Study Guide

to accompany

F.-C. Bagliano and G. Bertola

Models for Dynamic Macroeconomics
Oxford University Press, Oxford 2004.

Yu-Fu Chen

Economic Studies
School of Social Sciences
University of Dundee
3 Perth Road
Dundee DD1 4HN
United Kingdom
Email: [Link]@[Link]

Michael Funke

Department of Economics
Hamburg University
Von-Melle-Park 5
20146 Hamburg
GERMANY
Email: funke@[Link]

Dundee & Hamburg, .....

Copyright © 2007 Yu-Fu Chen, Michael Funke

 Chapter 1: Dynamic Consumption Theory

I. Motivational Questions and Exercises:

Exercise 1.1 (p. 3, footnote 1):

In what way does hyperbolic discounting differ from exponential discounting?

Solution: The conventional exponential discounting model weights utility in year T by the discount
factor F(T), where

(1.1) F (T ) = δ T

and δ = 1/(1+ρ). A popular alternative to exponential discounting is hyperbolic discounting, which

captures the commonly observed anomaly of declining impatience, i.e. the fact that discount factors
increase with delay. Figure (1.1) illustrates a hyperbolic discount function with declining impatience. It
is apparent that hyperbolic discounting places an especially large weight on immediate payoffs as
compared to deferred payoffs.

Figure 1.1: Hyperbolic Discount Function

Present
Value

Declining Impatience:
A B
>
B C
A

B
C
Delay (periods)
0 1 2

The hyperbolic formulation for discrete time periods yields discount factors F(T) given by

(1.2) {1 , λδ , λδ 2 , λδ 3 , ... , }
where 0 < λ < 1 and δ < 1. The discount factor terms involving δ are all multiplied by the parameter λ
except in the first period. Assuming additivity over time, an explicit example of hyperbolic utility is

u (c0 ) + u (c1) + u (c2 ) + u (c3) + ...

1 1 1
(1.3)
4 8 16

An alternative flexible formulation proposed by Loewenstein and Prelec (1992) takes the form

(1.4) F (T ) = (1+ ρT )−ω ρ

where ω, ρ > 0. A well-known problem of hyperbolic discounting is that it induces patterns of

inconsistency. What is dynamic consistency? Let´s look at a function to be maximised at some point in

2
the future and once again at a further point in the future. The tail for the initial time that corresponds to
the future actions must be the same as that when you consider the weighting of future actions from the
later point in time (a difference of a scale factor is allowed). Another way of stating this is that if you
maximise over the entire function, you get the same optimal path as you would if you set the first
element and then maximize over the rest of your choices. What is more, the ordering of the future paths
will be the same independent of whether you have not yet maximised the first element or not. When
dynamic consistency does not hold, then one is faced with a rather odd game. It is as if the person
playing the game today is different from the person playing the game tomorrow. Assume, for example,
the utility function (1.4). From t1 onwards, the future looks the same as it did from t0 onwards. Therefore
the hyperbolic utility function looks like

u (c1) + u (c 2 ) + u (c3) + ...

1 1
(1.5)
4 8

which cannot be transformed into the tail from the standpoint of the first period by multiplying the entire
function by any scale factor. On the contrary, exponential discounting characterizes the preferences of a
rational agent, meaning that the preferences are dynamically consistent.

Additional References:

Laibson, D. (1997) “Golden Eggs and Hyperbolic Discounting”, Quarterly Journal of Economics 112,
443-477.

Loewenstein, G. and D. Prelec (1992) „Anomalies in Intertemporal Choice: Evidence and an

Interpretation“, Quarterly Journal of Economics 107, 573-597.

Exercise 1.2:

Consider an American soldier who belongs to a battalion of troops committed to duty in Iraq. He realises
that while being stationed in Iraq he may die and therefore discounts his future according to

u (c0 ) + u (c1) + u (c2 ) + u (c3) + ...

1 1 1
(1.6)
4 8 16

Assume that he emerges from his tour of duty in Iraq alive and well. Once he has returned home, from
time t = 1 onwards, he discounts the future according to

u (c1) + u (c 2 ) + u (c3) + ...

1 1
(1.7)
2 4

Does dynamic consistency hold for this “going to the front” example?

Solution: Equations (1.6) and (1.7) may look hyperbolic, but they are not. Equation (1.7) can be
transformed from (1.6) simply by multiplying by ¼.

Exercise 1.3:

Verify that for the case of the CRRA utility function the absolute value of the elasticity of marginal
utility is γ.

Solution: The elasticity of marginal utility is defined as

3
(1.8)
du ′(c ) c
= =
(
u ′′(c )c − γc −γ −1 c
= −γ
)
dc u ′(c ) u ' (c ) c −γ

Exercise 1.4:

Suppose that the parameter of the CRRA utility function is γ = 1. What happens in this special case?

Solution: Consider the limit

f (γ )
(1.9) lim
γ → 1 g (γ )

and suppose that limγ →1 f(γ) and limγ →1 g(γ) are both zero. L´Hospital´s Rule provides a powerful
method for evaluating such limits and to determine indeterminate forms of type 0/0 or ∞/∞. According
to L´Hospital´s Rule

f (γ ) f ′(γ )
(1.10) lim = lim .
γ → 1 g (γ ) γ → 1 g ′(γ )

Therefore:

f ′(γ ) ⎛ − ln(c)c1−γ ⎞
(1.11) lim = lim ⎜ ⎟ = ln(c) .
γ → 1 g ′(γ ) γ → 1 ⎜⎝ −1 ⎟
⎠

Exercise 1.5:

The Arrow-Pratt measure of absolute risk aversion is defined as − u ′′(c) u ′(c) . The higher the curvature
of u (c) , the higher the risk aversion. Consider an agent with expected utility
E (U ) = κu (c1) + (1 − κ )u (c 2 ) where cS is the consumption level in state s = 1,2 and κ (1 − κ ) is the
corresponding probability assigned to state s. Provide a justification for the Arrow-Pratt measure of risk
aversion by showing that higher values of − u ′′(c ) u ′(c) imply indifference curves which are more
highly bowed.

Solution: An indifference curve I defines c2 as a function of c1

(1.12) c 2 = I (c1) .

The slope of the indifference curve in Figure (1.2) is given by

dc2 dI (c1) κu ′(c1) κu ′(c1)

(1.13) = =− =− .
dc1 dc1 (1 − κ )u ′(c2 ) (1 − κ )u ′(I (c1))
Differentiating (1.13) yields:

d 2 I (c1) κ ⎡ dI ⎤
(1.14) =− ⎢u ′′(c1)u ′(I (c1)) − u ′(c1)u ′′(I (c1)) ⎥ .
d c1 2
(1 − κ ) u ′(c 2) ⎣
2 d c1 ⎦

4
At the 45° degree line c1 = c 2 = I (c1) . Furthermore, dI (c1) d c1 = − κ (1 − κ ) , u ′′(c 2 ) = u ′′(I (c1)) = u ′′(c1)
and u ′(c 2 ) = u ′(I (c1)) = u ′(c1) . Thus, equation (1.14) can be rewritten as

d 2 I (c1) u ′′(c ) ⎡ κ ⎤ κ
= − 1+
u (c ) ⎣ (1 − κ ) ⎥⎦ (1 − κ )
⎢
(1.15) .
d c1 2 ′

Consider the two indifference curves I1 and I2 in Figure (1.2) at point P. Since c1 = c2 the slope of both
indifference curves is d I 1 d c1 = d I 2 d c1 = − κ (1 − κ ) . On the other hand, we have

d 2 I 2 d 2 I1
(1.16) > .
d c12 d c12

The inequality (1.16) is equivalent to

− u ′′1 (c ) − u ′′2 (c )
(1.17) <
u ′1 u ′2

and therefore an agent´s Arrow-Pratt measure of absolute risk aversion increases when his indifference
curves are more bowed inwards, implying that he is less willing to accept risky prospects.

Figure 1.2: Indifference Curves and Arrow-Pratt Measure of Absolute Risk Aversion

c1

I2
I1
45°
c2

Exercise 1.6:

Determine the degree of absolute prudence P (c ) ≡ −u ′′′(c ) / u ′′(c ) and relative prudence
p (c ) ≡ −u ′′′(c)c / u ′′(c ) for the quadratic utility function u (c) = c − (b 2) c 2, b > 0 , the exponential utility
( )
function u (c) = − 1 γ e−γc , γ > 0, and the isoelastic utility function u (c ) = c1−γ − 1 (1 − γ ) , respectively.

Solution: Kimball (1990) measures the strength of the precautionary saving motive. His theory of
precautionary savings is mathematically analogous to the Arrow-Pratt theory of risk aversion. Kimball
measures the absolute prudence by P (c ) ≡ −u ′′′(c ) / u ′′(c ) , and shows that precautionary saving becomes
larger as the value of absolute prudence gets larger.

The derivatives for the three instantaneous utility function are:

b 2
(1.18) u (c) = c − c ; u ′(c ) = 1 − bc ; u ′′(c) = −b ; u ′′′(c ) = 0 ; P(c) = 0 ; p(c) = 0 .
2

5
Since P (c) = p(c) = 0 , agents have no precautionary saving motive. The reason for this is due to the fact
that marginal utility is linear and thus unaffected by mean preserving increases in variance.

1
(1.19) u (c) = − e−γc ; u ′(c) = e−γc ; u ′′(c ) = −γ e−γc ; u ′′′(c ) = γ e−γc ; P (c) = γ ; p (c) = γc .
2
γ

For the exponential utility function, the absolute degree of prudence is constant and unrelated to
consumption while the relative degree of prudence is increasing in c.

c1−γ − 1 γ +1
(1.20) u (c ) = ; u ′(c) = c −γ ; u ′′(c) = −γ c −γ −1 ; u ′′′(c) = γ c −γ − 2 (γ + 1) ; P(c) = ;
1− γ c
p (c ) = γ + 1 .

For the isoelastic utility function, the absolute degree of prudence is decreasing in consumption, while
the relative degree of prudence is constant.

Additional Reference:

Kimball, M. (1990) “Precautionary Saving in the Small and the Large”, Econometrica 58, 53-73.

Exercise 1.7 (p. 5):

Illustrate the derivation of equation (1.5) of the textbook.

Solution: (a) The first (and more general) way is via dynamic programming. It is obvious that we can
rewrite the objective function as

⎡ ∞ u (c ) ⎤
(1.21) Max U t = Et ⎢u (ct ) + ∑ t +i
⎥.
ct + i
⎣⎢ i =1 (1 + ρ )i ⎦⎥
∞ u (ct +i ) 1 ∞ u (ct +1+i )
If we re-index ∑ from i = 0, instead of i = 1, we have ∑ , which can be written
i =1 (1 + ρ )i (1 + ρ ) i =0 (1 + ρ )i
U (.) . Thus,
1
as
(1 + ρ ) t +1

⎡ ⎤
(1.22) Max U t = Et ⎢u (ct ) + U t +1 (.)⎥ .
1
ct + i ⎣ (1 + ρ ) ⎦

Since c is the control variable and after the maximization is no longer a function of state variable A, we
find that Vt , the intertemporal utility after maximization, is a function of At and that Vt +1 is a function
of At +1 . Thus the objective function, in terms of V, becomes:

⎡ ⎤
(1.23) Vt ( At ) = Max ⎢u (ct ) + E t Vt +1 ( At +1 )⎥ ,
1
ct ⎣ (1 + ρ ) ⎦

which is the standard Bellman equation for discrete-time cases. The first order condition gives

6
 dE A
(1.24) u ′(ct ) + Et Vt′+1 ( At +1 ) t t +1 = 0 .
1
1+ ρ dct

dAt +1
We have = −1 , according to state equation: At +i +1 = (1 + r )At +i + yt +i − ct +i . Thus, we have
dct

(1.25) u ′(ct ) = E t Vt′+1 ( At +1 ) .

1
1+ ρ

The envelope theorem for Vt ( At ) ≡ U (c t ( At ), At ) for optimised c gives

dVt ( At ) ∂U (ct ( At ), At ) ∂u (ct ) ∂u ((1 + r )At + y t − At +1 )

(1.26) ≡ = = = u ′(ct )(1 + r ) .
dAt ∂At ∂At ∂At

The above equation also implies for the period t+1 that

(1.27) E t Vt′+1 ( At +1 ) = E t u ′(ct +1 )(1 + r ) .

Substituting into equation (1.25) gives

1+ r
(1.28) u ′(ct ) = E t u ′(ct +1 ) ,
1+ ρ

which is equation (1.5).

(b) Equation (1.5) can also be derived using the ordinary Lagrangian method. We can construct the
Lagrangian for the objective function

∞ u (ct +i ) ∞
(1.29) L = Et ∑ + Et ∑ λi ((1 + r )At +i − At +i +1 + y t +i − ct +i ) .
i =0 (1 + ρ ) i
i =0

It is obvious that we cannot compare the Lagrangian multipliers in different periods due to the
depreciation (discounting by r) of A, in terms of values at t. Thus, it is convenient to create an new
variable such that

(1.30) Μ i = (1 + r )i λi .

The reason for this trick is that the multipliers depreciate at a rate of r. The Lagrangian now becomes

⎡ ∞ ⎡ u (ct +i ) Μi ⎤⎤
(1.31) L = E t ⎢ ∑ ⎢ + ((1 + r ) At + i − At + i + + y t + i − c t + i ) ⎥⎥ .
⎢⎣i =0⎢⎣ (1 + ρ ) (1 + r )i
i 1
⎥⎦ ⎥⎦

The first order condition for the consumer at periods t and t+1 is

(1.32) u ′(ct ) = Μ 0 ,

E t u ′(ct +1 ) Μ 1
(1.33) = .
1+ ρ 1+ r

7
The multiplier should be comparable, and indeed equal, for the adjusted M so that Μ 0 = Μ 1 . Thus, we
obtain equation (1.5):

1+ r
(1.34) u ′(ct ) = E t u ′(ct +1 ) .
1+ ρ

Exercise 1.8 (p. 24):

Derive the Euler equation (1.44).

Solution: Expanding u ′(ct +1 ) around the value of ct for ct +1 by Taylor expansion up to the second
order yields

(1.35) u ′(ct +1 ) = u ′(ct ) + u ′′(ct )(ct +1 − ct ) + u ′′′(ct )(ct +1 − ct )2 .

1
2

Taking expectations on both sides gives

(1.36) Et u ′(ct +1 ) = u ′(ct ) + u ′′(ct )Et (ct +1 − ct ) + u ′′′(ct )Et (ct +1 − ct )2 .

1
2

By the Euler equation E t u ′(c t +1 ) = u ′(ct ) on p. 24, we then have equation (1.44).

Exercise 1.9 (p. 26):

Derive the first-order condition (1.47).

Solution: This is similar to equation (1.5) and it still uses the same objective function and difference
equation. Thus, it must yield the same results as equation (1.5).

1+ r
(1.37) u ′(ct ) = E t u ′(ct +1 ) .
1+ ρ

Substituting equation (1.45) into the above equation and using the assumption r = ρ at the end of p. 25
gives equation (1.47)

(
(1.38) e −γct = E t e −γct +1 . )
Exercise 1.10 (p. 28):

Derive the first-order condition (1.57).

b 2σ 2
Solution: The lognormal distribution is given by E e bx = e ( ) b μ+
2 [ ]
, if x ~ N μ , σ 2 (see the Derivation
[
below). Therefore, if ε ~ 0, σ ε , then
2
]
[ ]
(1.39) E t e −γθε = e −γθ 0 + (−γθ )
2
σ ε2 2
= eγ θ σε 2
2 2 2

8
Thus,

(1.40) K t =
1
γ
( [ ])
ln E t e −γθε =
1
ln⎜⎛ e γ
γ ⎝
θ σε 2 ⎞
2 2 2
⎟=
⎠ γ
1 γ 2θ 2σ ε2 γθ 2σ ε2
2
=
2

Derivation: Suppose a random variable x is normal with mean μ and standard deviation σ. Then,

∞ ⎡ (x − μ ) ⎤
E [exp(bx )] = ∫ exp(bx ) exp ⎢−
1
⎥ dx
σ 2π −∞ ⎣ 2σ 2 ⎦
exp(bμ ) ∞ ⎡ y2 ⎤
= ∫ exp ⎢by − ⎥ dy
σ 2π −∞ ⎢⎣ 2σ 2 ⎥⎦
⎛ 1 ⎞
(1.41) =
exp⎜ bμ + b 2σ 2 ⎟ ∞
⎝ 2
∫
⎡
⎠ exp ⎢− y − bσ (2
) ⎤⎥ dy
2

σ 2π −∞ ⎢ 2σ 2 ⎥
⎣ ⎦
⎛ 1 ⎞
exp⎜ bμ + b 2σ 2 ⎟
= ⎝ 2 ⎠
σ 2π
⎛ 1 ⎞
= exp⎜ bμ + b 2σ 2 ⎟.
⎝ 2 ⎠

The first line is merely the definition of the expectation using the standard formula for the normal
density. The second line transforms this into a new variable y = x − μ which is also normally
distributed. The third line then completes a square, while the fourth uses the fact y − bσ 2 is also normal
so its density integrates out to unity.

Exercise 1.11 [Blanchard and Fisher (1989), Problem …, pp. 289-290]:

Consumers maximise expected utility

⎡T −1 1 ⎤
(1.42) max E ⎢ ∑ e−γc ⎥
⎣ t =0 γ ⎦

subject to

(1.43) At +1 = At + yt − ct

and

(1.44) yt = yt −1 + et (
et ~ N 0, σ .
2
)
Derive the Euler equation marginal utility in period t = expected marginal utility in period t+1 for
consumption under rational expectations analytically and plot the relationship between consumption in
period t and period t+1 for different risk levels.

Solution: We can just use an ordinary Lagrangian to solve the system:

9
 T −1 ⎛ 1 ⎞ T −1
L = E ∑ ⎜ − ⎟e −αCt + E ∑ λi [ At − At +1 + Yt − C t ]
i =0 ⎝ α ⎠ i =0
(1.45)
T −1 ⎡⎛ 1 ⎞ ⎤
= E ∑ ⎢⎜ − ⎟e −αCt + λi [ At − At +1 + Yt − Ct ]⎥.
i = 0 ⎣⎝ α ⎠ ⎦

The first order condition for the consumer at periods t and after is,

(1.46) e −αC0 = λ 0 , Ee −αC1 = λ1 , Ee −αC2 = λ 2 , …,

The multipliers should be the same since there is no interest/discount rate for A. Thus

(1.47) Ee −αC1 = ... = Ee −αCT −1 = e −αC0 .

Alternatively, one can transform At +1 = At + Yt − Ct into the following equation:

T −1 T −1
(1.48) AT = A0 + ∑ Yt − ∑ Ct .
t =0 t =0

Since the consumer only lives for T periods, we need the transversality condition that AT = 0 . Thus, the
life-long budget constraint becomes

T −1 T −1
(1.49) A0 + ∑ Yt = ∑ Ct .
t =0 t =0

The Lagrangian now becomes

T −1 ⎛ 1 ⎞ ⎡ T −1 T −1 ⎤
(1.50) L = E ∑ ⎜ − ⎟e −αCt + λ ⎢ A0 + ∑ Yt − ∑ Ct ⎥ .
i =0 ⎝ α ⎠ ⎣ t =0 t =0 ⎦

The first order conditions are the same as those we already have:

(1.51) Ee −αC0 = Ee −αC1 = ... = Ee −αCT −1 = λ .

Equation (1.49) indicates that C and Y have the same error terms/random walk.

(1.52) Ct = ECt + ε t ,

( ) [ ]
where ε t ~ N 0, σ 2 and Ct ~ N ECt , σ 2 . From the first order conditions, for periods 0 and 1, we
have

(1.53) Ee −αC1 = e −αC0

The lognormal distribution (see Exercise 1.10) implies

[ ]
b2σ2

( )=e
bμ+
(1.54) E e bx 2
, if x ~ N μ , σ 2 .

[
Therefore, if Ct ~ N ECt , σ 2 , then ]

10
(1.55) [ ]
E e −αC1 = e −αEC1 +α σ
2 2
2
= e −αC0 .

Taking logs on both sides,

(1.56) − EC1 + ασ 2 2 = −C 0

and substituting (1.52) into (1.56) gives

ασ 2
(1.57) C1 = C 0 + + ε1 .
2

It is then straightforward to obtain

ασ 2
(1.58) C t +1 = C t + + ε t +1 .
2

Rewriting the budget constraint leads to

T −1 T −1
(1.59) At + ∑ Yτ = ∑ Cτ
τ =t τ =t

Taking expectations yields

T −1 T −1
(1.60) At + ∑ Et Yτ = ∑ Et Cτ .
τ =t τ =t

We know that

(1.61) E t Yt +1 = ... = E t YT −1 = Yt .

And from (1.58) we have

sασ 2
(1.62) E t C t + s = C t + .
2

Substituting (1.61) and (1.62) into (1.60) yields

ασ 2 (T − t )(T − t − 1)
(1.63) At + (T − t )Yt = (T − t )C t + .
2 2

Rearranging finally gives

⎛ 1 ⎞ α (T − t − 1)σ 2
(1.64) Ct = ⎜ ⎟ t
A + Yt − .
⎝T −t ⎠ 4

Equation (1.64) quantifies the sensitivity of consumption to the volatility level. The crux of the answer is
that the relationship has a rather simple form – the slope is proportional to the variance of et. The
proportionality factor is positive if utility is convex.

11
 Figure 1.3: Consumption Profiles and Risk

ct

High Risk
Moderate Risk
No Risk

t
t t+1

With convex marginal utility, increasing uncertainty about consumption in period t+1 increases expected
marginal utility. To restore the Euler equation, marginal utility today must go up and/or marginal utility
tomorrow must fall. This is achieved by reducing consumption today, i.e. increasing precautionary
saving. Put differently, the consumption path is steeper when agents are exposed to higher risk. Without
risk (σ² = 0), the consumer wants to smooth consumption perfectly over time. Thus, the slope of the ct
path is zero. The higher the risk is, however, the stronger is the precautionary saving motive, so the
slope becomes steeper as σ² rises.

Additional Reference:

Blanchard, O.J. and S. Fisher (1989) Lectures on Macroeconomics, Cambridge (MIT Press).

Exercise 1.12:

Consider aggregate consumption over time. How does the consumption profile change when the
(idiosyncratic) risk levels shifts to a higher level? Sketch the contour of consumption in a graph.

Solution: In general one should expect that prudent consumers save more when the risk increases.
Higher savings imply higher future wealth and therefore increased risk is expected to cause the growth
rate of consumption to rise. Furthermore, when risk is idiosyncratic, then an increase in risk increases
the actual, not only the expected, consumption profile.

12
 Figure 1.4: Aggregate Consumption and Increasing Risk
Aggregate
Consumption

Time

Shift to higher risk

Exercise 1.13:

During the far-reaching economic transition process in Mainland China both the government and SOEs
have, since the 1980s, reduced the social welfare net they provide. This expenditure has now become an
individual responsibility: (a) in urban areas only about half the population is covered by basic health
insurance, while in rural areas only about 20 percent of the population is covered by cooperative health
insurance programmes; (b) the share of China’s workforce in 2004 covered by unemployment insurance
is only about 15 percent; (c) the basic government pension scheme is extremely limited and provides a
pension equal to only 20 percent of average local wages; and finally (d) families are also responsible for
a significant share of education expenditure, since government expenditure on education amounts to
only 2 percent of GDP. What type of savings behaviour would you expect in light of the precautionary
savings motive during the reform period? How has the Chinese savings rate developed over time?

Solution: Given the high level of uncertainty [high σ in equation (1.64)], the savings rate can be
expected to be high in order to cope with life’s uncertainties such as unemployment or illness.
Furthermore families save because school fees are a large burden.

Figure 1.5: Gross savings in Percent of GDP in Mainland China, 1980 - 2004
44

42

40

38

36

34

32

30
80 82 84 86 88 90 92 94 96 98 00 02 04

Gross Savings in % of GDP

Since the beginning of the 1980s, the Chinese gross saving rate has oscillated around very high levels by
any international standard. By contrast, for example, in the U.S. consumers spent more than their
disposible income in 2005, i.e the saving rate was slightly negative.
The difference between China´s consumption-to-GDP ratio and that in other countries is even higher
when government consumption expenditures for education and healthcare are taken into account. While

13
some advanced countries, including Australia and Canada, have quite modest personal consumption
shares, they often reflect that households receive substantial publicly provided goods like education and
healthcare that are not included in measures of personal consumption in national accounts.

Table 1.1: Consumption Shares Across Countries

(in percent of GDP, 2004)
Personal Government Consumption Σ
Consumption on Education and Healthcare
US 70 10 80
UK 65 12 77
Australia 60 11 71
Canada 56 7 63
Korea 51 5 56
France 56 6 62
Germany 57 6 63
Italy 60 12 72
Japan 57 5 62
India 67 4 70
China 41 3 44
Source: Aziz, J. (2006) “Rebalancing China´s Economy: What Does Growth Theory Tell US?”, IMF Working
Paper WP/06/291, Washington.

Exercise 1.14 (p. 35):

In the textbook the subtractive form of a habit consumption function is used, in which outer utility is
derived from the difference between current consumption and the habit stock,
u (c, χ ) = (c − χ )1−γ (1 − γ ) where χ is the level of habit. Alternatively, assume now that the habit
consumption function is given by the multiplicative form

1−γ
⎛ c ⎞
⎜ ⎟
⎜ χφ ⎟
(1.65) u (c, χ ) = ⎝ ⎠
1− γ

(a) Interpret the parameter φ and (b) Demonstrate that the relative degree of risk aversion is still given
by γ.

Solution: (a) If φ = 0, the model collapses to the standard CRRA modelling approach and habits are
irrelevant. For φ = 1, consumers only care about how consumption compares to their habits, i.e. the level
of consumption is irrelevant. (b) The relative degree of risk aversion is

⎛ γ χ −φ
⎜
− ⎜−
(
cχ −φ
−γ
) ⎞
⎟
⎟c
⎜ c ⎟
⎝ ⎠ =γ .
( )
(1.66) −γ
χ −φ cχ −φ

Exercise 1.15 (Blanchard and Fisher (1989), Problem 5, pp. 309-310):

14
Consider a consumer facing a stochastic taste shock. The consumer maximises the intertemporal linear-
quadratic utility function

⎡ ∞ (a + et )Ct − bCt2 ⎤
(1.67) E⎢ ∑ ⎥
⎢⎣t =0 (1 + ρ )t ⎥⎦

subject to

(1.68) At +1 = (1 + r )( At + Y t − C t ) ,

where E[⋅] is the expectations operator, C is consumption, Y is labour income, A is private wealth, r is
the interest rate, and ρ is the discount rate. The disturbance term et captures stochastic taste shocks,
shocks to marginal utility. For simplicity assume ρ = r.

(a) Derive the first order conditions.

( )
(b) Assume that et follows the first-order autoregressive process et = α et −1 + ε t with ε t ~ N 0,σ 2 and 0
< α < 1. Characterise the resulting optimal consumption path.
( )
(c) Alternatively assume that et follows the random walk et = et −1 + ε t with ε t ~ N 0,σ 2 . Again
characterise the resulting optimal consumption path.

Solution:

Subtask (a): The Euler equation requires

(1.69) U ′(C t ) = E [U ′(C t +1) | t ]

Since (1.69) is valid for all t, we have

∂U (C t )
(1.70) U ′(C t ) = = (a + et ) − 2bC t
∂ Ct

Rearranging equation (1.70) leads to

(1.71) (a + et ) − 2bC t = E [(a + et +1) − 2bC t +1 | t ]

(1.72) (a + et ) − 2bC t = a + E [et +1 | t ] − 2bE [C t +1 | t ]
(1.73) et − 2bC t = E [et +1 | t ] − 2bE [C t +1 | t ]

(1.74) {E[C t +1 − C t | t ]} = 1
E[et +1 − et | t ] .
2b

Using backward iteration we finally have

(1.75) E [C t +1 | t ] = C t + E[et +1 − et | t ]
1
2b

(1.76) E[E[C t +1 | t ] | t − 1] = E [C t | t − 1] + E[E [et +1 − et | t ] | t − 1]

1
2b

15
(1.77) E[C t +1 | t − 1] = E[C t | t − 1] + E [et +1 − et | t − 1]
1
2b

(1.78) E [C t +1 | t − 1] = C t −1 + E [et − et −1 | t − 1] + E[et +1 − et | t − 1]

1 1
2b 2b

(1.79)

E [E [C t +1 | t − 1] | t − 2] = E [C t −1 | t − 2] + E[E[et − et −1 | t − 1] | t − 2] + E [E [et +1 − et | t − 1] | t − 2]
1 1
2b 2b

(1.80) E[C t +1 | t − 2] = C t −2 + E[et −1 − et −2 | t − 2] + E[et − et −1 | t − 2] + E[et +1 − et | t − 2]

1 1 1
2b 2b 2b

(1.81) E [C t +1 | 0] = C 0 + E [et +1 | 0]
1
2b

Since E [et +1 | t ] = 0 , we obtain E [C t +1 | t ] = C t , i.e. expected consumption is constant over time. The
budget constraint then implies that consumption at each point in time is equal to lifetime wealth divided
by the length of life.

Subtask (b): The dynamics of consumption can be derived in a conceptually straightforward manner,
although the algebra becomes rather lengthy and tedious. The law of iterated projection yields

(1.82) At +1 = (1 + r )( At + Y t − C t )

(1.83) At +1 = (1 + r )((1 + r )( At −1 + Y t −1 − C t −1) + Y t − C t )

(1.84) At +1 = (1 + r )((1 + r )((1 + r )( At − 2 + Y t − 2 − C t − 2 ) + Y t −1 − C t −1) + Y t − C t )

M
t
(1.85) At +1 = (1+ r )
t +1
A0 + ∑ (1+ r )
t +1−i
(Y i − C i )
i =0

At
As t → ∞ and assuming the transversality condition that lim = 0 gives us
t →∞ (1 + r )t
A∞ ∞
(1.86) = A0 + ∑ (1 + r )−i (Yi − Ci )
(1 + r )∞ i =0

∞ ∞
A0 + ∑ (1+ r ) Y i = ∑ (1+ r ) C i
−i −i
(1.87)
i =0 i =0

∞ ∞
∑ (1+ r ) E [C i | 0] = A0 + ∑ (1+ r ) E [Y i | 0]
−i −i
(1.88)
i =0 i =0

Substituting (1.81) into (1.88) gives

16
 ∞ ⎞ ∞
−i ⎛
∑ (1+ r ) ⎜ C 0 + E [ei | 0]⎟ = A0 + ∑ (1+ r )−iE [Y i | 0]
1
(1.89)
i =0 ⎝ 2b ⎠ i =0

−1 ∞
⎛ r ⎞ 1 ∞
∑ (1+ r ) E [ei | 0] = A0 + ∑ (1+ r ) E [Y i | 0]
−i −i
(1.90) ⎜ ⎟ C0 +
⎝ 1+ r ⎠ 2b i = 0 i = 0

⎛ r ⎞⎡ ∞ 1 ∞ ⎤
⎟ ⎢ A0 + ∑ (1+ r ) E [Y i | 0] − ∑ (1+ r ) E [ei | 0]⎥
−i −i
(1.91) C 0 = ⎜
⎝ 1 + r ⎠⎣ i =0 2b i =0 ⎦

Since (1.91) holds in all periods, we can finally write consumption in period t as

⎛ r ⎞⎡ ∞ 1 ∞ ⎤
⎟ ⎢ At + ∑ (1+ r ) E [Y t +i | t ] − ∑ (1+ r ) E [et +i | t ]⎥ .
−i −i
(1.92) C t = ⎜
⎝ 1 + r ⎠⎣ i =0 2b i =0 ⎦

Thus the change in consumption from t-1 to t is

⎡ ∞ 1 ∞ ⎤
At + ∑ (1+ r ) E [Y t +i | t ] − ∑ (1+ r ) E [et +i | t ] −
−i −i
⎢ b ⎥
⎛ r ⎞⎢ 2
(1.93) C t − (1 + r ) C t −1 = ⎜
i =0 i =0
⎟ ⎥
⎝ 1 + r ⎠ ⎢(1 + r )⎡ ∞ 1 ∞ ⎤⎥
⎢ At −1 + ∑ (1+ r ) E [Y t +i −1 | t − 1] − 2b ∑ (1+ r ) E [et +i −1 | t − 1]⎥ ⎥
−i −i
⎢
⎣ ⎣ i =0 i =0 ⎦⎦

⎡ ⎤
⎢ − (1 + r ) ⎥
⎢ At At −1 − (1 + r )Y t −1 + ⎥
⎛ r ⎞⎢ ∞ 1 ∞ ⎥
(1.94) C t − (1 + r ) C t −1 = ⎜ ⎟ ⎢ ∑ (1+ r ) E [Y t +i | t ] − ∑ (1+ r ) E [et +i | t ] −
−i −i
⎥
⎝ 1 + r ⎠ ⎢i = 0 2b i =0 ⎥
⎢⎡ ∞ + r ∞ ⎤⎥
⎢ ⎢ ∑ (1+ r ) E [Y t +i | t − 1] − ∑ (1+ r ) E [et +i −1 | t − 1]⎥ ⎥
−i 1 −i

⎣ ⎣i = 0 2b i =0 ⎦⎦

Using the fact that At − (1 + r ) At −1 − (1 + r )Y t −1 = −(1 + r ) C t −1 finally leads to

⎡∞ ⎤
∑ (1+ r ) (E [Y t +i | t ] − E [Y t +i | t − 1]) − ⎥
−i
⎢
⎛ r ⎞ ⎢i = 0 1
(1.95) C t − C t −1 = ⎜ ⎟ ⎥+ ret −1 .
⎝ 1 + r ⎠⎢ 1 ∞
( [ ] [ ])⎥ 2b
⎢⎣ 2b i∑ (1+ r ) et +i
−i
E | t − E et +i | t − 1 ⎥⎦
=0

The innovation et is assumed to follow the AR(1) process et = αet −1 + ε t . Through similar manipulations
as in Blanchard and Fisher (1989), p. 287 we have

(1.96)

⎛ ⎡∞ ⎤ ⎡∞ ⎤⎞ ∞
[ ] [ ]
∑ (1+ r ) ⎜⎜ E ⎢ ∑ α j ε t +i − j | t ⎥ − E ⎢ ∑ α j ε t +i − j | t − 1⎥ ⎟⎟ = ∑ (1+ r ) ∑ α j (E ε t +i − j | t − E ε t +i − j | t − 1 )
∞ ∞
−i −i
i =0 ⎝ ⎣ j =1 ⎦ ⎣ j =1 ⎦ ⎠ i =0 j =1

17
The expression in the last parenthesis is zero for j > i because the shocks for innovations prior to t-1 are
identical (the additional information in t is irrelevant because the shocks are already known in t-1).
Collecting the remaining terms gives us

∞ ∞ εt
∑ (1+ r ) α i ε t = ε t ∑ (1+ r ) α i =
−i −i
(1.97) .
t =0 t =0 α
(1 + r )
It is then straightforward to verify that the change in consumption in (1.95) simplifies to

⎛⎡ r ⎤ ⎞
⎜⎢ ⎥ ⎟
⎛ r ⎞⎡ ∞ ⎤ 1 ⎜ (1 + r ) ⎟
⎟ ⎢ ∑ (1+ r ) (E [Y t +i | t ] − E [Y t +i | t − 1])⎥ −
−i
(1.98) C t − C t −1 = ⎜ ⎢ ⎥ ε t − ret −1 ⎟
⎝ 1 + r ⎠ ⎣i = 0 ⎜
⎦ 2b ⎢1 − α ⎥
⎜⎜ ⎢ (1 + r ) ⎥ ⎟⎟
⎝⎣ ⎦ ⎠

Equation (1.98) is intuitive. It implies that measured consumption is correlated, i.e. the change in
consumption today provides us with some information as to what the change in consumption is likely to
be tomorrow.

Subtask (c): When et is assumed to follow a random walk, then α = 1. Using (1.98), the following
calculations then determine the dynamics of consumption:

⎛⎡ r ⎤ ⎞
⎜⎢ ⎥ ⎟
⎛ r ⎞⎡ ∞ ⎤ 1 ⎜ (1 + r )
(1.99) C t − C t −1 = ⎜ ⎟ ⎢ ∑ (1+ r ) (E [Y t +i | t ] − E [Y t +i | t − 1])⎥ −
−i
⎢ ⎥ ε t − ret −1 ⎟
⎝ 1 + r ⎠ ⎣i = 0 ⎜
⎦ 2b ⎢1 − 1 ⎥ ⎟
⎜⎜ ⎢ ⎟⎟
⎝ ⎣ (1 + r ) ⎦
⎥
⎠

⎛ r ⎞⎡ ∞ ⎤ 1
⎟ ⎢ ∑ (1+ r ) (E [Y t +i | t ] − E [Y t +i | t − 1])⎥ − (ε t − ret −1)
−i
(1.100) C t − C t −1 = ⎜
⎝ 1 + r ⎠ ⎣i = 0 ⎦ 2b

∞
Using the fact that et = et −1 + ε t = et − 2 + ε t −1 + ε t = L = ∑ ε t −i equation (1.100) can be rewritten as
i =0

⎛ r ⎞⎡ ∞ 1 ⎤ 1 ⎛ ∞ ⎞
(1.101) C t − C t −1 = ⎜ ⎟⎢ ∑ ( E [Y t + i | t ] − E [Y t + i | t − 1])⎥ − ⎜ ε t − r ∑ ε t −1−i ⎟
⎝ 1 + r ⎠ ⎢⎣i =0 (1+ r )i
⎥⎦ 2b ⎝ i =0 ⎠

When the error term follows a random walk, successive changes in consumption are uncorrelated.

Additional Reference:

Blanchard, O.J. and S. Fisher (1989) Lectures on Macroeconomics, Cambridge (MIT Press).

Exercise 1.16 (p. 41):

Suppose consumers have the following utility function for durable goods (there are no non-durable
goods like in Exercise 6 on p. 41 in the textbook):

18
 t
∞ ⎛ 1 ⎞
(1.102) U t = ∑ ⎜⎜ ⎟⎟ u (S t )
t =0 ⎝ 1+ ρ ⎠

where

(1.103) S t = (1 − δ ) S t −1 + ct ,

where St is the stock of durable goods, δ is the depreciation rate and ct is the purchase of new durable
goods. The accumulation constraint takes the form

(1.104) At +1 = (1 + r )( At + yt − ct ) .

We assume that future wage incomes yt are uncertain and consumer’s have access to financial markets
with no borrowing constraints.

(a) Set up the consumer’s utility maximisation problem and obtain the Euler equation.
⎛ 1+ r ⎞
(b) Show that for a quadratic utility function and ⎜⎜ ⎟⎟ = 1 , the innovations in consumption
⎝1 + ρ ⎠
follow a MA(1) process. Interpret the result.

Solution:

Subtask (a): The stock of durables in period t is

∞
S t = ∑ (1−δ ) ct − j .
j
(1.105)
j =0

The consumer’s problem can therefore be rewritten as

t
∞⎛ 1 ⎞ ⎛∞ ⎞
(1.106) max E 0 ∑ ⎜⎜ ⎟⎟ u ⎜ ∑ (1−δ ) j ct − j ⎟
ct , At +1 t =0⎝ 1+ ρ ⎠ ⎝ j =0 ⎠

subject to the accumulation constraint (1.104). The first-order condition of the Lagrangian with respect
to ct gives

j
∞ ⎛ 1 ⎞
(1.107) ∑ ⎜⎜ ⎟⎟ (1−δ ) j E t u ′(S t + j ) = λ t ,
j =0⎝ 1+ ρ ⎠

while the first-order condition for At+1 is

(1.108) λ t = (1 + r ) λ t +1

where λt is the associated multiplier for the restriction. Thus we have

19
(1.109)

j
∞ ⎛ 1 ⎞
∑ ⎜⎜ ⎟⎟ (1−δ ) j E t u ′(S t + j ) = (1+ r )λ t +1
j = 0⎝ 1+ ρ ⎠
j j
⎛ 1 ⎞ ⎛ 1+ r ⎞ ∞ ⎛ 1 ⎞
⎟⎟ (1−δ ) j E t u ′(S t + j ) = ⎜⎜ ⎟⎟ (1−δ ) j E t u ′(S t + j +1)
∞
⇔ ∑ ⎜⎜ ⎟⎟ ∑ ⎜⎜
j = 0⎝ 1+ ρ ⎠ ⎝ 1 + ρ ⎠ j =0⎝ 1+ ρ ⎠
j 2 j
⎛ 1 ⎞ ∞ ⎛ 1 ⎞ ⎛ 1 ⎞ ∞ ⎛ 1 ⎞
⇔ ⎜⎜ ⎟⎟(1 − δ ) ∑ ⎜⎜ ⎟⎟ (1−δ ) j E t u ′(S t + j +1) = ⎜⎜ ⎟⎟ (1 − δ )(1 + r ) ∑ ⎜⎜ ⎟⎟ (1−δ ) j E t u ′(S t + j + 2 )
⎝ 1 + ρ ⎠ j = 0⎝ 1+ ρ ⎠ ⎝ 1+ ρ ⎠ j = 0⎝ 1+ ρ ⎠
⎛ 1+ r ⎞
⇔ u ′(S t ) = ⎜⎜ ⎟⎟ E t u ′(S t +1) .
⎝1 + ρ ⎠

Subtask (b): With a quadratic utility function and r = ρ, marginal utility is linear and therefore (1.109)
becomes

(1.110) S t = E t S t +1 ⇔ S t +1 = S t + ε t ,

where E t −1 ε t = 0 . From (1.103) we have

(1.111) ct = S t − (1 − δ ) S t −1 ⇔ ct −1 = S t −1 − (1 − δ ) S t − 2 ⇔ Δct = ΔS t − (1 − δ ) ΔS t −1 ,

where Δ is the first-difference operator. From (1.110) we know that ΔS t = ε t for all t and thus

(1.112) Δct = ε t − (1 − δ )ε t −1 .

The intuition is clear. Let’s assume that a positive shock ε t > 0 occurs and therefore St is increasing.
Then consumers don’t have to buy as much durables in period t+1 as they still have the undepreciated
part they have obtained in period t. Finally, in the special case of full depreciation within one period
(δ = 1) we obtain the textbook result that consumption follows a random walk, i.e. Δct = ε t .

II. Software Tools

Software Exercise 1.1 [Alessie and Lusardi (1997)]:

According to Alessie and Lusardi (1997), the consumer maximizes the utility function

∞ 1 e −θ (cτ −γcτ −1 )
(1.113) max E t ∑ − ,
τ =t θ (1 + ρ )t −τ
subject to the following intertemporal budget constraint

∞ cτ ∞ yτ
(1.114) ∑ = (1 + r )At −1 + ∑ ,
τ =t (1 + r ) τ −t
τ =t (1 + r )τ −t
where c is the consumption, y is the non-capital income, At −1 is the non-human wealth at the end of
period t-1, θ is a positive parameter, ρ is the time preference rate, r is real interest rate, γ is a positive

20
parameter related to consumption at the previous period and 0 < γ < 1 . Habitual behaviour is usually
slow to change and therefore past behaviour exercises influences over current behaviour that is
“habitual”. The solution of the consumption process is determined by

γ ⎛ γ ⎞ r ∞ t −τ τ
(1.115) ct = ct −1 + ⎜⎜1 − ⎟⎟Y pt − ∑ (1 + r ) ∑ Γ j −1 ,
(1 + r ) ⎝ (1 + r ) ⎠ (1 + r ) τ =t +1 j =t +1

where

(1.116) Γ j −1 =
1
θ
(
ln E j −1 exp − θψ * w j ,)
⎛ γ ⎞ r ∞
∑ψ (1 + r ) ,
−t
(1.117) ψ * = ⎜⎜1 − ⎟
⎝ (1 + r ) ⎟⎠ (1 + r ) t =0 t

r ⎛ ∞ ⎞
⎜ (1 + r )At −1 + ∑ (1 + r ) Et yτ ⎟
t −τ
(1.118) Y pt =
(1 + r ) ⎝ τ =1 ⎠

and

(1.119) E t yτ − E t −1 yτ = ψ τ −t wt , ψ 0 = 1

where wt is an innovation disturbance. Non-capital income is assumed to follow a stochastic process

with MA(1) shocks,

(1.120) y t = y t −1 + wt ,

(1.121) wt = et + αet −1 .

Thus,

(1.122) y t +1 = y t + et +1 + αet = y t −1 + et + et +1 + α (et −1 + et ) ,

(1.123) y t + 2 = y t +1 + et + 2 + αet +1 = y t −1 + et + et +1 + et + 2 + α (et −1 + et + αet +1 ) .

From (1.119), we have

Et yt − Et −1 yt
(1.124) ψ0 =1= .
wt

Substituting (1.122) into (1.119) yields

21
 E t y t +1 − Et −1 y t +1
ψ1 =
wt
E t [ y t −1 + et + et +1 + α (et −1 + et )] − E t −1 [ y t −1 + et + et +1 + α (et −1 + et )]
(1.125) =
wt
E t [ y t −1 ] − E t −1 [ y t −1 ]
= = 1.
wt

Thus,

(1.126) ψ j = 1 , j = 0, …, ∞ .

Substituting (1.126) into (1.115) yields

⎛ γ ⎞ r ∞ ⎛ γ ⎞ r ∞
∑ψ t (1 + r ) = ⎜⎜1 − ∑ (1 + r ) ,
−t −t
(1.127) ψ * = ⎜⎜1 − ⎟⎟ ⎟⎟
⎝ (1 + r ) ⎠ (1 + r ) t =0 ⎝ (1 + r ) ⎠ (1 + r ) t =0

⎛ γ ⎞ r ⎛1+ r ⎞ ⎛ γ ⎞
(1.128) ψ * = ⎜⎜1 − ⎟⎟ ⎜ ⎟ = ⎜⎜1 − ⎟,
⎝ (1 + r ) ⎠ (1 + r ) ⎝ r ⎠ ⎝ (1 + r ) ⎟⎠
∞ ⎛1+ r ⎞
since ∑ (1 + r )−t = ⎜ ⎟.
t =0 ⎝ r ⎠

Substituting (1.128) into (1.116) yields

⎛ ⎛ γ ⎞ ⎞
⎟⎟(et + αet −1 )⎟⎟ .
1
(1.129) Γ j −1 = ln E j −1 exp⎜⎜ − θ ⎜⎜1 −
θ ⎝ ⎝ (1 + r ) ⎠ ⎠

b 2σ 2
The lognormal distribution (see Exercise 1.10) implies E e ( )= e
bx
b μ+
2 [ ]
, if x ~ N μ , σ 2 . Therefore,
if we assume that et ~ N [
0, σ e2 ] and i.i.d., then we obtain
⎛ ⎛ ⎞ ⎞⎟
γ ⎞ ⎛⎜ σ e2 α 2σ e2
( )
2 2
⎟ = θ ⎛⎜1 − γ ⎞⎟ 1 + α 2 σ e .
2
(1.130) Γ j −1
1 ⎜
= ln exp θ ⎜⎜1 −
2
⎟ +
θ ⎜ ⎝ (1 + r ) ⎟⎠ ⎜ 2 2 ⎟⎟ ⎜ (1 + r ) ⎟
⎝ ⎠ 2
⎝ ⎝ ⎠⎠

(1.122) and (1.123) imply that Et yτ = yt . Thus equation (1.118) becomes

r ⎛ ∞
t −τ ⎞
(1.131) Y pt = ⎜ (1 + r )At −1 + y t ∑ (1 + r ) ⎟ ,
(1 + r ) ⎝ τ =t ⎠

r ⎛ ⎛1 + r ⎞⎞
(1.132) Y pt = ⎜⎜ (1 + r )At −1 + y t ⎜ ⎟ ⎟⎟ = rAt −1 + y t .
(1 + r ) ⎝ ⎝ r ⎠⎠

Substituting (1.130) and (1.132) back into (1.115) gives

22
 σ 2 ∞ (τ − t )
( )
2
γ ⎛ γ ⎞ r ⎛ γ ⎞
(1.133) ct = ct −1 + ⎜⎜1 − ⎟⎟(rAt −1 + y t ) − θ ⎜⎜1 − ⎟⎟ 1 + α 2 e ∑ .
(1 + r ) ⎝ (1 + r ) ⎠ (1 + r ) ⎝ (1 + r ) ⎠ 2 τ =t +1 (1 + r )τ −t

The term of P = ∑
∞ (τ − t ) can be solved by the following procedure:
τ =t +1 (1 + r )
τ −t

1 2 3
(1.134) P= + + + ... ,
1 + r (1 + r ) 2
(1 + r )3

(1.135) (1 + r )P = 1 + 2
+
3
+ ....
(1 + r ) (1 + r )2
(1.134) – (1.135) yields

1 1 1+ r
(1.136) rP = 1 + + + .... = .
(1 + r ) (1 + r )2 r

Thus,

(1.137) P= ∑
∞ (τ − t ) =
1+ r
.
τ =t +1 (1 + r )
τ −t
r2

Substituting (1.137) into (1.133) finally gives

( )
2
γ ⎛ γ ⎞ 1 ⎛ γ ⎞ σ2
(1.138) ct = ct −1 + ⎜⎜1 − ⎟⎟(rAt −1 + y t ) − θ ⎜⎜1 − ⎟⎟ 1 + α 2 e .
(1 + r ) ⎝ (1 + r ) ⎠ r ⎝ (1 + r ) ⎠ 2

Matlab simulations:

There are two Matlab files for equation (1.138):

• HabitConsumption_a.m: Only one shock at period t for et ;

• HabitConsumption_b.m: Continuous shocks for et ;

The results of HabitConsumption_a.m:

Students need to change/modify the following input data

% beginning of input data

Ctm1 = 1.0; % C(t-1)
Ytm1 = 1.05; % Y(t-1)
Atm1 = 0.6; % A(t-1)
r = 0.1; % real interest rate, annual rate
gamma = 0.6;
theta = 0.5;
sigma = 0.08; % annual % s.d. of Y(t)
alpha = 0.4;
period = 50; % The periods of C(t) to be computed
shock = -2.0; % The ONLY shock in terms of Normal distribution: ~N(0,1)
% end of input data

23
For period t, the Matlab file computes the consumption at t via (1.138), based on the input data.
Equation (1.114) implies that

(1.139) At = (1 + r )At −1 + y t − ct ,

At the end of period t, the new wealth is computed based on equation (1.139). For period t+1, and
periods after, E [ y t +1 ] = y t due to the expected values of y process.

Input data:
r = 0.1000, sigma = 0.080, gamma = 0.600
theta = 0.500, T = 50.000 e(t) = -0.168
C(t-1) = 1.000, A(t-1) = 0.600, Y(t) = 1.0500

Results:
t C(t) A(t) Y(t)
t- 1 1.00000 0.60000 1.05000
t+ 0 0.97003 0.57197 0.88200
t+ 1 0.92950 0.53126 0.83160
t+ 2 0.90555 0.51044 0.83160
t+ 3 0.89153 0.50155 0.83160
t+ 4 0.88349 0.49982 0.83160
t+ 5 0.87902 0.50238 0.83160
t+ 6 0.87670 0.50753 0.83160
t+ 7 0.87566 0.51421 0.83160
t+ 8 0.87540 0.52183 0.83160
t+ 9 0.87561 0.53000 0.83160
t+10 0.87609 0.53851 0.83160
t+11 0.87674 0.54722 0.83160
t+12 0.87749 0.55605 0.83160
t+13 0.87830 0.56495 0.83160
t+14 0.87915 0.57389 0.83160
t+15 0.88002 0.58286 0.83160
t+16 0.88090 0.59184 0.83160
t+17 0.88179 0.60084 0.83160
t+18 0.88268 0.60984 0.83160
t+19 0.88358 0.61884 0.83160
t+20 0.88448 0.62785 0.83160
t+21 0.88538 0.63685 0.83160
t+22 0.88628 0.64586 0.83160
t+23 0.88718 0.65487 0.83160
t+24 0.88808 0.66388 0.83160
t+25 0.88898 0.67288 0.83160
t+26 0.88988 0.68189 0.83160
t+27 0.89078 0.69090 0.83160
t+28 0.89168 0.69991 0.83160
t+29 0.89258 0.70892 0.83160
t+30 0.89348 0.71793 0.83160
t+31 0.89438 0.72693 0.83160
t+32 0.89529 0.73594 0.83160
t+33 0.89619 0.74495 0.83160
t+34 0.89709 0.75396 0.83160
t+35 0.89799 0.76297 0.83160
t+36 0.89889 0.77198 0.83160
t+37 0.89979 0.78098 0.83160
t+38 0.90069 0.78999 0.83160
t+39 0.90159 0.79900 0.83160
t+40 0.90249 0.80801 0.83160
t+41 0.90339 0.81702 0.83160
t+42 0.90429 0.82602 0.83160
t+43 0.90519 0.83503 0.83160
t+44 0.90610 0.84404 0.83160
t+45 0.90700 0.85305 0.83160
t+46 0.90790 0.86206 0.83160
t+47 0.90880 0.87107 0.83160

24
 t+48 0.90970 0.88007 0.83160
t+49 0.91060 0.88908 0.83160
t+50 0.91150 0.89809 0.83160

The Matlab program also automatically provides us with a time-series figure for c(t), y(t), A(t):

Alessie and Lusardi (1997)

C:-, Y: ---, A: -.-
1.1

0.9

0.8
C, Y, A

0.7

0.6

0.5

0.4
0 5 10 15 20 25 30 35 40 45 50
Time

The results of HabitConsumption_b.m:

The input data needed for HabitConsumption_b.m is almost completely the same as
HabitConsumption_a.m, except that the Matlab file simulates the shocks of e(t) by the standard
normal distribution generator randn(1), which is based on N ~ (0,1). After inserting the input data

% beginning of input data

Ctm1 = 1.0; % C(t-1)
Ytm1 = 1.1; % Y(t-1)
Atm1 = 0.6; % A(t-1)
r = 0.1; % real interest rate, annual rate
gamma = 0.6;
theta = 0.5;
sigma = 0.1; % annual % s.d. of Y(t)
alpha = 0.7;
period = 50; % The periods of C(t) to be computed
% end of input data

the Matlab file gives the following results and figure:

Input data:
r = 0.1000, sigma = 0.080, gamma = 0.600
theta = 0.500, T = 50.000
C(t-1) = 1.000, A(t-1) = 0.600, Y(t) = 1.0500

Results:
t C(t) A(t) Y(t)
t- 1 1.00000 0.60000 1.05000
t+ 0 1.09534 0.72285 1.15819
t+ 1 1.17203 0.82331 1.20022
t+ 2 1.16001 0.81733 1.07169
t+ 3 1.10823 0.76362 0.97278

25
t+ 4 1.09559 0.75689 1.01251
t+ 5 1.06551 0.72924 0.96217
t+ 6 1.08241 0.75795 1.03821
t+ 7 1.12141 0.81318 1.10084
t+ 8 1.15365 0.86031 1.11946
t+ 9 1.19094 0.91350 1.15810
t+10 1.21701 0.95322 1.16538
t+11 1.20735 0.95005 1.10887
t+12 1.21105 0.96294 1.12893
t+13 1.27459 1.04761 1.26297
t+14 1.40951 1.21796 1.47509
t+15 1.48318 1.31480 1.45823
t+16 1.55549 1.41001 1.51922
t+17 1.55962 1.42340 1.43202
t+18 1.60719 1.48892 1.53037
t+19 1.66391 1.56542 1.59152
t+20 1.71131 1.63074 1.62009
t+21 1.84613 1.80096 1.85327
t+22 1.82768 1.78725 1.63388
t+23 1.76767 1.72367 1.52536
t+24 1.67446 1.62026 1.39868
t+25 1.51556 1.43801 1.17128
t+26 1.41604 1.32703 1.16126
t+27 1.35049 1.25681 1.14757
t+28 1.28788 1.19011 1.09550
t+29 1.24732 1.14987 1.08806
t+30 1.20747 1.11050 1.05311
t+31 1.15552 1.05658 0.99056
t+32 1.14774 1.05568 1.04118
t+33 1.12526 1.03715 1.00115
t+34 1.09041 1.00377 0.95332
t+35 1.08323 1.00358 0.98266
t+36 1.05746 0.98109 0.93461
t+37 1.03062 0.95733 0.90875
t+38 1.03553 0.97165 0.95412
t+39 1.05251 1.00046 0.98415
t+40 1.15758 1.13499 1.19206
t+41 1.20133 1.19592 1.14876
t+42 1.24807 1.26044 1.19300
t+43 1.32734 1.36400 1.30485
t+44 1.34921 1.39869 1.24750
t+45 1.32028 1.37241 1.15413
t+46 1.29542 1.35101 1.13677
t+47 1.28147 1.34271 1.13808
t+48 1.27010 1.33750 1.13062
t+49 1.19831 1.25979 0.98684
t+50 1.19108 1.25955 1.06487

Alessie and Lusardi (1997)

C:-, Y: ---, A: -.-
2

1.5
C, Y, A

0.5
0 5 10 15 20 25 30 35 40 45 50
Time

26
Please note that the results and figure should be different every time the Matlab file is run due to the
randomness of the shocks.

Further Software Exercises:

Show the impact of changes in the following parameters on the time-series C(t) and A(t) by running the
Matlab file HabitConsumption_a.m, and provide an economic interpretation of the results:

1. The habit persistence parameter γ ,

2. the real interest rate r,
3. the standard deviation σ ,
4. a random walk vs MA(1), and
5. the θ parameter.

Additional Reference:

Alessie, R. and A. Lusardi (1997) “Consumption, Saving and Habit Formation, Economics Letters 55,
103-108.

27
28