Precedent Transactions: What Buyers Have Actually Paid

A precedent transaction analysis asks one question: what have buyers actually paid for businesses like this one when control changed hands? Five completed deals in the same industrial segment, all invented, give a median of 9.5 times and a mean of 9.70 times. Applied to the Rs 2,88,00,00,000 of earnings before interest, tax, depreciation and amortisation (EBITDA) that Sankalp Industrial Systems Limited earned in Year 0, the median indicates Rs 27,36,00,00,000, and the analysis stops there.

The everyday version of this can be checked in an afternoon by anybody. Suppose the question is what a two bedroom flat in one particular colony is worth. There are two places to look. The first is the boards outside the property agents' shops, where flats are advertised at asking prices that get rewritten every few weeks and that nobody has yet paid. The second is the registry, where four flats in that colony actually changed hands last year: money moved, keys moved, and the price is on record. The second list is shorter, older and far harder to get hold of. Everybody who has ever bought a flat knows it is the better list.

A precedent transaction analysis is that same registry, applied to companies instead of flats. A precedent transactionA completed deal in which control of a comparable business changed hands. is a registry entry: a business changed hands, a buyer committed money it could not take back, and the price is a fact rather than a quote. The observations are real, and there are almost never enough of them. The strength of the method and every difficulty inside it come from that one source.

Everything below runs on one set of five invented deals and one invented company. Sankalp Industrial Systems Limited is a listed manufacturer of industrial valves, precision castings and the aftermarket parts and service that go with them. In Year 0, its last completed year, it earned revenue of Rs 12,00,00,00,000 and EBITDA of Rs 2,88,00,00,000. The EBITDA figure is the only thing about Sankalp the worked example needs, and it is held still throughout.

What does a precedent transaction analysis actually ask?

One question, and it is worth stating it in the narrowest form it will take: what have buyers actually paid for businesses like this one, when control of the whole business changed hands? Three restrictions are packed into that sentence. Buyers, not sellers and not observers. Actually paid, not offered and not discussed. Control of the whole business, not a parcel of shares in it. Strip any one of those away and the question being asked is a different one, and the number that comes back is a different number.

The method itself has three steps and none of them is complicated. Collect completed transactions in which control of a comparable business changed hands. Express each price the same way, as an enterprise value over that target's own EBITDA. Businesses of different sizes can then be laid beside one another. Then take a middle measure of the resulting multiples and apply it to the company being valued. The arithmetic is the easy half, and the whole craft of the method sits in deciding which deals are allowed into the set.

Why express the prices as multiples at all, rather than simply reporting what was paid? Because the rupee prices are not comparable with one another. In this set the five prices run from Rs 9,80,00,00,000 to Rs 34,65,00,00,000, so the largest is 3.54 times the smallest. The multiples run from 8.6 times to 11.4 times, so the highest is 1.33 times the lowest. Dividing each price by the earnings it bought strips out most of the difference in size and leaves behind the part that can travel from one business to another.

And it is worth being exact about what comes out at the far end. The output is an indicated valueThe output of applying a multiple from a set to the company being valued. It indicates; it does not offer and it does not conclude.. Nobody has made an offer, so an indicated value is not one. A handful of completed prices cannot support a conclusion as strong as this company is worth that figure, so an indicated value is not that either. An indicated value answers the narrow question above, carried across to a company that has not been sold, and it should be read with the question still attached.

Which deals belong in the set, and which do not?

There is a door, and then there are four tests. The door is the definition itself: control of the whole business changed hands. A purchase of a parcel of shares, however large the rupee amount attached to it, has not come through that door, and no adjustment made afterwards will turn its price into something comparable with a price paid for a whole company.

Test one is business mixWhat proportions of a company's revenue come from which kinds of activity: manufacturing, service, distribution and so on.. Does the target earn its revenue from the same kinds of activity, in roughly the same proportions, as the company being valued? Test two is size. Is it a comparable scale of business, or is it so much larger or smaller that the buyers involved and the reasons for buying are of a different kind altogether? Test three is completionThe point at which a transaction actually happened and money moved, as against being announced or agreed.. Did the transaction actually happen, with the money moving? Test four is recency. Was the price struck recently enough that it still describes conditions which hold now?

Each of the four is pass or fail rather than a matter of degree, and a deal that fails any one of them is answering a different question, so averaging it in mixes two questions into a single number. The discipline is harder than it sounds. A failing deal is very often the most interesting one in the pile, and the temptation is to include it with a footnote attached. A footnote does not survive into the median.

The three commonest candidates each fall at a different place on that diagram. A buyer acquires 30 per cent of a comparable business. Control did not change hands, so the deal never reaches test one and its price carries nothing at all about what control is worth. A deal is announced at a stated price and then abandoned. A price nobody paid is a price nobody paid, however firmly it was announced at the time, so the deal fails completion. A target in an adjacent segment selling to different customers fails business mix. The same argument arrives later about deal 5, turning up that time without anybody noticing it.

What do the five completed deals in this segment show?

Here is the set. Five completed transactions, all in the same industrial segment, each one a change of control of a whole business. Each deal is cited by its number, and the numbers are how the five deals are named from here on.

Read the two numeric columns against each other and the first useful thing about this set falls straight out: they do not line up. Deal 4 is the largest transaction in the set at Rs 34,65,00,00,000 and sits fourth by multiple. Deal 5 is the smallest at Rs 9,80,00,00,000 and sits first. Deal 1 is the second smallest and sits last of all. Size and multiple are two different facts about a transaction, and in this set the ordering by one is very nearly the reverse of the ordering by the other.

The second useful thing is the shape of the upper band. Four of the five deals sit close together at 8.6, 9.1, 9.5 and 9.9 times, a spread of 1.3 turns from the bottom of that group to the top of it. The fifth, deal 5, sits at 11.4 times, a turn and a half above the highest of the other four. One observation is further from its nearest neighbour than the whole of the rest of the set is wide. The gap drives almost everything that happens later, and it is the reason a median rather than a mean is the measure being carried forward.

The record carries one further fact about deal 4: it went to a second round with two bidders still in it. How a sale is conducted belongs to the conduct of a sale, covered separately. The second round is part of the record of the deal, and nothing whatever is built on it.

The dispersion in this set is the finding rather than a nuisance to be averaged away. Five buyers, in one segment, doing what looks from a distance like the same kind of thing, paid between 8.6 and 11.4 times. Each of them was buying a particular business at a particular moment for its own particular reasons, and the spread is a measurement of how much those particulars matter. A method that reports only the middle of that spread has thrown away the most honest thing it had.

What does the one financial buyer in the set establish?

Deal 1 is the only transaction in this set bought by a financial buyerA buyer acquiring for a return on its own capital over a holding period, rather than to run the business alongside another one., and it carries the lowest multiple in the set at 8.6 times. The record adds one detail: it bought the whole of the equity, with no cost saving underwritten and none claimed. The other four were bought by a strategic buyerA buyer already operating a business, buying for its own operating reasons..

Every reader who reaches that row does the same thing with it, and the move is worth catching in the act. The lowest multiple sits beside the only financial buyer, and a rule assembles itself: financial buyers pay less. The rule is tidy, it is intuitive, and this set cannot support it. One observation is a row in a table and not a finding, and saying so out loud is part of reporting the set honestly.

Making that claim properly would take several things. Several financial buyers, first, so there is something to average. Something to compare that average against, second. And third, some way of holding everything else still. Deal 1 differs from the other four in more than one respect: a different target, a different moment, a different set of reasons for buying. With one observation there is nothing to average, nothing to compare, and no way at all to separate the buyer type from everything else that was different about that transaction.

None of that makes the buyer type uninteresting. Buyer type is one of the more interesting things about a set of precedents, and the comparison between what a buyer already running a business in the segment pays and what a buyer acquiring for a return on its own capital pays is covered separately, where it can be given the space and the observations it needs. Deal 1 is recorded, labelled as one observation, and nothing whatever is built on it.

Why does the smallest deal carry the highest multiple?

Deal 5 is Tapti Service Partners Private Limited at 11.4 times, the highest multiple in the set and the smallest transaction in it at Rs 9,80,00,00,000. Deal 5 is also the only one of the five whose business is a genuinely different shape. Tapti is a pure aftermarket service business: it does not manufacture, and its revenue comes from servicing and supplying parts for equipment somebody else built.

Now put that beside the company being valued. Sankalp Industrial Systems Limited runs three divisions. Division 1, industrial valves, earns Rs 6,00,00,00,000 of the Rs 12,00,00,00,000 of revenue, being 50.0 per cent. Division 2, precision castings, earns Rs 4,20,00,00,000, being 35.0 per cent. Division 3, aftermarket parts and service, earns Rs 1,80,00,00,000, being 15.0 per cent, at a 30.0 per cent EBITDA margin. So Sankalp does have an aftermarket business, it is the best margin in the company, and it is one division out of three carrying fifteen rupees in every hundred of revenue.

Here is the everyday version of the same question. A workshop that sells and fits new tyres and a workshop that only does puncture repairs and wheel balancing are both in the tyre business, and nobody who has run either would price them the same way. Repeat customers, the cash cycle and how much of the work is booked in advance are all different. Taking the price paid for the second and using it to price the first is not an error of arithmetic; it quietly answers a different question.

So does deal 5 belong? There are two answers and each of them is respectable. For keeping it: it is a completed change of control in the same industrial segment, it came through the door and past the other three tests, and dropping observations because they are inconvenient is exactly how a set gets bent to fit an answer somebody already had. For dropping it: business mix is test one, a pure service business is not the mix of a company that earns 85 per cent of its revenue from valves and castings, and a test waived once is not a test.

Whether deal 5 belongs is a genuine question with two defensible answers, and ruling on it would hide a choice that belongs to the reader. The cost of each choice can be reported without ruling on it. Drop deal 5 and the median falls from 9.50 times to 9.30. The mean falls from 9.700 to 9.275. On Sankalp's Year 0 EBITDA the indicated enterprise value falls from Rs 27,36,00,00,000 to Rs 26,78,40,00,000. Anybody who is handed those two figures can make the decision for themselves.

Median or mean, and what separates the two here?

The medianThe middle observation once the values are put in order. It does not move when an extreme value is added or removed. of the five multiples is 9.5 times, which is deal 3, sitting third of five once they are in order. The meanThe arithmetic average. Every observation contributes to it, in proportion to how far from the middle it sits. is 9.70 times, being 48.5 divided by five. The gap between the two is 0.20 turns. On Sankalp's Year 0 EBITDA of Rs 2,88,00,00,000 the gap is Rs 57,60,00,000, the entire difference the choice of measure makes here, and it comes from one observation.

Where that comes from is visible in the spacing. The four lower multiples step up gently: 8.6 to 9.1 is half a turn, 9.1 to 9.5 is four tenths, 9.5 to 9.9 is four tenths. Then 9.9 to 11.4 is a turn and a half. A mean feels every one of those distances, so the one long step at the top pulls the average upward. A median never leaves its seat: it counts places rather than distances, and there are still two observations above it and two below it whatever the top one does.

The cleanest demonstration of the difference is not the one most people reach for. Take deal 3 out of the set, the median observation itself, and watch what happens to the median. Nothing happens to it. The four remaining multiples are 8.6, 9.1, 9.9 and 11.4. The middle pair, 9.1 and 9.9, average to 9.50, exactly where the median was before. The mean does move, from 9.700 to 9.750. Every observation contributes to a mean, and one of them has just left.

A median is a position in an ordering while a mean is a total shared out, and that single structural difference decides everything about how each behaves on a set this small. Neither is more correct in general. On a set with one observation sitting a turn and a half above the rest, the median is the measure that describes where the bulk of the set actually is, and that is why it is the one carried forward here. A reader shown only one measure cannot tell whether the two agree, so the mean is reported alongside the median rather than hidden.

What does the median indicate for the company being valued?

One multiplication. The median of the set is 9.5 times. Sankalp Industrial Systems Limited earned EBITDA of Rs 2,88,00,00,000 in Year 0, its last completed year. Multiply the two and the analysis indicates an enterprise value of Rs 27,36,00,00,000. The indicated enterprise value is the output, and there is nothing else to compute.

Two details in that sentence are doing more work than they look like they are. The first is Year 0. The earnings figure the multiple is applied to is the last completed year, not a forecast. Every multiple in the set was struck against the target's own earnings at the time, and mixing a completed year with a forecast year would compare two different things. The same Rs 2,88,00,00,000 is used in every state of the set below. When a figure moves, the multiple moved and the earnings did not.

The second is the word indicates. The multiplication is the easiest step and the choice of what to multiply by is the hardest. An indicated value rather than an answer comes out for that reason alone. Had the mean been carried forward instead, the same EBITDA would have indicated Rs 27,93,60,00,000. Nothing except the choice of measure created that Rs 57,60,00,000, and both figures come from the same five deals.

A single figure hides how wide the underlying evidence actually is. Take the lowest multiple in the set and Sankalp is indicated at Rs 24,76,80,00,000. Take the highest and it is indicated at Rs 32,83,20,00,000. The spread is Rs 8,06,40,00,000 across five completed transactions in one segment. Nothing about the median makes that spread go away; the median simply chooses a point inside it, and reporting the point without the spread tells a reader far less than they need.

How much does one deal move the answer when there are only five?

Everybody accepts in the abstract that five observations is thin. Almost nobody has a feel for how thin. The way to get one is to take each deal out of the set in turn and look at what the two measures do. The arithmetic takes five minutes and settles the question permanently. The result is set fragilityHow much the answer moves when a single observation is added to or removed from a small set., measured rather than asserted.

Two things come out of that table and both are worth carrying away. The first is that the median moves by at most 0.20 turns in either direction, running from 9.30 to 9.70 across all six states. The mean runs from 9.275 to 9.975. The median's whole band is 0.40 turns wide and the mean's is 0.70 turns wide. Measured as the largest single move away from the full set, the median shifts 0.20 turns at most, being Rs 57,60,00,000 on Sankalp. The mean shifts 0.425 turns, being Rs 1,22,40,00,000. Losing one observation out of five moves the mean more than twice as far as it moves the median, and the whole practical case for reporting a median on a set this size rests on that.

The second is the row for deal 3. Remove the median observation itself and the median does not move by anything at all, the cleanest possible demonstration of a median at work. The mean over the same removal moves from 9.700 to 9.750. A mean has no idea which observation was in the middle.

One arithmetic warning goes with that table. The table is the exact place a careful reader gets caught. Take the mean's move on removing deal 5 from the printed two decimal figures, 9.70 less 9.28, and the result is 0.42 turns. On Rs 2,88,00,00,000 of EBITDA that works out at Rs 1,20,96,00,000. Take it from the full values, 9.700 less 9.275, and it is 0.425 turns, being Rs 1,22,40,00,000. The two answers differ by Rs 1,44,00,000 for no reason except that the first one rounded twice. Rounding happens once, at the end, and one figure is never rebuilt out of another figure's printed form. The mean column above is printed to three decimal places for exactly this reason: on four or five observations each carrying one decimal, three decimals is exact rather than a false precision.

What does a completed price carry that a screen price does not?

Come back to the flat and the registry for a moment. The answer is the same in both places. A price on a screen and a price actually paid for control are not two measurements of one thing. The two prices are different things, and three differences between them matter enough to name.

The first is what was bought. A screen price is the price of a small parcel of shares. The parcel buys a share of the profits and a vote that changes nothing. A completed transaction bought the right to decide: who runs the business, what it does next, what it borrows, whether it is sold again. Nobody trading a hundred shares is paying for any of that, so a screen price cannot contain what it is worth.

The second is whose reasons are inside it. A screen price is set by whoever traded last, for reasons nobody records, and it is remade continuously all day. A completed transaction was struck by one identified buyer, at one moment, for its own purpose, after that buyer had looked at the business properly. The observation is richer and narrower at the same time: it says a great deal about one buyer, and nothing at all about whether the next one would think the same way.

The third is commitment. Money moved and it cannot be moved back. A screen price is an offer, and an offer can be withdrawn in a second by anybody who changes their mind. A completed price is the only kind of price about which nobody can later say they did not really mean it, and that is the entire reason a precedent set is worth assembling at all.

Every one of those three strengths arrives attached to a cost, and the costs are the reason the rest of this guide has been so careful. Completed transactions are scarce, so a set is small and one observation moves it. Each was struck at its own moment, so a set is a collection of snapshots rather than a picture of now. And each has one buyer's reasons baked into it. Baked-in reasons are exactly what was wanted and also exactly what cannot be separated out afterwards.

Why is a price struck at one moment not a price today?

Because a completed transaction is a photograph and not a live reading. The buyer in deal 3 decided what it was willing to pay under the conditions in front of it on the day it decided: what borrowing cost, what the order book looked like, what it believed about the next few years. None of those conditions is bound to hold now, and none of them is visible in the multiple. StalenessThe extent to which a price struck at an earlier moment no longer describes the conditions in front of a buyer today. is the risk that a set is silently reporting a world that has moved on.

Staleness needs no finance at all to see. A neighbour mentions what they paid for a flat in the same colony, and the first question anybody asks in reply is when. The same rupee figure means completely different things if the answer is last month or seven years ago, and nobody needs to be taught that. The instinct is the same one, and in a set of precedents it is the fourth test.

A precedent multiple has a date attached to it whether or not anybody prints the date, so a set that does not say when each price was struck is not a set a reader can judge. A set that runs in Year 0 and forward, as this one does, carries no calendar date against any deal, so recency can be named as a test but never applied. Where a real reader would find the date is in the public record. For a listed company in India, what must be disclosed about a change of control and when is a matter for the Securities and Exchange Board of India at sebi.gov.in; a company's own filings and its shareholding sit with the Ministry of Corporate Affairs at mca.gov.in. The date is findable; it must never be assumed.

Where does a precedent analysis stop?

A precedent analysis stops at Rs 27,36,00,00,000. The indicated value is the whole output. The analysis does not say the figure is right. The analysis does not say Sankalp Industrial Systems Limited is worth that. The analysis compares the figure with nothing, and it draws no conclusion from any comparison it has not made. Knowing where a method stops is part of knowing the method, and this one stops at a single indicated value with its assumptions attached.

Two questions naturally follow and neither is answered here. The first is how this figure sits against what similar listed businesses trade at on a screen, and what any difference between the two would mean. Precedent multiples do tend to sit above trading multiples, and there are respectable explanations for why. Setting the two side by side, and working through how much a gap between them can and cannot tell a reader, is a subject of its own and is covered separately. The second is whether the kind of buyer changes what gets paid. Answering that needs more observations than one financial buyer in a set of five, and it too is covered separately.

One common habit is worth naming plainly and refusing. Where a buyer pays more than a traded price, two causes are usually named: the buyer was acquiring control, and the buyer expected benefits from putting the two businesses together. Both are real causes. But a buyer signs one price and not two, so no record supports splitting the extra payment into a fixed share for one and a fixed share for the other, and no published deal record anywhere carries that split. The two can be named as causes; they can never be presented as portions of a number.