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.

174<br />

specify the parametric form <strong>of</strong> the p.d.f. for the price direction and the absolute size<br />

<strong>of</strong> the price changes.<br />

2.1 Dynamics <strong>of</strong> the price direction<br />

The parametric model for the direction <strong>of</strong> the transaction price change Di=j,(j=−1, 0, 1) is taken from the class <strong>of</strong> logistic ACM (autoregressive conditional mult<strong>in</strong>omial)<br />

models suggested by Russel and Engle (2002). In order to relate the<br />

probability ji ¼ Pr ½Di¼jjF i 1Š<br />

for the occurrence <strong>of</strong> price direction j to subsets<br />

<strong>of</strong> F i 1 (and further explanatory variables), we use a logistic l<strong>in</strong>k function. This<br />

leads to a mult<strong>in</strong>omial logit model <strong>of</strong> the form:<br />

ji ¼<br />

exp ji<br />

P 1<br />

j¼ 1 exp ji<br />

; j ¼ 1; 0; 1; (2.10)<br />

where Λji represents a function <strong>of</strong> some subset <strong>of</strong> F i 1 to be specified below. As a<br />

normaliz<strong>in</strong>g constra<strong>in</strong>t, we use Λ0i=0, ∀i.<br />

Due to the observed dynamic behavior <strong>of</strong> the transaction price changes<br />

associated with the bid-ask bounce or the volatility cluster<strong>in</strong>g, one can expect that<br />

the process <strong>of</strong> the price direction also exhibits serial dependence. In order to shed<br />

light on this serial dependence, which has to be taken <strong>in</strong>to account when modell<strong>in</strong>g<br />

the conditional distribution <strong>of</strong> the price direction, we def<strong>in</strong>e the follow<strong>in</strong>g state<br />

vector<br />

xi ¼ ðx1i; x1iÞ<br />

0 ð1; 0Þ<br />

¼<br />

0<br />

if Yi < 0<br />

ð0; 0Þ<br />

0<br />

if Yi ¼ 0<br />

ð0; 1Þ<br />

0<br />

8<br />

<<br />

(2.11)<br />

:<br />

if Yi > 0;<br />

and consider its correspond<strong>in</strong>g sample autocorrelation matrix. For a lag length ‘,<br />

this matrix is given by:<br />

with<br />

ðÞ¼D ‘<br />

1<br />

1<br />

ðÞ¼ ‘<br />

n ‘ 1<br />

ðÞD ‘<br />

1 ; ‘ ¼ 1; 2; ...; (2.12)<br />

X n<br />

ðxixÞ xi ‘ x<br />

i¼‘þ1<br />

ð Þ 0 :<br />

R. Liesenfeld et al.<br />

D denotes a diagonal matrix conta<strong>in</strong><strong>in</strong>g the standard deviations <strong>of</strong> x −1i and x 1i.<br />

Figure 4 below depicts the cross-correlation function <strong>of</strong> up to 30 lagged<br />

transactions. The significant, but not very large, first order cross-correlations provide<br />

empirical support for the existence <strong>of</strong> a bid-ask bounce: The probability <strong>of</strong> a<br />

price reduction is significantly (HAL) positively correlated with the price <strong>in</strong>crease<br />

<strong>in</strong> the previous period (upper right panel), and also significant (both stocks), a price<br />

<strong>in</strong>crease is more likely if a negative price change is observed for the previous<br />

transaction (lower left panel). The cross-correlation effects turn out to be asymmetric,<br />

<strong>in</strong> the sense that the correlation <strong>of</strong> a negative price change with a previous<br />

positive one is smaller than the effect vice versa. For HAL we observe negative first

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

Saved successfully!

Ooh no, something went wrong!