Confirmation Bias in Financial Analysis: How It Shows Up and How to Counter It

Confirmation bias is the pull toward evidence that agrees with a view already held and away from evidence that does not. In analytical work the bias shows up in the sources that get read, the questions that get asked, how conflicting facts get weighted, and what gets filed without action. Confirmation bias survives intelligence because it works before reasoning starts, so the counters that hold are structural: written kill criteria, a designated disagreer, a pre-mortem.

Underneath the answer sits one uncomfortable fact about how a mind handles a view it already holds. Think of a household that has half decided to buy a second-hand scooter from a neighbour. The mechanic says the clutch is a bit worn; the household hears "nothing serious". A cousin says the same model gave trouble; the household remembers that the cousin exaggerates. The price is a little high; the household notes that a new one costs three times as much. Nobody in that house is lying to anybody. Each fact arrived, was sorted, and the sorting happened before anyone sat down to think. The sorting is confirmation biasThe tendency to notice, seek, trust and remember information that fits a view already held, and to overlook or discount information that does not., and the finance version is the same shape with more zeros on it. The analyst's version shows up in four places, survives being clever, and yields only to counters that need nobody to notice anything. All four counters are visible in the February moment of the Kaveri Cold Chain case, where the assumption register, the decision log and the audit trail earn their keep.

What is confirmation bias, in an analyst's terms?

An analyst spends the day receiving evidence: reports, calls, filings, a line in a trade newsletter, a remark from a promoter. Long before that evidence reaches the part of the day called analysis, it has already been handled. Some of it feels important and gets read twice; some feels early, or thin, or off the point, and gets skimmed and set aside. The handling is fast and feels like judgement. Confirmation bias is the fact that the handling is not neutral: once a view exists, evidence that fits it feels weightier and evidence that does not feels like noise, and the sorting is finished before reasoning begins. The analyst then reasons carefully and honestly about the pile that survived, and reaches a well argued conclusion that was decided by the sorting.

Look at the picture below and follow one item of evidence through it. Ishaan Verma, an invented analyst at an invented investment team, recommended in January that the team lend Rs 25,00,00,000 to Kaveri Cold Chain Private Limited, an invented refrigerated warehousing business, on the view that a signed pharma contract would lift occupancy from 62 to 80 per cent within a year. The occupancy view now sits at the centre. Five items arrive over the following months. Three of them fit the view and pass through the sorter into the supporting tray, where they get read, forwarded and discussed. Two of them do not fit, and pass into the discard tray, where they get read once and filed. Notice where the reasoning box sits: after the trays, and connected only to the supporting one. From inside the process, nothing in the discard tray ever looked like something that needed weighing. Nothing in it was weighed.

How does confirmation bias show up in analytical work?

Because the bias shows up differently at each stage, it is easier to catch at four points along the work than as one event. The picture below lays an analysis out left to right, from choosing what to read through to what happens to the facts that disagree, and marks the symptom at each stage. The four symptoms serve as a checklist an analyst can run over a finished analysis, not as an accusation. The four symptoms are that sources get chosen for agreement, questions get framed so that the answer supports the view, conflicting facts get weighted lightly, and disconfirming evidence gets filed instead of acted on.

Take them one at a time with the Kaveri case. Sources: in the weeks after his view formed, Ishaan Verma read the pharma client's press releases, Kaveri's own occupancy reports and a regional logistics newsletter, and did not pull the state electricity regulator's tariff consultation. Questions: on the call with Kaveri's promoter he asked how quickly the pharma pallets would ramp, and did not ask who else was building refrigerated capacity nearby. Weighting: when the first quarter's occupancy came in two points up, that felt like the thesis working; when spillover was still at zero, that felt like early days. Filing: the February trade report was read carefully, judged early stage, and filed. Each of those four is a reasonable act on its own. The pattern is that every triage landed on the same side. Each act arrived alone, so the analyst inside the pattern could not see it as a pattern.

Why does confirmation bias survive intelligence and good intentions?

The part most people find hard to accept is this. A sharper analyst can spot a weak argument, so a sharper analyst might be expected to be safer. The expectation holds for other people's arguments. When the view is the analyst's own, the sharpness goes to work on the sorting side of the picture above, and it produces better reasons for the pile that survived. A less experienced analyst asked why the competitor report does not matter offers two thin reasons. A very good one offers a paragraph on permitting timelines, capital costs, the pharma client's switching costs and the difference between announced and built capacity, all of it correct, all of it produced to order. Confirmation bias survives intelligence because intelligence is applied after the sorting, so a more capable mind builds a more capable defence of whatever the sorting let through.

The habit has a name worth knowing: motivated reasoningReasoning that starts from the conclusion already wanted and works backward to arguments for it, while feeling from the inside like ordinary careful thinking.. Motivated reasoning is neither lying nor laziness; it is what careful thinking looks like when the destination was fixed before the journey began. Nor is good intent a shield. An analyst who genuinely wants to be objective, and who checks a box saying they have considered the other side, has still done the considering on evidence that reached the reasoning box, and the discard tray is not there. The picture below makes the point with two invented analysts. Both are asked to list reasons for and against the Kaveri view. The skill went where the motivation pointed it, so the stronger analyst finds far more reasons for and about the same number against.

Which counters actually work?

If the moment of bias is invisible from inside, then any counter that asks the analyst to notice it is asking for the one thing the bias removes. Being objective, keeping an open mind and considering the other side are all ruled out on those grounds. The counters that hold put a rule or a person into the process at a moment chosen in advance. Disconfirming evidence then gets its meeting whether or not it feels important on the day. Four counters work because none of them depends on the analyst seeing the bias: kill criteria written before the decision, a designated disagreer during it, a pre-mortem before committing, and disconfirming evidence sought first rather than last.

Take each in turn. Kill criteriaConditions written down before a decision, stating in advance which facts, if they arrive, will reopen or reverse it. Also called tripwires. are the most powerful of the four because they are written by the analyst on the day the view is clearest and least defended. In January, before the loan is approved, Ishaan Verma writes in his decision log: competitor announces refrigerated capacity within 30 kilometres, reopen the case; power tariff revision notified, rerun the model; spillover still zero at the end of the second quarter, reopen the spillover row. Now the February report is not a judgement call. The report matches a line written in January, and the line says reopen. A designated disagreerA named person given the job of arguing against a proposal in a meeting, whether or not they personally agree with it, so that the case against gets a full hearing. Sometimes called the devil's advocate. does during the decision what kill criteria do after it: someone is told, in advance and by name, that their job today is to argue against the loan, so the case against is made by a person who does not have to overcome their own view first. A pre-mortemAn exercise run before a decision is final: assume it is twelve months later and the decision has failed, then write down the most likely reasons why. Distinct from a post-mortem, which is run after the fact. asks the whole team to imagine it is December, the Kaveri loan has gone badly, and to write the story of why; the exercise gets people to name the competitor and the tariff while naming them still costs nothing. And disconfirming evidenceAny fact that, if true, makes the current view less likely to be right. The opposite of confirming evidence, which makes the view more likely. sought first simply reverses the order of the search: before reading anything that supports the view, spend an hour trying to find the strongest fact against it, so the discard tray is filled before the supporting tray, when it is still being read rather than defended against.

How does structure beat willpower here?

The household and the scooter make the same point. The household already believes it is being objective, so telling it to be more objective about the clutch changes nothing. A rule set before the neighbour ever named a price does change something: no second-hand vehicle is bought without a written report from a mechanic the household chose, and if the report lists more than two faults the household walks away. The rule does not need anyone to notice their own bias. The rule runs whether or not the household is feeling careful that morning. Structure beats willpower because willpower has to be exercised at the exact moment the bias has hidden the need for it, while a structure written earlier acts on the day without asking anyone to notice anything.

The picture below puts the two kinds of counter side by side, and the useful column is the third one, what each requires. Every willpower counter requires the same impossible thing: that the analyst notice, in real time, a sorting that has already happened outside their attention. Every structural counter requires something a colleague can check: a line in a log, a name on an agenda, a written sheet from a pre-mortem, a search done in a fixed order. Twelve months on, the review can therefore ask whether the counters were in place, and cannot usefully ask whether the analyst tried hard enough. Structure is also cheap. Three lines in a decision log cost Ishaan Verma ten minutes in January. Those ten minutes would have bought a meeting in February.

What did Ishaan Verma do with the February report?

The worked moment itself, told plainly and fairly, runs as follows. In January the investment team approves the Rs 25,00,00,000 loan to Kaveri Cold Chain on Ishaan Verma's recommendation. His assumption register has five rows: occupancy today 62 per cent, the pharma contract at 14 points of capacity, spillover of 4 further points, the tariff of Rs 1,150 per pallet-month held flat, and power cost of Rs 1,80,00,000 held flat. His decision log records what he decided, when and why, and its line on what would change his mind names the pharma contract slipping and nothing else: no competitor, no tariff. In February, a month after drawdown, a trade report crosses his desk noting that a competitor is planning a refrigerated facility about twenty kilometres from Kaveri. He reads it, properly. He judges the plan early stage. In February that is a defensible reading: nothing is built, no permit is public, and announced capacity in this sector often stays announced. He files the report. Nothing in his process tells him to reopen the case, add a row to the register or raise it at the monthly review, so he does none of the three.

The bias did not lie to Ishaan Verma about the report; it decided which facts were worth a meeting, and it decided in the direction of the view he already held. Look at what was missing rather than at what he did. His register had no row for competitor entry, so there was nowhere for the report to land. His log had no kill criterion for it, so there was no line that said reopen. No one on the team had been given the job of arguing against the loan, so nobody was hunting for exactly this. Twelve months on, the competitor is open, spillover has delivered 0 of the forecast 4 points, occupancy has reached 71 rather than 80, and profit before tax (PBT) is Rs 1,10,00,000 against a forecast of Rs 4,20,00,000. The loan is serviced on time. Whether the decision was sound is a separate question. The fact arrived in February and landed in December, and the ten months between were the price of a filing decision that felt, at the time, entirely reasonable.

How do lenders, analysts and investors actually use these counters?

A lender writes the kill criteria into the paperwork. When a bank sanctions a term loan, the sanction letter often carries monitoring triggers: an occupancy figure that, if breached in any quarter, requires the borrower to present a revised plan; a covenant on power cost as a share of revenue; a requirement to report new capacity within a stated radius. Each trigger is a kill criterion that a committee has moved out of one analyst's head and into a document a monitoring officer reads every quarter without needing to share the analyst's view. The credit committee itself is a standing designated disagreer, and the good ones name a member per case whose job is to make the case against.

An analyst uses the pre-mortem before publishing and the "what would change my mind" line inside the note. A research note that says the view is wrong if the pharma client's ramp slips past June, or if a second facility is announced within the catchment, has done in one sentence what a kill criterion does: it has told the reader, and the analyst's own future self, exactly which facts get a meeting. Practitioners do not counter confirmation bias by trying to feel differently about evidence; they counter it by deciding in advance which evidence gets acted on, and writing that decision where someone else can hold them to it.

A household uses the same trick at a smaller scale, and it is worth seeing because it shows the idea is not a finance trick. A couple paying instalments on a flat under construction can write, before the first payment, the condition on which they stop: if the builder misses the second milestone by more than three months, the couple pauses and takes advice. Written in advance, the rule runs on the day. Left to the day itself, the same couple will find, entirely honestly, that the delay is understandable and the builder is trying.

Kaveri Cold Chain Private Limited, Ishaan Verma and Devika Rao are invented.
Educational material. Not advice on any investment, tax, budget or market position.