Boxiong Wang

This research project uses linked New Zealand administrative data to study how major disasters reshape residential mobility, population distribution, and patterns of adjustment over time. Current work focuses on the Christchurch earthquakes, including migration out of and into affected areas, the persistence of relocation, and where people move after leaving Christchurch. The project is designed as a broader research program that can support a series of studies on post-disaster adaptation, resilience, and recovery.

Emergency administrative restraint is often interpreted as evidence of government capacity and fiscal resilience. We develop a sequential model in which implementation capacity is accumulated before a shock, while visible shortcuts and unfunded recovery liability are determined afterward. When the capacity gap exceeds a bailout-proof buffer, a government can use fewer visible shortcuts yet remain less fiscally resilient. We study this diagnostic problem across Chinese prefectures. Objective precipitation disasters provide external administrative stress, and we estimate how the causal response to that stress varies across predetermined administrative-state measures. Purchaser-network breadth and supplier-network breadth exhibit the most consistent positive shock-response heterogeneity; routine-procurement continuity and workflow stability behave differently, revealing multidimensional observable state. Conditional on physical burden and visible emergency response, the state measures add information for expenditure growth and budget imbalance under the primary finite-cluster inference, although conservative wild-cluster sensitivity weakens precision. Official disaster-support records validate realized upper-level fiscal support as a real institutional margin. The administrative-state measures are not structural capacity: the design identifies heterogeneous causal responses to external stress, not the effect of changing latent implementation capacity.

Banks monitor all credits but devote intensive independent review to only part of the portfolio. We study how improvements in review precision affect aggregate credit information when the bank chooses which credits receive intensive review. For every fixed allocation, greater precision improves inference about a common credit condition. Endogenous deployment can overturn this benchmark. Better review changes the private value of resolving losses for workout decisions and can redirect scarce review from a credit with high systematic loading to one with lower loading. Information about the same common condition then falls, even though each fixed allocation becomes more informative. Exposure-specific risk allows private decision value to diverge from common-state information value. We give primitive sufficient conditions for the target reversal and establish the same ranking pattern in an open four-exposure, two-slot environment. At the first information-reducing reallocation, the more informative plan remains socially preferred, and a deployment-sensitive score implements it locally.

Service-quality systems often evaluate performance only for requests that produce completed service. When completion is endogenous, this convention can make the monitored metric improve while request-level access deteriorates. We study this selective-observability problem in wheelchair-accessible ride-hailing. We combine public New York City completed-trip records with regulator-held monthly request-funnel counts, linking conditional response time to the share of submitted requests that reach completion. For Uber, mean completed response time falls from 5.482 to 4.952 minutes and the completed within-10 share rises from 90.124% to 94.099% between January–March and August–December 2025, while completion falls from 90.857% to 77.678%. A minimal model shows why a tighter completion-conditioned target can induce both more service effort and a lower completion cutoff, and why a completion guardrail constrains the latter margin when it binds. A nested-information result then formalizes when a regulator should move from inexpensive completion-only monitoring to aggregate-funnel review and, if decision losses justify the cost, to rule-complete records. The evidence does not identify the 2025 rule as the cause, operator intent, or official compliance. It shows that completion-only and joint wait-completion monitors can produce opposite operational assessments. Conditional targets should therefore be paired with extensive-margin monitoring when completion deterioration could materially change the decision.