ABSTRACT
In the above discussion we assumed that the spectral density matrices f(>.) and g(>.) associated with TI 1 and TI2 are known. However they are never known in practice, and must be inferred from samples. If the training samples of II 1 and II2 are available, we can make estimators f(>.) and g(>.) for f(>.) and g(>.), respectively, from them. Then the plug-in version of Dn is given by
( 5.27) We choose rr, if DH > 0, and choose rr2 if DH :::; 0. The asymptotics of DH were illuminated in Kakizawa (1996).
