Jerry Parker: The Dangers of Over-Optimization

There is a version of trading research that feels productive and destroys performance. You look at the backtest. The numbers are compelling. You add a filter, smooth the equity curve, adjust a parameter, and the numbers get better. Then you go live and the system produces results you have never seen before, and not in a good way. Jerry Parker has been through exactly this. He has run Chesapeake Capital for over 35 years, made money almost every year for the first decade, then nearly broke what was working by trying to make it better.

This is part three of a three-part Trading Triumphs conversation with Jerry, a Turtle Trader trained by Richard Dennis in 1983. In this segment, Jerry covers over-optimization, what it actually costs, and how he rethought his approach to research from the ground up.

Watch the full episode below, then read on for the complete breakdown.

Table of Contents

Ten years of making money with simple rules

Jerry’s first four years were under Richard Dennis. After that, he ran Chesapeake Capital with the same basic approach for roughly eight to ten more years and made money almost every year. The systems were simple: breakout entries, breakout exits, ATR-based stop losses. One entry rule, one exit rule, a stop. No more.

That consistency impressed clients. Money came in. Then came the temptation that hits every trader who strings together a run of success: what if we made it better? Jerry and his team started researching ways to improve performance. They looked at what other traders were doing. Profit objectives. Reducing positions based on volatility. Adjusting parameters to produce smoother equity curves.

The backtests looked great. The new approaches had better numbers than the original systems. The team was confident. They implemented the changes.

Within three to four months, they saw performance they had never seen in any backtest. Four consecutive losing months. Relative performance that was genuinely bad by any measure. Nothing in the historical data had predicted this. The system had been optimised into something fragile.

Recognising the mistake and undoing it

Jerry’s response to this situation is worth noting. He didn’t double down on the research. He didn’t rationalise the drawdown as temporary noise. He recognised it as a signal that something was wrong with the approach, not just the current environment.

He acknowledged that many traders who go deep into backtesting become so invested in the process that they can’t abandon their latest research even when live results contradict it. The research becomes an identity. Admitting it failed means admitting the process was flawed. Jerry describes getting to exactly the opposite conclusion: we went too far. The original approach was correct. The improvements were the problem.

The fix was to go back to basics. Add markets for diversification. Move toward longer time frames. Keep the rules simple. Do not introduce additional filters, conditions, or parameter adjustments aimed at improving equity curve aesthetics. The improvements that actually matter come from more markets and more diversification, not from more complexity in the rules themselves.

What Jerry does instead of optimization

Jerry’s current approach to research is to extract as little as possible from the backtest, not as much as possible. His framework is built around a simple principle: here is a good place to buy, here is a good place to sell, and that is enough. The parameters around entry and exit are chosen from a range that makes conceptual sense, not optimised to find the single best setting.

He runs multiple systems across a wide range of parameter values, from the shortest timeframe he is comfortable with to the longest, then trades systems at various points within that range without obsessing about which specific setting is best. The reason is straightforward: when you look at 40 or 50 years of data across many markets, systems from the shortest to the longest all make roughly the same amount of money. That pattern is not a coincidence. It is telling you something about the nature of trends and why trying to find the single best parameter is mostly noise.

What he focuses on instead: the trade statistics. Average win. Average loss. Average trade. Win percentage. Those numbers, taken across a robust dataset, tell him whether the edge is real. The equity curve, the drawdowns, the Sharpe ratio, the volatility of returns, those he largely ignores during the design process, because trying to optimise them is how the over-optimisation trap starts.

The problem with smoothing equity curves

Jerry made a direct point about Sharpe ratios and smooth equity curves: they are where over-optimisation begins. Whenever you try to reduce drawdown, increase the Sharpe ratio, or smooth the ride for investors, you are making decisions that improve the historical appearance of the system at the cost of its live robustness.

A system optimised to produce a smooth equity curve will look excellent in the backtest. In live trading, it will encounter situations the optimisation did not account for, and the results will be worse than a simpler, less-optimised approach would have produced. The smoother backtest equity curve is a red flag, not a positive signal, unless it emerged naturally from a simple system with few parameters.

The right response to a volatile equity curve is not to adjust the parameters until the curve looks better. The right response is to accept the volatility, trade at appropriate leverage for your drawdown tolerance, and let the system run. The drawdowns and whipsaws are part of the process. Trying to remove them from the backtest removes them from your confidence intervals too, which means you are unprepared when they appear in live trading.

The case for multiple simple systems over one complex system

Rather than building one highly optimised system with many rules, Jerry’s preference is multiple simple systems with very few rules each. A 50-day breakout system. A 75-day breakout system. A 100-day breakout system. Each with one entry, one exit, and a stop. Run them all simultaneously across a wide universe of markets.

This structure achieves diversification without requiring optimisation. The different systems will enter and exit at different times, producing exits with meaningfully different P&L profiles even when the entries were on the same day. That difference in exits is the actual source of diversification across correlated systems, not the entries. Jerry addressed the correlation question directly: yes, all the systems are likely long the same markets during strong trends. That is not a problem. They are all making money. The diversification comes when those trends end and the different exit rules produce different outcomes.

For traders with limited capital who can only run one system, Jerry’s advice is to start with a medium-to-long-term trend following approach. Something simple. Get comfortable with it. Let it work. As capital grows, add a second system with a different timeframe. The path from one system to four or five systems is gradual, but it starts with having one good simple system and following it consistently.

The consistency principle: the actual edge

Jerry’s clearest statement in this conversation was about what actually makes a system work. Not the parameters. Not the entry and exit rules. Not the choice of markets. Those things matter, but they are table stakes. The real edge, the thing that makes a mediocre systematic approach outperform a sophisticated discretionary one, is always doing the same thing. Consistency.

He described this as a superpower. If you trade smaller during drawdowns rather than stopping some trades, you can still do all the trades your system signals. That is the right approach. If you cut some trades during drawdowns to manage risk, you are breaking the statistical foundation of the system. The system needs all the trades to produce its expected performance. Skipping trades during rough patches means you will be underinvested when the big trends arrive.

The cutback rule, as Jerry calls it, is about reducing leverage, not reducing trade frequency. Cut your position sizes during drawdowns if you need to. But keep taking every signal. The moment you start deciding which trades to skip, you are adding discretion to a system that was designed to be systematic, and the backtest results no longer apply to what you are actually doing.

Leaving fate in the hands of the markets

Jerry ended the conversation with something that lands differently after 40 years of experience. His big lesson from trend following is that the one thing you need to be successful, large sustained trends, is completely outside your control. You cannot manufacture trends. You cannot force the market to move. You can only position yourself correctly and wait.

The control you do have is entirely about process: did you follow the system, in the right way, on every signal? That is it. Everything else is the market’s job. Jerry is explicit that he prefers a mediocre systematic approach executed with total consistency over a sophisticated approach that can break. A system that is followed precisely, with simple rules and wide market coverage, will capture the trends when they come. The only way to miss them is to be out of the market or not fully invested when they arrive.

His firms houses, boats, and standard of living came from hanging onto long-term trends and not getting out too quickly. The one complaint he hears most from other traders: “I got out too soon.” That is the opposite of the mistake most people expect to make. It is also the most expensive one.

Key principles for avoiding over-optimization

If live performance immediately diverges from the backtest in a way you’ve never seen, the system is probably over-optimised. Don’t wait a year to decide.
Equity curve smoothness is a red flag during design, not a goal. Simple systems with more markets are more robust than complex systems with smooth backtests.
Take as little from the backtest as possible. A good entry point, a reasonable exit, a stop. Done.
Diversify through multiple systems across wide market universes, not through adding filters and conditions to a single system.
During drawdowns, reduce leverage. Don’t reduce trade frequency. The system needs all its trades.
The edge comes from always doing the same thing, not from having the best parameters.

Related episodes

Get the show notes & transcript

Want to build robust trading systems that hold up when you go live? Subscribe to the Better System Trader podcast for weekly interviews with the world’s top systematic traders and quantitative researchers.