Kevin Davey has been trading for more than 25 years, finished in the top three of the World Cup Championship of Futures Trading three times — including first place — and has been trading full-time since around 2007 or 2008. His book, Entry and Exit Confessions of a Champion Trader, documents 52 strategies covering entries and exits in practical detail.
In this episode, he covers how to evaluate entries and exits systematically, why popular approaches often underperform, and what actually separates a robust entry from one that just looks good in a backtest.
Watch the full episode below, then read on for the complete breakdown.
Why entries get all the attention
Entries dominate most trading discussions. Kevin’s explanation: entries feel like control. You choose when and where to get in. Exits, by contrast, feel like the market does what it wants to you. That asymmetry in perceived control is why traders obsess over entries and spend far less effort on exits — even though exits are where most of the money is actually made or lost.
The practical implication is that traders often spend time optimizing entries that barely matter while ignoring exit design, which has a much larger effect on system performance.
The n-bar exit method for testing entries
Kevin’s core method for evaluating whether an entry has any signal at all: attach a neutral n-bar time exit. No stop, no target — just hold for one bar, then two bars, then three, and track the average profit at each point.
If the entry has a genuine edge, you should see the n-bar profitability start positive and stay positive as you extend the hold. If profitability is erratic or consistently negative, the entry has no signal — regardless of how it looks in a fully optimized backtest.
This technique strips out the influence of stop and target placement. It isolates the entry itself. If a signal can’t produce positive expectancy with a neutral exit, layering on an optimized exit won’t fix it.
Entry logic: does it need to make sense?
Kevin prefers entries with a logical basis. He wants to understand why a pattern should work before he tests it. The reason isn’t philosophical — it’s practical. A pattern with a conceptual reason behind it is easier to stick with during drawdowns. If you only know that something worked historically, losing confidence during a losing stretch is much easier.
That said, he’s encountered entries that passed all his testing for reasons he didn’t fully understand — including one that originated from a coding mistake. His approach: let the results prove themselves in real time before discarding something just because the logic is unclear. If it keeps working forward, the logic question becomes less important.
Indicator categories and what tends to work
Kevin’s testing found that different indicator categories have different base rates of success:
- Momentum indicators (e.g., close minus N-bars close) tend to work more often than not
- RSI also shows up frequently as a useful input
- Pattern-based entries, which use no optimizable parameter, reduce the risk of fitting — no lookback period to over-optimise
- Most other indicators, tested individually with realistic trading costs, fell in a range of 45 to 55% accuracy — and once slippage and commissions were added, most of those lost money
The implication is not that most indicators are worthless. It is that combining them, or using them in specific contexts, may produce something usable — but testing each one in isolation first is a necessary step. Don’t assume an indicator that looks interesting in a chart review will actually produce edge in a systematic test.
The dangerous belief: if it can be optimised, it must be
Kevin identifies over-optimisation as the single biggest failure mode in system development — and it is especially dangerous around entries and exits because both have many optimisable parameters.
His warning: creating a great-looking backtest is not difficult. Given enough degrees of freedom, any historical series can be fit. The skill is not in producing high backtest returns — it is in producing returns that hold out-of-sample. These are almost opposite skills.
A heuristic he uses: the moment you start asking yourself whether you have over-optimised, you have already passed the point. That internal question arriving earlier than it once did is a sign of experience. Newer traders ask it much later, or never.
He is also skeptical of the arms race for faster computers. The need to run thousands of optimisation iterations is not a hardware problem — it is a signal that the development process has too many degrees of freedom from the start.
Entry symmetry as a curve-fitting safeguard
For long-short strategies, Kevin prefers symmetric entries: the same parameter for long and short. If the long entry uses a 10-bar breakout, the short entry also uses a 10-bar breakout.
Markets can fall faster than they rise, and some arguments exist for asymmetric entries on stock indices. Kevin acknowledges this but still defaults to symmetry — not because it always produces the best backtest, but because it prevents parameter multiplication. A separate long parameter and short parameter doubles the optimisable inputs immediately. Symmetry keeps the system simpler and reduces the chance of fitting a historical anomaly.
Market regimes: a useful filter with a big cost
Kevin has experimented with regime filters — trading only long when price is above a 200-bar moving average, for example. Simple regime rules can help. More elaborate multi-regime frameworks are harder to implement than they sound.
His observation on regime-switching models: they tend to be late at both ends. In a bull market, the indicators don’t trigger until 20% in. And at the end, they stay bullish 20% too long. So instead of capturing the full move, you capture around 60% — and during the parts you missed, you were likely positioned the wrong way. The net benefit often does not justify the added complexity and the new optimisable parameters the regime model introduces.
Exits: more important, less discussed
Kevin’s position on exits: they tend to matter more than entries, but receive less attention because they feel less controllable. His research and workshop testing produced a result he hadn’t anticipated: some of the best exits were entry signals used as exit triggers — not as stop-and-reverse signals, just plain exits.
For example, if a 10-bar breakout high is a long entry signal, using that same signal as a short exit (without reversing) often outperformed conventional exit approaches. The implication is that the book’s 52 entry methods are also a source of 52 potential exit ideas — a dimension most traders haven’t considered.
Exit complexity: simple usually wins
Kevin’s own systems are typically simple on the exit side. Most use a fixed stop loss and nothing else. Some use a stop-and-reverse approach where he stays continuously in the market. He avoids combining multiple exit types — stop loss, trailing stop, profit target, and break-even stop simultaneously — because each adds optimisable parameters, and the tendency to adjust them “just to see what happens” is what produces over-fitted systems.
His rule: every exit variable you add is one more thing you will eventually be tempted to optimise. Keep the count low from the start.
The interaction problem
Testing entries and exits independently only gets you so far. Kevin’s consistent finding: the same entry can perform very differently depending on the exit, and the same exit can work well with some entries and poorly with others. The interaction between the two is where a lot of the real system behaviour lives.
This means there is no shortcut to testing the combination. An entry that looks strong with an n-bar exit may underperform once a real stop is introduced. An exit that works well in one configuration may hurt performance in another. The system must be evaluated as a whole, not as a collection of independently tested components.
Key takeaways
- Use the n-bar exit method to isolate whether an entry has signal. If it can’t produce positive expectancy without an optimised exit, the entry has no edge.
- Most indicators tested individually lose money once trading costs are included. The bar for a useful entry is higher than it appears.
- Symmetric long-short entries reduce parameter count and curve-fitting risk, even if they sacrifice some theoretical performance.
- Entry conditions can be used as exits. This is an under-explored source of exit ideas.
- When you start asking whether you have over-optimised, you already have.
- Exit design matters more than most traders think, but complexity in exits tends to backfire. Simple exits hold up better out-of-sample.
Related episodes
- How to spot a trading scam: Kevin Davey
- How to detect a failing trading strategy: Kevin Davey
- How to avoid curve fitting: Jeff Swanson
Want to learn more about building better trading systems? Subscribe to the Better System Trader podcast for weekly interviews with the world’s top systematic traders and quantitative researchers.
