Estimation of unidentifiable finite mixture binary models and test statistics for specifying the number of components

2016 Fiscal Year Final Research Report

• Project/Area Number: 26780137
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Economic statistics
• Research Institution: Shinshu University (2016); Hiroshima International University (2014-2015)
• Principal Investigator: MASUHARA Hiroaki 信州大学, 学術研究院社会科学系, 准教授 (10419153)
• Project Period (FY): 2014-04-01 – 2017-03-31
• Keywords: 有限混合モデル / 識別性 / 検定統計量

Outline of Final Research Achievements

We analyze two problems in a finite mixture model in which two or more distributions are additively mixed. First, the finite mixture model can not be distinguished in a binary variable taking values of 0 or 1. We demonstrate that it can be identified in a simultaneous equation model with a continuous variable following the normal distribution. Second, the finite mixture models can not be tested how many distributions it is composed of. Since the likelihood increases endlessly due to over-identification, we propose a Vuong test using Laplace approximation. In Monte-Carlo simulation, Laplace approximation is preferable.

Free Research Field

応用計量経済学