25.06.2013 Views

statistique, théorie et gestion de portefeuille - Docs at ISFA

statistique, théorie et gestion de portefeuille - Docs at ISFA

statistique, théorie et gestion de portefeuille - Docs at ISFA

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

where E0[·] <strong>de</strong>notes the expect<strong>at</strong>ion with respect to the power-law distribution f0. We have<br />

<br />

x<br />

c E0<br />

u<br />

<br />

E0 ln<br />

=<br />

b<br />

, and c < b,<br />

b − c<br />

(121)<br />

x<br />

<br />

u<br />

= 1<br />

<br />

x<br />

c E0 ln<br />

u<br />

,<br />

b<br />

(122)<br />

x<br />

<br />

u<br />

=<br />

b<br />

.<br />

(b − c) 2 (123)<br />

Thus, we easily obtain th<strong>at</strong> the unique solution of (119) is c = 0, and equ<strong>at</strong>ion (120) does not make sense<br />

any more. So, un<strong>de</strong>r H0, ĉ goes to zero and dˆ is not well <strong>de</strong>fined. Thus, Wald test cannot be performed un<strong>de</strong>r<br />

such a null hypothesis. We must find another way to test (SE) against H0.<br />

In this goal, we remark th<strong>at</strong> the quantity<br />

ˆηT = ĉ<br />

<br />

ĉ<br />

u<br />

+ 1<br />

dˆ<br />

101<br />

(124)<br />

is still well <strong>de</strong>fined. Using (120), it is easy to show th<strong>at</strong>, as T goes to infinity, this quantity goes to b,<br />

wh<strong>at</strong>ever ĉ being positive or equal to zero. For positive c, this is obvious from (120) and (121), while for<br />

c = 0, expanding (118) around ĉ = 0 yields<br />

ĉ<br />

ĉ u<br />

dˆ<br />

=<br />

<br />

1<br />

T<br />

T<br />

∑<br />

i=1<br />

ln xi<br />

u<br />

−1<br />

(125)<br />

= ˆb → b. (126)<br />

In or<strong>de</strong>r to test the <strong>de</strong>scriptive power of the (SE) mo<strong>de</strong>l against the null hypothesis th<strong>at</strong> the true mo<strong>de</strong>l is the<br />

Par<strong>et</strong>o mo<strong>de</strong>l, we can consi<strong>de</strong>r the st<strong>at</strong>istic<br />

<br />

ˆηT<br />

ζT = T − 1 , (127)<br />

ˆb<br />

which asymptoticaly follows a χ 2 -distribution with one <strong>de</strong>gree of freedom. In<strong>de</strong>ed expanding the quantity<br />

(xi/u) ĉ in power series around c = 0 gives<br />

which allows us to g<strong>et</strong><br />

where<br />

<br />

xi<br />

ĉ<br />

∼= 1 + ĉ · log(<br />

u<br />

xi<br />

u<br />

<br />

) + ĉ2<br />

2 · log2 <br />

xi<br />

+<br />

u<br />

ĉ3<br />

6 · log3 <br />

xi<br />

+ ··· , as ĉ → 0, (128)<br />

u<br />

1<br />

T ∑ <br />

xi<br />

ĉ<br />

∼= 1 + ĉ · S1 +<br />

u<br />

ĉ2<br />

2 · S2 + ĉ3<br />

3 S3, (129)<br />

1<br />

T ∑ <br />

xi<br />

c <br />

xi<br />

log ∼= S1 + ĉ · S2 +<br />

u u<br />

ĉ2<br />

2 S3, (130)<br />

S1 = 1<br />

T<br />

S2 = 1<br />

T<br />

S3 = 1<br />

T<br />

T<br />

∑<br />

i=1<br />

T<br />

∑<br />

i=1<br />

T<br />

∑<br />

i=1<br />

log<br />

xi<br />

u<br />

log 2 xi<br />

u<br />

log 3 xi<br />

u<br />

37<br />

<br />

, (131)<br />

<br />

, (132)<br />

<br />

. (133)<br />

(134)

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!