A low win rate is not a red flag in a backtest

The golden cross study on this site wins 19.59 percent of its trades. That is not a typo. Buy when the 50 day moving average crosses above the 200 day average, hold the trend, and across 220 US large caps from 2015 to 2024 the trade closes positive fewer than one time in five. The same study returned an average of 9.02 percent per ticker over the decade, the best result of the four strategies compared in this piece.

The three weeks tight study, a breakout out of a narrow consolidation, wins 48.66 percent of its trades, more than double golden cross. Its return over the same universe and the same ten years was 2.66 percent. A strategy that wins less than half as often produced more than three times the return. If win rate told you what you needed to know, this pair of numbers would not exist.

Four strategies, one universe, one decade

All four numbers below come from the same 220 US large caps and the same window, January 2015 to December 2024, so the comparison is not confounded by different stocks or different years. NR7 volatility breakout wins 10.15 percent of trades and returned negative 0.09 percent, a loss. Golden cross wins 19.59 percent and returned 9.02 percent. Three weeks tight wins 48.66 percent and returned 2.66 percent. Williams %R reversal wins 69.13 percent and returned 27.21 percent.

The number that tracks return better

Tenachine reports a second figure alongside win rate on every study page: average risk to reward, defined as gross profit divided by gross loss, pooled across every trade in the study. It is not an average of each ticker's own ratio, it is the total dollars won divided by the total dollars lost across the whole run. For the same four strategies it reads 0.6964 for NR7, 2.3193 for golden cross, 1.1979 for three weeks tight, and 1.3531 for Williams %R reversal.

A ratio below 1.0 means the losing trades cost more in total than the winning trades made, whatever the win rate says. NR7 sits at 0.6964. Its win rate, 10.15 percent, is not the reason it lost money. The reason is that its wins were not big enough, or its losses were too big, relative to how rarely it won. Win rate tells you how often. Risk to reward tells you how much. A backtest needs both before it means anything.

Why a low win rate can still work

Trend following strategies like golden cross are built around exactly this trade-off on purpose. Most attempts to catch a trend fail early and get cut for a small loss. The rare attempt that catches a real multi-year move keeps running and keeps compounding. Nineteen and a half small losses and cheap exits for every big winner is the design, not a flaw in it. The 2.32 risk to reward figure is the evidence that the design worked over this particular decade, not a guarantee that it will keep working.

Mean reversion and short-hold strategies like three weeks tight and Williams %R reversal work the other way round. They aim to be right often, on smaller moves, and accept that any single win is unlikely to be dramatic. A 48.66 percent or 69.13 percent win rate is normal for that style. It is not evidence of a better strategy than golden cross, it is evidence of a different one, answering a different question about the same market.

The part win rate does not cover

Williams %R reversal is the best performer of the four by return, 27.21 percent, and also has the highest win rate, 69.13 percent. It also carries the worst average maximum drawdown of the four, negative 31.88 percent, more than three times golden cross's negative 9.35 percent and roughly five times three weeks tight's negative 6.48 percent. A high win rate did not protect it from a rough stretch. Win rate measures how often a trade closes positive. It says nothing about how deep the account went underwater on the way to that result, and a strategy with an appealing headline return can still be uncomfortable, or unfinanceable, to actually hold through.

Coverage differs across these four too, and it matters before comparing anything else. Golden cross completed 193 of 220 tickers, NR7 completed 200, Williams %R reversal completed 214, and three weeks tight completed only 142, the thinnest sample of the four. A win rate and a return calculated over 142 completed runs out of 220 rests on a smaller base than one calculated over 214, and that gap is worth remembering before treating any of these numbers as precise.

How to actually use a win rate

1. Never read win rate by itself. Put it next to the risk to reward ratio from the same run before drawing any conclusion.
2. Check which side of 1.0 the risk to reward ratio falls on. Below 1.0, the strategy is giving back more than it wins even if it wins often.
3. Look at the drawdown separately. A high win rate does not buy protection from a bad stretch, as Williams %R reversal shows here.
4. Check how many tickers actually completed a run. A ratio built on 142 completions carries more noise than one built on 214.
5. Only then ask whether the win rate fits the style of strategy. A trend follower with a high win rate would be unusual. A mean reversion strategy with a low one would be too.

Why Tenachine shows both numbers together

Tenachine takes a strategy written in plain English, generates the backtest code, and runs it across a universe of US stocks rather than one favourite ticker. Every study page reports win rate and risk to reward as separate figures next to each other, along with drawdown and coverage, because a win rate on its own invites exactly the misreading this piece walks through. The four studies compared here, golden cross, NR7 volatility breakout, three weeks tight and Williams %R reversal, are not a ranking of which rule to trade. They are four honest answers to four different questions, over the same stocks and the same decade, and the win rate column is the one column that would have misled you about all four if you had read it alone.

None of this is a forecast. A decade in which trend following paid off and a narrow breakout rule barely broke even is a fact about 2015 to 2024, not a promise about the years after it. Past performance does not predict future results.