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.
The performance analysis <strong>of</strong> chart patterns 217<br />
extrema. The f<strong>in</strong>al step consists to scan the extrema sequence to identify an<br />
eventual chart pattern. If the same sequence <strong>of</strong> extremum was observed <strong>in</strong> more<br />
than one w<strong>in</strong>dow, only the first sequence is reta<strong>in</strong>ed for the recognition study to<br />
avoid the duplication <strong>of</strong> results.<br />
B.2. M2<br />
M2 is the extrema detection method built on <strong>high</strong> and low prices. Accord<strong>in</strong>g to this<br />
method, maxima and m<strong>in</strong>ima have to be detected onto separate curves. Maxima on<br />
<strong>high</strong> prices curve and m<strong>in</strong>ima on the low one.<br />
Let Ht and Lt (t =1,...,T), be respectively the series for the <strong>high</strong> and the how<br />
prices, and k a w<strong>in</strong>dow conta<strong>in</strong><strong>in</strong>g l regularly spaced time <strong>in</strong>tervals such that:<br />
Hj;k fHtjktkþl1g (21)<br />
Lj;k fLtjktkþl1g; (22)<br />
j =1,...,l and k =1,...,T−l+1. We smooth these series through the kernel estimator<br />
detailed <strong>in</strong> Appendix A to obta<strong>in</strong> ^mh XHj;k and ^mh XLj;k . We detect maxima<br />
on the former series and m<strong>in</strong>ima on the latter one <strong>in</strong> order to construct two<br />
separate extrema series max ^mhðXH j;k Þ and m<strong>in</strong> ^mhðXL j;k Þ such that:<br />
max ^mhðXH j;k Þ ¼ ^mh XHj;k S ^m 0<br />
h<br />
m<strong>in</strong> ^mhðXL j;k Þ ¼ ^mh XLj;k S ^m 0<br />
h<br />
XHj;k ¼þ1; S ^m0<br />
h<br />
XLj;k ¼ 1; S ^m0<br />
h<br />
XHjþ1;k ¼ 1<br />
XLjþ1;k ¼þ1 ;<br />
where S(x) is the sign function def<strong>in</strong>ed <strong>in</strong> the previous Section.<br />
We record the moments for such maxima and m<strong>in</strong>ima, denoted respectively by<br />
tM ^m XHj;k h and tm ^m XLj;k h and we project them on the orig<strong>in</strong>al <strong>high</strong> and how<br />
curves to deduce the orig<strong>in</strong>al extrema series maxHj;k and m<strong>in</strong>Lj;k , such that:<br />
maxHj;k ¼ max H tM ð ^mhðXH j;k ÞÞ 1;k; H tM ð ^mhðXH j;k ÞÞ;k; H tM ð ^mhðXH j;k ÞÞþ1;k<br />
m<strong>in</strong>Lj;k ¼ m<strong>in</strong> L tmð ^mhðXL j;k ÞÞ 1;k; L tmð ^mhðXH j;k ÞÞ;k; L tmð ^mhðXH j;k ÞÞþ1;k :<br />
However, this method does not guarantee alternate occurrences <strong>of</strong> maxima<br />
and m<strong>in</strong>ima. It is easy to observe, <strong>in</strong> the same w<strong>in</strong>dow k, the occurrence <strong>of</strong> two<br />
consecutive m<strong>in</strong>ima on the low series before observ<strong>in</strong>g a maximum on <strong>high</strong> series.<br />
To resolve this problem we start by record<strong>in</strong>g the moments for the selected maxima<br />
on <strong>high</strong> curve, tM(Hj,k), and m<strong>in</strong>ima <strong>in</strong> low curve, tm(Lj,k). Then we select, for<br />
w<strong>in</strong>dow k the first extremum from these two series, E1,k , and its relative moment,<br />
, such that:<br />
tE1;k<br />
tE1;k<br />
¼ m<strong>in</strong><br />
t<br />
tM Hj;k ; tm Hj;k (23)<br />
E1;k ¼ maxH j;k [ m<strong>in</strong>Lj;k j ¼ tE1;k : (24)