8
$\begingroup$

Could you please help with the proof of the proposition:

Let $G(\cdot)$ be bounded, continuous, strictly increasing function on $\mathbb{R}.$ Let ${\xi_t},\, t\in\mathbb{Z}_+$ be i.i.d random variables.

Would it be true to say that we have something like the uniform Law of large numbers here? Why? $$ \sup_{x\in\mathbb{R}}\;\left|\;n^{-1}\sum_{t=1}^n G(x+\xi_t) - \mathbf{E}G(x+\xi_1)\;\right| \stackrel{P}\longrightarrow 0, \quad n\to\infty. $$

Thanks in advance!

$\endgroup$
  • $\begingroup$ This thread mathoverflow.net/questions/121238/… could interest you. $\endgroup$ – Davide Giraudo Jul 9 '13 at 19:41
  • $\begingroup$ Thanks, but unfortunately they're talking about compacts there $\endgroup$ – Dan Jul 10 '13 at 11:31
  • 1
    $\begingroup$ An idea could be the following: after having rescaled $G$, we can assume that $G$ is a cumulative distribution function. Consider $(\eta_i,i\in\Bbb Z)$ an i.i.d. sequence of random variables independent of $(\xi_i,i\in\Bbb Z)$, say on an other probability space $(\Omega',\mathcal F',\mu')$. Then we can use Glivenko-Cantelli's theorem for the product space and the i.i.d. sequence $(\xi_i-\eta_i,i\in\Bbb Z)$. $\endgroup$ – Davide Giraudo Jul 10 '13 at 13:02
1
$\begingroup$

Probably an application of Glivenko-Cantelli's theorem as I suggested in the comments will work.

Assume that the $\xi_t$ are defined on a probability space $(\Omega,\mathcal F,\mu)$. We can assume that $G$ is the cumulative distribution function of a real valued random variable. Let $(\eta_t,t\in\Bbb Z)$ be a collection of independent random variables with cumulative distribution function $G$, defined on $(\Omega',\mathcal G,\nu)$. Using Glivenko-Cantelli's theorem for the product space and the family of i.i.d. random variables $(\xi_t-\eta_t,t\in\Bbb Z)$ gives $$\sup_{u\in\Bbb R}\left|\frac 1n\sum_{i=1}^n\chi_{\{\eta_i(\omega')-\xi_i(\omega)\leqslant u\}}-\mu\otimes\nu\{\eta_i(\omega')-\xi_i(\omega)\leqslant u\}\right|\to 0$$ for $\mu\otimes\nu$ almost every $(\omega,\omega')$. Then integrate with respect to $\nu$.

An alternative approach would use the result of the paper

Ramon van Handel, The universal Glivenko-Cantelli property, Probab. Th. Rel. Fields 155, 911-934 (2013)

which is available here. We consider $\mathcal F:=\{t\mapsto G(x+t), x\in\Bbb R)\}$, which is a Glivenko-Cantelli class as $G$ is bounded. But it seems quite overkill for this problem, since their result is more general.

$\endgroup$
  • 1
    $\begingroup$ Thanks for the article! Will think about it $\endgroup$ – Dan Jul 11 '13 at 13:34

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.