A public record of trading strategies I tested and killed.
Write-ups: snipergoate.substack.com · X: @snipergoate
The method, short version: every study is pre-registered. Hypothesis, sample, significance gates, and kill criteria get written down and locked before the data gets a vote. Significance bars scale with the number of things tried, because a permutation sweep is a multiple-comparisons machine. Everything is scored at brutal retail costs (buy the ask, sell the bid) and confirmed on data the rule never saw. What fails is killed, documented, and published here with code and cell-level receipts.
Why publish the kills? Because the incentives of trading content all point the other way. Winners get published, losers get buried, and the survivors' parameters are for sale. A public graveyard with receipts is harder to fake than a track record. And negative results are most of what honest research produces.
| Campaign | The belief on trial | Verdict |
|---|---|---|
| The TA Files (8 parts) | Technical analysis predicts price | 31,600 backtests, 0 survived. Closed permanently. |
More campaigns queued. One kill at a time, receipts every time.
Each campaign folder ships code/make_figs.py (Python 3 plus matplotlib), the
derived cell-level statistics it reads (receipts/), and the pre-commitment
document (REGISTRATION.md). Raw vendor market data is not redistributed. The
receipts are derived aggregates sufficient to verify every claim and figure.
Research journal, not investment advice. Everything here documents strategies I tested and killed.