2001issue C081-7
Hourly Donchian and Turtle rules locked after one walk-forward window
This archive article restates a foreign-exchange evaluation that counted Donchian lookbacks in hours, applied a Turtle trading size-and-stop rule, and froze every constant after one calendar window so the next window could only confirm whether the same entry, exit, and abstention rules still fired.
- The evaluation treated entry, exit, position size, and money management as one procedure and chose its constants on a training window before applying them to a later window.
- A long or short opened when an hourly bar made a higher high or lower low than the prior n-entry hours, and it closed when price made a lower low or higher high than the prior n-exit hours.
- Each trade used a constant 10,000 US dollars of size, and a money-management stop ended the trade if the loss reached 2 percent of capital.
- A shared 326-hour entry lookback was locked for the Swiss franc and the euro after the yen was dropped, with lookbacks tuned on January 1991 through December 1994 data and then applied unchanged to January 1995 through December 1998 data.
One procedure on hourly bars
Tick prints without volume were rebuilt as one-hour open-high-low-close bars, and a trading day was defined as 15 hours from 6 am to 9 pm.
The evaluation treated entry, exit, position size, and money management as one procedure and chose its constants on a training window before applying them to a later window.
Hourly Donchian entry and exit
Donchian breakout was stated on those hourly bars. A long or short is opened when an hourly bar makes a higher high or lower low than the prior n-entry hours, and it is closed when price makes a lower low or higher high than the prior n-exit hours.
Exit lookback is defined as C times entry lookback, and C of one-half was kept after other fractions from one-sixteenth to one were checked on the same currency set.
Turtle size and the money-management stop
Turtle trading supplied the size and stop inside the same procedure. Each trade used a constant 10,000 US dollars of size, and a money-management stop ended the trade if the loss reached 2 percent of capital.
Walk-forward freeze and the shared lookback
Walk-forward analysis was the freeze step. Lookbacks were tuned on January 1991 through December 1994 data and then applied unchanged to January 1995 through December 1998 data.
The yen was dropped after its in-sample lower confidence limits on average profit and loss crossed zero, and a shared 326-hour entry lookback was locked for the Swiss franc and the euro.
The locked 326-hour hourly lookback is a little more than 21 days and sits near a 20-day daily-bar setting used with the same 2 percent stop in an earlier Turtle study.
Trade frequency and combined books
With n-entry at 326 hours, the in-sample Swiss-franc and euro series produced about 40 trades, roughly one per month. Raising the entry lookback from 180 to 420 hours reduced Swiss-franc trades from about 70 to 30.
Return-to-risk was defined as annualized rate of return divided by maximum peak-to-valley drawdown, and the Swiss-franc and euro books were combined to study diversification, with monthly profit-and-loss correlations of 0.31 in sample and 0.45 out of sample.
All readings on this track · 6 readings
- 1992Turtle trading as one testable Donchian breakout procedure
- 2001Hourly Donchian and Turtle rules locked after one walk-forward window
- 2001Critiquing Turtle trend following as a drawdown budget
- 2010Evaluate Turtle-style bands as one mechanical procedure
- 2012Quiet-tape cycle clocks for index futures
- 2020Constructing Turtle-style dual Donchian trend-bands