Beware the "E" in the P/E ratio fine print

https://www.wsj.com/finance/stocks/think-the-s-p-500-looks-cheap-read-the-fine-print-fa888f6e?mod=hp_lead_pos10

Think the S&P 500 Looks Cheap? Read the Fine Print

The index’s P/E ratio depends on who is defining the ‘E’

By Jonathan Weil, The Wall Street Journal, July 13, 2026

Wall Street pros are predicting stellar S&P 500 earnings growth this year and beyond. But there’s a problem with the underlying numbers. They are often deeply distorted…

If you pick an “E” that uses actual net income for the past four quarters, the index may seem pricey at about 29 times earnings.

More commonly, though, Wall Street analysts steer investors to some form of alternative earnings. Viewed this way, the index may trade for around 22 times earnings using analysts’ 2026 estimates, except the adjustments that analysts make routinely ignore regular, real-world expenses such as stock-based pay or restructuring costs…

Many Wall Street analysts don’t calculate GAAP earnings projections, even when asked explicitly for them in survey forms. So the consensus forecasts become a hodgepodge of estimates, with some tracking net income and some using earnings before bad stuff…


[end quote]

The rest of the article compares earnings growth forecasts provided by various analysts using GAAP and non-GAAP numbers. It’s confusing because it’s a hodge podge of different methods.

Wall Street analysts and major index providers often favor operating earnings (pro-forma or “headline” earnings), which intentionally strip out “one-time” or “unusual” costs.

The Valuation Gap: Because operating earnings exclude bad news like massive write-offs, aggregate operating earnings for the S&P 500 are almost always higher than as-reported earnings.

The “Fine Print”: If a valuation model uses operating earnings, the market will appear cheaper (lower P/E ratio) than if it uses Shiller’s stricter as-reported earnings.

Leaving out depreciation from earnings has a dramatic impact on AI. The chips are run 24/7 at high temperatures that physically wear out the microscopic solder points. The attrition is tremendous - about 1/4 are fried in 3 years. But the hyperscaler companies are depreciating the chips over 5 years instead of their physical life expectancy of 3 years. This makes non-GAAP earnings billions of dollars higher.

I prefer to look at the CAPE, the cyclically adjusted price-to-earnings ratio.
Cyclically adjusted price-to-earnings ratio - Wikipedia.

Using average earnings over the last decade helps to smooth out the impact of business cycles and other events and gives a better picture of a company’s sustainable earning power.

But many investors invest in the S&P 500 Index. The S&P 500 is a constantly-changing index that is selected by a committee at Standard & Poors. New, growing companies are added continually and failing ones weeded out.

The continuous rotation of companies within the S&P 500 introduces a structural phenomenon that systematically biases the long-term CAPE ratio upward, making modern stock markets appear more expensive relative to historical averages than they might actually be. The rotation ensures that the CAPE ratio will almost always flash an “overvalued” warning signal during prolonged periods of economic innovation and growth. The fastest-growing companies may not have had significant earnings in the past 10 years before their most recent growth. The fastest-growing companies may not have had significant earnings in the past 10 years before their most recent growth so their CAPE (10 year average) looks high.

Is the S&P500 in a bubble despite the earnings forecasts for 2026-2027?

Most of the earnings comes from within a circular “AI ecosystem.” Real end-users are only about 15% of the earnings. Earnings estimates are really based on building the AI infrastructure. Gemini calculates that the end-user spending will break even with investments around 2035. That’s assuming a very high growth rate of end-user spending from the current level of 30% to 35% CAGR – ignoring the more typical S-shaped curve of initial rapid adoption followed by plateau.

If the deep-pocketed hyperscalers like Microsoft, Alphabet, Amazon and Meta realize this and begin to slow down their spending the entire ecosystem share prices will crater. Everything is priced for perfection. That doesn’t even count cheaper or more advanced technologies from China already in the works to make up for the embargo of the most advanced chips.

Wendy

3 Likes

I read a long long time ago that P/E or EPS, etc., was highly fungible. A better ratio for valuation was P/S (Price to Sales) because it was a lot harder to fudge sales figures. Although I’m sure that can be done too.

5 Likes

And on that point:

Revenue Growth is the only silver lining I see in this market.

4 Likes

@Hawkwin thank you for sharing this chart.

  1. SPX is rising faster than Earnings Growth.
  2. 40% of SPX is tech-based, especially the AI hyperscalers.
  3. The AI-related sales are at least 85% within the “AI ecosystem” - that is, AI suppliers of capex and software to each other, not to end users. It will take many years, if ever, that sales to end users will equal the AI ecosystem spending. Kind of like the years it took after 2000 for the internet to catch up with the infrastructure.

Wendy

2 Likes

At the end of the day, you don’t take home your sales. You take home the net earnings. So that’s where the focus needs to be.

Sure, P/S can be interesting, but it’s better applied to a single company over time to get some insight into how the business is being run over time. At most, you could compare it within a pretty tightly defined industry.

And as a CPA, I can assure you there are a dozen ways from Sunday that sales can be manipulated from quarter to quarter and from year to year. Some are legit, some aren’t. Retailers and wholesalers run month end, quarter end, and year end sales to book more sales. Of course, the price cuts implicit in those sales affect profitability, but hey, we got more sales booked!

There are no shortcuts in financial analysis. You have to dig deep into a company - not just the most recent quarter or year, but back several years to get an idea of what is normal for a company and what is out of the ordinary. You can’t look at any single metric and expect to have a successful analysis.

Just like a doctor isn’t going to look only at blood pressure or total cholesterol or just a physical exam to diagnose a patient. The doctor is going to look at many data points to arrive at a diagnosis. Sometimes, those data points need to be followed over time.

Analysis for an investment isn’t looking at any one metric. You need to look at many metrics, and you need to follow them over some period of time.

--Peter

PS - Yes, earnings for a quarter or a year can be manipulated by management. But earnings over longer periods of time is much simpler: cash in minus cash out. Over several years, that figure simply can’t be manipulated.

10 Likes

At the same time…

JPMorgan’s results kick off what analysts expect will be another strong earnings season for big banks.

DB2

1 Like

P/S ratios vary widely by industry, so they are most useful when comparing competitors.

DB2

1 Like

Sales from quarter to quarter can easily be fudged. In fact, most smart corporations manage it closely. I prefer to use cash flow. Cash flow can’t be fudged (aside from outright fraud, of course), cash is cash is cash. So you could, for example, look at price to cash flow as one measure.

1 Like

Enron did a pretty good job of it for five years. Ten, if you count the groundwork laid in getting mark-to-market accounting approved by the SEC.

2 Likes

@MarkR @ptheland how would cash flow account for depreciation which impacts the bottom line?

According to Gemini, the AI chips are run at high temperatures and will start to fail in large numbers in about 3 years but the depreciation schedules say 5 years. These are physical failures and the chips will need to be replaced. This affects reported earnings apart from cash flow this year since cash flow doesn’t include depreciation but bottom-line earnings do.

Wendy

Interesting as I suspect the whole technology will be changing at least that fast, if not faster. In telecom, Bell Labs/WeCo built switching gear with supposed 50 year life expectancies, but each level was soon obsolete, as new ideas, needs came along pushed the old aside… I question the monster data centers being built, by the time they are online, most likely that tech will also hear footsteps.. Maybe even faster with AI helpers… Interesting times…

4 Likes

Remember that depreciation is just an artificial way of spreading the cost of a long (-ish) lived asset over a number of years. While management can try to goose earnings by using a long life for depreciation, most assets (other than real estate) are basically worthless after about 5 years. Perhaps 10 or even 20 years for some specialized machinery (like generators in a dam).

Go to the extreme. Over the life of a business (birth to death) earnings are exactly cash in less cash out (other than cash returned to owners as profits).

As you look at smaller and smaller time periods, cash flow starts to slightly distort the profits of the business. Expenses are paid that benefit multiple periods, so various means are used to spread those costs over longer time periods. That is what depreciation and amortization do. Accounts receivable and payable do the same thing - we actually sold (or purchased) an item, but have received (or paid) the cash yet. Loans are either repaid or defaulted on and effectively become income.

The longer the time period you look at, the closer GAAP income comes to cash in less cash out.

--Peter

2 Likes

There are many articles about the short lifespan (1-3 years) of data center GPUs. This shouldn’t be so speculative. The NYT tells us that back in 2022 there were already 2700 data centers in the US alone. What has been their performance?

DB2

2 Likes

I had Gemini generate the P/FCF for several companies:

Company Ticker P/FCF Ratio (TTM) Context & What’s Driving the Number
Meta Platforms META ~36.0 Meta remains highly efficient at spinning off cash from its family of apps, keeping its valuation multiple comparatively reasonable despite aggressive spending on Reality Labs and AI hardware.
Microsoft MSFT ~43.0 Reflects a premium valuation. While Microsoft generates massive operational cash flow, its massive Azure and AI infrastructure buildout absorbs a highly significant portion of that capital.
NVIDIA NVDA ~43.4 NVIDIA’s explosive free cash flow growth has actually kept its FCF multiple remarkably in line with software giants like Microsoft, despite the stock’s massive price appreciation.
Alphabet GOOGL ~67.1 Google’s core advertising engine is highly cash-generative, but aggressive, multi-billion dollar CapEx cycles for Google Cloud and AI data centers have temporarily pinched FCF, pushing the multiple higher.
Amazon AMZN Highly Elevated / Negative Amazon’s P/FCF fluctuates dramatically. In years of hyper-aggressive global data center and logistics buildouts (projected to hit record highs), CapEx matching or exceeding operating cash flow can make its P/FCF appear highly distorted or negative.
2 Likes

You should take a look outside the tech industry. I’m not surprised these 5 are all fairly similar - save for Amazon.

Operating earnings is the cleanest and better metrics. Note, it includes depreciation, SBC and excludes one-time expenses like write-down. Those who claim you need to include write-down’s and prefer GAAP income, and often claiming they are purists, don’t want to talk about GAAP also has many phantom income, for ex: stock price fluctuations, companies recording gain on selling an unit, etc. The one-time write-down’s do not impact a business on-going ability to generate operating income/ profit.

What is more important is whatever metrics you use, use it consistently. Who really bring these kind of arguments are those who “feels” markets are expensive and when their narrative is not helping them (like current earnings growth), they need new arguments…

A much simpler metrics I use is when the earnings are powerful, you should demand lower PE and when they are troughing tolerate higher PE. But many folks who complain about high multiples will also cry when earnings are troughing in a recession…

Being in the permabear company is like living in bad neighbourhood, it only brings misery.

1 Like