On the Effect of Prior Assumptions in Bayesian Model Averaging with Applications to Growth Regression

Ley, Eduardo ; Steel, Mark F. J.

World Bank, Washington, DC

关键词: AREA;    BAYES FACTOR;    BINOMIAL DISTRIBUTION;    CLASSIFICATION;    COVARIANCE;   

DOI  :  10.1596/1813-9450-4238
RP-ID  :  WPS4238

学科分类：社会科学、人文和艺术（综合）

来源: World Bank Open Knowledge Repository

【 摘 要 】

This paper examines the problem ofvariable selection in linear regression models. Bayesianmodel averaging has become an important tool in empiricalsettings with large numbers of potential regressors andrelatively limited numbers of observations. The paperanalyzes the effect of a variety of prior assumptions on theinference concerning model size, posterior inclusionprobabilities of regressors, and predictive performance. Theanalysis illustrates these issues in the context ofcross-country growth regressions using three datasets with41 to 67 potential drivers of growth and 72 to 93observations. The results favor particular prior structuresfor use in this and related contexts.