Glossary

The handful of terms the studies use, in plain words, each with its source.

Disposition effect: The tendency to sell positions that are up more readily than positions that are down. Named by Shefrin and Statman (1985); measured on brokerage records by Odean (1998). Source: Shefrin & Statman (1985); Odean (1998)
PGR — proportion of gains realised: In Odean’s method: on the days a trader sells something, the share of their winning positions they actually sold, realised gains ÷ (realised gains + paper gains). The figures in this site’s public-panel study count every day a position is held instead, so there PGR is a daily rate: the share of held winning positions closed that day. Levels from the two conventions are not comparable; ratios are approximately comparable. Source: Odean (1998)
PLR — proportion of losses realised: The same share for losing positions: realised losses ÷ (realised losses + paper losses). If PGR is bigger than PLR, gains are being realised more readily than losses. The same two conventions apply as for PGR. Source: Odean (1998)
Paper gain / paper loss: A position that is up (or down) and was not sold that day. In Odean’s method only days on which the trader sold something else count; in this site’s public-panel study every held day counts. Source: Odean (1998)
Realised gain / realised loss: A position sold that day while it was up (or down). Source: Odean (1998)
Qualifying sale day: A day on which an account closed at least one position. In Odean’s method and in the client script (partner_disposition.py), only these days count, so a trader who rarely trades is not scored as disciplined. The public-panel study counts every held day instead; see PGR. Source: partner/ README
Orphan close: A closed position with no earlier day on which it was held as a paper position. A few are normal (positions opened and closed the same day); a majority means the export did not join correctly, and the script refuses it. Source: partner/ README
Denominator: Exactly who or what was counted underneath a percentage. Two honest studies can report very different levels because their denominators differ, which is why the ratio is the thing to compare. Source: this site’s studies
95% interval: A range that, by a stated method, would contain the true value in 95% of repeated samples. It shows how precisely the data pins a number down. Source: NIST/SEMATECH e-Handbook
t statistic: An estimate divided by its standard error. Large values mean the estimate is far from zero relative to its noise; with very large samples it gets large even for small effects, so it is not a measure of size. Source: NIST/SEMATECH e-Handbook
Form 4: The filing a company insider makes with the SEC after buying or selling the company’s shares. Source: Investor.gov