ABSTRACT
 
This talk, although directly targets at testing continuous-time diffusion 
models, has a running theme on how to validate a statistical model in 
general (what to look for and how to formulate a test procedure). The 
special characteristics of diffusion models, dependent and continuous in 
time, provide some spice and excitement, but more importantly allow us to 
wonder beyond the specific task.
 
Ait-Sahalia (1996, Reviews of Financial Studies) proposed a test for 
diffusion models, which is based on a kernel estimator of the stationary 
density function. Pritsker (1998, RFS) conducted a detailed follow-up study 
and finds severe size distortion of the test proposed by Ait-Sahalia and in 
particular 2755 years of data are required in order for the kernel density 
estimator to attain a level of accuracy achieved with 22 years of 
independent data. The kernel method has been taken to be responsible for 
the problem, which leads to a belief that the kernel method is incapable of 
capturing the dependence induced by a continuous-time diffusion model!
 
I will show in the talk that the kernel method is still an effective method 
after some alterations and additions.
 
COFFEE: 3:45 p.m., 104 Snedecor Hall
 
Seminar schedules and abstracts are available via WWW: 
http://www.stat.iastate.edu/