Statistical inferences by semiparametric empirical likelihood methods

Hitomi和Nishiyama(2005)通过蒙特卡罗模拟研究了经验似然估计的偏差，其中标准线性模型设置被用作DGP，并与ET和GMM估计进行了比较。通常情况下，如果估计并插入干扰参数，感兴趣的参数的渐近方差往往很大，但也有相反的情况。它们提供了发生跳跃的充要条件。Nishiyama和Robinson(2005)得到了半参数平均导数的一个有效的二阶Edgeworth展开式，并检验了它与Bootstrap分布的差异。Nishiyama，Liu和Sueishi(2005)在矩条件下比较了包含非参数泛函的半参数模型的EL和GMM估计量，Morimune和Hoshino提出在数据包含异类观测的情况下使用Bootstrap。研究了Cressie-Read功率发散检验的小样本功率特性。

Hitomi studied the bias of the empirical likelihood estimators by a Monte Carlo simulation, where standard linear model setup is used as the DGP, also comparing with the ET and GMM estimators.Hitomi and Nishiyama (2005) investigated how a puzzling situation in terms of asymptotic variance occurs in semiparametric settings. It is normally the case that if nuisance parameters are estimated and plugged in, the asymptotic variance of the parameters of interest tends to be large, but there are cases the opposite is true. They provided a necessary and sufficient condition when it happens.Nishiyama and Robinson (2005) obtained a valid Edgeworth expansion of order two for the semiparametric averaged derivatives and checked the difference from the Bootstrap distribution.Liu and Nishiyama studied the performance of empirical likelihood estimators of diffusion processes with jumps by simulation. It is shown that EL performs better than the GMM for some parameters.Nishiyama, Liu and Sueishi (2005) compared EL, GMM estimators for semiparametric models which include nonparametric functionals in the moment condition.Morimune and Hoshino proposed to use Bootstrap when data includes heterogeneous observations.Tanizaki et.al. studied small sample power properties of Cressie-Read power divergence test.