Essays on the econometric analysis of cross-sectional dependence

2019-04-09This dissertation contributes to the econometric analysis of cross-sectional dependence in the framework of factor, spatial and network analysis. Chapter 1 considers panel data models with cross-sectional dependence arising from both spatial autocorrelation and unobserved common factors. It derives identification conditions and proposes estimation methods that employ cross-sectional averages as factor proxies, including the 2SLS, Best 2SLS, and GMM estimations. The proposed estimators are robust to unknown heteroskedasticity and serial correlation in the errors, unrequired to estimate the number of unknown factors, and computationally tractable. I establish the asymptotic distributions of these estimators and compare their consistency and efficiency properties. An empirical application finds strong evidence of spatial dependence of real house price changes across 377 Metropolitan Statistical Areas in the US from 1975Q1 to 2014Q4. The results also reveal that population and income growth have significantly positive direct and spillover effects on house price changes. Chapter 2 (co-authored with M. Hashem Pesaran) considers production and price networks with unobserved common factors, and derives an exact expression for the rate at which aggregate fluctuations vary with the dimension of the network. It introduces the notions of strongly and weakly dominant and non-dominant units, and shows that at most a finite number of units in the network can be strongly dominant. The pervasiveness of a network is measured by the degree of dominance of the most pervasive unit in the network, and is shown to be equivalent to the inverse of the shape parameter of the power law fitted to the network outdegrees. New cross-section and panel extremum estimators for the degree of dominance of individual units in the network are proposed and their asymptotic properties investigated. An application to US input-output tables spanning the period 1972 to 2007 suggests that no sector in the US economy is strongly dominant. The most dominant sector turns out to be the wholesale trade with an estimated degree of dominance ranging from 0.72 to 0.82. Estimation and inference in the spatial econometrics literature are carried out assuming that the matrix of spatial or network connections have uniformly bounded absolute column sums in the number of cross-section units, n. In Chapter 3, M. Hashem Pesaran and I consider spatial models where this restriction is relaxed. We begin by extending the GMM estimator introduced by Lee (2007) and deriving the asymptotic properties. We then propose a Bias-Corrected Method of Moments (BMM) estimator that avoids the problem of weak instruments by self-instrumenting the spatially lagged dependent variable and correcting the bias of the moment conditions. Both estimators are consistent and asymptotically normal, depending on the rate at which the maximum column sum of the weights matrix rises with n. The estimation methods are applied to examine the inflation spillovers across more than 300 industries in the US from 1997 to 2014