摘要：This paper investigates empirical models of the US short term interest rate. It
make use of a combination of classical non-parametric methods and of parametric bayesian
methods. In a fi rst step, it investigates the shape of drift and volatility functions using non
parametric tools. The paper then develops a bayesian approach to model selection based on
the minimisation of the Hellinger distance between the posterior predictive density of a
discretised model and a non-parametric estimation of the data density. A discretisation of
various parametric formulations are then estimated, ranging between constant elasticity of
variance to switching regimes.
