Implied loss rates in the stratosphere compared to other models and historical experience
Now the emerging consensus says that the government’s stress tests weren’t stressful enough. Elizabeth Warren says so. So does Goldman Sachs. Nouriel Roubini, as well.Â
Hogwash. The odds are vanishingly small that any of the tested banks will actually report losses in the coming seven quarter even close to the losses that the stress case implies. Forecasters who’ve taken to using the stress case as their base case for the banks (Goldman, this means you) are meaningfully underestimating banks’ profitability for the next two years.
The stress-test skeptics (try saying that ten times fast) make several analytical mistakes. First, they hang their entire case on a single metric: the unemployment rate. It’s already hit 9.4%, they note, which is ahead of the government’s 8.9% stress assumption. That means the 10.3% stress assumption for 2010 is too low as well.Â
Maybe, maybe not. Yes, unemployment has shot up faster than the government assumed, but signs of economic recovery are nonetheless arriving thick and fast. (The latest: unemployment claims came in better than expected yet again this morning, while retail sales were stronger than expected, as well, and business inventories lower.) At the rate things are going, that 10.3% stress bogey for joblessness next year looks less likely now than it did when the government came up with it.
But more to the point, unemployment isn’t the accurate indicator of loan losses—particularly commercial loan losses—that everyone seems to think. To rely on that on that one number as a reason to doubt the entire stress test is a huge, huge stretch.Â
Secondly, the stress test itself, meanwhile, was a ham-handed effort that didn’t come close to adequately identifying (and accounting for) which banks have already been aggressive in recognizing and charging off their problem loans. It essentially assumed they’ve all been dragging their heels. Which they have not.
Third, the test ignored the way the chargeoff cycle works. In particular, it didn’t recognize that as the cycle turns down, the weakest credits go nonperforming earliest and produce the greatest losses. Rather, the test assumes early-cycle loss rates will persist through the entire cycle. They won’t. Result: the stress test’s loss rate estimates are too high.Â
Fourth, the stress test loss assumptions for the 19 big banks can’t be extrapolated to the industry as a whole, as some analysts have done, as a way to calculate industrywide loss estimates. Why? Because the loan portfolios of the country’s other 8,000 banks, in aggregate, aren’t comparable to the big banks’. Their loan books are much less risky. For example, the smaller banks’ commercial real estate exposure contains a higher percentage of owner-occupied CRE loans than the big banks’. Loss rates on owner-occupied tend to run closer to those of commercial loans rather than non-owner occupied CRE.
Finally, if you try to model the numbers quarter by quarter, the ramp-up of chargeoffs the skeptics seem to expect over the next seven quarters simply doesn’t pass the smell test.Â
I could go on, but let me stop there. Against all this, the skeptics point to a bloated unemployment rate (and ignore signs of economic recovery). But if you go through the numbers you’ll see that the stress case spits out loss numbers that are much higher than other models’, or what history would imply. To see how wide the difference can be, let’s look at one of the stress-tested banks in particular, Fifth Third. The chart below shows Fifth Third’s quarterly net chargeoff rate over the past nine quarters, along with three different projections of chargeoffs over the next seven quarters.
The lowest line is our own base case, the middle line is our own stress case, while highest line is the quarterly level of losses implied by the Fed’s stress test. (The Fed did not forecast quarterly losses; we simply smoothed the estimated total losses over the next seven quarters.)Â
(The spike in reported chargeoffs in the fourth quarter of 2008, by the way, is the result of an aggressive yearend writedown of the company’s most troubled credits. The bulk of the chargeoffs related to loans in the company’s most distressed geographies, Florida and Michigan. Loans in those two states made up only 28% of the company’s total loans, yet contributed 66% of fourth quarter NCOs. So Fifth Third hasn’t exactly been sitting on its hands.)
Anyway, as you can see, our base case loss forecast has the future quarterly losses peaking at just over 3% of loans, annualized, or 6 times “normalized” level. Our stress level has losses peaking at 5%, while the Fed’s stress test puts losses at 6% (or 12 times the company’s normalized loss rate.)Â
Anything’s possible, of course, but given our knowledge of Fifth Third’s credit management practices (like, say, that fourth-quarter cleanup) and its remaining portfolio, actual losses will likely come in somewhere between our base case and our Stress case. The Fed’s stress case is apt to be way off—as it was designed to be.
It goes without saying that this makes a huge difference in evaluating Fifth Third as an investment. The estimate of future losses will determine estimates of future book value, as well as the timing and level of the company’s future normalized earnings. We believe analysts who use the Fed’s stress-case loss forecast as their base case are seriously underestimating Fifth Third’s (and most others banks’) earnings power in the next several quarters.Â
Now let’s take a look at the total industry. Some analysts have taken the average cumulative loss forecast for the 19 tested banks and applied it to the entire industry. I’ve already said why I think that’s an analytical mistake. But let’s look anyway at what would happen to quarterly losses under this assumption.
Under the stress test, the average 2009-2010 cumulative loss assumption for the 19 banks comes to 9% of total loans. The next chart shows the industry’s quarterly net chargeoff level for the last nine quarters, as well as two projections, S&P’s and the Fed’s stress case. (I smoothed the Fed’s annual numbers into quarterly estimates.) S&P’s loss forecast, revised last week has cumulative losses in these two years of 6% versus the Fed’s stress test at 9%.
It is hard for me to believe that either loss projection will actually occur. Under the S&P forecast, the industry’s quarterly level of net chargeoffs (in dollars) would have to rise by a factor of 2.1 times from the most recent quarter. Under the stress test, losses would have to rise by a factor of 3.3 times. That would be more than 10 times our estimate of normalized losses. It’s possible, I suppose, but highly unlikely--particularly given the economy’s signs of stabilization over the last two months.
My points are simply that the estimated level of losses projected by the Fed’s stress test will likely prove to be significantly too high, even if one of the assumptions the Fed used (peak unemployment rate) proves to be too low. As a result, bank book values will hold up better than expected in the next seven quarters, earnings will be better than expected, normalized earnings will resume sooner than expected, and the level of “normalized” earnings will be higher than expected. Analysts who cling to the stress case as a base case are making a serious mistake, and will be causght flat-footed once bank fundamentals turn in earnest.
What do you think? Let me know!
Thomas Brown
www.bankstocks.com
Now the emerging consensus says that the government’s stress tests weren’t stressful enough. Elizabeth Warren says so. So does Goldman Sachs. Nouriel Roubini, as well.Â
Hogwash. The odds are vanishingly small that any of the tested banks will actually report losses in the coming seven quarter even close to the losses that the stress case implies. Forecasters who’ve taken to using the stress case as their base case for the banks (Goldman, this means you) are meaningfully underestimating banks’ profitability for the next two years.
The stress-test skeptics (try saying that ten times fast) make several analytical mistakes. First, they hang their entire case on a single metric: the unemployment rate. It’s already hit 9.4%, they note, which is ahead of the government’s 8.9% stress assumption. That means the 10.3% stress assumption for 2010 is too low as well.Â
Maybe, maybe not. Yes, unemployment has shot up faster than the government assumed, but signs of economic recovery are nonetheless arriving thick and fast. (The latest: unemployment claims came in better than expected yet again this morning, while retail sales were stronger than expected, as well, and business inventories lower.) At the rate things are going, that 10.3% stress bogey for joblessness next year looks less likely now than it did when the government came up with it.
But more to the point, unemployment isn’t the accurate indicator of loan losses—particularly commercial loan losses—that everyone seems to think. To rely on that on that one number as a reason to doubt the entire stress test is a huge, huge stretch.Â
Secondly, the stress test itself, meanwhile, was a ham-handed effort that didn’t come close to adequately identifying (and accounting for) which banks have already been aggressive in recognizing and charging off their problem loans. It essentially assumed they’ve all been dragging their heels. Which they have not.
Third, the test ignored the way the chargeoff cycle works. In particular, it didn’t recognize that as the cycle turns down, the weakest credits go nonperforming earliest and produce the greatest losses. Rather, the test assumes early-cycle loss rates will persist through the entire cycle. They won’t. Result: the stress test’s loss rate estimates are too high.Â
Fourth, the stress test loss assumptions for the 19 big banks can’t be extrapolated to the industry as a whole, as some analysts have done, as a way to calculate industrywide loss estimates. Why? Because the loan portfolios of the country’s other 8,000 banks, in aggregate, aren’t comparable to the big banks’. Their loan books are much less risky. For example, the smaller banks’ commercial real estate exposure contains a higher percentage of owner-occupied CRE loans than the big banks’. Loss rates on owner-occupied tend to run closer to those of commercial loans rather than non-owner occupied CRE.
Finally, if you try to model the numbers quarter by quarter, the ramp-up of chargeoffs the skeptics seem to expect over the next seven quarters simply doesn’t pass the smell test.Â
I could go on, but let me stop there. Against all this, the skeptics point to a bloated unemployment rate (and ignore signs of economic recovery). But if you go through the numbers you’ll see that the stress case spits out loss numbers that are much higher than other models’, or what history would imply. To see how wide the difference can be, let’s look at one of the stress-tested banks in particular, Fifth Third. The chart below shows Fifth Third’s quarterly net chargeoff rate over the past nine quarters, along with three different projections of chargeoffs over the next seven quarters.
The lowest line is our own base case, the middle line is our own stress case, while highest line is the quarterly level of losses implied by the Fed’s stress test. (The Fed did not forecast quarterly losses; we simply smoothed the estimated total losses over the next seven quarters.)Â
(The spike in reported chargeoffs in the fourth quarter of 2008, by the way, is the result of an aggressive yearend writedown of the company’s most troubled credits. The bulk of the chargeoffs related to loans in the company’s most distressed geographies, Florida and Michigan. Loans in those two states made up only 28% of the company’s total loans, yet contributed 66% of fourth quarter NCOs. So Fifth Third hasn’t exactly been sitting on its hands.)
Anyway, as you can see, our base case loss forecast has the future quarterly losses peaking at just over 3% of loans, annualized, or 6 times “normalized” level. Our stress level has losses peaking at 5%, while the Fed’s stress test puts losses at 6% (or 12 times the company’s normalized loss rate.)Â
Anything’s possible, of course, but given our knowledge of Fifth Third’s credit management practices (like, say, that fourth-quarter cleanup) and its remaining portfolio, actual losses will likely come in somewhere between our base case and our Stress case. The Fed’s stress case is apt to be way off—as it was designed to be.
It goes without saying that this makes a huge difference in evaluating Fifth Third as an investment. The estimate of future losses will determine estimates of future book value, as well as the timing and level of the company’s future normalized earnings. We believe analysts who use the Fed’s stress-case loss forecast as their base case are seriously underestimating Fifth Third’s (and most others banks’) earnings power in the next several quarters.Â
Now let’s take a look at the total industry. Some analysts have taken the average cumulative loss forecast for the 19 tested banks and applied it to the entire industry. I’ve already said why I think that’s an analytical mistake. But let’s look anyway at what would happen to quarterly losses under this assumption.
Under the stress test, the average 2009-2010 cumulative loss assumption for the 19 banks comes to 9% of total loans. The next chart shows the industry’s quarterly net chargeoff level for the last nine quarters, as well as two projections, S&P’s and the Fed’s stress case. (I smoothed the Fed’s annual numbers into quarterly estimates.) S&P’s loss forecast, revised last week has cumulative losses in these two years of 6% versus the Fed’s stress test at 9%.
It is hard for me to believe that either loss projection will actually occur. Under the S&P forecast, the industry’s quarterly level of net chargeoffs (in dollars) would have to rise by a factor of 2.1 times from the most recent quarter. Under the stress test, losses would have to rise by a factor of 3.3 times. That would be more than 10 times our estimate of normalized losses. It’s possible, I suppose, but highly unlikely--particularly given the economy’s signs of stabilization over the last two months.
My points are simply that the estimated level of losses projected by the Fed’s stress test will likely prove to be significantly too high, even if one of the assumptions the Fed used (peak unemployment rate) proves to be too low. As a result, bank book values will hold up better than expected in the next seven quarters, earnings will be better than expected, normalized earnings will resume sooner than expected, and the level of “normalized” earnings will be higher than expected. Analysts who cling to the stress case as a base case are making a serious mistake, and will be causght flat-footed once bank fundamentals turn in earnest.
What do you think? Let me know!
Thomas Brown
www.bankstocks.com
