Seven ways to trade gold, ranked honestly. Three of them lose money.

A trend-following strategy. It takes the midpoint of each bar, builds a band above and below it at three times the recent volatility (ATR over 10 bars), and locks that band in the trend direction so it only loosens, never tightens. That locked band is the trailing stop. When price closes above it, the trend reads up and the strategy goes long; when price breaks the band, it exits to cash. Long-only, so the worst case is sitting flat, never short.

Same trend-persistence premise as a breakout system, but the stop trails on volatility (ATR) rather than on a fixed-bar low. That adapts the stop distance to the current regime: tighter in calm markets, wider in volatile ones.

HL2   = (high + low) / 2
ATR   = Wilder's ATR(period=10)
upper = HL2 + 3 * ATR
lower = HL2 - 3 * ATR
# Final bands lock in the trend direction:
fu[t] = min(upper[t], fu[t-1])  if close[t-1] <= fu[t-1]  else upper[t]
fl[t] = max(lower[t], fl[t-1])  if close[t-1] >= fl[t-1]  else lower[t]
# Direction flip on close vs prior locked band:
if close[t] > fu[t-1]: direction = +1
if close[t] < fl[t-1]: direction = -1
Supertrend line = fl when direction = +1, fu when direction = -1

• A short-window M15 Sharpe looks high (about 1.7 gross, lower after realistic cost) but is bull-window-flattered. The trustworthy full-cycle figure is the H4 22-year 0.89, not the M15 demo number.
• The recursive band lock means the line can sit very far from price in long calm trends, so stop distance can become uncomfortably wide.
• ATR over 10 bars on a fast timeframe reacts quickly to news spikes and can flip out near a real CPI or NFP print.

The classic Turtle-style channel breakout, long-only. Enter when price closes above the highest high of the last 20 closed bars; exit when it closes below the lowest low of the last 15 bars. The exit channel is the stop, so there is no separate stop-loss.

Trend persistence: a new 20-bar high is statistically more likely to be followed by another up-move than by mean reversion, especially in commodities with positive drift. The payoff is asymmetric, many small stop-outs and a few large winners that ride the channel up.

upper_entry = max(high[t-20 : t])   # excludes current bar (no look-ahead)
lower_exit  = min(low[t-15  : t])
if flat and close[t] > upper_entry: -> LONG
if long and close[t] < lower_exit:  -> FLAT

• Whipsaw periods (range-bound months) bleed via repeated false breakouts.
• Bull-flatter bias: the 22-year window is mostly up, so absolute return looks great, but the strategy cannot profit in a sustained bear, it just sits flat.
• Donchian channel exits give back open profit before flipping flat. This is intentional, but it feels worse than a trailing stop.

The benchmark every gold strategy must beat. Always long, never trades. The Sharpe and max-drawdown numbers come from the gross spot price curve with no leverage and zero turnover cost.

Not a strategy. It is the bar to clear: if a long-only trend follower cannot deliver a comparable return at materially less drawdown, the strategy is not earning its complexity.

position[t] = +1 for all t
return[t]   = close[t]/close[t-1] - 1

No parameters. Buy-and-hold has nothing to tune.

• Full exposure to gold's structural drawdowns (the 2011 to 2015 cycle was -45%).
• No regime filter, so you stay long right through the macro top.
• Carry cost is zero in a backtest but nonzero with a real broker (swap fees on a long XAUUSD CFD).

Classic fast/slow Simple Moving Average crossover. Go long on a golden cross (the fast average crossing up over the slow one). The short side, a death-cross short, is not enabled here.

The oldest documented trend filter. SMA crossovers are slower than EMA or MACD but also less noisy, good in slow-moving regimes and bad in fast ones.

fast = SMA(close, 20)
slow = SMA(close, 50)
golden cross: fast crosses up over slow -> LONG
death cross : fast crosses down under slow -> FLAT (in LO mode)

• Lag: SMAs weight every bar in the window equally, so the cross fires well after the trend has started.
• Whipsaw in ranges, with repeated crosses and no follow-through.

Classic Wilder RSI(14) mean-reversion. Buy when RSI drops below 30 (oversold), short when RSI rises above 70 (overbought), and exit when RSI returns to 50.

Markets overshoot in both directions; extreme oscillator readings are supposed to precede a reversion to the mean. It is the textbook strategy in most retail trading books, which is part of why we keep it on the board as a control.

delta = close.diff()
gain  = max(delta, 0)
loss  = -min(delta, 0)
avg_gain = Wilder-EMA(gain, 14)
avg_loss = Wilder-EMA(loss, 14)
RS  = avg_gain / avg_loss
RSI = 100 - 100 / (1 + RS)
if flat and RSI < 30: -> LONG
if flat and RSI > 70: -> SHORT
if long and RSI >= 50: -> EXIT
if short and RSI <= 50: -> EXIT

• Trends in gold can keep RSI overbought (above 70) for weeks at a time, so you stay short and bleed.
• Mean reversion as a thesis assumes a stationary regime; gold is not stationary.

Bollinger mean reversion. Buy when close drops below the lower band (2 sigma below the mean), short when close rises above the upper band, and exit at the middle band. Same indicator as Bollinger breakout, with the opposite trade direction.

Standard-deviation bands measure how unusual a move is. The reversion thesis says unusual moves snap back; the breakout thesis says they continue. Both cannot be right on the same asset.

mid   = SMA(close, 20)
stdev = STDEV(close, 20)
upper = mid + 2.0 * stdev
lower = mid - 2.0 * stdev
if flat and close < lower: -> LONG
if flat and close > upper: -> SHORT
if long and close >= mid:  -> EXIT
if short and close <= mid: -> EXIT

• Same trend-versus-reversion mismatch as RSI on gold.

Z-score reversion, long-only. When close is 2 standard deviations or more below its 20-bar mean, go long; exit when the z-score returns to 0. Mathematically equivalent to a Bollinger lower-band reversion, exposed through a different parameterization.

Same as Bollinger reversion: statistically extreme positions revert to the mean. Long-only avoids the catastrophic short side that kills long-and-short reversion on gold.

mu    = SMA(close, 20)
sigma = STDEV(close, 20)
z     = (close - mu) / sigma
if flat and z < -2.0: -> LONG
if long and z >= 0:   -> EXIT

• Same as Bollinger reversion: the snap back does not snap on gold.