Job market paper. We study how the distribution of permanent labor income (PLI) affects the transmission of fiscal policy. Using US household survey data, we show that the hand-to-mouth share falls from 64% in the bottom PLI quintile to 22% in the top one. In a HANK model with a non-homothetic bequest motive that reproduces this gradient, the multiplier of a fiscal package depends at least as much on who receives the payout and who finances it as on how long the debt lasts. Applied to the One Big Beautiful Bill Act, the model gives a cumulative multiplier of −0.06, against 0.57 in a standard HANK model with the same aggregate MPC.

Under review, Journal of Monetary Economics. Higher permanent labor income inequality (PLI) raises saving, since households with higher PLI save a larger share of it. For asset markets to clear, prices must adjust: how does this feed back on wealth inequality? If the saving finances capital, interest rates fall, wages rise, and poorer households gain. If it raises asset prices instead, the capital gains accrue to the rich. In a heterogeneous-agent model with non-homothetic wealth preferences and imperfect competition, markups weaken the first channel and let the second dominate for five decades along the transition.

We study the role of expectations regarding aggregate risk in two standard HANK models, using a novel deep learning approach that approximates their solutions in moving-average form. In a simple Intertemporal Keynesian Cross model, we find strong sign, size and state dependence in the impulse responses to deficit-financed transfer shocks. These nonlinearities are, however, already present in the nonlinear perfect-foresight solution. Aggregate risk slightly lowers the MPC, as fewer households are hand-to-mouth. Households also stay further from the borrowing constraint, which dampens the response to negative fiscal shocks. In a larger HANK model, the nonlinearities are weaker, and the dynamics under aggregate risk and under perfect foresight again coincide.

We introduce a novel deep learning algorithm for solving large life-cycle models with both continuous and discrete choices. This allows us to simultaneously account for both labor supply choices with human capital accumulation, portfolio choices with a risky and a risk-free asset, and housing and mortgage choices. We work directly on the Bellman equation and approximate both value and policy functions with neural networks, and use a simulated training sample instead of tensor product grids. This substantially alleviates the curse of dimensionality. We confirm that we can accurately solve a large life-cycle model in 3 hours on a single GPU. An accompanying easy-to-use software package implements the method.

Despite governments’ commitments to limit global warming to 1.5°C, there is still investment in carbon-intensive capital. This paper uses a growth model featuring irreversible investment, capacity utilisation, clean and polluting capital to study this apparent paradox. It shows that current investment in polluting capital and CO2 emissions are coherent with expectations of a future carbon tax, if investors also expect a bailout of polluting capital. This result implies that governments’ credibility can play an important role in reducing the cost of implementing an optimal carbon tax by committing not to bail out.

We study why monetary tightening can push small but very productive and fast-growing firms out of the market. These firms are highly leveraged during their growth period and are thus highly sensitive to changes in interest rates. Monetary tightening can therefore worsen the selection of firms, with long-term effects on growth and productivity.