Intertemporal Consumption, Consumption CAPM

(CCAPM) and the Equity Premium Puzzle (EPP)

• In this section we look at:

1. Intertemporal consumption
2. The CCAPM
3. The EPP
4. Evidence on EPP

1
 Optimizing Consumption
• Consider a consumer with a utility function:
U ( co ,c1 ,......,cT )

• The consumer tries to maximize the utility

subject to a constraint that the present value of
his lifetime consumption cannot exceed the PV
of his lifetime income. The budget constraint is:
T
ct yt
 (1  r)t  0 (1  r)t
2
• Thus the optimization problem is:
• Max U ( co ,c1 ,......, cT )

subject to,
T
ct yt
 (1  r )t  0 (1  r )t

3
 For a two-period case we have:
Max
U ( co ,c1 )
subject to

c1 y1
co   y0 
1 r 1 r

4
 Two-Period Consumption Case

C1

y1+(1+r)y0

y1 A

Slope = -(1+r)
B
c1*
U0
y0+y1/(1+r)

0
y0 c0* C0
5
• The general optimization problem is:
• Max U ( co ,c1 ,......,cT )

subject to

T
ct yt
 (1  r )t  0 (1  r )t

6
Assume a logarithmic utility function which is additively
separable. That is,
In(c1 ) In(ct ) In(cT )
U (c0 , c1 ,.., ct ,...cT )  In(c0 )   .........   ...... 
1  (1   )t (1   )T
T
In( ct )

0 (1   )
t

• Budget constraint is: ct T

yt
 t 0 (1  r )t  0 (1  r )t
T

• Where δ is the subjective discount factor reflecting the

rate at which the individual weights future consumption
relative to the present.
7
So we seek to maximize: T
In( ct )
0 ( 1   )t
T
ct yt
Subject to:
 T
t 0
(1  r ) t

t 0 (1  r )
t

8
 T
In( ct ) T yt T
ct 
L     t 
0 (1   )  0 1  r  0 (1  r ) 
t t

• FOCs: L 1
    0.............( 1 )
co c0
L 1 1 
   0......( 2 )
ct 1    ct ( 1  r )
t t

L 1 1 
   0..............( 3 )
cT 1    cT ( 1  r )
T T

L  T yt T
ct 
  t 
 t
 0.............( 4 )
  0 1  r  0 ( 1  r ) 
9
• To get the consumption path we shall start by comparing time t
consumption to time t-1 consumption.
L 1 1  1 1 
  0  ...............( A)
ct 1 1    ct 1 (1  r )
t 1 t 1
1    ct 1 (1  r )
t 1 t 1

L 1 1  1 1 
  0  ...............................( B)
ct 1    ct (1  r )
t t
1    ct (1  r )
t t

Dividing equation A by equation B yields :

1   t ct  (1  r )t
1    ct 1 (1  r )t 1
t 1

 1   
t t 1 ct
.  (1  r )t t 1
ct 1
ct (1  r )  1 r 
  OR ct   ct 1
ct 1 1    1  
10
• In general for any two adjacent periods we would
have: c  1 r 
t
 
ct 1 1  

 1 r 
• Or ct   ct 1
1  

• Or more generally: u ( ct ) 1  

u ( ct 1 ) 1  r
11
• The above expression can be rewritten as follows:

U ' (ct ) 1  

U ' (ct 1 ) 1  r
1 
 U ' (ct )  U ' (ct 1 )
1 r
 1 r 
 U ' (ct 1 )  U ' (ct ) 
1  
• The last expression is important. It relates loss from present
sacrifice to gains from future consumption.
12
• The last expression says that if I reduce my period t-1
consumption by one unit then my loss would be U’(Ct-1)
and my future consumption would increase by:
1  r  U ' (c )
(1   )
t

• The market interest rate r measures the return on

additional savings.

• The discount rate δ gives the individual’s loss from

waiting to consume. It measures patience. An impatient
individual has a very high discount rate.

13
 Implications from the FOCs
(a). The slope of the consumption path depends on r relative to δ.
This gives us time profiles of consumption as shown below:
• If r > δ, then : ct  (1  r )  1  ct  1  c  c
ct 1 1   
t t 1
ct 1
It pays to save now and consume more later.
ct (1  r ) ct
• r < δ, then :  1  1  ct  ct 1
ct 1 1    ct 1

It pays to consume more now (save less now) and consume less later.
ct (1  r ) ct
• r = δ, then :   1   1  ct  ct 1
ct 1 1    ct 1
• current and future consumption equal.
14
 Implications

ct

r 
r 
r 
0 T Time

15
2. Along an optimal consumption path, ct-1 is a
good indicator for ct. This comes from:

 1 r 
ct   ct 1
1  

16
 CCAPM and the Equity Premium
Puzzle
• In this section we look at:
• The CCAPM
• The EPP
• Evidence on EPP

• We follow Romer (2001) closely in this

section (chapter 7).

17
• We now want to show an important result which
we took for granted earlier on.
• We now want to better understand where the
following came from:
•
• 1    U ' (ct )

(1  r ) U ' (ct 1 )

• To show the above we shall use a more general

utility function.
• We assume that the utility is no longer
logarithmic (we still assume that it is additive).
18
• Max U ( co ,c1 ,......,cT )
s.t.
T
ct yt
 ( 1  r )t  0 ( 1  r )t

T
U (ct )
Where U (c0 , c1 ,...., ct ,....cT )  
t 0 (1   ) t

19
 T
U (ct )  T yt T
ct 
L     t 
0 (1   ) t
 0 1  r t
0 (1  r ) 
L U ' (ct 1 )  U ' (ct 1 ) (1  r )t 1
  0   ..........(C )
ct 1 1    t 1
(1  r ) t 1
1    t 1
1
L U ' (ct )  U ' (ct ) (1  r )t
  0   ..................( D )
ct 1    (1  r )
t t
1    1
t

Equating C and D y ields :

U ' (ct 1 ) (1  r )t 1 U ' (ct ) (1  r )t

1    t 1
1 1   t 1

1    (1  r )t 1
t

U ' (ct )
1    (1  r )
t 1 t
U ' (ct 1 )


1  
t  ( t 1)
(1  r )t 1t

U ' (ct )
1 1 U ' (ct 1 )


1  

U ' (ct )
(1  r ) U ' (ct 1 )
20
 The Consumption CAPM (CCAPM)
• Our aim in this section is to derive the CCAPM
equation.

• We will have to assume a function for the utility

function.

• The CCAPM from a Quadratic consumption

function.

21
• Recall from the last section an optimising consumer’s FOCs yield:
U (Ct ) 1  r

U (Ct 1 ) 1  
• Let ri be the return on assets i.

• Reducing current consumption by one unit (saving) increases

future consumption. If the individual is optimising then the
current loss from savings is equal to expected future gain.

• Thus:
1
U (Ct )  Et [(1  rt 1 )U (Ct 1 )]
i

1 
22
• Recall the expectation of the product of two variables equals
the product of their expectations plus their covariance.
• That is,
• Cov(X,Y) = E[{X - E(X)}{Y - E(Y)}]
• = E(XY) - E(X)E(Y)
•  E(XY) = Cov(X,Y) + E(X)E(Y)

• Ignoring the subscripts:

• E(1+rt+1)E[U’(Ct+1)] = Cov[(1+rt+1, U’(Ct+1)] + E(1+rt+1)E[U’(Ct+1)]

• Thus: 1

U (Ct )  Et [(1  rt 1 )U (Ct 1 )]
i

1 
• Becomes:
U (Ct ) 
1
1 
Et (1  rti1 ) Et {U (Ct 1 )}  Covt (1  rti1 ,U (Ct 1 ))
23
 U (Ct ) 
1
1 

Et (1  rti1 ) Et {U (Ct 1 )}  Covt (1  rti1 ,U (Ct 1 )) 
• Assuming a quadratic function: U(C) = C – (aC2)/2 , U’(C) = 1 – aC

U (Ct ) 
1
1 

Et (1  rti1 ) Et {U (Ct 1 )}  Covt (1  rti1 , 1  aCt 1 ) 
• Lets closely look at the last term: Cov(1+rt+1, 1-aCt+1)

• We know that Cov(X+a,Y+b) = Cov(X, Y), since a and b are constants.

• Also Cov(aX, bY) = abCov(X,Y).

• In our case Cov(1+rt+1, 1-aC) = Cov(rt+1, -aCt+1)

• = (1).(-a).Cov(rt+1, Ct+1) = -aCov(rt+1, Ct+1)
• Thus:
U (Ct ) 
1
1 

Et (1  rti1 ) Et {U (Ct 1 )}  aCovt (1  rti1 , Ct 1 ) 
24
• Solving for expected returns on assets yields:

Et (1  r ) 
i
t 1
1
Et [U (Ct 1 )]

U (Ct )(1   )  aCov(1  rti1, Ct 1 ) 

25
 Expected Return for a Risk-Free Asset
• We can calculate the above equation in the case of a risk free

aCov(1  r 
asset. Recall that:
i
t 1 , Ct 1 )  0
• Since rt+1 is now our rf ignoring subscripts:
Cov ( r , C )
 rc 
 r C
 rc r  C  Cov( r , C )
for risk - free asset  r  0
• Thus:  Cov ( r , C )  0

1  rf 
1
U (Ct )(1   )
Et [U (Ct 1 )]
• To get the risk premium we subtracting the above equation from
risky equation yields:
aCov (1  rt 1 , Ct 1 )
i
E (r )  rf 
i
t 1
Et [U (Ct 1 ]
26
• Thus the CCAPM equation is:
aCov(1  r , Ct 1 )
i
E (r )  rf 
i
t 1
t 1
Et [U (Ct 1 ]

aCov(1  r , Ct 1 )
i
E (r )  rf 
i
t 1
t 1
Et [U (Ct 1 ]
• Where rf is the risk-free rate.
• The covariance between asset returns and consumption
is known as its consumption beta.
• Central prediction from the CCAPM is: The premiums
that assets offer are proportional to the consumption
beta.
27
 CCAPM: An Alternative Proof
• In above proof we used a quadratic consumption function. Let’s try
a more general case that does not use a quadratic function.
• Model Setup: A representative investor maximizes lifetime utility
that depends only on current consumption and future
consumption. The expected utility from future consumption is
discounted by a discount factor, β. In a two-period model, lifetime
utility is given by: U (Ct , Ct 1 )  U (Ct )  Et [U (Ct 1 )]
• We assume that the investor’s endowment is given by (et , et+1).
That is, at time t the investor receives an exogenous endowment of
resources, denoted by et, which is allocated between consumption
and assets; at time t+1 the investor receives et+1.
• The number of assets that he will buy is denoted by λ and the price
for each unit of the assets is pt. At time t + 1 assets pay a dividend
dt+1 that can be used, along with a new exogenous endowment of
resources (et+1), for consumption purposes in period t + 1.
28
The investor’s problem is:

Max U (Ct , Ct 1 )  U (Ct )  Et [U (Ct 1 )]



Ct  pt  et ............................(1)

Ct 1  et 1   ( pt 1  dt 1 )...............................(2)

29
 Substituting the two constraints into the objective
function, we have:

U (Ct , Ct 1 )  U (et  pt )  Et [U (et 1   ( pt 1  dt 1 ))]

Differentiating with respect to λ yields a first-order condition:

U (Ct )( pt )  EtU (Ct 1 )( pt 1  dt 1 )  0

Dividing all terms by U'(Ct) and adding pt to both sides of
the equation, we get:
U (Ct 1 ) 
pt  Et  ( pt 1  dt 1 )
 U (Ct ) 
U (Ct 1 ) ( pt 1  d t 1 ) 
1  Et  
 U (C t ) pt 
30
 U (Ct 1 )
Let’s define the ratio of marginal utilities as: mt ,t 1 
U (Ct )

 
This changes the above expression to: 1  Et mt ,t 1 pt 1  dt 1 
 pt 

Using the fact that the gross rate of return on a risky asset is
defined as: pt 1  dt 1 we obtain:
1  rj ,t 1 
pt


1  Et mt ,t 1 (1  rj ,t 1 ) 

31
 For any two random variables X and Y, E(XY) = E(X)E(Y) + Cov(X, Y), so

 
Et mt ,t 1 (1  rj ,t 1 )  Et (mt ,t 1 ) Et (1  rj ,t 1 )  Cov(mt ,t 1,1  rj ,t 1 )

1   [ Et (mt ,t 1 ) Et (1  rj ,t 1 )  Cov(mt ,t 1 ,1  rj ,t 1 )]...................(4)

 1   [ Et (mt ,t 1 ) Et (1  rj ,t 1 )]   [Cov(mt ,t 1 ,1  rj ,t 1 )]
1  Covt (mt ,t 1 ,1  rj ,t 1 )
  Et (1  rj ,t 1 )
Et (mt ,t 1 )
1 Covt (mt ,t 1 ,1  rj ,t 1 )
 Et (1  rj ,t 1 )   ..........................(4' )
Et (mt ,t 1 ) Et (mt ,t 1 )

32
If we follow the above procedure for a risk-free asset, then we
must note that a risk-free asset always gives a certain return,
lets denote this by rf, t+1. In the case of a risk-free asset,
Covt(mt, t+1, 1 + rf, t+1) = 0 and equation (4) simplifies to:
1  Et (mt ,t 1 ) Et (1  rf ,t 1 )
1
 1  rf ,t 1  ....................................................(4' ' )
Et (mt ,t 1 )
Subtracting equation (4’’) from equation (4’) yields:

1
Et (rj , t 1 )  rf , t 1   Covt (mt , t 1 ,1  rj , t 1 )
Et (mt , t 1 )
1
Et (rj , t 1 )  rf , t 1   Covt [U (Ct 1 ), 1  rj , t 1 ]
Et [U (Ct 1 )]
33
 1
Et (rj , t 1 )  rf , t 1   Covt [U (Ct 1 ), 1  rj , t 1 ]
Et [U (Ct 1 )]

1
Et (rj , t 1 )  rf , t 1  Covt [U (Ct 1 ), 1  rj , t 1 ]
Et [U (Ct 1 )]
• Above equation shows that the expected excess returns on two
assets i and j will differ only because the covariance of their
respective returns with (the marginal utility of) consumption differ.
• An asset whose return has a negative covariance with the marginal
utility of consumption has a positive covariance with consumption
(thanks to diminishing marginal utility).
• NB: ↑C→↓U'(C)→r↑ (i.e., an increase in consumption reduces
marginal utility of consumption [↑C→↓U'(C)] and if marginal
utility and the return are negatively related a decrease in marginal
utility increases r [so ↓U'(C)→r↑].
• 34
• So an asset whose return has a negative covariance with the
marginal utility of consumption has a positive covariance with
consumption. Such an asset must offer a high expected rate of
return in order for investors to hold this asset.

• Since this asset covaries positively with consumption, it is not

very useful in so far as it does not allow for consumption
smoothing. In other words, it does not offer a high return when
times are bad.

• Basically an asset with a negative covariance with the marginal

utility of consumption, is riskier and investors requires a higher
premium to hold it.
•

35
 The Equity Premium Puzzle (EPP)
• The term EPP refers to the inability of standard neoclassical
economic theory to reconcile the historically large realized
premium of stock market return over the risk free rate, with its
low covariability with aggregate consumption growth.

• EPP - coined by Mehra and Prescott (1985).

• That is, the high equity risk premium (i.e., difference between
return on stock and on government bonds) is ridiculously high
such that reconciling such premiums with classical economic
theory would require extremely risk averse individuals.

36
• The equity risk premium can be stated as:
E (r )  rf   [Cov(r , g )]
• θ is a risk aversion measure, g is the growth in consumption.
• According to theory the risk premia of financial assets are
explained by their covariance with per capita consumption
growth (g).

• Mehra and Prescott(1985) used US historical data and found

the average annual yield on the S&P Index to be around 7%
and the average yield on short-term gvt. debt ≈ 1%.

• This implies an EP = 7 – 1 = 6%.

37
• Standard economic models on capital markets and risk
preferences (e.g., CCAPM) suggests equity premium of less than
0.05%.

• The gap between the observed value of the EP and the values
predicted by CCAPM and other similar models is too huge as to
create the EPP.

• They observed that in order to reconcile the huge difference

between return on equity to government bonds in the US,
individuals must have ridiculously high risk aversion according
to standard economic models.

• The equity risk premium found by Mehra and Prescott (1985)

and by Mankiw and Zeldes (1991) is around 6%, while a
reasonable one that is in line with the economic models is
around 0.5% (Mehra and Prescott, 1985).
38
• In their seminal paper Mehra and Prescott (1985) argue that it is
difficult to reconcile observed data with the above equation. Such a
failure creates the famous Equity Premium Puzzle.
• Mankiw and Zeldes (1991) - Used US data for the period 1890-1979
to calculate the difference between average return on the stock
market and the risk free rate(return on short-term gvt. debt). They
found an equity premium = 6% = 0.06.
• That is, . For that period:
E (r )  r  0.06
f
 g  3.6%  0.036
 r  16.7%  0.167
Recall  rg 
Cov (r , g ) rg  0.40
 r g
  rg r g  Cov (r , g )
Cov (r , g )   rg r g  (0.40)(0.036)(0.167)
 0.0024
• Solving for theta: 0.06 = θ(0.0024), which implies that θ = 25. Which
is too high!!!! 39
 4. Calculating the Equity Premium from
Economic Models
• Assume that individuals have a CRRA utility function. So
we max the following utility function:
T
1 Ct1
U  subject to :
t 1 (1   ) 1  
t

T T
Ct Yt

t 1 1  r 
t

t 1 1  r 
t

T
1 C 1
 T Yt T
Ct 
L t
    t
t 1 1    1     t 1 (1  r ) t 1 1  r  
t t

40
 L Ct 
  0
Ct 1    1  r 
t t


L C 
 t 1
 0
Ct 1 1   t 1
1  r t 1

• The above equations yield; 

C 1 r
t


C t 1 1 

  1  r  
C   Ct 1 .................... * *
1  
t
41
• Inserting expected returns in **:

C t


1
1 
  
Et 1  rti1 Ct1 

 Ct 1 

 
1     Et 1  rt 1  
i

 Ct 

42
• Now let the growth rate of consumption from
t to t+1 be: c Ct 1  Ct Ct 1
gt 1   1
Ct Ct
• So:
Ct 1
gt 1  1 
c

Ct
 

  C
So, 1    Et 1  rt 1  , becomes :
i t 1

 Ct 


1    Et (1  r )( g  1)
i c
t 1


43
 • Taking a second order Taylor approximation of
the RHS around r = g = 0 yields:

E (1  r )( g
t
i c
t 1  1) 

 1
(1  r )( gi c
t 1  1)  1  r  g  gr   (  1) g 2
2


E (1  r )( g
i c
t 1  1) 
 1
 1  E (r )  E ( g )  E ( gr )   (  1) E ( g 2 )
2

44
 • Recall Cov(X,Y)=E[{X-E(X)}{Y-E(Y)}]
• =E(XY)-E(X)E(Y)
• Thus:
• E(XY) = Cov(X,Y) + E(X)E(Y)
• E(gr) = Cov(g,r) + E(g)E(r)
• E(g2) = Cov(g,g) + E(g)E(g) = Var(g) + [E(g)]2
• So;
 
E (1  r i )( g tc1  1)   1  E (r )  E ( g )   [ E ( g ) E (r )  Cov (r , g )]
1
  (  1){[ E ( g )]2  Var ( g )}
2

45
 • So:

E( gr )  Cov( g , r )  E( g ) E(r )
• So: E ( g )  Cov ( g , g )  E ( g ) E ( g )
2

 Var ( g )  [ E ( g )] 2

  1
E (1  r i )( g tc1  1)   1  E (r )  E ( g )   [ E ( g ) E (r )  Cov(r , g )]   (  1){[ E ( g )]2  Var ( g )}
2

46
• For short periods the E(g)E(r) and [E(g)]2 terms are small
so they are approximately zero. Omitting them yields:

 i c
t 1 
E (1  r )( g  1)  1  E (r )  E ( g )   [Cov(r , g )]
1
  (  1){Var ( g )}
2

• Including the LHS as well yields:

1
1    1  E (r )  E ( g )   [Cov(r , g )]   (  1){Var ( g )}
2
1
  E (r )  E ( g )   [Cov(r , g )]   (  1){Var ( g )}
2
47
 • Solving for E(r) yields:
1
E (r )    E ( g )   [Cov(r , g )]   (  1){Var ( g )}.......(* * *)
2
• Let’s consider the above equation for the case of a risk
free asset:
Recall  XY 
Cov( X , Y )   XY  X  Y  Cov( X , Y )
 XY

Cov ( r , g )
So;  rg 
 r g
  rg r g  Cov ( r , g )
Cov ( r , g )   rg r g  0
since  r  0 48
• So;
1
r    E ( g )   (  1){Var ( g )}
2

• Subtracting the r bar equation from the E(R)

equation yields:
E(r )  r   [Cov(r, g )]

• The above equation is the equity premium

equation.

49
 Possible explanations of the EPP
Denial of the puzzle – there is no equity premium (EP).
• There are a number of reasons why some would argue that there is no
equity premium.
• It has been found that in the past 40 years (1969 -2009) there has been
no EP in the US stocks.
• Maybe the representative consumer is much more risk averse than is
normally supposed- not much support for this.
• The general consensus in the extant literature is that the CRRA
coefficient is about 1 and not 25 as is suggested by the data.

Low number of data points: the period 1900–2005 provides only 3.5
independent 30-year windows, which is a very low number from
which to conclude out-performance (over 30-year periods).

50
• Survivor bias - Most studies on this have focused on the US. It
is possible that there is selection bias. The US market was the
most successful stock market in the 20th century. Other
countries' markets displayed substantially lower long-run
returns. Picking the best observation (US) from a sample leads
to upwardly biased estimates of the premium.

• Imperfections in the model of risk aversion – try relaxing the

assumption of expected utility maximization. Epstein and Zin
(1990) did this and found a significant reduction in the EP.

51
• Maximizers vs. Satisficers - We can also relax the assumption
that the consumer fully optimize and instead assume that the
consumer adopts satisficing rules rather than fully optimising.

• Let’s try to differentiate between maximizers and satisficers. A

maximizers need to be assured that every purchase or decision
was the best that could be made. Yet how can anyone truly
know that any given option is absolutely the best possible? The
only way to know is to check out all the alternatives.

• On the other hand, a satisficer has criteria and standards.

He/she searches until he/she finds an item that meets those
standards, and at that point, he/she stops.

52