Are Banks Ready for CECL?

The roadmap to a successful CECL banking program using DataRobot

We are currently in the midst of one of the most significant regulatory changes in the history of the global financial system. New accounting regulations and standards will soon fundamentally change how financial institutions must estimate expected loss.

These new changes will impact the entire financial system. The new regulations apply to all banks, savings associations, credit unions, and financial institution holding companies (regardless of size) that currently file regulatory financial reports conforming to GAAP standards. For simplification, we’ve used “bank” to indicate all of the affected financial institutions by this change.

In this blog series I will review how banks can turn their current expected loss processes reliant on backward-looking credit indicators into more accurate forecasts by using forward-looking methods driven by predictive models created with machine learning.

I'll provide a path to show how investments in strategic technology such as automated machine learning will spur rapid regulatory compliance and promote broader improvement to the business by both reducing operational costs and driving bottom-line growth.

 

What is expected loss? And why do we need to forecast it?

There are many fundamental things that banks must manage, such as balancing loan-receivables and the demand for deposits. Adding to the complexity is that not all loans that banks provide will be paid back in full. Or, some loans may become delinquent or even default entirely, which, if left unchecked, would create a loss for the bank on its expected income.

Therefore, to ensure sufficient mitigation of such risks, regulators require that banks prepare for these potential losses by provisioning a portion of expected income to cover potential losses. In the event of a loan loss, the bank can then use the allowance of provisions set aside to cover the loss.

 

This level of criticality for expected loss provisioning has rightfully drawn the attention of regulators around the world.

The relationship between loan loss provisioning and capital is significant and it requires careful consideration to ensure it is set appropriately. If loss provisions ultimately prove too high, then the bank will unnecessarily hold reserves for more than what is needed to cover its losses, preventing the bank from releasing that capital elsewhere. On the other hand, if reserves prove inadequate to account for higher than expected loan losses, then the bank’s financials will suffer once these losses are realized. Under such circumstances, it is safe to say that shareholders and senior management will not be happy with the subsequent increase in capital costs incurred by the firm.

This level of criticality for expected loss provisioning has rightfully drawn the attention of regulators around the world.

 

Financial Accounting (bear with me -- this is important!)

The current Generally Accepted Accounting Principles (GAAP) requires an “incurred loss” methodology for recognizing credit losses based on historic charge-offs. Simply put, this method delays the recognition of a loss until it’s probable that the loss has been incurred. The subjectivity around what deems a threshold to be “probable” directly affects when a credit loss can be recognized. This subjectivity ultimately led to severe criticism of the incurred loss method because it restricts the ability to record credit losses that may be expected, but do not yet meet the “probable” threshold.

The recent global financial crisis of 2008 exploited this concern. In the lead-up to the financial crisis, the stock market was reacting to estimates of expected credit losses based on forward-looking information and devaluing banks before accounting losses could even be recognized! This highlighted a critical weakness in the accounting standards that regulators needed to address to prevent a similar global banking crisis from ever occurring again.

Fast forward and we are now in the midst of a revolutionary and unprecedented change in international accounting standards and regulation. The Financial Accounting Standards Board (FASB) and the International Accounting Standards Board (IASB) have recently overhauled the accounting requirements for loss provisioning and credit impairment. The new standards are based on an “expected loss” method. Unlike the incurred loss method that is based on backward-looking loss rates, the expected loss method applies when the loss has not yet occurred, but its occurrence is probable. In other words, the loss of future-flow is expected with some probability.

 

The Opportunity

The new expected loss standards require that banks use information about past events (i.e., historical data) and "reasonable and supportable” forecasts when estimating expected credit losses. Although this is a huge change to the current incurred loss standards, it also provides a unique opportunity because the new standards do not prescribe how lenders choose to make the estimate, but only that the forecasts must be “reasonable and supportable.” This gives banks the flexibility to implement the best models and methodologies to forecast expected loss for their portfolio, as long as the forecasts can be proven to be reasonable and supportable.

This means that accurate and transparent models for predicting expected loss forecasts should be at the core of successful compliance programs. But, with such ambiguity around the process, how can banks reasonably test hundreds of different modeling methodologies to guarantee that they implement the best one? The answer is closer than you may think.

Different modeling methods are used on different types of loans or assets — and different models are sometimes even combined to use on one asset type — to estimate expected credit losses. However, the varying types of modeling methods also vary in their methodological complexity. Below are some examples of common models that we see across the industry.

At the most simplistic end of the modeling spectrum, we have the Discounted Cash Flow (DCF) analysis. On the most complex end of the modeling spectrum, we have the Probability of Default (PD) method. And, in the middle of the road, we have methods such as Average Charge-Off, Vintage Analysis, Static Pool Analysis, and Migration methods such as Roll Rate Analysis.

In our next blog in this series, we will dive into each of these modeling methods in more detail, and focus on how to automate your CECL modeling process. This will include not only the automation of the development of highly accurate machine learning models to predict expected loss, but also the automation of the documentation and deployment of the chosen models. The end result is a reduction of model risk, while also greatly improving operational efficiency and model accuracy.

The bottom line is that your expected loss forecasts will be highly accurate, reasonable, and easily supportable to auditors and regulators, quickly aligning your CECL compliance process to regulatory expectation.

More on this topic: 

New call-to-action

 

About the Author

As the head of Model Risk Management at DataRobot, Seph Mard is responsible for model risk management, model validation, and model governance product management and strategy, as well as services. Seph is leading the initiative to bring AI-driven solutions into the model risk management industry by leveraging DataRobot’s superior automated machine learning technology and product offering.