An Edgeworth expansion for symmetric statistics (Q1359428)

Let \(X_1,X_2, \dots,X_n\) be i.i.d. random variables taking values in a measurable space and \(T=t(X_1,X_2, \dots,X_n)\) be the symmetric statistic. Write \[ T=ET+ \sum_{i=1}NT_i\;+ \sum_{1\leq i<j\leq n} T_{i,j} +R \] where \(T_i=E (T\mid X_i)-ET\), \(T_{i,j}= E(T\mid X_i,X_j)- E(T\mid X_i)- E(T\mid X_j)+ET\). The authors prove that for \(T\) an Edgeworth expansion with remainder \(O(N^{-1})\) is valid under the Cramér condition \[ \limsup_{| t|\to \infty} \bigl| E\exp\{it\sqrt NT_1\}\bigr|<1 \tag{C} \] and some moment conditions for all parts of the statistic \(T\). They also give an example to show that under the condition (C) and arbitrarily restrictive moment conditions it is generally impossible to obtain an Edgeworth expansion with remainder term \(o(N^{-1})\), even for \(U\)-statistics of order 2.