[Abstract]
　We derive the finite sample bias and mean squared error of the fully aggregated estimator (FAE) for the stationary AR(1) model with intercept, which is proposed by Han, Phillips, and Sul (2011, Econometric Theory). Our analytical results show why the FAE is less biased than the ordinary least square estimator in finite sample case and is not biased by non-normality of error distribution and by intercept term at least O(1/T ), where T is sample size. We also propose a second order unbiased FAE using the analytical result. Finally, we examine the Monte Carlo simulation and show that it is consistent with the theoretical results.

[Abstract]
　This paper analyzes impacts of an outdated but sticky economic circumstance in the Japanese private nursing home market. Under this circumstance, homes cover all longevity risks of the residents. To measure its impacts, I compare consumers’ lifetime payments with and without this circumstance. This analysis provides implications for consumer welfare as a policy evaluation and longevity risk premium that homes assume. For the prediction analysis, I construct a structural econometric model that fits industrial organization of the nursing home market. Then, I propose a flexible prediction technique using a nonparametric Bayesian approach. My prediction analysis finds an excess payment under the circumstance that can be compensate only if a consumer live longer than thirty years in a nursing home. This results indicate rich policy implications for sustainable policy of elderly care in Japan.

要旨：City size distributions are known to approximately follow Zipf’s law, or more generally a common power law (CPL), across many countries. The most popular explanation for this regularity has been given by random growth models (e.g., Gabaix, 1999; Duranton, 2006, Rossi-Hanseberg and Wright, 2007), in which the power law results from the common iid growth rate of all cities. Since each subsample of cities has exactly the same properties, it also follows the same power law asymptotically without any role played by geography. The objective of the present paper is to establish counter-evidence from the US and Japan that the geography does indeed matter for CPL to hold among different regions and countries. Specifically, we show that random subsamples of cities taken from geographically connected regions are significantly more consistent with CPL than fully random subsamples. Furthermore, the sets of cities in economically meaningful regions (based on interregional trade) are even more significantly consistent with CPL. Finally, we present preliminary evidence that our finding is consistent with a class of central place models (Hsu, 2013, Akamatsu et al., 2013) in which sizes and locations of cities are determined through the spatial coordination of many industries subject to different degrees of scale economies.

[Abstract]
　Factor loading structural change tests proposed by Breitung and Eickmeier (2011) exhibit nonmonotonic power when a portion of the factor loadings have structural changes occuring at common dates. To investigate this phenomenon, we develop spurious factor representations of the factor model with m common breaks. We show that, under a local alternative asymptotic framework in which the magnitudes of breaks shrink at rate max{N,T}^(1/2), the original factor space is consistently estimated and the tests have power. However, if we apply the fixed alternative asymptotic framework in which the break magnitudes are fixed, then the common breaks in the factor loadings are considered as spurious factors with time invariant loadings, hence the tests lose power. Monte Carlo simulations clearly illustrate these findings. This paper casts a new light on the well known nonmonotonic power problem in structural change testing since it now occurs by identification failure between the common factors and the factor loadings. Furthermore, none of the SupWald, the SupLR, and the SupLM structural change tests can avoid this problem.