August 19, 2026 · Markets, Investing, Data

Stop caring about all-time highs

Every few weeks a headline announces that the S&P 500 has closed at an all-time high, and somebody decides to wait. Wait for a dip, wait for the froth to come off, wait for a level that feels less like the top. The instinct sounds reasonable. It just rests on a mistaken picture of what a record is.

A record is the normal state of a rising line

Start from the economy rather than the chart. A stock index is a claim on the earnings of a few hundred companies. Those companies retain part of what they earn and reinvest it. The population they sell to grows. Output per hour worked grows. And on top of all of that, central banks deliberately aim for prices to rise by about two percent a year, forever — which means the same real earnings are counted in more dollars with every year that passes.

Put those together and the index has an upward drift built in. Not a guarantee about any given year; a drift. And here is the part that gets skipped: a line that drifts upward spends most of its life at the highest point it has ever reached. It has to. The only way for a rising line not to be at a record is for it to be climbing back out of a fall — and the falls are the interruption, not the baseline.

Nominal US GDP sets a record in most quarters. Nobody writes about it, because everyone understands that a growing economy producing a record amount of output is an ordinary Tuesday. The index is exactly the same fact, measured to more decimal places and delivered with more adrenaline.

Every window looks the same

The chart below is the S&P Composite from 1871 to today, measured monthly. Drag the start year and the view refits to whatever is left. Shave off the first century. Then the next fifty years. Then start it the year you were born.

The shape survives every cut. Whichever decade you start from, you get the same picture: a line that climbs to the right, appears to climb faster the further out it goes, and is dotted with the months where it stood higher than it ever had. The dots are not rare events. They are most of what the line does.

The share of months at a new high actually rises as the window shortens: 18 % of all months since 1871, 31 % since 1950, roughly half over the past ten years. Partly that is because a short window has a lower bar to clear. Mostly it is because the further you are from the last crash, the more of the sample sits above everything that came before.

The acceleration is an illusion, and that is the whole point

Now switch the chart to a logarithmic axis. The curve straightens into something close to a line.

That is the tell. On an ordinary, linear axis, equal percentage gains draw bigger and bigger steps, because 10 % of 5,000 is a hundred times the points that 10 % of 50 was. The “faster and faster” rise is not the market accelerating. It is the same rate of growth, drawn against a bigger number. On a logarithmic axis, where equal percentage moves take equal vertical space, the acceleration disappears and a steady slope shows up in its place.

Which is why a record number, on its own, tells you nothing about whether the market is expensive. The level is the accumulated product of every prior year of growth. It has to be the largest number in the series; that is the only thing a compounding series can produce. Standing at 7,000 rather than 700 says something about how long compounding has been running, not about what you are paying for a dollar of earnings. Read a record as information and you have mistaken the arithmetic of compounding for news.

To care about a record is to make a forecast

Here is the part worth being precise about. An all-time high is a fact about the past and nothing else: today’s price is above every previous price. That sentence contains no information about tomorrow — unless you add an assumption that the level a series has reached tells you something about where it goes next.

That assumption is a forecast, and it is a very specific one: past prices predict future returns. Anyone who sells or waits because of a record is making that forecast — usually without noticing that they have made one.

It is a testable claim, so test it. Using the same 1871–2026 monthly series, split every month into two piles: the ones that set a record and the ones that did not. Then look at what the following twelve months delivered, in real total return.

There is no penalty. If anything the record months came out marginally ahead, and lost money slightly less often. The historic data does get consulted here — and it says the opposite of what the person invoking it believes.

What a record actually tells you

It tells you one thing, and it points backwards: nothing bad has happened lately. Records arrive in clusters, because they can only occur when the market is not climbing back out of something. Long stretches without one are not calm — they are the aftermath. The S&P went 299 months, from September 1929 to September 1954, without a new nominal high. Twenty-five years. That drought is the thing that deserves respect, and you will notice it is not a record that causes it.

The honest caveats

Three of them, so this does not read as a sales pitch.

1. 01

   Much of the record-setting is inflation. Switch the chart to inflation-adjusted: the same series sets a record in only about a tenth of its months, and the real price is up roughly 64× since 1871, against 1,700× nominal. Nominal records are partly just the money shrinking.

2. 02

   “Records are normal” is not the same as “valuation does not matter.” What you pay relative to earnings has historically mattered a great deal for long-run returns. It just is not measured by the index level. Price-to-earnings is a ratio; the record is a number. They are not the same instrument.

3. 03

   None of this says the next twelve months will be good. It says the record itself is not the reason to think they will not be. The real risk lives in your time horizon, in whether you can sit through a 50 % drawdown without selling, and in what you pay in fees — not in the headline.

So

A record high is what a growing economy looks like when you plot it. It shows up every few weeks, it has shown up in every historical window you can cut, and it has never been a signal in the data. Treating it as one means quietly betting that past prices forecast future prices — a bet the past itself declines to support.

There are good reasons to be careful with money. This is not one of them.

Not investment advice. This is a piece about how a chart behaves, written by a software developer, not a licensed adviser. Every figure in the text is computed from the dataset sitting under the chart.