Implementing Genetic Algorithms in Finance

Applications

Nihaar Sinha

Abstract

This project investigates the use of genetic algorithms in a financial application of

portfolio optimization.  Genetic algorithms have long been used in optimization problems

as well as the financial sector.  Banks and hedge funds pay millions of dollars to

programmers who can develop the most accurate optimization algorithms.  What this

project does is try to replicate that development on a very small scale- using only three

different companies and 8 total shares in the portfolio.  The genetic algorithm considers a

number of factors in coming up with the optimized solution.  These factors include an

evaluation of the price to earnings ratio of the stock, the yield on the stock, and a special

weight determining how diversified the portfolio is.  While the metrics used are to an

extent crude, they serve the purpose of demonstrating how a concept such as genetic

algorithms can be used in the field of finance.

What are Genetic Algorithms?

Search technique used to find exact and approx. solutions

Used in optimization and search problems

“Evolutionary algorithim”

Inheritance, Crossover, Mutation, Crossover

Four phases: initialization, selection, reproduction, termination

Genetic Algorithm

http://www.meteck.org/IMAGES/image001.gif

Similar Research

Genetic Algorithms have been used in finance for years

“Genetic Algorithm Optimisation for Finance and Investment” by Robert Pereira is a good paper to explain fundamentals of financial genetic algorithms

This project is similar in that it uses a simple example to describe applied mathematics

Reading in Data

Source code is imported from the Internet

Data is parsed for keywords

Last Trade: “</small><big><b><span id="yfs_l10_ibm">

P/E :“</span>:</th><td class="yfnc_tabledata1">

Reading in Data (cont.)

Parsing data means source must be permanent

Source utilized is Yahoo! Finance

Snapshot of what the Yahoo! Finance interface looks like

http://finance.yahoo.com/q?s=IBM

Genetic Algorithm

User picks 3 stocks

Stocks are assigned “point value” based on 3 metrics

Metrics are P/E closeness to 16, yield, and diversification

Genetic Algorithm (cont.)

Random combinations of 8 shares amongst 3 stocks are generated

Combinations are ranked, and bottom half are eliminated

Top half are crossed over and re-ranked

Process continues until one solution remains

10/29/09

Class Structure

The Main method prompts for the stock symbols and sends them to the Rank class for data mining and generation

Rank class imports data, calculates it, and formats it

Main method receives the data and runs the genetic algorithm

Problems Run Into

How exactly to parse through all the data and find the relevant information

How to organize the program to make it less cluttered

How to form combinations that can be crossed over

Training the algorithm to return the ideal result every time

Solutions

After inspecting the source, relevant data is always followed by a specific tag

User prompted for information through JDialog boxes

Combinations turned into permutations, with 8 total shares

Ideal result can be pre-derived, and algorithm can then be coached to find the right solution.

How has my hypothesis changed?

Focus has shifted to applying genetic algorithms in finance

Concentration on educational purposes and helping explain how applied mathematics play a role in finance

Testing has shown that the best way to import data is through importing the source code and not trying to download files

Progress since 2nd Quarter

• Program now includes genetic algorithm

• Better class structure and code organization

• Several more complex data structures introduced

• Matrices

• Arrays

• Genetic Algorithm