Popular trading ideas, put through a proper test on real data: no hindsight, trading costs included, all the major pairs, and a straight verdict at the end. Including the ones we use ourselves.
Verdicts are "supported", "not supported" or "mixed", based on whether the idea made money after costs, held up in both halves of the data, and worked across most pairs rather than one.
Our own trend label, tested on seven majors over twenty years. Following it lost about 1.2% a year after costs, and its gains came mainly in 2008.
Extreme COT positioning, tested on seven currencies since 2004. Fading it roughly broke even (-0.1% a year) with a painful 2008, and on the euro the crowd was usually right.
The textbook high-yield v low-yield strategy since 2004. The interest is real (+1.5% a year net), but 2008 took 25%, and a typical retail swap mark-up turns it negative.
Fifteen markets since 2004. Real pivots held 64.2% of the time; decoy levels at the same distances held 64.1%. No meaningful difference.
Yields and currencies do move together in the same month (correlations +0.22 to +0.54), but the gap's change told us nothing about the next month, and trading it lost money after costs.
Brokers publish what percentage of their clients are long or short, and it's often sold as a contrarian signal. We'd love to test it, but there's no free, reliable history of that data to test against. Rather than guess, we're saying so. If that changes, it goes to the top of the list.