【1108期】 11月24日计量经济学学术讲座:Sparse Weak Factor Models: Identification, Estimation and Inference (章永辉,副教授,中国人民大学)

时间:2022-11-15

【主题】Sparse Weak Factor Models: Identification, Estimation and Inference

【报告人】章永辉(副教授,中国人民大学经济学院) 

【时间】20221124日周四下午14:30-16:00

【地点】腾讯会议,会议ID603 252 130,密码:240626

语言】中文

【摘要】In this paper, we study the identification, estimation and inference of weak factor models, where the factor loadings are sparse among individuals and different factors have different degrees of loading sparsity. For a strong factor model, it is well established that the principal component (PC) estimate can provide consistent estimators for factors and factor loadings only up to some rotation. One may take it for granted that rotational indeterminacy may render failure of identification for weaker factors with mixed strengths. However, we show that the relevant rotation matrix in estimating factor loadings has an approximate block upper triangular structure. This finding has three important implications for the choice of estimation methods: (i) the PC estimate for a weaker factor is free of the contamination of relatively stronger factors and then keeps an approximately sparse structure automatically; (ii) there is no need to use regularization techniques such as -norm or -norm penalties to purse a sparse estimate for factor loadings; (iii) there is no need to use pre-screening methods to choose relevant individuals before estimating the factor model. These finding justifies the use of PC method in estimating weak factor models. To determine the number of factors in sparse weak factor models, we propose a criterion based on singular value thresholding as Bai and Ng (2019) and establish its consistency. In contrast with Bailey, Kapetanios and Pesaran (2021) where only the strength of the strongest factor can be identified and estimated, we can provide consistent estimators of factor strengths for all weak factors based on the PC estimator. We further provide valid confidence intervals for robust diffusion index where a forecast is constructed by a time series regression augmented by possible weak factors. We demonstrate the finite sample performance of our estimator via Monte Carlo simulations. The empirical application shows our method is appealing in forecasting macroeconomic indicators from FRED-QD.

报告人简介】章永辉,中国人民大学经济学院计量与数量经济学系副教授。2013年获得新加坡管理大学经济学博士学位。2013年至今在中国人民大学经济学院工作。主要研究领域为理论计量经济学,半参数/非参数计量,面板数据模型等。学术论文发表在Journal of EconometricsEconometrics Journal, Econometric Reviews等学术期刊上。


返回原图
/