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.

Order aggressiveness and order book dynamics 135<br />

addresses the question <strong>of</strong> where and when it is likely that an order will be placed<br />

given the current state <strong>of</strong> the market.<br />

In order to reduce the impact <strong>of</strong> noise <strong>in</strong> the data and to allow for a better<br />

identification <strong>of</strong> systematic relationships, we explicitly focus on most aggressive<br />

market orders, limit orders and cancellations. On top <strong>of</strong> the common classification<br />

scheme proposed by Biais et al. (1995), we select only those orders whose volumes<br />

are substantially larger than the average order volume. 3 Accord<strong>in</strong>g to previous<br />

studies <strong>in</strong> this field, these are the most aggressive and <strong>in</strong>terest<strong>in</strong>g orders. Note that<br />

this classification scheme also applies to cancellations, and we def<strong>in</strong>e a<br />

cancellation as aggressive whenever a large volume is cancelled. In this sense,<br />

our approach can be seen as an extension <strong>of</strong> the study by Coppejans and Domowitz<br />

(2002) who also focus on the arrival rate <strong>of</strong> trades, limit orders and cancellations.<br />

However, they do not explicitly study <strong>high</strong> volume orders but consider all<br />

<strong>in</strong>com<strong>in</strong>g orders. Moreover, as they analyze the <strong>in</strong>dividual processes separately<br />

us<strong>in</strong>g a generalized version <strong>of</strong> Engle and Russell’s (1998) (univariate) autoregressive<br />

conditional duration (ACD) model, their framework does not allow for<br />

any multivariate <strong>in</strong>terdependencies between the <strong>in</strong>dividual processes.<br />

In this sett<strong>in</strong>g, we state the follow<strong>in</strong>g research questions: (1) Can we confirm<br />

previous results regard<strong>in</strong>g the determ<strong>in</strong>ants <strong>of</strong> order aggressiveness and traders’<br />

order submission strategies when the multivariate dynamics <strong>of</strong> limit order books<br />

are fully taken <strong>in</strong>to account? (2) How strong are the (dynamic) <strong>in</strong>terdependencies<br />

between the <strong>in</strong>dividual processes and how important is it to account for the order<br />

book dynamics? (3) After modell<strong>in</strong>g the multivariate dynamics, what is the<br />

additional explanatory power <strong>of</strong> order book variables? (4) Can we confirm<br />

theoretical predictions regard<strong>in</strong>g the impact <strong>of</strong> order book variables on traders’<br />

order submission strategies?<br />

Our analysis is based on order book data from the five most liquid stocks traded<br />

on the Australian Stock Exchange (ASX) dur<strong>in</strong>g the period July–August 2002. By<br />

replicat<strong>in</strong>g the electronic trad<strong>in</strong>g at the ASX, we reconstruct the complete order<br />

book at each <strong>in</strong>stant <strong>of</strong> time. The order arrival <strong>in</strong>tensities are modelled us<strong>in</strong>g a sixdimensional<br />

version <strong>of</strong> the autoregressive conditional <strong>in</strong>tensity (ACI) model<br />

<strong>in</strong>troduced by Russell (1999), where we <strong>in</strong>clude explanatory variables that capture<br />

the current state <strong>of</strong> the order book as well as <strong>recent</strong> changes <strong>in</strong> the book.<br />

It turns out that market depth, the queued volume, the bid-ask spread, <strong>recent</strong><br />

volatility, as well as <strong>recent</strong> changes <strong>in</strong> both the order flow and the price play an<br />

important role <strong>in</strong> expla<strong>in</strong><strong>in</strong>g the determ<strong>in</strong>ants <strong>of</strong> order aggressiveness. We show<br />

that the impact <strong>of</strong> these variables is quite stable over a cross-section <strong>of</strong> stocks.<br />

Moreover, these results hold irrespective <strong>of</strong> the specification <strong>of</strong> the model<br />

dynamics. Confirm<strong>in</strong>g the results <strong>of</strong> Coppejans and Domowitz (2002) we also<br />

observe that the arrival rates <strong>of</strong> market orders and limit orders can behave quite<br />

differently <strong>in</strong> their dependence <strong>of</strong> the state <strong>of</strong> the order book. Therefore, a limit<br />

order should not necessarily be considered simply as a less aggressive version <strong>of</strong> a<br />

market order. This f<strong>in</strong>d<strong>in</strong>g motivates modell<strong>in</strong>g the <strong>in</strong>dividual processes <strong>in</strong> a<br />

multivariate sett<strong>in</strong>g. Moreover, we f<strong>in</strong>d clear evidence for multivariate dynamics<br />

and <strong>in</strong>terdependencies between the <strong>in</strong>dividual processes.<br />

It is also shown that the <strong>in</strong>clusion <strong>of</strong> order book variables clearly improves the<br />

goodness-<strong>of</strong>-fit <strong>of</strong> the model. In addition, we demonstrate that a model that<br />

3 For more details, see Sect. 4.3.

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

Saved successfully!

Ooh no, something went wrong!