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

Create successful ePaper yourself

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

140<br />

The <strong>in</strong>novation term ei is computed from the <strong>in</strong>tegrated <strong>in</strong>tensity function<br />

associated with the most <strong>recent</strong>ly observed process. Hence,<br />

0<br />

1<br />

ei ¼ XK<br />

k¼1<br />

@1<br />

Z k t<br />

Nk ðÞ ti tk Nk ðÞ ti 1<br />

k<br />

ðs; F sÞds<br />

Ay k i<br />

: (5)<br />

Under fairly weak assumptions, the <strong>in</strong>tegrated <strong>in</strong>tensity function corresponds to an<br />

i.i.d. standard exponential variate. 7 Therefore, e i is a random mixture <strong>of</strong><br />

exponential variates. For this reason, weak stationarity <strong>of</strong> the model depends on<br />

the eigenvalues <strong>of</strong> the matrix B. If these lie <strong>in</strong>side the unit circle, the process e i is<br />

weakly stationary.<br />

S<strong>in</strong>ce the <strong>in</strong>tensity function has left-cont<strong>in</strong>uous sample paths, Ψ i also has to be a<br />

left-cont<strong>in</strong>uous function and predeterm<strong>in</strong>ed, at least <strong>in</strong>stantaneously before the<br />

arrival <strong>of</strong> a new event. Therefore, Ψi is known <strong>in</strong>stantaneously after the occurrence<br />

<strong>of</strong> ti−1 and does not change until ti. Then kðt;F tÞ changes between ti−1 and ti only<br />

k k<br />

as a determ<strong>in</strong>istic function <strong>of</strong> time accord<strong>in</strong>g to the functions l0 (t) and s (t).<br />

The basel<strong>in</strong>e <strong>in</strong>tensity function l 0 k (t) is specified <strong>in</strong> terms <strong>of</strong> the backward re-<br />

currence times xkðÞ¼t t t k<br />

M k , k =1,...,K, <strong>of</strong> all processes and may be specified<br />

ðÞ t<br />

us<strong>in</strong>g a Weibull-type parameterization depend<strong>in</strong>g on the parameters ω k k<br />

and pr ,<br />

k<br />

0 t<br />

YK<br />

ðÞ¼exp !k<br />

r¼1<br />

x r ðÞ t<br />

pkr 1 ; p k r > 0 : (6)<br />

Moreover, the seasonality functions s k (t) are parameterized as l<strong>in</strong>ear spl<strong>in</strong>e<br />

functions given by 8<br />

s k ðÞ¼1 t þ XS<br />

j¼1<br />

k<br />

j t j 1<br />

ft> jg<br />

; (7)<br />

where τj, j=1,..., S, denote the S nodes with<strong>in</strong> a trad<strong>in</strong>g day and νj k the correspond<strong>in</strong>g<br />

parameters.<br />

As a result, by denot<strong>in</strong>g W as the data matrix consist<strong>in</strong>g <strong>of</strong> all po<strong>in</strong>ts and<br />

explanatory variables and denot<strong>in</strong>g θ as the parameter vector <strong>of</strong> the model, the loglikelihood<br />

function <strong>of</strong> the multivariate ACI model is given by<br />

ln LðW; Þ ¼ XK<br />

X n<br />

k¼1 i¼1<br />

Z ti<br />

ti 1<br />

k<br />

ðs; F sÞds<br />

þ y k i ln k ti;F ti ; (8)<br />

where t 0=0 and n denotes the number <strong>of</strong> po<strong>in</strong>ts <strong>of</strong> the pooled process. Under correct<br />

specification <strong>of</strong> the model, the result<strong>in</strong>g k-type ACI residuals<br />

^" k i ¼<br />

Z k ti t k i 1<br />

^ k ðs;FsÞds A. D. Hall, N. Hautsch<br />

7 See Brémaud (1981) and Bowsher (2002).<br />

8 In order to identify the constant ω k , s k (t) is set to one at the beg<strong>in</strong>n<strong>in</strong>g <strong>of</strong> a trad<strong>in</strong>g day.

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

Saved successfully!

Ooh no, something went wrong!