A thesis becomes useful at the moment it names what would refute it. This is how to write that line so it is checkable rather than rhetorical.
12 min read · Level 2
The band between "what I expect" and "what would make me wrong" is where the honesty of a thesis actually lives, and I think most investors set it far too narrow on purpose. A single hard threshold looks rigorous and is easy to renegotiate on the day, because a number breached by a small margin always has a reason attached to it - a definitional quirk, a one-off, a delayed report. A band forces you to say in advance how much of a shortfall is noise and how much is refutation, which is a genuinely hard judgement and the one you are being paid to make. My own view is that if the band feels uncomfortably wide when you write it, it is probably about right, because you are admitting how little precision the evidence will actually support. The investors I have seen handle disconfirmation well are not the ones with the sharpest triggers; they are the ones who wrote down beforehand how much ambiguity they were prepared to tolerate, and then did not move the line when the ambiguity arrived wearing a persuasive face.
A falsifier is a sentence of the form: if I observe X by date D, I will conclude this belief was mistaken. It is not a risk, not a worry, and not a list of things that could go wrong. It is one observable event that, if it happens, ends the argument.
Most attempts fail in one of four ways, and all four are common enough to be worth naming:
Unobservable "I am wrong if the market does not really want this." Nobody can check what a market really wants. Undated "I am wrong if growth stalls." Stalls when? A belief with no clock never expires. Unfalsifiable "I am wrong if the thesis does not play out." This is the belief restated with a negative in front of it. Too cheap "I am wrong if the company fails." True of every investment. A falsifier that fires only on total loss tells you nothing while there is still time to act.
A usable falsifier has three properties. It is observable - a fact you could look up or a thing you could see happen. It is dated - there is a moment when you must check. And it is specific to this belief rather than to investing in general, which means a different thesis would have a different falsifier.
Here is the same claim written badly and then well:
Weak "I believe embedded payments will grow. I am wrong if they do not." Strong "I believe vertical software companies will capture payment processing from generic processors, because the software already sits in the transaction flow. By the end of next year I expect the three largest companies in this category to report more than a third of revenue from payments. If payment revenue is still under a tenth of the total, the capture is not happening at the speed my reasoning requires, and I will treat the belief as mistaken rather than early."
The last clause is the hardest thing in this lesson to write and the most important. Mistaken rather than early is the distinction that separates a belief that can be tested from one that cannot, because "early" is always available as an explanation and it is the explanation that allows a thesis to survive indefinitely without ever being right.
Because confirmation bias does not feel like bias. It feels like paying attention. Once you hold a belief, the evidence that supports it is more interesting, more memorable, and easier to find, and you will find it without any dishonesty at all - the world produces enough of everything that a search will succeed.
A falsifier written in advance is the one tool that reliably interrupts this, and it works because of when it is written rather than what it says. Before you hold a position, you can think clearly about what would change your mind. Afterwards you cannot, because the answer has consequences. Writing it early is a way of using the clearer version of yourself to constrain the more invested one.
It also converts a vague unease into a scheduled decision. Investors who hold positions long past the point of doubt rarely ignored the evidence - more often they kept re-deciding, weekly, with the same inconclusive information, and each re-decision defaulted to holding because nothing had definitively changed. A dated falsifier replaces an infinite series of small evasions with one appointment.
The last reason is that a belief with a stated refutation is one you can be argued out of by somebody else. That sounds like a cost, and it is the point. An investor who cannot be argued out of anything has removed the only external error-correction available to them.
Educational, not personalised financial advice. Nothing in the Investor Academy is a recommendation to buy, sell or hold anything — what is right for you depends on your own circumstances and the rules where you live.
The belief. An investor thinks specialist lenders will take share from banks in small-business lending in one market, because banks have retreated from loans under a certain size and the fixed cost of underwriting is falling.
The first attempt at a falsifier. "I am wrong if specialist lenders do not grow." Useless: unobservable in practice, undated, and it restates the belief.
The second attempt. "I am wrong if bank share of small-business lending in this market is flat or rising two years from now." Better - observable from published statistics, dated, specific. But it has a flaw the investor catches on rereading: the thesis is about loans under a certain size, and the published statistic covers all small-business lending, most of it larger. The falsifier could fail to fire while the belief was already dead in the segment that matters.
The third attempt, which they keep. "By 31 December two years from now, I expect the share of new small-business loans under 250,000 originated by non-bank lenders in this market to exceed 25%, from about 15% today. If it is below 18%, the shift is not happening at the speed my reasoning requires, and I will treat the belief as mistaken rather than early. I will check the regulator's quarterly series each January."
Why that version works. There is a number, a source, a date and a threshold. There is a band between 18% and 25% where the investor has deliberately left themselves room to judge rather than pretending precision they do not have. And the phrase mistaken rather than early closes the escape hatch in advance, at the only time it can be closed honestly.
What it caught. Two years later the figure is 19%. The belief was directionally right and materially slower than the reasoning required - which, for a position sized on the fast version, is the same as being wrong. The investor reduces rather than abandons, and writes a new thesis with a slower clock. Without the number they would have described 15% to 19% as validation, because it is, directionally.
Writing a falsifier that can only fire once the loss has already happened. "I am wrong if the position is down 70%" is not a falsifier, it is a description of a bad outcome that has already occurred. A good falsifier fires on the mechanism rather than the price, and it fires while there is still time to act - which usually means it refers to something in the world rather than something in your account.
Do not start from risks. Start from the counterfactual: if this belief is simply wrong, what will be observably true in two years that is not true now? The answer to that question is your falsifier, and approaching it this way avoids the usual trap of listing worries instead of tests.
Name the statistic, where it is published, the threshold and the day you will look. If no such statistic exists, that is important information about your thesis: you have a belief you cannot monitor, which means you will end up judging it by the price of your own position, which is circular.
Set the level you expect and the level below which you will treat the belief as mistaken, and accept that between them lies judgement. False precision invites you to dismiss a real signal on a technicality, and a single hard threshold tends to get quietly renegotiated on the day it is breached.
Write the words: if this observation occurs, I will treat the belief as mistaken rather than early. Every thesis that dies slowly dies of that word, and the only moment you can honestly rule it out is before you own anything.
Apply it retrospectively. If it would have fired constantly it is too sensitive; if nothing in the last two years would have moved it, it is probably unobservable. This takes ten minutes and catches most unusable falsifiers before they cost anything.
Write the falsifier at the same sitting as the belief, before any position exists, and rewrite it whenever the belief changes materially. It matters most for slow-moving theses - infrastructure shifts, regulatory change, behavioural adoption - because those are the ones where "early" is genuinely plausible for years and therefore always available as a reason to do nothing.
Do not attach a falsifier to a position you hold for diversification rather than for a view. A broad holding taken precisely because you are not forecasting anything has no belief to refute, and inventing one turns a deliberately passive decision into something you will be tempted to trade. Falsifiers belong to claims, and only some holdings are claims.
An investor writes a clean, dated, numeric falsifier. The date arrives and the number comes in below the threshold. Rather than acting on it, they examine the measurement: the definition changed slightly, one large participant was reclassified, the reporting period was short. Each observation is factually correct. The conclusion drawn is that the falsifier did not really fire, and a revised version is written with a later date. Two years later the same thing happens again. Nothing dishonest occurred at any point, and the belief has now been immune to evidence for four years.
Decide in advance who checks the falsifier and write down what counts as a measurement dispute, or the check becomes a negotiation with yourself that you will win every time.
A belief about adoption is written with a falsifier about adoption rates, which duly continue rising. Meanwhile the cost of acquiring each new customer in that market triples, quietly, in the accounts of every company in the category. The thesis was about the destination and the falsifier watched the destination, so it never fired - but the economics of getting there had already broken, and the load-bearing assumption that was actually wrong was never the one being monitored.
List the load-bearing assumptions separately and check whether your falsifier watches the one most likely to fail, not the one easiest to measure.
Create a free account to read the rest and apply it to your own investing.