Outlier
What makes a value an outlier, and when should a researcher keep it rather than throw it away?
In plain words
An outlier is a value that sits far from the rest of a set of observations, such as a single hour, day or week whose move dwarfs everything around it. It may be genuine or a data error, and deciding which is part of the work.
Why it matters
Discarding outliers can make a market look calmer and safer than it is, since the largest moves are usually the ones that clear positions. Keeping them without checking can let a bad tick distort an average. The first step is asking whether the value could really have happened.
In a real week: BTC, week of 16 March 2020
In the week of 16 March 2020, BTC ranged 55.33% between a low of 4442.12 on Monday and a high of 6900.0 on Friday, with realised volatility at 165.6% annualised. The largest 1h move was -7.67% at 10:00 UTC on Monday, and the best day, 19 March, gained 14.26% against a worst day of -6.18%. Volume of 6.72 billion was 1.87 times the prior 12-week average. The data flags no figure as faulty; one possible reading is that the market really moved this far that week.
A common mistake
Treating any extreme number as bad data. A value should be checked against related measurements before removal; if several independent measures agree, the outlier is the market, not the record.