Moving Average Crossover Strategy: Complete Rules and Testing

How the signal works

Two averages are computed, one faster and one slower. When the faster crosses above the slower, recent prices have risen relative to the longer-term average, which is a mechanical definition of an emerging uptrend. The reverse cross defines an emerging downtrend.

fast = SMA(close, 50)
slow = SMA(close, 200)

Long  when fast crosses above slow
Flat or short when fast crosses below slow

Common pairs:
   10 / 30    fast, many signals, high whipsaw
   20 / 50    moderate
   50 / 200   slow, the "golden cross"
   9 / 21     common in shorter timeframes

The pair determines holding period more than anything else.

Note that the crossover is mathematically equivalent to a smoothed momentum measure: the fast average exceeds the slow when the average of recent returns over the fast window is positive relative to the slower window. This is why the crossover and other trend measures give similar results, and why combining several of them adds little.

A complete, testable rule set

Universe

15 to 30 liquid, uncorrelated instruments: index ETFs, sector ETFs, major FX pairs, or futures across asset classes. A single instrument is not enough.

Timeframe

Daily bars, signals on the close, orders at the next open.

Entry long

EMA(50) crosses above EMA(200), and the EMA(200) is higher than it was 20 bars ago. The slope condition removes many flat-market signals.

Entry short

The mirror image. Omit for long-only equity implementations.

Volatility filter

Skip entries when ATR(20) divided by price is in the lowest 20 percent of its two-year range, which indicates a compressed, directionless market.

Initial stop

Entry minus 3 x ATR(20), which is wider than the crossover exit and only triggers on sharp adverse moves.

Exit

The opposite crossover, or the stop, whichever comes first.

Position size

Risk 0.4 percent of equity per position using the ATR stop distance. Maximum 10 concurrent positions, maximum 3 per asset class.

Re-entry

Allowed on a new crossover only. No discretionary re-entries after a stop.

The golden cross and death cross, examined

The 50-day crossing above the 200-day is called a golden cross, and the reverse a death cross. These receive substantial media attention, which is the main reason to know them.

• Tested performance is unremarkable. On broad indices, golden crosses have preceded both continued advances and immediate declines. The signal identifies that a trend has already been underway rather than predicting a new one.
• Signal frequency is very low. A few signals per decade on a single index means the sample is far too small to evaluate on one instrument.
• The lag is substantial. By the time a 50-day crosses a 200-day, a significant portion of the move has occurred. This is inherent, not a flaw.
• The attention creates a modest self-fulfilling element. Widely reported signals can attract flows, though this effect is small relative to the noise.
• As a regime filter it is more useful than as a signal. Being above or below the 200-day is a better-tested condition than the crossing event itself.

Managing whipsaws

Diversification is the most effective of these by a wide margin. A crossover system on one instrument will spend long stretches losing; the same system across twenty uncorrelated markets has some instruments trending at almost any time, which is what makes the approach viable.

Testing it without fooling yourself

1. 1

   Test a grid of parameter pairs, not a single pair

   Fast lookbacks from 10 to 100 against slow from 50 to 300. Plot the results as a surface. A robust system shows a broad region of acceptable performance, not an isolated peak.

2. 2

   Include costs and next-open execution

   Signals generated on the close and executed at the next open must account for the gap between them plus spread and commission.

3. 3

   Test across many instruments

   Twenty markets over twenty years provides far more independent evidence than one market over eighty.

4. 4

   Report drawdown duration prominently

   Crossover systems spend long periods underwater. The time-underwater statistic determines whether you would have kept trading it.

5. 5

   Compare against the obvious alternative

   Benchmark against buy and hold and against a simple 200-day filter. A crossover system that underperforms a filter is not worth the extra complexity.

6. 6

   Check the sensitivity to the start date

   Shift the test window by a few months and re-run. Results that change substantially indicate the sample was dominated by a small number of trades.