Bayesian econometric analysis of semiparametirc model

在本研究项目中，我们构建了金融时间序列、宏观经济时间序列和社会经济面板数据的贝叶斯计量模型，并提出了基于马尔可夫链蒙特卡罗(MCMC)方法的有效估计方法。Omori考虑了具有杠杆效应、跳跃和重尾误差分布的随机波动率模型，提出了两种基于MCMC的高效估计方法。进一步，将模型推广到多元因素随机波动率模型。石田对日经225的超高频数据计算的已实现波动率进行了实证研究，发现已实现波动率的时间序列存在均值回归、沿记忆特性和时变波动性。Wago估计了J…犯罪率面板数据的时空模型更多的APAN使用贝叶斯方法并进行了模型选择。渡边对已实现波动率(称为资产收益波动率的非参数估计器)进行了文献调查，并表明已实现波动率对预测未来的波动率很有用。日经225指数的无模型隐含波动率是使用期权价格进行分析的。此外，Watanabe使用日收益估计ARCH模型和使用已实现波动率估计ARFIMAX模型(具有长记忆性)，并比较了两种模型在预测波动率和评估VaR方面的性能。小泉考虑了结合Probit模型和Poisson回归模型的内生切换模型的替代规格，并利用MCMC推导了有效的估计方法。在此基础上，提出了基于数据增强的潜在伽马变量随机前沿模型的MCMC估计方法。OGA使用贝叶斯方法将马尔可夫切换模型应用于不同的综合指数，并提出了一个模型来检测日本经济周期中衰退和扩张的非对称性。较少

In this research project, we construct Bayesian econometric models for financial time series, macro-economic time series and socio-economic panel data, and proposed efficient estimation methods using Markov chain Monte Carlo (MCMC) methods. Omori considered stochastic volatility models with leverage effects, jumps and heavy-tailed error distributions and proposed two highly efficient estimation methods using MCMC. Further, the models are extended to the multivariate factor stochastic volatility models. Omori also derived the effective MCMC acceleration step to improve the convergence rate of Gibbs sampler for the well-known sample selection models.Ishida conducted empirical studies of the realized volatilities computed from ultra high frequency data for Nikkei 225, and found that there exist a mean reversion, along memory property and a time-varying volatility in the time series of realized volatilities.Wago estimated the spatio-temporal model for the panel data of the crime rates in J … More apan using Bayesian approach and conducted model selections. Watanabe conducted a literature survey on the realized volatility (known as a nonparametric estimator of volatility of asset returns) and showed that realized volatilities are useful to predict future volatilities. Model-free implied volatilities for the Nikkei 225 stock index are analyzed using option prices. Furthermore, Watanabe estimated ARCH models using daily returns and ARFIMAX models (with long memory property) using realized volatilities, and compared the model performances in predicting the volatilities and the evaluation of VaR.Kozumi considered alternative specifications of endogenous switching models which combine probit models and Poisson regression models, and derived efficient estimation methods using MCMC. Further, he proposed the MCMC estimation method for stochastic frontier model with latent gamma variables using the data augmentation. Oga applied Markov switching model to the differenced composite indices using Bayesian approach, and also proposed a model to detect asymmetry in recessions and expansions in business cycles in Japan. Less