In this article, we propose a novel, multilevel, dynamic factor model, to determine endogenously clustered regions for the investigation of regional clustering and synchronization of provincial business fluctuations in China. The parameter identification and model estimation was conducted using the Markov Chain Monte Carlo method. We then conducted an empirical study of the provincial business fluctuations in China (31 Chinese provinces are considered except Hong Kong, Macau, and Taiwan due to the data unavailability), which were sampled from January 2000 to December 2015. Our results indicated that these provinces could be clustered into four regions: leading, coincident, lagging, and overshooting. In comparison with traditional geographical divisions, this novel clustering into four regions enabled the regional business cycle synchronization to be more accurately captured. Within the four regional clusters it was possible to identify substantial heterogeneities among regional business cycle fluctuations, especially during the periods of the 2008 financial crisis and the 'four-trillion economic stimulus plan'.

This paper considers a semiparametric panel data model with heterogeneous coefficients and individual-specific trending functions, where the random errors are assumed to be serially correlated and cross-sectionally dependent. We propose mean group estimators for the coefficients and trending functions involved in the model. It can be shown that the proposed estimators can achieve an asymptotic consistency with rates of root−N T and root−N T h, respectively as (N, T ) → (∞, ∞), where N is allowed to increase faster than T . Furthermore, a statistic for testing homogeneous coefficients is constructed based on the difference between the mean group estimator and a pooled estimator. Its asymptotic distributions are established under both the null and a sequence of local alternatives, even if the difference between these estimators vanishes considerably fast (can achieve root-N T 2 rate at most under the null) and consistent estimator available for the covariance matrix is not required explicitly. The finite sample performance of the proposed estimators together with the size and local power properties of the test are demonstrated by simulated data examples, and an empirical application with the OECD health care expenditure dataset is also provided.

This paper considers a semiparametric panel data model with heterogeneous coefficients and individual-specific trending functions, where the random errors are assumed to be serially correlated and cross-sectionally dependent. We propose mean group estimators for the coefficients and trending functions involved in the model. It can be shown that the proposed estimators can achieve an asymptotic consistency with rates of root−N T and root−N T h, respectively as (N, T) → (∞, ∞), where N is allowed to increase faster than T. Furthermore, a statistic for testing homogeneous coefficients is constructed based on the difference between the mean group estimator and a pooled estimator. Its asymptotic distributions are established under both the null and a sequence of local alternatives, even if the difference between these estimators vanishes considerably fast (can achieve root-N T 2 rate at most under the null) and consistent estimator available for the covariance matrix is not required explicitly. The finite sample performance of the proposed estimators together with the size and local power properties of the test are demonstrated by simulated data examples, and an empirical application with the OECD health care expenditure dataset is also provided.