Separating movement, churn, and trading cost

Some charts appear to sustain a direction while shaking a trade along the way. Could we describe that behaviour without confusing a large move with a noisy one?

The study defined the question at an hourly holding scale. A market that repeatedly gives back the previous hour's move differs from one that makes several-hour legs in both directions, even when both finish a session near their starting point.

We compared the share of consecutive hours moving in the same direction, the relationship between neighbouring returns, and displacement relative to the total path over eight hours. Mid prices reduced the mechanical bounce that can arise when a series alternates between bid and ask.

A separate trade simulation measured shake-outs and adverse excursions, with and without the study's spread charge. Currency-leg comparisons examined how co-moving currencies could leave a smaller movement in the pair between them.

Results were computed per instrument and split by sample half and session. A single pooled score would have hidden differences and unstable rankings.

Some pairs differed visibly at the extremes of the hourly persistence rankings. The middle of the ranking was much less stable. An instrument's appearance also depended on the timescale and session being measured.

Much of the difference in the simulated trading-cost ranking came from spread relative to the distance traded. Removing spread made the double-stop-out comparison substantially flatter. Some relationship between churn and deeper excursions remained, but it was smaller than a simple clean-versus-noisy label suggested.

When two currency legs moved together, their common movement partly cancelled in the cross. That made the pair smaller. Whether that smaller pair was expensive still depended on the spread charged for trading it.

Recent clean or churny days did not establish a useful next-trade timing signal in these tests. A description of the past path did not automatically predict the next one.

Later pair-selection rankings and live-fill comparisons are excluded from this public finding. Their calculations contain unresolved defects, and the small live-fill sample could not establish a reliable performance effect. We do not use them to recommend particular pairs or to claim a trading improvement.

This is a comparison of one broker's hourly FX data and specified simulations. Lower-timeframe behaviour needs its own evidence. The proposed explanations involving participant behaviour were not directly observed in order-flow data.

A noise measure is most useful when its unit and clock are explicit. It supplies a way to describe what a chart is doing; it does not decide whether a particular trader has an advantage there.