Strategy Report Card — Donchian 20 Breakout (GLD)

VERDICT: REJECT — no robust edge

Do not trade. The edge does not survive out-of-sample + costs.

A classic Donchian channel breakout strategy. It calculates the highest and lowest close over the past 20 bars (explicitly excluding the current bar via [1] offset). When the current close exceeds the prior 20-bar high it enters long; when it falls below the prior 20-bar low it enters short. It holds the position until the opposite signal fires, making it a trend-following system suited to trending market regimes.

Family donchian_breakout · direction long_short · params period=20

Translation warnings

The script uses [1] offset on highest/lowest, meaning it compares today's close to the channel computed from the previous 20 bars (not including today). The built-in donchian_breakout template may include the current bar in its rolling window; verify the engine's implementation matches this one-bar lag exactly.

Metric	Original	Improved +none+flat ★
OOS Sharpe	-0.045	-0.045
Stress Sharpe (3x cost)	-0.149	-0.149
Ann return	-0.7%	-0.7%
Max drawdown	-34.9%	-34.9%
Ann turnover	14.9	14.9
PSR (P[Sharpe>0])	0.453	0.453
Bootstrap p(mean<=0)	0.554	0.554

Recommended: improved — OOS Sharpe -0.045, PSR 0.453, bootstrap p 0.554.

Breadth — does the same edge hold elsewhere?

BROAD — edge holds across many assets and timeframes (low overfit risk)

Across assets: 5/8 positive (5/8 under stress), median Sharpe 0.404

Asset	Sharpe	Stress
SPY	0.55	0.42
QQQ	0.94	0.85
IWM	-0.14	-0.20
EEM	-0.02	-0.10
EWZ	0.25	0.21
GLD	-0.05	-0.15
AAPL	0.61	0.55
MSFT	0.71	0.63

Timeframe	Sharpe	Stress
D	-0.05	-0.15
W	0.63	0.62
M	0.27	0.26

Want your strategy checked like this — out-of-sample, net of cost?

Request a free analysis of any strategy →

Out-of-sample, net of cost. PSR / bootstrap / stress-test robustness is what a TradingView backtest cannot show. Not investment advice.