Why a daily timing evaluation depends on interval, lookbacks, and the fitting objective

The daily procedure as specified

A daily Mechanical trading system can be specified from a utilities-to-broad-market Relative Strength Index series plus separate up-side and down-side lookbacks. That Relative Strength Index series, read with those lookbacks, is the Trend filter inside the procedure. The Mechanical trading system is the full rule set: when a buy is allowed, when a sell is allowed, and when the system may stand aside.

The series used to select parameters

A log transform of the target price series is used so successive differences approximate percentage change and can serve as the series on which parameters are selected. The Relative Strength Index construction itself is restated in log differences so unequal long-run growth of the two averages does not distort the magnitude of the indicator.

Optimizing on summed percentage outcomes, rather than raw price differences, is presented as a way to stop later, larger-level observations from dominating earlier equal-percentage moves. The fitting objective is therefore part of the model, not a neutral scoring step added after the Trend filter is already fixed.

Independent lookbacks and the holding rule

Buy and sell lookbacks are specified independently (UpLag versus DnLag) because the two signal types are treated as different procedures. A shared lookback would have forced the buy rule and the sell rule to share one memory length. Independent lookbacks do not.

Why a locked hold was removed

A minimum locked holding period after a buy is treated as an unacceptable real-world constraint and is removed so an exit can occur without a forced wait. That choice changes the Mechanical trading system under evaluation: an exit may arrive as soon as the sell lookback fires, rather than after a required pause.

What the evaluation can claim

Editorial reading: daily sampling chooses the observation grid for the Trend filter. Independent UpLag and DnLag lookbacks mean the buy rule and the sell rule are not required to share a memory length. The log restatement and the summed-percentage objective decide which moves dominate parameter choice.

Those three design decisions are part of the object under test. The historical evaluation can therefore speak only to this daily Mechanical trading system as specified. It does not, on its own, speak to a different sampling interval, a single shared lookback, or an objective based on raw price differences.