EFFICIENT ESTIMATION OF A DISTRIBUTION FUNCTION UNDER QUADRANT DEPENDENCE

Let X-1, X-2, ..., be real-valued random variables forming a strictly stationary sequence, and satisfying the basic requirement of being eit her pairwise positively quadrant dependent or pairwise negatively quad rant dependent. Let F be the marginal distribution function of the Xis , which is estimated by the empirical distribution function F-n, and a lso by a smooth kernel-type estimate (F) over cap(n), by means of the segment X-1, ..., X-n. These estimates are compared on the basis of th eir mean squared errors (MSE). The main results of this paper are the following. Under certain regularity conditions, the optimal bandwidth On the MSE sense) is determined, and is found to be the same as that i n the independent identically distributed case. It is also shown that nMSE(F-n(t)) and nMSE((F) over cap(t)) tend to the same constant, as n --> infinity, so that one can not discriminate between the two estima tes on the basis of the MSE. Next, if i(n) = min {k is an element of { 1, 2, ...}; MSE (F-k(t)) less than or equal to MSE (F-n(t))}, then it is proved that i(n)/n tends to 1, as n --> infinity. Thus, once again, one can not choose one estimate over the other in terms of their asym ptotic relative efficiency. If, however, the squared bias of (F) over cap(n)(t) tends to 0 sufficiently fast, or equivalently, the bandwidth h(n) satisfies the requirement that nh(n)(3) --> 0, as n --> infinity , it is shown that, for a suitable choice of the kernel, (i(n) - n)/(n h(n)) tends to a positive number, as n --> infinity. It follows that t he deficiency of F-n(t) with respect to (F) over cap(n)(t), i(n) - n, is substantial, and, actually, tends to infinity, as n --> infinity. I n terms of deficiency, the smooth estimate (F) over cap(n)(t) is prefe rable to the empirical distribution function F-n(t).