t-statistic for a strategy's mean return
The sample mean divided by its estimated standard error. Because the volatility is estimated from the same data, it follows a t distribution rather than a normal.
01The formula
- the t-statistic
- the average daily return
- the sample standard deviation of daily returns
- the number of days
- Student's t distribution with n − 1 degrees of freedom
- the null hypothesis: the true mean return is zero
- the daily Sharpe ratio, x̄/s
02See it move
03Where it is used
- Quant researcherReports the t-statistic of every backtested signal and asks for well above 2 before trading it.
- Portfolio managerJudges whether a manager's track record shows skill, given its length.
- Risk managerTests whether a desk's average P&L is distinguishable from zero.
04Why traders use it
It links the Sharpe ratio to evidence: t is the daily Sharpe times √n. A high Sharpe over a short history is weak evidence; the t-statistic says exactly how weak.
05Worked example
An average daily return of 0.10% with volatility 1.0% over 250 days gives t = 1.581. The 5% critical value is 1.970, so the p-value of 11.51% does not reject 'no edge', even with an annualised Sharpe of 1.59.
06How it is built
- Estimate the standard error with the sample volatility: .
- Divide: .
- Because is noisy, has fatter tails than a normal: it follows when the returns are normal and .
- Rewrite it as : the daily Sharpe ratio times .
07The common mistake
Testing many strategies and reporting the best t. If you try 20 ideas, one of them will usually show t above 2 by luck alone.
08In Python and LaTeX
import numpy as np from scipy.stats import t as tdist xbar, s, n = 0.001, 0.01, 250 t = xbar / (s/np.sqrt(n)) print(round(t, 3), round(tdist.isf(0.025, n-1), 3), round(2*tdist.sf(abs(t), n-1), 4))
t = \frac{\bar{x}}{s / \sqrt{n}} \sim t_{n-1} \ \text{under } H_0: \mu = 0, \qquad t = \widehat{\text{SR}}_{\text{daily}}\sqrt{n}09Where it fits
Rather not code the maths yourself? Describe your trading idea and let AI build the bot.
Build a bot with AgenKitThe worked example and the model run the same code, so every number matches. Models simplify markets on purpose. This is education, not financial advice.