Constructing a three-rule moving-average system

From maxims to a testable procedure

Editorial framing: this article treats system construction as a translation workshop. Three familiar trader maxims are converted into one inspectable trend-following procedure. Moving-average lag and a high-low-trigger split are used as the built-in loss, profit, and abstention machinery. Historical comparison is examined only after that procedure is complete.

Mechanical-system construction is presented as three stages: state a theory of how trades might work, convert that theory into explicit rules, then evaluate the rules. A mechanical-trading-system is a complete, testable procedure that states when to enter, when to exit, and when to stand aside from explicit market inputs.

The theory to be quantified

The theory being quantified is the trio of cutting losses, allowing winning trades to continue, and trading with the trend. Those statements remain a theory until they become explicit rules.

A simple moving average and the unfiltered rule

A simple-moving-average is the arithmetic mean of prices over a lookback. Prices of 105, 100, and 103 produce a three-period average of 102.667. A moving-average is a lookback mean of ordered prices that updates as the newest observation replaces the oldest and can be computed on closes, highs, or lows.

The unfiltered moving-average rule buys when price closes above the average and sells when price closes below it.

A longer lookback such as 200 days adjusts slowly and ignores minor counter-moves, while a 10-day average tracks price more closely and is more exposed to whipsaws. A 20-day average sits between those extremes. A whipsaw is a repeated opposite crossing of price around an average that reverses a position before a move can develop.

Lag as the loss and hold mechanism

Lag is the delay between current price and a moving average built from past observations. Because a moving average lags price, a long entered on a close above it is either closed when price recrosses below or held while the advance continues, and is later closed when fading momentum lets the average catch up and a downward swing breaks it.

Editorial reading: that same lag is the construction's way of cutting losses and allowing winning trades to continue. It is not added after the rules are written. It is already inside the average.

A trend-filter on the close average

Trend is read from the moving average of closes: in a 10-day window the average rises when today's close replaces a smaller close from 10 days earlier and falls when it replaces a larger one.

The finished procedure takes long entries only while the close average is rising and short entries only while it is falling, so signals are filtered to the quantified trend. The trend-filter is a comparison of today's close average with yesterday's that allows a signal only when the average is moving in the trade's direction. Trend-following here means taking longs only while a defined trend measure is rising and shorts only while that measure is falling.

A high-low-trigger without optimized thresholds

To reduce whipsaws without optimizing thresholds, the long trigger is a close above an x-day average of highs and the short trigger is a close below an x-day average of lows. A long exits on a close below the low average and a short exits on a close above the high average.

The high-low-trigger is a non-optimized construction that buys against an average of highs and sells against an average of lows so random noise has room to trade. Editorial reading: the space between those two averages is also the abstention machinery. Price can move there without forcing an entry.

Evaluation comes last

Once the rules are explicit, the third stage is to evaluate them. The constructed rules were evaluated with lookbacks from 10 to 50 in steps of 10 on Treasury bonds, the dollar index, and silver from January 1991 through December 1993, allowing $50 for commission and slippage, and were compared with a high-low-filter-only variant and an unfiltered simple crossover.

This article does not treat that historical workflow as a present-day result.