Recent Advances of Exception Mining in Stock Market - Evolving Application Domains of Data Warehousing and Mining - page 224

Table 2. Examples of original alerts

No. Date Time Type Threshold Value

8 13/06/2005 9:30:00 Price Change Rate 0.04 0.04

18 24/05/2006 9:30:03 Turnover Rate 1000000 2167700

23 21/11/2005 9:25:00 Price Change Rate 0.04 0.05

24 27/03/2006 9:25:00 Price Change Rate 0.04 0.1

33 27/03/2006 9:25:05 Price Change Rate 0.08 0.1

time series X, and one is [{P 1 , 97%}, {P 23 , 91%},
{P 9 , 96%}] on time series Y and [{P 33 , 87%},
{P 3 , 92%}, {P 9 , 97%}] on time series Z. Then all
the points are ranked as [{P 1 , 98%}, {P 9 , 97%},
{P 3 , 92%}, {P 23 , 91%}, {P 33 , 87%}] by taking the
maximum of the probability for each point. The last
step is to determine the number of final outliers.
We choose the top k outliers from the ranked list as
the final outliers. The value of k can be set based
on the specific application, If we set k =3, then the
final outliers are P 1 , P 9 and P 3 .
The proposed V-BOMM and P-BOMM use
the principle curve algorithm as the kernel. The
principle curve is applied on the three individual
time series. The computation complexity of
principle curve is O ( n 2 ) (Zhang & Wang 2003).
Therefore, the computation complexity of our
proposed OMM is also O ( n 2 ).

Based on the relevant financial knowledge, we
constructed daily price return, daily price range
and daily trade amount as the three time series and
each of them was assigned with the same weight
on experiments. The daily amount was the original
attribute of raw data, but the daily price return was
calculated out from the raw data:
Price Return = ( P 1 - P 2 ) / P 2 , (3)
where P 1 is the current closing price and P 2 is
the previous closing price. The daily price range
was calculated with the following equation:
Daily Price Range = P 3 - P 4 , (4)
where P 3 is the daily highest price and P 4 is
the daily lowest price.
We chose the real Alerts generated by China
stock exchange during 1 June 2004 to 3 Mar 2006
as a benchmark, which is the same as that of our
experimental data. Rules were set by surveillance
staff to detect the abnormal movement of turnover
exception and price change rate exception. Some
examples of rules are shown in Table 2.
and VOMM trade

Application Background and data

The Shanghai Stock Exchange (SSE) is a Chinese
stock exchange based in Shanghai, with a mar-
ket capitalization of nearly US$3.02 trillion in
2007 making it the largest exchange in mainland
China. It is a non-profit organization directly
administered by the China Securities Regulatory
Commission (CSRC). The proposed OMM was
tested and evaluated with the data from SSE. The
data are daily trade records in SSE from 1 June
2004 to 3 Mar 2006. It has the 425 trading days.
The attributes of the data sets include the high-
est price, the lowest price, the closing price and
trade amounts.

k

.

_ amounts

• Stage 2: Run Voting-based OMM on Daily

Price Return data, Daily Price Range data
and Trade Amount data respectively, and
choose the top k ( k =60,50,40,30,20,10)
samples as outliers on each measure. The
outcomes are vectors named as V-BOMM k
( k =60,50,40,30,20,10)

• Stage 3: Run the Probability-based OMM

on the three data respectively, and choose
the top k ( k =60,50,40,30,20,10) samples as