A bandwidth selection for kernel density estimation of functions of random variables

Abstract In this investigation, the problem of estimating the probability density function of a function of m independent identically distributed random variables, g ( X 1 , X 2 ,…, X m ) is considered. The choice of the bandwidth in the kernel density estimation is very important. Several approaches are known for the choice of bandwidth in the kernel smoothing methods for the case m =1 and g is the identity. In this study we will derive the bandwidth using the least square cross validation and the contrast methods. We will compare between the two methods using Monte Carlo simulation and using an example from the real life.