Académique Documents
Professionnel Documents
Culture Documents
Volker Hahn
University of Konstanz
Tutorials
matlab (see ``MATLAB for Economics and Econometrics A Beginners
Guide'' by John Frain, available online)
no midterm exam, teams of at most three students have to hand in
problem sets every week, 20\% of the overall grade
Classical Economics
Keynesianism
Monetarism
1 Introduction
2 Solow Model
3 Dynamic Optimization
Cake-eating problem
Stationary dynamic programming
The Contraction Mapping Theorem and its applications
Generalizations of stationary dynamic programming
5 Ricardian Equivalence
7 RBC Models
Idea
Model
The social planner's problem
HP-filter
Competitive equilibrium
Assessment
Log-linearization
Critique of RBC models
8 Money
Money-in-the utility approach
Cash-in-advance models
Search theory and money
New Keynesian Economics
Where has money gone?
Time inconsistency
Real GDP per worker and capital per worker roughly grow at a
constant rate in many developed countries (e.g. the US over the last
200 years)
Both grow at similar rates. As a result, the ratio between capital and
real GDP is approximately constant over time.
The real rate of return on capital and the real interest rate are
approximately constant over time.
Cobb-Douglas function: F (AL, K) = (AL)1 - \alpha K \alpha where 0 \leq \alpha \leq 1
It is straightforward to check that the Cobb-Douglas production
function satisfies our assumptions about F (AL, K).
\partial F (AL,K)
marginal product of effective labor:
\partial AL = (1 - \alpha ) F (AL,K)
AL
\partial F (AL,K)
marginal product of capital:
\partial K = \alpha F (AL,K)
K
f (k) = k \alpha
Nt+1 - Nt = nNt
Nt = N0 (1 + n)t
The efficiency of labor (or level of usable knowledge), At , grows at a
constant rate g
At = A0 (1 + g)t
Recall that for F (AL, K) = (AL)1 - \alpha K \alpha the production function in
intensive form
\alpha
is given by f (k) = k .
This implies
Otherwise the capital stock in intensive form would shrink because ...
...a fraction of the capital stock depreciates (\delta ).
...effective labor grows at rate (1 + n)(1 + g) - 1 = g + n + ng .
\alpha
If actual investment s(kt ) is equal to break-even investment, kt is
constant over time.
Definition
Balanced Growth Path: On the balanced growth path kt is constant.
The BGP (balanced growth path) capital stock is given by the solution
to sk \alpha = [\delta + g + n + ng] k
In the following, we will use X \ast to denote the value of X on the BGP.
\Bigl( \Bigr) 1
1 - \alpha
k \ast = s
n+g+ng+\delta
(kt)
( + n + g + ng)kt
skt
kt
kt+1 kt
kt
a permanent increase in the saving rate s increases Y\widehat t\ast , but not its
growth rate g \Rightarrow level effect
\infty
\sum
V (s0 ) = \mathrm{m}\mathrm{a}\infty \mathrm{x} \beta t u(st - st+1 )
\{ st \} t=1
t=0
subject to
st+1 \in \Gamma (st ), \forall t \geq 0
\Gamma (st ) = [0, st ]
Suppose you had already solved the problem from period t+1 on \Rightarrow
V (st+1 ).
Then the cake-eating problem in period t is
First-order condition:
Envelope theorem:
V \prime (st ) = u\prime (st - st+1 )
Combining yields the so-called Euler equation:
In consumption levels:
ct = \beta t c0
Let X be the set of possible values that the state x can take.
\bigl\{ \bigr\}
\Phi (xt ) := \{ xs \} \infty \prime
s=t : xs\prime +1 \in \Gamma (xs\prime )\forall s \geq t
\infty
\sum
W (x0 ) = \mathrm{s}\mathrm{u}\mathrm{p} \beta t u(xt , xt+1 )
\bfx \in \Phi (x0 ) t=0
5 Is V differentiable?
Under somewhat more restrictive assumptions (for example \Gamma (x) must
not be an open interval (a, b) and should be bounded), we obtain
uniqueness and existence of solution V to FE and the existence of
policy functions that attain V.
Proof: By the contraction mapping theorem, yet to be introduced.
Operator T
(T V )(x) := \mathrm{s}\mathrm{u}\mathrm{p} \{ u(x, y) + \beta V (y)\}
y\in \Gamma (x)
Operator T takes the function V as input and spits out a new function
TV .
A solution V \ast to the functional equation is a fixed point of the
operator T, i.e. satisfies
V \ast = T V \ast
1 the operator T has exactly one fixed point v \ast \in S and
1 Monotonicity: If f, g \in B(X) such that f (x) \leq g(x) for all x \in X ,
then (T f )(x) \leq (T g)(x) for all x \in X .
2 Discounting: Let the function f + a, for f \in B(X) and a \in \BbbR + , be
defined as (f + a)(x) := f (x) + a. There exists \beta \in (0, 1) such that
for all f \in B(X), a \geq 0 and all x \in X
Note that an upper boundary for the distance from the fixed point can
be calculated in the following way:
| | V \ast - T n V | | \leq 1
1 - \beta | | T
n+1 V - T n V | | < \varepsilon
1 - \beta .
Hahn (U. Konstanz) Advanced Macroeconomics I October 23, 2016 52
3.4 Dynamic Optimization: Generalizations of stationary
dynamic programming
Again, similar theorems exist that allow, in particular, the use of Euler
equation and transversality condition to characterize solutions.
\int \infty
e - \rho t u(c(t)) dt
t=0
subject to
\. = - c(t), s(t) \geq 0
s(t) \forall t \geq 0.
by choosing a consumption path c(t) for an initial cake size s(0) = 1 .
In the Solow model, the firm simply invests all available inputs (labor
and capital). In the RCK-Model, the firm chooses its inputs to
maximize profits.
In the RCK-model, prices adjust such that the demand and supply of
labor and capital respectively are equal. In the Solow model, all
resources are used by assumption.
Social Planner
A social planner is a hypothetical agent who can choose an arbitrary
feasible allocation (i.e. sequences of consumption and capital) to maximize
welfare (where welfare is just the utility of the representative household in
our case).
\Bigl( relative
the coefficient of \Bigr) \Bigl(
risk \Bigr)
aversion is defined as
\widehat t u\prime \prime C
RRA = - C \widehat t /u\prime C
\widehat t
for a CRRA-utility function this coefficient is \sigma
The parameter \sigma measures the concavity of the CRRA-utility function.
For very large values, the function becomes ``more concave''. The
household tends to smooth consumption over time.
\Bigl\{ \Bigr\}
V\~t (Kt ) = \mathrm{m}\mathrm{a}\mathrm{x} \widehat t ) + \beta V\~t+1 (Kt+1 )
u(C
\~ t (Kt )
Kt+1 \in \Gamma
s.t. \widehat t
Kt+1 = Kt (1 - \delta ) + F (Kt , At Nt ) - Ct , Ct /Nt = C
where
\~ t (Kt ) := [0, Kt (1 - \delta ) + F (Kt , At Nt )]
\Gamma
Note that Kt is not guaranteed to remain in a compact set over time.
Note that \~ t (Kt ) also depends on time because At depends on time.
\Gamma
1
u\prime (C\^t\ast ) = \beta u\prime (C\^t+1
\ast
)(1 - \delta + f \prime (kt+1
\ast
)) \forall t \geq 0
(1 + n)
2 Transversality condition:
1 1
= \beta \ast \forall t \geq 0
\^
Ct \ast \^
Ct+1
Hence per-capita consumption must be of the form
C\^0\ast = (1 - \beta )K
\^ 0 .
\infty
\sum
\beta t u(Ct )
t=0
subject to
Lagrangean:
\infty
\sum \Bigl( \Bigl[ \Bigr] \Bigr)
\scrL = \beta t u(Ct ) - \lambda t Kt+1 - Kt (1 - \delta ) - F (Kt ) + Ct
t=0
First-order conditions:
\Bigl( \Bigr)
\beta u\prime (Ct+1 ) (1 - \delta ) + F \prime (Kt+1 ) = u\prime (Ct )
\bigl\{ \bigr\}
V (k) = \mathrm{m}\mathrm{a}\mathrm{x} \mathrm{l}\mathrm{n}(f (k) - k + ) + \beta V (k + )
k+ \in [0,k\alpha ]
4 Guess a functional form for the value function and apply the method
of undetermined coefficients.
....
...
Repeat until the difference between two vectors Vs and V s+1 is lower
than some given \varepsilon (to compute the difference between the vectors, we
can use the maximum absolute difference for all pairs of corresponding
entries).
Note: The policy function in a particular step is given by the values
of j that yield the maximum value for different values of i.
Hahn (U. Konstanz) Advanced Macroeconomics I October 23, 2016 76
4.3 Neoclassical Growth Theory: Competitive equilibrium
All contracts can be enforced (for example, firms always pay back their
loans).
No information asymmetries.
The household owns the firm (it receives the firm's profits).
The household can rent out one unit of the good (capital) from
period t to t + 1 to the firm and receives a compensation of rt+1 in
period t + 1.
rt : real interest rate
\infty
\sum
\widehat 0 , C
U (C \widehat 1 , ...) = \widehat t )
\beta t u(C
t=0
\Biggl( \tau
\Biggr)
\prod 1
\mathrm{l}\mathrm{i}\mathrm{m} \Omega \tau \geq 0
\tau \rightarrow \infty 1 + r\tau \prime
\tau \prime =1
\infty
\Biggl( \tau
\Biggr)
\sum \prod 1
C\tau
1 + r\tau \prime
\tau =0 \tau \prime =1
\infty
\Biggl( \tau
\Biggr)
\sum \prod 1
\leq \Omega 0 (1 + r0 ) + [L\tau w\tau + profits of firm in \tau ]
1 + r\tau \prime
\tau =0 \tau \prime =1
3 The firm pays rt \Omega t (in addition to the principal \Omega t ) to the household,
which is the owner of capital.
4 The firm pays wt Lt to the household, which has supplied its labor.
Definition
For given K0 = \Omega 0 and r0 , a competitive equilibrium is a path of
consumption, assets, capital, labor, real wages, and real interest rates for
all periods t = 0, 1, 2, ... such that
All markets clear for all periods (labor markets, goods markets, capital
markets).
First-order conditions:
\partial F (Kt , At Lt )
= rt + \delta , \forall t \geq 1,
\partial Kt
\partial F (Kt , At Lt )
= wt , \forall t \geq 0
\partial Lt
\partial F (Kt , At Lt )
= rt + \delta
\partial Kt
or, equivalently
\partial f (kt )
= rt + \delta
\partial kt
Combining with the household's Euler equation yields:
\prime
\widehat t ) = 1 + f (kt+1 ) - \delta \cdot \beta u \widehat (C
uC\widehat (C \widehat t+1 )
1+n C
This is the same condition that we obtained for the social planner's
problem!
\Biggl( \tau
\Biggr)
\prod 1
\mathrm{l}\mathrm{i}\mathrm{m} \Omega \tau = 0
\tau \rightarrow \infty 1 + r\tau \prime
\tau \prime =1
\bigl[ \bigr]
\mathrm{l}\mathrm{i}\mathrm{m} \beta t u\prime (C\^t )At 1 - \delta + f \prime (kt ) kt = 0.
t\rightarrow \infty
Definition
An allocation is Pareto-efficient if it is not possible to raise the utility of
one agent without lowering the utility of at least one other agent.
\partial Yt
recall that wt = \partial Lt = (1 - \alpha ) LYtt and rt = \partial Yt
\partial Kt
Yt
- \delta = \alpha K t
- \delta
Thus 1 - \alpha corresponds to the fraction of total income Yt that accrues
to labor
Our findings are in line with Kaldor's growth facts about income
shares.
Definition
On the BGP Yt , Ct and Kt grow at constant rates.
Note that such a BGP, if it exists, will be in line with Kaldor's growth
facts.
Note that such a BGP, if it exists, may be only one possible solution.
\biggl( \biggr)
ct+1 1 1 + f \prime (kt+1 ) - \delta 1/\sigma
= \beta
ct 1+g 1+n
If (1 + g)\sigma (1 + n) > \beta (1 - \delta ), (1 + g)\sigma (1 + n) = \beta (1 + f \prime (k) - \delta ) has
a unique solution for k (due to the Inada conditions and because f (k)
\prime
Consumption is given by
\bigl[ \bigr]
\widehat \ast = At c\ast = At k \ast (1 - \delta ) + f (k \ast ) - k \ast (1 + n)(1 + g)
C t t t t t+1
\bigl[ \bigr]
\mathrm{l}\mathrm{i}\mathrm{m} \beta t u\prime (C\^t\ast )At 1 - \delta + f \prime (kt\ast ) kt\ast = 0.
t\rightarrow \infty
\Bigl( \Bigr) - \sigma
Due to u\prime (C\^t ) = C\^t and C\^t\ast = At c\ast = (1 + g)t A0 c\ast , the
Now we can determine the saving rate on the BGP for y = k \alpha
c = (1 - s)y , y = k \alpha , c = ((1 - \delta ) - (1 + n)(1 + g)) k + f (k), and
(1 + g)\sigma (1 + n) = \beta (1 + f \prime (k) - \delta ) yield
(1 + n)(1 + g) - 1 + \delta
s = \alpha 1 \sigma
\beta (1 + n)(1 + g) - 1 + \delta
Due to \beta (1 + g)1 - \sigma < 1 and (1 + n)(1 + g) - 1 + \delta > 0, s \in (0, 1).
difference between Ramsey-Cass-Koopmans model and Solow model
1 Solow model
saving rate is assumed to be constant
2 Ramsey-Cass-Koopmans model
saving rate is derived from household's optimization and a function of
the model's parameters
the saving rate is constant (on the BGP!)
Hahn (U. Konstanz) Advanced Macroeconomics I October 23, 2016 97
4.5 Neoclassical Growth Theory: Transition dynamics
\biggl( \biggr)
ct+1 1 1 + f \prime (kt+1 ) - \delta 1/\sigma
= \beta
ct 1+g 1+n
For given (ct , kt ), these equations specify values for (ct+1 , kt+1 ).
In the long run, the system converges along the saddle path to the
BGP, irrespective of the initial value of k.
The initial value of
the state variable k (k0 ) is exogenously given.
the jump variable c (c0 ) is determined by the only value of c0 that
allows for convergence to the BGP along the saddle path.
Log-linear approximation:
\biggl( \biggr)
xt - x\ast xt - x\ast
\~t = \mathrm{l}\mathrm{n}(xt ) - \mathrm{l}\mathrm{n}(x\ast ) = \mathrm{l}\mathrm{n}(xt /x\ast ) = \mathrm{l}\mathrm{n} 1 +
x \approx
x\ast x\ast
1 \~ c\ast
k\~t+1 = kt - \ast c\~t
\beta k
c\~t+1 - c\~t = - \beta (1 - \alpha )\alpha (k \ast )\alpha - 1 k\~t+1
k\~t+1 = ak k\~t
c\~t = ac k\~t
k\~t+1 and c\~t are expressed as functions of the only state variable k\~t
Hahn (U. Konstanz) Advanced Macroeconomics I October 23, 2016 111
ln(ct+1 )ln(ct )=0
ln(c)
A
ln(k ) ln(k)
1 \~ c\ast
k\~t+1 = kt - \ast c\~t
\beta k
c\~t+1 - c\~t = - \beta (1 - \alpha )\alpha (k \ast )\alpha - 1 k\~t+1
a = - (\lambda 1 + \lambda 2 )
b = \lambda 1 \lambda 2
Hahn (U. Konstanz) Advanced Macroeconomics I October 23, 2016 115
The characteristic equation
Re-arranging yields
\biggl( \biggr)
b
a = - \lambda 1 +
\lambda 1
a\lambda 1 = - \lambda 21 - b
and thus
\lambda 21 + a\lambda 1 + b = 0
Similarly, \lambda 2 also fulfills
\lambda 22 + a\lambda 2 + b = 0
The equation
\lambda 2 + a\lambda + b = 0
is called the characteristic equation of (1 + a\scrL + b\scrL 2 )k\~ = 0.
is
k\~t = \alpha 1 \lambda t1 + \alpha 2 \lambda t2 ,
where \alpha 1 and \alpha 2 are coefficients that are yet to be determined
We can verify that k\~t = \alpha 1 \lambda t1 + \alpha 2 \lambda t2 is a solution by observing
\mathrm{l}\mathrm{i}\mathrm{m} k\~t = 0
t\rightarrow \infty
J = T - 1 AT
- 1
A = T
\biggl( JT \biggr)
\lambda 1 \cdot
J =
0 \lambda 2
vt+1 = Jvt , vt = J t v0
k\~t and \~
lt evolve according to
\biggl( \biggr) \biggl( \biggr) \biggl( t \biggr)
k\~t v1,t t \lambda 1 v1,0
\~lt = T v2,t = T J v0 = T \lambda t2 v2,0
Growth rates are also very different across countries (even over a long
period of time)
For \alpha \approx 1/3, per-capita income differences of a factor ten can be
explained by capital per capita differences of 103 = 1000 \Rightarrow
implausible
1 - \alpha
1 1 A \alpha
r = \alpha - \delta = \alpha 1 - \alpha - \delta = \alpha 1 - \alpha - \delta
k 1 - \alpha y \alpha Y\^ \alpha
1 - \alpha
As
\alpha =2 for \alpha = 1/3, the rates of return to capital r differ by a
factor of more
2
than 10 = 100 \Rightarrow implausible
Possible conclusions
1 A is different across countries
2 more fundamental change to the model required (like different
production function)
Two countries with the same values for s, n, \delta , \alpha and the same
technology level At should eventually converge to the same value of
per capita income
Absolute convergence
All countries converge to the same BGP. A country with lower output
per-capita thus has a higher growth rate of per-capita GDP (for some
time).
Conditional convergence
All countries converge to different BGPs. The BGP values depend on (are
conditional on) the country's values for n, s and the technology level.
\alpha \alpha
\mathrm{l}\mathrm{n}(Y /L)it = \mathrm{l}\mathrm{n}(Ai ) + gt + \mathrm{l}\mathrm{n}(si ) - \mathrm{l}\mathrm{n}(ni + \delta + g)
1 - \alpha 1 - \alpha
countries are indexed by i.
Assumptions on technology:
The level of technology may differ across countries.
The growth rate of technology is the same.
Kt+1 = Kt + sK Yt - \delta K Kt
Ht+1 = Ht + sH Yt - \delta H Ht
\delta H and \delta K are the respective depreciation rates.
k
h_{t+1}h_t=0
k_{t+1}k_t=0
\Biggl( \Biggr) 1
s\beta H s1 - \beta 1 - \alpha - \beta
k \ast = K
\delta + n + g + ng
\Biggl( \Biggr) 1
s\alpha K s1 - \alpha 1 - \alpha - \beta
h\ast = H
\delta + n + g + ng
\beta can be estimated by comparing the wages for unskilled workers and
the average wage.
1
typical values are \alpha \approx \beta \approx 3.
\alpha
\mathrm{l}\mathrm{n}(Y /L)it = \mathrm{l}\mathrm{n}(Ai ) + gt + \mathrm{l}\mathrm{n}(sK,i )
1 - \alpha - \beta
\beta \alpha + \beta
+ \mathrm{l}\mathrm{n}(sH,i ) - \mathrm{l}\mathrm{n}(ni + \delta + g)
1 - \alpha - \beta 1 - \alpha - \beta
Y = AN 1 - \alpha K \alpha where A = Total factor productivity (TFP); \alpha \approx 1/3,
corresponds to the share of income that accrues to capital owners
dY = (dA)N 1 - \alpha K \alpha + A(1 - \alpha )(dN )N - \alpha K \alpha + AN 1 - \alpha \alpha (dK)K \alpha - 1
Y
Divide by and introduce growth rates dY /Y = gY , dN/N = gN ,
dA/A = gA and dK/K = gK
We obtain gY = gA + (1 - \alpha )gN + \alpha gK
By computing this equation for a country, we can decompose a
country's output growth into growth induced by technological change
and growth that is a consequence of capital accumulation and
population growth
Bt+1 + Tt = (1 + rt )Bt + Gt
\Biggl( \tau
\Biggr)
\prod 1
\mathrm{l}\mathrm{i}\mathrm{m} B\tau \leq 0
\tau \rightarrow \infty 1 + r\tau \prime
\tau \prime =1
\infty
\Biggl( \tau
\Biggr)
\sum \prod 1
B0 (1 + r0 ) + G\tau
1 + r\tau \prime
\tau =0 \tau \prime =1
\infty
\Biggl( \tau
\Biggr)
\sum \prod 1
= T\tau
1 + r\tau \prime
\tau =0 \tau \prime =1
\infty
\Biggl( \tau
\Biggr)
\sum \prod 1
[C\tau + T\tau ]
1 + r\tau \prime
\tau =0 \tau \prime =1
\infty
\Biggl( \tau
\Biggr)
\sum \prod 1
\leq \Omega 0 (1 + r0 ) + [L\tau w\tau + profits of firm in \tau ]
1 + r\tau \prime
\tau =0 \tau \prime =1
All markets clear (labor markets, goods markets, capital markets) for
all periods.
\infty
\Biggl( \tau
\Biggr)
\sum \prod 1
T\tau
1 + r\tau \prime
\tau =0 \tau \prime =1
distortionary taxes
finite lifetime
credit constraints
Like in the Solow model, suppose the saving rate s is exogenous and
constant
A is constant
endogenous growth
\widehat t+1 - K
(K \widehat t )(1 + n) = sAK
\widehat t - (\delta + n)K
\widehat t
(kt)
( + n + g + ng)kt
skt
kt
d
AKt
d
sAKt
d
( + n)Kt
d
Kt
Hahn (U. Konstanz) Advanced Macroeconomics I October 23, 2016 174
6.2 AK model: Socially optimal solution
For a given initial value K0 , the social planner faces the problem:
\mathrm{m}\mathrm{a}\mathrm{x} \widehat t , C
U (C \widehat t+1 , ...)
C
\widehat 0 ,C
\widehat 1 ,...,K1 ,K2 ,...
where \widehat t
C is given by \widehat t
Kt+1 = (1 - \delta )Kt + AKt - Nt C
The per-period \widehat
utility function u(Ct ) is assumed to be of the CRRA
type.
Notes:
Kt is guaranteed to remain in a compact set only if 1 - \delta + A \leq 1 (this
parameter region is not very relevant).
\~ t :=
K1
By using
(1 - \delta +A)t Kt , the maximization problem can be
transformed into a maximization problem where ``the
functional-equation operator'' is a contraction mapping for
\Bigl( \Bigr) 1 - \sigma
1+A - \delta
\beta
1+n <1 and the state variable can be restricted to a
compact set.
Thus we obtain
\widehat t+1 \biggl( \beta (1 - \delta + A) \biggr) \sigma
1
C
=
C \widehat t (1 + n)
\Bigl( \Bigr) 1
\beta (1+A - \delta ) \sigma
With \chi := (1+n) , we obtain
\widehat t = \chi t C
C \widehat 0
t - 1
\sum
\widehat t = \kappa t K
\widehat 0 - 1 \prime
\widehat t\prime
K \kappa t - t C
1 + A - \delta \prime
t =0
\bigl( \bigr) t
t \widehat \kappa t 1 - \chi \kappa \widehat
= \kappa K0 - \cdot C0
1 + A - \delta 1 - \chi \kappa
\^ t = \chi t K
K \^ 0
It is now immediate to compute the saving rate
st = (AKt - Ct )/(AKt ).
As Kt and Ct grow at the same rate, the saving rate is constant (it
will be computed later).
In equilibrium
a representative firm maximizes its profits (subject to technological
feasibility) by choosing the amount of capital used in production in all
periods t \geq 1.
a representative household maximizes its utility (subject to its flow
budget constraints for all periods t \geq 0 and the no-Ponzi condition).
the goods and capital markets clear in all periods t \geq 0.
\biggl( \biggr) 1
\widehat t+1 = \beta (1 + r) \sigma
\widehat t
C C
1+n
\biggl( \biggr) 1
\widehat t+1 = \beta (1 + A - \delta ) \sigma
\widehat t
C C
(1 + n)
\Bigl( \Bigr) 1
\beta (1+A - \delta ) \sigma 1+A - \delta
Let us introduce \chi := (1+n) and \kappa := 1+n
C \widehat t = \chi t C
\widehat 0
\Bigl( \Bigr)
\widehat t+1 = \kappa \Omega
\Omega \widehat t - 1 \widehat
1+A - \delta Ct
\widehat t
\Omega can be written as
t - 1
\sum
\widehat t = \kappa t \Omega
\widehat 0 - 1 \prime
\widehat t\prime
\Omega \kappa t - t C
1 + A - \delta \prime
t =0
\bigl( \bigr) t
t \widehat \kappa t 1 - \chi \kappa \widehat
= \kappa \Omega 0 - \cdot C0
1 + A - \delta 1 - \chi \kappa
Now we can compute the value of the capital stock over time
Because the capital market must clear at each point in time \widehat t = \Omega
K \widehat t
\forall t \geq 0.
K \widehat t = \chi t \Omega
\widehat 0
Because Y\widehat t = AK
\widehat t = \chi t A\Omega
\widehat 0
Thus the growth rate of per-capita output is
\Bigl( \Bigr) 1
\beta (1+A - \delta ) \sigma
gY\widehat = \chi - 1 = 1+n - 1
The growth rate is endogenously determined
Y\widehat t - C \widehat t
st :=
Y\widehat t
1 \Bigl[ \chi \Bigr]
= 1 - 1 - (1 + A - \delta )
A \Biggl[ \kappa \Biggr]
\biggl( \biggr) 1 - \sigma
1 1 + A - \delta \sigma 1
= 1 - 1 - \beta \sigma (1 + A - \delta )
A 1+n
The household obtains a net income of r(1 - \tau ) per unit of savings.
Y\^t - C\^t - G
\^ t
st =
Y\^t
1
1 - \delta [\beta (1 + (A - \delta )(1 - \tau ))] \sigma
= - + 1 - \sigma
A A(1 + n) \sigma
An increase in the tax rate lowers the growth rate and the saving rate.
An increase in the saving rate has a long-run impact on the level and
growth rate of per-capita output
They are ``real'' in the sense that nominal rigidities do not play a role.
For simplicity, we set the depreciation rate \delta to one (together with the
particular form of the utility function we employ, this will allow for an
analytical solution).
Kt+1 = Kt (1 - \delta ) + Yt - Ct = Yt - Ct
\mathrm{l}\mathrm{n} At = A\~t ,
where
A\~t = \rho A\~t - 1 + \varepsilon t .
Note that A\~t follows a Markov process, which allows for a recursive
representation of the social planner's problem and for a recursive
competitive equilibrium.
We assume
- 1 < \rho < 1
\varepsilon t are i.i.d and N (0, \sigma \varepsilon 2 ), where \sigma \varepsilon > 0
A\~0 is exogenously given (like K0 )
Hahn (U. Konstanz) Advanced Macroeconomics I October 23, 2016 212
7.3 RBC Models: The social planner's problem
subject to
subject to
\biggl[ \biggr]
\partial u (Kt , Kt+1 , Lt , At ) \partial u (Kt+1 , Kt+2 , Lt+1 , At+1 )
+ \beta \BbbE t = 0,
\partial Kt+1 \partial Kt+1
\biggl[ \biggr]
1 1 \alpha - 1
= \beta \BbbE t \cdot (At+1 Lt+1 )1 - \alpha \alpha Kt+1
Ct Ct+1
2 The first-order condition (``Euler equation'') with respect to labor yields
1 b
(1 - \alpha )L - \alpha
t At
1 - \alpha \alpha
Kt \cdot =
Ct 1 - Lt
3 We have already specified the economy's resource constraint
\biggl[ \biggr]
1 1 Yt+1
= \alpha \beta \BbbE t \cdot
Ct Ct+1 Kt+1
Yt 1 b
(1 - \alpha ) =
Ct Lt 1 - Lt
Kt+1 = Yt - Ct
Using the third equation to replace Kt+1 in the first equation yields
the following system of difference equations:
\biggl[ \biggr]
Yt - Ct Yt+1
= \alpha \beta \BbbE t
Ct Ct+1
Yt 1 b
(1 - \alpha ) =
Ct Lt 1 - Lt
\biggl[ \biggr]
Yt - Ct Yt+1
= \alpha \beta \BbbE t
Ct Ct+1
Yt 1 b
(1 - \alpha ) =
Ct Lt 1 - Lt
fortunately has a rather simple solution.
The solution involves a constant saving rate and constant labor (not
only in the steady state!)
Yt - Ct
st = = \alpha \beta
Yt
1 - \alpha
Lt =
1 - \alpha + b(1 - \alpha \beta )
1 Why is the saving rate constant (in the social planner's solution)?
We define
the steady-state level of capital Kt : K
the log deviation of capital from its steady-state level
\~ t := \mathrm{l}\mathrm{n} Kt - \mathrm{l}\mathrm{n} K
K
1
K = s 1 - \alpha L, (s and L have been computed already)
\~ t = 1
K (1 - \alpha )\scrL \varepsilon t
(1 - \alpha \scrL )(1 - \rho \scrL )
1 \alpha - \alpha \rho \scrL + \alpha \rho \scrL - \rho
= (1 - \alpha )\scrL \varepsilon t
\alpha - \rho (1 - \alpha \scrL )(1 - \rho \scrL )
\biggl( \biggr)
1 \alpha \rho
= - (1 - \alpha )\scrL \varepsilon t
\alpha - \rho 1 - \alpha \scrL 1 - \rho \scrL
The (a?) solution is
\infty
\sum
\~ t = (1 - \alpha ) \alpha i+1 - \rho i+1
K \varepsilon t - 1 - i
\alpha - \rho
i=0
\infty
\sum \alpha i+1 - \rho i+1
Y\~t = (1 - \alpha ) \varepsilon t - i
\alpha - \rho
i=0
\Biggl\{
0 for t<0
Y\~t = t+1 t+1
(1 - \alpha ) \alpha \alpha - \rho
- \rho for t \geq 0
Note: We will see that the first theorem of welfare holds. Hence the
dynamic response of the economy in the competitive equilibrium will
be the same.
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 1 2 3 4 5 6 7 8 9
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 1 2 3 4 5 6 7 8 9
Suppose we had data on log GDP for t = 1, ...T , i.e. time series data
Y1+ , Y2+ , Y3+ , ... YT+ .
How can we decompose this time series into
1 a trend component (equivalent to steady-state levels in our model) and
2 a cyclical component (deviation from the trend, equivalent to the
variables with `` \sim '')?
...
Hahn (U. Konstanz) Advanced Macroeconomics I October 23, 2016 230
Recursive Competitive Equilibrium (continued)
...
The representative firm chooses Kt+1 and Ldt to maximize its profits,
taking the price functions r(At , \bfK t ), w(At , \bfK t ) as given.
All markets clear, e.g. \bfK t = \bfOmega t .
Note: Bold variables are aggregates:
Because individual agents are small, they cannot affect bold variables.
\partial F (Kt , At Lt )
w(At , \bfK t ) =
\partial Lt
\partial F (Kt , At Lt )
r(At , \bfK t ) + \delta = , where \delta = 1.
\partial Kt
\biggl( \biggr)
\partial ut \bigl( \bigr) \partial ut+1
0= - \Delta C + \beta \BbbE t (1 + r(At+1 , \bfK t+1 )) \Delta C
\partial Ct \partial Ct+1
\biggl( \biggr)
1 1
= \beta \BbbE t (1 + r(At+1 , \bfK t+1 ))
Ct Ct+1
\partial ut \partial ut
\Delta L + \Delta Lw(At , \bfK t ) = 0
\partial Lt \partial Ct
b 1
= \cdot w(At , \bfK t )
1 - Lt Ct
Thus we conclude
Proposition
The first theorem of welfare holds for the RBC model.
As a next step, all equations describing the dynamics of the system are
log-linearized.
1 \~ t
\approx 1 - X
1+X\~ t
\Bigl( \Bigr)
f (Xt , Yt , Zt ) = f X(1 + X \~ t ), Y (1 + Y\~t ), Z(1 + Z\~t )
\bigl( \bigr) \bigl( \bigr)
\approx f X, Y , Z + fX X, Y , Z X X \~ t
\bigl( \bigr) \bigl( \bigr)
+fY X, Y , Z Y Y\~t + fZ X, Y , Z Z Z\~t
Heterogeneity is disregarded.
What is the size of the factor (1 + \lambda ) with which the stochastic
consumption stream would have to be multiplied for the agent to be
indifferent between
the perfectly safe consumption stream, engineered by monetary policy
the stochastic consumption stream, multiplied by (1 + \lambda )
\lambda is a measure of the maximum welfare gains that could be achieved
by stabilization policy.
Functions of Money
unit of account
store of value
medium of exchange
This is a shortcut for a more detailed model in which money allows for
economizing on transaction costs.
households, rent out capital to firms, endowed with one unit of labor
per period; labor is supplied inelastically
\infty
\sum
\beta t u(ct , mt ), 0 < \beta < 1
t=0
Pt \omega t = Tt + Wt + Mt - 1 + It - 1 Bt - 1
\geq Pt ct + Mt + Bt
Tt = Mt - Mt - 1
\mu , the gross money growth rate is exogenously given and constant
(\mu \geq 1).
The firms issue bonds Bt that pay a gross nominal rate of return It
(gross real rate: Rt ) to finance capital purchases.
fk (kt , lt ) + 1 - \delta = Rt
and
fl (kt , lt ) = Wt /Pt
mt - 1 + It - 1 bt - 1
\omega t = wt + \tau t + \geq ct + mt + bt
\Pi t
where
m, b, \tau , w: analogous to M , B , T , w, but in real terms
\Pi t : gross rate of inflation, \Pi t := 1 + \pi t = PPt - 1
t
mt + It bt
\omega t+1 = wt+1 + \tau t+1 +
\Pi t+1
mt
\omega t+1 \leq \tau t+1 + wt+1 + + Rt (\omega t - ct - mt )
\Pi t+1
It
where Rt := \Pi t+1 is the gross real rate of return.
\bigl\{ \bigr\}
V (\omega t ) = \mathrm{m}\mathrm{a}\mathrm{x} u(ct , mt ) + \beta V (\omega t+1 )
ct ,mt
mt
s.t. \omega t+1 = \tau t+1 + wt+1 + + Rt (\omega t - ct - mt )
\Pi t+1
With the help of the above expressions for \omega t+1 , the first-order
conditions can be stated as:
1
um (ct , mt ) + \beta uc (ct+1 , mt+1 ) = uc (ct , mt )
\Pi t+1
Using the two equations at the top of the slide, we can also write
um (ct , mt ) It - 1
=
uc (ct , mt ) It
V\omega (\omega ss ) = V\omega (\omega t ) = \beta Rt V\omega (\omega t+1 ) = V\omega (\omega ss ) \Rightarrow \beta Rss = 1.
The steady-state capital stock can be computed with the help of:
1
fk (k ss ) + 1 - \delta = Rss =
\beta
fl (k ss , 1) = wss
\mu
I ss =
\beta
mss + I ss bss
\tau ss + wss + = css + mss + bss
\Pi ss
\Bigl( \Bigr)
With the help of \tau ss = mss 1 - \mu 1 , bss = k ss , \Pi ss = \mu , and
fk (k ss , 1)k ss + fl (k ss , 1)lss = f (k ss , 1)
css = f (k ss ) - \delta k ss
Hahn (U. Konstanz) Advanced Macroeconomics I October 23, 2016 265
Existence and Uniqueness of Steady State
\biggl( \biggr)
\beta
um (css , mss ) = 1 - uc (css , mss )
\mu
\biggl( \biggr)
ss \beta
\phi m (m ) = 1 - vc (css )
\mu
Our assumptions on \phi and v guarantee that this equation has a unique
solution.
1
um (ct , mt ) + \beta uc (ct+1 , mt+1 ) = uc (ct , mt )
\Pi t+1
\beta
vc (css ) = vc (css , mt ) - \phi m (mt )
\Pi t+1
Multiplying through by mt and noting that mt /\Pi t+1 = mt+1 /\mu yields
or, equivalently
\mu
mt+1 = (vc (css ) - \phi m (mt ))mt
\beta vc (css )
Definition (Neutrality)
Money is neutral if the real steady-state levels of consumption, capital,
labor, and output do not depend on the level of money.
This is plausible because only real money holdings enter into the utility
function.
Definition (Superneutrality)
Money is superneutral if the real steady-state levels of consumption, capital,
labor, and output do not depend on the gross growth rate of money \mu .
Superneutrality holds.
Now the optimal money growth rate \mu can be determined as the value
of \mu that maximizes the household's utility in the steady state.
As consumption does not depend on \mu , the optimal value of \mu satisfies:
Friedman Rule
The so-called Friedman rule is optimal, i.e. a constant negative growth rate
of money such that iss = 0.
It implies deflation (1 = Rss \Pi ss = 1/\beta \cdot \Pi ss ) \Rightarrow \Pi ss = \beta < 1.
\mu ss = \Pi ss < 1, i.e. the nominal stock of money is declining.
um (ct , mt ) It - 1
=
uc (ct , mt ) It
Definition (Seigniorage)
The revenues from printing money are seigniorage.
Assume that
log deviations of TFP from its long run level
deviations of the nominal money growth rate from the long-run
money-growth rate.
Pt \omega t = Tt + Mt - 1 + It - 1 Bt - 1
\geq Pt ct + Mt + Bt + Pt kt
Tt \leq Mt - Mt - 1
Cash-In-Advance Constraint
An agent can only acquire goods in a particular period if he has acquired a
sufficient quantity of money in the previous period.
\bigl\{ \bigr\}
V (\omega t , mt - 1 ) = \mathrm{m}\mathrm{a}\mathrm{x} u(ct ) + \beta V (\omega t+1 , mt )
ct ,mt
s.t. the budget constraint and the CIA constraint
Like in the MIU model, the real budget constraint in t+1 can be
written as
mt
\omega t+1 \leq \tau t+1 + + Rt (\omega t - ct - mt )
\Pi t+1
Recall that the CIA constraint is
Mt - 1 mt - 1
ct \leq + \tau t = + \tau t
Pt \Pi t
Using the above expression for \omega t+1 , we can write the first-order
conditions as:
Dividing by Pt yields
Iterating yields
\infty
\sum \mu t+i /Pt+i
1
= \beta i
Pt \lambda t
i=1
yields
\infty \partial V (\omega t+i ,mt+i - 1 )
1 \sum \partial Mt+i - 1
= \beta i
Pt \lambda t
i=1
\mu t+1
it = It - 1 =
\lambda t+1
1
Rss =
\beta
1
iss \approx - 1 + \pi ss
\beta
The steady-state capital stock is implicitly given by
1
fk (k ss ) = R - 1 + \delta = - 1 + \delta
\beta
Money is superneutral.
In addition
\beta Vm i
=
uc 1+i
Inflation equals the growth rate of the nominal money stock
Mt - Mt - 1
\Theta ss := Mt - 1
\pi ss = \Theta ss
Hahn (U. Konstanz) Advanced Macroeconomics I October 23, 2016 284
Extensions
Trading a good is costly, the receiver of a good incurs fixed costs \epsilon
(\epsilon < U ).
In each period t = 0, 1, 2, ... the agents meet pairwise and randomly.
Only agents who meet can trade.
The value function depends on this state variable (V (g) and V (m))
Let \pi be the probability of a particular agent in state g being willing
to trade his good for money.
He takes \Pi , the respective probability for the other agents, as given
Bellman equations:
Intuition: The more likely it is that the other agents will accept my
money in the future (the higher \Pi is), the more attractive is it for me
to accept money in a transaction.
The Phillips curve is derived from the optimal price setting decisions
of monopolistically competitive firms
Different approaches
Taylor-price setting: overlapping prices (Taylor 1979, 1980)
Pre-determined prices
Calvo-price-setting: in each period, a firm may adjust the price of its
output only with probability \lambda (Calvo 1983).
Rotemberg (1982): quadratic costs of price adjustment
p\ast t = pt + \alpha yt
Suppose a firm that may adjust its price xt in the following way
\infty
\sum
xt = (1 - (1 - \lambda )\beta ) (1 - \lambda )j \beta j Et p\ast t+j
j=0
where we have taken into account that the price xt will remain
unchanged until period t+j with probability (1 - \lambda )j .
The factor 1 - \sum
(1 - \lambda )\beta is a normalization,
(1 - (1 - \lambda )\beta ) \infty j j
j=0 (1 - \lambda ) \beta = 1
\infty
\sum
xt = (1 - (1 - \lambda )\beta ) (1 - \lambda )j \beta j Et p\ast t+j
j=0
\infty
\sum
= (1 - (1 - \lambda )\beta )p\ast t + \lambda (1 - \beta ) (1 - \lambda )j \beta j Et p\ast t+j
j=1
\infty
\sum
= (1 - (1 - \lambda )\beta )p\ast t + \lambda (1 - \beta )(1 - \lambda )\beta (1 - \lambda )j \beta j Et p\ast t+j+1
j=0
= (1 - (1 - \lambda )\beta )p\ast t + (1 - \lambda )\beta \BbbE t xt+1
\infty
\sum
pt = \lambda (1 - \lambda )i xt - i
i=0
\infty
\sum
pt = \lambda (1 - \lambda )i xt - i
i=0
\infty
\sum
= \lambda (1 - \lambda )i xt - i + \lambda xt
i=1
\infty
\sum
= \lambda (1 - \lambda ) (1 - \lambda )i xt - i - 1 + \lambda xt
i=0
= (1 - \lambda )pt - 1 + \lambda xt
The current price level equals a weighted average of the previous period's
price level and the prices set by the firms who updated their prices.
Important characteristics:
\infty \biggl(
\sum \biggr) j
\lambda 2
\pi t = \alpha Et [yt+j ]
1 - \lambda
j=0
The New Keynesian model implies that the price level is given by the
previous period's price level and the expected marginal costs in this
and future periods.
IS curve
yt = - \gamma (it - Et [\pi t+1 ]) + Et [yt+1 ]
The IS curve is similar to the one in the IS-LM model in that output is
a monotonically decreasing function of the real interest rate.
\Biggl[ \infty
\Biggr]
\sum
yt = - \gamma Et (it+i - \pi t+i+1 )
i=0
\sigma t =\rho \sigma \sigma t - 1 + \varepsilon t \varepsilon \sim N (0, \sigma \varepsilon 2 )
\xi t =\rho \xi \xi t - 1 + \mu t \mu \sim N (0, \sigma \mu 2 )
Due to the Fisher equation, this calls for a more than one-to-one
increase in nominal interest rates in response to an increase in inflation
expectations.
Some empirical evaluation shows that central banks also seem to put
some weight on their previous interest-rate decisions.
Monetary policy has an impact on output and inflation only via the
interest rate.
That central banks choose interest rates rather than the money supply
is more realistic (exception: quantitative easing?)
Why can central banks choose nominal interest rates?
Hint: They are the monopoly supplier of currency and reserves (base
money).
\infty
\sum \bigl[ \bigr]
Lt = Et (\pi t+j )2 + a(yt+j )2
j=0
Two objectives
1 stabilize inflation around 0
2 stabilize output gap
Discretionary policy: In each period, the central bank chooses \pi t and
yt , subject to the Phillips curve, taking expectations about its own
future policy as given.
\kappa 2
\pi t = - \pi t + Et [\pi t+1 ] + \xi t
a
1
\pi t = 2 [Et [\pi t+1 ] + \xi t ]
1 + \kappa a
\infty
\Biggl( \Biggr) j
1 \sum 1
\pi t = \kappa 2 \kappa 2
Et [\xi t+j ]
1+ a j=0 1+ a
\infty
\Biggl( \Biggr) j
1 \sum 1
= \kappa 2 2 (\rho \xi )j \xi t
1+ a j=0 1 + \kappa a
\Bigl( 2
\Bigr)
1 1 + \kappa a
= \Bigl( \Bigr) \Bigl( \Bigr) \xi t
\kappa 2 \kappa 2
1+ a 1+ a - \rho \xi
a
= \xi t
a(1 - \rho \xi ) + \kappa 2
This implies:
\kappa \kappa
yt = - \pi t = - \xi t
a a(1 - \rho \xi ) + \kappa 2
Here:
two targets: output and inflation
one instrument: interest rate
Intuition:
A supply shock drives inflation up and output down (or vice versa).
Monetary policy can only shift inflation and output into the same
direction (lower interest rates, e.g., raise output and inflation)
Hence the central bank cannot stabilize supply shocks perfectly.
Inserting \pi t \kappa + ayt = 0 into the IS curve, applying the Phillips curve
and re-arranging, yields
with f2 > 0
Optimal policy can be described by a forward-looking Taylor rule.
\biggl[ \biggr]
a j
\mathrm{l}\mathrm{i}\mathrm{m} Et [\pi t+j ] = \mathrm{l}\mathrm{i}\mathrm{m} (\rho \xi ) \xi t = 0
j\rightarrow \infty j\rightarrow \infty a(1 - \rho \xi ) + \kappa 2
Strict inflation targeting, i.e. a policy where Et [\pi t+j ] = 0 \forall j \geq 1 is
not optimal in general because it leads to a high output variance.
Intuition
If inflation expectations are high, this will lead to high inflation today.
The central bank should promise that inflation will be low in the future.
If this announcement is credible, the central bank will be better able to
stabilize current inflation.
In the future, the central bank has an incentive to renege on its
promise.
\infty
\sum \bigl[ \bigr]
Lt = Et (\pi t+j )2 + a(yt+j - \Delta )2
j=0
\kappa 2
\pi t = - \pi t + \kappa \Delta + Et [\pi t+1 ] + \xi t
a
1
\pi t = 2 [\kappa \Delta + Et [\pi t+1 ] + \xi t ]
1 + \kappa a
Iterating recursively
a a
\pi t = 2
\xi t + \Delta
a(1 - \rho \xi ) + \kappa \kappa
\kappa
yt = - \xi t
a(1 - \rho \xi ) + \kappa 2
For \Delta = 0, this solution coincides with the solution previously derived.
a
For positive \Delta , inflation is systematically higher and equal to
\kappa \Delta \Rightarrow
inflation bias
reputation building
sticky wages