recent developments in high frequency financial ... - Index of
recent developments in high frequency financial ... - Index of
recent developments in high frequency financial ... - Index of
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
168<br />
Keywords F<strong>in</strong>ancial transaction prices . Autoregressive conditional mult<strong>in</strong>omial<br />
model . GLARMA . Count data . Market microstructure effects<br />
JEL Classification C22 . C25 . G10<br />
1 Introduction<br />
R. Liesenfeld et al.<br />
F<strong>in</strong>ancial transaction data, <strong>of</strong>ten called ultra <strong>high</strong> <strong>frequency</strong> data, is marked by two<br />
ma<strong>in</strong> features: the irregularity <strong>of</strong> time <strong>in</strong>tervals and the discreteness <strong>of</strong> price<br />
changes. Based on the sem<strong>in</strong>al work by Russell and Engle (1998) and Engle<br />
(2000), a large body <strong>of</strong> studies has been centered around the further development<br />
<strong>of</strong> autoregressive conditional duration (ACD) models <strong>in</strong> order to characterize the<br />
transaction <strong>in</strong>tensities. This paper is concerned with appropriately modell<strong>in</strong>g the<br />
discreteness <strong>of</strong> the price process at the transaction level with<strong>in</strong> a count data<br />
framework. While quantal response approaches seem to be more suitable <strong>in</strong> modell<strong>in</strong>g<br />
price changes on the transaction level if the outcome space consists only <strong>of</strong><br />
a few possible outcomes, the approach presented here is particularly designed for<br />
shares where the possible outcome space for the price changes is a larger range<br />
<strong>of</strong> <strong>in</strong>teger values. This holds for most <strong>of</strong> the assets traded on European asset<br />
markets. But our approach is also attractive for the analysis <strong>of</strong> transaction price<br />
movements at more liquid markets such as the NYSE. With the decimalization on<br />
the 29th January 2001, the m<strong>in</strong>imum tick size at the NYSE was reduced from 1/16th<br />
<strong>of</strong> a US-Dollar to one cent for stocks sell<strong>in</strong>g at prices greater than or equal to US<br />
$1. This leads to larger price jumps <strong>in</strong> ticks (the smallest possible price change) and<br />
a larger range <strong>of</strong> observable discrete trade-by-trade price jumps. For example, <strong>in</strong><br />
the month before decimalization for the IBM stocks, 95% <strong>of</strong> all price changes at the<br />
tick level were <strong>in</strong> a range <strong>of</strong> ±3 ticks, while this range changed to ±10 ticks <strong>in</strong> the<br />
month thereafter.<br />
S<strong>in</strong>ce transaction price changes are quoted as multiples <strong>of</strong> a smallest divisor, the<br />
use <strong>of</strong> cont<strong>in</strong>uous distributions to characterize price changes is far from be<strong>in</strong>g<br />
appropriate <strong>in</strong> particular for markets with <strong>high</strong> transaction <strong>in</strong>tensities. Accord<strong>in</strong>gly,<br />
Hausman et al. (1992) proposed an ordered probit model with conditional heteroscedasticity<br />
to analyze stock price movements at the NYSE. The same approach is<br />
used by Bollerslev and Melv<strong>in</strong> (1994) to model the bid-ask spread at FX-markets.<br />
Contrary to the older round<strong>in</strong>g approaches by Ball (1988), Cho and Frees (1988)<br />
and Harris (1990), condition<strong>in</strong>g <strong>in</strong>formation can be <strong>in</strong>corporated <strong>in</strong> the ordered<br />
response models quite easily. A drawback <strong>of</strong> the ordered probit approach is that<br />
the parameters result from a threshold cross<strong>in</strong>g latent variable model, where the<br />
underly<strong>in</strong>g cont<strong>in</strong>uous latent dependent variable has to be given some more or<br />
less arbitrary economic <strong>in</strong>terpretation (e.g., latent price pressure). Moreover, s<strong>in</strong>ce<br />
the parameters are only identified up to a factor <strong>of</strong> proportionality, the estimates<br />
<strong>of</strong> the moments <strong>of</strong> the latent price variable are only identifiable us<strong>in</strong>g additional<br />
identify<strong>in</strong>g restrictions.<br />
An alternative to the ordered response models is the autoregressive conditional<br />
mult<strong>in</strong>omial (ACM) model proposed by Russel and Engle (2002). Similar to the<br />
ordered response models, this approach also rests on the assumption that the distribution<br />
<strong>of</strong> observed transaction price changes is discrete with a f<strong>in</strong>ite number <strong>of</strong><br />
outcomes. A drawback <strong>of</strong> the ACM model is the necessity that all potential out-