Estimating international risk-sharing in the presence of endogeneity.
Abstract
Over six chapters, this thesis explores how to estimate risk-sharing when output is
not exogenous. The thesis starts with a survey of the current literature and how it
estimates risk-sharing. This survey is then followed by risk-sharing estimations based
on a panel of 24 countries over the period of 1970-2007. The estimation approaches
applied include the literature's Classical and Level approaches, as well as alternative
estimation approaches that provide robust parameter estimates when the literature
commonly assumed output exogeneity is dropped. These alternative estimators
consist of procedures using instrumental variables ranging from first differenced two-stage
least squares, a dynamic generalized method of moments estimation, and an
instrumental variables estimation using an instrument derived from a structural
vector autoregressive model. Also, a Monte Carlo Simulation is undertaken to show
the severity of the bias inherent in the Classical estimation method, as well as
to show the performance of the proposed alternative methods. When output is
endogenous, the Classical estimation method is found to underestimate risk-sharing,
while the best performing alternative approaches are concluded to be the Level
approach and the instrumental variables estimation approach using an instrument
derived from a structural vector autoregressive model. This thesis contributes to the
risk-sharing literature by discussing and quantifying the bias the Classical estimation
approach suffers from due to output endogeneity. It also contributes by adapting
estimation methods from other fields that allow consistent estimations of risk-sharing
parameters in the presence of endogeneity bias, and by analyzing the performance
of these asymptotic panel estimators in the specific context of the panel dimensions
commonly found in the risk-sharing literature.