Application of KDE in finance

Professionals in quantitative finance are often faced with a difficult choice when having to estimate the probability of a financial event: either taking frequencies of observed time series as best estimators, or fitting a parametric distribution. In this paper, we consider a refinement of this first option, which retains the major advantage of not making constraining parametric assumptions, while mitigating common drawbacks: Kernel Density Estimation. We show how the method can be applied for the purpose of estimating Value-at-Risk. We show how this technique can be applied to quantile estimation, and present a practical application for the estimation of volatile balances in the modeling of Non-Maturing Deposits, both for current and savings accounts.Antoine Gustin, Dr. Christopher Van Weverberg and Pr. François Ducuroir