Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

Make your likes visible on Facebook?

Connect your Facebook account to Prezi and let your likes appear on your timeline.
You can change this under Settings & Account at any time.

No, thanks

A WAVELET-BASED HYBRID APPROACH TO FORECAST STOCK MARKET:

No description
by

ali mehr

on 4 September 2014

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of A WAVELET-BASED HYBRID APPROACH TO FORECAST STOCK MARKET:

What we are going to do?
Forecasting

future
market index

and direction of Stock Market with
three approaches
in order to find a better way to forecast Malaysia market
A little about Machine Learning Systems
A key part of economy of a country is
Why: Posses a Key role in
growth
of an industry.
Research Objectives
1)
To select the
most effective economical and fundamental
variables by an intelligence data mining system;

2)
To study the
co-movement
between the target stock market and selected markets in order to find the most effective markets on the target market;

3)
To examine the
accuracy
of the selected
fundamental
indicators,
selected markets
and selected
technical
indicators
separately
in forecasting the target market index by proposed
hybrid intelligence system
;

Research Objectives
4)
To examine the accuracy of
combination of all
selected fundamental indicators, selected markets and selected technical indicators in forecasting target market indices by proposed
hybrid intelligence system
;

5)
To
evaluate the accuracy
of all forecasted results in order to find out which one gives the best forecast of the market,
technical-alone, fundamental-alone
or
combination of both
;

Background of study
Regarding financial markets
three
most significant artificial intelligence applicability
fields
are: “credit evaluation”, “portfolio management” and “
financial forecasting and planning
”.
Technical, Fundamental
OR
Combination?
Most of the researchers working on technical analysis are from
"Computing"
or
"Information Systems"
departments.
Research Questions
1)
What are the
most effective economical
and fundamental variables?

2)
How is the
co-movement
of the target stock market and
selected markets
? And which of them have the most effect on the target market index?

3)
How accurate the selected
fundamental
indicators,
selected markets
and selected
technical
indicators
forecast
the target market index
separately
?

4)
How accurate the
combination of all
selected fundamental indicators, selected markets and selected technical indicators
forecast
the target market index?

A WAVELET-BASED
HYBRID APPROACH
TO
FORECAST STOCK MARKET
:
TECHNICALS,
FUNDAMENTALS AND
CO-MOVEMENT OF MARKETS

Why:
Increase fund
for firms by individuals purchase of shares.
is a
financial time series
and has "
nonlinear nature
".
So forecasting stock market is a
"complicated mission"
.
"Machine Learning Systems"
or "Artificial Intelligence Systems" have this
ability
of
mapping
and
forecasting
nonlinear trends.
What causes this complication: Political, economical and psychological
variables
.
What indicators people mostly use to study:
Technical

and
Fundamental
Founded on
previous
Financial times series data. (
Historical data
)
Founded on
economical
and
financial
condition and
firm performance
.
Note: According to the literature study on
historical
and
nonlinear
data is
AIs expertise
.
Regarding to the literature: People mostly use Technical
OR
Fundamental, and moreover most of them apply technical variables
OVER

fundamental !?
(looking to this later)
So how about
Combination
of
Technical and Fundamental analysis?
"stock market"
Why? They mostly use financial time series to
test
their
"Novel AI methodology"
!
Useful for short term and
medium term
, Soni (2011).
Mostly Finance, Economics and sometimes computing and IS departments work in this field.
Useful for
medium term
and long term trading, Soni (2011).
Fundamental analysis result is a more suitable
"Voting machine"
than
Weighing machine
, Graham et al. (2004).
This is where we as
finance researchers

should
go in
, to make an
opportunity
to
combine
technical and fundamental analysis, which is interesting for
fund managers
,
investors
more importantly
government
.
We are going to add
another factor

in this combination to make it more interesting
with the aim of
achieving more
accurate
and
reliable
result.
Chavan and Patil (2013) and Adebiyi et al. (2012 a,b)
believes

that "
Hybridized
parameters give
better result
that applying
only single
type of input variables.", but Chandwani (2014) and Chen et al. (2011)
does not believe
in that.

This
problem of combining
technical and fundamental variables
still needed
to be studied, but in addition to that ...



3rd Approach!?
Select the best
Technical indicators
from the literature
Select the best
Fundamental variables
and use intelligent data mining technique to gain the most effective indicators on the target market
Select markets which possess "
CO-MOVEMENT
" with Malaysia stock market
The
tendency
of two markets to
move
in
parallel
. This is where
correlation
comes to help!

- Opening price
- Closing price
- Highest price
- Lowest price
- Trading Volume


Applying indicators:
To extract the data pattern and trend easier.

- Stochastic %K
- Stochastic %D
- A/D Oscillator
- Relative Strength Index (RSI)
- MACD



- GDP
- Short term interest rate
- Long term interest rate
- Industrial production
- Government consumption
- Consumer price index
- Gross national product
- Crude oil price
- Private consumption

- New York Dow Jones average index
-
Japan
stock market index (NIKKEI)
-
Korea
stock exchange (KOSPI)
- Stock Exchange of
Thailand
SET Index
-
Jakarta
Stock Exchange Composite Index
-
Shanghai
Stock Exchange Composite Index
-
Taiwan
Stock Exchange Weighted Index
-
Hong Kong
Hang Seng Index
-
Bombay
Stock Exchange (BSE) Index
- FTSE Straits Times Index (STI)
We apply our
Model
to get the forecasting result of these 3 approaches
separately
, result
1
, result
2
and result
3
.
Then we

combine all results
with
"
nonlinear ensemble

using Principle Component Analysis
(
PCA
)
technique" - Yu et al. (2005) and Cheng et al. (2010) -to get the
final result
, and set up
comparison
of all forecasting results at the end.
There is no need for any
assumptions
in these systems compare to traditional methods because system will be
constructed from
input data.
There is no need to change the
numerical
data to
nominal data
in these systems. The numeric nature of these systems is
superior
to nominal nature of
traditional
manipulation techniques because in conventional techniques numeric data must be
converted into nominal
values before they can be used as input, and therefore there are the problems of
losing information
, inappropriate data intervals and different conversion methods leading to different mining results.
When
new data
comes to the existing machine learning system, there is
no need
to put all the
previous data
in the system, because the system has already
constructed
by previous data.
Another advantage of these systems is that they
learn
during the
process
.

That's why the
key part
to work with these systems is
how to train them
.


The whole system consists of
preprocessing
,
forecasting model
and
evaluation
.

(ANNs) Mostly 3 layers: Input Layer, Hidden Layer, Output Layer
These systems are like
a baby
, you should train them how to act with input data.
This part of these system is called
Hidden Layers
(in ANN).
Now, it is
done by all historical
data, and it has already forecast
next day value
.
Tomorrow
comes and you give the
actual value
to the system,
it will say:

hmmmm, I thought so ...
(
or even:
I said so ..)
Again it will learn from
the new data to
improve
itself
, and so on ...
Now, consider a machine learning system (
baby
) which is
under training
by our rules.
We put
historical data
(e.g. daily data) into the system.
The system is trained to analyze data
day by day
, and forecast
next day
value.

It will become more
accurate
to forecast next day value, day by day.
Co-movement between KLCI and Dow Jones
As you can see in some parts there are
parallel movements
between KLCI and Dow Jones.

We try to establish a model in order to use these kind of co-movements for a
more reliable
forecast of future market index.
Artificial intelligence methods can be applied in three types:

- Single intelligent technique
- Expert System Applications
-
Hybrid Intelligent Systems
(HIS)
There are lots of these models and algorithms like:

Artificial Neural Networks (ANN), Rule-based systems technique, Neuro-Fuzzy, Feedforward Neural Network,
Backpropagation Nueral Networks (BPNN)
, Hidden Markov Model, Wavelet Transform,
Support Vector Machine (SVM)
, Support Vector Regression, Fuzzy Ruled System, Adaptive Neuro Fuzzy Inference System, Autoregressive Conditional Interval,
Wavelet Transform
, Recurrent Neural Network, Genetic Algorithm, Artificial Fish Swarm Algorithm, Bee Colony Algorithm, Adaptive Bacterial Foraging Algorithm, and etc.
Table 1:
Hybrid Intelligent Systems

compared with

conventional systems

in financial planning and forecasting
Table 2:
comparison between
a HIS
and one or
some single intelligent models
results in financial planning and forecasting
Table 3:
comparison among
some HISs
and even
some single intelligent methods
in the same study and data
Sample Design and Hybrid Intelligent Model of the Study
Last
twenty years
technical and fundamental (economical) data, 1994 -2014, of
Malaysia
stock market, as target market, will be studied with the
same period
of
selected markets
like the US, Japan, China, India, Singapore, Hong-Kong, Thailand, Taiwan, Indonesia and South Korea.
The
exchange rate
of the target and selected markets will be studied as
input variables
too.

Monthly
technical variables and
quarterly
fundamental variables will be collected as input for this study.
Wavelet Analysis
Wavelets allow
complex information
such as
patterns
to be decomposed into
elementary forms
at different positions and scales and subsequently reconstructed with high precision.

1) One of the
main advantages
of wavelets is that they offer a simultaneous localization in
time
and
frequency
domain.

2) The minimum requirement of
normalization
of data.

3) Wavelets have the great advantage of being able to
separate the fine details in a signal
. Very
small wavelets
can be used to isolate very
fine details
in a signal, while very
large wavelets
can identify
coarse details
. (That's why we can use it in short term and long term data)

4) The second main advantage of wavelets is that, using fast wavelet transform, it is computationally very fast.

5) A wavelet transform can be used to decompose a signal into component
wavelets.

6) Wavelet theory is capable of revealing aspects of data that other signal analysis techniques miss the aspects like trends, breakdown points, and discontinuities in higher derivatives and self-similarity.

7) It can often compress or de-noise a signal without appreciable degradation.
Support Vector Machine (SVM)
SVM is a
supervised learning algorithm
for
Pattern Recognition
and Regression Estimation – Non Parametric.

Remarkable characteristics of SVMs:

1) Good
generalization
performance

2) Based on a
strong
and nice
Theory

3) Generally
avoids over-fitting

4) It is
robust to noise


5) Generalize well even in high dimensional spaces under small training set conditions

6) It has a simple geometrical interpretation in a high-dimensional feature space that is nonlinearly related to input space

7) Absence of local minima: Training SMV is equivalent to solving a linearly constrained quadratic programming problem. Hence the solution of SVMs is
unique and globally optimal.

Back-Propagation Neural Network (BPNN)
In this study, one of the widely used ANN models, the back-propagation neural network (BPNN), is used for time series forecasting.

Advantages of BPNN:

1) Neural networks are very effective for
solving

multiple class
classification
.

2) A neural network can be seen as machine that is
designed
to
model the way in which the
brain perform
a particular task.


3) Provides
flexible mapping
between inputs and outputs.

4) Neural networks have been used to identify system as well as data classification technique with significant success rate.

5) Neural network can adjust themselves to data without any explicit function. It is function approximation technique in that network can approximate any function.



Wavelet - BPNN
Wavelet - SVM
Nonlinear Ensemble PCA Technique
Hsieh et al. (2010), Chen et al. (2001), Wu (2010), Lora et al. (2007),
Zhang and Dong (2001)
, Pindoriya and Singh(2008).
Cao and Tay (2001), Boyacioglua and Avci (2010),
Tang et al. (2009)
, Zhang et al. (2004)
Yu et al. (2005)
, Jolliffe (1986), Karhunen and Joutsensalo (1995)
Thanks for your attention
and patience.
Co-movement of markets (monthly)
Scope of the Study
In this study we use data from Malaysia and the markets that have co-movement with Malaysia (we call these markets, selected markets) and use an intelligent data mining technique to select the most effective fundamental indicators in the target market. Then we apply fundamental indicators, indices of selected markets and technical indicators separately in order to forecast the future market index of the target market. From this step, we would have 3 different forecasts of the Malaysian market index. Next step is to combine all mentioned indicators (fundamental, technical and selected markets) in order to apply them in forecasting future market index. The result of this step will be compared with the other 3 results to see whether our hybrid intelligence system is successful to combine the ability of all indicators together and get a better result.







Justification
As the first significance of this study we can mention the introduction of an improved solution for combination of technical and fundamental variables in order to forecast the target market index more accurately. Secondly, this study will combine not only technical and fundamental but also comoving markets indices with the data collection. This attempt itself will be a novel attempt as hybrid dataset to combine these three approaches together, in order to study stock market index. Third significance of this study will be preprocessing and selecting the best dataset in term of forecasting target market and getting highest impact on the market index. Moreover, the methodology that this study will be done by is a novel hybrid intelligent system which itself will be a remarkable feature of this research.








Expected Contribution
Forecasting stock market index with lesser data and simpler approach is always in the interest of investors and fund managers. Moreover, the result of this study will be a useful and extraordinary help not only to the government section in order to map the market, analyze their interrelation with other countries, enhance economical factors where it needed, and even modify their regulation and diplomacies by accurate forecast of future market index, but also to individual and corporate fund managers and investors in order to trace the future of the market by the most accurate tools. We hope this study achieve an accurate result and be a good help to individual traders, corporate investors and government sectors.

University Malaya
Faculty of Business and Accountancy

Department of Finance and Banking

Supervisor:
Dr. Mohamed Shikh Abubaker Al-Baity

Student:
Mohammadali Mehralizadeh CHA120016

Sep 2014
Full transcript