Counterfactual Reasoning in Finance: What Would Have Happened Otherwise

A counterfactual is the honest picture of what would have happened had a decision gone the other way. Every judgement of a decision secretly compares its outcome to one, so the comparison should be built openly, from the information available at the time, not from what is now known. A counterfactual differs from a scenario in direction: a scenario looks forward from a decision, a counterfactual looks sideways from an outcome.

An outcomeWhat actually happened after a decision: the profit earned, the loan repaid, the price reached. A result, not a judgement of the result. by itself is a number with nothing to stand beside. Rs 1,10,00,000 of profit, a loan repaid on time, a fund up 9 per cent: none of these is good or bad until the question compared to what is answered, and the "compared to what" is always the road not taken. The road not taken was never travelled, so it cannot be looked up or remembered. The road has to be reconstructed, and the only honest materials for the reconstruction are the facts that were on the desk when the decision was made. Inside every "that went well" sits a hidden comparison. The road not taken can be built for a loan, a fund or a company from decision-date facts alone, and keeping later knowledge out of it is what separates a counterfactualThe picture of what would have happened had the decision gone the other way. From "counter to fact": a world that did not occur, reconstructed for comparison. from a scenario.

What is a counterfactual, and why is every judgement secretly one?

A small case first. A traveller takes the train to a meeting instead of driving, and arrives on time. Was that a good call? Arriving on time cannot settle it. If the road was clear and the drive would have taken half the time, the train was a slow choice that happened to work. If there was an accident on the highway, the train saved the meeting. The mind, the moment it asks "good call?", goes looking for the drive that was not made. The ghost journey is the counterfactual, and it gets consulted without anyone deciding to consult it.

Every verdict on a decision is a comparison between the outcome that arrived and an outcome that did not, and the second half of that comparison is usually left unspoken. "That investment doubled" is only praise if the money would have done less elsewhere. "The vendor's sales rose after he moved his cart" is only a good move if sales would not have risen on the old corner too. In a festival week they might well have. The wedding caterer who ordered for 500 and fed 420 with food to spare is judged, silently, against the caterer who ordered for 420 and ran out. In each case the person judging holds a picture of what would have happened otherwise, and the quality of the judgement depends entirely on the quality of that picture. A counterfactual is simply that picture made explicit, written down, and built with care rather than assumed in a flash.

The finance version is Ishaan Verma, an invented analyst at an invented investment team, who recommended in January that the team lend Rs 25,00,00,000 to Kaveri Cold Chain Private Limited, an invented refrigerated-warehousing business. Twelve months later Kaveri Cold Chain reported a profit before tax of Rs 1,10,00,000 and had serviced the loan on time. Around the table someone says "that went well". Hold that sentence for a moment. Inside it sits a comparison nobody has stated: went well compared to what the team’s Rs 25,00,00,000 would otherwise have done. Until that other road is drawn, "went well" is a feeling, not a finding.

Why can an outcome not be judged without one?

Kaveri Cold Chain's Rs 1,10,00,000 of profit before tax cannot be judged with nothing beside it. The profit is a positive number, so it is not a loss. Beyond that it says nothing about the decision that produced it. Put something beside it and the same number changes character. Beside January's starting point, break-even, it is a gain of Rs 1,10,00,000 and the loan looks like it lifted the business. Beside Ishaan Verma's own thesis, a profit before tax of Rs 4,20,00,000, it is a shortfall of Rs 3,10,00,000 and the loan looks like it fell well short. Same rupees, opposite readings, and the only thing that moved was the baselineThe reference point an outcome is measured against. Change the baseline and the same outcome reads as a gain or a shortfall..

"Good" and "bad" are relations, not properties, and a relation needs two ends, so an outcome without a counterfactual is uninterpretable. The same need is why a household that reports savings of Rs 40,000 this year cannot know whether the year went well until it asks what a normal year saves, and why a fund manager reporting 9 per cent cannot be praised or blamed until it is known what the money would have earned sitting somewhere plainer. The trap is that a positive number feels like an answer. A positive number is not an answer. A positive number is one end of a comparison waiting for its other end, and when nobody supplies that other end deliberately, the mind supplies one of its own, silently and usually flatteringly.

How is an honest counterfactual built?

If the other end of the comparison has to be reconstructed, the question is what it may be built from. The everyday version runs like this. Someone leaves the house without an umbrella and gets soaked. Was leaving it a mistake? Only if the sky at eight in the morning gave reason to carry one. The eight o’clock sky was all there was, so the decision belongs to that sky and not to the rain at noon. The same rule, applied to money, is what separates a counterfactual from a story told backwards.

An honest counterfactual is built in four moves, and every one of them is anchored to the decision date rather than to today. First, fix the date: January is when the recommendation was made, so the counterfactual for the Kaveri Cold Chain loan is a January object. Second, list what was actually on the desk that day, with sources, and nothing else. Third, name the alternative that was genuinely available then, not an alternative invented afterwards to make the outcome look good or bad. Fourth, carry that alternative forward using only the January facts and ask what it would most plausibly have produced. The four moves produce the road not taken as January could have seen it, and only that version of the road can fairly be set beside the outcome.

Ishaan Verma’s assumption register is what makes the first two moves possible. The register therefore matters more than the model built on it. Five inputs, each with a source and a date: current occupancy 62 per cent from Kaveri Cold Chain's own December occupancy report; a signed pharma contract for 14 per cent of capacity, dated January; a further 4 points of occupancy expected from spillover, recorded as Ishaan Verma's own estimate with no external source; a tariff of Rs 1,150 per pallet per month from the January rate card, held flat; and power cost of Rs 1,80,00,000 from the last audited accounts, held flat. His decision log names the alternative the team was weighing the same week: renewing a Rs 25,00,00,000 working-capital line to Malnad Grain Storage, an invented long-standing borrower with a clean repayment record, at a lower rate and with far less that could go wrong. Because all of that was written down in January, the counterfactual can be built from January without a single later fact.

Where does hindsight corrupt it?

The trap is one almost everyone walks into. Twelve months on, a reviewer knows things January did not. A competitor opened near Kaveri Cold Chain in month four, May, and took roughly 5 points of the existing occupancy. A power tariff revision in month nine, October, pushed the year's power cost from the Rs 1,80,00,000 in the register to Rs 2,30,00,000. Spillover, forecast at 4 points, delivered none. The road not taken then has to be built with those facts already in the reviewer's head. Without meaning to, the reviewer draws a January in which the competitor was obvious and the tariff risk was glaring, and concludes the loan should never have been made. The January the reviewer draws is not a counterfactual. The drawing is hindsightJudging a past decision with facts learned after it was made. Once an outcome is known it feels as though it was always predictable. The feeling is false. wearing a counterfactual's clothes.

Hindsight corrupts a counterfactual by letting knowledge acquired after the decision date leak backwards into the reconstruction of the alternative. The leak is quiet. A fact known now feels like a fact always known. Its effect on the verdict is severe. Judged on January's facts, the loan and the renewal were both defensible and the loan was the riskier road with the higher return. Judged with May and October smuggled in, the loan becomes an obvious mistake, and Ishaan Verma is a poor analyst. The second verdict feels sharper and is worthless. Grading a January decision against a December exam paper settles nothing about January. Hindsight also produces a lesson nobody can use. January could not act on "do not lend to a business whose competitor will open in four months", having no way to know.

The umbrella holds the whole rule, so take it up again. Getting soaked at noon does not make the eight o'clock decision wrong. Learning from the morning means going back to the eight o'clock sky and asking what it showed: were there clouds ignored, was there a forecast left unchecked. A cloud ignored and a forecast unchecked are lessons January can use. In the Kaveri Cold Chain register there is exactly one such lesson, and it needs no later fact to find it: the spillover input, 4 points of occupancy with no source behind it, was the weakest line in the register in January and it delivered zero.

One refinement remains, and it is where careful people still slip. The information date is set by the decision under judgement, not by the calendar. In February, month one, a trade report noting the competitor's plan for a nearby facility landed on Ishaan Verma's desk, and he filed it without reopening the case. The February report is inadmissible against the January recommendation, already made a month earlier. Against the February decision not to reopen the case the same report is fully admissible: in February it sat on the desk. Same fact, two decisions, two different verdicts on whether it counts. Every decision has its own information date, and a review that keeps them straight can be hard on the February filing while staying fair to January.

What is the difference between a counterfactual and a scenario?

Counterfactuals and scenarios get confused constantly, and the confusion matters because the two are used at different moments for different jobs. The wedding caterer returns, on the morning of the order. If 500 come, this much food is needed; if 420 come, this much; if 600, that much. Each of those orders is a scenario: several roads fanning forward from a decision not yet made. The same caterer the day after the wedding, with 80 plates of food left over, reasons differently: an order for 420 would have run out by the fourth course, and an order for 460 would have been about right. The order for 460 is a counterfactual: one road reaching sideways from a known outcome to the order that was not placed.

A scenario looks forward from a decision to several possible futures; a counterfactual looks sideways from an actual outcome to the one road not taken. Scenarios exist before the outcome and are tools for choosing. A counterfactual exists after the outcome and is a tool for judging. In January, Ishaan Verma's model carried scenarios: occupancy reaching 80, occupancy stalling at 76, occupancy falling if the pharma client left. Twelve months later, with occupancy at 71 and profit before tax at Rs 1,10,00,000, the counterfactual is a different object: the Malnad Grain Storage renewal carried forward on January's facts and set beside what actually happened. Notice that a scenario can be wrong and still have been well made. A counterfactual is never a prediction at all. A counterfactual is a comparison, and its only failure is being built with facts that arrived too late.

What is the counterfactual for the Kaveri loan?

The moves do not change with the subject, so the same construction would serve a fund or a company. Built properly, it runs as follows. The decision date is January. The facts are the five register inputs. The alternative is the Malnad Grain Storage renewal that the decision log names. Carrying that alternative forward on January's facts: Rs 25,00,00,000 lent to a long-standing borrower with a clean record, priced lower than the Kaveri Cold Chain loan, with little in the register that could go wrong. The two loans are priced illustratively at 12 per cent and 9.5 per cent, so the taken road earned Rs 3,00,00,000 of interest in the year and the declined road would have earned Rs 2,37,50,000. The difference, Rs 62,50,000, is what the team was paid for taking the riskier road.

Set beside the road not taken, the Kaveri Cold Chain decision reads as a risky bet that landed narrowly, not as the clean win it looks like on its own. The good news is that the taken road delivered its interest and its capital on time. But it delivered them off a business that kept only Rs 1,10,00,000 of the Rs 4,20,00,000 it promised, about 26 per cent, having lost 9 of the 18 forecast points of occupancy and absorbed Rs 50,00,000 of unbudgeted power cost. The cushion between the team's money and trouble was thin, and it was thin for reasons that January could not see. The declined road would have earned Rs 62,50,000 less and required none of that luck. The comparison is not a verdict that the loan was wrong. On January’s facts both roads were defensible. The comparison is a verdict about how much of the outcome was decision and how much was fortune, and the split only becomes visible with the second road drawn.

Where did the Rs 3,10,00,000 shortfall come from, and does the counterfactual need to know? The bridge below reads from left to right. The thesis of Rs 4,20,00,000 loses Rs 2,60,00,000 to the 9 points of occupancy that never came, made up of the 4 spillover points that delivered zero and the roughly 5 existing points the competitor took, and loses a further Rs 50,00,000 to the power tariff revision, arriving at Rs 1,10,00,000. The decomposition is the material of a post-mortem, set out under the post-mortem. For the counterfactual the decomposition plays a different role: it shows how much of the taken road’s outcome depended on things January could not see. The size of that dependence is what makes the landing narrow rather than clean.

How do lenders, fund managers and households actually use this?

A lender's credit committee uses the counterfactual in two directions. Looking at loans it made, the committee asks what the same capital would have earned on the next-best deal that week. A well-run desk therefore records the alternative in the decision log the way Ishaan Verma did, so the comparison can be built later without inventing it. Looking at loans it declined, the better desks track what happened to the borrower afterwards. A declined loan that went on to perform elsewhere is the counterfactual for the decline, and without that record a committee only ever learns from the loans it said yes to. The second habit is rare, and the honest lessons hide there.

A fund manager lives with a counterfactual that is written down in advance: the benchmark. A fund up 9 per cent has been judged against nothing until it is known what the plainest alternative, the index or a deposit, would have done, and the whole apparatus of attributionBreaking down a result into the decisions that produced it: how much came from what was chosen, how much from timing, how much from the market moving anyway. is an attempt to separate what the manager's decisions added from what the market would have handed anyone. Practitioners rarely ask whether an outcome was good; they ask what the alternative would have produced, and how much of the gap was skill rather than fortune. The same discipline separates decision qualityHow well a decision was made given what was knowable at the time. Distinct from outcome quality: how well things turned out. from outcome quality: a good decision can lose and a poor one can win, and only the road not taken, built at the decision date, tells the two apart.

A household does this every time it reaches for a should-have. The loan rate should have been fixed; the old scooter should have been sold earlier; the money should have been kept in the deposit. Each is a counterfactual, and each is usually built with hindsight, using the rate that came later or the price the scooter fetched after the fact. The honest version asks what the household knew on the day: what the bank was quoting, what a scooter was fetching then, what the deposit was paying then. Asked that way, some of the should-haves dissolve into could-not-have-knowns. The dissolving is kinder and, more to the point, true. In research on financial products, the same expectation applies formally: an analyst's recorded reasoning is what lets a reader later judge the recommendation on the facts available at the time. Conduct rules for research analysts, published by the Securities and Exchange Board of India (SEBI), require the basis of a recommendation to be documented for that reason.

Kaveri Cold Chain Private Limited, Malnad Grain Storage, Ishaan Verma and the investment team are invented.
Educational material. Not advice on any investment, tax, budget or market position.