Bank of America at this juncture epitomizes a never-before-seen disparity between an economy that just underwent the fastest, deepest dive ever, and consumers who keep paying their bills as if the good times never stopped rolling. Like all of the nation’s lenders, BofA is tasked with forecasting the trend, that more than any other, will determine the course for the recovery: how high the coming wave of losses on credit cards, car loans, mortgages, and the like will rise once the government’s unprecedented support in the form of cash payments, extra jobless benefits, and big subsidies to small business run out.

BofA’s second-quarter report, released on July 17, indeed forecasts a big wave of defaults to come. In the first half of 2020, America’s second largest bank by assets to JPMorgan Chase took by far the largest loan loss provisions for any six-month period since the first two quarters of 2011, when the aftermath of the financial crisis was still pummeling its credit and capital. Those additions to its reserves hammered what had been a run of stellar earnings through 2019.

But BofA’s projections are more a shot in the dark than at the onset of any other steep recession. The reason is twofold. First, the origins of this downturn are so unusual that its length and depth are far harder to forecast than for previous recessions. Second, BofA is trying to project a deterioration in credit that, five months into the crisis, hasn’t started yet––in part thanks to the never-before-seen level of government support to families and businesses, as well as forbearance on payments from the banks. So far, the consumer is defying 11.5% unemployment and the quickest, sharpest drop in GDP since the Great Depression, to keep making payments on loans just as regularly as before the crisis erupted. If you’d gone to sleep on Valentine’s Day, and just awoke to see BofA’s still terrific second-quarter numbers for bad loans, you’d never know that the economy was in free fall.

That disconnect is flummoxing CEO Brian Moynihan, who stated on the conference call, “We are seeing nothing that is consistent with an 11% unemployment rate in actual consumer payment behavior. And that has to do with the stimulus…That’s helping substantially.” Moynihan also notes that BofA is navigating “the most tumultuous period since the Great Depression,” and that the dizzying crosscurrents pose huge challenges in forecasting consumer credit.

Although it’s never been more difficult to foresee where the consumer is headed, BofA provides one of the best guides in its second-quarter report. Its view is also a reasonable road map for the overall economy, since its course depends, more than anything else, on the credit performance of America’s families, and companies in airlines, hospitality, and other industries that rely on their spending.

BofA is reckoning that the wage and salary earners will endure a long voyage on rough seas. It’s taking big provisions based on that conservative outlook. A sweeping new accounting rule should make the estimates for loan losses that BofA and other big banks are making right now the most comprehensive forecasts ever on consumer and corporate credit. The regime requires that the banks predict losses much farther into the future than under the previous rules, and take the entire hit upfront instead of spreading out the pain quarter after quarter.

So let’s examine where BofA is building reserves, and by how much, and what that tells us about how it’s handicapping the fate of the consumer.

Big provisions for losses, decent profits

Until March of this year, the banks were experiencing a golden period of profitability––chiefly because consumers were making their credit card, mortgage, and other payments more reliably than ever before, and corporate borrowers were just as steady. Then, the COVID-19 outbreak in March transformed the credit outlook overnight from sunny to dark, hammering their earnings. BofA’s profits dropped 45% in the first quarter versus the first quarter of last year, from $7.3 billion to $4 billion. In the second quarter, it suffered a steeper, year-over-year slide of 52% to $3.5 billion, slashing total earnings in half so far this year. The biggest driver by far was an explosion in provisions for bad loans. That burden rose from minuscule $1.9 billion in the first two quarters of last year to $9.9 billion.

Still, BofA managed to book $7.5 billion in profits through the June quarter, or $15 billion on an annualized basis. The Fed’s emergency rate reductions curbed its interest income to $11 billion, an 11% drop from the second quarter of 2019. But its deposits soared as customers poured cash into their checking balances in a flight to safety, and its Global Markets franchise boosted profits by 96% to $2.18 billion as traders profited from a surge in investors’ growing hunger for the relative security of bonds. In recent years, BofA’s tight expense management has helped surmount super-low rates to keep profits growing. That discipline delivered again in the second quarter, as costs barely budged from the same period a year ago.

The metric Moynihan most stresses is pretax, pre-provision income, since it measures the bank’s long-term earning power across good and bad periods for credit. Those are also the profits available for bolstering reserves in tough times like today. The bigger the pretax, pre-provision earnings, the greater the strength for weathering the COVID-19 hurricane. For the first half, BofA booked $18.2 billion in that category, only 7% below the $19.6 billion recorded in the same period last year. In other words, excluding the big new provisions, BofA’s underlying profits remained almost as strong in the pandemic as in the flush times that extended through early spring.

A new rule makes BofA front-load the hit from future defaults

Those new provisions are the result of a rule change the Financial Accounting Standards Board quietly implemented on Jan. 1, 2020. Previously, when it came to credit card loans, the rules didn’t require estimating losses on current but endangered loans (i.e., customers regularly paying interest but likely to default) beyond a year ahead.

The current expected credit losses (CECL) standard transformed that methodology by mandating that banks forecast potential losses for the entire life of the loan. “Now they’re provisioning not just for the loans that have gone delinquent, but the ones that are current and may go bad at any time in the future,” says David Fanger, a senior VP at Moody’s Investors Service. For example, if a card loan is now current but BofA projects a high probability that it will default in 18 months, BofA is required to book a loss now, whereas under the “incurred” model, it would have booked the provision a few quarters hence, when the borrower actually stops paying interest. The idea is to front-load all the provisions based on the best estimate of all losses to come for current and delinquent loans alike.

As Fortune previously reported, under the old rule based on the accrual model, provisions tended to build gradually as a deepening recession caused more and more defaults. “But the new regime means that banks are taking much larger provisions based on their best guess at the moment of what total losses will be on still current loans,” says Bain Rumohr, senior director of North American banks at Fitch Ratings. CECL’s arrival coincided with the COVID-19 crisis, and the combination is forcing banks to take the entire hit right now for all estimated future losses on all their loans caused by the crisis.

That’s a tall order for two reasons. First, the pandemic started as a health and not an economic crisis so the shape of the recovery is far harder to predict than the comeback from a normal recession. Second, it’s extremely difficult to predict how much of the consumer’s excellent performance so far stems from the big waves of government support, and how much that gold star record will deteriorate once the federal dollars stop flowing. “What the banks are attempting to do is figure out what level of loss comes through when the programs phase out,” says Rumohr.

It’s remarkable that BofA’s portfolio as of the second quarter is looking almost as strong as during the flush times that lasted until February. In part, that’s because it’s been offering broad forbearance to its customers. Since March 16, it has extended 1.8 million payment deferrals, mostly on credit card loans, to its customers. In a good sign, requests for deferrals pretty much ceased by late June. BofA reports that 60% of credit card holders who’ve gotten forbearance have made at least one payment since. But since the cardholders with deferrals owe $7.6 billion, that still leaves a large number of customers, and large balances, at risk.

“There’s a notable disconnect between what you see so far in their delinquency and charge-off rates, and what you would typically see with high unemployment,” says Rumohr. In its consumer lending book, which includes $84 billion in credit card loans, its nonperforming credit has actually dropped from $3 billion to $2.2 billion in the past year, lowering the share from 0.67% to 0.49%. In commercial lending, the nonperformers total $2.2 billion, or just 0.41% of the portfolio.

It’s the same story with charge-offs, loans that have previously been expensed but are recognized as uncollectible and removed from loan loss reserves. That number for the entire bank was just $1.1 billion in the second quarter, a mere $200 million more than in the second quarter of 2019 when the credit outlook was terrific. “Looking at those delinquencies and charge-off levels, you wouldn’t think there’s a crisis going on,” says Fanger.

Looking ahead

It’s hard to imagine a starker contrast than the divide between customers’ current performance, and how BofA sees the future. Since the close of 2019, it has doubled allowances for loan losses from $9.4 billion to $19.4 billion. The biggest increase came in cards, where reserves mushroomed two-and-a-half times from $3.7 billion to $9.25 billion, rising from 3.8% of the portfolio to 11%. The second biggest jump came in commercial real estate. “That’s another area where the virus is causing a lot of uncertainty because it’s damaging the retail and hospitality market,” says Fanger. Real estate reserves more than doubled to $2.2 billion. That category and cards account for two-thirds of the total increase in allowances for credit losses.

The question is whether BofA is making assumptions that truly reflect the extent, and probable duration, of the downturn. Moynihan shared the assumptions on the economy that the bank is employing to estimate loan losses at two weeks ago. He’s assuming that unemployment ends the year at 10%, just a one-point improvement from the current rate, and joblessness remains high at 9% through the first half of 2021, easing to 7.5% by year-end.

It’s instructive to compare that outlook with the “most adverse” scenario that the Fed used in the latest stress test, completed in June. That test is based on conditions that prevailed before the outbreak, but is consistent with a V-shaped recovery from the crisis. The Fed’s recession scenario posits that the unemployment rate rises to 10% over six quarters, then drifts back to 8% over the next 18 months, meaning that joblessness remains elevated well beyond mid-2023. Under that scenario, BofA would have sufficient reserves today to cover 45% of future losses. The Fed deems that level BofA is holding sufficient capital to cover the projected charge-offs and still maintain a strong capital buffer.

BofA is betting that although the COVID-19 downturn is more sudden and severe than the Fed’s pre-pandemic, worst-case test, it will cause less long-term damage because it will pass much faster. In effect, BofA is saying that it’s already taken all the reserves it will need to cover damage from the pandemic, and that the hit will be about half as severe as the Fed’s test case where joblessness remains stubbornly high three years hence.

Keep in mind that CECL requires BofA to front-load all the loan losses it anticipates going forward. If BofA’s forecasts are correct, that’s great news for the bank and the U.S. economy. But the pandemic crisis is the ultimate in moving targets. “So if the COVID crisis is extremely long-lasting, it’s possible that the reserves won’t be big enough, and BofA will be forced to keep increasing provisions,” says Fanger. Whether BofA indeed has big enough allowances to get through the crisis will also depend on how many more consumers default once the banks end forbearance, and the government stimulus ceases.

It’s encouraging, however, that BofA has increased reserves dramatically, in theory to the point where current allowances will cover all future losses, and still booked over $7 billion in profits in the past two quarters. It has also suspended all share buybacks, so it’s replenishing capital at the same time that it’s adding to reserves. BofA also deserves credit for keeping its credit card portfolio at much more modest levels than its rivals, potentially limiting the extra reserves it will need to take if the recovery is slower than the already lengthy comeback it predicts. Moynihan proudly noted that BofA is now carrying half the loans to cardholders it did at the start of the Great Recession.

BofA is heading into this recession a lot stronger than the last time and, in part because of CECL, is being what looks like extra prudent in bracing for future losses. Here’s the problem: We’re in dangerous, uncharted waters, where the dragons that no model can predict may be lurking.