econ 747 – lecture 5:

incomplete markets, heterogeneous agents

and precautionary savings

Thomas Drechsel

University of Maryland

Spring 2022
recap from previous lecture

I We built a benchmark complete markets model

I Heterogeneity (idiosyncratic risk) could be insured away
I We started looking at some form of market incompleteness
I So far, no meaningful interaction between market incompleteness and
heterogeneity of agents

1 / 37
this lecture

I Introduce idiosyncratic risk and incomplete markets in combination

I Recall that we need both for financial frictions to matter

I In today’s lecture, the focus will be primarily on heterogeneity, the market

incompleteness (financial friction) will be ad-hoc
I We will learn today how precautionary savings are a consequence of market
incompleteness

2 / 37
references

I A good textbook reference for this lecture is the Ljungqvist and Sargent text book
Recursive Macroeconomic Theory (2nd edition), Chapters 16 and 17

I As usual, I will provide additional references throughout

3 / 37
overview

1. Precautionary savings in partial equilibrium

2. Incomplete markets models: baseline setting

3. Models with incomplete markets, heterogeneous agents and precautionary savings

3.1 Pure credit economy: Huggett (1993)
3.2 Adding capital: Aiyagari (1994)
3.3 Adding aggregate risk: Krusell and Smith (1998)

4. The latest generation of heterogeneous agent models (“HANK”)

4 / 37
precautionary savings in partial equilibrium
why start in partial equilibrium?

I We want to build intuition for the basic decision margins of individual agents

I Recall Lucas tree model: take consumption patterns as given, price assets (GE)

I Here: take asset prices as given, study agents’ consumption behavior (PE)

5 / 37
a simple pe setting
I Assume single household, endowment income, no uncertainty

I Budget constraint and no-Ponzi condition

ct + bt+1 = Rbt + yt
bt+1
lim = 0
t→∞ Rt

I Euler equation for risk-free bond is

u0 (ct ) = βRu0 (ct+1 )

I Consumption path will be:
I Constant if βR = 1
I Increasing if βR > 1
I Decreasing if βR < 1

6 / 37
the permanent income hypothesis
I In this world, current income has no effect on consumption path, only lifetime
income (permanent income) does

I How to see this formally? Consolidate the budget constraint and impose no-Ponzi
condition:

Rbt = ct − yt + bt+1
ct+1 − yt+1 bt+2
Rbt = ct − yt + +
R R
...
∞
X ct+j − yt+j
Rbt =
Rj
j=0

7 / 37
the permanent income hypothesis
I Suppose we have log utility and βR = 1 so that ct+1 = ct

I In that case we get

∞ ∞
X 1 X yt+j
Rbt = ct j
+
R Rj
j=0 j=0
 
∞
R  X yt+j 
ct = Rbt +
R−1 Rj
j=0

I Current consumption depends only on lifetime income (asset wealth and all future
endowment income)
I Current income changes (that leave permanent income the same) do not change
consumption
8 / 37
testing the pih empirically
I Macro data
I Hall (1978): assume quadratic utility ⇒ Et ct+1 = ct ; no current information (other
than current consumption) should predict future consumption; confirmed for
disposable income, rejected for stock prices
I Wilcox (1989): look at pre-announced increases in social security benefit; do not rise
consumption immediately, so PIH rejected; possible explanation: liquidity constraints

I Micro data
I Zeldes (1989a): tests for liquidity constraint explanation; PIH holds for “rich”
households
I Shea (1995): consider households with long-term union contracts; consumption
responds to predictable wage movements; responds asymetrically to
increase/decreases

9 / 37
precautionary savings

I With uncertainty, current consumption depends only on expected lifetime income

I “Precautionary savings” = difference between consumption

1. when income is certain
2. when income is uncertain but has same mean as in 1.

10 / 37
precautionary savings

I Look at case βR = 1 and and compare certainty to uncertainty case:

u0 (ct ) = u0 (ct+1 )
u0 (c∗t ) = Et u0 (c∗t+1 )
 

Suppose Et c∗t+1 = ct+1

 
I

I Precautionary savings would mean c∗t < ct . Is this the case?

11 / 37
precautionary savings
Use result Et u0 (c∗t+1 ) > u0 Et c∗t+1
   

I

I This holds iff u000 > 0 (based on Jensen’s inequality)

I u000 > 0 means marginal utility is convex
I See Zeldes (1989b)

Substitute assumption Et c∗t+1 = ct+1 into the right hand side of the above
 
I
inequality

Et u0 (c∗t+1 ) > u0 (ct+1 )

 

u0 (c∗t ) > u0 (ct )

c∗t < ct

(if u00 < 0 i.e. function u0 is decreasing)

12 / 37
incomplete markets models: a baseline setting
environment

I Similar to previous lecture, with some changes in notation

I Households i = 1, ..., I, ex ante identical, but ex-post heterogeneous
I Preferences:
∞
X
Ui,t = Et β t−τ u(ci,τ )
τ =t

I Incomplete markets: only risk free bonds

ci,t + ai,t+1 ≤ (1 + r)ai,t + si,t w

13 / 37
environment

I si,t is employment status or productivity

I si,t follows m state Markov process with transition matrix Pm×m which gives
transition probabilities between states k and ` as P(si,t = sk , si,t+1 = s` )
I Suppose assets can be chosen from grid A = {a1 , ..., an }
(Careful: here the superscript denotes the realization of the state, not the history)
I No aggregate uncertainty
I Assume for now that β(1 + r) < 1, treat r and w as parameters

14 / 37
remarks

I Defining s and a over discrete spaces will be helpful when we take about solution
methods for these models further below

I Important note: in this setting, there are actual individual savings (no “illusion of
choice”) because there are incomplete markets and heterogeneity

15 / 37
bellman equation of agent i

m
( )
X
k k ` 0 `
V (a, s ) = max
0
u(c) + β P(s , s )V (a , s )
c,a ∈A
`=1

subject to

c + a0 = (1 + r)a + sk w

(I have dropped subscript i for ease of notation)

16 / 37
what’s the solution

I What are the policy functions in this economy?

17 / 37
what’s the solution

I What are the policy functions in this economy?

a0 = ga (a, s)
c = gc (a, s)

17 / 37
what’s the solution

I What are the policy functions in this economy?

a0 = ga (a, s)
c = gc (a, s)

I With a discrete state space, the policy functions are also transition matrices

I They are the same for each agent, because agents are ex-ante identical

17 / 37
wealth-employment distribution

I In this economy, we can define the unconditional distribution of states for a given
agent recursively

I Define λt (a, s) = P rob(at = a, st = s). We get

X X
λt+1 (a0 , s0 ) = λt (a, s)P(s, s0 )
s {a:a0 =g(a,s)}

I A stationary distribution satisfies

λ(a, s) = λt+1 (a, s) = λt (a, s)

18 / 37
interpretation of stationary distribution

I The fraction of time that an infinitely lived agent spends in state (a, s)

I Fraction of households in state (a, s) in a given period in a stationary equilibrium

I In the models we study below, the initial distribution of agents over individual
state variables (a, s) remains constant over time even though the state of the
individual household is a stochastic process

19 / 37
introducing incomplete markets and heterogeneity

I I will run you through some specific models that build on the above framework

I In this lecture, the market incompleteness we look at is relatively ad-hoc (a simple

exogenous borrowing limit) and the focus is an the heterogeneity aspect

I In the future lectures, we will have more elaborate frictions and but will then
typically make the heterogeneity aspect simpler
I e.g. one saver and one borrower

20 / 37
pure credit economy: Huggett (1993)
hugget environemnt

I Two changes to framework above

1. We determine r in equilibrium
2. We add a borrowing limit

I Borrowing limit is exogenously assumed to be

a > −φ

ws1
I Could be natural debt limit (φ = r ) or exogenous parameter that is more
restrictive (more on this later)

21 / 37
competitive equilibrium

I Given φ and P, a stationary competitive equilibrium is an interest rate r, a policy

function a0 = g(a, s) and a stationary distribution λ(a, s), such that

1. g(a, s) solves households’ maximization problem, taking r as given

2. λ(a, s) is implied by P and g(a, s)
3. Asset markets clear
X
λ(a, s)g(a, s) = 0
a,s

22 / 37
computational solution algorithm

1. Guess an interest rate rj

2. Given rj , solve household’s problem for g j (a, s)

3. Use g j (a, s) and P to compute λj (a, s)

4. Compute excess demand of savings

X
ej ≡ λj (a, s)g j (a, s)
a,s

5. If |ej | < tol, stop

Otherwise, if ej > 0, set rj+1 < rj ; if ej < 0, set rj+1 > rj

23 / 37
adding capital: Aiyagari (1994)
environment: production economy

I Households:
∞
X
max β t u(ct )
t=0

subject to
ct + kt+1 = (rt + 1 − δ)kt + wt st

(again, I have omitted subscript i for the households’ problem)

I Representative firm:

max AF (Kt , Nt ) − rt Kt − wt Nt

24 / 37
environment: production economy
I Firm FOCs:

∂F
wt = A
∂Nt
∂F
rt = A
∂Kt

I Household policy function

kt+1 = g(st , kt )

I A constant, so no aggregate risk

25 / 37
competitive equilibrium

I Given P, a stationary competitive equilibrium is a set of prices (r, w), a policy

function k 0 = g(k, s), a stationary distribution λ(k, s) and an aggregate allocation
(K, N ) such that

1. g(k, s) solves households’ maximization problem, taking prices as given

2. λ(k, s) is implied by P and g(k, s)
3. Firms maximize profits
4. Labor markets clear
Nt = ξ ∞ s̄

5. Capital markets clear

X
λ(k, s)g(k, s) = K
k,s

26 / 37
computational solution algorithm
1. Guess a capital stock K j

2. Given K j , solve firm’s problem for rj , wj

3. Given rj , wj solve households problem for g j (k, s)

4. Use g j (k, s) and P to compute λj (k, s)

5. Calculate implied aggregate capital stock

X
K̃ j ≡ λj (k, s)g j (K, s)
k,s

6. If |K̃ j − K j | < tol, stop

Otherwise, if K̃ j − K j > 0, set rj+1 < rj ; if K̃ j − K j < 0, set rj+1 > rj

27 / 37
putting things together

I In Bewley setting, there is a precautionary savings motive for each individual

agent. However, aggregate savings are still zero in equilibrium. Tightening
borrowing limit will be reflected in lower interest rates.

I The Aiyagari setting, there is a positive supply of aggregate savings, because

savings are in capital. In this setting, precautionary motives will lead to lower
interest rates (returns) and a higher capital stock.

I Can put things together in one diagram ...

28 / 37
aggregate precautionary savings

29 / 37
adding aggregate risk: Krusell and Smith (1998)
aggregate risk

I Suppose At subject to shocks (aggregate risk)

I In this case, no invariant distribution λ(k, s)

I Since the distribution in a given period matters for prices, each agent will want to
know the entire distribution

I Insight of Krusell and Smith: can approximate distribution with a finite set of
moments (the mean)

30 / 37
insights of krusell and smith

I Not that much lost with a representative agent model

I Model with uninsurable idiosyncratic risk displays far less dispersion and skewness
in wealth than we see in US data

I The framework remains a very influential baseline in macroeconomics!

31 / 37
the latest generation of heterogeneous agent
macro models
the world according to hank

I As reviewed in Lecture 2, modern quantitative DSGE models usually feature New

Keynesian elements (e.g. nominal rigidities) around a neoclassical core
I The models studied above add substantial complexity to the neoclassical core,
through heterogeneity (idiosyncratic risk) and incomplete markets
I The very latest generation of macro models bridges this more complex core again
with New Keynesian features
I Heterogeneous Agent New Keynesian (“HANK”) models
I See e.g. Kaplan, Moll, and Violante (2018)

32 / 37
from micro to macro

I HANK models allow to study empirically realistic wealth and income distributions
I A key element is the marginal propensity to consume out of transitory income
(the PIH does not hold)
I Distributional features of the economy:
I Are an object of interest
I Matter meaningfully for aggregate dynamics

33 / 37
from micro to macro

I In the first HANK models, strong emphasis on matching micro moments

I More recently, emphasis on also quantitatively matching/explaining aggregate

dynamics, e.g. responses to different types of shocks
I E.g. Auclert, Rognlie, and Straub (2020) and Bayer, Born, and Luetticke (2020)

34 / 37
a picture from a former umd student

I From Lee (2020): “Quantitative Easing and Inequality”

I The graph shows the welfare effects of QE in terms of consumption equivalents
for different income groups and decomposed into different sources
35 / 37
remarks and further reading

I In this course, we focus in more detail on financial frictions and will typically have
relatively simple layers of heterogeneity
I Important building blocks of HANK models are related to what we study
I Different types of market incompleteness
I E.g. often multiple assets with different degrees of liquidity and different returns

I Highly recommended further reading on the HANK literature:

I Ben Moll’s teaching material (available on his website)

36 / 37
wrapping up
wrapping up

I We have introduced idiosyncratic risk and incomplete markets in combination

I We got to know some canonical heterogeneous agent models

I From now on, we will start going deeper into the formulation of financial frictions,
and study how they matter for macroeconomic dynamics

37 / 37
Bibliography

Aiyagari, S. R. (1994): “Uninsured idiosyncratic risk and aggregate saving,” The Quarterly Journal of Economics, 109, 659–684.
Auclert, A., M. Rognlie, and L. Straub (2020): “Micro jumps, macro humps: Monetary policy and business cycles in an estimated HANK
model,” Tech. rep., National Bureau of Economic Research.
Bayer, C., B. Born, and R. Luetticke (2020): “Shocks, frictions, and inequality in US business cycles,” .
Hall, R. E. (1978): “Stochastic Implications of the Life Cycle-Permanent Income Hypothesis: Theory and Evidence,” Journal of Political Economy,
86, 971–987.
Huggett, M. (1993): “The risk-free rate in heterogeneous-agent incomplete-insurance economies,” Journal of economic Dynamics and Control, 17,
953–969.
Kaplan, G., B. Moll, and G. L. Violante (2018): “Monetary policy according to HANK,” American Economic Review, 108, 697–743.
Krusell, P. and A. A. Smith, Jr (1998): “Income and wealth heterogeneity in the macroeconomy,” Journal of political Economy, 106, 867–896.
Lee, D. (2020): “Quantitative Easing and Inequality,” Job Market Paper.
Shea, J. (1995): “Union Contracts and the Life-Cycle/Permanent-Income Hypothesis,” The American Economic Review, 85, 186–200.
Wilcox, D. W. (1989): “Social security benefits, consumption expenditure, and the life cycle hypothesis,” Journal of Political Economy, 97,
288–304.
Zeldes, S. P. (1989a): “Consumption and Liquidity Constraints: An Empirical Investigation,” Journal of Political Economy, 97, 305–346.
——— (1989b): “Optimal Consumption with Stochastic Income: Deviations from Certainty Equivalence,” The Quarterly Journal of Economics, 104,
275–298.