贝叶斯分位数结构方程模型的变分近似推断OA北大核心CHSSCDCSSCICSTPCD

Structural equation model(SEM)is an important tool for analyzing the relationship between latent variables.Al-though quantile SEM(QSEM)provides a comprehensive analysis of the relationship between latent and observed variables at dif-ferent quantiles,the computational efficiency based on MCMC estimation is not high.In view of this,the paper mainly discusses variational inference for Bayesian QSEM(BQSEM).A QSEM based on regularized priors is constructed,and the variational poste-rior estimates of the QSEM and the regularized QSEM(RQSEM)are given(MFVB_QSEM and MFVB_RQSEM),respectively.The variance estimation of posterior distribution is improved by using a Bootstrap method.Simulation results show that this variational inference is significantly faster than MCMC estimation while providing reliability inference.Furthermore,when it is applied to study the determinants of capital structure in Chinese listed companies,the research results show that the determinants of the capi-tal structure of Chinese listed companies exhibit different impacts across different quantiles,with profitability and liquidity being the most important determinants.