Confirmation Bias in Investing: How You Fool Yourself

Be honest about how the last hour of "research" actually went. You already liked the stock. You opened the concall summary, the annual report, a couple of analyst notes. And as you read, the good lines lit up — the order book "robust," the guidance "raised," the management "confident" — and the awkward lines slid past, half-read. You finished more convinced than you started, and you called that diligence. It was the opposite. You went looking for a reason to do what you'd already decided to do, and the market is generous: it will hand a reason to anyone who only wants one.

That is confirmation bias — the mind's habit of seeking, weighting, and remembering evidence that supports what it already believes, while quietly discounting everything that doesn't. In investing it is the most dangerous bias of all, because it disguises itself as the cure. Reading more feels like reducing risk. But if you read only to confirm, more reading just builds a taller, better-defended wrong opinion. And the disconfirming evidence is usually sitting right there in the primary source: across 1,519 listed Indian companies tracked on Inve, 13,280 management commitments are on the record, and 934 of them — one in fourteen — were simply never mentioned again on any later call (Inve data, 2026-06-24). The bull skips the ones that went quiet. The bear skips the ones that got delivered. Both call it research.

This piece is about the one discipline that breaks it — Munger's "iron prescription" — and how to run your research through it instead of around it. We'll ground it in two real Indian concalls, because the trap is much easier to see in someone else's stock than in your own.

How does confirmation bias hide inside ordinary research?

It hides because it doesn't feel like bias. It feels like agreeing with the facts. The trick the mind plays is in the selection and the weighting, not the fabrication — you're not making anything up, you're just choosing what to look at and how hard to look.

Three mechanisms do the work. Selective search: you type "why X is a great business" into the search bar, not "why X will disappoint." The query decides the answer before you read a word. Selective weighting: the supporting fact ("revenue beat") gets full marks; the contradicting fact ("management dodged the margin question again") gets explained away as a one-off, a conservative quarter, noise. Selective memory: a month later you remember the order book and have forgotten the dodge entirely. Each step is small and reasonable on its own. Stacked, they manufacture conviction out of a balanced reality.

The incentive layer makes it worse, and it's worth naming. Much of what you read while "researching" was written by someone paid to be bullish — sell-side analysts whose firms want banking relationships, IR teams whose job is the story, finfluencers paid per view to be exciting. Munger's line is the whole defence: don't ask the barber whether you need a haircut. When the easiest evidence to find is produced by people paid to confirm what you already hope, confirmation bias and incentive bias hold hands and walk you straight off the ledge.

Here is the homely version, committed to fully. Reading a concall to confirm your thesis is like a lawyer who only interviews witnesses for his own client. Every answer helps the case; the file looks airtight; he walks into court certain he'll win. He has not built a strong case — he has built a one-sided one, and the other side's witnesses, the ones he refused to call, are sitting in the gallery waiting. The market is the opposing counsel, and it has read the whole witness list. The investor who only called friendly witnesses is the one who gets surprised in cross-examination — which, in the market, is called a drawdown.

What is Munger's iron prescription, and why is it the cure?

Charlie Munger gave the most demanding standard in investing for whether you're allowed to hold an opinion at all: "I'm not entitled to have an opinion on this subject unless I can state the arguments against my position better than the people who support it."

Read that twice, because the bar is higher than it first looks. It is not "consider the other side." It is not "list a few risks." It is: build the strongest possible case against your own thesis — stronger than its most articulate real-world opponent would build — and only if your conviction survives that are you entitled to it. Until you can do that, you don't have a view. You have a preference wearing a view's clothes.

This is the exact antidote to confirmation bias, because it inverts the search. Confirmation bias asks "what supports my thesis?" The iron prescription forces "what is the most devastating true thing a smart bear could say about this stock, and can I answer it?" You cannot run that exercise by selectively reading the good lines. You are now hunting the dodges, the ghosted guidance, the receding targets — not to dismiss them, but to see if they sink you. If they don't, your conviction is earned. If they do, the bias just saved you from a position you'd have entered blind.

It is hard, deliberately. Stating the bear case better than the bears means understanding it from the inside, not strawmanning it. Most "I considered the risks" sections in research notes fail this instantly — they list risks they've already decided don't matter. The iron prescription only counts if, for a moment, you genuinely scare yourself.

What does the evidence look like when you actually go hunting?

The two companies below are illustrations of how to read a concall against your own bias — not a view on either stock, and certainly not a buy or sell call. They're useful precisely because the disconfirming evidence is there in plain sight, and a committed bull would have read straight past it.

Eternal (Zomato): the target that got quietly redefined. Walk through the food-delivery growth guidance the way a bull would have lived it. Q1 FY25: management guides 20%+ CAGR, and the business is running above it at 27–28% (Inve data, Q1 FY25). Easy to own — the number is beating the guide. Q2 FY25, reiterated. Then Q3 FY25, growth slows and the guidance goes "at risk." Then the moment a confirming reader skips. On the Q4 FY25 call (1 May 2025), with that quarter's growth down to 16%, CFO Akshant Goyal reframes the whole target: "20%, therefore, is more a long-term 4-5 year CAGR guidance. I'm just clarifying that even in the past, we have stated that it's not an immediate every year growth guidance" (Eternal Q4 FY25 concall). The goalpost didn't move a little — the goalpost stopped being annual. Growth then fell to 13% the next quarter (Inve data, Q1 FY26), and by Q4 FY26 the 20% figure simply wasn't reiterated at all (Inve data, Q4 FY26).

Here's the said-vs-did table the bull's memory edits out:

Now the dodge — the witness the bull never calls. On the Q3 FY26 call (21 January 2026), analyst Nikhil Choudhary asks the obvious question: when does food-delivery growth re-accelerate? Management's answer, graded evasive by Concall AI: "Very hard to say, Nikhil. These things keep changing for reasons which are beyond our control. We don't want to venture and take a guess here on how this moves" (Eternal Q3 FY26 concall). Read that as the bull and it's reassuring humility. Read it as the bear, next to a 20% target that quietly became "4–5 year," and it's the same management that was happy to put a number on growth when the number was 28% declining to put one on it now. The confirming reader remembers "20% long-term aspiration intact." The disconfirming reader remembers a specific target that became elastic exactly as the actual number fell from 28% to 13% — and a CFO who, when pressed for a fresh one, wouldn't venture a guess.

Adani Enterprises: the target that vanished. This is the cleaner pattern — a hard, headline number that simply stopped being said. In Q4 FY24, management's stated goal was a "10 GW green hydrogen integrated manufacturing ecosystem by FY26" (Inve data, Q4 FY24). By Q1 FY25 the timeline was pushed out to "closer to FY28." Then, across the Q4 FY25, Q2 FY26 and Q3 FY26 calls, the 10 GW target was not mentioned or updated at all — management's commentary moved to the 6 GW solar cell and module capacity instead (Inve data, Q3 FY26). The number didn't get missed in public; it got retired in silence.

That silence is the disconfirming evidence, and it is the single easiest thing for a holder to never notice — because an absence doesn't light up the way a fresh headline does. Nobody issues a press release that says "we have stopped talking about the goal we set." You only catch it if you line the calls up side by side and ask, deliberately, what did they used to volunteer that they've gone quiet on? That is the question confirmation bias is built to suppress.

What is Munger's iron prescription, applied to these?

Take Eternal as the test case, because it's the harder one — there's a genuinely good business in there, which is exactly when the bias is most expensive. The iron prescription says: before you let yourself own the bull thesis ("dominant marketplace, Blinkit optionality, food delivery is a long-term 20% compounder"), state the bear case better than a real bear would. It sounds like this: the one hard number management was willing to commit to — 20% food-delivery growth — got redefined from annual to "4–5 year CAGR" the same quarter growth fell to 16%, then slid to 13%, and when an analyst asked point-blank about re-acceleration the CFO declined to venture a guess; a management that meters its specificity to its results is telling you something, and the burden is on the bull to explain why this time the elastic target snaps back (and beware anchoring to the original 20% the same way).

If you can state that, fairly, and your thesis still holds — because you've decided the Blinkit economics or the long-run delivery TAM outweigh a soft patch in one segment's growth — then you've earned the position. If you can't state it, or you find yourself reaching for "management is just being conservative," you were confirming, not researching. The point is not that the bear is right. The point is that until you can make the bear's case in your own words, you don't yet know whether you're right either.

Where do you find the disconfirming evidence you're avoiding?

In the primary source, not the commentary — and by "primary source" here we mean the company's own original disclosure (the concall transcript itself), not the NISM sense of primary research (facility visits, talking to customers and suppliers). Third-party commentary is where confirmation bias is easiest to feed; the concall is the one place a company is cross-examined in public, and the disconfirming evidence is sitting in the parts your bias skips.

The Q&A, read for the dodges. Prepared remarks are written by IR to confirm the bull case; skip them for this exercise. Go to the Q&A and find the question an analyst keeps returning to — receivables, the margin walk, the underperforming segment — and read management's answer as the bear would. A direct, specific answer is genuine disconfirmation of the bear case (good for the bull). A pivot to "very hard to say… we don't want to venture a guess" — exactly Eternal's Q3 FY26 reply on food-delivery acceleration — is disconfirmation of the bull case, and it's precisely what your bias filtered out the first time. Across Inve's tracked calls, an evasive-graded answer like that shows up in 932 transcripts spanning 617 companies (Inve data, 2026-06-24); each one is a witness the bull refused to call.

The guidance walk, read for the drift. Don't read this quarter's target in isolation — that's where confirmation lives. Read the last four-to-six quarters' targets in sequence, the way the table above does for Eternal. A number that only ever moves down, then gets redefined, then disappears, is the disconfirming evidence the bull thesis cannot survive without an answer. Of the commitments tracked on Inve, 934 have been ghosted and 534 companies — roughly one in three — carry at least one quietly-dropped target (Inve data, 2026-06-24). Adani's 10 GW hydrogen goal is one of them, and a held position is precisely where you're least likely to have noticed your own company's ghost.

The metric that went quiet, read as a confession. Find a number management used to volunteer and check whether it's still being said — Adani's "10 GW by FY26" is the textbook case: stated as a headline, pushed to FY28, then dropped without a word. And cross-check the headline profit against the cash, since quality of earnings is where a flattering story quietly diverges from reality. Things that stop being mentioned when they'd be embarrassing are the cleanest disconfirmation there is — and the easiest for a committed bull to never notice, because absence doesn't light up the way a good headline does.

How do you build research that tests instead of confirms?

You change the question you walk in with, and you build a record that won't let you forget the inconvenient parts. Confirmation bias thrives on selective memory; a written, dated record of what management actually committed to is its natural enemy.

Start from the bear, every time. Before you read a single bullish source on a stock you own or want, write the strongest case against it in your own words — the iron prescription as a literal first step, not an afterthought. Only then read, and read to see whether the bear case holds. This one reversal does more than any amount of "balanced" reading, because it points the search engine of your attention at the evidence you'd otherwise skip.

Track the commitments, so you can't un-remember. Selective memory is half the bias; the fix is to not rely on memory (why a Promise Tracker beats memory). Inve's Promise Tracker holds each commitment's original wording against its current verdict — Achieved, On Track, Missed, or Ghosted — so a target like Eternal's "20% CAGR" or Adani's "10 GW by FY26" stays on the record with the quarter it went quiet, and the concall summaries surface the graded Q&A and the dodges per quarter. The disconfirming evidence is on the record whether or not your bias wanted to keep it. Across a 15–25 stock portfolio, that written record is what stops a held position from quietly becoming a thesis you've stopped testing — nobody re-reads a year of transcripts for ten companies from memory.

Assign the verdict before you re-read, then check it. Write down what would break your thesis — "if the margin target is missed again or the growth guidance gets redefined once more, I'm wrong" — and then go look, specifically, for those things. Pre-committing to the disconfirming signal makes it much harder to explain away when it arrives.

Where we could be wrong: the opposite failure is real too, and it's symmetric. A redefined target is not always a confession — sometimes a 20% guide genuinely should be read over four or five years, and a company that drops a hydrogen target may have made the right capital-allocation call to chase solar instead. Hunt only for reasons to be bearish and you'll never hold anything long enough to compound; disconfirmation-seeking can curdle into a permanent, paralysed pessimism that misses every good business over a manageable flaw. The iron prescription cuts both ways: state the bear case better than the bears and the bull case better than the bulls, then weigh them honestly. The goal is not to talk yourself out of every stock. It is to make sure that when you stay in, it's because the thesis survived a real test — not because you only ever called your own witnesses.

What does honest research actually feel like?

It feels, briefly, worse. Done right, research should make you less comfortable at least some of the time — because you went looking for the thing that would prove you wrong and, occasionally, found it. The hour that ends with you more certain than you started, every single time, is not research. It is confirmation with extra steps.

The investor who can state the case against his own best idea better than its critics — and hold it anyway, because it survived — owns something most market participants never have: a conviction that has actually been tested. Everyone else just has a preference they've spent a few hours decorating with friendly evidence. So before you add to that position you're sure about, ask the owner's question, not the trader's: what would have to be true, five years out, for this to work — and have I gone looking, as hard as a bear would, for the evidence that it won't? If you haven't, you don't own a thesis yet. You own a hope you've been careful not to test.

Frequently asked questions