20.11.2012 Views

recent developments in high frequency financial ... - Index of

recent developments in high frequency financial ... - Index of

recent developments in high frequency financial ... - Index of

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.

300 V. Voev<br />

where �kl(t) is the (k, l) element <strong>of</strong> the �(t) process. With this notation, we will<br />

now <strong>in</strong>terpret the ‘true’ covariance matrix as<br />

� t+1<br />

�t+1 =<br />

t<br />

�(u)du. (18)<br />

Thus the covariance matrix at time t + 1 is the <strong>in</strong>crement <strong>of</strong> the <strong>in</strong>tegrated<br />

covariance matrix <strong>of</strong> the cont<strong>in</strong>uous local mart<strong>in</strong>gale from time t to time t + 1.<br />

The realized covariance as def<strong>in</strong>ed <strong>in</strong> expression (1) consistently estimates �t+1 as<br />

given <strong>in</strong> Eq. (18). Furthermore, Barndorff-Nielsen and Shephard (2004) show that<br />

under a set <strong>of</strong> regularity conditions the realized covariation matrix follows asymptotically,<br />

as M →∞, the normal law with N × N matrix <strong>of</strong> means � t+1<br />

t �(u)du.<br />

The asymptotic covariance <strong>of</strong><br />

�<br />

�<br />

√ M<br />

� RC<br />

t+1 −<br />

� t+1<br />

t<br />

�(u)du<br />

is �t+1, aN 2 × N 2 array with elements<br />

�� t+1<br />

�<br />

�t+1 = {�kk ′(u)�ll ′(u) + �kl ′(u)�lk ′(u)} du<br />

t<br />

k,k ′ ,l,l ′ =1,...,N<br />

Of course, this matrix is s<strong>in</strong>gular due to the equality <strong>of</strong> the covariances <strong>in</strong> the<br />

<strong>in</strong>tegrated covariance matrix. This can easily be avoided by consider<strong>in</strong>g only its<br />

unique lower triangular elements, but for our purposes it will be more convenient<br />

to work with the full matrix. The result above is not useful for <strong>in</strong>ference, s<strong>in</strong>ce the<br />

matrix �t+1 is not known. Barndorff-Nielsen and Shephard (2004) show that a<br />

consistent, positive semi-def<strong>in</strong>ite estimator is given by a random N 2 ×N 2 matrix:<br />

Ht+1 =<br />

M�<br />

j=1<br />

xj,t+1x ′ j,t+1<br />

− 1<br />

2<br />

M−1 �<br />

j=1<br />

�<br />

xj,t+1x ′ j+1,t+1 + xj+1,t+1x ′ �<br />

j,t+1 , (19)<br />

�<br />

where xj,t+1 = vec r j r t+ M<br />

′<br />

t+ j<br />

�<br />

M<br />

and the vec operator stacks the columns <strong>of</strong> a<br />

matrix <strong>in</strong>to a vector. It holds that MHt+1<br />

p<br />

→ �t+1 with M →∞.<br />

With the knowledge <strong>of</strong> this matrix, we can comb<strong>in</strong>e the asymptotic results for<br />

the realized covariance, with the result <strong>in</strong> Eq. (7) to compute the estimates for πt,<br />

ρt and νt.<br />

For the equicorrelated matrix F we have that4 �<br />

fij =¯r σ (RC)<br />

ii σ (RC)<br />

jj , where ¯r<br />

is the average value <strong>of</strong> all pairwise correlations, implied by the realized covariance<br />

matrix, and σ (RC)<br />

is the (i,j) element <strong>of</strong> the realized covariance matrix. Thus �,<br />

ij<br />

the population equicorrelated matrix, has a typical element φij =¯̺ √ σiiσjj , where<br />

σij is the (i,j) <strong>of</strong> the true covariance matrix � and ¯̺ is the average correlation<br />

4 In the follow<strong>in</strong>g exposition, the time <strong>in</strong>dex is suppressed for notational convenience.<br />

.

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

Saved successfully!

Ooh no, something went wrong!