基于多重尝试Metropolis算法的非线性DSGE模型贝叶斯推断

本文提出一种改进的多重尝试Metropolis算法，用于非线性动态随机一般均衡模型的贝叶斯参数估计和模型选择。多重尝试策略通过每次迭代抽取多个尝试点的方法来提高算法的混合速率，新方法中提出使用近似的方法提高计算速度，并通过接收概率调整偏差。数值实验表明新方法在相同的计算时间内具有更高的估计效率。最后，本文比较了具有不同货币政策设定的模型对中国经济数据的拟合效果，发现中国数据更加支持具有时变通胀目标的模型。

Bayesian Inference for Nonlinear DSGE Model via Multiple-try Metropolis Algorithm

This paper proposes a multiple-try-based Metropolis algorithm for Bayesian parameter estimation and model selection in nonlinear dynamic stochastic general equilibrium (DSGE) models. Multiple-try method generates multiple trial points in each iteration to achieve faster mixing rate. The new approach proposes to use approximation to save computational cost, the approximation bias is adjusted through the acceptance rate. Simulation results demonstrate effectiveness of the proposed algorithm. The algorithm is applied to estimate DSGE models with different monetary policy settings using the macroeconomic data of China. It shows that the data favors the model with time-varying inflation target.