 2018, but the remaining 100 or so people from the project team will continue to do improvements on things like forecasting and linking to IFRS 9 stress-testing, which will continue to be key elements to develop,” he says.
As banks move forward with their plans, it will be important to understand that impairment under IFRS 9 is a performance indicator, and thus affects their P&L. If the ECL is not calculated efficiently, it would affect a firm’s entire strategy and performance of different business lines.
Many banks focus on modeling, but they should realize there are also operational issues, says EY’s So. “How are they going to cut over the data to the new accounting system? What about banks that is in multi locations?” he says.
Regardless of whether banks are aiming to meet the January 2018 go-live date or are heading toward the 2019 deferred date (depending on their country of domicile), banks need to consider how IFRS 9 will impact their business both operationally and from a budgeting standpoint.
“The better firms think about the dress rehearsal, but this is not the case for many banks yet. Also, those that are subsidiaries should not wait around and rely on the head office for implementation processes and modelling if the head office jurisdiction has a different adoption timeline,” So says.
This article was originally published by the Waters Technology.

We use cookies in order to give you the best possible experience on our website. By continuing to use this site, you agree to our use of cookies.