1: \begin{proof}[Outline of the Proof]

2: Expansions of the risk functions corresponding to \eqref{thm1} when the distributions of $x(i)$, $i=1,\ldots,N$, and $y$ are the same

3: are obtained by \citet{K96biometrika} for curved exponential families by using differential geometrical notions

4: and by \citet{H98as} for general models under rigorous regularity conditions.

5: \citet{FKA04sjs} obtained several related results when when the distributions of $x(i)$, $i=1,\ldots,N$, and $y$ are different.

6: The expansion \eqref{thm1} is shown by lengthy calculations parallel to those in \citet{K96biometrika} and \citet{H98as}

7: by using the results such as \eqref{FKA-1}, \eqref{FKA-2}, and \eqref{FKA-3} obtained by \citet{FKA04sjs}.

8: \end{proof}