Regime-dependent long and short rules in mechanical systems

A combined book can hide two markets

A mechanical-trading-system is a fully specified set of entry, exit and abstention rules that can be tested as one procedure. That single procedure still has two sides. A reader asked whether mechanical tests should be run separately for long and short positions because up and down markets are often described as having distinct structure.

Editorial reading: one rulebook can hide two markets when long and short trades are scored together. A long-short-split-test treats each side as its own sample. Trend-following enters and holds with the prevailing directional move and should be judged where that move produces the entry. Mean-reversion fades an excursion toward a reference level once that excursion is defined by rules, and should be judged where the defined excursion produces the fade. Regime-dependence is the name for that split.

Read every threshold as a number

A published mechanical buy rule used an advance-decline-ratio that had to be greater than or equal to the numeric value one on every day of a stated lookback-period. Clarifying that rule required reading one as a threshold value, not as a label, so the condition is an advance-decline-ratio at or above 1 for the entire lookback.

Readers asked for implementations of published rules in more than one programming environment. That request shows a mechanical-trading-system is only testable after its logic is fully specified in executable form.

Equal oscillator bands do not imply equal sides

The same letter noted that most oscillators place overbought and oversold bands at equal distances and that many systems are tested with long and short rules applied together. That letter argued a market with a long-term upward bias would likely need different parameters for long positions than for short positions.

Editorial reading: symmetric oscillator bands and a combined long-short book assume the two sides share structure. The letter’s upward-bias point is a reason to drop that assumption and run a long-short-split-test.

A pattern is not a rule until it can be run

Encoding chart formations as a single software line was described as difficult because pattern identification, marking and projection resist reduction to one simple coded rule.

Editorial reading: until identification, marking and projection are written as executable conditions, a chart formation is not yet a mechanical-trading-system, and neither a trend-following nor a mean-reversion reading of it can be tested under a stated regime.