Income and Wealth Distribution in Macroeconomics: A Continuous-Time Approach

Questions & answers

Q1. What is the paper’s basic methodological move, and why does it matter beyond mathematical elegance?

The paper “recasts the standard incomplete market model of Aiyagari (1994), Bewley (1986) and Huggett (1993) in continuous time,” and shows this transforms the model “into a system of partial differential equations” (Introduction, p. 1). This matters because “the literature has suffered from a dearth of theoretical and analytical results. Little is known about the properties of consumption and saving behavior in the presence of borrowing constraints, those of the resulting wealth distribution, and equilibrium uniqueness,” with “most studies rel[ying] on purely numerical analyses” that are “often difficult and costly, particularly if the question at hand requires solving for the economy’s transition dynamics” (Introduction, p. 1).

Q2. What is the coupled HJB-KF (“Mean Field Game”) system, and how do its two equations relate to each other?

The HJB equation “characterizes the optimal choices of a single atomistic individual who takes the evolution of the distribution and hence prices as given,” while the KF equation “characteriz[es] the evolution of the distribution, given optimal choices of individuals”; the two are coupled because “optimal consumption and saving depend on the interest rate which is determined in equilibrium and hence depends on the wealth distribution” (Introduction, p. 1). The paper highlights that the two equations “run in opposite directions in time: the Kolmogorov Forward equation runs forward… and looks backwards – it answers the question ‘given the wealth distribution today… what is the wealth distribution tomorrow?’ In contrast, the [HJB equation] runs backwards and looks forward – it answers the question ‘given an individual’s valuation of income and wealth tomorrow, how much will she save today’” (Section 1.3, p. 11).

Q3. Why does moving to continuous time make borrowing constraints analytically much easier to handle?

In continuous time, “the borrowing constraint never binds in the interior of the state space and only shows up in a boundary condition. The consumption first-order condition always holds with equality, thereby sidestepping any complications due to ‘occasionally binding constraints’” (Introduction, p. 3). This is explicitly contrasted with discrete time, “where there is typically a critical level of wealth, strictly bigger than the borrowing constraint, such that the constraint binds for all lower levels of wealth” (footnote 8, p. 3) – so continuous time removes an entire region of distorted first-order conditions that discrete-time treatments must handle separately.

Q4. What is the paper’s headline analytic result about how poor households behave near the borrowing constraint?

Proposition 1 shows that, under a mild condition on risk aversion at low consumption (Assumption 1), as wealth approaches the borrowing constraint from above, the low-income type’s saving policy and consumption policy both behave according to an explicit square-root law in the distance from the constraint, governed by a “speed” parameter ν₁ that is an explicit function of the interest-rate/discount-rate gap, risk aversion (via the IES), and the transition intensity into the low-income state (Section 2.2, pp. 13-14, equations 19-21). Corollary 1 draws out the implication that “the derivatives of type 1’s consumption and saving policy functions become unbounded at the borrowing constraint… [which] has an important implication, namely that individuals hit the borrowing constraint in finite time” – with an explicit “hitting time” formula proportional to the square root of the initial distance from the constraint divided by ν₁ (Section 2.2, pp. 14-15).

Q5. How does this connect to marginal propensities to consume, a quantity central to a large body of applied work?

The paper shows that the same speed parameter ν₁ that governs how fast a poor household approaches the borrowing constraint also determines her marginal propensity to consume (MPC) out of a windfall income gain (Section 2.4), and the introduction summarizes the comparative statics directly: “this MPC is higher the lower is the interest rate relative to the rate of time preference, the more willing to intertemporally substitute individuals are, or the higher is the likelihood of getting a high income draw; it is non-monotone in the income received in low-income states (e.g. unemployment benefits)” (Introduction, p. 3). The paper explicitly flags that “understanding the theoretical determinants of MPCs is… important for a large body of applied work” on fiscal stimulus, monetary transmission, and the effects of credit crunches (Introduction, footnote 5, p. 3).

Q6. What does the paper prove about the shape of the stationary wealth distribution?

Proposition 3 gives a closed-form solution for the stationary wealth distribution in the special case with two income types, and a direct corollary of households hitting the borrowing constraint in finite time is that the wealth distribution “features a Dirac point mass at this constraint” (Introduction, p. 3; Section 2.3). The paper further notes that this point mass can propagate: “if there is a Dirac mass at [the constraint], there must also be a Dirac mass” at the wealth level households transition to upon receiving a high-income draw, “and so on,” so the point mass “spreads” through parts of the state space reachable from the constraint (Section 2.3, p. 26).

Q7. Under what condition does the paper prove the stationary equilibrium is unique, and why does uniqueness matter?

Proposition 5 proves uniqueness “assume[ing] that the intertemporal elasticity of substitution is weakly greater than one for all consumption levels” and that the borrowing limit is a strict no-borrowing constraint: under this condition, individual saving is strictly increasing in the interest rate for every household, so aggregate saving S(r) is strictly increasing and can equal the fixed bond supply at only one interest rate (Section 2.6, pp. 29-30, drawing on a decomposition of substitution and income effects due to Olivi 2017). The introduction explains the stakes: “without a uniqueness result the economy could, in principle, be subject to poverty traps and history dependence” (Introduction, p. 3) – a possibility the neoclassical growth model’s representative-agent counterpart rules out trivially but that is not obvious once households are heterogeneous and constrained.

Q8. Beyond the theoretical results, what is the paper’s numerical contribution, and how general is it?

The paper develops “a simple, efficient and portable algorithm for numerically solving for equilibria in a wide class of heterogeneous agent models, including – but not limited to – the Aiyagari-Bewley-Huggett model,” based on a finite-difference method for the coupled HJB-KF system (Abstract; Introduction, p. 2). The authors emphasize portability: because it uses “viscosity solutions and finite difference methods… designed to handle non-differentiable and non-convex problems,” the same algorithm applies “without change” to environments discrete-time methods handle only with difficulty, illustrated in the paper with a housing/mortgage model whose down-payment constraint creates poverty traps and multiple stationary distributions (Introduction, p. 4, Section 4.3). The paper explicitly notes that this machinery became the foundation for later extensions to aggregate uncertainty (Ahn, Kaplan, Moll, Winberry, and Wolf 2017 – itself part of this same reading list) and to multiple assets with adjustment costs (Kaplan, Moll, and Violante 2016) (Introduction, footnote and p. 5).