POMDP and Its Application in Finance

1
POMDP and Its Application in Finance
CIEG Workshop

• Enlu Zhou, University of Maryland

April 23, 2007
2
What is POMDP ?

• Markov Decision Process (MDP)
• Playing chess
• Partially Observable Markov Decision Process
  (POMDP)
• Playing Poker

3
How to play a poker game ?

• Observe
• Update knowledge
• Look-ahead
• Act optimally (but myopically)

4
How to solve a POMDP ?

• The above method is one way to generate a
  suboptimal solution.
• Generally,
• Filtering (observe and update knowledge)
• Dynamic Programming (Look-ahead)
• Value Iteration, Policy Iteration, ...

5
POMDP in Finance

• Portfolio Optimization
• Hedging
• Problems arise naturally in many financial
  models, where some model coefficients
  (volatility, drift, ...) are stochastic and not
  observable.

6
About My Research

• Current
• Designing a novel algorithm for continuous-state
  POMDPs and developing theoretical analysis of
  error bounds
• running simulation on toy problems
• Future
• Apply it to finance problems

7
About this workshop

• My knowledge of cyberinfrastructure is almost
  none. /blush
• I hope at the end of the workshop,
  supercomputers dont seem so mysterious to me any
  more and I can do some programming on it.
• I also hope to learn from all of you about your
  research.

8

• Thank you !