Practical estimation of multivariate densities using wavelet methods

This paper describes a practical method for estimating multivariate densities using wavelets. As in kernel methods, wavelet methods depend on two types of parameters. On the one hand we have a functional parameter: the wavelet Ø (comparable to the kernel K ) and on the other hand we have a smoothing parameter: the resolution index (comparable to the bandwidth h ). Classically, we determine the resolution index with a cross‐validation method. The advantage of wavelet methods compared to kernel methods is that we have a technique for choosing the wavelet Ø among a fixed family. Moreover, the wavelets method simplifies significantly both the theoretical and the practical computations.