recent developments in high frequency financial ... - Index of
recent developments in high frequency financial ... - Index of
recent developments in high frequency financial ... - Index of
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Modell<strong>in</strong>g f<strong>in</strong>ancial transaction price movements: a dynamic <strong>in</strong>teger count data model 187<br />
hypothesis <strong>of</strong> Diamond and Verrecchia (1987) who predict a negative correlation<br />
between price changes and the time between transactions. Hence, one would<br />
expect asymmetric effects on the probabilities <strong>of</strong> a certa<strong>in</strong> price reaction. 14<br />
In a similar way, the transaction rate and the transaction volume are <strong>in</strong>troduced<br />
<strong>in</strong>to the model as condition<strong>in</strong>g <strong>in</strong>formation for the size <strong>of</strong> the price changes (2.23):<br />
ln !i ¼ d0Di þ d1Di 1 þ v 0Vi þ v1Vi 1 þ t0Ti þ t1Ti 1 þ i: (3.3)<br />
Note that the conditional distribution <strong>of</strong> the price change Pr Yi ¼ yijZi; F y;z<br />
h i<br />
ð Þ<br />
result<strong>in</strong>g from the specifications (3.2) and (3.3) does not explicitly rest on a<br />
structural theoretical model for the jo<strong>in</strong>t process <strong>of</strong> price changes, volume, and<br />
transaction rates, which would treat each <strong>of</strong> these variables as an endogenous<br />
quantity. Equations (3.2) and (3.3), rather, reflect ad-hoc assumptions with respect<br />
to the distribution <strong>of</strong> the price changes conditional on volume and transaction rate<br />
(as it could result from a jo<strong>in</strong>t distribution <strong>of</strong> these variables). Correspond<strong>in</strong>gly,<br />
the estimated relations cannot be <strong>in</strong>terpreted as structural economic relations.<br />
Nevertheless, the augmented ICH model can serve as an <strong>in</strong>strument for captur<strong>in</strong>g<br />
and quantify<strong>in</strong>g the relationship between important marks <strong>of</strong> the trad<strong>in</strong>g process.<br />
This allows us to shed light on the empirical relevance <strong>of</strong> the theoretical implications<br />
sketched above.<br />
Tables 3 and 4 conta<strong>in</strong> the estimation results for the augmented ICH model with<br />
the transaction rate and trad<strong>in</strong>g volume as additional covariates. For the two submodels,<br />
we have chosen the same order <strong>of</strong> the process that was found to be optimal<br />
for the pure time series specification. In the price direction model (Table 3) both log<br />
trad<strong>in</strong>g volume Vi and the log time between transactions T i have a significantly<br />
positive impact on the log–odds ratios (i.e., the probability that the transaction price<br />
changes <strong>in</strong>creases with the size <strong>of</strong> the transaction volume and the time between<br />
transactions). S<strong>in</strong>ce the probability <strong>of</strong> a nonzero price change can be <strong>in</strong>terpreted as<br />
a specific measure <strong>of</strong> price volatility, it implies that low transaction rates go along<br />
with <strong>high</strong>er price volatility. This provides empirical support for the implications <strong>of</strong><br />
the model proposed by Diamond and Verrecchia (1987), where no transactions<br />
<strong>in</strong>dicate bad news, which contradicts the theoretical implications <strong>of</strong> Easley and<br />
O’Hara (1992) where no transactions <strong>in</strong>dicate lack <strong>of</strong> news <strong>in</strong> the market. Our<br />
f<strong>in</strong>d<strong>in</strong>g that <strong>high</strong> transaction volumes are positively correlated with volatility is<br />
consistent with the implication <strong>of</strong> the model proposed by Easley and O’Hara<br />
(1987), where large volumes correspond to the existence <strong>of</strong> additional news <strong>in</strong> the<br />
market. The effect <strong>of</strong> volume on the probability <strong>of</strong> a price change is partly compensated<br />
for by the subsequent transaction. The effect <strong>of</strong> the transaction time on the<br />
probability <strong>of</strong> a price change is asymmetric <strong>in</strong> the sense that the major reaction for<br />
negative price change occurs immediately, while parts <strong>of</strong> the reaction on log–odds<br />
ratio for a positive price change occurs also with the subsequent transaction. This<br />
<strong>in</strong>terest<strong>in</strong>g reaction pattern holds for JBX and HAL.<br />
F<strong>in</strong>ally, the <strong>in</strong>clusion <strong>of</strong> the microstructure variables greatly improves the value<br />
<strong>of</strong> the Schwarz criterion, but worsens the dynamic properties <strong>of</strong> the model as<br />
<strong>in</strong>dicated by the Q-statistics. Our empirical results for the direction <strong>of</strong> the price<br />
changes are <strong>in</strong> accordance with those put forth by Rydberg and Shephard (2003) for<br />
14 The LR-test clearly rejects the null hypothesis <strong>of</strong> symmetric price reactions.<br />
i 1