The Turtle Rules

The most famous trend-following experiment in history

In 1983, commodity traders Richard Dennis and William Eckhardt made a bet. Dennis believed trading could be taught as a set of rules; Eckhardt thought it was innate. They ran an experiment: recruit a small group of novices, teach them Dennis's system, give them real capital. Over the next four years the group — nicknamed "the Turtles" — reportedly made enormous profits, proving Dennis right (via the experiment) and creating the single most-documented mechanical trading system in history.

Kaufman introduces them:

During the mid-1980s the group known as the Turtles, founded by Richard Dennis and Bill Eckhardt, was the biggest trading sensation… The method is based on an N-day breakout.

The specifics leaked out through Curtis Faith's 2003 Original Turtle Trading Rules and Michael Covel's 2007 The Complete TurtleTrader. Neither book is in our reference library; Kaufman reproduces the key mechanics.

Built on Donchian

Before the Turtles, Richard Donchian had a 4-Week Rule for trading commodity breakouts (which we covered in the Keltner/Donchian lesson). Kaufman on lineage:

The 40/20 channel breakout, also called Donchian Channels, was the earliest N-day breakout that is recorded. — Kaufman

The Turtles ran two parallel Donchian systems with asymmetric entry/exit periods. That's the structural insight: enter on a long-window breakout, exit on a shorter-window reverse.

The two systems

System 1 (S1) — aggressive

• Long entry: intraday high exceeds the 20-day high
• Long exit: intraday low falls below the 10-day low
• (Mirror for shorts)
• Filter: skip an S1 entry if the previous S1 trade was profitable. Unless the prior trade was a 2N-loss (where N = 20-day ATR) — then take it.

The skip filter's purpose: avoid entering after a winning trade in the same direction, which often marks exhaustion. The 2N-loss override catches the case where you've just taken a full stop and the market reverses — the next signal is potentially a fresh trend.

System 2 (S2) — conservative

• Long entry: intraday high exceeds the 55-day high
• Long exit: intraday low falls below the 20-day low
• (Mirror for shorts)
• No skip filter.

S2 is slower, misses more noise, and catches major multi-month trends. Turtles ran both systems simultaneously — most winning trades came from S2; S1 captured earlier entries on the trades S2 would have missed.

Position sizing — the N framework

This is where the Turtles' approach gets interesting. Kaufman on "N":

N = 20-day average true range.

One unit sizes a position so that a 1N price move = 1% of portfolio equity. Which means a full 2N stop loss = 2% portfolio risk per trade.

Let's work an example:

• Account: $100,000
• Target risk per unit: 1% = $1,000
• Stock ATR(20) = $2.50
• 1 unit size = $1,000 / $2.50 = 400 shares

Entry at $50. Stop at $50 − 2 × $2.50 = $45. Risk if stopped: 400 × $5 = $2,000 = 2% of account. ✓

Pyramiding (adding units)

On favorable movement of 0.5N from the prior unit's entry, add another unit. Max 5 units per market (Kaufman: "A maximum of 5 units are allowed" at kaufman.txt:9031).

So in a strongly trending market you'd add at $51.25, $52.50, $53.75, $55.00 — each 0.5N = $1.25 above the prior.

Every unit's stop trails the newest one at its 2N-below point. The entire stack exits together when the 10-day (S1) or 20-day (S2) Donchian low is breached.

Diversification limits

Kaufman reproduces the categories:

• Per single market
• Per correlated-market cluster (energies, metals, currencies, short rates)
• Per less-correlated group
• Per net long/short direction

The specific numeric caps (often cited as 4 single / 6 correlated / 10 directional) are not in Kaufman — they appear in Covel/Faith. If you see those numbers, they trace to those sources, not to Kaufman's direct reproduction.

Drawdown rule

Kaufman verbatim (kaufman.txt:9041):

For every 10% drawdown in the portfolio, measured from the peak equity, the position size was cut by 20%… For every 6⅔% recovery, 10% was added back.

Built-in deleveraging: when the system is losing, trade smaller. When recovering, scale back up. Most retail traders skip this and blow up during regime changes.

Play with it

Toggle S1 (20/10) vs S2 (55/20) and market regime. Notice:

• S1 in trending markets: many entries, many exits, choppy performance. Asymmetric PnL — small losses but occasional huge winners.
• S2 in the same trend: fewer entries, cleaner rides. Win rate low (~40%) but winners large.
• Either system in sideways: whipsaw death. Channels flip, stops trigger, no runway.

Kaufman's 38-year test on copper

Kaufman tested the Turtle system on copper from 1980–2017:

The slower system [S2] showed steady profits for the entire 38 years… The faster system [S1] made profits only in the early 1980s — the period in which Dennis would have decided to form the Turtles.

Striking detail: the fast system stopped working right after the Turtles were assembled. Dennis built S1 during a regime that was ending. S2 kept producing for four decades. This is both validation (the long-window breakout really works on long-horizon data) and warning (parameters optimized on one regime don't survive the next).

Why people say it "stopped working"

The claim that trend-following degraded after 2000 is common. Kaufman's mechanism-level explanation (kaufman.txt:39693):

Increased volume from a broad set of participants introduces more noise into short-term price movement… a trend system will take longer to recognize a trend change. It will enter later and exit later, resulting in a reduction in net profits. Over a long test interval, this will show a noticeable decay in performance.

Translation: more participants = more noise = more whipsaw-like behavior at breakouts. The signal hasn't disappeared — its SNR degraded. Systems still work; win rates and profit magnitudes are smaller than the original era produced.

Common mistakes

• Running partial rule sets. Kaufman: "It is not clear how much discretion was allowed in following the signals… Underlying the process was the imperative that you can't miss the trade." The Turtles' edge depended on taking every qualifying signal. Cherry-picking kills the edge.
• Skipping the drawdown rule. The 10%/20% deleveraging is what kept Turtles alive through drawdowns. Without it, a 30% drawdown can cascade to 50%.
• Using a single instrument. Turtles traded a diversified basket. A single-market trend-following system has brutal equity curves.
• Over-optimizing the N values. 20/55/10/20 are Turtle canon. Grid-searching N on your own market is curve-fitting — exactly what Dennis's original formulation warned against.
• Ignoring that it's a long-horizon system. S2 can underperform for 12+ months in the wrong regime. If you can't tolerate that, you can't run Turtle-style trend-following.

Quick check

What you now know

• Two Donchian-based breakout systems run in parallel: S1 (20-day entry, 10-day exit, with skip filter), S2 (55-day entry, 20-day exit, no filter).
• N = 20-day ATR is the position-sizing unit. 1 unit sized so 1N move = 1% portfolio. 2N stop = 2% risk per trade.
• Pyramid up to 5 units at 0.5N intervals; all units share a trailing stop.
• Drawdown rule: −10% portfolio = −20% position size, until +6.67% recovery adds 10% back.
• Kaufman's copper test: S2 profitable for 38 years; S1 stopped working right after the Turtles formed.
• Post-2000 degradation is real but mechanism-level (increased noise), not a failure of the core idea.
• Partial systems, over-optimization, and single-market runs all kill the edge. Turtle-style only works as a complete system on a diversified basket.

Next: Mean Reversion Basics — the philosophical opposite of everything the Turtles did.