Confirmation and the Story After the Purchase

The order the evidence arrives in

Confirmation bias is usually described as a preference for agreeing information, which makes it sound like a taste. The mechanism is more specific and more uncomfortable: a person who has already acted has a stake in the decision being right, and the search for information is shaped by that stake. What gets read, which comparison gets made, which peer is chosen for the multiple, which quarter is called a turning point — each of these is a choice, and a choice made after a purchase is made by someone with a position to defend. The asymmetry is easiest to see in the sourcing rather than in the reasoning. A holder reads the bear case looking for its weak points, and an observer reads it looking for its argument. A holder selects the peer group that makes the company look best-disciplined, and an observer selects the peers that share its economics. Neither is lying and neither is necessarily wrong; the difference is that one of them has already answered the question they are now investigating. This is why the bias survives intelligence: it does not degrade the analysis, it decides what gets analysed. There is a second-order version that is worth naming because it is the one that costs the most. Because a defended thesis tends to accumulate evidence and shed doubt, the holder’s confidence rises with time rather than with the quality of the case. Two quarters of ordinary results, each one absorbed through a bull frame, produce a position that feels better supported than it did when it was opened — and a conviction level that has not been paid for by anything. The honest reading of accumulating evidence is the one that asks what share of it was available before the decision, and what has genuinely changed. One report, two theses — Revenue misses by 4%: Read as a cyclical wobble, or as demand cracking · Gross margin beats by 3%: Read as pricing power, or as a one-off mix shift ← · The reading chosen: Depends on the position, not on the facts ← · What would settle it: A question asked before the position existed A testable habit: write the one fact that would change your mind and the date you will check it. If a fact appears that is not on the list, it is allowed not to change anything — the list is what keeps the argument from being rewritten after each report.

Asking for the falsifier

The practical countermeasure is to make the disconfirming question part of the research rather than an act of virtue. Before a position exists, write down what would have to be true for the thesis to be wrong, in terms specific enough to check: a margin level, a growth rate, a competitor launch, a refinancing date. The value of writing it before the purchase is that the same person is doing the writing, with no position to protect and no result to explain. The second move is to read the strongest opposing case as its author would write it — not hunting for the sentence that can be dismissed, but stating the argument in a form its holder would accept. That exercise is uncomfortable in proportion to how much is riding on the position, which is the point: the discomfort is the signal that the search is being organised by the conclusion. The version of this that survives contact with a busy schedule is smaller and more mechanical: for each holding, name one thing that would have to be true for the bear case to be right, and check it on a date. The third move is social rather than procedural. A thesis that is never argued against is a thesis that has not been tested, so the useful habit is to have the argument somewhere it can lose: a written note that others can read, a recorded view that can be checked later, or a conversation with someone who owns nothing. What all three have in common is that the thesis becomes a claim rather than a summary; a claim can be wrong, and being wrong on the record is what makes the next claim more careful. • Write the falsifier before the position, in terms specific enough to check. • Read the strongest opposing case as its author would write it. • For each holding, name one thing the bear case needs and check it on a date. • Put the view where it can lose — written, recorded or argued with someone who owns nothing. The sign that a thesis has stopped being a thesis is that every new fact makes it stronger. Real evidence is mixed, and a position whose supporting case has grown for two quarters without a single uncomfortable development in the notes has not been confirmed — it has been curated.

Where you go looking, and what that decides

Confirmation is mostly not a failure of reasoning — it is a failure of **search**. The reasoning is usually fine; the inputs are selected. Once a position exists, the natural information flow is one-sided in ways that are invisible from inside it. The feed you read is whatever you chose to follow, and it is populated by people who hold the same view because that is what you found interesting before you had a position. The company’s own materials are written by people paid to present it favourably. The analyst who agrees is quoted; the analyst who disagrees is dismissed as not understanding the business. None of those steps feels like bias, because none of them involves ignoring a fact. The cheap structural fix is to decide **who you would have to hear from** before you need to hear it, and then go there deliberately. For a long position that means the short thesis — the write-up, the bear case, the person who has done the work on the other side — read in full rather than skimmed for flaws. For a valuation it means the reverse of your own assumption: if your base case is twelve percent growth, look at what the price implies and at what the company would have to do to miss. The test of whether a piece of research is real work is whether it could change your mind. If nothing you read today could, you were not researching. There is a social version of the same trap worth naming separately, because it is stronger than most people expect. Holding the same asset as your friends converts a disagreeing argument into a disagreement with people, which raises the cost of changing your mind well beyond the money. That is not a reason to avoid talking about positions; it is a reason to notice when the reason you are confident has changed from the business to the group. A thesis that requires the social group to stay intact is a thesis with an extra, invisible risk attached. • Bias enters through search, not through logic — the inputs are selected before the reasoning starts. • Decide in advance whose contrary view you will actually read, then read it in full. • If no available evidence could move you, what you are doing is not research. • Sharing a position with a group raises the social cost of changing your mind. A practical discipline: write the disconfirming case first, in one paragraph, before the bull case. It is much easier to be honest about what would break the idea when you do not yet have a position to defend.

Building a process that is allowed to disagree with you

Knowing that confirmation bias exists changes very little, because the bias does not announce itself — it operates through the questions you think to ask. What changes behaviour is a procedure that asks the uncomfortable questions on your behalf, at a fixed time, whether or not you are in the mood. The interventions with the best support share one feature: they make the opposing case concrete rather than abstract. Simply reminding yourself to be open-minded does nothing measurable. Being made to write out the strongest version of the other side, in full, and to specify what evidence would change your mind, does. That is the finding behind the “consider the opposite” line of work, and it is the reason a written falsifier beats a general intention to stay humble. A workable process has four pieces. The **falsifier**, written before the position exists: the specific observation that would show the thesis is wrong, which turns a vague confidence into a claim that can fail. The **pre-mortem**: a short written answer to “it is a year from now and this went badly — what happened?”, which surfaces risks the optimistic framing never generates. The **red team**: someone, or a structured exercise, whose assigned job is to argue the other side, because an argument with no assigned advocate is an argument that does not happen. And the **checklist**, applied identically to every candidate, so that the evidence assembled for the tenth idea is the same kind of evidence assembled for the first. The fourth piece is the one most often skipped and the one that matters most in practice: a scheduled review that compares the thesis against what actually happened, using the falsifier rather than the price. Without it, the process produces documents; with it, it produces learning, because the question is not whether the position made money but whether the mechanism behaved as described. Two failure modes are worth naming, because both feel like rigour. The first is the red team that is really a straw man — arguing against a version of the thesis weaker than the one you hold, which produces confidence rather than doubt. The second is the review that becomes a scorecard of outcomes, which rewards a badly reasoned winner and punishes a well-reasoned loser and teaches the wrong lesson in both directions. • Concrete opposition changes behaviour; general open-mindedness does not. • Write the falsifier before the position, and the pre-mortem before the optimism hardens. • Assign somebody the other side, or the argument does not get made. • Review the mechanism, not the outcome — that is what makes the record informative. A practical test of any research process: look at the last three positions and find the written falsifier for each. If you cannot find one, the process has no way of telling you that you were wrong, which means it can only ever confirm.

Sources

• Confirmation bias and hypothesis testingKahneman, “Thinking, Fast and Slow”
• Belief perseverance and the psychology of predictionRoss, Lepper & Hubbard (1975), “Perseverance in self-perception”
• Disconfirming evidence in investment researchMauboussin, “Think Twice”