The Test Bench › gate

Does Donchian Channel Breakout (20/55-day) actually work?

MARGINAL

Real but small -- not worth what it costs to run.

The claim

A classic breakout/trend system that beats the market — Richard Donchian / trend-following tradition

What we did

Tested against Buy & hold SPY, same window, SPY 1993-2026, prior-day, OOS 2014+.

The result

Result: -0.05 dSharpe (vs buy-hold); OOS a mirage

+6-7%/yr vs buy-hold +10.8%; worst drawdown -55%->-37%; dSharpe -0.02 to -0.07 (the OOS bump sits on a negative in-sample)

Verdict

Buy the N-day high, sell the N-day low — the original trend-following channel. As an S&P long/cash timer (20- or 55-day), it cuts the worst drawdown (-55% -> -37%) but hands back most of the return (+6-7%/yr vs +10.8% buy-and-hold) and shows no full-sample risk-adjusted edge. The flattering out-of-sample number sits on a negative in-sample — the data-mining tell. It was built for a DIVERSIFIED basket of FUTURES (leverage, shorting, many uncorrelated markets), where trend-following has a real record; a single mean-reverting equity index is the wrong instrument, and it shows.

What the test did and did not do

tested out of sample
point-in-time data
checked for look-ahead
net of real costs

Almost nobody publishing a backtest says which of these they did. We list the ones we failed as well as the ones we passed.

Tested 2026-06-23. The same treatment, applied to every other system, is on the test bench.

← Curve 10Y2Y>0 All findings Equal-Weight vs Cap-Weight Divergence →

Written with AI assistance from our own test database. The numbers are computed, not composed.
General market commentary for education, not personalised investment advice. Backtested results are hypothetical. Trading carries a real risk of loss.