Selection bias
How does picking which examples to show quietly decide the conclusion before any analysis begins?
In plain words
Selection bias arises when the set of observations you choose is not a fair sample of the whole, so the pattern you find reflects the choosing as much as the market. A week picked for being dramatic, or a run of weeks ending at a convenient point, can support almost any story.
Why it matters
Risk judgments built on hand-picked episodes understate how bad conditions can get and overstate how reliable a strategy looks. The honest test is whether the conclusion still holds when every eligible week is counted, including the unremarkable and the uncomfortable ones.
In a real week: BTC, week of 14 December 2020
Week of 14 December 2020: BTC returned 22.32%, with a 27.87% range and a return percentile of 99.0 versus its 148 weeks of history. Its prior four weeks were 15.4, -1.24, 6.46 and -0.95%, split two up and two down. Quoting only the strong up weeks would hide the mixed tape the gain emerged from.
A common mistake
Assuming a striking example is representative. One exceptional week, however well documented, says little about frequency unless the full population of weeks is counted alongside it.