Book summary “Way of the Turtle”: summary

Introduction

You want to know whether successful trading requires innate talent or can be systematically taught to anyone. Legendary commodity trader Richard Dennis and his partner William Eckhardt debated this exact question in the early 1980s. Dennis believed great traders were made through strict rules, while Eckhardt argued that trading ability was an inborn trait. To settle the argument, Dennis recruited ordinary people, trained them in his proprietary trend-following methods, and gave them millions of dollars to trade. These novices became known as the Turtles.

A former Wunderkind trader reveals the mechanical systems, rigorous volatility-adjusted position sizing, and psychological discipline behind Richard Dennis's legendary Turtle trading experiment.

The book outlines the exact mechanics behind this legendary experiment and shows you how to apply systematic trading principles. You will discover how to build an edge, manage risk, and conquer your emotions.

• Can trading skills really be taught to complete beginners?
• How do you measure market volatility to size positions correctly?
• What distinguishes a robust trading system from a fragile backtest?

1. The Nature versus Nurture Experiment

The legendary Turtle experiment began as a wager between two top traders to prove whether great trading skills are born or can be taught to anyone. Richard Dennis believed he could train almost anyone to become a successful trader. His partner William Eckhardt believed trading greatness was a matter of nature rather than nurture.

Great trading skills can be taught to ordinary people through strict mechanical rules and psychological discipline.

To settle the debate, Dennis and Eckhardt took out large advertisements in the Wall Street Journal, Barron's, and the New York Times. Over 1,000 applicants applied for the position.

The Wager

Dennis and Eckhardt screened candidates using a rigorous 40-question test and extensive interviews. They selected individuals with backgrounds in gaming, linguistics, chess, professional gambling, and programming.

• 40 applicants were granted an interview;
• 13 applicants were initially chosen.

The famous moniker given to Dennis's trainees originated from a visit to a turtle farm in Singapore. While standing at a turtle farm in Singapore, Richard Dennis reportedly blurted out that they were going to raise traders like turtles are raised there.

Screening and Selection

Before joining the Turtles at age 19, the author spent two years programming Apple II computers and testing trading strategies using historical data. The author worked part-time after school for Harvard Investment Service in Harvard, Massachusetts, writing 30 to 40 different programs that tested trading systems in 1983.

Training for the first group of Turtles was held during the last two weeks of the year. In the second year, a new crop was trained in only one week. After a one-month trial trading period, Dennis evaluated performance and gave the author a $2 million account to trade for the duration of the program, which was the largest account given to any Turtle.

Over four and a half years, the Turtles earned an average return of over 80 percent per year.

2. Trading with an Edge

A trading edge relies on exploitable statistical advantages and positive mathematical expectation rather than prediction.

Without an edge, transaction costs like commissions, slippage, and data fees will cause your account to lose money. You derive edges from portfolio selection, entry signals, and exit signals. These advantages stem from recurring market behaviors driven by cognitive biases.

Expectation and Edge

Expectation answers what happens if a strategy is executed repeatedly over the long run. By calculating average wins versus risks, you learn to ignore short-term losses and avoid outcome bias. Losses are merely the cost of doing business in a positive expectation system.

A 0.2 expectation means you make 20 cents for every dollar risked. For example, you calculate $3,000 in total risk for 10 contracts of gold entered long at $350 with a stop at $320 and a 100-ounce contract size.

The E-Ratio Formula

The E-ratio measures the edge of an entry signal by comparing volatility-adjusted maximum favorable excursion against maximum adverse excursion over a specific time frame. Maximum favorable excursion measures good price movement in the trade direction, while maximum adverse excursion measures movement in the bad direction. You normalize these across markets using the average true range and divide by the total number of signals.

To calculate the E-ratio:

• Compute the MFE and MAE for the specified time frame.
• Divide each of them by the ATR at entry to adjust for volatility and normalize across markets.
• Sum each of these values separately and divide by the total number of signals to get the average volatility-adjusted MFE and MAE.
• Divide the average volatility-adjusted MFE by the average volatility-adjusted MAE to get the E-ratio.

A random entry tested over 10 years showed an E5-ratio of 1.01, an E10-ratio of 1.005, and an E50-ratio of 0.997, confirming that random entries have no edge. By contrast, you find a 0.99 E5-ratio and a 1.0 E10-ratio for the Donchian channel breakout sample. You achieve a 1.20 E70-ratio for 20-day breakouts, a 1.27 E70-ratio for 70,000 random entries filtered by the trend portfolio filter, and a 1.33 E70-ratio when combining a trend filter with a breakout.

3. Turtle Money Management

Volatility normalization and strict position limits control the risk of ruin and protect against catastrophic drawdowns.

Money management is the art of balancing the risk of ruin against profit potential by choosing proper trade and position sizes. Proper money management keeps the risk of ruin at an acceptable level and ensures survival through bad periods, avoiding excessive leverage that leads to extended drawdowns or sudden blowups. Betting $500 out of a $1,000 pocket on a favorable 50 percent dice game results in a 25 percent risk of ruin within just two rolls.

A complete mechanical trading system removes subjective emotional decisions and covers every aspect required for trading. By automating decisions such as markets, position sizing, entries, stops, exits, and tactics, a trader can maintain consistency and follow rules religiously even during long losing streaks.

The N Factor and Volatility

Position sizing must normalize dollar volatility across different markets using a volatility measure called N.

N represents the 20-day exponential moving average of the true range, known today as Average True Range, and ensures that trades in different markets carry the same chance for a specific dollar gain or loss. Richard Dennis and Bill Eckhardt created unit sizes where 1-ATR of price movement equaled 1 percent of account size, ensuring daily fluctuations and dollar-term risks were similar across all markets. For a $1 million account, if 1-ATR for a market equaled $10,000, the unit size was set so that the dollar movement per contract matched the 1 percent risk target.

You compute your position sizes using precise mechanical rules:

• Compute daily true range using maximum of H minus L, H minus PDC, or PDC minus L
• Compute N using the formula involving previous day's N and current day's true range
• Determine dollar volatility by multiplying N by dollars per point
• Calculate unit size by dividing 1 percent of account equity by dollar volatility
• Look at the dollar amount represented by 1-ATR of price movement in a particular market.
• Divide $10,000 by that 1-ATR dollar amount for every $1 million in trading capital to determine the unit size.

Portfolio Risk Limits

Rules governing total portfolio risk use position limits across single and correlated markets to prevent catastrophic drawdowns.

The Turtles enforced limits on single markets, closely correlated groups, loosely correlated groups, and total net direction to limit exposure during extreme events. Because markets become highly correlated during major price shocks and adverse trends, strict caps on units per market and total direction prevent staggering losses. During the October 1987 crash, the Turtle limits on units kept losses manageable and saved Richard Dennis over $100 million. Some participants still lost 40 to 60 percent account equity in a single day after the October crash when loaded short in interest-rate futures before the Federal Reserve lowered interest rates overnight.

You must follow strict aggregate limits to protect your capital:

• Limit single markets to a maximum of 4 units
• Limit closely correlated markets to a maximum of 6 units in one direction
• Limit loosely correlated markets to a maximum of 10 units in one direction
• Limit total units in one direction long or short to a maximum of 12 units
• Put on a maximum of 4 units per market.
• Put on a maximum of 6 units in markets that are highly correlated.
• Put on a maximum of 10 units in any given direction, which could be raised to 12 if there were positions in uncorrelated markets.

4. Breakout Systems and Exits

Mechanical channel breakouts capture infrequent major trends while holding through large retracements.

Trend-following systems rely on basic building blocks like breakouts, moving averages, volatility channels, time-based exits, and simple lookbacks to identify market states. These building blocks indicate when a market transitions from a stable state to a trending state or back. You can enter and exit trends without relying on complex, over-optimized secret formulas.

Donchian Channel Breakouts

The Turtles utilized Donchian channel breakout systems that generated entries based on historical price extremes over intermediate and long-term lookback periods. You determine highs and lows using a 20-day period for System 1 or a 60-day period for System 2 at the end of each day. You buy if a market exceeds the highest price for the preceding specified number of days, or sell short if it breaks below the lowest price.

• System 1 Entry: enter when price exceeds by a single tick the high or low of the preceding 20 days, ignoring the signal if the previous breakout resulted in a winning trade
• System 2 Entry: enter when price exceeds by a single tick the high or low of the preceding 55 days, taking all signals regardless of prior wins
• Adding Units: enter single-unit long positions at breakouts and add at half N intervals up to the maximum permitted units

Trend Exits and Time Stops

Profitable exits require holding through large retracements to capture infrequent large trends rather than taking profits early. System 1 exits occurred on a 10-day low for longs or high for shorts, while System 2 exits used a 20-day low or high, requiring immense discipline as profits evaporate. For trend followers, drawdowns stem from giving back accumulated profits during trend reversals rather than initial entry risk.

• Exit all units in a position when the price hits a 10-day low for long positions or a 10-day high for short positions under System 1
• Exit all units when the price hits a 20-day low for long positions or a 20-day high for short positions under System 2

5. Taming the Turtle Mind

Markets trend and create profit opportunities because human participants consistently and systematically make irrational decisions under uncertainty. Behavioral finance demonstrates that individuals under duress make poor assessments of risk and probabilities. Winning traders exploit these predictable psychological errors rather than relying on the flawed economic assumption of rational actor theory.

Cognitive biases destroy traders unless they learn to think in probabilities and abandon the desire to be right.

Specific psychological biases like loss aversion, the sunk cost effect, and the disposition effect ruin traders by causing them to abandon rules and hold losing positions. Loss aversion makes losses feel twice as painful as equivalent gains. The sunk cost effect and disposition effect prevent traders from cutting losses early, forcing them to hold bad trades until accounts are devastated.

Silicon Valley engineers watched their high-tech stock options drop to a tenth or hundredth of previous values in 2000 because they anchored on recent highs and refused to face reality.

Behavioral Finance and Biases

Winning traders abandon the desire to predict the future or be right, freeing themselves from outcome bias by expecting most individual trades to lose. Beginners view winning trades as proof of intelligence and losing trades as personal failures. Turtles do not care about being right and view the future as unknowable in specifics but foreseeable in character because human emotion does not change. By expecting most trades to lose, they separate trade outcomes from their self-worth and stick to their plans.

• Trade in the present: Do not dwell on the past or try to predict the future.
• Think in terms of probabilities, not prediction: Focus on methods where probabilities favor success over the long run.
• Take responsibility for your own trades: Don't blame mistakes on others, markets, or brokers, but learn from them.

Recency bias causes traders to overweight recent events and abandon proven systems during temporary rough patches just before profitable trends appear. Traders often look at a single market during a stagnant period and mistakenly conclude it is untradable, missing massive upcoming trends because they are blinded by recent losses.

In 1998, Curtis Faith taught his trading methods to a friend, warning him to execute all trades religiously. In February 1999, the friend refused to trade cocoa because he had lost money previously. Historical testing of cocoa breakout trades showed 17 losing trades in a row before a very sizable winning trend started in November 1998.

Thinking in Probabilities

Successful traders think using probability distributions and R-multiples, relying on large winning trades to outweigh frequent small losses. Like insurance companies that profit by covering risks for less than the probable payout, trend-following systems generate many losing trades offset by massive winning trades. Chuck Branscomb invented the R-multiple concept to compare trades by dividing profit by the amount risked.

Buying a contract of August gold at $450 with a stop at $440 risks $1,000. Earning a $5,000 profit on that trade makes it a 5R trade because the profit is five times the risked money.

Traders often reject simple, profitable rules because their egos crave secret or complicated knowledge to feel superior. Richard Dennis taught simple principles that made over $100 million, yet the Turtles initially struggled because they doubted that wealth could come from straightforward rules. The ego demands secret knowledge and the ability to predict, which makes discretionary trading appealing but dangerous.

Some Turtles believed Curtis Faith had been given secret information by Richard Dennis simply because Curtis was making money while they were losing.

6. Backtests, Lies, and Robustness

Backtest metrics often deceive due to overfitting, parameter optimization, and random market effects.

Discrepancies between historical simulations and actual performance stem from four major sources:

• trader effects
• random effects
• the optimization paradox
• overfitting

When many participants notice a popular system, their anticipatory orders can front-run the logic and ruin the original edge. A popular system traded with several hundred million dollars experienced a severe drawdown when other traders figured out its closing price triggers. Pure chance also creates massive performance variance, making it difficult to separate lucky managers from skilled ones based solely on track records.

The optimization paradox occurs because parameter optimization increases a successful backtest while decreasing the predictive accuracy of its metrics. Overfitting compounds this issue by tuning systems to isolated historical events, which creates performance cliffs that fail in actual trading. Curtis Faith tested 4,000 variations of a rule that lightened positions during specific days in September and July, artificially boosting returns from 45.7 percent to 58.2 percent based purely on past drawdown coincidence.

Robust Measures and Monte Carlo

Traditional performance measures are too sensitive to minor data changes and start dates, making them non-robust. Adjusting the start and end dates of a Triple Moving Average test by removing one month from the beginning and two months from the end caused the return to jump from 43.2 percent to 46.2 percent. Regressed Annual Return and R-Cubed provide stable alternatives by accounting for linear regression and drawdown duration.

Parameter scrambling tests system robustness by shifting key parameter values away from optimal settings to observe performance degradation.

Using the Bollinger Breakout system, moving from optimal values of 350 days and -0.8 exit threshold down to 250 days and 0.0 threshold decreased RAR% from 59 percent to 58 percent and R-Cubed from 3.67 to 2.18. You should follow specific steps to test system stability:

• Take a few system parameters and change them by a considerable amount, say, 20 to 25 percent of their value.
• Pick a point that is considerably down the side of the optimization curves.
• Look at the results for this test to evaluate stability.

Monte Carlo simulation generates alternative historical trading universes using random numbers to test system robustness against sequence variations. Richard Feynman utilized random numbers during the Manhattan Project to model neutron fission characteristics of uranium without needing to predict individual atom behavior.

7. Building Bulletproof Systems

Robust trading programs are founded on two core natural attributes: diversity and simplicity, designed to survive an unknowable future.

Nature uses diversity to insulate populations from radical changes. Simple organisms prove hardier during major climate shifts due to lack of environmental specialization. Just as simple viruses survive environmental catastrophes better than complex species, simple trading rules are less dependent on specific market conditions.

Diversity across uncorrelated markets and distinct trading systems insulates a portfolio against radical structural changes. Markets divide into three distinct behavioral classes:

• fundamentals-driven markets;
• speculator-driven markets;
• aggregated derivative markets.

Rich prohibited the Turtles from trading coffee in early 1985 due to volume concerns. This restriction caused them to miss a massive 1985 coffee trade that would have yielded a 280 percent return on a $5 million account. The March 1986 coffee contract had a value of N equal to 1.29 cents at entry. The unit size traded for a $5 million account was 103 contracts. The total long position size across four units reached 412 contracts. The approximate profit per contract on the missed coffee trade was $34,000, totaling $14 million in aggregate profit.

Market Classes and Diversity

Combining non-correlated or negatively correlated trading systems significantly improves risk-adjusted returns compared to running any single system alone. Running systems that perform well in opposing market conditions smooths out drawdowns because one system profits while the other suffers losses. Combining a better Bollinger Breakout sub-system with an RAR% of 38.2 percent and an R-Cubed of 1.19 with a worse sub-system of 14.5 percent RAR% and 0.41 R-Cubed resulted in a combined RAR% of 61.2 percent and an R-Cubed of 5.20. You need a minimum capital of $100,000 to trade a long-term trend-following futures system with basic diversification, and $1 million to adequately trade four or five different diversified systems.

You should apply these practical rules when designing your portfolio:

• Trade or exclude markets according to liquidity and market class only.
• Include as many markets as possible that do not correlate with one another to maximize portfolio diversification.

Epilogue and Final Lessons

Most people fail to follow their dreams because they default to predetermined career tracks out of a fear of failing. Society and companies set up predetermined tracks that individuals enter by default rather than conscious choice. Stepping off these tracks requires conscious effort and a willingness to risk failure, which is a necessary prerequisite for genuine learning.

Adults often struggle to learn new skills, like languages, because they are terrified of making mistakes and sounding foolish. Students studying Spanish in Buenos Aires who ignored how they sounded and conversed immediately learned to speak fluent Spanish, while those with years of formal study remained unable to converse.

10 Key Ideas

1. Great trading skills can be taught to ordinary people through strict mechanical rules and psychological discipline.
2. A trading edge relies on exploitable statistical advantages and positive mathematical expectation rather than prediction.
3. Position sizing must normalize dollar volatility across different markets using a volatility measure called N.
4. Rules governing total portfolio risk use position limits across single and correlated markets to prevent catastrophic drawdowns.
5. Mechanical channel breakouts capture infrequent major trends while holding through large retracements.
6. Cognitive biases destroy traders unless they learn to think in probabilities and abandon the desire to be right.
7. Backtest metrics often deceive due to overfitting, parameter optimization, and random market effects.
8. Parameter scrambling tests system robustness by shifting key parameter values away from optimal settings to observe performance degradation.
9. Robust trading programs are founded on two core natural attributes: diversity and simplicity, designed to survive an unknowable future.
10. Combining non-correlated or negatively correlated trading systems significantly improves risk-adjusted returns compared to running any single system alone.