Auditing stochastic crossovers with moving-average baselines

The reading that was locked

The stochastic oscillator was specified as a scaled reading of where the latest close sits inside the high-low range of the prior N bars. High values sit near the range top and low values sit near the range floor. The testers treated that reading as valid on any regular sampling interval, not only daily bars.

The crossover that could reverse the book

The evaluated rule did not reverse on the raw reading. A long or short position flipped only after a shorter moving average of the stochastic oscillator crossed a longer moving average and then remained on the new side for a stated confirmation count.

The search used four integers: the range length, the short and long smoothers of the stochastic oscillator, and the confirmation count. The same structure was run on both daily and weekly samples.

A volatility-scaled book and a transaction-cost load

The historical simulation used a large hypothetical multi-market book so contract counts could be scaled to volatility differences rather than holding one contract in every market. That design is a volatility-scaled book: quieter and wilder markets do not enter the simulation at equal unit size.

Each simulated round turn was charged 150 units as a transaction-cost load meant to stand in for commission plus typical entry and exit slippage, not commission alone. The testers argued that a charge of at least 100 units was already a reasonable floor.

What the daily and weekly search showed

Over the 1976-1984 window, six of sixteen weekly parameter sets finished with a loss. Sixteen of twenty-two daily sets finished with a gain that the testers still described as mediocre in most cases.

The strongest stochastic-oscillator sets used much slower smoothers than the commonly advertised very short dual averages. Combinations that stayed similarly fast were described as a persistent drain once the transaction-cost load was included. Fading those fast signals was also said to fail under the same cost load.

A plain price crossover as the comparison family

A non-optimized 10/30-day price moving-average crossover was presented as the stronger family in a like-for-like parameter scatter. Sharpe ratio and a gain-to-retracement ratio were the comparison scores, so ranking did not rest on raw gain alone.

Editorial comment: a searched four-integer grid leaves a parameter-hindsight gap between the single best lookback combination found after the test and the result a trader would have obtained without knowing that combination in advance. The unfitted price crossover is the foil that keeps that gap in view.

How far the testers would take the result

The testers limited the finding to this particular stochastic-crossover recipe looking poor to mediocre. They noted that no single test can prove an indicator useless in every form. They also warned against assuming the measure has value before the intended application is tested historically.