Walk-forward evaluation of stop overlays on average crossovers

Two decisions, not one search

The archive workflow separated the entry rule from the overlay that bounded or timed it. A triple-average-offset procedure was chosen first on a completed year. Stop and fill-timing overlays were then applied to that already locked rule and scored on the next unused quarter.

The same three-average rule was also run without a money-management overlay so that each overlay could be compared with an otherwise identical benchmark.

The locked entry rule

The triple-average-offset ranking used a short, an intermediate and a long simple average. The rule went long only when the short average was above the intermediate and the intermediate was above the long. It went short on the inverse ranking.

It did not stay continuously positioned. After entry it returned to flat when a later short-versus-intermediate cross appeared.

Average lengths were taken from a predeclared search grid of short, intermediate and long spans. That search belonged to the training year, not to the unused quarter.

The walk-forward-window

Each walk-forward-window was a completed optimization year followed by one unused quarter. An in-sample-lock selected one parameter combination on the training year with a predeclared objective and then held that combination fixed on the unused quarter.

Both segments then advanced by one quarter and the search was repeated. The unused quarter therefore received only parameters that had already been locked.

Stop and fill-timing overlays

Several stop-loss constructions were overlaid on the same locked entry rule. An entry-stop closed the position if a precommitted adverse-move distance from the fill was reached. A trailing-stop matched that bound on an immediate adverse move and otherwise ratcheted with the most favorable daily settlement. A profit-stop closed the position once a precommitted favorable-move distance from the fill was reached. One search joined an entry-stop with a profit-stop.

A fill-timing-overlay constrained whether a generated signal was executed on the close, on the next open, or only after a confirming or contrary price move.

The standalone entry-stop search added a grid of stop distances to the already defined average-length grid. The combined parameter set was still locked on the training year before it was scored on unused time.

Fills, rollovers and the unused-quarter score

Default fills were taken at the next session's open after a signal. Expiration-month rollovers were closed on the first session of the delivery month. Other overlays required a same-session close or a next-open fill only after a favorable or unfavorable price confirmation.

The unused-quarter score was quarterly net return on a contract. That figure subtracted a round-turn charge and interest on exchange margin at the prevailing Treasury bill rate. There was no interest credit on equity above the exchange minimum margin.

Those overlays were compared with a benchmark that used the same triple-average-offset rule and no money-management overlay.