Application of Neural Networks in Financial
Data Mining
Defu Zhang, Qingshan Jiang, Xin Li
Abstract—This paper deals with the application of a well-known
neural network technique, multi-layer back-propagation (BP) neural
network, in financial data mining. A modified neural network
forecasting model is presented, and an intelligent mining system is
developed. The system can forecast the buying and selling signs
according to the prediction of future trends to stock market, and
provide decision-making for stock investors. The simulation result of
seven years to Shanghai Composite Index shows that the return
achieved by this mining sys-tem is about three times as large as that
achieved by the buy and hold strategy, so it is advantageous to apply
neural networks to forecast financial time series, the different investors
could benefit from it.
Keywords—
data mining, neural network, stock forecasting.
I. I
NTRODUCTION
EURAL networks are a class of generalized nonlinear
nonparametric models inspired by studies of the human
brain. Their main advantage is that they can approximate any
nonlinear function to an arbitrary degree of accuracy with a
suitable number of hidden units [1]. Neural networks get their
intelligence from learning process, and then this intelligence
makes them have the capability of auto-adaptability, association
and memory to perform certain tasks. For a more detailed
description of neural networks, the interested readers are
referred to the papers in [2, 3, 4].
Financial forecasting is of considerable practical interest.
Due to neural networks can mine valuable information from a
mass of history information and be efficiently used in financial
areas, so the applications of neural networks to financial
forecasting have been very popular over the last few years [5, 6,
7, 8, 9, 10]. Some researches [11, 12] show that neural networks
performed better than conventional statistical approaches in
financial forecasting and are an excellent data mining tool.
However, a large number of published researches only pay
attention to giving the experimental simulating results and
seldom give detailed description from modeling to forecasting
to trading, and say nothing of giving their critical techniques. In
addition, despite neural networks have many advantages,
however, they still meet some problems, for example,
overfitting and poor explanation capability and so on. Although
many researchers pay special attention to avoid those problems,
the result is still dissatisfactory. In this paper, all mentioned
above will be further investigated.
Manuscript received November 3, 2004. This work was supported in part by
Xiamen University under Grant X01122.
D. F. Zhang is with the Department of Computer Science, 361005, Fujian,
China (Tel: 0086-0592-5918207; fax: 0086-0592-2183502; e-mail: dfzhang@
xmu.edu.cn).
Though the efficient market hypothesis [13], which is
currently the most popular view on market behavior, states that
no mining system can achieve consistently average returns
exceeding the average returns of the market indices as a whole.
However, many research results have shown that the mining
system based on neural networks can outperform market if it is
properly designed [14]. In this paper, the theory of neural
networks is briefly discussed, and a neural network model to
forecast future trends of stock time series is researched. Some
improved strategies are presented. At last, an intelligent mining
system is developed. The simulation results to Shanghai
Composite Index show that neural networks can be applied to
make profit for different investors.
II. F
INANCIAL TIME SERIES FORECASTING
A time series is a sequence of vectors
, where
x represents
past value which varies with the time
),,,,,(
12
ttitntt
xxxxX
it
t , such as stock price. One
of the aims in the paper is to predict the future trends on the base
of past price patterns . It can be formally stated as follows:
find a function
, so that:
t
X
RRF
n
o:
~
),,,,,(
12
ttitntt
xxxxFx ,
~
where
t
x is a predict value.
1
~
t
x can be predicted after x
t
is
predicted based on the following function
)
~
,,,,,,(
~
12)1(1 tttitntt
xxxxxFx
.
~~~
Similarly,
132
,,,
dttt
xxx can be predicted. The neural
networks of performing prediction are to approximate
. The
trained network is then used to predict the price of future
days.
d
It is well known that stock market is nonlinear dynamical
system, and is affected mainly by factors like interest rates,
inflation, economic growth, political situation and so on.
Although there is correlation, dependency and interaction
between these factors is obvious, their relation is rather difficult
to express in mathematical function, for example
. So
forecasting financial markets is a really challenging task. It is
well known that the closing price of stock market is one of the
most important factors and includes a lot of useful information,
so the closing price time series is selected to predict the future
trends of stock market.
TRANSACTIONS ON ENGINEERING, COMPUTING AND TECHNOLOGY V1 DECEMBER 2004 ISSN 1305-5313
ENFORMATIKA V1 2004 ISSN 1305-5313 392 © 2004 WORLD ENFORMATIKA SOCIETY