RESEARCH INTEREST

Econometrics, Panel Data Models, Factor Models

PAPERS

Canonical Correlation-based Model Selection for the Multilevel Factor (with In Choi and Yongcheol Shin,  Journal of Econometrics)

Abstract: We develop a novel approach based on the canonical correlation analysis to identify the number of the global factors in the multilevel factor model. We propose the two consistent selection criteria, the canonical correlations difference (CCD) and the modified canonical correlations (MCC). Via Monte Carlo simulations, we show that CCD and MCC select the number of global factors correctly even in small samples, and they are robust to the presence of serially correlated and weakly cross-sectionally correlated idiosyncratic errors as well as the correlated local factors. Finally, we demonstrate the utility of our approach with an application to the multilevel asset pricing model for the stock return data in 12 industries in the U.S.

Generalised Canonical Correlation Estimation for the Multilevel Factor Model (with Yongcheol Shin, submitted)

Abstract: We develop a novel approach based on generalised canonical correlation (GCC) to consistently estimate the multilevel factor model from high dimensional panel data. Our approach is robust to the non-zero correlation between local factors of different blocks and is valid even when some blocks share the same local factors. Moreover, our approach achieves consistency without iteration and is easy to implement. As a by-product, the number of global factors can be determined by GCC. Asymptotic distributions of the factors and loadings are derived. Via Monte Carlo simulations, we show the dominating performance of GCC compared to existing approaches. Finally, we apply our approach to an empirical study of the housing market in the UK using data at the local authority level.

Estimation and Inference for a Multi-dimensional Panel Data Model with Multilevel Factors

Abstract: This paper considers a multi-dimensional panel data model with multilevel factors when the numbers of cross-sections and time observations are large. We develop a multilevel iterative principal component (MIPC) method for estimation by iteratively updating between the slope coefficients and factors, given one another. Under a finite number of blocks, our approach is able to produce consistent estimates of the slope coefficients, factors, and loadings. We also propose a model selection criteria based on the eigenvalue ratios to determine the numbers of factors. Given consistent factor estimates from each block, we apply the generalised canonical correlation (GCC) estimation to separately identifying the global and local factors. We show the consistency of our estimates and establish the asymptotic normality of the bias-corrected estimator for the slope coefficients. The Monte Carlo simulation demonstrates good finite sample performance of MIPC compared to IPC in the presence of multilevel factor structure. In an empirical application, our model is applied to an analysis of the energy consumption and economic growth nexus using a cross-country panel data categorised by regions.

Industry Factors in Asset Pricing Models

Abstract: Industry comovement of the stock returns has been widely documented in the literature, suggesting the existence of the industry factors. In this article, we construct different types of industry factors as pricing factors. In our in-sample and out-of-sample test, we find that the APT model with the market and industry factors estimated from a multilevel factor model achieves superior performance.