You have 250 daily returns with a mean of 0.05%. Is that mean distinguishable from zero? Draw 250 returns at random from your own sample, with replacement, compute the mean, and repeat 10,000 times. If 8% of those means come out below zero, your one-sided p-value is about 0.08.
The appeal is that it makes no assumption of normality, which matters because returns have fat-tails and the textbook t-test understates uncertainty on short samples.
The limitation is that it can only reuse what you have. A bootstrap of ten calm years cannot produce a crash, so bootstrapped drawdown estimates from a benign sample are optimistic by construction.
Related: block-bootstrap, monte-carlo-simulation