Relative Valuation: Pricing a Company Against Its Peers

Relative valuation prices a company by what the market pays today for similar companies. Choose a peer set, choose a multiple, take one statistic across the peers, and apply it to the subject company's own denominator. The median of six invented peers, 7.8 times, applied to Sankalp Industrial Systems Limited's Rs 2,88,00,00,000 of earnings before interest, tax, depreciation and amortisation (EBITDA), gives an enterprise value of Rs 22,46,40,00,000. All four choices are assumptions, and the multiple hides every one of them.

The awkward part of this method is not the arithmetic. Begin on a street rather than in a spreadsheet. Outside, the difficulty registers faster. Suppose a neighbour is selling a small sweet shop and asks what it is worth. There are no accounts, no forecast and no patience for either. So the answer arrives the way it arrives for almost everybody. The tea stall two doors down changed hands last year for about four times what it cleared in a year, and a bigger shop near the bus stand went for about six times. The neighbour's shop clears roughly Rs 6,00,000 a year. Somewhere around Rs 30,00,000, and the figure feels reasonable to say.

The sweet shop answer contains the whole of this method. A set of businesses was picked and judged similar, a way of measuring them was picked, one number was picked out of the range those businesses gave, and it was multiplied by something belonging to the shop in front. Four decisions, made in about eight seconds, and the answer that came out the other end looks like a fact. Every relative valuation ever produced, on any company of any size, is that same sequence of four decisions with better vocabulary attached. The better vocabulary arrives with the four named steps. The discomfort is warranted by how quickly those four decisions went past.

What is relative valuation actually claiming?

Relative valuation claims something narrower than most people who use it believe. Relative valuationPricing a company from what the market pays for similar companies. says: here is what buyers and sellers are currently paying for businesses I have decided resemble this one, and if that is what they pay for those, this is roughly what they would pay for this. The claim describes prices in a market on a day. The claim never becomes a statement about what a business is worth in itself.

The distinction between a price and a worth sounds like hair splitting until what it rules out becomes clear. If every business in a sector is being priced on the same assumption and that assumption turns out to be wrong, a relative valuation carries the wrong assumption faithfully into the answer and reports no error at all. A relative valuation was never checking anything, so it cannot report an error. The method was copying. Relative valuation reports what the market is currently paying for a certain kind of business; it does not report whether the market is right to pay it.

The other approach in the trade does the opposite. An intrinsic model forecasts the cash a business will produce, decides what rate to bring that cash back at, and builds a value from the ground up without asking anybody what a similar company sold for. Intrinsic valuation is covered separately; its answer for Sankalp Industrial Systems Limited is Rs 21,28,13,79,094. There is also a set of prices actually paid by buyers who took control of whole companies, and a way of valuing a group by valuing its divisions one at a time, and both of those are covered separately as well.

What does a single multiple compress into one number?

A multipleA value divided by a measure of what produces it, such as EBITDA or earnings. is a price divided by whatever produces that price. Enterprise value divided by EBITDA. Share price divided by earnings per share. The division looks trivial and it is not. A great deal disappears inside it. When somebody pays Rs 20,70,00,00,000 for a business, that price already contains their view of how fast it grows, how much capital that growth eats, what the capital costs, how likely the whole thing is to go wrong, and how long any of it lasts. Divide the price by one year of EBITDA and all of it is still in there. None of it is legible.

Aswath Damodaran makes this argument better than anybody, and it is his: a multiple is a compressed valuation, and reading one is reading a valuation with all the working removed. An intrinsic model spreads its assumptions down a schedule where each one can be disagreed with on its own line. A multiple folds those same assumptions into a single number, and the number offers no handle to disagree with. The compression is exactly what makes a multiple fast, and exactly what makes it dangerous, and there is no version of the method that delivers the speed without the blindness.

What are the four choices between a set of peers and a value?

Four, in a fixed order, and the arithmetic only starts after the fourth. Choose who counts as a peer. Choose which multiple to read off them. Choose which single statistic summarises the set. Apply that statistic to the subject company's own figure. Every one of those is a judgement, three of them are made before a calculator is touched, and the fourth is the only one anybody ever checks.

Notice what that ordering means for anybody reviewing the work. A multiplication is nearly always correct. So a reviewer opens the sheet, sees the multiplication, checks it and finds it correct. The three decisions that actually determined the answer left no arithmetic behind them to check. In relative valuation the reviewable part of the work is the part that never goes wrong, and the part that decides the answer leaves no trace a reviewer can test.

Step one: who is even in the peer set?

Everything downstream is bounded by this. A peer setThe named companies whose prices are being borrowed for the subject company. is a list of companies somebody decided resemble the one being valued, and the statistic in step three can only ever pick from the numbers this list supplied. Six sleepy industrial businesses, and no arrangement of the later steps can produce a fast-growing company's price. One fast-growing business added to a set of five sleepy ones, and the later steps have been handed a number they were never going to ignore.

Building such a list by screen, and the argument about whether any one company belongs in it, are covered separately. The method needs a set to work on, so the set is locked here: six invented companies, numbered, alongside the subject company.

Peer 6 is the interesting one. Peer 6 grows about twice as fast as anything else in the set, carries net cash rather than net debt, and holds an EBITDA margin six percentage points above the next highest. Peer 6 is also the smallest: its Rs 5,00,00,00,000 of revenue is 41.67 per cent of Sankalp Industrial Systems Limited's Rs 12,00,00,00,000, and 62.50 per cent of the smallest other peer, peer 5 at Rs 8,00,00,00,000. The two comparisons are different, and quoting the wrong one has started arguments. So the base belongs with either figure whenever it is quoted. Whether peer 6 belongs in this set at all is a real question with two defensible answers, and it is covered under the selection of comparable companies.

Step two: which multiple is borrowed off them?

Every peer above carries more than one. The table shows enterprise valueThe value of the operating business to every provider of capital, lenders included. against EBITDAEarnings before interest, tax, depreciation and amortisation., but the same six companies also carry enterprise value against revenue, and price against earnings per share, and several others. Peer 6 reads 13.8 times on EBITDA, 4.14 times on revenue and 28.0 times on earnings. The three readings describe one price, and the three do not rank the set the same way.

Which of them to use, and the rule that decides which numerator may sit above which denominator, are covered under the individual multiples. Only two things matter at this altitude: the choice exists, and the choice moves the answer. Two analysts can read the same six companies on the same day and pick different multiples. Both valuations are honest, the two disagree, and neither contains an arithmetic error. The choice here is fixed at enterprise value to EBITDA and stays fixed, so everything that moves from here moves for one reason and not two.

Step three: which statistic is taken across the peers?

There are now six numbers and one is needed. Two candidates are usual: the medianThe middle value once the observations are put in order. and the meanThe arithmetic average of the observations, being their sum divided by how many there are.. A third option, honest and almost never taken, is to refuse to pick either and carry the whole range forward. The shape matters more than either statistic, so put the six in order first.

The shape of the ranked set is the whole of step three. The step from peer 5 to peer 6 is 5.1 turns. The entire spread from peer 1 to peer 5 is 2.1 turns. A turnOne whole unit of a multiple, so a move from 7.8 times to 8.8 times is one turn. is the natural unit here because a rupee gap means nothing across companies of different size. The choice of statistic is really a choice about how much weight to give one company that does not look like the other five. The median gives it one vote out of six. The mean gives it a vote weighted by how far away it stands.

Why is the median the default rather than the average?

Because of what each one is counting. A median counts positions: line the six up, take the middle. With an even count there is no single middle. The median is then the average of the third and the fourth, 7.6 and 8.0, giving 7.8 times. A mean counts distances: it adds every observation's full value, so an observation standing a long way out contributes its whole distance rather than a single vote. The six sum to 51.8, and 51.8 over 6 is 8.63 times. The project record carries that figure rounded to two decimals.

Now do the test. Take peer 6 out of the average alone. The mean falls from 8.63 times to 7.60 times. The fall of 1.03 times is more than a full turn. The median of the six was 7.8 times and it is still 7.8 times. Peer 6 was never the middle of anything, and moving it or removing it from an average was never going to disturb which observation sits in the centre of the queue. One company out of six moves the mean by more than a whole turn and moves the median by nothing at all, and that single fact is the entire reason the median is the working default.

A second fact is worth having, and it looks like a coincidence until its work becomes visible. Dropping peer 6 from the set altogether, rather than only from the average, leaves five observations whose median is 7.6 times and whose mean is exactly 7.60 times. The two measures land on the same figure. Take away the one observation that was pulling the average away from the middle, and the average and the middle agree. No demonstration of what an outlier does to a mean is cleaner. Whether that removal is the right thing to do is a different question, covered under the selection of comparable companies.

Step four: whose denominator does the multiple get applied to?

The subject company's own, always, and nobody else's. The peers supplied the multiple and that is the entire extent of their contribution. The peer median of 7.8 times, multiplied by Sankalp Industrial Systems Limited's own Year 0 EBITDA of Rs 2,88,00,00,000, gives an indicated enterprise value of Rs 22,46,40,00,000.

The rule sounds too obvious to state until somebody gets it wrong under time pressure. Getting it wrong happens more often than expected when a spreadsheet has seven columns and one of them is the wrong one. A peer's multiple applied to that same peer's EBITDA reproduces that peer's own enterprise value, already sitting in the table, and establishes precisely nothing about the company in question. The multiple comes from the peers and the denominator comes from the subject company. Mixing the two produces an answer that is arithmetically perfect and completely empty. A further rule governs which kind of numerator may sit above which kind of denominator, so an enterprise value is not divided by a figure belonging only to shareholders. The consistency rule is covered under the individual multiples.

What does the worked instance actually produce?

Two answers from the same six companies and the same denominator, differing only in which statistic somebody preferred. At the median of 7.8 times, Rs 2,88,00,00,000 of EBITDA gives an enterprise value of Rs 22,46,40,00,000. At the mean of 8.63 times, the same Rs 2,88,00,00,000 gives Rs 24,86,40,00,000. The distance between them is Rs 2,40,00,00,000.

Hold on to what did and did not happen in that sentence. Nothing about Sankalp Industrial Systems Limited changed. Not a rupee of revenue, not a percentage point of margin, not a machine, not a customer. The company was identical in both calculations and the answer moved by Rs 2,40,00,00,000 because one person prefers medians and another prefers averages. The Rs 2,40,00,00,000 gap is not a valuation range produced by uncertainty about the business; it is a valuation range produced by a preference about statistics, and the two get written down in the same font.

Precision matters here, and this is exactly where a calculation can quietly go wrong. The mean of the six is 8.633333 times. The mean is printed as 8.63, the precision the peer multiples themselves carry. Every figure derived from it is computed on the unrounded value. The unrounded value is why the mean applied to Rs 2,88,00,00,000 gives Rs 24,86,40,00,000, and why the gap against the median is exactly Rs 2,40,00,00,000. A figure derived from a rounded input carries that rounding into everything downstream of it. Round for display and compute on the full value, never the other way round. The other way round here gives Rs 24,85,44,00,000 and a gap of Rs 2,39,04,00,000, adrift by Rs 96,00,000. Neither of those is bad arithmetic on its own; what breaks a model is mixing the two inside the same set of numbers.

How wide is the answer when only the multiple changes?

Wider than most readers expect, and this is the figure worth carrying away. Leaving statistics aside for a moment, the six observed multiples do something on their own. Each one applied in turn to the same unchanged Rs 2,88,00,00,000: peer 1's 6.6 times gives Rs 19,00,80,00,000, and peer 6's 13.8 times gives Rs 39,74,40,00,000. The spread between them is Rs 20,73,60,00,000, or 109.09 per cent of the low end.

Choosing which of six companies to copy opens a range wider than the whole traded enterprise value of the company being valued. Sankalp Industrial Systems Limited's own traded enterprise value is Rs 22,40,00,00,000, and the spread that step one and step three between them can open is Rs 20,73,60,00,000. The decision about who is comparable is not a preliminary before the valuation. On this set it is worth more than every other choice combined.

What does the market method give that an intrinsic model does not?

Three things, and they are real rather than consolation prizes. The first is observability. Every input in the peer table above came from something somebody actually paid. An intrinsic model's inputs are somebody's forecast of what has not happened yet. The second is speed. A peer table can be assembled and applied in an afternoon; a model that forecasts five years of cash flow and then argues about what happens afterwards cannot. The third, and the one people underrate, is that a multiple is a shared language. An operator in this industry, told that businesses like theirs change hands at around eight times, knows exactly what is meant.

The cost is precisely the mirror of the benefit. Because the inputs are observed, the answer inherits whatever the market currently believes, including its mistakes, and reports no error when it does. Because it is fast, the assumptions never get written down, so nobody can go back and check which one was doing the work. An intrinsic model states what was assumed and can therefore be wrong in a way that can be found; a multiple states what somebody paid and can be wrong in a way nobody can locate. Neither of those is a defect to be fixed. Multiples and intrinsic models are two different instruments. The trade uses both side by side, precisely to make a disagreement between them visible.

What can a multiple never establish?

Whether the price it borrowed was a sensible one. The soundness of the borrowed price is the boundary of the whole method, and the boundary does not move however carefully the four steps are executed. Sankalp Industrial Systems Limited trades at 7.78 times its Year 0 EBITDA, being Rs 22,40,00,00,000 over Rs 2,88,00,00,000. The median of these six invented peers is 7.8 times. The two figures are close, Rs 6,40,00,000 apart on the enterprise value, or 0.29 per cent.

Now watch how strong the pull is to say one more sentence. Everything about the shape of that comparison invites a verdict, and there is no verdict available. One thing has been established: on one measure, against one invented set of six companies, on one day, this company's multiple sits near the middle of that set. Nothing establishes that the company is worth what it trades at, nor that it is worth more, nor less, and no rearrangement of these numbers will establish any of those things. A company priced below its peers is either priced for something the multiple does not capture or it is not, and the multiple alone cannot distinguish between those two situations.

The second of those possibilities contains a great deal. A lower multiple can mean a business that grows more slowly, or earns less on each rupee it reinvests, or carries more debt, or depends on three customers, or is about to lose a contract nobody outside knows about yet. A lower multiple can equally mean nothing at all except that fewer people have looked. The compression that made the number fast also threw away the very information needed to separate those cases. So the multiple registers the same 7.6 times in every one of them. A multiple is therefore a question and not an answer: it marks where to start asking, and the asking is a different job entirely.