How to Read a Backtest: A High Win Rate Does Not Mean a Good Strategy

The backtest finishes and you get a row of numbers. Most people look at win rate first and total return second, then draw a conclusion. Those happen to be the two most misleading numbers on the panel.

Win rate alone carries no information

A strategy can win 90% of its trades and lose money for years, or win 35% and compound steadily. What decides the direction is expectancy:

expectancy = win rate × average win - (1 - win rate) × average loss

Two extreme but very common examples:

Strategy A wins three trades out of four. It feels wonderful to trade and the account shrinks steadily. Strategy B loses two out of three, makes you question your life choices while watching it, and is the one that works.

Trend following is structurally a Strategy B. Its win rate typically lands between 35% and 45%, because it places a bet at every plausible start of a trend and most plausible trends never materialise. When you see a low win rate, the correct next action is to look at the profit factor, not to close the tab.

Two panel metrics speak directly to this:

• Profit factor — gross profit divided by gross loss. Below 1.0 the strategy loses money; 1.0–1.3 will most likely not survive real transaction costs; above 1.5 is worth further study.
• Average holding period — in trend following, winning trades should be meaningfully longer than losing ones. If the two are similar, the stop or exit rule is truncating trends along with the noise.

Sample size: twelve trades prove nothing

This is the most thoroughly ignored item on the list. If a report contains a dozen trades, then however good the result looks, it is a record of luck rather than a property of the strategy.

A blunt rule of thumb: do not trust conclusions from fewer than 30 trades; statistical meaning starts somewhere north of 100. The reason is direct — trend-following returns are heavily concentrated in a handful of large winners, so the smaller the sample, the more the result depends on whether this particular slice of history happened to contain them. When the same parameters flip sign after you shift the start date, the usual explanation is not an unstable strategy but a sample too small to support the word "stable."

If a symbol produces too few trades over the available window, there are two legitimate responses: extend the window, or run the same rules across a basket of symbols and check whether the conclusion holds. The illegitimate response is to keep adjusting parameters until those twelve trades look good.

Max drawdown: whether the account survives long enough

Total return tells you the destination. Max drawdown tells you how hard the road is. People quit on the road, not at the destination.

The arithmetic of drawdowns is asymmetric, and it is worth memorising:

A strategy with 20% annualised return and a 55% max drawdown looks excellent on paper. Now picture it honestly: the account falls from $1,000,000 to $450,000 and stays there for over a year. In which month do you switch it off? If the answer is "I would switch it off," then your real annualised return for this strategy is whatever it was on the day you quit — not the 20% in the report.

Look at drawdown duration as well (how long from the peak until a new high). Depth is survivable; length usually is not.

You must compare against a benchmark

A return figure with no benchmark is meaningless. Through 2020 and 2021 almost any long-biased strategy produced a beautiful curve, because the underlying went up a great deal.

That is what excess return on the panel is for: strategy return minus buy-and-hold over the same window. If excess return is negative, you took on timing risk, transaction costs and the risk of missing moves, and got a worse result than doing nothing. At that point the strategy's value can only come from another dimension — typically a materially smaller max drawdown. That is in fact the usual shape of trend following: give up some of the upside, buy a shallower drawdown. Whether that trade is worth making requires looking at return and drawdown together, never one alone.

The panel also shows the benchmark's max drawdown, and putting the two side by side is the fastest judgment available:

• Higher return, shallower drawdown → worth pursuing.
• Lower return, materially shallower drawdown → a risk-preference trade-off; depends on what you want.
• Lower return, deeper drawdown → drop it, and do not try to rescue it with parameters.

Do not treat the Sharpe ratio as a universal yardstick

The Sharpe ratio divides return by volatility, which implicitly assumes returns are normally distributed. Trend-following returns are strongly right-skewed and fat-tailed. Scoring a trend strategy with Sharpe understates it systematically, because large upside moves land in the denominator too.

Sharpe is fine for comparing strategies of similar style. Across styles, drawdown and excess return are more reliable.

A checklist you can actually follow

After every run, ask six questions in order:

1. Are there enough trades (≥30, ideally ≥100)?
2. What is the profit factor? Is a low win rate compensated by it?
3. Can I genuinely sit through that max drawdown? How long did it last?
4. Is excess return over buy-and-hold positive or negative? Is the drawdown shallower?
5. Are commissions and slippage included? Does the conclusion survive one notch higher cost?
6. Does the conclusion point the same way on a different symbol and a different window?

If any one of the six has no answer, what you are holding is not a conclusion. It is a chart.

For why question 6 is mandatory, and why "tune until it looks good" is self-deception, see Three Traps in Backtesting: Survivorship Bias, Look-Ahead Bias, Overfitting.