Portfolio selection using nearest neighbor

Description

Implement the portfolio selction defined by Gyorfi. At each instant in time, look back for a set of nearest neighbors, then find the best CRP for the time steps following the nearest neighbors and use the corresponding weights for the next time step. This is done for a set of parameters, defining so-called experts. The final performance is the (weighted) average across all experts.

Usage

1
nn.gyorfi(x, pls, ks, qkl=NULL)

Arguments

x

Time series of relative prices

pls

pls is either a scalar or a sequence. If its a scalar, it is the value L as defined in Gyorfi article, and the pl sequence is 0.02 + 0.05 * (l-1)/(L-1), l in 1:L. Otherwise pls is the sequence of pl values itself.

ks

ks is either a scalar or a sequence. If its a scalar, it is the value K as defined in Gyorfi article, and the k sequence is 1:K. Otherwise ks is the sequence of k values itself.

qkl

A priori weights for the experts, this is a matrix (K,L) that should sum to 1. The NULL default results in an uniform loading.

Details

Straightforward implementation of the nearest neighbors portfolio algorithm.

Value

Return a list containing in order the set of weights and wealth evolution corresponding to all combinations of pl and k values.

Author(s)

Marc Delvaux

References

L. Gyorfi, F. Udina and H. Walk, "Nonparametric nearest neighbor based empirical portfolio selection strategies", Statistics & Decisions, 26:145-157, 2008.