A strategy backtests at 54% win rate. Live, it settles at 51%. Most traders look at that three-point gap and shrug it off as noise.
It isn't noise. It's the difference between an account that survives a decade and one that doesn't make it through the year.
The math nobody runs
Here's the part nobody explains well. Risk of ruin doesn't move in a straight line with win rate. It behaves more like a number raised to a power, and the power is roughly how many risk units your capital can absorb before you're wiped out.
A small drop in win rate doesn't shave a little off your odds of survival. It compounds, trade after trade, until the odds look nothing like where you started.
I've seen the table that makes this concrete. At a 40% win rate with 5% risked per trade, the probability of blowing the account rounds to 100% — it's basically a matter of time. Drop the risk to 2% and lift the win rate to 50%, and ruin probability falls to around 20%. Push to a 60% win rate at 1% risk per trade, and it drops under 1%.
Small moves in either input — win rate or position size — don't nudge the outcome a little. They swing it by orders of magnitude.
This is the same reason the Kelly criterion punishes overconfidence so hard. Kelly sizing scales directly with your estimated edge, and if your real edge is smaller than the number you plugged in, you're sizing for a strategy you don't actually have. Half-Kelly exists because everyone's edge estimate is wrong in one direction or another, and the fix underneath it is the same one that shows up everywhere in this business: size for the numbers you can prove, not the ones you hope are true.
Where people get it backwards
One mistake shows up before any of the others. Traders treat win rate as a fixed number instead of a moving target. A point or two of drift feels too small to act on, so it gets filed under noise instead of triggering any recalculation. By the time three or four bad weeks in a row make the pattern impossible to ignore, the account has already taken the hit that recalculating early would have avoided.
Win rate alone doesn't tell the story either. A 51% win rate with a 2:1 payoff is a different animal than a 51% win rate with a 1:1 payoff, and conflating the two is how people misjudge their own edge.
The costliest mistake is treating the backtest number as a floor. It's almost always a ceiling. Slippage, latency, and a parameter set that was quietly chosen because it worked don't lower your expectancy a little — they eat directly into the one input risk of ruin is most sensitive to. A backtest that shows 54% was never promising you 54% live — it was showing you the best 54% a curve-fitting process could find in the data you gave it.
The fix isn't complicated. It's just something almost nobody actually does. Every time live win rate lands more than a point or two below backtest, don't ask whether that's normal. Ask what your risk of ruin looks like at the new number, at your current position size.
One of our own strategies backtested at 54% over four years of data and settled closer to 51% across its first six months live. Nothing was broken — no bug, no bad fills, no data error. That's just what markets do to a backtest. We cut position size by a third the week the gap showed up, and left it there until the sample got big enough to trust.
None of this means small edges aren't tradeable — some of the best systematic strategies run on win rates in the 30s and 40s and survive because their sizing respects how thin the edge really is. If you're running an automated strategy and have never calculated your risk of ruin at your actual live numbers, that's worth an afternoon before a drawdown does the calculation for you. It takes less time than the drawdown that follows from skipping it. We publish live-versus-backtest divergence for every bot at v33systematic.com, so the gap is visible before your capital is on the line, not after.