Les deux définitions sont proches, mais pas exactement les mêmes. Une différence réside dans la nécessité pour le taux de survie d'avoir une limite.
Pour la plupart de cette réponse, j'ignorerai les critères pour que la distribution soit continue, symétrique et de variance finie, car ceux-ci sont faciles à accomplir une fois que nous avons trouvé une distribution à queue lourde à variance finie qui n'est pas à queue longue.
Une distribution est à queue lourde lorsque pour tout t > 0 ,Ft>0
∫RetxdF(x)=∞.(1)
Une distribution avec fonction de survie est à longue queue lorsquegF= 1 - F
limx→∞GF(x+1)GF(x)=1.(2)
Les distributions à longue queue sont lourdes. De plus, comme n'est pas croissant, la limite du rapport ( 2 ) ne peut pas dépasser 1 . S'il existe et est inférieur à 1 , alors G diminue exponentiellement - et cela permettra l'intégrale ( 1 )G(2)11G(1) de converger.
La seule façon de montrer une distribution à queue lourde qui n'est pas à longue queue, alors, est de modifier une distribution à longue queue pour que continue de tenir tandis que ( 2 ) est violé. Il est facile de bousiller une limite: changez-la dans une infinité d'endroits qui divergent à l'infini. Cela prendra cependant du temps avec F , qui doit rester croissant et cadlag. Une façon consiste à introduire des sauts vers le haut dans F , ce qui fera sauter G vers le bas, abaissant le rapport G F ( x + 1 ) / G F ( x )(1)(2)FFGGF(x+1)/GF(x) . A cette fin, définissons une transformation Tu qui transforme en une autre fonction de distribution valide tout en créant un saut soudain à la valeur u , disons un saut à mi-chemin de F ( u ) à 1 :FuF(u)1
Tu[F](x)={F(x)12(1−F(x))+F(x)u<xu≥x
Cela ne modifie aucune propriété de base de : TFTu[F] is still a distribution function.
The effect on GF is to make it drop by a factor of 1/2 at u. Therefore, since G is non-decreasing, then whenever u−1≤x<u,
GTu[F](x+1)GTu[F](x)≤12.
If we pick an increasing and diverging sequence of ui, i=1,2,…, and apply each Tui in succession, it determines a sequence of distributions Fi with F0=F and
Fi+1=Tui[Fi]
for i≥1. After the ith step, Fi(x),Fi+1(x),… all remain the same for x<ui. Consequently the sequence of Fi(x) is a nondecreasing, bounded, pointwise sequence of distribution functions, implying its limit
F∞=limi→∞Fi
is a distribution function. By construction, it is not long-tailed because there are infinitely many points at which its survival ratio GF∞(x+1)/GF∞(x)) drops to 1/2 or below, showing it cannot have 1 as a limit.
This plot shows a survival function G(x)=x−1/5 that has been cut down in this manner at points u1≈12.9,u2≈40.5,u3≈101.6,…. Note the logarithmic vertical axis.
The hope is to be able to choose (ui) so that F∞ remains heavy-tailed. We know, because F is heavy-tailed, that there are numbers 0=u0<u1<u2<⋯<un⋯ for which
∫uiui−1ex/idF(x)≥2i−1
for every i≥1. The reason for the 2i−1 on the right is that the probabilities assigned by F to values up to ui have been successively cut in half i−1 times. That procedure, when dF(x) is replaced by dFj(x) for any j≥i, will reduce 2i−1 to 1, but no lower.
This is a plot of xf(x) for densities f corresponding to the previous survival function and its "cut down" version. The areas under this curve contribute to the expectation. The area from 1 to u1 is 1; the area from u1 to u2 is 2, which when cut down (to the lower blue portion) becomes an area of 1; the area from u2 to u3 is 4, which when cut down becomes an area of 1, and so on. Thus, the area under each successive "stair step" to the right is 1.
Let us pick such a sequence (ui) to define F∞. We can check that it remains heavy-tailed by picking t=1/n for some whole number n and applying the construction:
∫RetxdF∞(x)=∫Rex/ndF∞(x)=∑i=1∞∫uiui−1ex/ndF∞(x)≥∑i=n+1∞∫uiui−1ex/ndF∞(x)≥∑i=n+1∞∫uiui−1ex/idF∞(x)=∑i=n+1∞∫uiui−1ex/idFi(x)≥∑i=n+1∞1,
which still diverges. Since t is arbitrarily small, this demonstrates that F∞ remains heavy-tailed, even though its long-tailed property has been destroyed.
This is a plot of the survival ratio G(x+1)/G(x) for the cut down distribution. Like the ratio of the original G, it tends toward an upper accumulation value of 1--but for unit-width intervals terminating at the ui, the ratio suddenly drops to only half of what it originally was. These drops, although becoming less and less frequent as x increases, occur infinitely often and therefore prevent the ratio from approaching 1 in the limit.
If you would like a continuous, symmetric, zero-mean, unit-variance example, begin with a finite-variance long-tailed distribution. F(x)=1−x−p (for x>0) will do, provided p>1; so would a Student t distribution for any degrees of freedom exceeding 2. The moments of F∞ cannot exceed those of F, whence it too has finite variance. "Mollify" it via convolution with a nice smooth distribution, such as a Gaussian: this will make it continuous but will not destroy its heavy tail (obviously) nor the absence of a long tail (not quite as obvious, but it becomes obvious if you change the Gaussian to, say, a Beta distribution whose support is compact).
Symmetrize the result--which I will still call F∞--by defining
Fs(x)=12(1+sgn(x)F∞(|x|))
for all x∈R. Its variance will remain finite, so it can be standardized to the desired distribution.