High-Dimensional Macroeconomic Forecasting Using Message Passing Algorithms
提出两种方法：将时变系数回归模型转化为高维静态回归问题，并用贝叶斯分层先验压缩系数；引入因子图和消息传递框架设计高效估计算法，用于预测美国通胀。
This paper proposes two distinct contributions to econometric analysis of\nlarge information sets and structural instabilities. First, it treats a\nregression model with time-varying coefficients, stochastic volatility and\nexogenous predictors, as an equivalent high-dimensional static regression\nproblem with thousands of covariates. Inference in this specification proceeds\nusing Bayesian hierarchical priors that shrink the high-dimensional vector of\ncoefficients either towards zero or time-invariance. Second, it introduces the\nframeworks of factor graphs and message passing as a means of designing\nefficient Bayesian estimation algorithms. In particular, a Generalized\nApproximate Message Passing (GAMP) algorithm is derived that has low\nalgorithmic complexity and is trivially parallelizable. The result is a\ncomprehensive methodology that can be used to estimate time-varying parameter\nregressions with arbitrarily large number of exogenous predictors. In a\nforecasting exercise for U.S. price inflation this methodology is shown to work\nvery well.\n