Forecasting might not be top of mind as you prepare for CECL. Understanding the standard, deciding whether to handle the transition internally or engage third party assistance, gathering data, pooling, choosing a methodology – those issues drive our concerns well before we encounter a “reasonable and supportable” forecast.
The forecasting piece is, in large part, what makes CECL different from incurred loss estimating. The Great Recession revealed the insufficiency of being able to reserve only for a probable or already incurred loss event; those years proved the incurred loss model “too little, too late.” The FASB determined we needed a forward-looking component to be prepared for an economic downturn when it comes – and not only a Great Recession, but a local economic jolt, like the closing of a factory that employs a significant portion of the local population. CECL allows us to do that, even to the extent of reacting to a rumor, be it from a credible source.
Ultimately we are forecasting why the future is different from the past, be it better or worse. We do that using economic factors. External factors can be broad in scope, such as political turmoil or a trade war. Or they can be local, such as a weather event or the opening of a new business that will increase employment in your area. And they can reflect internal factors, like losing a quality loan officer to another institution, or signing new loan talent, or the closing of a competing institution.
Here are a few things to consider when developing the forecasting piece of your CECL process:
- Initially, identify the variables that will determine expected losses. What specific factors apply to your population, your institution? Hone in on the most important factors; don’t complicate matters with too many variables. Remember, sometimes less is more. Accordingly, I challenge you to find any group of factors collectively more predictable and closely correlated with losses than simply looking to unemployment rate alone. Can it be done? Sure. But, for decades, unemployment rate has demonstrated a strong correlation to loan performance and, in and of itself, represents a collective representation of many other economic factors we might otherwise also consider. To that end, beware of double counting the impact of related variables. We often advise our MST clients to set their standards high, to error on the side of omitting what might be relevant, as opposed to adding a factor that could prove not to be.
- Analyze the factors you have identified as meaningful to confirm correlations. Determine their impact by understanding the impact they have had on losses in the past. Determining correlations between historical losses and external variables can help determine the directionality of a factor, as well as exclude factors that might not have impact. The MST Virtual Economist compares economic variables against your loan performance in a matter of seconds and will indicate the level of correlation of that variable.
- Remember that correlations do not indicate causation. For example, the stock market has climbed more often in years after the National League has won the World Series, but it wouldn’t be wise to base investment decisions for a coming year based on which league prevails. Beware of spurious correlations.
Where you determine causation, seek out the data you’ll need to support your forecast. You can do that yourself or look to external sources for forecasting. For example, for unemployment rates or GDP, consult the reports of the Federal Open Market Committee or an economic agency in your state or a local university. There’s a lot of free data available, as well as more expensive services. As to your internal factors, of course, that information comes from your own experience – product lines or changes in the lending team, for example.
- Determine the level of impact, how these factors will influence your portfolio. This is more art than science. How do you quantify the impact? You might leverage experience from the past based on your loss history. No losses? Leverage peer data.
- Incorporate the impact of your factors into the model itself. How you forecast will be based on the methodology you decide to run. You might overlay it with your qualitative factors if you run a simple cohort model, for example, or it might be ingrained or embedded in a more sophisticated model, such as within a PD/LGD transition matrix.
One common and justified concern we often encounter throughout our wide-spread advisory engagements is how far in advance institutions should forecast. Your forecast is limited to the contractual life of the assets and generally further reduced to how far into the future these forecasts can be reasonably supported. Most institutions are using a one- to two-year period and reverting to history beyond that forecast able period.
Remember few “experts” predicted the Great Recession – at least the magnitude to which it was experienced. Most of the economic gurus of the day missed it. So can you be any better at predicting the future? Your regulators won’t expect you to have a crystal ball. It’s always better to acknowledge that you don’t know instead of acting like you do know then being proven that you didn’t.