Published Trading Strategies Backtest: 20 Tests With Costs, None Passed

We ran a published trading strategies backtest with real retail costs, and none of the twenty tests passed: one was a candidate worth following, nineteen failed, and the mark had been written down before the first run. Many of the patterns in the papers are there before costs. A trade every day pays the spread every day, and that is larger than the effect.

This was run on 10 October 2026, and no window, pair or cost was changed after seeing a result. The chart shows all twenty against the mark, and the rules and results are in the tables below.

What was the pass mark for this published trading strategies backtest?

It was written on 10 October 2026, before any of them ran. Eighteen tests were planned, so one reaching a t-statistic of 2 by luck was likely. A test passed only if, after costs, all three of these held:

1. A t-statistic of the net return of 2.8 or more, which is 5% one-sided shared between the tests.
2. Net return above zero in each half of the sample taken alone.
3. Net return above zero in at least 60% of calendar years.

A net t of 2.0 to 2.8 that met the other two was a candidate: worth following on data it has not seen, not worth money. Everything else failed, and nothing was re-tuned after a fail. Two gold tests were added afterwards and before they ran, with the mark unchanged.

What was tested, on what data, and at what cost?

Twenty tests of sixteen rules, because four fixing windows are each judged twice: on the dollar against seven currencies in equal parts, and on EUR/USD alone. These rules come from published work, not from our courses, so they are stated in full.

The sources the study names, read at source:

• Fixing windows: Krohn, Mueller and Whelan, Foreign Exchange Fixings and Returns Around the Clock.
• Europe and US hours: Breedon and Ranaldo, Intraday Patterns in FX Returns and Order Flow.
• Fed day: Mueller, Tahbaz-Salehi and Vedolin, Exchange Rates and Monetary Policy Uncertainty.
• Month-end fix: Melvin and Prins, Equity Hedging and Exchange Rates at the London 4 p.m. Fix.

The gold tests are the rule published by Speck and others, as the study names it. We could not find that work at its source, so it has no citation here. Carry, trend and momentum come from the wide literature on them, and the study names no paper for those six tests.

Did any of the published strategies make money after costs?

A few stayed above zero, and none by enough. Results are net percent a year, because these tests have no stop, so R does not apply. Tests above zero are given in words.

The pre-Tokyo test moves its start from 17:00 in the paper to 18:00, because the rollover hour cannot be traded at retail.

Why did these academic trading strategies fail?

Mostly because the pattern is smaller than the cost of collecting it.

The patterns are real before costs. Gold made its whole rise outside London hours, as the rule says, and was about flat between the auctions, where the rule expects a fall. EUR/USD fell in Europe's hours and rose in America's. Trading either one every day pays the spread every day, and the spread is larger. That is the plainest lesson here, and it is the same one our test of the stop hunt as a mechanical rule taught us.

Carry, trend and momentum made nothing on the majors since 2006. Carry lost 28% in 2008, and has earned about what financing costs since.

Fed day and month-end are too small and too few. Their t-statistics of 1.7 and 1.9 cannot be told from luck, on 94 and 140 trading days. The month-end effect was also below zero in the second half. Trading an FOMC day carries its own risks, and a tiny average does not make them smaller.

The one candidate is fading. Short EUR/USD in Europe's hours was up in ten years of twelve, and the two down years are the last two: -3.2% in 2025 and -1.5% in 2026 to October. Its t-statistic of 2.35 is below the 2.8 that this many tests call for. We did not move its hours to find a better year.

What does this test not prove?

It does not prove that the papers are wrong or that these effects are gone from every market.

• It tests retail collection, not the pattern. A desk paying far less than a retail spread faces a different sum, and we did not test that.
• The mark is strict on purpose. A 2.35 would clear the usual bar of 2 if it were the only test ever run. It was not.
• It is a set of chosen rules. Twenty tests are not all of published research. Most of that literature is on stocks, which we did not test.
• The markets are narrow. Seven majors and gold intraday, eight currencies monthly. Emerging market carry is not here.
• It uses one source for each kind of data. Another broker's spreads and swaps will differ, and the financing markup is an assumption.
• It does not test any later refinement of these rules. Nothing was re-tuned, which keeps the test clean and also means no improved version got a chance.

A rule that fails after costs has told you something a rule with no costs stated never can. Why strategies fail backtests draws out the common causes, and the archive of our tests lists the others, including our own setup and Gann, on the research page.

What changed when this page came back?

This page was taken down on 31 October 2026 and put back on 11 November 2026. Its figures were rerun on 11 November 2026 with scripts/seo-research-figures.mjs, which reads the study's RESULTS.md. All twenty rows matched: none passed, one candidate, nineteen failed. No finding was reworded to read better than the numbers.

Course 3 of ForexiaPRO, Reverse Price Action, teaches our read of the chart step by step, and our own test of that read as a rule lost too.