Essays on the Welfare Implications of Fiscal Policies over the Business Cycle in Heterogeneous-Agent Models

Abstract

This dissertation investigates the welfare implications of fiscal policy in business cycle models with rich household heterogeneity. It demonstrates that understanding heterogeneity in households’ consumption–saving behavior—both empirically and theoretically—is essential for designing effective fiscal policy. To address the theoretical challenges, the study also proposes a computational method for efficiently solving heterogeneous-agent models with aggregate shocks, which are workhorses of modern macroeconomic analysis. Chapter 1 examines the optimal design of countercyclical transfers. One-time stimulus checks are widely used during recessions, but expectations of future transfers alter households’ precautionary saving. I develop a two-asset heterogeneous-agent model with aggregate shocks, solved globally using the method in Chapter 3. The optimal policy provides an additional $1,800 per recessionary year relative to the baseline, with welfare gains from reduced consumption risk outweighing lower long-run capital. Effects are highly uneven: indebted and poor Hand-to-Mouth households benefit most, wealthy Hand-to-Mouth households lose, and Savers are neutral. Computationally, I show that standard forecasting-rule solutions can misestimate aggregate capital in recessions, leading to understated transfer cyclicality and welfare gains. Chapter 2 evaluates Canada’s Tax-Free Savings Account (TFSA), introduced in 2009. Using administrative tax data in a regression discontinuity design, I find that TFSAs raised the share of households with liquid savings by 3.5 percentage points. A structural model shows welfare gains of 0.2859% in consumption-equivalent terms. The policy encourages liquid saving, which increases long-run capital and consumption while reducing volatility. The main insurance value arises from smoothing individual, rather than aggregate, consumption fluctuations. Chapter 3 develops a new global solution method for heterogeneous-agent models with aggregate shocks. The algorithm blends finite difference methods with deep learning to overcome the curse of dimensionality. It iteratively updates guesses of derivatives and decision rules, solves value functions, and uses machine learning to extract cross-state sensitivities. The method is flexible and applicable to a wide range of models, particularly for studying fiscal and monetary policies where inequality interacts with aggregate outcomes. Its usefulness is illustrated by applications to the classic Aiyagari (1994) and Krusell and Smith (1998) models.