This document describes a method for financial market prediction and portfolio optimization using fuzzy decision trees. It involves screening stocks using technical indicators to classify them, then ranking them using fundamental indicators. Three stocks are selected - one showing an uptrend, downtrend, and steady state. Historical stock data is gathered and technical indicators are determined. A piecewise linear representation method and stepwise regression analysis are used for feature selection and data preprocessing. Fuzzy rules and a decision tree are generated using genetic algorithms to determine stock price decisions and transactions based on trends and indicators. The result is a collaborative trading model to detect stock turning points.
5. STOCK RANKING
β’
Corporate Evaluation using Fundamental Indicators
β’ Profitability β Returns on Assets and Equity
β’ Management Performance β Assets and Inventories Turnover
β’ Capital Structure β Assets to Liabilities, Liabilities to Equity
β’ Sales, Profit, Transaction Volume, Marginal Account
6. STOCK SELECTION
β’
Select 3 different stocks β one each showing uptrend, downtrend, and steady
state
β’
Attempt to display different profit making strategies in stock trading
β’
All subsequent processes are applied on these 3 stocks
8. TRAINING PHASE
β’
Gather Historical Stock data
β’
Obtain financial time series and price charts from data
β’
Determine technical indicators and momentum oscillators from charts
11. PIECEWISE LINEAR REPRESENTATION
METHOD
β’
Mining of trading points
β’ Points of begin (P) and end (Q) on a term of closing prices in the
ascending order of dates
β’ Point K having longest straight line distance between P and Q
β’ K is the turning point resulting in 2 segments.
β’ Apply recursively in the resulting segments till minimum distance
threshold
12. PIECEWISE LINEAR REPRESENTATION
METHOD
β’
Trading signals transformation
β’ Convert PLR segments into trading signals
β’ Uptrend segment
β’ I <= L/2 : 0.5 β (I β 1) / L
β’ I <= L/2 : I / L β 0.5
β’ Downtrend segment
β’ I <= L/2 : 0.5 + (I β 1) / L
β’ I <= L/2 : 1.5 β I / L
β’ Ranges from 0 to 1
β’ Can also act as a potential technical indicator
14. STEPWISE REGRESSION ANALYSIS METHOD
β’
Data Preprocessing for Feature Selection
β’ Used to select important factors which affect forecasting results
β’ Sort out affecting variables to leave more influential ones in the model
β’ Adding or removing factors to find the fittest combination, decided by Ftest statistical value (takes into account the PLR)
16. FUZZY RULES AND DECISION TREES
β’
Fuzzification
β’ Set of indicators selected by SRA fed into data fuzzification module
β’ This module transforms technical indicators to fuzzy values
β’ Adopt triangular and trapezoidal membership functions for the module
β’ Output decision is obtained as a Gaussian membership function
18. FUZZY RULES AND DECISION TREES
β’
Defuzzification
β’ Output from fuzzy inference scheme is transformed into a meaningful
decision
β’ Implemented using the popular Center of Area (COA) methods in the
Fuzzy Control Moduleβs algorithm
19. FUZZY RULES AND DECISION TREES
β’
Examples of Fuzzy decision rules
β’ If MACD above signal line, then BUY
β’ If RSI increases above 70, then market is BULLISH
β’ If Price increases above BB upper then market is BULLISH
β’ If MACD is LOW and RSI upper goes HIGH to LOW, then SELL
β’ If MACD is HIGH and CCI upper goes LOW to HIGH, then BUY
21. GENETIC ALGORITHMS AND REFINEMENT
β’
Evolving the decision tree using GA
β’
Fitness function set as forecasting accuracy of the model
Selection
Crossover
Mutation
Replace
Termination
22. RESULT
β’
Decision of Stock price and transaction will be determined by the decision
tree on the basis of trends and indicators
β’
Uptrend if hike in price is greater than 0.5%
β’
Downtrend if fall in price is less than -0.5%
β’
Steady state / hold if y is between -0.5% and 0.5%
23. CREDITS
β’
A Collaborative Trading Model by Support Vector Regression and TS Fuzzy
Rule for Daily Stock Turning Points Detection β Wu, Chang, Chang, Zhang
β’
Evolving and Clustering Fuzzy Decision Trees for Financial Time Series Data
Forecasting β Lai, Fan, Huang, Chang
β’
A Fuzzy Logic Based Trading System β Chueng, Keymak
β’
Nigerian Stock Market Investment using a Fuzzy Strategy β
Neenwi, Kabari, Asagba
β’
Common Stock Portfolio Selection: A multiple criteria Decision making
Methodology and an application to the Athens Stock Exchange β
Xidonas, Askounis, Psarras