A parameter-varying universal portfolio using a cyclic constrained-search algorithm

Abstract

The performance of the Dirichlet universal portfolio can be improved by varying the parameter vector periodically after a fixed number of days, which is known as a trading period. After a trading period, a new parametric vector is chosen using a cyclic constrained-search algorithm that improve upon the wealth achieved in the previous period. The algorithm is run on some selected stock -data sets from the local stock exchange. Empirically, it is shown that higher returns in wealth are achieved for the parameter-varying universal portfolio over the constant-parameter universal portfolio.