The arithmetic nobody runs
A win rate is an estimate, not a fact. Thirty-one trades is a sample, and samples carry uncertainty. The standard error of a proportion is:
SE = sqrt( p × (1 − p) / n )
Plug in p = 0.77 and n = 31: SE = sqrt(0.77 × 0.23 / 31) ≈ 7.5%. The 95% confidence interval is roughly ±1.96 standard errors — 77% ± 14.8 points. So your “77% strategy” is, statistically, anywhere from 62% to 92%.
Read that range again. The bottom of it is 62%. You cannot tell, from 31 trades, whether you own a 77% machine or a 63% coin with a good month behind it.
Why the bottom of the range changes everything
For premium-selling strategies — straddles, strangles, iron condors, the whole income family — the economics live and die on the win rate, because the payoff is asymmetric: many small wins, occasional large losses.
At 77%, a typical short strangle with a 1:2.5 win/loss size ratio makes money comfortably. At 63%, the same strategy is roughly break-even before costs — and losing after STT, brokerage and slippage. Same strategy, same backtest: the difference between “deploy it” and “delete it” sits entirely inside a confidence interval most traders never compute.
How many trades do you actually need?
There is no magic number, but there are honest thresholds:
- ~100 trades — you're allowed to have an opinion. The 95% band narrows to about ±8 points.
- 200–300 trades — you're allowed to size up. The band tightens to ±5 points or better.
- More than one market regime — non-negotiable. Thirty-one trades from one calm quarter tell you how the strategy behaves in calm quarters. 2018's grind, March 2020's collapse, 2022's chop — a strategy that hasn't seen a regime change hasn't been tested; it's been flattered.
And count trades, not days. “Backtested over one year” sounds rigorous, but it means nothing if the strategy only fired 24 times.
Even enough trades can lie
Sample size fixes the statistical problem. Four structural errors still inflate results — look-ahead bias, survivorship bias, ignored Indian costs, and curve-fitting — and none of them throws an error. We take those apart, with current NSE/BSE numbers, in the honest backtest guide.
A checklist you can run today
Before you trust any backtest — yours or anyone's:
- Find n. If the trade count isn't shown, that's your first red flag.
- Compute the band: SE = sqrt(p(1−p)/n). Double it — that's roughly your ±.
- Ask what the strategy earns at the bottom of that band, after costs.
- Check the dates: does the window include at least one regime you'd hate?
- If n is under 100: collect more evidence. That's not caution — that's arithmetic.
Where this is built in
Every Algoshastra verdict shows the trade count next to the result and feeds it into a trust score — a small sample caps the score at “low confidence” no matter how good the headline number looks, and the verdict says not enough evidence instead of flashing a green light. You describe a strategy in plain English, it backtests on real NIFTY/SENSEX per-strike data with the full Indian cost stack, and the confidence context comes standard — including when the honest answer is that the strategy loses money.
Common questions
How many trades does a backtest need to be reliable?
There is no magic number, but useful thresholds exist: around 100 trades the 95% confidence band on a win rate narrows to roughly ±8 percentage points — enough to form an opinion. At 200–300 trades it tightens to ±5 points or better — enough to justify sizing up. Below ~100 trades, the honest conclusion is 'not enough evidence yet'. The trades should also span more than one market regime, not one calm stretch.
Is a 77% win rate good for a trading strategy?
The number alone cannot say. Win rate is an estimate with uncertainty that depends on the trade count. Over 31 trades, a reported 77% win rate has a 95% confidence interval of roughly 62% to 92% — statistically indistinguishable from a 63% strategy. For premium-selling option strategies with asymmetric payoffs, the difference between 77% and 63% is often the difference between profitable and losing after costs.
How do I calculate the confidence interval of a win rate?
Use the standard error of a proportion: SE = sqrt(p × (1 − p) / n), where p is the win rate and n the number of trades. The approximate 95% confidence interval is p ± 1.96 × SE. Example: p = 0.77, n = 31 gives SE ≈ 7.5%, so the interval is roughly 62% to 92%.
Why does my strategy win in backtests but lose in live trading?
The two most common reasons are statistical and structural. Statistical: the backtest had too few trades, so the reported edge was inside the noise band. Structural: the backtest flattered the strategy — look-ahead bias, survivorship bias, missing Indian costs (STT, brokerage, slippage), or curve-fitting parameters to the past. Both must be ruled out before trusting a result.
Read next
The honest frame
Investment in securities market are subject to market risks. Read all the related documents carefully before investing.
Backtested results are hypothetical, do not represent actual trading, and are not indicative of future results. This article is educational and is not investment advice or a recommendation; Algoshastra is a strategy-building and testing tool, not a registered investment adviser or research analyst. Past or backtested performance does not guarantee future returns.