The imposition of CECL reporting, beginning in 2020, presents the US banking industry with unprecedented analytical and administrative challenges. It is no surprise that banks admit to ever-increasing expenses as they prepare. Banks also face the prospect of even more expense later – unnecessary expense – if the analytical systems they adopt turn out to be excessively punitive in ALLL translation. Recent turmoil in the marketplace suggests that one prominent class of analytics – the so-called “Merton models” – may threaten present and prospective users with just such expense. Merton models apply market signals such as share prices, stock option volatilities, CDS spreads and bond prices, according to proprietary formulas, in order to estimate the probability of default (PD) of the company in focus. Major credit rating agencies present such models as their CECL solutions on public obligors.
- CECL’s life-of-loan estimated credit loss (ECL) makes the PD exercise more challenging than ever before.
- Mistakes in determining PD over term become far more expensive than ever before.
- CECL’s scale and cadence can be properly served in only one way.
- Volatility’s return to the marketplace, along with the evolution in its messaging, calls for a newfound skepticism towards relevant Merton models.
CECL Versus CCAR & DFAST – How do the regulatory goals diverge?
Regulators and auditors have cautioned banks to be careful about planning to deliver exactly what CECL calls for: Expected Loss that is “reasonable and supportable.” Bankers who have been otherwise consumed in recent years by the stress-based orientations of DFAST and CCAR must be on guard against reflexively applying stresses in their CECL exercises beyond what they can actually foresee over their specific loan tenors. Even when bankers revert to historical statistics for loan tenors beyond their own predictive confidence intervals, as CECL will require, those statistics, if undifferentiated, may underserve. Rather, bankers should take care to select statistics to use on new originations that are reflective of their latest underwriting standards, when those standards are higher than in the past.
Quantitative Probabilities of Default – Why will they serve CECL so well?
Quantitative term probabilities of default (Term PDs) that are properly grounded – that is, “reasonable and supportable” – offer the prospect of relief from, or at least authoritative challenge to, incumbent methodologies that may hit ALLL too hard on the C&I front. There are several systems on offer in the C&I space. The RapidRatings approach involves an in-depth examination of the obligor’s financial statements according to any one of 22 global industry sets, an approach which we find more predictive than geographically-based orientations and which permits identical application to public and private companies of whatever size and location.
Volatility – What kind of signal has it now become for corporate health?
The case for fundamentals versus market signals in quantitative analysis would seem to have strengthened appreciably this month, as volatility exploded overnight and as a 10% correction took hold in just a few days. Have default prospects across Corporate America really surged as acutely and as instantaneously as the Merton models now tell us? Is present-day volatility even the same signal that it was when the Merton models first appeared in the late 1980s, before volatility became a tradable asset class and well before the stampede of CTAs, variable annuitants, risk parity funds and others into the space?
CECL’s Mandate – How can estimated losses be kept “reasonable and supportable?”
RapidRatings believes that quantitative analytics are essential to surmounting CECL. They alone fully enable automation. They alone present a solution to the scale and frequency of the CECL representations that publicly held BHCs will soon have to make to the world. Where fundamentals govern the quantitative exercise, as they do at RR, runaway volatility in the marketplace is kept from becoming runaway volatility in the ALLL.