Donchian Channel Strategy: The Original Breakout System

What the channel is

Upper channel = highest high of the last N bars
Lower channel = lowest low of the last N bars
Middle        = (upper + lower) / 2

That is all.  No averaging, no weighting, no smoothing.

Signals in the classic form:
   Buy    when close exceeds the upper channel of the last N bars
   Exit   when close falls below the lower channel of the last M bars
          where M is typically half of N

The middle line is sometimes used as a trailing exit,
which is more conservative than waiting for the lower channel.

The absence of parameters beyond N is what makes the channel valuable for research. A moving average system has a type, a length, and often a second length. A Donchian system has one number, which means far less scope for fitting historical noise.

A complete Donchian system

Universe

20 to 40 liquid instruments across asset classes. This is not optional: single-instrument channel systems spend years unprofitable.

Timeframe

Daily bars, evaluated on the close, executed at the next open.

Entry long

Close above the highest close of the past 55 bars. Using closes rather than intraday highs substantially reduces false signals.

Entry short

Close below the lowest close of the past 55 bars.

Trend filter

Optional but recommended: require the 100-period moving average to be sloping in the direction of the trade.

Initial stop

Entry minus 2.5 x ATR(20) for longs.

Exit

Close below the lowest close of the past 20 bars, or the ATR stop, whichever comes first.

Position size

Risk 0.4 percent of equity per position using the ATR stop distance. Maximum 12 positions, maximum 4 per sector.

Pyramiding (optional)

Add one unit for every 0.5 ATR of favourable movement, up to 3 additional units, raising stops so total risk stays constant.

Choosing the lookback

The important finding from decades of testing is that most lookbacks in the 20 to 100 range produce broadly similar risk-adjusted results across a diversified portfolio. That plateau is evidence that the underlying effect is real. If your testing shows one lookback dramatically outperforming its neighbours, that is evidence of noise, not discovery.

What actually improves a channel system

1. Diversification across uncorrelated markets. By far the largest improvement. A channel system on one instrument is unusable; on thirty it becomes viable.
2. Volatility-based position sizing. Equalises risk contribution across markets with very different price levels and volatilities.
3. A volatility contraction filter. Requiring compressed volatility before the breakout improves the quality of signals, because expansion from compression is more likely to persist.
4. Using closes rather than intraday extremes. Reduces spike-triggered entries meaningfully.
5. A longer-term trend filter. Taking only breakouts aligned with a 100 or 200-period trend reduces countertrend whipsaws.
6. Pyramiding with constant total risk. Adds exposure to trades that are working without increasing the risk budget.

Notice that optimising the lookback does not appear on this list. That is deliberate: parameter tuning is the least effective of the available improvements and the most likely to produce a fitted result.

What to expect