What the Judge Answered to Old Trading Rules: 37 Proofs, 0 Survived
As of 31 October 2026, the Lab's judge had returned 37 proofs on 9 old trading rules: 0 survived, 21 mirages, 16 inconclusive. These figures are already published on this site: this page gathers them with their source, their date and their limits. It adds no new calculation.
Page published on 2 November 2026. Research on past data, not investment advice.
The figures and their sources
| Figure | What it measures | Date | Source |
|---|---|---|---|
| 37 proofs: 0 survived, 21 mirages, 16 inconclusive | 9 old rules frozen before the test | 31 October 2026 | Faber article, update |
| 1,306 trials, no survivor | An AI searching during one session, across 7 families | 12 October 2026 | Same article |
| 0 "survives" in 350 sessions | Automated research on pure noise, a control and not old rules | 11 October 2026 | MCP guide, details |
| +172.7% then −37% | Champion of a genetic algorithm on Bitcoin, training then unseen data | 16 August 2026 | Post-mortem |
The 37 proofs in detail
| Rule | Markets | Verdicts | Published reading |
|---|---|---|---|
| Faber 10 months | S&P, US, world ex-US, gold | 4 inconclusive | Lower gain, much smaller drawdown |
| 12-month momentum | 4 regions, gold, 3 currencies | 3 mirages, 5 inconclusive | Below buy and hold everywhere |
| Halloween | Japan, Asia, North America | 1 mirage, 2 inconclusive | Below buy and hold |
| Turn of the month | US, Europe, Japan, Asia | 2 mirages, 2 inconclusive | 219 trades: fees crush it |
| Dual momentum | US versus ex-US | Inconclusive | +131% versus +282% |
| Sector momentum 12-1 | 10 US sectors | Inconclusive | +3,605% versus +1,897% |
| Low volatility | 10 US sectors | Inconclusive | +1,042% versus +1,791% |
| Connors RSI(2) | Europe, Japan, Asia | 3 mirages | +12%, +32%, −26% |
| Turtle S1 and S2 | 6 currencies | 12 mirages | A loss on 11 of 12 |
One example: Faber, since July 2007
Strategy versus buy and hold, 0.1% fees per fill, on the two series using month-end returns:
| Market | Last data point | Gain | Maximum drawdown |
|---|---|---|---|
| US equities (Ken French) | August 2026 | +429% / +665% | 19% / 50% |
| Developed ex-US (Ken French) | August 2026 | +107% / +170% | 26% / 56% |
Faber stays inconclusive: it does not beat the market anywhere. The drawdown is measured on this one period, and the judge does not count it in its verdict. The S&P 500 and gold rows of the original article are not repeated here: those series are monthly averages, with a bias flagged on 31 October.
The protocol
- Frozen before the test. On 12 October 2026: nine rules, their markets and a control.
- Judged on 31 October 2026 by the Lab 253 judge, with 10 bps of fees and no parameter sweep.
- Reference. Buy and hold, on the same bars. The judge's method describes the sealed period and the controls.
The limits
- 37 proofs are not 37 ideas. They cover 9 rules, some on several markets.
- The control was not judged. The SMA 200 exceeds the engine's compute budget.
- A Lab defect affected three rules. The rotations (dual momentum, sector momentum, low volatility) were judged on the last 30% of their history, not from the pre-registered date.
- Halloween Europe is missing. That band, already opened once, was not replayed.
- Sector momentum needs care. It is the only rule above its reference, from September 1996 to August 2026, and its verdict stays inconclusive. At 30 bps of fees the gap is no longer significant; across 4 regions (2015-2026) the rule fails: +162% versus +200%. Ken French portfolios are not tradable funds. Across 37 proofs, one isolated result at this level can come from chance.
- These periods are now open. They can no longer serve as an untouched test for these rule families.
- The judge has had defects, published. Its current version is 254; these proofs come from 253.
- The 350 sessions are synthetic noise. They show that the protocol lets no false positive through on that noise (95% interval: 0 to 1.1%), not that a real rule fails. On a clear planted signal, the loop confirms it in only 8% of sessions.
- The +172.7% does not come from an old rule. It is the result of a GPU search, cited for the gap between training and unseen data.
This result does not say these rules will never work elsewhere. It says that on these markets and periods, fees included, none obtained "survives". To test a script yourself: the online judge or the tool comparison.
FAQ
How many old rules survived the judge?
None. As of 31 October 2026, the Lab 253 judge returned 37 proofs on 9 rules frozen on 12 October: 0 survived, 21 mirages, 16 inconclusive, with 10 bps of fees and no parameter sweep.
Does inconclusive mean the rule is wrong?
No. It means the judge could not conclude that it beats buy and hold over the period tested. It says nothing about other markets or the next period.
Do these figures say what to buy or sell?
No. These are research results on past data, with their limits. They are neither a forecast nor advice.
Research on past data, not investment advice. Past results do not guarantee future results.
⚠️ Disclaimer — This article is for informational and educational purposes only. It does not constitute investment advice or a buy/sell recommendation. Past performance does not guarantee future results. Strategy Arena is an educational simulator with virtual capital. Always do your own research before making investment decisions.