TY - JOUR
T1 - Stochastic generalized method of moments
AU - Yin, Guosheng
AU - Ma, Yanyuan
AU - Liang, Faming
AU - Yuan, Ying
N1 - Funding Information:
We thank the referees, associate editor, and editor for many insightful suggestions which strengthened the work immensely. Yin’s research was supported by a grant from the Research Grants Council of Hong Kong, Ma’s research was supported by a US NSF grant, Liang’s research was supported by grants from US NSF (DMS-1007457 and CMMI-0926803) and King Abdullah University of Science and Technology (KUS-C1-016-04), and Yuan’s research was supported by a U.S. National Cancer Institute R01 grant (R01CA154591-01A1).
PY - 2011
Y1 - 2011
N2 - The generalized method of moments (GMM) is a very popular estimation and inference procedure based on moment conditions. When likelihood-based methods are difficult to implement, one can often derive various moment conditions and construct the GMM objective function. However, minimization of the objective function in the GMM may be challenging, especially over a large parameter space. Due to the special structure of the GMM, we propose a new sampling-based algorithm, the stochastic GMM sampler, which replaces the multivariate minimization problem by a series of conditional sampling procedures. We develop the theoretical properties of the proposed iterative Monte Carlo method, and demonstrate its superior performance over other GMM estimation procedures in simulation studies. As an illustration, we apply the stochastic GMM sampler to a Medfly life longevity study. Supplemental materials for the article are available online.
AB - The generalized method of moments (GMM) is a very popular estimation and inference procedure based on moment conditions. When likelihood-based methods are difficult to implement, one can often derive various moment conditions and construct the GMM objective function. However, minimization of the objective function in the GMM may be challenging, especially over a large parameter space. Due to the special structure of the GMM, we propose a new sampling-based algorithm, the stochastic GMM sampler, which replaces the multivariate minimization problem by a series of conditional sampling procedures. We develop the theoretical properties of the proposed iterative Monte Carlo method, and demonstrate its superior performance over other GMM estimation procedures in simulation studies. As an illustration, we apply the stochastic GMM sampler to a Medfly life longevity study. Supplemental materials for the article are available online.
KW - Generalized linear model
KW - Gibbs sampling
KW - Iterative monte carlo
KW - Markov chain monte carlo
KW - Metropolis algorithm
KW - Moment condition
UR - http://www.scopus.com/inward/record.url?scp=80053377942&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=80053377942&partnerID=8YFLogxK
U2 - 10.1198/jcgs.2011.09210
DO - 10.1198/jcgs.2011.09210
M3 - Article
C2 - 22375093
AN - SCOPUS:80053377942
SN - 1061-8600
VL - 20
SP - 714
EP - 727
JO - Journal of Computational and Graphical Statistics
JF - Journal of Computational and Graphical Statistics
IS - 3
ER -