02.12.2012 Views

Applications of state space models in finance

Applications of state space models in finance

Applications of state space models in finance

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

76 5 Conditional heteroskedasticity <strong>models</strong><br />

This section summarizes some <strong>of</strong> the most important multivariate conditional heteroskedasticity<br />

<strong>models</strong>. For those topics and the many extensions that are not covered<br />

here, the reader is referred to Bauwens et al. (2003) and Asai et al. (2006) who provide<br />

surveys <strong>of</strong> multivariate GARCH and multivariate SV <strong>models</strong>, respectively.<br />

5.3.1 Multivariate GARCH<br />

A general multivariate GARCH model for an N-dimensional process ɛt|Ωt−1 is given by<br />

ɛt = ztH 1/2<br />

t , (5.45)<br />

where zt is an N-dimensional IID process with zero mean and the identity matrix<br />

IN as covariance matrix. These properties <strong>of</strong> zt together with Equation (5.45) imply<br />

that E(ɛt|Ωt−1) = 0 and E(ɛtɛ ′ t|Ωt−1) = Ht. For illustrative purposes, only the case<br />

with N = 2 will be considered <strong>in</strong> the follow<strong>in</strong>g. In the bivariate case, the conditional<br />

covariance matrix is given by<br />

Ht =<br />

� h11,t h12,t<br />

h21,t h22,t<br />

�<br />

, (5.46)<br />

where Ht depends on lagged errors ɛt−1 and on lagged conditional covariance matrices<br />

Ht−1. The most <strong>in</strong>fluential parameterizations <strong>of</strong> H t can be summarized as follows.<br />

5.3.1.1 The vech model<br />

The most general representation <strong>of</strong> H t−1 is the vech model as proposed by Bollerslev<br />

et al. (1988). By employ<strong>in</strong>g the vech( ) operator, which vertically stacks the matrix<br />

elements on or below the pr<strong>in</strong>cipal diagonal and thus transforms an N × N matrix <strong>in</strong>to<br />

an N(N + 1)/2 × 1 vector, all non-redundant elements <strong>of</strong> H t are stacked <strong>in</strong>to a column<br />

vector:<br />

vech(Ht) = ω ∗ + Γ ∗ vech(ɛt−1ɛ ′ t−1 ) + ∆∗ vech(Ht−1), (5.47)<br />

where ω ∗ = vech(Ω) is a N(N + 1)/2 × 1 parameter vector and Γ ∗ and ∆ ∗ are N(N +<br />

1)/2 × N(N + 1)/2 matrices. In the bivariate case, Equation (5.47) becomes<br />

⎡<br />

⎣<br />

h11,t<br />

h21,t<br />

h22,t<br />

⎤<br />

⎡<br />

⎦ = ⎣<br />

ω ∗ 11<br />

ω ∗ 21<br />

ω ∗ 22<br />

⎤<br />

⎡<br />

⎦ + ⎣<br />

⎡<br />

+ ⎣<br />

γ ∗ 11 γ ∗ 12 γ ∗ 13<br />

γ ∗ 21 γ ∗ 22 γ ∗ 23<br />

γ ∗ 31 γ ∗ 32 γ ∗ 33<br />

δ ∗ 11 δ ∗ 12 δ ∗ 13<br />

δ ∗ 21 δ ∗ 22 δ ∗ 23<br />

δ ∗ 31 δ ∗ 32 δ ∗ 33<br />

⎤ ⎡<br />

⎦ ⎣<br />

⎤ ⎡<br />

⎦ ⎣<br />

ɛ 2 1,t−1<br />

ɛ2,t−1ɛ1,t−1<br />

ɛ2 2,t−1<br />

h11,t−1<br />

h21,t−1<br />

h22,t−1<br />

⎤<br />

⎤<br />

⎦<br />

⎦ . (5.48)<br />

Despite its flexibility, the vech model has two major drawbacks: to guarantee positive<br />

def<strong>in</strong>iteness <strong>of</strong> H t it is necessary to impose further constra<strong>in</strong>ts on Γ ∗ and ∆ ∗ ; see<br />

Engle and Kroner (1995) for a discussion. Besides, overall N(N + 1)/2 + N 2 (N + 1) 2 /2<br />

parameters have to be estimated. As this number grows at a polynomial rate with<br />

<strong>in</strong>creas<strong>in</strong>g N, estimation <strong>of</strong> this general model may become quite cumbersome without<br />

further restrictions.

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

Saved successfully!

Ooh no, something went wrong!