Identifying latent grouped patterns in panel data models with interactive fixed effects

We consider the estimation of latent grouped patterns in dynamic panel data models with interactive fixed effects. We assume that the individual slope coefficients are homogeneous within a group and heterogeneous across groups but each individual's group membership is unknown to the researcher. We consider penalized principal component (PPC) estimation by extending the penalized-profile-likelihood-based C-Lasso of Su, Shi, and Phillips (2016) to panel data models with cross section dependence. Given the correct number of groups, we show that the C-Lasso can achieve simultaneous classification and estimation in a single step and exhibit the desirable property of uniform classification consistency. The C-Lasso-based PPC estimators of the group-specific parameters also have the oracle property. BIC-type information criteria are proposed to choose the numbers of factors and groups consistently and to select the data-driven tuning parameter. Simulations are conducted to demonstrate the finite-sample performance of the proposed method. We apply our C-Lasso to study the persistence of housing prices in China's large and medium-sized cities in the last decade and identify three groups.