Thesis Discipline: Holding a View Without Defending It

Thesis discipline is holding a view firmly enough to act on and loosely enough to drop, and it is decided before the evidence arrives rather than after. The view is written down, then the two or three assumptions it rests on, then what would contradict each one. Thesis risk then becomes measurable: it is the chance that one of those named assumptions is wrong.

What is this price asking a buyer to assume?

What does thesis discipline actually mean, and what is it not?

The calculator valued nothing. Given a price and four choices, it returned what the price requires of earnings: at the illustrative Rs 486/- for Sarvani Coatings Limited, invented, with the analyst's own required returnThe annual return an investor decides is needed before the money would be better used elsewhere. A preference, not a measurement, and one built up separately. set at 12 per cent and a 25 times rating assumed for the end of year five, roughly 24.2 per cent yearly earnings growth. The rest is the part no calculator can do: holding that requirement as a view, and keeping the record that makes holding it auditable.

The word carries the wrong associations. Discipline here is not conviction, not the ability to sit still while a price moves against the holder, not humility, and not a temperament somebody either has or does not have. Each of those describes a person, and none of them can be checked by anybody, including the person.

Thesis discipline is narrower and far more boring. Thesis discipline is an ordering. The conditions under which the view changes are written before the evidence that would trigger them exists, and everything else follows from that single ordering. Nothing about the conditions is clever. The timing is what makes them work.

The same ordering runs through ordinary life. A household decides in January that the second mobile connection goes if it ever costs more than Rs 400/- in a month. In January nobody knows which month will be the expensive one, so the rule has teeth. When the same household decides in June instead, holding a Rs 460/- bill, whether that counts as too much, there is a reason it was high that month, and there is always a reason. The June decision is not worse because the household is weak. The evidence and the rule simply arrived in the wrong order.

Why does defending a view feel like the honest thing to do?

Because of what happens the moment a view leaves the notebook. Shown to one colleague, it stops being a claim under test and becomes a position being held. Nothing about the analyst has changed, and yet everything arriving afterwards arrives already sorted into the parts that help and the parts that need explaining. The sorting never announces itself: nobody thinks the thought "I will ignore this." The unhelpful item simply arrives with a reason attached, and the reason is usually correct. The quarter was seasonal. The comparison base was odd. One customer pushed an order into the next period. The sorting happens to careful people. The remedy therefore has to be procedural rather than a resolution to be more open minded. Being told to try harder gives an honest analyst nothing to do on Monday morning.

The everyday version: a landlord tells relatives at a wedding that the new tenant is completely reliable. Two months later the rent arrives eleven days late. The landlord recalls that the tenant mentioned a delayed salary, notes that the previous tenant was worse, and files it as a one off. Every step of that is reasonable. The missing item is a line written on the day the tenant moved in saying what would count as unreliable. Such a line would have been written by somebody who did not yet have a view to protect.

What is Thesis Risk, and what makes it measurable?

Thesis risk is the chance that one of the assumptions carrying the view is wrong. One word is doing the work: the assumptions, definite, particular, countable. Thesis risk is not a mood about how confident anybody feels. Thesis risk is a property of a list, and if there is no list there is no property.

Which is why the naming comes first. Once it is written down that the view rests on three things and not on forty, each can be weighed on its own: which of them does the most damage if it is wrong, and which of them anybody could actually watch. Neither question can even be asked about an assumption nobody has written down.

A view with no written assumptions has no thesis risk that anybody can state, and that is a very different thing from having none. The risk is fully present and simply invisible, including to the person carrying it. Consider a shopkeeper who buys ninety per cent of the shop's stock from one wholesaler two streets away. Nothing has ever gone wrong, so the shopkeeper says the shop has no supplier risk. The concentration exists whether or not anybody has written it down. Writing it down does not create it; writing it down is the moment it becomes something arithmetic can be done about.

What has to be written down before any evidence arrives?

Four items, and the whole thing fits on half a sheet of paper. The view, in one sentence. The two or three assumptions it depends on. For each of them, what would contradict it. And the date.

The first two are the parts people manage: writing a view in one sentence is uncomfortable but doable, and listing the assumptions underneath it is what a careful reader does anyway. The third item is where almost every file falls down, and it is the item that does all the work.

Contradicting an assumption takes an observable: a specific thing a reader could see, arriving from somewhere specific, in a form that settles something. Not a feeling, not a direction, not a trend. If the assumption is that the gross margin gain persists, an answer such as margins get worse is useless. There is no year in which nothing gets slightly worse. A segment disclosureA breakdown in the published accounts showing revenue and profitability for each separately reported part of a business. The contents are set by accounting standards, covered in the accounting material., or a volume and realisationRevenue divided by the units actually sold, so the average price a maker realised per unit. Built and used in the earnings material. split separating pricing from mix, is an observable: either it turns up in a filing or it does not, and if it turns up it says something.

The observable is the item people leave out, and leaving it out is what turns a written thesis back into an unwritten one. An assumption with nothing watching it can never be contradicted, so it will still be sitting there, unchanged and unchallenged, four years from now. Going out and hunting for the evidence that would break a held view is a separate practice, covered separately. Only the ordering matters here: the disconfirming evidenceEvidence that would count against a view rather than for it. How to go looking for it deliberately is a practice of its own, developed in the sequence that follows this one. is described on the same day the assumption is, before anybody knows whether it will ever turn up.

What separates a revision from a drift?

A revision is a dated change that names the evidence which caused it and the assumption it touched. Three parts, all present, or it is not a revision. A drift is a view that differs from the one held last quarter with no recorded reason for the difference.

The distinction needs a name because of an asymmetry: a revision is visible and a drift is not. A revision leaves a line in the file with a date on it, findable again in a year. A drift has no moment at which it happened, no line to look up, and no entry to be embarrassed by. Drift is far more common than revision, and it is invisible unless the earlier version was written down. A file without dates cannot show that any drifting occurred.

The street version is a vegetable seller who never writes prices down. Asked in April what a kilo of onions cost in January, the seller gives a confident answer that is off by a third, and is not being dishonest. There is nothing to check against. The seller two stalls down chalks the price on a board each morning and rubs it out at night, and can say exactly what happened. The board existed at the time, so memory did not have to do the work.

What has an update that changed nothing actually produced?

The accounts are read, the margin line checked, the disclosure the observable named looked for, and at the end of the day nothing changes. The day feels wasted, and it was not.

Establishing that no named assumption was touched is a result. A file of unchanged views with dated checks against them is stronger evidence of discipline than a file full of revisions. The intuition runs the other way. Frequent revision looks responsive, and it may well be. Frequent revision also looks exactly like a file belonging to somebody who never wrote an observable, so every arriving item might mean something and nothing can be ruled out.

A security guard writes in the register at two in the morning that nothing happened. The entry is not a blank. A named person has signed a statement that the gate was watched at that hour and was clear, and a register full of those entries is what makes the one entry that is not blank worth something. A register with no two o'clock entries says nothing about two o'clock, in either direction.

When is a view dropped rather than adjusted?

A view is dropped when the assumption carrying it has been contradicted. Not when the price has moved against it, and not when holding it has become uncomfortable. Only one of those three triggers is about the business.

The hard case sits in between. Adjusting an assumption downwards just far enough to keep the view alive is drift in its hardest form, and it is far easier to see in somebody else's file than in one's own. The assumption said the gross margin gain persists. The evidence is awkward but not fatal. So the assumption quietly becomes that a good part of the gain persists, the view survives, and nothing is recorded as having changed. From where the analyst is sitting, nothing has.

The test that catches it is mechanical, and mechanical is the point. Compare the words in the assumption today with the words on the day it was written. Not the conclusion, the words. If they have moved and no dated entry explains the move, the view was not adjusted. The view was propped up.

What does the written record have to contain to be worth keeping?

Five things, and nothing else. The original view in its original words. The assumptions as first written. The observable attached to each. Every dated revision. And the evidence behind each one, named specifically enough that somebody else could go and look at the same thing.

The shape is not tidiness and it is not compliance. A record kept this way preserves what was thought before the outcome was known, so it is the only arrangement that lets an analyst find their own errors later. A summary written at the end of the year is written by somebody who already knows the answer, and it will be a clean, confident and quietly wrong account of what actually happened in March.

The whole of it, worked on one invented issuer

Here the procedure is carried out once, on Sarvani Coatings Limited, using only figures already established in the earlier sequences and the calculator above at its opening settings. The procedure ends in a statement of what the price requires and where the exposure sits, never in a figure of what the shares are worth.

The view, in one sentence. At the illustrative Rs 486/- against published year three earnings per share of Rs 11.58/-, on a 12 per cent yearly return and a 25 times rating at the end of year five, both of them the reader's own choices, the price carries about 24.2 per cent compound annual growthA single yearly rate which, applied repeatedly, turns a starting figure into an ending figure over a stated number of years. Built in the quantitative methods material. in earnings for five years, while the most recent published year delivered revenue growth of 13.9 per cent with the source of the gross marginRevenue less the cost of materials, expressed as a share of revenue. Where it sits on the profit ladder is settled in the accounting material. gain still unresolved.

The ladder behind that sentence, in four steps. Rs 486/- taken forward five years at the required 12 per cent has to reach Rs 856.50/-. Rated at 25 times, a price like that needs earnings of Rs 34.26/- a share. Set against the published Rs 11.58/-, earnings have to become 2.96 times what they are, and the single yearly rate that does that in five years is 24.2 per cent. Each step is one multiplication or one division that can be checked, and not one of them is a forecast.

The 24.2 per cent and the 13.9 per cent are not the same kind of number, so one bridge has to be stated before they can be set beside each other. The first is an earnings growth rate; the second is a revenue growth rate, year three against year two, on revenue of Rs 2,415 crore against Rs 2,120 crore. The two rates line up only if the margin holds flat and nothing below the operating line moves. Year three earnings actually grew 41.1 per cent, from Rs 197 crore to Rs 278 crore. The 41.1 per cent clears 24.2 per cent easily and still cannot be projected forward. Most of what produced it was a single year move in gross margin, 44.0 per cent to 46.0 per cent, a gain of 2.0 points. A margin that steps up once has stepped up once. The margin has not begun to grow.

The assumptions, named. Three, and the view rests on nothing else. First, that some material part of the 3.0 point gross margin gain from year one to year three persists. Second, that the share of the fieldHow large a slice of the whole described sector one maker accounts for, measured on revenue. Constructed in the sector research material. keeps moving, having moved 0.13 percentage points last year in a field growing 11.0 per cent. And third, that the exit ratingThe multiple of earnings other buyers are assumed to apply at the end of the holding period. The exit rating is a guess about a future crowd, not an assumption about the business. five years out is near 25 times.

The observables, one line each. For the first: a segment disclosure, or a volume and realisation split that separates pricing from mix. A disclosure of either kind may arrive and may never arrive. For the second: the share figure, computed against the same sector base every year and arriving annually whether or not it is convenient. For the third: nothing. No filing, disclosure or statistic anybody could watch would contradict an assumption about how other buyers will price these earnings in five years. The blank is not an oversight here. The blank is the entry.

The thesis risk, stated. Three named assumptions, of which one is watched by a figure that arrives every year, one by a disclosure that may never come, and one by nothing at all. The sentence is the measurement. The sentence says where the exposure sits, which part of it reading can reduce, and which part no amount of work can.

Now stress it, with three arrivals in the year after the published record. The next result shows gross margin holding at 46.0 per cent and no segment disclosure appears: assumption one is neither confirmed nor contradicted, because the observable did not arrive, so nothing is revised and the date is recorded. A volume and realisation split is filed instead, showing realisation flat while input cost per unit keeps rising: assumption one is contradicted directly, the view changes, and the change names the evidence. Or the price falls twelve per cent with no disclosure of any kind: nothing is touched. Other people changing their minds affects none of the three named assumptions, and a reader who revises here has drifted. Only the middle arrival produced a revision, and the file records all three.