Postdoc - 02, Linda Chamakh (polytechnique): Explicit Mean-Embeddings for Financial Portfolio

Minimum divergence portfolios have been recently introduced to find portfolio allocation matching investor distribution preference by minimising the divergence between the investor's target distribution and the portfolio empirical distribution. We tackle this problem taking the Maximum Mean Discrepancy as divergence function. In this presentation, we present our main results, namely the analytical computability of the MMD for various pairs of (kernel, target distribution) functions and the fact that such analytical knowledge leads to faster convergence of MMD estimators.