Density estimation based on pointwise mutual information

he purpose of this article is to develop a new bivariate density estimation method based on the decomposition of joint density into pointwise mutual information and marginal densities. The pointwise mutual information and product of marginal densities are estimated by bivariate kernel density estimators with shuffled data. Our method is defined as a product of the marginal densities and pointwise mutual information. Monte-Carlo simulations indicate that this estimation method provides good finite sample performance for weak dependent data.