Lecture 2: Consumption

Tiago Cavalcanti1

1 University of Cambridge

E200: Macroeconomics
Cambridge
Michaelmas 2022
Main reading:

Romer, ch. 8

Lecture 1 Outline:
I Consumption Theory
I Permanent Income Hypothesis
I Consumption under uncertainty
I Empirical tests
I Borrowing constraints
Why Do We Study Consumption?

I Consumption is the largest component of GDP

I Consumption, savings, and ultimately investments are the main

drivers of the business cycle

I Consumption models constitute one of the main building blocks

of any macroeconomic model

I The effects of different reforms & policies depend on their

impact on consumption
I Social security reforms
I Tax reforms
I Cash transfers
Milton Friedman (1912-2006), Nobel Prize Winner, 1976
Historical Development

1. Keynesian consumption function:

I Essential component of IS (investment-saving) curve;
I Ct = α + βYt
I Yt is disposable income
I β is the marginal propensity to consume. Determines various
multipliers;
I ’Miopia’ property.

2. Friedman (1957) Permanent Income Hypothesis (PIH):

Ct = γt (Wt + Ht ), Wt is the market wealth; and Ht is the human
wealth;
I Consumption-saving decision is a forward looking and long
horizon planning problem
I Consumption should only adjust to changes of the permanent
component of income.
I Also see Modigliani and Brumberg (1954) life cycle theory.
Finite Horizon Case

I Time is discrete: t = 0, 1, ..., T;

I Utility function: U = u(c0 ) + βu(c1 ) + ... + β T u(cT );

I β ∈ [0, 1] : Time preference factor;

I u0 (·) > 0, u00 (·) < 0;

I Example:

c1−σ − 1
u(c) = , σ > 0. And lim u(c) = ln(c).
1−σ σ→1
Household’s Problem

The household’s problem is:

T
X
max β t u(ct ),
{ct ,at+1 }Tt=0
t=0

Subject to:

ct + at+1 = yt + (1 + r)at , ∀t = 0, 1, 2, ..., T.

a0 is given,

aT+1
≥ 0, (no-Ponzi games).
(1 + r)T+1
Notice that:
c0 + a1 = y0 + (1 + r)a0 ,
c1 − y1 a2
c1 + a2 = y1 + (1 + r)a1 ⇒ a1 = + . Then,
1+r 1+r
c1 a2 y1
c0 + + = (1 + r)a0 + y0 + ,
1+r 1+r 1+r
c2 − y2 a3
c2 + a3 = y2 + (1 + r)a2 ⇒ a2 = + , Then,
1+r 1+r
c1 c2 a3 y1 y2
c0 + + 2
+ 2
= (1+r)a0 +y0 + + .
1 + r (1 + r) (1 + r) 1 + r (1 + r)2
By repeated substitution, we have the present value budge constraint:
T T
X ct aT+1 X yt
t
+ T
= (1 + r)a 0 + .
(1 + r) (1 + r) (1 + r)t
t=0 t=0
Solution to Household’s Problem I
T
X T
X
t
L= β u(ct ) + λt [yt + (1 + r)at − ct − at+1 ],
t=0 t=0

First-order conditions:

∂L
= β t u0 (ct ) − λt = 0,
∂ct
∂L
= −λt + (1 + r)λt+1 = 0,
∂at+1
∂L
= −λT ≤ 0, aT+1 ≥ 0, λT aT+1 = 0
∂aT+1
Then:

u0 (ct )
= (1 + r) ∀ t = 0, 1, 2, ..., T. (Euler)
βu0 (ct+1 )
Solution to Household’s Problem II

Euler equation:
u0 (ct )
= (1 + r)
βu0 (ct+1 )
Complementary-slackness condition implies that:
aT+1
= 0, guarantees no-Ponzi games
(1 + r)T

Therefore, the present value budget constraint implies that:

T T
X ct X yt
t
= a0 (1 + r) + , a0 given.
(1 + r) (1 + r)t
t=0 t=0
Solution to Household’s Problem III
c1−θ
Let u(c) = 1−θ , then the Euler equation becomes:
1
ct+1 = (β(1 + r)) θ ct .

I Elasticity of substitution between consumption at two points in

time:
d(ct+1 /ct )/(ct+1 /ct ) d ln(ct+1 /ct ) 1
EIS = − d(1+r)
= = .
d ln(1 + r) θ
(1+r)

I When θ → 0, then any change in r generates large effects on

ct+1
ct . (“low preferences for consumption smoothing”).

I When θ → ∞, then any change in r generates small effects on

ct+1
ct . (“high preferences for consumption smoothing”).
Solution to Household’s Problem IV
Notice that:
1 1 t
c1 = (β(1 + r)) θ c0 , c2 = (β(1 + r)) θ c1 ⇒ ct = (β(1 + r)) θ c0 .

Using the present value budget constraint, we have that:

T 1
!t T
X [β(1 + r)] θ X yt
c0 = a0 (1 + r) + ,
(1 + r) (1 + r)t
t=0 t=0

1
[β(1+r)] θ
If b = (1+r) < 1 ⇔ β < (1 + r)θ−1 , then:
1
1 − [β(1+r)]
θ " T
#
(1+r)
X yt
c0 = T+1 (1 + r)a0 + .
(1 + r)t
 1
[β(1+r)] θ t=0
1− (1+r)
Solution to Household’s Problem V
I As T → ∞, then:
 
1
! Financial wealth ∞ 
[β(1 + r)] θ  z }| { X yt 
c0 = 1 −  a0 (1 + r) + t
.
(1 + r)  (1 + r) 
 t=0 
| {z } | {z }
Eat a fraction of life-time income
I Recall that, ln(1 − b) ≈ −b for b < 1. If β = 1
1+ρ , then in logs:
1
[β(1+r)] θ
(1+r) ≈ θ1 (r − ρ) − r, and −b = r − θ1 (r − ρ).
" ∞
 #
1 X yt
c0 = r − (r − ρ) a0 (1 + r) + .
θ (1 + r)t
t=0

If θ → 0 and r > ρ postpone consumption;

If θ → ∞, then consume r of total wealth;
Savings Under Uncertainty I

X∞
max Et { β j u(ct+j )}, subject to
at+j+1 ,ct+j
j=0
at+j+1 + ct+j = (1 + r)at+j + yt+j ∀ j,
at+1+J
lim Et ≥ 0.
J→∞ (1 + r)J
Assume: yt+j is a random variable. At t the agent know yt .

∞
X ∞
X
j
L = Et { β u(ct+j ) + λt+j [(1 + r)at+j + yt+j − ct+j − at+j+1 ]},
j=0 j=0

∂L ∂L
= u0 (ct ) − λt = 0, = βEt [u0 (ct+1 )] − Et [λt+1 ] = 0,
∂ct ∂ct+1
∂L
= −λt + (1 + r)Et [λt+1 ] = 0,
∂at+1
Savings Under Uncertainty II

I Euler equation:

u0 (ct ) = β(1 + r)Et [u0 (ct+1 )].

I If β(1 + r) = 1 ⇒ u0 (ct ) = Et [u0 (ct+1 )]

I Suppose that variance of future consumption rise, then if u0 (c) is

strictly convex, then Et [u0 (ct+1 )] rises.

I By the Euler equation u0 (c) has to rise and ct falls (precautionary

savings).
Savings Under Uncertainty III

I Example: Quadratic utility function.

α2 2 α1
u(c) = α1 c − c , ⇒ u0 (c) = α1 − α2 c, c < .
2 α2
α1 − α2 ct = Et [α1 − α2 ct+1 ] ⇒ ct = Et [ct+1 ].
I Consumption follows a martingale and changes of consumption
can’t be predicted:

ct+1 = ct + µt+1 , Et [µt+1 ] = 0,

⇒ Et [∆ct+1 ] = 0.
Savings Under Uncertainty IV
Present value budget constraint:
   
∞ ∞
X 1 X 1
Et  ct+j  = at (1 + r) + Et  yt+j  .
(1 + r)j (1 + r)j
j=0 j=0

ct = Et [ct+1 ] and ct+1 = Et+1 [ct+2 ]

By the law of iterated expectations: Et [Et+1 (·)] = Et [·], then:

ct = Et [ct+1 ] ⇒ ct = Et [Et+1 (ct+2 )] = Et [ct+2 ] ... ct = Et [ct+2 ],

 
  ∞
1 1 X 1
ct 1 + + + ... = at (1+r)+Et  yt+j  .
(1 + r) (1 + r)2 (1 + r)j
j=0
  
∞
r X 1
ct = at (1 + r) + Et  yt+j  ,
1+r (1 + r)j
j=0
Savings under uncertainty IV
At any period t:
  
∞
r X 1
ct = at (1 + r) + Et  yt+j  ,
1+r (1 + r)j
j=0
  
∞
r X 1
ct = rat + Et  yt+j  = r(at + Ht ) = yPt ,
1+r (1 + r)j
j=0

at : Financial wealth; and Ht : Human wealth. Consumption in t is then

the annuity value of total wealth.

If σy increases but does not change the expected value, then ct

remains the same (Certainty Equivalence Principle). It holds for a
quadratic utility function, but it is not a general result.
Savings under uncertainty V
  
∞
r X 1
ct = at (1 + r) + Et  yt+j  ,
1+r (1 + r)j
j=0

and ct + at+1 = (1 + r)at + yt or (1 + r)at = ct + at+1 − yt

 
∞
r X 1
ct = rat+1 + Et  yt+1+j  . Since
1+r (1 + r)j
j=0

  
∞
r X 1
ct+1 = at+1 (1 + r) + Et+1  yt+j+1  ,
1+r (1 + r)j
j=0

∞ ∞
r X 1 X 1
ct+1 −ct = {Et+1 [ j
yt+1+j ]−Et [ yt+j+1 ]}.
1+r (1 + r) (1 + r)j
j=0 j=0
 ∞ ∞
r X 1 X 1
ct+1 −ct = {Et+1 [ yt+1+j ]−Et [ yt+j+1 ]}.
1+r (1 + r)j (1 + r)j
j=0 j=0

Interpretation:
I Consumption changes because new information arrives at date
t + 1;
I The new information arrives between t and t + 1 is yt+1 ;
I yt+1 can be decomposed into two parts:

yt+1 =Et [yt+1 ] + yt+1 − Et [yt+1 ] .

| {z } | {z }
Forecast at t Shock: new information
I Consumption should only react towards the new information
Robert Hall’s (1978) random walk test
I Basis of the test: Et (ct+1 ) = ct .

I Conditioned on present consumption, no other variable should

help to predict future consumption.
I Empirical Strategy: Run regression,
ct = γ0 + γ1 ct−1 + γ2 zt−1 + ηt , where zt−1 could be any lagged
variables.
I Data: Aggregate consumption, GDP and others.

I H0 : γ1 = 1 and γ2 = 0.

I Results: Null hypothesis is rejected

I consumption lagged more than one period and lagged income
have coefficients close to zero,
I but lagged return on stocks has prediction power on current
consumption.
Flavin’s (1981) excess sensitivity test I
I Assume that income process is a AR(q) process;
I yt+1 = α1 yt + α2 yt−1 + ... + αq yt+1−q + υt+1 .
I υt+1 is the new information at time t + 1.
I Under PIH, ct+1 − ct = ∆ct+1 = θυt+1 and ∆ct+1 should not
react towards Et [yt+1 ]
I Data: aggregate consumption and GDP

I Empirical Strategy:
I First run the income regression and get:

υt+1 = yt+1 − (α1 yt + α2 yt−1 + ... + αq yt+1−q )

I Run the consumption regression:

∆ct+1 = β0 + β1 ∆yt+1 + β2 ∆yt + θυt+1 + ηt

I β1 and β2 excess sensitivity parameters.

Flavin’s (1981) excess sensitivity test II

I ∆ct+1 = β0 + β1 ∆yt+1 + β2 ∆yt + θυt+1 + ηt

I PIH ⇒ H0 : β1 = 0 and β2 = 0;
I Consumption should react only to unexpected change in income
θ 6= 0.

I Results: βi 6= 0 and β1 = 0.355. Consumption reacts strongly

towards predicted income changes.

I Campbell and Mankiw (1989) found that excess sensitivity exists

for many countries other than US.

I What is wrong with this theory of consumption?

Saving is a buffer to income variations
I Frequent and short term income variations.
I due to job loss, illness, or other unforeseen events;
I Households usually build buffer stock in the form of saving
account in banks (more liquid) to prevent a dramatic fall in their
consumption when such an unfortunate event does occur. This is
precautionary motive of saving.

I Earning variations during life-cycle; age-earning profile is

hump-shaped;
I Over the life-cycle, consumption tracks income; excess
sensitivity;
I With perfect credit market, i.e., without borrowing limit,
household consumption should be more less constant. We should
observe young households borrow.
I That is not what we observe in the data. This suggests existence
of borrowing limits
Life-time consumpt. and income (Carroll, 1997, QJE)
Borrowing constraint

P∞ t u(c
I maxat+1 ,ct E0 { t=0 β t )}, subject to

at+1 + ct = (1 + r)at + yt ∀ t, at+1 ≥ −φ, φ[0, ∞).

I Euler eq. when at+1 > −φ: u0 (ct ) = β(1 + r)Et [u0 (ct+1 )].

I Euler eq. when at+1 = −φ: u0 (ct ) ≥ β(1 + r)Et [u0 (ct+1 )].
I Constrained household has lower consumption today and would
like to reduce future consumption to increase today’s
consumption but can’t.

I Point: The consumption growth rate of constrained households is

on average higher than that of unconstrained households.
Zeldes’ (1989) test of existence of liquidity constraint
I Data: PSID (Panel Study of Income Dynamics). Split the sample
into three groups:
I Constrained group has low wealth-to-income ratio;
I Unconstrained group has high wealth-to-income ratio;
I Omit the middle group.

I Maintained hypothesis: the two groups are identical in every

other aspect other than one group is temporary short on cash.
I Empirical Strategy:
I estimate preference parameter β and θ using data of
unconstrained group;
I Use the estimates and data of constrained group to compute the
following:
λt = β(1 + rt )Et [c−θ −θ
t+1 ] − ct .
I Test: H0 : λt = 0.
I Result: λt > 0. Liquidity constraint exists.
Next time:

Real Business Cycles