Research
Every study below names the data it used and links its script, or says when the script is still to be published. Published whether or not the result was interesting.
New here? Start with the disposition-effect study, then the two negative results.
FEATURED · 6 OCT 2026
76.1%
About 3 in 4 traders realised gains more readily than losses: 3,547 of 4,659. Public data: Chinese A-share social-trading portfolios, 2016–18.
Open the chart full size (SVG, new tab) ↗
Per trader then averaged: ratio 1.61 (chart). Pooled across every observation: ratio 1.47 (PGR 0.0785, PLR 0.0535).
Update, 17 Oct 2026: the direction survives when holding age is held fixed; the size depends on how “gain” is timed, and a V-shaped selling pattern is not ruled out. Robustness update →
Chinese A-shares, Xueqiu social-trading portfolios, Jun 2016 – Mar 2018 · CC0 panel, Jin, Li & Zhu (2021) · reproduce.py · published 6 Oct 2026 · reviewed 17 Oct 2026
ALSO PUBLISHED · NEGATIVE RESULT · 11 OCT 2026
The insider-buying edge was mostly 2008.
Forty years of research said insiders beat the market. In 470,754 SEC Form 4 records, take out 2008–09 and an average 21-day edge above 0.62% is ruled out.
All studies
-
76.1% of 4,659 retail traders realised gains more readily than losses
A PGR/PLR replication of Odean (1998) on free CC0 data — 4,731 traders, 2,780,879 trader-stock-lot-days — and the three caveats that move the result. Robustness update, 17 Oct 2026: the direction survives a holding-age control; the size depends on how “gain” is timed, and a V-shaped selling pattern is not ruled out.
replication
-
The insider-buying edge was mostly 2008
Forty years of research said insiders beat the market. In 470,754 SEC Form 4 records, 2006–2026, two years supplied 94% of the excess return of the era that looked like it worked. Take them out and nothing is left.
negative resultinsider filings
-
A pre-registered replication of An et al. (2024) on the same CC0 panel. Its headline, that the preference for selling winners was about twice as strong when the rest of the portfolio was down (ratio 2.041, now corrected), is not supported: comparing each trader with themselves at the same holding day, the ratio is 1.03 (95% CI 0.95–1.13). Correction logged.
correctedpre-registeredreplication
-
A pre-registered test of volatility-scaled momentum failed its own kill criteria at five positions
The mechanism passed on 45.7 years of factor data (Sharpe 0.78, t 5.25); the five-position version with costs did not, and no money was committed. A negative research result, not a strategy or a signal.
negative resultpre-registered
-
We tested 26 swing-trading strategies under pre-registered rules. None survived.
Three monthly ETF rules and 23 daily swing rules on 22 ETFs, each sealed with a hash before any result was computed and judged once on a locked holdout, net of a 0.20% round-trip cost. None passed; 0 of 32 across the project. A negative research result, not a strategy or a signal.
negative resultpre-registered
-
We tested 50 trading strategies, from swing trades to day trades. None survived.
The swing study’s 32 strategies plus 18 new rules: 7 investing, 7 overnight and 4 day-trading, each sealed with a hash before any result was computed. 0 of 50 across the project; 0 of 44 under sealed pre-registration. The opening-range breakout on QQQ made +3.56 bp per trade before costs and lost it to 5.47 bp of costs. A negative research result, not a strategy or a signal.
negative resultpre-registered
Sample trader report
What the coming trader report looks like, run on one real trader from the same public CC0 panel as the disposition-effect study, chosen by a rule fixed before any individual result was looked at. Read the sample report →
In progress
Next study in progress: the disposition effect among on-chain perpetual-futures traders on Hyperliquid. Pre-registered before any results were computed; published whatever it finds.
Get the next study by email the day it’s published — with the script, and whichever way the result goes.