Database Reference
In-Depth Information
Variance-Based Outlier Mining (VOMM)
market is very complicated and the anomalies
may affect and/or be reflected in many measures,
including closing price, volume, price range,
depth, spread, trading, etc. Therefore, it is far
from enough to detect outliers from closing price
only, and multiple measures in stock market need
to be considered.
Qi and Wang (2004) proposed a Variance-based
Outlier Mining Model (VOMM), a general outlier
model based on principal curve (Zhang & Wang
2003) to find outliers on daily closing price. The
design of VOMM aims to solve the outlier min-
ing problem where outliers are highly intermixed
with normal data. In VOMM, the information of
data is decomposed into normal and abnormal
components according to their variances. With
minimal loss of normal information in the VOMM,
outliers are viewed as the top k samples holding
maximal abnormal information in a dataset. The
principal curve is a smooth nonparametric curve
passing through the “middle” of the dataset and
describes the normal information with a nonlinear
summary of the data.
In stock market, the daily closing price is af-
fected not only by daily random fluctuation, but
also long-term trend. If we consider the long-
term trend as normal information and the daily
random fluctuation as abnormal information, it
is difficult to separate the abnormal information
from the normal information. Therefore, VOMM
is an appropriate approach to handle this issue.
Qi and Wang (2004) applied VOMM to analyze
the daily INDEXSH (Integrate Index in Shanghai
Stock Exchange of China) during the period of 1st
January 1998 to 31st December 2001. To evaluate
the experimental results, the stock analyst was
asked to detect the outliers of INDEXSH in the
same period of time. The authors also collected
the significant events happened in this period of
time to evaluate the results. The results showed
that VOMM is feasible to identify the outliers of
INDEXSH. In addition, the authors compared
VOMM with the Gaussian model (Smith 1981)
and GARCH model (Franses & Dijk 2000) on
the same datasets, and the results indicate that
VOMM perform better than Gaussian model and
GARCH model.
As a general outlier mining model, VOMM is
applicable in many applications. However, stock
case-Based reasoning for
Market Surveillance
Case-based reasoning (CBR) is the process of
solving new problems based on the solutions of
similar past problems (Buta & Barletta 1991). CBR
builds a case database to store the past problems
and their solutions. Whenever an input is coming,
a case-based system will search its case database
for an existing case that matches the input. If a
past case is identified to exactly match the input
problem, the system will immediately provide a
solution for the problem. If, on the other hand,
there are no past cases matching the input, sys-
tem will try to retrieve a case that is similar to
the input situation but not exactly appropriate to
provide as a complete solution. The case-based
system must then find and modify those small
portions of the retrieved case that do not meet
the input specifications. This will also provide a
complete solution, but it generates a new case that
can be automatically added to the case database.
Then, the system is updated when new cases are
stored in case database. Normally, to build a case
database, terms and classified cases need to be
defined by experts.
Buta & Barletta (1991) presents a CBR ap-
proach for market surveillance. Their approach
aims to evaluate trading patterns in the context of
additional company, industry, and market data, and
identify possibly suspicious trades. By using CBR,
the time cost of analyzing suspicious stock trade
is reduced significantly. They also introduced an
intelligent market monitor (IMM) for processing
transactions. It is a market surveillance applica-
tion at the Toronto Stock Exchange (TSE). Their
Search WWH ::
Custom Search