Comparable Company Analysis: Choosing the Peer Set

A comparable company is one the market prices on the same terms as the subject. Four dimensions decide that: business mix, size, growth, and capital intensity with the risk that comes with it. Across the six numbered peers the multiple rises in step with margin, growth and return on capital, and not with size. Whether the sixth peer belongs in the set at all has two defensible answers.

Consider two catering businesses in the same city. The difficulty here is not arithmetic, and it shows itself faster over food than over a spreadsheet. The first does about forty weddings a year. Rented halls, hired staff, a lorry-load of vessels, three months that pay for the other nine, and a thin sliver left at the end. The second feeds the canteens of eight offices on annual contracts. Same city, same suppliers, and almost everything else different: money arrives every month rather than in a monsoon-shaped lump, the staff are permanent, the equipment sits in one place, and what is left at the end of a year is a much fatter sliver on a much smaller turnover.

Suppose the second business changed hands last year at six times its annual operating profit, and the question is what the first one is worth. The objection arrives before it can be named. Both are catering. Both would sit under the same line in a directory. Nobody who has stood in either kitchen believes one price says much about the other. The disbelief that one price says much about the other is right, and the work is to turn it from an instinct into four things that can be checked.

Borrowing another company's price is a defensible act rather than a lazy one only under one condition. A peer set is the list of businesses judged to be priced on the same terms as the subject. Six invented companies make that decision visible. Five of them line up beautifully. The sixth starts an argument with no settled answer. Both sides of it are set out below, and neither wins.

What does comparable actually mean here?

Not similar. Similar is a word about products, and a peer set is not a list of businesses making the same thing. A comparable companyA company the market prices on the same terms as the subject. is one where the market, looking at it, answers the same question it would answer about the subject and reaches the same kind of answer. Two businesses can turn out identical products and be priced on entirely different terms; two businesses in unrelated lines can be priced on almost identical ones.

Return to the caterers. Both cook. If likeness of product settled the matter they would be a matched pair. But when somebody prices the second one, what is really being priced is a stream of monthly contract revenue that rises slowly, needs very little new equipment to rise at all, and comes from eight customers who could each leave at three months of notice. When somebody prices the first, they are pricing a seasonal, labour-heavy business with high turnover, thin margins, and a working capital swing every wedding season. Two prices, two questions, one product.

Comparable means the pricing question is the same, and that question is made of a small number of things that can, with effort, be observed from outside. The reframing turns an argument about judgement into a checklist. The same reframing explains why peer sets built at speed go wrong so reliably: they are assembled from the businesses that come to mind when somebody says the name of an industry, and coming to mind is not one of the things the market prices.

One more clarification before the four dimensions arrive. A peer set is not a claim that the subject deserves the peer price. The claim is only that the peer prices are the relevant evidence. Whether those prices are believed, and what they say once they are, is a separate question. Choosing the peer set is choosing which question gets asked, and everything downstream inherits the choice.

Why four dimensions, and why these four?

Because of what a multiple is made of, settled earlier in this sequence. A multiple is a whole valuation folded into a single number. When somebody pays a price for a business, that price already holds their view of how fast the cash will rise, what a rupee of new capital put into the business will earn, and how likely the whole arrangement is to disappoint. Divide the price by one year of profit and none of those three can be read back out, but every one of them is still in there.

So the comparability test writes itself. Two companies deserve the same multiple when those three things are the same for both. The trouble is that not one of the three can be observed directly. Nobody publishes a line called expected growth in perpetuity, and nobody publishes their own risk. Companies publish accounts, and markets publish a traded price. So the work proceeds on proxies: things that can be seen and that move with the things that cannot.

The four dimensions come from that gap. Business mixWhat the revenue is made of, here valves, castings and aftermarket service. stands in for what the cash flow is made of and therefore how durable it is. Size stands in for a cluster of practical differences that have little to do with pricing directly. Forecast revenue growthThe rate revenue is expected to rise at over the near term. stands in for growth, the only one of the three with a near-direct observable. And capital intensityHow much invested capital it takes to produce a rupee of profit. alongside leverage stands in for the remaining two at once: what growth costs to buy, and how much risk is stacked behind the same operating cash. Koller, Goedhart and Wessels put growth, return on invested capital and value into one expression, and these four are that expression turned outward into things a reader can look up.

The four dimensions are proxies rather than causes, and a proxy is checked rather than trusted. The proxies are imperfect stand-ins for quantities nobody publishes, so a peer matched on all four can still deserve a different multiple. Imperfect proxies are not a defect in the method. The method is being honest about what it can and cannot see.

What does business mix change, and why is margin its visible face?

Business mix names the parts revenue is actually made of, and it is the dimension people skip because it takes reading rather than sorting. Sankalp Industrial Systems Limited, the manufacturer this subject area uses throughout, runs three lines. Industrial valves bring in Rs 6,00,00,00,000 of revenue at a 23.0 per cent margin on earnings before interest, tax, depreciation and amortisation (EBITDA). Precision castings bring in Rs 4,20,00,00,000 at 25.0 per cent. Aftermarket parts and service bring in Rs 1,80,00,00,000 at 30.0 per cent. The three lines add to Rs 12,00,00,00,000 of revenue, exactly the consolidated figure.

The three lines behave differently in ways that matter to anybody pricing the whole. Valve revenue arrives when customers build things. Castings revenue arrives when other manufacturers order. Aftermarket revenue arrives because machines already installed keep wearing out, a different kind of promise entirely. A company that is one-third aftermarket is priced differently from an otherwise identical company that is one-tenth aftermarket, and no amount of matching on revenue will discover that.

A business whose revenue mix has shifted almost always shows it in the profit line before it shows it anywhere else, so margin is the visible face of mix. Margin is not a perfect reading. Two companies at the same margin can get there by opposite routes, one through pricing power and one through squeezing costs, and those two are not priced the same. But margin is published, mix often is not in usable detail, and a four point difference in margin across two companies in the same line of work is a signal to go and read rather than a number to average away.

Across the six invented peers the margins run 20.0, 21.0, 23.0, 24.5, 26.0 and 30.0 per cent, in peer number order, and Sankalp Industrial Systems Limited sits at 24.0 per cent, between peer 3 Kaimur Industrial Limited and peer 4 Girnar Precision Limited. Hold that ordering in mind. The same ordering reappears three more times.

What does size actually change about a business?

A great deal, and almost none of it is what people assume. A Rs 5,00,00,00,000 supplier cannot underwrite a contract that a Rs 15,00,00,00,000 supplier can, so size changes which customers a business can serve. Size changes the borrowing that can be arranged and what it costs. Size changes how many separate things can go wrong before the whole business is in trouble. Size changes whether losing one customer is an inconvenience or an emergency. A household running on one salary against a household running on three makes the same point: the second is not richer per person, it is simply harder to knock over.

All of that is real, and every bit of it argues that size belongs in the comparability test. But the argument repays careful reading. Size settles which business is under examination, not the price paid per rupee of profit. A very small supplier and a very large one may be doing different work with different customers under different risks, in which case they fail the other three dimensions too and would be excluded on those. Size is how the difference is noticed; the other three are why it is acted on.

Size is the screening dimension, not the pricing dimension, and confusing those two is the most common structural error in this whole method. The six invented peers make the point without any argument at all: ordered by revenue they read peer 6, peer 5, peer 1, peer 3, peer 4, peer 2, and their multiples read 13.8, 8.7, 6.6, 7.6, 8.0 and 7.1 times. The relationship is not weak. There is no relationship at all. The smallest company in the set carries the highest multiple and the largest carries close to the lowest.

Why does the multiple follow growth more closely than anything else?

Because growth is the one thing inside a valuation that has a near-direct observable, and because of where it sits in the arithmetic. Two companies with the same profit today and different growth rates are two different streams of cash, and every extra year of separation widens the gap. The valuation compresses all of that into the multiple, so a difference in expected growth arrives in the multiple more completely than a difference in almost anything else.

Look at the two ends of the invented set. Peer 1 Aravalli Flow Controls Limited is expected to grow revenue at 4.0 per cent and trades at 6.6 times EBITDA. Peer 5 Shivalik Systems Limited is expected to grow at 9.5 per cent and trades at 8.7 times. Everything else about them is closer than that gap suggests: both make industrial equipment, both carry ordinary balance sheets, both sit in the Rs 8,00,00,00,000 to Rs 9,00,00,00,000 revenue range. The 2.1 turns between them is very largely a difference of view about the next several years.

Growth is also the dimension where a small error does the most damage, and a growth error has to be controlled in the peer set rather than corrected later. A set whose average expected growth is three points above the subject's cannot be repaired by any amount of care with the arithmetic afterwards. The borrowed multiple carries the peers' growth assumption, not the subject's, and there is no line in the calculation where it can be taken back out. The compression argument is doing its work here, and it is why the choice of peers matters more than any other choice in this method.

How do capital intensity and leverage carry the risk?

Return on invested capitalOperating profit after tax against the capital employed to earn it. answers a question that growth alone cannot: what does the growth cost? Two companies both growing at 8.0 per cent are not equally valuable if one of them has to put back thirty paise of every rupee it earns to get there and the other has to put back sixty. The first is producing cash the owner can keep; the second is producing a construction programme. A street vendor whose only equipment is a cart can double her stall with one more cart. A workshop with a press line cannot double without a second press line, and the press line has to be paid for first.

Across the invented peers the returns on invested capital run 11.0, 12.5, 14.0, 15.5, 17.0 and 24.0 per cent, in peer number order once again, with Sankalp Industrial Systems Limited at 15.0 per cent, between peer 3 and peer 4. The ordering is identical to the margin ordering and the growth ordering. A constructed set built to hold a lesson looks exactly like that.

Leverage is the last piece, and it is the one readers most often think belongs somewhere else. Net debt to EBITDABorrowings less cash against a year of EBITDA, a measure of leverage. across the six runs 1.8, 2.2, 1.5, 1.2 and 0.6 times for peers 1 to 5, and peer 6 Nallamala Components Limited carries net cashCash exceeding borrowings, so the figure is negative leverage. of 0.4 times EBITDA, alone in the set. Sankalp Industrial Systems Limited sits at 1.67 times, between peer 3 at 1.5 and peer 1 at 1.8.

Leverage belongs in a comparability test because it changes the risk sitting behind identical operating cash, and risk is a third of what any multiple compresses. Two companies producing exactly the same Rs 2,88,00,00,000 of EBITDA are not equally safe if one of them owes twice as much as the other. Sorted from least borrowed to most, the six read peer 6, peer 5, peer 4, peer 3, peer 1, peer 2, and their multiples read 13.8, 8.7, 8.0, 7.6, 6.6 and 7.1 times: falling steadily, with exactly one step out of five going the wrong way, between peer 1 and peer 2. The inversion is named rather than smoothed away. A constructed set that ran perfectly on all four dimensions would teach something false about how real sets behave.

Do the six invented peers actually line up?

The six do line up, and the shape of that alignment is the most useful single fact about the set. Ordering the six by EBITDA margin gives peers 1, 2, 3, 4, 5, 6. Ordering them by forecast revenue growth gives peers 1, 2, 3, 4, 5, 6. Ordering them by return on invested capital gives peers 1, 2, 3, 4, 5, 6. In all three cases the multiple order is the same order: 6.6, 7.1, 7.6, 8.0, 8.7 and 13.8 times, rising without a single step backwards.

A word of caution belongs right here, before the picture makes the point look stronger than it is. The six companies were constructed and their figures were chosen. A real set of six industrial manufacturers would not line up this neatly on three dimensions at once, and a real set that did would invite suspicion of the sample before belief in the finding. The constructed set does not demonstrate that peers always behave this way. The set demonstrates what an analyst hopes to see, and therefore what a set that fails to show it is reporting.

A well-built peer set looks like a gradientAn ordering in which one measure rises steadily with another. rather than a cloud, and a set that looks like a cloud is usually reporting that it was matched on the wrong thing. That is the diagnostic use of this whole exercise. When the peers are plotted against growth and the points scatter with no shape at all, the honest reading is that whatever the list was sorted on is not what the market prices.

Which dimension does the multiple refuse to follow?

Size, completely. Setting the four orderings out together is worth the space. Seen beside one another they stop the argument being a matter of opinion.

Sorted by EBITDA margin, ascending, the six read peers 1, 2, 3, 4, 5, 6, and their multiples read 6.6, 7.1, 7.6, 8.0, 8.7, 13.8 times: five steps, none of them backwards. Sorted by forecast revenue growth, ascending, the order is identical and so is the result. Sorted by revenue, ascending, the six read peers 6, 5, 1, 3, 4, 2, and the multiples read 13.8, 8.7, 6.6, 7.6, 8.0, 7.1 times: three of the five steps break the run. Sorted by leverage, least borrowed first, the six read peers 6, 5, 4, 3, 1, 2, and the multiples read 13.8, 8.7, 8.0, 7.6, 6.6, 7.1 times: falling in step, with one break in five, between peer 1 and peer 2.

Three of the four dimensions reproduce the multiple order exactly and the fourth destroys it. And notice what kind of fact it is. The finding does not say size is unimportant. The finding says size is not what is being priced. A set assembled on size has been matched on the dimension carrying the least pricing information and left unmatched on the three carrying the most. Because the arithmetic never sees the dimensions at all, nothing in it afterwards will ever reveal the error.

Where does the subject sit on each of the four?

In four different places. Asking about all four at once is the point. Sankalp Industrial Systems Limited earns Rs 12,00,00,00,000 of revenue at a 24.0 per cent EBITDA margin, giving EBITDA of Rs 2,88,00,00,000. Its forecast revenue growth is 10.0 per cent. Its return on invested capital is 15.0 per cent. Its net debt runs at 1.67 times EBITDA. Its shares are priced so that the traded enterprise value is Rs 22,40,00,00,000. Against that EBITDA the multiple is 7.78 times.

Now place it. On margin, 24.0 per cent sits between peer 3 at 23.0 and peer 4 at 24.5. On return on invested capital, 15.0 per cent sits between peer 3 at 14.0 and peer 4 at 15.5. On revenue, Rs 12,00,00,00,000 sits between peer 3 at Rs 11,00,00,00,000 and peer 4 at Rs 13,50,00,00,000. On leverage, 1.67 times sits between peer 3 at 1.5 and peer 1 at 1.8. And its multiple of 7.78 times sits between peer 3 at 7.6 and peer 4 at 8.0.

Four of those five readings agree with each other beautifully. The fifth does not. On forecast revenue growth, 10.0 per cent sits above peer 5 at 9.5 per cent and below peer 6 at 18.0 per cent, a materially higher position than anywhere else it lands. Its position on growth and its position on the multiple do not match, and that mismatch is a question rather than a finding.

Care matters here. The conclusion that suggests itself at this point is not available. A company sitting below where one dimension alone would place it is either being priced for something the multiple does not capture, or it is not, and a comparison across five ratios cannot separate those two situations. Perhaps the market doubts the growth. Perhaps the growth is expected to cost more capital than the peers' growth costs. Perhaps the mix is read as less durable. Perhaps nothing at all, and the reading is noise inside a set of six invented companies. The comparison has found a place to go and read, and the honest output of the whole exercise is a question with the assumptions named beside it, not a conclusion about the price.

Is the sixth peer a comparable company or not?

The four dimensions lead to one real argument rather than a teaching prop with an obvious answer. Peer 6 Nallamala Components Limited trades at 13.8 times EBITDA against a set that otherwise runs from 6.6 to 8.7 times. Everything about it is at the far end of every dimension.

State the differences precisely. A loose sentence about the size comparison in particular says something false. Its revenue is Rs 5,00,00,00,000. Against Sankalp Industrial Systems Limited's Rs 12,00,00,00,000 that is 41.67 per cent. Against the smallest of the other five peers, peer 5 Shivalik Systems Limited at Rs 8,00,00,00,000, it is 62.50 per cent. The two ratios run against two different bases, and a sentence quoting one without naming its base is not a rounding error but a wrong statement. Its forecast revenue growth of 18.0 per cent is nearly twice peer 5's 9.5 per cent and four and a half times peer 1's 4.0 per cent. Its 30.0 per cent EBITDA margin is four points above the next highest, peer 5 at 26.0, and ten points above the lowest, peer 1 at 20.0. And it carries net cash of 0.4 times EBITDA, alone in a set where everybody else borrows.

Answer one: it is not a comparable company

A peer set exists to hold the four dimensions roughly constant. Only then does the multiple compare like with like. Peer 6 holds none of them. Peer 6 is a fraction of the subject's size, it grows at a rate the subject does not approach, it earns a return on capital nine points higher, and its balance sheet runs the opposite way round. The company fails every one of the four tests.

And the consequence is not abstract. Its 13.8 times is a price for a growth and return profile that Sankalp Industrial Systems Limited does not have. Importing that multiple imports an assumption about the subject that nothing in the subject's own figures supports. The arithmetic backs the concern: with peer 6 in the set the mean is 8.63 times, and without it the mean is 7.60 times. More than a full turn of enterprise value per rupee of EBITDA, produced by one observation in six.

Answer two: it is the end of a gradient, not a point off one

The four dimensions are continuous rather than categorical, and peer 6 does not sit off the pattern the other five trace. Peer 6 sits at the end of it. Every dimension moves in exactly the direction it moves for peers 1 through 5, only further: higher margin, higher growth, higher return on capital, less borrowing, higher multiple. There is no discontinuity anywhere in the sequence. If the five make a line, the sixth is on the line.

There is a harder version of this argument and it deserves stating plainly. Excluding a company because its multiple is high is choosing the answer before the work is done. If the top of the range goes on the ground that it is far from the subject, consistency asks why the bottom stays: peer 1 is forecast to grow at 4.0 per cent against the subject's 10.0 per cent, a gap of six points, as large a departure on that dimension as several plausible exclusions. A rule that removes the observation pulling the answer up and keeps the observation pulling it down is not a rule about comparability at all.

How much is the argument actually worth?

Far less than it feels like, on the statistic that gets used, and this is the deflating fact that ought to travel with the whole dispute. Drop peer 6 from the set entirely and the median falls from 7.8 times to 7.6 times. The median moves by 2.56 per cent. On the same removal the mean falls from 8.63 times to 7.60 times, a move of 11.97 per cent.

Precision matters here more than it looks. The mean of the six is 8.633333 and is printed above as 8.63, but every figure worked out from it is worked out on the unrounded number. The mean move is therefore 11.97 per cent and not 11.94. Taking 1.03 off a printed 8.63 gives 11.94, and taking 1.033333 off 8.633333 gives 11.97. The difference sounds trivial and is not. The same rounding, carried into an enterprise value, is worth Rs 96,00,000.

One more thing falls out of the removal, and it looks like a coincidence without quite being one. On the five remaining companies the median is 7.6 times and the mean is exactly 7.60 times. The two figures coincide, and the coincidence is the cleanest possible demonstration of what the outlier was doing to the average in the first place. A median counts positions and gives every company one vote; a mean counts distances and gives the company furthest away the loudest one. Take the far company out and the two measures, more than a turn apart before, land on the same figure.

What is the treatment of a peer that cannot be decided?

Three things, and none of them is a judgement about peer 6. The treatment of an undecidable name is a matter of craft rather than opinion, and on that there is a rule.

First, state which answer was taken. Not in a footnote and not in a cell comment: in the same place the statistic appears. Anybody reading the output then knows a decision was made and knows which way it went. A peer set with an unstated exclusion is a peer set nobody can review.

Second, give a reason that is not about the company's multiple. Giving a reason away from the multiple is the discipline that separates a judgement from a preference. Excluded because its forecast revenue growth is 18.0 per cent against a set otherwise running 4.0 to 9.5 is a reason. Excluded because 13.8 times looked wrong is the answer wearing a reason's clothes. The test is simple: with the multiple never seen, would the reason still stand? If not, it is not a reason.

Third, report the statistic both ways. Median 7.8 times with peer 6 in, 7.6 times with it out. Mean 8.63 times with it in, 7.60 times with it out. Reporting both is what converts an unsettled argument into information the reader can use, and it costs one extra line. The reader who disagrees with the call can now act on their own view without rebuilding the analysis, and the reader who agrees has learned how much the call was worth. On the median here that is a quarter of a turn.