Portfolio optimization using variable-size genetic network programming

By Fabian Köhnke and Christian Borgelt

Outline

  1. Genetic Network Programming
  2. Variable-Size GNP
  3. Example Simulation
  4. Simulation Study

Genetic Network Programming (GNP)

G for Genetic

N for Network

N for Network

P for Programming

Let's run a GNP...

Let's run a GNP...

Variable-size genetic network programming

Challenges

Challenges

Challenges

Challenges

Challenges

Challenges

The novel Mutation Operator

The novel Mutation Operator

The novel Mutation Operator

Simulation Results

Simulation Study of 7 different GNP runs with 24 validations each and the shown combination of judgment and processing nodes. The values show the results without the new mutation operator.
JN:1 PN:2 JN:5 PN:5 JN:10 PN:10 JN:15 PN:15 JN:25 PN:25 JN:50 PN:50 JN:100 PN:100
Profit 41.99 28.32 45.15 69.65 30.65 49.25 35.39
Mean Nodes 3.00 10.00 20.00 30.00 50.00 100.00 200.00

Results with the new mutation operator.
JN:1 PN:2 JN:5 PN:5 JN:10 PN:10 JN:15 PN:15 JN:25 PN:25 JN:50 PN:50 JN:100 PN:100
Profit 47.99 32.65 97.85 78.61 55.25 46.05 50.11
Mean Nodes 22.08 27.06 32.48 31.88 24.50 58.28 70.79

Network Example

Node Numbers

Future Research

Appendix

Faulty Crossover

Mutation

Algo