Optimal taxation in a large-scale overlapping generations economy with both idiosyncratic and aggregate risks Hangjun He

Format:

Book

Thesis/Dissertation

Author/Creator:

He, Hangjun, author.

Contributor:

University of Pennsylvania. Applied Mathematics and Computational Science., degree granting institution.

Language:

English

Subjects (All):

Applied mathematics.

0364.

0511.

0501.

Local Subjects:

Applied mathematics.

0364.

0511.

0501.

Genre:

Academic theses

Physical Description:

1 online resource (99 pages)

Contained In:

Dissertations Abstracts International 87-12A

Place of Publication:

Ann Arbor : ProQuest Dissertations and Theses, 2026

Language Note:

English

Summary:

The problem of optimal taxation becomes substantially more difficult as one moves to richer fiscal instruments and more realistic macroeconomic environments. This dissertation studies optimal taxation in a large-scale overlapping-generations economy with both idiosyncratic and aggregate risks, allowing for flexible nonlinear taxes and transfers defined on multivariate household observables. The analysis makes two contributions. The first contribution is computational. We develop an analytical characterization of the marginal steady-state economy with respect to fiscal policy parameters and use it to compute the gradient of social welfare. The key difficulty is that, in an economy with aggregate shocks, household value and policy functions depend not only on individual states but also on equilibrium aggregate dynamics, which themselves depend on the cross-sectional distribution of households. We show how this fixed-point structure can be differentiated by combining a backward recursion for marginal household decisions and value functions with a forward recursion for the marginal population distribution. A central innovation is to treat the coefficients in the Krusell--Smith forecasting rule as differentiable functions of the fiscal policy parameters. This makes it possible to recover the general-equilibrium effect of policy changes on household expectations and thereby construct analytical gradients in an OLG economy with aggregate risk. The resulting method supports gradient-based optimization and is implemented with parallel computing techniques to handle the model's high-dimensional state space. The second contribution is theoretical. We develop a framework of sufficient variables for taxation. Because observable tax instruments are often functionally dependent in equilibrium, a tax system defined on a high-dimensional policy domain may admit an equivalent representation on a lower-dimensional domain. We show that the set of sufficient variables depends both on the underlying economic environment and on the class of tax functions under consideration. In the example economy, these results imply that some tax instruments are redundant once equilibrium relationships among states, decisions, and income realizations are taken into account. Quantitatively, the dissertation shows that analytical gradients are substantially faster and more stable than numerical differentiation in the large-scale OLG economy studied here. We then solve for optimal nonlinear fiscal systems parameterized by a neural network and compare policy domains of different dimensions. The numerical results indicate large welfare gains relative to the benchmark fiscal system and support the sufficient-variables theory: once taxes condition on a sufficiently informative set of observables, enlarging the formal list of tax instruments provides little additional welfare improvement

Notes:

Source: Dissertations Abstracts International, Volume: 87-12, Section: A.

Advisors: Smetters, Kent A. Committee members: Auerbach, Alan J.; Fernández-Villaverde, Jesús

Ph.D. University of Pennsylvania 2026

Vendor supplied data

Local Notes:

School code: 0175

ISBN:

9798247973447

Access Restriction:

Restricted for use by site license