Past exam of the mathematics course of the University of Cambridge / 2003 / iii / Paper 43 / 2

The empirical distribution gives each observed value mass $1/n$, so its empirical distribution function isFor each of $B$ independent replications, draw $n$ values with replacement from this empirical distribution, recompute the same estimator ${\hat{\theta }}^{\ast (b)}$, and use their sample standard deviation:This estimates the conditional bootstrap standard error; dividing it by $\sqrt{B}$ would instead estimate simulation uncertainty in the bootstrap mean and is not the requested sampling standard error.

When the original observations are distinct, the first resampled index may be arbitrary. After $j$ distinct indices have been drawn, the next index avoids them with probability $1-j/n$. Multiplying conditional probabilities givesHere $P_{\ast }$ denotes probability conditional on the data.

For $0\le t\le \theta$, independence and the uniform distribution give $P(X_{(n)}\le t)=(t/\theta {)}^n$. Consequently, for fixed $x\ge 0$ and $n>x$,For $x<0$ the survival probability is one. Thus the scaled endpoint gap converges in distribution to a unit-rate exponential law; it is not exactly exponential at finite $n$.

The plug-in estimator principle replaces the unknown sampling distribution by its empirical distribution and evaluates its population functional at that fitted distribution. Here the fitted support endpoint is $\hat{\theta }=X_{(n)}$, also the maximum-likelihood estimate: the likelihood is ${\theta }^{-n}{1}_{\{\theta \ge X_{(n)}\}}$ and decreases over its allowed support. For an empirical bootstrap sample $Y_1,\dots ,Y_n$, the corresponding gap isAlmost surely the original maximum is unique. The bootstrap gap equals zero precisely when this observation is selected at least once, soThe true limiting exponential distribution has no atom at zero. This nonvanishing discrepancy proves bootstrap failure for a uniform endpoint: the bootstrap distribution cannot converge to the correct limit, even though the endpoint estimate itself is consistent. It fails to recreate the unsampled population tail beyond the observed maximum. A parametric bootstrap drawing from $U[0,X_{(n)}]$, rather than from the discrete empirical sample, reproduces the scaled-gap law here.