Factor IV Estimation in Conditional Moment Models with an Application to Inflation Dynamics
提出一种新的集成条件矩估计量，直接利用因子条件矩限制，无需先参数化或估计这些限制，适用于工具变量数可能超过样本量的时间序列数据，并用于估计美国新凯恩斯菲利普斯曲线。
Abstract In a conditional moment model, we develop a new integrated conditional moment (ICM) estimator which directly exploits factor-based conditional moment restrictions without having to first parametrize, or estimate such restrictions. We focus on a time series framework where the large number of available instruments and associated lags is driven by a relatively small number of unobserved factors. We build on the ICM principle originally proposed by Bierens (1982) and combine it with information reduction methods to handle the large number of potential instruments which may exceed the sample size. Under the maintained validity of the true factors, but not that of observed instruments, and standard regularity assumptions, our estimator is consistent, asymptotically normally distributed, and easy to compute. In our simulation studies, we document its reliability and power in cases where the underlying relationship between the endogenous variables and the instruments may be heterogeneous, non-linear, or even unstable over time. Our estimation of the New Keynesian Phillips curve with U.S. data reveals that forward- and backward-looking behaviors are quantitatively equally as important, while the driver’s role is nil.