Light tails: Gibbs conditional principle under extreme deviation

Abstract: Let

X_{1}, .., X_{n}

denote an i.i.d. sample with light tail distribution and

S_{1}^{n}

denote the sum of its terms; let

a_{n}

be a real sequence going to infinity with

n .

In a previous paper (cite{BoniaCao}) it is proved that as

n i g h t a r r o w i n f t y

, given

l e f t (S_{1}^{n} / n > a_{n} i g h t)

all terms

X_{i_{e x t}}

concentrate around

a_{n}

with probability going to 1. This paper explores the asymptotic distribution of

X_{1}

under the conditioning events

l e f t (S_{1}^{n} / n = a_{n} i g h t)

and

l e f t (S_{1}^{n} / n g e q a_{n} i g h t)

. It is proved that under some regulatity property, the asymptotic conditional distribution of

X_{1}

given

l e f t (S_{1}^{n} / n = a_{n} i g h t)

can be approximated in variation norm by the tilted distribution at point

a_{n}

, extending therefore the classical LDP case developed in Diaconis and Freedman (1988) . Also under

l e f t (S_{1}^{n} / n g e q a_{n} i g h t)

the dominating point property holds. It also considers the case when the

X_{i}

's are

m a t h b b R^{d} -

valued,

f

is a real valued function defined on

m a t h b b R^{d}

and the conditioning event writes

l e f t (U_{1}^{n} / n = a_{n} i g h t)

or

l e f t (U_{1}^{n} / n g e q a_{n} i g h t)

with

U_{1}^{n} := l e f t (f (X_{1}) + .. + f (X_{n}) i g h t) / n

and

f (X_{1})

has a light tail distribution

.

As a by-product some attention is paid to the estimation of high level sets of functions.

This page was built for publication: Light tails: Gibbs conditional principle under extreme deviation