Most strategies begin as an observation: prices seem to keep moving after big news, or certain assets tend to rebound after sharp falls. These observations are useful starting points but are too vague to test. The first step is to turn them into a hypothesis.

Ingredients of a good hypothesis

  • Universe: which instruments, over which period?
  • Signal: exactly what is measured, and when?
  • Action: what position is taken, and for how long?
  • Expected effect: what result would support the idea — and what would refute it?

An example

Vague: “Stocks with good news go up.” Testable: “Among large-cap US stocks from 2010 to 2024, those whose news sentiment score rises by more than two standard deviations relative to their own 90-day history outperform their sector over the following five trading days, after costs.”

Why an economic rationale helps

A hypothesis with a plausible reason — investors under-react to information, or some participants are forced sellers — is more likely to persist than a pattern found by searching data. Patterns without a rationale are frequently statistical accidents.

Write the hypothesis down before looking at results. This simple habit is one of the strongest defences against fooling yourself.

Check your understanding

Why write the hypothesis before testing?

To prevent adjusting the question to fit the results, which leads to overfitting.

What four elements define a testable hypothesis?

Universe, signal, action and expected effect (including what would refute it).