Relative Valuation: P/E, EV/EBITDA, P/S, P/B

Match the numerator to the claim

The single rule that prevents most multiple errors is that the numerator and the denominator must describe the same claim. Price and market capitalisation are claims on the equity, after lenders are paid, so they pair with metrics that belong to the equity: earnings, book value, sales in a business with little debt. Enterprise value is a claim on the whole business, so it pairs with metrics that exist before financing: EBITDA, EBIT, revenue, and free cash flow to the firm. Break the rule and the multiple carries the capital structure as a hidden variable. The price-to-earnings ratio of a heavily levered company is depressed by interest and by the debt that produced it, so two companies with identical operating performance trade at different P/Es for a reason that has nothing to do with value. EV/EBITDA removes that distortion, which is exactly why it is the default in leveraged industries and in acquisitions, where the buyer will refinance the debt anyway. Each multiple then answers a narrow question. Price-to-earnings is the market price of a dollar of accounting profit, useful for a stable, lightly levered business with clean earnings. EV/EBITDA is the price of a dollar of operating cash-like earnings, useful across capital structures and for comparing capital intensity, and it is flattered by heavy depreciation because it adds back a cost that some businesses genuinely incur every year. Price-to-sales is a check on the revenue side and is the least informative on its own, because it says nothing about margins. Price-to-book is the right tool for banks and insurers, whose assets are financial and closer to market value, and much less useful for a business whose value is intangible. The same company, four lenses — Market capitalisation, then enterprise value: $11,780m, then $13,110m with net debt added · EV/EBITDA: 17.0× against a 12.0× peer median · Price to earnings: 35.4× · Price implied by the peer multiple: $41.63 ← A negative denominator makes a multiple meaningless, not infinite. A company with negative EBITDA has no EV/EBITDA, and the peer table that quietly drops the row is telling you the comparison does not work for this kind of business.

Choosing peers, and reading what the multiple already assumes

A multiple is only as good as the peer set, and the peer set is a judgement. The honest approach is to define the comparison along the dimensions that drive the multiple — growth, operating margin, capital intensity and risk — and then ask whether the target is genuinely comparable on each. Industry classification is a starting point and often a bad one: two companies in the same sector can have completely different margins and reinvestment needs, and the multiple difference between them is then a fact about their economics rather than a market mistake. It is worth knowing what the ratio is made of, because it explains every multiple difference you will find. Rearranged, a price-to-earnings ratio is the payout ratio divided by the difference between the cost of equity and growth. So a company with a higher multiple must have one of three things: faster growth, lower risk, or a higher payout. There is no fourth option, and the arithmetic turns a vague argument about cheapness into a specific claim you can test. That is the discipline this lesson is really teaching. When the peer median implies $41.63 against a $62.00 price, the productive question is not whether the shares are expensive but which of the three inputs the market is being more generous about — and whether you agree. The company may be growing faster, or levered in a way that makes the equity riskier and therefore deserving of a lower multiple, or retaining all of its earnings to fund a return above the cost of capital, which justifies a higher one. A multiple never answers that question; it only asks it in a form that can be checked against the statements. Cross-check the direction of travel. A multiple that has compressed while the fundamentals held is a different situation from the same multiple on deteriorating fundamentals, and the price chart alone will not tell you which you are looking at.

Where every multiple breaks

Each multiple is a shortcut that assumes something specific about the business, and each one fails in a recognisable way. Learning the failure mode matters more than learning the formula, because the formula takes a minute and the failure costs a thesis. **P/E** breaks first on capital structure: a levered company and an unlevered one with identical operations produce different earnings per share, so the cheaper P/E may simply be the more indebted business. It also breaks on one-off items, since the denominator is a single year. **EV/EBITDA** fixes the leverage problem and introduces a worse one: it treats depreciation as noise, which is true for a software company and nonsense for an airline whose aircraft wear out. **P/B** works only where the assets are financial or replaceable, and reads absurdly for a business whose value is a brand. **P/S** ignores profitability entirely, so it is a useful screen for young companies and a trap for anything mature. There is a second-order failure that matters more than any of these: the multiple embeds a cycle position. A steelmaker at 6× at the top of a cycle is more expensive than the same company at 12× at the bottom, because the denominator is about to halve. Normalising the earnings — taking an average margin across a full cycle — is the repair, and it is the single step that separates a comps table from a comps argument. The same company, four multiples, four stories — 12× earnings with debt at 3× EBITDA: cheap until you notice the interest cover · 7× EV/EBITDA, capital-intensive: the depreciation is real; EBITDA flatters it ← · 1.9× book, asset-light: the assets that matter are not on the statement · 0.8× sales, 2% margins: a screen that says nothing about profitability · 6× at a cycle peak versus 12× at the trough: the cheap one is the expensive one ← A comps table is only as good as its normalisation. If the peers are at different points in their own cycles, you are comparing four different years and calling it a valuation.

A multiple is shorthand for two assumptions

A price-to-earnings ratio looks like an observation about the market and is actually a compressed valuation model. Rearranged, a stable-growth valuation implies that the multiple equals the **payout ratio divided by the difference between the required return and the growth rate**. Read that way, every multiple is a statement about three things at once — how much of the earnings is paid out, how fast the earnings grow, and what return an owner demands — and the same multiple can come from completely different combinations. A company with a high payout and low growth trades at the same multiple as one with a low payout and high growth, and the two have very different risk. This is why “it trades at fifteen times” answers almost nothing on its own: the useful question is which of the three inputs the market is being generous about. The decomposition also explains the most common misuse of comps, which is comparing a multiple across companies with different balance sheets. Earnings belong to shareholders and sit below the interest line, so any comparison of P/E across companies with different leverage is comparing the same business at different capital structures. Moving up to enterprise value over operating earnings fixes the numerator and the denominator at the same level, which is why it is the standard multiple for comparing a set of real businesses. The exception proves the rule: for banks and insurers, debt is raw material rather than financing, so book value and its multiple return as the right lens. One more property is worth carrying because it decides how much to trust any screen. Multiples are much better at **explaining** prices in the past than at **predicting** returns in the future. Cross-sectionally, a cheap-looking set usually does earn a modest premium over long periods, but the dispersion within the cheap set is enormous, and most of it comes from the fact that the market is pricing something the multiple cannot see — a deteriorating business, a cyclical peak, an accounting difference. A multiple is a question about what the price assumes, not an answer about what will happen; the answer only comes from the reverse-DCF exercise of asking what has to be true for the multiple to be deserved. The same 15× from three different places — High payout, low growth, average return required: Utility-like: the multiple is mostly the payout · Low payout, high growth, average return required: Growth is doing the work, and it is the assumption to attack · Average payout and growth, low required return: The discount rate — usually a rates story, not a company story ←

Cleaning the denominator, or the multiple is noise

The note at the end of the previous read said a comps table is only as good as its normalisation, and this is the work that sentence is asking for. The numerator of a multiple is a market price and is not in dispute; the denominator is a choice, and four choices have to be made the same way for every company in the table. **Trailing or forward?** A trailing multiple uses what the company reported; a forward multiple uses consensus estimates, which means a “cheap” forward P/E can be cheap because the price is low or because the estimates are high — and estimates get revised. Both are legitimate, but a table that mixes them is comparing a fact with a forecast. **Reported or adjusted?** The same company can trade at twenty times GAAP earnings and fifteen times its own adjusted figure, and the five-point gap is the reconciliation: the exclusion that recurs every year is an operating cost with a flattering label, so the honest move is to normalise the peers to the same treatment rather than to whichever number is lowest. The third choice is the one that does the most damage, because it is invisible unless you make it deliberately: **where in the cycle does this earnings figure sit?** A cyclical business at the top of its cycle reports an enormous denominator, so its P/E looks cheap precisely when it is most expensive — the market is capitalising peak earnings that are about to fall. At the bottom the arithmetic inverts and the same company looks wildly expensive on the trough number. The repair is mid-cycle earnings: take a full-cycle average margin — or an average of the last seven to ten years, which span a cycle — and apply it to current revenue, then build the multiple on that. The pattern to remember is uncomfortable and short: for cyclicals, the multiple has to be computed on a number the company has not yet reported. The fourth is the share count, and it is where adjusted figures quietly become fiction. Diluted shares include options and unvested restricted stock, and if a company excludes stock-based compensation from its adjusted earnings while still counting the shares, it has taken the expense out of the numerator and left the dilution in the denominator only once. Worse, a forecast that grows earnings per share while issuing shares can show growth that is entirely arithmetic. So: use diluted shares, keep the same share-count convention across the peer set, and treat negatively denominated multiples as undefined rather than small. When earnings are negative or near zero, the P/E has no meaning at all — switch to EV/Sales, EV/EBITDA or price to book, and say so in the table rather than printing a number, because a multiple of two hundred on a penny of earnings is not a valuation, it is a division by almost nothing. One company, four multiples, four different stories — $120 on trailing GAAP EPS of $10: 12× — but $10 is a peak-cycle figure · $120 on mid-cycle EPS of $5: 24× — the same price, read honestly ← · $120 on adjusted EPS of $8: 15× — the gap is the recurring exclusion · $120 on trough EPS of $1: 120× — the multiple is a division by nearly nothing ← The mechanical test before any comps table is built: write the four conventions on the table itself — trailing or forward, reported or adjusted, mid-cycle or as-reported, diluted shares — and confirm every row uses the same four. A table without the conventions printed is not a valuation, it is a set of divisions that happen to be in the same file.

Sources

• Multiples, comparables and the determinants of each ratioDamodaran, "Investment Valuation"; Liu, Nissim & Thomas (2002), "Equity Valuation Using Multiples"
• Enterprise value and the equity bridgeStandard valuation practice; Koller, Goedhart & Wessels, "Valuation"
• Why multiples must be matched to the same claim on the businessStandard valuation practice; CFA Institute equity curriculum