Evaluation of stocahstic models via regularized information criteria

2015 Fiscal Year Final Research Report

• Project/Area Number: 25540013
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: Statistical science
• Research Institution: Kyushu University
• Principal Investigator: Nishii Ryuei 九州大学, マス・フォア・インダストリ研究所, 教授 (40127684)
• Project Period (FY): 2013-04-01 – 2016-03-31
• Keywords: モデル評価 / カルバック ライブラー情報量 / 決定係数 / 期待対数尤度 / 情報量基準

Outline of Final Research Achievements

The information criterion AIC proposed for evaluations of statistical models has contributed greatly to statistics and the application fields. AIC which is an unbiased estimate of the expected log likelihood, however, is not an absolute scale like the coefficient of determination in regression analysis. In this research, AIC-coefficient of determination is introduced based on the difference of AIC's of a current model and the simplest model for driving an absolute scale for the model evaluation. In addition, the heteroscedastic coefficient of determination is introduced in regression analysis with heteroscedasticity. Furthermore, a two stage procedure is introduced for model estimation and adaptive information criteria of regression models based on spatio-temporal data.

Free Research Field

統計モデリング