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.
100<br />
S. Frey, J. Grammig<br />
quotes on each market side are used for the construction <strong>of</strong> break even and update<br />
conditions. Table 3 reports the estimation results for a specification that does not<br />
rely on a parametric assumption on the distribution <strong>of</strong> the market order size when<br />
construct<strong>in</strong>g the marg<strong>in</strong>al break even and the update conditions as described <strong>in</strong><br />
section 3.2.1. The results reported <strong>in</strong> Table 4 are obta<strong>in</strong>ed when us<strong>in</strong>g the average<br />
break even and update conditions (11) and (12) for GMM estimation, while<br />
ma<strong>in</strong>ta<strong>in</strong><strong>in</strong>g the parametric assumption (2) about the trade size distribution. For<br />
each parametric specification, the moment condition <strong>of</strong> Eq. (6) is employed, too.<br />
The full set <strong>of</strong> eight update conditions is exploited.<br />
The estimates reported <strong>in</strong> Table 3 show that abandon<strong>in</strong>g the parametric<br />
assumption concern<strong>in</strong>g the market order size distribution improves the results only<br />
marg<strong>in</strong>ally. For four <strong>of</strong> the thirty stocks the model is not rejected at the 1%<br />
significance level. Hasbrouck’s (2004) conjecture that the distributional assumption<br />
might be responsible for the model’s empirical failure is therefore not supported.<br />
Generally, the estimates <strong>of</strong> the adverse selection components, transaction<br />
costs and drift parameters do not change dramatically compared to the basel<strong>in</strong>e<br />
specification. The transaction cost estimates rema<strong>in</strong> negative.<br />
Ma<strong>in</strong>ta<strong>in</strong><strong>in</strong>g the distributional assumption, but us<strong>in</strong>g average break even<br />
conditions <strong>in</strong>stead <strong>of</strong> marg<strong>in</strong>al break even conditions, considerably improves the<br />
empirical performance. Table 4 shows that we have model non-rejection for 22 out<br />
<strong>of</strong> the 30 stocks at the 1% significance level. With a s<strong>in</strong>gle exception the estimates<br />
<strong>of</strong> the marg<strong>in</strong>al transaction cost parameter γ are positive for those stocks for which<br />
the model is not rejected at 1% significance level. The size <strong>of</strong> the implied<br />
transaction cost estimates are broadly comparable with the relative realized spreads<br />
figures reported <strong>in</strong> Table 1. For example, the estimation results imply that<br />
transaction costs account for 0.013% <strong>of</strong> the euro value <strong>of</strong> a median sized<br />
DaimlerChrsyler trade. This value is quite comparable with the average relative<br />
realized spread which amounts to 0.010% (see Table 1).<br />
Figure 1 shows graphically the improved empirical performance delivered by<br />
the revised methodology. The figure depicts means an medians <strong>of</strong> implied and<br />
observed ask side price schedules <strong>of</strong> four selected stocks. The results obta<strong>in</strong>ed from<br />
the basel<strong>in</strong>e estimation which uses marg<strong>in</strong>al moment conditions confirm the<br />
disturb<strong>in</strong>g f<strong>in</strong>d<strong>in</strong>gs reported <strong>in</strong> Såndas’ (2001). The price schedules implied by the<br />
model estimates are below the observed price schedules at all relevant volumes.<br />
The economically implausible negative price discount at small volumes is caused<br />
by the negative transaction costs estimates. This suggests that the model is not only<br />
rejected on the grounds <strong>of</strong> statistical significance, but that fundamentally fails to<br />
expla<strong>in</strong> the data. The model does a bad job even <strong>in</strong> describ<strong>in</strong>g the ‘average’ state <strong>of</strong><br />
the order book. However, Fig. 1 shows that when work<strong>in</strong>g with average break even<br />
and update conditions the empirical performance <strong>of</strong> the model is considerably<br />
improved. Especially the median observed price schedules correspond closely to<br />
those implied by the model. 14<br />
14 We have also estimated a specification that comb<strong>in</strong>es the nonparametric approach towards<br />
trades sizes and average moment conditions, but the results (not reported) are not improved<br />
compared to the parametric version. In this version, the model is not rejected for 16 out <strong>of</strong> 30<br />
stocks. The follow<strong>in</strong>g analysis therefore focuses on the parametric specification us<strong>in</strong>g average<br />
moment conditions.