What is median relative error?

Median relative error is the midpoint of comparable relative errors and describes a typical case.

Plain meaning

Relative error = |predicted lots ÷ official lots − 1|. After sorting those errors, half are at or below the median. It is less pulled by extremes than an average, and it is not the same as P90, which describes harder cases.

Why it matters

Use the median for a typical case, time-weighted error for recent fit, and P90 for the tail. The three numbers answer different questions.

How to read it

Lower is better. A low median with a high P90 means typical tiers are acceptable but a few are far off. If both are high, treat any single-IPO figure more conservatively.

How it appears here

Allotment history can show median, time-weighted and P90 together. The headline time-weighted figure is an overall recent-sample summary, not one IPO's win rate.

Common mistakes

Treating the median as a guarantee for every tier.
Looking only at the median without P90.
Reading historical median error as a forecast interval for the next IPO.