On the Coverage Bound Problem of Empirical Likelihood Methods For Time Series (1401.5002v2)

Abstract: The upper bounds on the coverage probabilities of the confidence regions based on blockwise empirical likelihood [Kitamura (1997)] and nonstandard expansive empirical likelihood [Nordman et al. (2013)] methods for time series data are investigated via studying the probability for the violation of the convex hull constraint. The large sample bounds are derived on the basis of the pivotal limit of the blockwise empirical log-likelihood ratio obtained under the fixed-b asymptotics, which has been recently shown to provide a more accurate approximation to the finite sample distribution than the conventional chi-square approximation. Our theoretical and numerical findings suggest that both the finite sample and large sample upper bounds for coverage probabilities are strictly less than one and the blockwise empirical likelihood confidence region can exhibit serious undercoverage when (i) the dimension of moment conditions is moderate or large; (ii) the time series dependence is positively strong; or (iii) the block size is large relative to sample size. A similar finite sample coverage problem occurs for the nonstandard expansive empirical likelihood. To alleviate the coverage bound problem, we propose to penalize both empirical likelihood methods by relaxing the convex hull constraint. Numerical simulations and data illustration demonstrate the effectiveness of our proposed remedies in terms of delivering confidence sets with more accurate coverage.