Pattern Discovery of Fuzzy Time Series for Financial Prediction

1
Pattern Discovery of Fuzzy Time Series for
Financial Prediction

• Chiung-Hon Leon Lee,
• Alan Liu, Member, IEEE, and Wen-Sung Chen
• Presenter Bob Crichton

2
Problem

• Investors want to maximize profit from stock
  sales
• Need to know when to buy and sell

3
Some Other Methods Used For Financial Prediction

• Neural Networks
• Genetic Algorithms
• NeuroFuzzy
• Classification and Regression Tree
• Naïve Bayes
• Fuzzy Time Series

4
Whats Wrong With Other Methods?

• Training of systems is not trivial, results
  cannot be re-used
• Systems are Black Boxes
• Models may need tuning, Investors do not have
  background knowledge to do so

5
Whats Wrong With Other Methods?

• Gap between prediction results investment
  decisions
• Investors are more concerned with reversal
  patterns than the actual price

6
Authors proposal

• Knowledge-based method, transfers data to
• Comprehensible rules
• Visual patterns

7
How to represent time series data?

• Symbolic Fuzzy Linguistic Variables
• Computation Load is reduced
• Linguistic variables can be comprehensible to
  investors

8
Japanese Candlestick Theory
9
Color Definitions

• If open-close gt 0 then the body color is BLACK
• If open-close lt 0 then the body color is WHITE
• If open-close 0 then the body color is CROSS

10
Example Candlestick Chart
11
Modeling the Candlestick Pattern

• Whats important?
• Lengths of shadow and body
• Imprecise, i.e. short, long
• Opening and closing values in relation to
  previous time period
• Both use Fuzzy Linguistic variables to
  describe/model

12
Membership Function For Shadow and Body Length
13
Membership Function For Open and Close Styles
14
Pattern Recognition Problems

• Sensing Problem
• Acquisition of measured values, i.e. recording
  stock prices over time
• Feature Extraction Problem
• Extract characteristic features from input data,
  i.e. candlestick lengths
• Pattern Classification Problem
• Must determine optimal decision procedures

15
Fuzzy Sets for TAIEX

• A1 (EXTREME DECREASE)
• A2 (LARGE DECREASE)
• A3 (NORMAL DECREASE)
• A4 (SMALL DECREASE)
• A5 (SMALL INCREASE)
• A6 (NORMAL INCREASE)
• A7 (LARGE INCREASE)
• A8 (EXTREME INCREASE)

16
TAIEX Data, Variations, and Fuzzy Sets
17
TAIEX Forecasted Results
18
System Prototype
19
Authors Conclusions

• Fuzzy Candlestick patterns can be used to
  increase efficiency of KD of financial time
  series.
• Using system, investors can
• Save and share investment experience
• Increase efficiency of investment strategies

20
Future work

• Implement system on large scale

21
Any Questions?

• ?????