I failed to prove the exercise 27.15. in Billingsley's textbook.

Let $X_1,X_2, \dots$ be i.i.d. positive random variables with expectation $m$ and variance $\sigma^2 \in (0,\infty)$.

If $N_t = \max \{n : S_n \leq t \}$, then $$\frac{S_{N_t} -t}{\sqrt{t}} \Rightarrow 0.$$

I already proved that $\frac{t}{N_t} \rightarrow m$ as $t \rightarrow \infty$ and the following facts (what author gave me as hints):

If $X_i$ are i.i.d. with $\mathbb{E}X^2 < \infty$, then $n \, \mathbb{P}(|X_1| \geq \varepsilon \sqrt{n}) \rightarrow 0$ and $\displaystyle n^{-\frac{1}{2}} \max_{1\leq k \leq n} |X_k| \rightarrow 0$ in probability.

Any help will be appreciated.



Since $S_{N_t}\le t$ and $S_{N_t+1}>t$, we have $$\mathbb P(|S_{N_t}-t|>\sqrt{t}\varepsilon)=\mathbb P(S_{N_t}<t-\sqrt{t}\varepsilon,S_{N_t+1}>t)\le\mathbb P(X_{N_t+1}>\sqrt{t}\varepsilon).$$ Let $M_t=N_t+1$ and observe that $\frac{t}{M_t}\to m$. For $0<\delta<m$ we have $$\mathbb P(X_{M_t}>\sqrt{t}\varepsilon)\le\mathbb P\left(\left|\frac{t}{M_t}-m\right|\ge\delta\right)+\mathbb P\left(X_{M_t}>\sqrt{M_t}\varepsilon\sqrt{m-\delta},\,\left|\frac{t}{M_t}-m\right|<\delta\right).$$ Let $\gamma=\varepsilon\sqrt{m-\delta}$ and $n=\lfloor\frac{t}{m+\delta}\rfloor$. If we further assume that $\delta<\frac{m}2$, then it follows that $n\le M_t< 3n$ on the event $\{|\frac{t}{M_t}-m|<\delta\}$. This implies that the second term above is bounded by $$\mathbb P(\max_{n\le k<3n}X_k>\sqrt{n}\gamma)\le2n\mathbb P(X_1>\sqrt{n}\gamma).$$ Since $n\to\infty$ as $t\to\infty$, putting everything together shows that $\frac{S_{N_t}-t}{\sqrt{t}}\to0$ in probability. Note that the second hint wasn't explicitly used, but the idea behind it was important in our final estimate of the second term.

  • $\begingroup$ Thank you very much! I read your answer and get critical ideas to prove this. However, I proved this one by assuming $\delta < \frac{m}{2}$ not $\delta > \frac{m}{2}$. $\endgroup$ – cdamle May 3 '17 at 14:03
  • $\begingroup$ You're right, that was a typo. I will edit. $\endgroup$ – Jason May 3 '17 at 14:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.