TITLE:     Universal Portfolios in the Stock Market

 

SPEAKER:   David Julian [Stanford University)

 

DATE:      2-3 P.M., Tuesday, January 21, 2003

 

LOCATION:  Half Dome, 3L (PA)

 

HOST:      Vinay Deolalikar

 

-----------------------------------------------------------------------

 

ABSTRACT:

In data compression there are on-line universal algorithms, such as Lempel-Ziv, which compress a source with an unknown distribution from a family of distributions (almost) as well as if the distribution were known. Similarly, universal portfolios generate a return (almost) as large as the best return from a family of constant rebalanced portfolio strategies. Universal portfolios were first proposed by Tom Cover in his 1991 paper "Universal Portfolios". Since then a number of theoretical properties have been found, including worst case lower bounds by Erik Ordentlich and Tom Cover.  

In this talk we will review the universal portfolio setup and some of the key properties. We will then compare the theoretical and empirical performance based on back-testing from actual stock market data. This will highlight several properties of the universal portfolio and lead to refinements in some theoretical expressions.  The back-testing results will also show that in some situations the universal portfolio is capable of explosive performance. However, the computational complexity is one of the obstacles to implementing the universal portfolio in practice. So we will conclude by exploring some ways to more efficiently compute the universal portfolio.