IFRS 9 and CECL were designed with two outcomes in mind: to ensure sufficient reserves on the eve of a recession and to prevent restricted lending from curtailing a nascent recovery.
After the global financial crisis of the late 2000s, it became clear that allowances under the incurred loss framework were too reactive. Banks entered the recession with reserves that were too lean given the tsunami of losses that was approaching. Allowances then ballooned as the recession worsened, causing banks to restrict their lending behavior just when a steady flow of credit was needed the most.
From a public policy perspective, the exact opposite pattern is the desirable one. The ideal situation would be for allowances to peak on the first day of a new recession.
Debate about the procyclicality of allowances under CECL and IFRS 9 is well established. There are severe doubts about whether the new model-based approaches will actually fix the problem. Here, I want to take a step back and see if it’s possible to design a loan loss reserve system that could only act in a counter-cyclical manner.
In the economy, there are a number of features described as “automatic stabilizers,” including the combined tax and welfare system. As the economy lurches into recession, business profitability declines and unemployment rates increase. Tax collection from payrolls and corporations then decrease, and government spending increases, as more people join the welfare rolls. The shift toward a larger budget deficit then “leans against” the cycle, cushioning the blow of the initial shock and acting to lessen the depth of the recession.
The fact that long-term interest rates rise in a boom and fall in a recession is another prime example of an automatic stabilizer. The key point is that neither legislation nor executive action is required for these forces to operate. Similarly, they do not rely on the vagaries of a model’s predictions. They will simply occur as the economic cycle ebbs and flows, without any intervention in the process whatsoever.
Booms and Busts
Thinking about booms and busts in banking, several obvious regularities emerge. In a broad-based boom, strong growth in lending will be widespread across the banking book. Credit cards will be well utilized, and we will see consumers taking on bigger mortgages and more frequent auto loans. Moreover, to sate the high level of retail demand, business lending will also typically be elevated, as investments are made to boost the production of goods and services.
These trends are then reversed in recession, often alarmingly so.
If proportional reserves were calculated as a function of the current industry loan growth rate – higher in booms and lower in recessions – reserves would then act like an automatic stabilizer. What this means is that at times when the industry is booming, the proportion of the outstanding loan balance reserved for potential losses would have to rise. This would happen irrespective of whether economists viewed the ongoing expansion as sustainable or not.
When calculated as a function of the current industry growth rate, reserves would act like a tax on lending, reducing the amplitude of cycles that are rather prevalent in the banking universe.
Sometimes, by way of contrast, booms are product specific. The U.S. auto finance sector, for example, witnessed double digit growth in loans and leases from 2010 through early 2016, before slowing to more mundane levels in recent quarters. Since then, the proportion of loans in default has risen by around 30%, even though the economy has continued to grow strongly in an environment of low unemployment (see Chart 1, below).
While proportional auto loan allowances based on macro forecasts (like those seen in CECL and IFRS 9) might not have moved during the boom, those based on industry growth rates would have increased rapidly. Now that growth has fallen below the long-term rate, proportional reserves would currently be much lower, easing the impact of the slowdown in lending on the industry.
Modeling Options: Pros and Cons
The next question is whether it is possible to incorporate this way of thinking into a framework that is more consistent with the tenets of IFRS 9 and CECL. To explore this, we investigate the upturn in U.S. commercial and industrial (C&I) loan losses that began in early 2015.
We begin by using a Hodrick-Prescott filter on C&I loans outstanding, extracting the cyclical component. Our first model (#1) uses only lags of this component to predict the rate of nonperforming loans. Our second model (#2) augments this specification with a range of standard macroeconomic variables, including oil prices and various interest rates. Our third model (#3) excludes the lagged C&I loan growth component and considers only macroeconomic data.
We assume perfect foresight of the economy and make out-of-sample predictions from the start of 2015 to Q2 2018. The results are depicted, below, in Chart 2.
The model based only on past lending growth (Model #1) is too pessimistic (implying that associated reserves would be too high), while the model based only on macro data (Model #3) is too optimistic. Nevertheless, both models do predict an increase in nonperforming loans through the relevant period. The best performing model (Model #2), by a margin, is that which uses both past lending growth and macroeconomic data.
The simple conclusion from this is that past lending growth could be a useful predictor of future losses, especially if combined with other metrics.
The first model (containing only lagged lending growth) also has a lot of advantages. Since it only relies on forecasts of one variable, it is the model that is least reliant on economists making accurate prognostications.
The other models require more treacherous economic forecasts to be made, including the prediction of a precipitous fall in oil prices (which, admittedly, began in 2014). It is this event that is normally collared as the culprit in the rise of nonperforming C&I loans.
The most accurate model (Model #2) also suffers from the criticism that forecasts of many input variables are required. Even so, when we flatline all the economic data at Q4 2014 levels, we find only a slight deterioration in the performance of the model. It turns out that the lagged lending growth variables are still doing most of the heavy lifting.
It may be a bridge too far to suggest that banks should base loan loss allowances on current lending growth rates alone. However, it does seem like this mechanism would work to make reserves truly countercyclical. As a compromise measure, we suggest that CECL and IFRS 9 analysts consider using past industry lending growth as a key driver in their models.