Thesis Drivers: The Few Variables the View Depends On

A thesis driver is one of the two or three variables that decide whether a view turns out right. Two things together make a variable one: moving it moves the answer materially, and nobody yet knows how it will behave. The worked exercise below rests on three. A view resting on twelve variables is usually resting on none of them.

How to move one assumption and watch the result, and how a forecast gets built in the order it gets built, are both already settled. One question those two skills leave hanging is this: out of the forty or fifty numbers now sitting in a model, which handful is actually carrying the view. The ranking below takes a worked exercise, tests its candidates in the open, throws most of them out, and ends with three variables and a short list of things to watch. Two ideas sit underneath the work: a view is a small set of claims rather than a general impression, and a load bearing input is found by moving it rather than by arguing about it.

So what makes a variable a driver?

Start away from finance. A household is deciding whether it can afford a flat. Dozens of numbers are involved: the stamp duty, the society maintenance, the cost of a new geyser, the interest rate, whether the elder son gets into a government college or a private one. Ask which of those decide the answer and the list collapses fast. The geyser is Rs 12,000/- once. The maintenance is known and small. Two of them decide it: the interest rate and the college fee. Each is large enough to change the answer, and neither has been settled yet.

The whole test has two halves. A variable is a driver only when moving it moves the answer materially and its behaviour is genuinely unresolved, and a variable that fails either half is not a driver however important it feels. Both halves are doing real work, and each of them kills a different kind of candidate.

Fail the first half and what remains is a variable nobody can predict that barely touches the outcome. Whether the geyser costs Rs 12,000/- or Rs 18,000/- is genuinely unknown and genuinely irrelevant. Watching it is a way of feeling busy.

A candidate that fails the second half is more dangerous, and the danger is that it looks exactly like a driver. The line is large. The line moves the answer. The behaviour is simply already settled. The household knows its rent to the rupee. The rent is enormous, and there is nothing left to find out about it, so the rent is not a driver. In a model this shows up as the line somebody spent three days on, built beautifully, and which cannot change the conclusion no matter how it lands.

Why should the list be short?

The usual answer given for keeping the list short is tidiness, and the usual answer is wrong. Three variables are not better than twelve because three is neater. Three variables are better because of what happens six months later, when the shares have moved and somebody asks why.

A view resting on three variables gives that question an answer. There were three things to watch, all three have been watched, and the analyst can say which one moved and whether it moved as expected. The view has been checkable the whole time, and checkable by anybody, not only by its author.

A view resting on twelve gives the same question no answer at all. When the price moves there is no saying which of the twelve is speaking. Eleven of them moved a little too, and nothing separates them. The discipline is not about being neat, it is the difference between a claim somebody can monitor and a claim nobody can, including its author.

There is a second cost, and it is the more expensive one. A view with twelve supports can never be shown wrong. No single observation contradicts twelve things at once, so whatever arrives in the next quarter, some part of the list survives and the view survives with it. A view that always survives sounds like robustness. A view that always survives is the opposite. A claim that no observation can damage is a claim that no observation can improve either, so the person holding it learns nothing from a year of watching, whether the shares rise or fall.

How are they actually found?

By moving things, not by arguing about them. The method has three steps and takes about twenty minutes on a model already open.

Step one. Each candidate is taken on its own, moved by an amount stated out loud, and what happens to the answer is written down. The amount is always stated. A move of unstated size produces a number that cannot be compared with anything. Step two. The recorded effects are ranked. Step three, and this is the step people skip: every variable whose behaviour is already known is struck out, however high it ranked. The survivors of both steps are almost always two to four items, and they are seldom the rows that took longest to build.

Single variable sensitivity, dragging the gross margin and watching earnings per share respond, is covered under sensitivity, and moving a coherent set of variables together rather than one at a time is covered under scenario analysis. A ranking asks for something different, an act of comparison rather than of dialling. The arithmetic below is worked in the open, at a level anybody can repeat with a calculator.

Here is that ranking run on the worked exercise. Every move below is made on the published third year, where revenue was Rs 2,415 crore and gross profit Rs 1,111 crore, giving a gross margin of 46.0 per cent, and where profit after tax was Rs 278 crore for earnings per share of Rs 11.58/- on 24.00 crore shares. Each move is taken one at a time with everything else held; revenue moves carry the published gross margin and leave the costs below the gross line alone; and each pre tax figure is taken down to a per share number at the rate the accounts actually bore.

The tax rate is worth one sentence of its own. Careless work starts here. The effective tax rateThe tax charge sitting in the accounts divided by the profit before tax, so the rate the company actually bore rather than the rate written in the statute. here is Rs 93 crore of tax on Rs 371 crore of profit before taxThe profit line struck after operating costs, depreciation and interest but before any tax charge. The composition of the line is set out in the accounting material., which is 25.0674 per cent. The rate prints as 25.1 in a summary table, and 25.1 is not the figure to compute with. The two absolutes are divided directly instead. On a figure this size the difference is small; the habit is not.

Read the last column downward and the shape is unmistakable. The top of the list is worth six to ten times the bottom of it. One point of gross margin is Rs 24.15 crore of gross profit, about Rs 18.10 crore after tax, about Rs 0.75/- on 24.00 crore shares, or 6.5 per cent of the year's earnings per share. A tenth of other income is Rs 3.80 crore, about Rs 2.85 crore after tax, about Rs 0.12/- a share, roughly 1.0 per cent of earnings. The margin point is worth about six and a half times the other income move, and the tail of the list is flat enough that arguing about the order down there is wasted breath.

Notice the red bar. Employee cost ranks second, comfortably ahead of everything except the margin, and it is not a driver. Step three removes it: the headcount and the increment cycle are known at the start of the year, so however large the line is, there is nothing about it left to find out. Ranking high gets a variable into the second round. Ranking high does not get it through.

Why does a sensitivity need its year attached?

One more thing before the survivors. Quoting a ranking figure without its year is the most common way a good figure gets quoted into meaninglessness. Every number in the table above was computed on the published year. Move the base year and every number in it changes.

Take the worked exercise's own forward case. The forward case runs volume growth of 6.0 per cent with realisation adding another 3.0, giving revenue of Rs 2,415 crore times 1.06 times 1.03, or about Rs 2,636.70 crore. One point of gross margin on that year is Rs 26.37 crore rather than Rs 24.15 crore, and about Rs 0.82/- a share rather than about Rs 0.75/-. Same move, same company, same model, Rs 0.07/- apart, purely because of which year it was run on. A sensitivityThe change in an output for a stated change in one input, quoted together with the size of the move and the year it was computed on. quoted without its base year is not yet a number, and the first question to ask about any such figure is which year produced it.

The first driver: does gross margin hold at 46.0 per cent?

Gross margin clears the first half by a distance, as the table shows. The second half of the test is what earns it the top place. Over two years the gross margin rose 3.0 points, from 43.0 per cent to 46.0 per cent, and the published statements cannot say why. Three explanations fit the same evidence and nothing on the face of the accounts separates them.

The first is a pricing environment across the whole field that let every maker price ahead of its input costs. The second is Sarvani Coatings pricing better than its rivals. The third is a mix shiftA change in the proportions of what gets sold. Selling more of the dearer line lifts revenue per unit and cost per unit at the same time. towards industrial work, which lifts realisationRevenue divided by the units sold, so what the seller actually got for each tin. Realisation stands in for price wherever price itself is not disclosed. and input intensity together and can look exactly like better pricing.

The field evidence narrows this and then refuses to settle it. Across that identical two year window the two peers moved as well: 2.4 points at Nandivarman Paints Limited, 3.6 at Kesaria Surface Solutions Limited, set against 3.0 at Sarvani Coatings. All three rose, and a shared condition explains that pattern well. Sarvani Coatings sits neatly between the two peers, so no figure in the field evidence separates its own pricing from its mix. The period on all three of those figures matters: they are two year moves, and the one year move from the second to the third year is 2.0 points, from 44.0 to 46.0 per cent. Two year and one year figures may not be set beside each other, and the per unit work belongs to the one year window.

Gross margin is the first driver because it moves the answer more than anything else in the model and because, after all the evidence that exists has been read, it remains an open question. Every driver has that shape: the large number that nobody can close.

The second driver: does volume growth stay above the field's 4.5 per cent?

Volume looks smaller on the ranking, and it is: one point of volume growth is about Rs 22.78 crore of revenue and about Rs 0.33/- a share on the published year. Volume is a driver anyway, for a reason the ranking cannot show.

The margin question and the volume question are different claims about different things, and they can move in opposite directions. Sarvani Coatings grew volume 6.0 per cent against a field at 4.5 per cent, so it was 1.5 percentage points ahead. The 1.5 point gap decides whether the company is growing with the field or ahead of it. A firm can hold its margin beautifully while its volume slips back to the field's rate, and the two together produce a very different year from either one alone. Volume qualifies as a driver because whether the company grows ahead of its field is a separate claim from whether it keeps its margin, and neither answers the other.

The second half of the test is satisfied without argument. Nobody knows next year's volume growth, for the field or for the company, and the record shows how wide the range is: field volume growth over five years ran 2.1, 6.8, minus 1.4, 7.2 and 4.5 per cent. One of those five years was a contraction inside a growing field. Anybody willing to state next year's number to a decimal is guessing in a suit.

The third driver: does the 0.13 point share gain repeat?

The share gain is the one that gets described badly, so its size has to be attached every single time. Sarvani Coatings grew revenue 13.9 per cent against a field at 11.0 per cent, so its share of that field moved from 4.87 per cent to 5.00 per cent. The gain is 0.13 percentage points. The gain is real. The gain is also small, and a note that says the company is taking share without stating that number has quietly turned a modest fact into a claim the record does not support.

Now the part that surprises people, and it is arithmetic rather than opinion. The same 0.13 points worked in rupees instead of points: holding the earlier year's share of the field flat and applying it to this year's field of Rs 48,300 crore gives about Rs 2,353.93 crore of revenue. The company actually did Rs 2,415 crore. The difference, the revenue the share gain itself brought in, is about Rs 61.07 crore. At the published 46.0 per cent gross margin that is about Rs 28.09 crore of gross profit, about Rs 21.05 crore after tax, about Rs 0.88/- a share.

Compare that with the Rs 0.75/- from a whole point of gross margin. A gain small enough to need a decimal place in percentage terms is worth more in rupees a share than a full point of gross margin, which is why the size travels with it in both directions and never only in the flattering one. Note carefully what the two figures are, because they are not the same kind of move: one is a single unit move of one line, the other is an entire driver moving at once. The two are set side by side to show that small in points is not the same as small in rupees, not to rank one against the other.

Everything else, and why calling it detail is not an insult

Look back at what got struck out. Depreciation, other income, the effective tax rate, the working capital cycleThe stretch of days between paying for materials and collecting from customers. The cycle decides when cash arrives, not how much profit is reported.. None of those lines is trivial, and forecasting them well is real work. Each is detail in one specific sense and only that one.

Each of these can be forecast wrongly without the view being wrong, and that single property is precisely what makes a line detail rather than a driver. Put depreciation Rs 9 crore out and the earnings per share moves about Rs 0.29/-, the view is untouched, and every reason for holding it is exactly as strong as before. Put the gross margin one point out and there is no view left to hold.

The distinction matters for how the model gets built. Detail lines still have to be right. A model that is wrong everywhere is not trusted anywhere, and the errors do accumulate. Monitoring attention through the year is what a detail line does not get. Each is forecast once, sanity checked at each release, and the reading time goes on the three lines that can actually change the answer.

How is a driver told apart from something that merely moves alongside?

There is a test and it takes ten seconds: state, in one sentence, how the variable reaches the earnings line. If the sentence can be stated, and it contains an actual chain of events rather than the words moves with or is associated with, the variable is a driver. If it cannot, the variable is a pattern.

Gross margin passes trivially: a higher margin means more gross profit on the same revenue. The extra gross profit flows through unchanged operating costs to a bigger profit before tax and a bigger figure per share. Every link is a thing that happens. The same test applied to a variable that has tracked this company's earnings for five years, say the monsoon arriving early, fails: the correlation can be described in detail, but what happens next cannot be stated without inventing it.

A driver has a stateable mechanism that reaches the earnings line and a pattern has only a history, and the difference matters because a view built on a pattern fails without any warning at all. That is the real danger. A driver that turns against the view announces itself: the margin prints lower, the volume prints slower, and it shows up in the release. A pattern simply stops holding one day, and nothing in the filings says so. There was never a mechanism to break.

What is done with a driver that cannot be observed?

Here is the uncomfortable case, and the worked exercise supplies an honest one. Whether Sarvani Coatings has its own pricing power, as opposed to sitting in a field where everyone could price ahead of costs, decides the first driver outright. Pricing power is also disclosed absolutely nowhere. Pricing power is not a line in the statements, not in the notes, and no filing anywhere contains it.

Two wrong answers present themselves. The first is to pretend pricing power is observable, usually by calling the margin improvement itself the evidence of it. Such reasoning assumes the very thing in question. The second is to drop the driver quietly on the ground that it cannot be measured. Dropping it is worse: the view still rests on it, and now nobody has written that down.

The correct response is to name the observable thing that stands in for the driver and state in advance what reading would be taken from it, so an unobservable variable still produces a testable expectation. Here the stand in is realisation per unit read against input cost per unit, both recomputable from published figures. If realisation keeps outrunning input cost while the field's margins flatten, that is a reading in one direction. If realisation moves with the field's, that is a reading in the other. Neither settles it. Both are better than an unwritten assumption.

Then it goes on the list. A monitoring list is just the drivers written out with the observation that would speak to each, and the discipline of writing it is what turns three sentences into three things that can actually be done next quarter.