donttrade: studies and corrections

Correction: Site-wide wording and repository READMEs (19 Oct 2026)

2026-10-19T00:00:00Z

What was wrong: (1) The repository’s README.md and pdde/README.md still gave the portfolio-driven disposition effect headline corrected on 17 Oct 2026 (“It replicates”, ratio 2.041; “Result: replicated”). (2) Pages said the scripts behind every headline figure are public; they are public for 4 of the 6 studies. (3) The sample report called its comparison group of 2,262 traders “with at least 30 sell-days”, the same words it used for the 2,264 traders it chose from, and the script it links to, samples/sample_trader_report.py, was not in the repository. (4) Two studies described the author’s own re-computations as “independent passes”. (5) The specimen memo still read “Cohen’s d of 0.49 is the honest magnitude”, wording changed on the study on 17 Oct 2026, and “Both are above.” had lost its “1”. (6) The homepage refund line left out two conditions in the terms: the same script version and the same export. (7) Before 14 Oct 2026, requirements.txt allowed pandas>=1.3, with which a clean checkout did not run, and reproduce.py printed “levels differ by 4-6x; the ratio moves <5%”. What changed: (1) Both READMEs now carry a dated correction (17 Oct 2026) at those statements, linking the corrected study and this log; the original figures are kept, marked as superseded. (2) Now “scripts public for 4 of 6”; the other 2 are listed, with their gaps, on each study. (3) The 2,262 group now reads “with at least 30 sell-days and both a PGR and a PLR defined”; the 2,264 sentence is unchanged. The script is now in the repository and prints the same wording. (4) Now “re-computed three separate times by the author”; the heading “On verification” is now “How this was checked”. (5) Now “the honest magnitude of this measure”, with a link to the study’s robustness update, and “Both are above 1”. (6) Now “(same script version, same export)”; /pricing/ and /for-firms/ use the same conditions as the terms. No price or refund term changed. (7) On 14 Oct 2026, requirements.txt was changed to pandas>=2.2 and the printed claim to “levels differ by nearly 4x; the ratio moves <7%”. That change was not logged here at the time; it is now. Items 1–6 change wording only; no computed number changed.

Study: Fifty trading strategies across four styles, a combined negative result (17 Oct 2026)

2026-10-17T00:00:00Z

A new study; its code is to be published with it. Published whether or not the result was interesting.

Correction: Portfolio-driven disposition effect (17 Oct 2026)

2026-10-17T00:00:00Z

What was wrong: The headline said retail traders’ preference for selling winners was “about twice as strong on days their other holdings were mostly down” (ratio 2.041). A pre-registered robustness test that compares each trader with themselves at the same holding day gives 1.03 (95% CI 0.95–1.13), and the specification the study used reports an effect of similar size in synthetic panels that contain none. The headline claim is not supported. What changed: The title, summary, list entries, page description and social card now say the headline did not survive a stricter test. A dated correction box sits at the top of the study and the robustness update is in its body; the original figure is kept there, marked as corrected. The replication verdict recorded on 10 Oct 2026 is kept as the record of what was run; it no longer supports the headline.

Study: Volatility-scaled momentum, a pre-registered negative result (16 Oct 2026)

2026-10-16T00:00:00Z

A new study, with its script and data linked. Published whether or not the result was interesting.

Study: Swing-trading strategies under sealed pre-registration, a negative result (16 Oct 2026)

2026-10-16T00:00:00Z

A new study; its code is to be published with it. Published whether or not the result was interesting.

Study: Portfolio-driven disposition effect: the headline did not survive a stricter test (corrected 17 Oct 2026)

2026-10-16T00:00:00Z

A new study, with its script and data linked. Published whether or not the result was interesting.

Correction: Volatility-scaled momentum (16 Oct 2026)

2026-10-16T00:00:00Z

What was wrong: volmom/reproduce_volmom.py labelled per-side breakeven costs “round-trip”, printed K7 FAIL for all three specifications, and described the 10% drawdown limit as applied. What changed: The script now prints per-side and round-trip breakevens, compares the round trip with K7’s 0.50% floor (S1 pass, S2 fail, S3 pass) and says the limit is counted, not applied. In volmom/README.md the S2 round trip now reads 0.467% (the script’s figure; 0.468% was a doubling of a rounded number), the holdout coverage reads 84.8% as the script prints on a fresh download (85.0% could not be reproduced), and the replication window is labelled November 1927 – December 2011. No computed number changed; the verdict, no specification passes all four, is unchanged.

Correction: Disposition effect; portfolio-driven disposition effect (16 Oct 2026)

2026-10-16T00:00:00Z

What was wrong: reproduce.py and pdde/reproduce_pdde.py downloaded the CC0 panel with Python’s default client, which Harvard Dataverse refused with HTTP 403, so a clean checkout could not fetch the data. What changed: Both scripts now send an explicit User-Agent and write to a temporary file first; the downloads in volmom/ send the same User-Agent. Re-run end to end from fresh downloads: no figure changed.

Study: Insider-buying study, a negative result (11 Oct 2026)

2026-10-11T00:00:00Z

A new study, with its script and data linked. Published whether or not the result was interesting.

Correction: Insider-buying edge (11 Oct 2026)

2026-10-11T00:00:00Z

What was wrong: An earlier draft claimed the 63-day horizon told the same story as the 21-day one (2017–2026: −0.41%, t −0.71). What changed: Withdrawn. On the corrected screen the 63-day figure is +0.93% (t 1.45), the opposite sign, and the sign of that 2,715-event result is set by 66 observations. No claim is made about the three-month horizon.

Correction: Insider-buying edge (11 Oct 2026)

2026-10-11T00:00:00Z

What was wrong: An earlier version of the pipeline claimed to exclude 10b5-1 plan transactions. It never did: the flag lives in the SEC's SUBMISSION table and the code looked for it in the transactions table, where it does not exist, and the reader silently dropped the missing column. What changed: The filter is now applied where the field actually is. It exists only from 2023 Q1 (13 of 81 quarters) and removes 3.89% of filing-owner records in those quarters. The two eras are therefore not filtered alike, and the post says so.

Study: Disposition-effect replication (6 Oct 2026)

2026-10-06T00:00:00Z

A new study, with its script and data linked. Published whether or not the result was interesting.