ReportsMonitoringPortfolioMethodologyPricingBlogAboutStart for free
Back to blog

Analysis

Valuation Multiples Only Mean Something Against History

Valuation multiples carry nothing at face value. What counts is where the reading sits against the company's own band: window, rolling median, four states.

14 min read
See a sample report

This is a method, not a recommendation. Nothing here, or anywhere else on Taufolio, is investment advice. Treat every example as a starting point for your own research.

Is this company expensive? The face value of a multiple does not answer that, and it is the first thing to accept before valuation multiples start meaning anything at all. A price-to-earnings ratio of 40 can be a low valuation. One of 8 can be a high one. The only thing that settles it is where today's reading sits against the way that same multiple has behaved in that same company over the past several years.

That is not a way of dodging the question. It is the answer, delivered in a different unit from the one most readers expect: instead of a number you get a position, and a position takes half an hour to compute in a spreadsheet.

Before you start, though, make sure the question is due yet. Valuation is the last gate rather than the first. It earns its place only after profitability and financial safety have been cleared, because the cheapest company on the market is often simply the worst one. The order of the work is laid out in the piece on analysing a company before you invest.

Why the face value says nothing

Take two companies in one industry doing nearly the same thing. The first sells its software as a subscription, the second sells the same capability as a one-off licence. The first will trade higher, and not for a year but for a decade, because its revenue repeats itself while the second company has to win its revenue again every year.

Comparing their multiples directly therefore measures no price at all. It measures a business model.

The same thing happens across industries, only more visibly. A car maker and a software house differ not in which is cheaper but in how much capital each has to tie up to earn the same unit of profit. The multiple registers that as a difference in price when it is a difference in business.

Does that make multiples useless? Quite the opposite, as long as you treat them as what they are.

Which gives the first conclusion, one that looks like a detail and rearranges the whole analysis: a multiple is not a measurement, it is a time series. One number pulled out of that series means nothing, in the same way one temperature reading means nothing until you know whether it was taken in July or in January. Price to earnings shows it most sharply, being the multiple that says the opposite of what it looks like whenever a one-off lands in the denominator.

Valuation multiples are read against a company's own history

The best reference point for a company's multiple is the same multiple in the same company one, three and five years ago. The reason is plain. Everything that permanently lifts or depresses its valuation against other companies cancels out when the comparison is with itself: the revenue model, the capital intensity, the quality of the balance sheet, the market's habits around that name.

What remains is exactly what you wanted to measure: whether the market is paying more or less for this business today than it usually pays.

A problem that sinks most peer comparisons also disappears on the way. You no longer have to argue that two companies are comparable, because you are comparing one company with itself, and that is the only pair where comparability holds by definition.

The method charges an entry fee, and it should be paid knowingly. It needs history, so it does nothing for a recent listing, and it expires when the business has changed enough that the band describes a different firm under the same name. Selling half the operation, or making an acquisition the size of the company, is exactly such a moment. At that point the question of financial safety and of servicing the new debt comes back first.

How to draw the band in a spreadsheet

The whole procedure has five steps and lives in one file:

  1. Pick one multiple and stay with it. Price to earnings suits a company with a stable margin; price to sales suits one whose earnings jump around or turn negative.
  2. Pick a window. Five years is the floor, ten is often better, three is usually too few.
  3. Build the series: the closing price of each month divided by the matching per-share line from the filings, computed over the last four reported quarters.
  4. Add a rolling median over three years. A shorter one is jagged, a longer one smooths away precisely what you are looking for.
  5. Mark two edges: the level the reading repeatedly reached from below, and the one it repeatedly came down to. Not the all-time high and low, but the levels it kept turning back from.

Step five is the only one that calls for judgement, which is why it gets a chapter of its own below. The first four are arithmetic.

Four states instead of one number

Once the band exists, I stop quoting the value of the multiple and start quoting a state. There are four of them, and they cover every conversation about valuation I have ever needed to have:

Below the band. Lower than the reading ever came down to across the window. Either the market sees something your numbers do not, or the denominator is about to be revised down. The first thing to check is not the valuation but the forecasts.

Lower half. Between the bottom edge and the rolling median. The most ordinary state in the world and the one that needs the least interpretation.

Upper half. Between the median and the top edge. The market is paying more than usual and usually has a reason, which you should be able to name before you accept it as a good one.

Above the band. Higher than the reading ever reached across the window. This state demands an argument of its own, because it says the market has adopted a new assumption about this company.

A state says less than a number, and that is the best thing about it. A number implies a precision valuation does not have, while a state shows exactly as much as the series shows.

What is the state for, then? For one thing only: setting the height of the bar for the rest of the analysis. A company above its band has to answer the questions about its business better than a company in the lower half, because the market has already priced that better answer in.

What a state does not do is say anything about where the price goes next, and it has no such ambition.

Three numbers from one day

If anyone still believes you can say "it is expensive" without naming a window, one measurement across the whole market settles it. On 4 September 2026 the forward twelve-month price-to-earnings ratio for the broad US index stood at 19.5, against a five-year average of 19.8 and a ten-year average of 19.0 (weekly earnings review for the index, 4 September 2026).

One day, one index, two contradictory answers. Against five years the market sits below its own norm, against ten years above it.

This is not a defect in the measurement. It is the whole content of it. "Expensive" is not a property of a reading but a relation between a reading and a chosen window, so the window travels with the answer the way a unit travels with a number.

The same rule disarms the sentence that turns up in market conversations more than any other, the claim that shares are nearly always expensive. It is neither true nor false, because it cannot be checked until somebody names the measure, the window and the level at which expensive begins. Name those three and it stops being an opinion and becomes a reading, and a reading can be tested.

What about comparing with the industry?

It has its place, in a narrower range than the one it is usually applied to.

An honest comparison of two companies' multiples needs similar profitability, similar forecast momentum and similar financial safety, and that full set is rare. Without it, the gap between the multiples measures the gap between the businesses.

There are two situations, though, where the peer group is the only thing available. The first is a recent listing with no history of its own. The second is a question its own history never asks: whether this management is doing better than the managements working in the same conditions.

So I treat the two comparisons as different tools rather than as rivals. Its own history answers the question about price and the peer group answers the question about choice. Whether money can be made in this industry at all is a third question, and a more important one than either.

The edges are judgement calls

Here is the weakest point of the method, and I would rather raise it myself than wait for somebody to find it. Step five, drawing the two edges, has no rule. Suppose two people look at the same chart: one puts the top edge at 16, the other at 20, and both have arguments.

It sounds like an objection that dismantles the whole approach.

Except that every valuation method contains a judgement call, and the only real difference is whether you can see it.

In a discounted cash flow model the judgement calls are the discount rate and the terminal growth rate, and the effect of changing them disappears inside a spreadsheet that emits one precise-looking number. Here the judgement call is a line drawn on a chart and it stays in front of you the whole time. A model in which changing one input doubles the output is an instrument for testing sensitivity rather than a generator of prices, and that cuts both ways.

I think a visible judgement call is cheaper than a hidden one. Two people with edges at 16 and 20 will agree on the state in most cases, because the reading rarely sits exactly between them. And when it does, that is the information: this company is being valued ambiguously today.

How often should you recompute it?

Less often than instinct suggests. The band moves at the pace of years, because it is built from sixty monthly points, so one quarter shifts it by an amount you need a magnifying glass to see.

Two other things move fast: the price, daily, and the denominator, at every results release and every forecast revision. So I recompute the position once a quarter, alongside the report, and add one new point to the series rather than rebuilding the band from scratch. Redrawing the whole band every week produces the sensation of work and no new information.

There is one exception and it needs watching, because it spoils the series quietly. When a company changes how it presents its results, or buys or sells a large part of the business, the old series stops describing the same thing. The band then has to be computed from the moment of the change, accepting that it will be too short for a few years.

What the method will not show

The first thing is a cyclical company, and it is the trap that costs the most. Such a company's multiple is at its lowest at the top of the cycle, when the earnings in the denominator are at a record and about to fall. A reading below the band then means the opposite of what it looks like, so with commodity producers, chemicals and capital equipment I ask about the denominator first and about the position second.

Second: a multiple can fall without the price falling.

The denominator holds a forecast revised every few weeks, so a company can grow into its own valuation with no correction at all. Every move in the band splits into two components, price and forecast, and it pays to know which of them did the work. Plenty of things move a share price between results that have nothing to do with the company.

Third: with a loss, an earnings-based multiple simply does not exist, and with earnings near zero it takes values that look like signals and are not.

What remains is a sturdier denominator, sales or operating cash, with the caveat that price to sales ignores the margin and misleads whenever the margin moves. A company that lifted its margin from a few per cent to the mid teens looks steadily more expensive on such a chart, even though a unit of its profit costs exactly what it used to.

The fourth is the most insidious, because it looks like the method succeeding. A band measures the market's habits rather than value, so a company that spent five years being systematically overpaid for has a high band, and touching the bottom of it does not make the shares a bargain. The band answers the question of what the market usually pays for this, not what it is worth, and confusing those two questions is the one mistake capable of invalidating the whole exercise.

The fifth concerns the chart itself and is easily mistaken for a signal. A change of fiscal year, or a switch to a different revenue presentation, creates a step in the series. That step is an accounting change rather than a market event, so before reading the position I check that the band has no such break inside it.

And the last one, common to all valuation work: it disqualifies, it does not qualify. A high valuation is an argument against, and a low one is not an argument for. The list of questions about the business itself stands alongside and does not get one item shorter, as the set of twenty-five questions worth asking any company makes plain.

What to do with this in the next week

Take one company and build a band for it. You need a spreadsheet, five years of month-end prices and one per-share line from the filings, computed over the last four quarters and rolled forward every quarter.

Three columns are enough: date, multiple, rolling median over thirty-six months. On the chart, mark the two levels the reading kept turning back from, write down beside them why you chose those levels, and read which of the four states the company is in today. Record the window too, because without it the result cannot be reproduced a year from now.

Then repeat for a second company in the same industry and compare not the multiples but the states. That is the one industry comparison that does not require the two companies to be alike.

One closing sentence about the inverse calculation, because it goes further than everything above. Take today's price, assume the multiple returns to its own median, and the equation can be solved for the denominator instead, which tells you what growth in sales or earnings is already priced into today's quote. That is still a description of today's price rather than a forecast, and that is exactly why it is useful.

In Taufolio this part of the work is done by the valuation section of a Full report. It sets the company's multiples against their own history, says plainly where in that history today's reading sits, and links every figure to its source document. What the section looks like on a real company is in the sample reports.

And if you only have time for one window rather than two, take the longer one. The shorter one will always tell you that today looks normal.

Frequently asked questions

None of them on their own. A multiple is the price divided by one line from the filings, so its face value only says what you pay per unit of earnings, sales or cash, never whether that is a lot. The information arrives when you place today's reading against the way that same multiple has behaved in that same company over the past five to ten years.
Five years is a sensible floor and ten is often better, while three is usually too few, because three years can sit entirely inside one phase of the market. Past ten years a different problem appears: a different era of interest rates and cost structures enters the series, so the band describes a company that no longer exists. The window changes the answer, which is why you record it next to the reading.
With its own history, for as long as that history exists. A peer comparison needs similar profitability and similar forecast momentum, and when those conditions are missing the gap between two multiples measures the gap between two business models rather than a difference in price. The peer group earns its place in one case: when a company has been listed too briefly to have a band of its own.
Because the market pays for repeatability. A company with subscription revenue is easier to forecast and trades higher than a company selling the same thing once, and the gap persists for years rather than quarters. Comparing their multiples directly therefore measures the business model, not which of the two is cheaper today.
Not by itself. A cyclical company's multiple is at its lowest at the top of the cycle, when the earnings in the denominator are at a record and about to fall, so the lowest reading can be the worst moment. A low multiple is a question rather than an answer: check whether the denominator is durable before you conclude anything from it.
It means the denominator rose. The denominator holds a revenue or earnings forecast that gets revised every few weeks, so a company can grow into its own valuation with no correction in the price at all. That is why every move in a multiple splits into two components, and why it is worth knowing which of them did the work.
  • fundamentals
  • research
  • methodology
Share:XLinkedIn

Posts are produced with AI tools and go through editorial review by the Taufolio team before publishing. How we write and correct posts

Cookies and analytics

Essential cookies keep you signed in. Analytics runs only if you agree.