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
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
2.1. Bulles r<strong>at</strong>ionnelles multi-dimensionnelles <strong>et</strong> queues épaisses 39<br />
Q UANTITATIVE F INANCE Multi-dimensional r<strong>at</strong>ional bubbles<br />
5. Consequences for r<strong>at</strong>ional expect<strong>at</strong>ion<br />
bubbles<br />
We have seen in section 3.3 from proposition 1 th<strong>at</strong>, as a<br />
result of the no-arbitrage condition, the spectral radius of the<br />
m<strong>at</strong>rix EP[A] = 1<br />
δ Id is gre<strong>at</strong>er than 1. As a consequence, by<br />
applic<strong>at</strong>ion of the converse of proposition 2, this proves th<strong>at</strong><br />
the tail in<strong>de</strong>x κ1 of the distribution of (X) is smaller than 1.<br />
Using the same arguments as in Lux and Sorn<strong>et</strong>te (1999),<br />
th<strong>at</strong> we do not recall here, it can be shown th<strong>at</strong> the distribution<br />
of price differences and price r<strong>et</strong>urns follows, <strong>at</strong> least over<br />
an exten<strong>de</strong>d range of large r<strong>et</strong>urns, a power law distribution<br />
whose exponent remains lower than 1. This result generalizes<br />
to arbitrary d-dimensional processes the result of Lux and<br />
Sorn<strong>et</strong>te (1999). As a consequence, d-dimensional r<strong>at</strong>ional<br />
expect<strong>at</strong>ion bubbles linking several ass<strong>et</strong>s suffer from the<br />
same discrepancy compared to empirical d<strong>at</strong>a as the onedimensional<br />
bubbles. It therefore appears th<strong>at</strong> accounting for<br />
possible <strong>de</strong>pen<strong>de</strong>nces b<strong>et</strong>ween bubbles is not sufficient to cure<br />
the Blanchard and W<strong>at</strong>son mo<strong>de</strong>l: a linear multi-dimensional<br />
bubble dynamics such as (16) is hardly reconcilable with some<br />
of the most fundamental stylized facts of financial d<strong>at</strong>a <strong>at</strong> a very<br />
elementary level.<br />
This result does not rely on the diagonal property of the<br />
m<strong>at</strong>rices E[At] but only on the value of its spectral radius<br />
imposed by the no-arbitrage condition. In other words, the fact<br />
th<strong>at</strong> the introduction of <strong>de</strong>pen<strong>de</strong>nces b<strong>et</strong>ween ass<strong>et</strong> bubbles is<br />
not sufficient to cure the mo<strong>de</strong>l can be traced to the constraint<br />
introduced by the no-arbitrage condition, which imposes th<strong>at</strong><br />
the averages of the off-diagonal terms of the m<strong>at</strong>rices At<br />
vanish. As we indic<strong>at</strong>ed before, this implies zero correl<strong>at</strong>ions<br />
(but not the absence of <strong>de</strong>pen<strong>de</strong>nce) b<strong>et</strong>ween ass<strong>et</strong> bubbles.<br />
It thus seems th<strong>at</strong> the multi-dimensional generaliz<strong>at</strong>ion is<br />
constrained so much by the no-arbitrage condition th<strong>at</strong> the<br />
multi-dimensional bubble mo<strong>de</strong>l almost reduces to an average<br />
one-dimensional mo<strong>de</strong>l. With this insight, our present result<br />
generalizing th<strong>at</strong> of Lux and Sorn<strong>et</strong>te (1999) is n<strong>at</strong>ural.<br />
To our knowledge, there are two possible remedies. The<br />
first one is based on the r<strong>at</strong>ional bubble and crash mo<strong>de</strong>l<br />
of Johansen <strong>et</strong> al (1999, 2000) which abandons the linear<br />
stochastic dynamics (8) in favour of an essentially arbitrary<br />
and nonlinear dynamics controlled by a crash hazard r<strong>at</strong>e.<br />
A jump process for crashes is ad<strong>de</strong>d to the process, with a<br />
crash hazard r<strong>at</strong>e evolving with time such th<strong>at</strong> the r<strong>at</strong>ional<br />
expect<strong>at</strong>ion condition is ensured. This mo<strong>de</strong>l is squarely<br />
based on the r<strong>at</strong>ional expect<strong>at</strong>ion framework and shows th<strong>at</strong><br />
changing the dynamics of the Blanchard and W<strong>at</strong>son mo<strong>de</strong>l<br />
allows s<strong>at</strong>isfying results to be obtained, as the corresponding<br />
r<strong>et</strong>urn distributions can be ma<strong>de</strong> to exhibit reasonable f<strong>at</strong> tails.<br />
The second solution (Sorn<strong>et</strong>te 2001) requires the existence<br />
of an average exponential growth of the fundamental price <strong>at</strong><br />
some r<strong>et</strong>urn r<strong>at</strong>e rf > 0 larger than the discount r<strong>at</strong>e. With<br />
the condition th<strong>at</strong> the price fluctu<strong>at</strong>ions associ<strong>at</strong>ed with bubbles<br />
must on average grow with the mean mark<strong>et</strong> r<strong>et</strong>urn rf , it can be<br />
shown th<strong>at</strong> the exponent of the power law tail of the r<strong>et</strong>urns is<br />
no more boun<strong>de</strong>d by 1 as soon as rf is larger than the discount<br />
r<strong>at</strong>e r and can take essentially arbitrary values. This second<br />
approach amounts to abandoning the r<strong>at</strong>ional pricing theory<br />
(6) with (4) and keeping only the no-arbitrage constraint (7)<br />
on the bubble component. This hypothesis may be true in the<br />
case of a firm, som<strong>et</strong>imes in the case of an industry (railways in<br />
the 19th century, oil and computer in the 20th for instance), but<br />
is hard to <strong>de</strong>fend in the case of an economy as a whole. The<br />
real long run interest r<strong>at</strong>e in the US is approxim<strong>at</strong>ely 3.5%,<br />
and the real r<strong>at</strong>e of growth of profits since World War II is<br />
about 2.1%. Thus, for the economy as a whole, the discounted<br />
sum is always finite. It would be interesting to investig<strong>at</strong>e the<br />
interplay b<strong>et</strong>ween inter-<strong>de</strong>pen<strong>de</strong>nce b<strong>et</strong>ween several bubbles<br />
and this exponential growth mo<strong>de</strong>l.<br />
As a final positive remark balancing our neg<strong>at</strong>ive result,<br />
we stress th<strong>at</strong> the stochastic multiplic<strong>at</strong>ive multi-ass<strong>et</strong> r<strong>at</strong>ional<br />
bubble mo<strong>de</strong>l presented here provi<strong>de</strong>s a n<strong>at</strong>ural mechanism for<br />
the existence of a ‘universal’ tail exponent µ across different<br />
mark<strong>et</strong>s.<br />
Acknowledgments<br />
We acknowledge helpful discussions and exchanges with<br />
J P Laurent, T Lux, V Pisarenko and M Taqqu. We also<br />
thank T Mikosch for providing access to Le Page (1983) and<br />
V Pisarenko for a critical reading of the manuscript. Any<br />
remaining error is ours.<br />
Appendix A. Proof of proposition 1 on<br />
the no-arbitrage condition<br />
L<strong>et</strong> t be the value <strong>at</strong> time t of any self-financing portfolio:<br />
t = W ′<br />
t Xt, (A.1)<br />
where W ′<br />
t = (W1,...,Wd) is the vector whose components<br />
are the weights of the different ass<strong>et</strong>s and the prime <strong>de</strong>notes<br />
the transpose. The no-free-lunch condition reads:<br />
t = δ · EQ[t+1|Ft] ∀{t}t0. (A.2)<br />
Therefore, for all self-financing str<strong>at</strong>egies (Wt), wehave<br />
W ′<br />
<br />
t+1 EQ[A] − 1<br />
δ Id<br />
<br />
Xt = 0 ∀ Xt ∈ R d , (A.3)<br />
where we have used the fact th<strong>at</strong> (Wt+1) is (Ft)-measurable<br />
and th<strong>at</strong> the sequence of m<strong>at</strong>rices {At} is i.i.d.<br />
The str<strong>at</strong>egy W ′<br />
t = (0,...,0, 1, 0,...,0) (1 in ith<br />
position), for all t, is self-financing and implies<br />
(ai1,ai2,...,aii − 1<br />
δ ,...,aid)<br />
·(X (1)<br />
t ,X (2)<br />
t ,...,X (i)<br />
t ,...,X (d)<br />
t ) ′ = 0, (A.4)<br />
for all Xt ∈ Rd . We have called aij the (i, j)th coefficient of<br />
the m<strong>at</strong>rix EQ[A]. As a consequence,<br />
(ai1,ai2,...,aii − 1<br />
δ ,...,aid) = 0 ∀i, (A.5)<br />
and<br />
EQ[A] = 1<br />
δ Id. (A.6)<br />
539