To paraphrase investment legend Jeremy Grantham, bull markets get craziest in their final leg. The surge we're now witnessing from already superexpensive heights signals that the end may be near.

The S&P 500 looks headed for another gangbusters performance for the week ending Feb. 5. By the close on Friday, the index had risen 4.7% to notch still another record at 3887. Wall Street banks almost universally predict the good times to keep rolling; Goldman Sachs foresees another 10.6% jump to 4300 by year-end. Obviously, it's tempting to jump on a train whose momentum not only refuses to flag, but keeps building.

Investors, however, should consider how much they have to lose if, by the standard metrics, stock prices simply shift from what appear to be excessively pricey levels to just moderately pricey. It's a big, scary number.

Let's look at four versions of the price/earnings multiple used to measure whether stocks are a bargain or overpriced. They're all measures of what investors pay for each dollar in earnings. What varies is how best to measure earnings.

Current P/E

We'll start with the S&P's current P/E. Right now, it's 40.3 based on $96.45 in GAAP net earnings per share over the past four quarters ended in December; I'm including the analysts’ estimates for Q4, since only half the members had reported when S&P posted its most recent data. That multiple is huge, but it's not terribly relevant. It's artificially inflated by the midyear collapse in profits driven by the COVID crisis.

P/E before COVID

So what's a better measure? I'll suggest three. The first is the P/E based on what the big-caps were earning before the cataclysm. You could argue that those are "normalized" earnings. Since the Congressional Budget Office now predicts that national income will regain 2019 levels in mid-2021, it's not daft to posit that profits will return to where they left off before the deluge. At the close of 2019, S&P EPS stood at an all-time high of $139.47. If profit indeed rescales those peaks sometime this year, a reliable P/E would be 28 (the S&P's current "price" of 3887 divided by earnings of $139.47).

Getting back to almost $140 per share would be a steep, hard climb. Even so, paying $28 for each dollar in profits is really, really expensive. A 28 P/E is 15% above the highest reading since the 2009 crash that decimated profits (and inflated P/Es, as we're witnessing today). That was the peak of $24.34 posted in September of 2016.

CAPE

But is $139.47 per share really the right benchmark for profits? Not according to our third yardstick, the famous cyclically adjusted price/earnings multiple, or CAPE, developed by the great Yale economist Robert Shiller. Earnings follow an erratic course. In years where they're flying high, they make P/Es look artificially low, and when they're temporarily depressed, the S&P looks like a bargain, but really isn't. To correct for those swings, Shiller uses not the last 12-month earnings, but a 10-year average, adjusted for inflation.

The CAPE formula shows that pre-COVID earnings were unusually high versus history, and therefore probably due to reset at more normal levels, as measured, for example, as a share of GDP or revenues. Today, CAPE earnings measure $109. Hence, Shiller's P/E (3887 over $109) is 35.7. It has been that lofty only once in history, prior to the tech collapse in 2000. Every time it has even approached those heights, a steep selloff followed.

Adjusted CAPE

To calculate how much stocks could fall, it's best to adjust the CAPE. That's because it uses "real," or inflation adjusted, profits. The earnings investors put in their pockets grow with overall prices, because the companies whose stocks they own are selling their cars and groceries at prices that rise each year. On average over 10 years, profits would be 7.5% to 10% higher if inflation were included. We'll pick the 10% that's more favorable to the bulls.

Add 10%, and CAPE profits come to $120. That's a good estimate for where earnings will settle once national income returns to its pre-pandemic plateau. What P/E should apply? Since 1990, the S&P multiple has averaged 21.8. That's far higher than the 17.1 reading since 1950, but since low-20s P/Es have lasted a long time, it's at least reasonable to expect them to remain in that range, though a return to earlier norms is a risk.

If earnings reset at $120, and the multiple is 21.8, the S&P would sell at around 2620. That is 1270 points, or 33%, below today's record 3887. The numbers show a significant risk that a diversified basket of big-cap stocks could drop by one-third. Keep in mind that in March, the S&P cratered to 2237, almost 400 points below my best estimate of its fair value.

The bulls have a point in arguing that earnings will jump way beyond the 2019 record this year, and soar from there. But it's not a strong argument. Earnings have never grown to the sky before. The more stock prices keep spiking from towering to more towering peaks, the more likely they are to tumble. If you're fully invested in stocks, the crash could cost you a third of the portfolio that the Wall Street pros are saying can only grow from here.