Sample Size Requirements for Detecting a Real Edge

Confirming a modest, genuine Sharpe ratio can take decades of clean data, using the same arithmetic behind its confidence interval.

Detecting a real trading edge with reasonable statistical confidence takes far more data than most backtests provide, and the required amount of data can be estimated directly from the target Sharpe ratio using the same standard error approximation that governs its confidence interval. A strategy with a modest but genuine edge can require decades of clean, out-of-sample history before the data alone can rule out that the result is noise.

Using the approximation that the standard error of an estimated Sharpe ratio near zero is about the square root of one divided by T, where T is the number of years of data, the number of years required to detect a true Sharpe ratio with 80 percent power at a one-sided 5 percent significance level is approximately the quantity 1.645 plus 0.84, divided by the target Sharpe ratio, all squared. The constant 1.645 is the one-sided 5 percent critical value, and 0.84 is the value needed to reach 80 percent power, so their sum is about 2.49.

For a true annual Sharpe ratio of 1.0, the required T is approximately 2.49 squared, which is about 6.2 years. For a true Sharpe ratio of 0.5, half the size, the required T roughly quadruples to about 24.7 years, since halving the target squares the multiplier on the sample size needed. For a true Sharpe ratio of 2.0, a genuinely exceptional and rare figure, the required T falls to about 1.5 years.

The calculation above uses the simplified, near-zero approximation for the standard error, and the fuller formula, which includes a term proportional to the Sharpe ratio squared, pushes the true required sample size higher for any Sharpe ratio that is not close to zero. It also assumes independent, normally distributed annual returns, an assumption real return series violate through skewness and fat tails, both of which tend to widen the true standard error further. Every figure above should be read as an optimistic floor rather than a guarantee.

None of these adjustments change the basic conclusion, only the exact number: a moderate, genuine edge takes years to decades of clean data to confirm statistically, and any specific figure quoted for a specific Sharpe ratio should be read as an order of magnitude rather than a precise deadline.