"Stocks fall as oil climbs." "Tech rallies on rate-cut hopes." Financial news needs a reason for every move, and it needs it fast. The problem is that two things happening at the same time is not evidence that one caused the other. On any given day, dozens of stories are published while prices move for many reasons at once: fund flows, options expiries, positioning, and news that was already expected.
Confusing correlation with causation is not an academic mistake. It leads investors to react to the wrong signal, to build strategies on relationships that break without warning, and to feel more certain than the evidence allows.
Correlation describes. Causation explains.
Correlation measures how two variables move together. Causation means that a change in one actually produces a change in the other. A correlation can be real and stable without being causal: ice-cream sales and swimming accidents both rise in summer, but neither causes the other — hot weather drives both.
Markets are full of these hidden third variables. Two stocks can rise together because investors are buying an entire sector, not because of anything either company announced. A currency and a commodity can move in lockstep because both respond to interest-rate expectations.
The famous coincidences
Search long enough and you will find patterns that look predictive. The best known is the "Super Bowl indicator", which claimed that the league of the winning team foretold the stock market's direction for the year. It worked for a surprising number of years — and then it didn't, because there was never any mechanism connecting football to corporate earnings.
With enough data, some patterns will always appear by chance. The question is never "do these two things move together?" but "is there a reason they should?"
This is why trading signals discovered by mining historical data so often fail on new data. The more relationships you test, the more coincidences you find.
A five-question checklist
When you read an explanation for a market move — in the news, on social media, or in an AI-generated summary — run through these questions:
- Is there a plausible mechanism? Can you describe, step by step, how the event would change expected cash flows, interest rates or risk? If the explanation is only "investors reacted to the news", there is no mechanism yet.
- Did the timing line up? Did the move start when the news became public, or before? A price that was already falling for an hour before a headline was not caused by the headline.
- Was it a surprise? Markets move on the difference between what happened and what was expected. News that was fully anticipated is usually already priced in.
- What else could explain it? Look for confounders: did the whole sector move? Was there macro data at the same time? Is it earnings season?
- Is there second-level evidence? Beyond the price itself, is there supporting data — trading volume, options activity, a quantified company statement, or a similar past event with a similar reaction?
What a real mechanism looks like
A mechanism is a chain of cause and effect that could, in principle, be checked. For example, an energy supply shock can move growth stocks through a sequence like this:
- eventEnergy supply cut announced
- effect 1Higher inflation expectations
- effect 2Higher bond yields
- price impactLower valuations for long-duration growth stocks
Each link can be tested: did inflation expectations actually rise? Did yields move? Did the most rate-sensitive stocks fall the most? If a link is missing, the story is weaker than it sounds.
Why this matters for AI-generated analysis
Language models are extremely good at producing fluent explanations. That is exactly the risk: a model can generate a convincing causal story for any price move, whether or not the data supports it. Phrases like "reflecting growing optimism" or "as the news attracted investor attention" sound analytical but explain nothing.
That is why every key event in a CausifyMarket report separates the causal read from the factual summary, lists the confounders considered, and assigns an explicit confidence level. When the data cannot separate an event from its confounders, the report says so — see why we sometimes say "undetermined".
The takeaway
You do not need statistics software to think causally. You need the habit of asking how, when, compared to what, and what else. Explanations that survive those questions deserve your attention. Explanations that don't are just headlines.