9 Consumption-Based Models

In this section wewill derivea typical consumption-based asset pricing model in dis-
crete time with dynamic programming. These models have been developed both by
macroeconomists and nancial economists. Macroeconomists want to understand the
behavior of consumption under uncertainty whilenancial economists want to under-
stand asset returns. Consumption and asset returns turn out to beintricately related
to eachother. The model that we will examine is similar to that of Lucas.
10
The
consumption-based model has motivated lots of empirical research, including the de-
velopment of GMM.
11
9.1 Assumptions and Notation
To begin, wemakeseveral assumptions:
1. Thereis a representativeagent.
2. Markets areperfect - no rationingconstraints, transactions costs, etcetera.
3. Thereexists a (real) riskless asset.
4. Individuals can freely borrowand lend at therisk-freerate.
5. labor incomeis given exogenously and it is diversiable.
6. Therepresentativeinvestor has additively separableVon Neumann-Morgenstern
utility.
7. Therepresentativeinvestor lives until period T.
Wealso need to describethenotation to beused:
10
Lucas, R. (1978), \ Asset Prices in an ExchangeEconomy," Econometrica, 46, 1429-45.
11
Hansen, L. and K. Singleton(1982), \ Generalized Instrumental VariablesEstimationof Nonlinear
Rational Expectations Models," Econometrica, 50, 1269-1288.
69
u(c
t
) is theagent's (undiscounted) utility of consumption at period t.
is therepresentativeagent's discount factor for futureutility.
E
t
is theexpectations operator conditional on all information availableat t.
A
t
is theagents total wealth in period t.
y
t
is theagent's exogenous labor incomein period t.
r
it+1
is thereturn on asset i fromt to t +1. It is not known until t +1:
r
0t+1
=r
f t
is therisk-freereturn fromt to t +1. It is known at t:
w
it
is thefraction of theagent's wealth invested in asset i fromt to t +1. It is
chosen in period t:
9.2 Derivation
Under theseassumptions, therepresentativeinvestor solves theproblem,
max
c
t
;w
i t
E
0
T
X
t=0

t
u(c
t
); (160)
s:t: A
t+1
=(A
t
+y
t
c
t
)
"
n
X
i=0
(1+r
it+1
)w
it
#
; (161)
n
X
i=0
w
it
=1: (162)
Thetermin brackets in equation (161) is just thereturn on theagent's portfolio. The
budget constraint just saysthat wealthnext periodisequal tothewealththat westart
with thisperiod minuswhat weconsumepluswhat weearn timesour portfolioreturn.
The additive utility assumption expressed in (160) is signicant. Assuming additive
separability ties together theconcepts of risk aversion and intertemporal substitution.
A big point in thework of Epstien and Zin (cited earlier) is that wemay want to be
70
abletoidentify risk aversion and intertemporal substitution separately. Usingthefact
that r
0t+1
=r
f t
, wecan combine(161) and (162) to obtain
A
t+1
=(A
t
+y
t
c
t
)
"
1+r
f t
+
n
X
i=1
(r
it+1
r
f t
)w
it
#
: (163)
Nowthesummation termis theexcess return on our portfolio of risky assets. Wecan
express thevaluefunction in period t as
V
t
(A
t
) =max
ct ;w
i t
E
t
T
X
s=t

s t
u(c
s
); s:t: (163); (164)
or as thebellman equation,
V
t
(A
t
) =max
ct ;w
i t
[u(c
t
) +E
t
V
t+1
(A
t+1
)]; s:t: (163): (165)
Wecan substitute(163) into (165) to get thenal objectivefunction:
V
t
(A
t
) =max
ct ;w
i t
(
u(c
t
) +E
t
V
t+1

(A
t
+y
t
c
t
)
"
1+r
f t
+
n
X
i=1
(r
it+1
r
f t
)w
it
#! )
:
(166)
Givenour objectivefunction, wecanderiverst order conditionsfor therepresentative
investor. Therst order condition for consumption is,
FOC for c
t
: u
0
(c
t
) =E
t

V
0
t+1
(A
t+1
) [1+r
f t
+
P
n
i=1
(r
it+1
r
f t
)w
it
]

;
t =1;2; 3;: : :; T:
(167)
This condition characterizes the optimal consumption-savings path for our agent. It
says that sheshould always set themarginal utility of consumption today equal tothe
discountedexpectedreturnof her portfoliotimesthemarginal valueof wealthnext pe-
riod. Thisdemonstratesanother typeof consumptionsmoothingbehavior predictedby
economic models. In intertemporal models, agents smooth their consumptions across
71
time. It would probably not beoptimal, for example, for our agent toconsume90%of
her wealth in period 0 and leave the remaining 10%for therest of her lifespan. The
rst order condition for portfolio weights look like,
FOC for w
it
: E
t
[V
0
t+1
(A
t+1
)(r
it+1
r
f t
)] =0; i =1;2; 3;: : :; N; t =1; 2;3; :: : ; T:
(168)
Besides theserst order conditons, wecan also usetheenvelopecondition,
V
0
t
(A
t
) =E
t

V
0
t+1
(A
t+1
)
"
1+r
f t
+
n
X
i=1
(r
it+1
r
f t
)w
it
#!
: (169)
Combining(167) and (169), weobtain
V
0
t
(A
t
) =u
0
(c
t
); (170)
so theFOC for theportfolio weights, (168), can bewritten as
E
t
[u
0
(c
t+1
)(r
it+1
r
f t
)] =0; i =1;2; 3;: : :; N; t =1; 2;3; : :: ; T: (171)
Weimmediately recognizethat (171) is of theform
E
t
[M
t+1
r
e
it+1
] =0; (172)
and we are done. The consumption-based model basically implies that the pricing
kernel, M
t+1
, isequal tothemarginal utilityof consumptioninthenext period, u
0
(c
t+1
).
How do we interpret this model's results? Where are the demand and supply
curves? Therepresentativeagent in this model is actually settingprices and choosing
consumption simultaneously to make markets clear. We could probably derive de-
mand and supply curves for this model, but their forms would be messy. Like most
representativeagent models, this model predicts no trading.
72
Thisasset pricingmodel providessomeintuitionabout what amodel withheteroge-
neousagents and incompletemarkets might look like. Therst order conditions of the
representativeagent arethesameconditions that any ordinary consumer would solve
in thesamecircumstances. Thus, wecan think of consumers choosingtheir consump-
tions to meet theconditions above. If weaggregateconsumers in someway without a
representativeagent, it is not clear what sort of equilibriumwewill get.
Models similar to this havebeen developed with therst order conditons of rms
rather thanconsumers. Production-basedasset pricingmodelsbasically solveasimilar
problemin which rms aremaximizingthepresent valueof futurecash ows with the
constraints that output is produced by aproduction function and capital in period t is
equal todepreciated capital fromperiod t 1plus investment. Thesemodels basically
stipulate that expected stock returns should be set equal to the expected returns on
investment.
73
9.3 Homework Problems
1. Derive the consumption based model without assuming the existence of a risk
freeasset. You will haveto add a constraint that werelaxed in class. You may
want to givethat constraint a lagrangemultiplier and proceed.
2. Supposethat therepresentativeagent has quadratic utility. What sort of asset
pricingmodel results (thisis afamous model - you may not knowthenameyet).
Nowsupposethat therepresentativeagent's consumption is a linear function of
his return. What sort of asset pricingmodel results?
3. Supposethat =
1
1+r
f t
andsupposethat youhavenolabor income. Usingall the
rst order conditions wederived, derivean expression that relates consumption
today tofutureconsumption only (you should get portfolioreturns tocancel out
somehow). What does this expression mean in economic terms?
74