Lookback, threshold, and position-capacity for a stiffness trend-filter

Three layers before the filter is complete

A stiffness-style trend-filter is incomplete until three layers are specified together: the moving-average lookback that shapes the series, the stiffness-threshold and sampling interval that turn the reading into entries and exits, and the position-capacity queue that decides which of those signals are allowed to become positions.

Editorial reading: the plot on a chart is not yet a mechanical-trading-system. The system exists only when the filter rules and the capital, lot, and queue constraints are joined as one testable procedure.

Moving-average and stiffness-period

Two lookbacks define the series. The moving-average is a lookback smoother applied to ordered price observations. The stiffness-period is the second lookback on the stiffness calculation. Both are treated as free parameters.

The stiffness series had not been tested on intraday bars in the documented work. Any short-interval use was described as requiring parameter-retuning of the moving-average and stiffness-period inputs to the new sampling interval, the instrument's volatility, and prevailing market conditions.

On a 30-minute futures example the stiffness plot used a moving-average length of 100 and a stiffness-period of 70.

Before treating the filter as a system, the documented procedure was to optimize those parameters after a visual-range-estimate: a chart-side adjustment of the moving-average and stiffness-period controls used to sketch a plausible search band.

Threshold and sampling interval

The same 30-minute example used a stiffness-threshold of 90. A mechanical long entry was taken when the reading crossed above 90, and the position was exited when the reading crossed back below 90.

Editorial reading: the sampling interval belongs in the specification, because a change of bar length means the same numeric lookbacks no longer describe the same trend-filter.

Position-capacity and the surplus-signal-queue

The portfolio test used 200000 of starting capital, a 10000 maximum trade size, a 10-share round-lot, and a 1000 minimum position value. It did not use leverage, margin, or pyramiding.

Those capital rules create position-capacity: an implicit limit on concurrent holdings from starting capital and maximum trade size, before any ranking function is added. Dividing starting capital by maximum trade size implied 20 names.

When more names signaled than cash or that implied limit allowed, later entries were ignored until an exit freed capital. That surplus-signal-queue skips unused entries rather than financing them with leverage or pyramiding. A ranking score was not applied, so larger eligible size was preferred only by the platform default.

Because the round-lot rule kept each fill below the 10000 size cap, the number of concurrent holdings could exceed that 20-name implication.