1987issue C091-6
Objective-ranking can make nearby-contract rules look busy or idle
A historical nearby-contract study of Standard & Poor's 500 futures ranked the same indicator settings on five separate objectives. Editorial reading: treat that multi-objective grid as a diagnostic of search bias, because total profit, average winning trades, and average losing trades can make one rule family look hyperactive, one-sided, or almost idle.
- The archive simulated the nearby-contract in Standard & Poor's 500 futures from 2 September 1982 through 1 December 1986, rolling the March, June, September, and December contracts on the first session of each delivery month.
- Settings were chosen by objective-ranking on overall profit, short-side profit, long-side profit, average winning trade, and average losing trade, not by one blended score.
- Ranking by smallest average losing trade left some range-oscillator and RSI settings with no trades, while ranking by largest average winning trade selected a four-day RSI setting with only three trades.
- Editorial reading: the same family of rules can look hyperactive, one-sided, or almost idle depending on which ranking objective is used.
The nearby-contract sample
The archive simulated nearby Standard & Poor's 500 futures from 2 September 1982 through 1 December 1986. It used the March, June, September, and December contracts and treated the nearby-contract as the front-month futures contract, rolled on the first session of the delivery month.
Simulated trades were filled at the open and charged a 100-dollar round-turn commission.
Two search designs
The range-oscillator in this study is a daily oscillator that scales the high-versus-prior-close move by the largest of the day's range, the high-versus-prior-close move, and the prior-close-versus-low move. It replaced an earlier high-minus-prior-close over high-minus-low ratio with a version that divides the high-versus-prior-close component by whichever of those three quantities is largest.
The dual-average-scan is a crossover search over one shorter and one longer moving-average window. The short window ran from 2 to 15 days in one-day steps and the long window ran from 6 to 60 days in three-day steps, producing 266 combinations.
The net-profit comparison covered seven technical indicators.
Five separate ranking objectives
Objective-ranking means selecting indicator settings by one chosen backtest statistic at a time rather than by a single blended score. Each indicator was ranked separately on overall profit, short-side profit, long-side profit, average winning trade, and average losing trade.
When ranking by smallest average losing trade, some range-oscillator and RSI parameter sets produced no trades in the sample. Ranking by largest average winning trade selected a four-day RSI setting that generated only three trades over the sample.
Reading the grid as a bias check
Editorial reading: ranking on overall profit, long-side profit, or short-side profit can highlight active or one-sided books. Ranking on average winning trade or average losing trade can highlight settings that barely trade, including a four-day RSI choice with only three trades and range-oscillator or RSI choices with none. The five-way table is therefore a check on how the chosen objective, not only the rule family, shapes what looks promising.