Minimum track record length
Two years of data and a Sharpe of 1.0 is not evidence of anything. This is the formula that says so precisely, and the number it returns is longer than almost anyone expects.
The formula
From Bailey and Lopez de Prado. With an annualised Sharpe SR, a benchmarkSR*, skew g3, kurtosis g4 and confidence level a:
MinTRL = 1 + [ 1 - g3·SR + (g4-1)/4 · SR² ] · ( Za / (SR - SR*) )²
Every figure below is computed from that expression when this page is built, at 95 percent confidence against a benchmark of zero. Nothing here is copied from a paper's table.
How many years you need
| Annual Sharpe | Normal returns | Skew -1, kurtosis 6 |
|---|---|---|
| 0.50 | 13.2 years | 20.6 years |
| 0.75 | 7.2 years | 12.8 years |
| 1.00 | 5.1 years | 9.8 years |
| 1.50 | 3.6 years | 7.4 years |
| 2.00 | 3.0 years | 6.4 years |
| 3.00 | 2.7 years | 5.6 years |
The first column is the optimistic case, and it is already uncomfortable: a Sharpe of 1.0 needs about 5.1 years before you can say with 95 percent confidence that it is above zero. The second column is what most real strategies look like, and it pushes the same result to roughly 9.8 years.
The uncomfortable consequence
Most retail backtests run on five to ten years of data and report Sharpe ratios between 0.5 and 1.5. Read the table against that: a large share of published backtests are too short to support their own conclusion, before anyone even asks how many variants were tested.
This is also why long histories are worth paying for, and why a strategy that only works on recent data is a harder sell than it looks. See theranking for how much history each platform bundles.
Doing it yourself
Four inputs, all of which you already have: your annualised Sharpe, the skew and kurtosis of your return series, and a confidence level. Use 1.645 for 95 percent. The one judgement call is the benchmark: zero asks whether you have any edge at all, while setting it to the Sharpe of buying and holding asks the sharper question of whether you are beating the obvious alternative, and it lengthens the requirement considerably.
A caution worth stating plainly: passing this test is necessary, not sufficient. It only says your record is long enough to be readable. Whether the number itself survives the search that produced it is adifferent question.
Frequently asked questions
What is minimum track record length?
The shortest record over which an observed Sharpe ratio can be declared greater than a benchmark at a chosen confidence level. It comes from Bailey and Lopez de Prado, and it answers the question people usually settle by intuition: is this track long enough to mean anything?
Why does a higher Sharpe need less history?
Because the signal is larger relative to the noise. The requirement falls quickly at first and then flattens: going from 0.5 to 1.0 saves years, going from 2.0 to 3.0 saves months. Beyond a Sharpe of about 2 the record length stops being the binding constraint and the trial count takes over.
Does this apply to a backtest or only to live trading?
The arithmetic is identical, but the interpretation is not. On live results it tells you whether the record is long enough. On a backtest it is a floor rather than a bar: a backtest that fails the length test is definitely too short, while one that passes still has to answer for how many variants were tried, which this formula does not see.
Why does skew make such a difference?
Because the Sharpe ratio only looks at mean and standard deviation, so it is blind to a return stream that grinds upward and occasionally falls off a cliff. The formula puts skew and kurtosis back in. At a Sharpe of 1.0, normal returns need about 5.1 years, while returns with skew of -1 and kurtosis of 6 need about 9.8. Most option-selling and carry strategies sit in the second column.
Do any backtesting platforms compute this?
It is not part of the standard reporting on any of the 20 platforms we assessed, which is why the table on this page exists. The formula needs four inputs you already have, so a spreadsheet is enough.
Related
Educational content only, not investment advice. The table describes a published statistic and is not a judgement on any specific strategy.