Integrating a piecewise linear representation method and a neural network model for stock trading points prediction
1. Integrating a Piecewise Linear Representation
Method and a Neural Network Model for
Stock Trading Points Prediction
Pei-Chann Chang, Chin-Yuan Fan, and Chen-Hao Liu
TSMCC.2008
Presenter: Yu Hsiang Huang
Date: 2011-12-30
3. Introduction
• Stock market
– Highly nonlinear dynamic system
• Interest rates, inflation rate, economic environments, political issues…
• Most resent research
– Derive accurate models
– Predict the future price of stock movement
• In this paper
– Trading decision
• Buy/Sell points
– Critical role to make a profit
4. IPLR Model
Candidate Stocks Screening
Selected stock
GA
Technical indexes SRA Related input variable
no PLR Turning point Trading signal
Reach Expect output input
number of
generation ? Calculate
profit
Test BP(train)
yes
End Trading decision
Related input variables BP Buy/sell
Related input variables
5. Genetic Algorithm
Initialization Randomly generate initial population
50
1 0 … 0 1 10
0.8
0.1
Selection Fitness function roulette-wheel selection
Tournament selection
Reproduction Crossover Two-point Mutation
genetic diversity
Termination # of generation , reach the best fitness value , …
6. IPLR Model
Candidate Stocks Screening
Selected stock
GA
PLR Turning point Trading signal
10. Piecewise Linear Representation
Derive the trading signal
Tradition
Up Down : 1 [sell]
Down UP : 0 [buy]
Not quite related to the
price variation
14. Stepwise Regression Algorithm
Apply by SPSS (Statistic Package for Social Science)
Calculate the significant value S
yes no
X1
X2 Y
X3
X4 no
Last X ?
X5
Xp yes
Output
15. IPLR Model
Candidate Stocks Screening
Selected stock
GA
Technical indexes SRA Related input variable
PLR Turning point Trading signal
Expect output input
BP(train)
17. IPLR Model
Candidate Stocks Screening
Selected stock
GA
Technical indexes SRA Related input variable
PLR Turning point Trading signal
Expect output input
Test BP(train)
Trading decision
Related input variables BP Buy/sell
18. Back-propagation Network
Trading decision
Test data input to BP
Change of the trading signal pass through the boundary value:
Change is upward sell
Change is downward buy
Boundary value : 0.508
19. IPLR Model
Candidate Stocks Screening
Selected stock
GA
Technical indexes SRA Related input variable
no PLR Turning point Trading signal
Reach Expect output input
number of
generation ? Calculate
profit
Test BP(train)
yes
End Trading decision
Related input variables BP Buy/sell
Related input variables
20. Experimental results
Historic data : from 2004/01/02 to 2006/04/12
Training data : 2004/01/02 to 2005/09/30
Testing data : 2005/10/1 to 2006/04/12
Up-trend : 30-day moving average cross over 90-day moving average
Down-trend : 30-day moving average cross down 90-day moving average
Steady : no major tendency of 30-day moving average with 90-day moving average
23. Conclusion
• Trading decision > determine stock price itself
• IPLR
– PLR : find turning point
– GA : improve the threshold value for PLR
– BPN : train the connection of the model
– Significant amount of profit
• Clustering of financial time series data
• A different forecasting model
– SVM , FNN,…
• A similar training pattern