The Real Business Cycle Model

Irasema Alonso

February 2025

Irasema Alonso The Real Business Cycle Model February 2025 1 / 48

Real Business Cycle Theory

The RBC theory was introduced by Finn Kydland and Edward

Prescott in the early 80s and it is based on the Solow growth
model.
“Time to Build and Aggregate Fluctuations” (1982) proposed a
theory of business cycle fluctuations far from the Keynesian
tradition.
They showed that technology shocks, i.e., deviations of the level
of technology from its growing trend, could be an important
cause of output fluctuations.
Like in the basic Solow model, there is no technological growth,
but instead the TFP parameter (the A) is random.
This is one way of capturing how shocks hit the economy and
”propagate”, i.e., influence the paths of the macroeconomic
variables when optimizing firms and households respond to
these fluctuations in the level of technology.
Irasema Alonso The Real Business Cycle Model February 2025 2 / 48
Many factors can lead to changes in total factor productivity: any
change implying that an economy can produce more aggregate
output with the same factors of inputs.
Factors that increase TFP include good weather, technological
innovations, the easing of government regulation, decreases in
the relative price of energy.
The RBC model tries to account for the statistical properties one
sees in the data on deviations of important macroeconomic
variables from their trends – the business cycle phenomenon.
Will the model exhibit the features we discussed (e.g.,
consumption and investment and output comove, consumption
fluctuates less than output, etc.)?

Irasema Alonso The Real Business Cycle Model February 2025 3 / 48

Steps

First I explain how the model is set up and works.

Then I show how it is calibrated: how the values of the different
parameters of the model are chosen.
Once this is done, one puts the model to work: one generates
data from it, looks at the statistical properties of the (artificial)
data and confront them to the actual data.

Irasema Alonso The Real Business Cycle Model February 2025 4 / 48

The model – a simple special case of K&P’s 1982
setup
It builds on a stochastic version of the Solow growth model.
The formulation here describes a process where the technology
level and population size do not exhibit growth.
Time is discrete and infinite: 0, 1, 2, · · ·
There is one type of good at each date, yt , which can be used for
consumption or investment.
ct + xt = yt .

The production function exhibits constant returns to scale in k

and l:
yt = F (zt , kt , lt )
where zt is a technology parameter. Key assumption:{zt } is a
stochastic process.
Irasema Alonso The Real Business Cycle Model February 2025 5 / 48
Law of motion for capital:

kt+1 = (1 − δ)kt + xt

The production function is given by:

yt = zt ktα lt1−α

The stochastic technology parameter, log zt , follows an AR(1)

process:
log zt+1 = ρ log zt + t+1
where ρ ∈ (0, 1) indicates positive autocorrelation and t is
identically and independently distributed over time (iid), normal
with zero mean and positive variance σ 2 .

Irasema Alonso The Real Business Cycle Model February 2025 6 / 48

There is a large number of (identical) consumers.
Each consumer is infinitely lived and derives utility from
consumption and leisure.
His/her preferences from the perspective of time 0 are given by:
∞
X
E β t u(ct , 1 − lt )
t=0

with 0 < β < 1. β is the subjective time discount factor. u is

increasing and concave.
The total time endowment is 1 and lt is the time spent working.
Each consumer chooses quantities and takes prices as given.

Irasema Alonso The Real Business Cycle Model February 2025 7 / 48

The variables yt , ct , and so on, are stochastic: they are driven by
the stochastic process for technology zt .
The consumer’s budget constraint reads

ct + xt = rt kt + wt lt ,

where rt is the market return to capital and wt the wage rate.

rt and wt are stochastic processes.
Substituting for xt :

ct + kt+1 = (1 + rt − δ)kt + wt lt

Irasema Alonso The Real Business Cycle Model February 2025 8 / 48

The consumer maximizes the utility function subject to the
budget constraint holding at all dates.
∞
X
max Et β t u(ct , 1 − lt )
ct ,lt ,kt+1
t=t

subject to
ct + kt+1 = (1 + rt − δ)kt + wt lt .
Firms maximize profits under perfect competition:

max F (zt , kt , lt ) − rt kt − wt lt .
kt ,lt

Equilibrium profits are zero.

Irasema Alonso The Real Business Cycle Model February 2025 9 / 48

The FOCs for consumer and firm optimization govern how the
model works.
Consumer’s intertemporal FOC which determines the
consumption-saving choice:

u1 (ct , 1 − lt ) = βE[u1 (ct+1 , 1 − lt+1 )(1 + rt+1 − δ)].

The marginal utility loss of lower consumption in period t must

equal the expected discounted value of the return to higher
investment, in terms of next period’s utility.
Consumer’s intratemporal FOC which determines the
labor-leisure choice:

u1 (ct , 1 − lt )wt = u2 (ct , 1 − lt ).

The marginal loss of working one more unit of time in terms of

lost leisure must equal the wage times the marginal utility gain of
higher earnings.

Irasema Alonso The Real Business Cycle Model February 2025 10 / 48

Firms’ profit maximization imply:

rt = F2 (zt , kt , lt )

and
wt = F3 (zt , kt , lt )

The equilibrium of this model takes the form of stochastic

processes for quantities and prices that satisfy all the
equilibrium conditions (consumer’s utility maximization, firms’
profit maximization, and market clearing).
Kydland and Prescott solved the social planner’s problem and
used recursive analysis where equilibrium processes are
expressed as functions of the economy’s state variables.
A state variable is a variable that is predetermined at time t and
that matters to outcomes.
In this case, the state variables are the current shock and the
capital stock: (zt , kt ).
Irasema Alonso The Real Business Cycle Model February 2025 11 / 48
Recursive formulation
The problem of the consumer in sequential form is:
∞
X (ct v(1 − lt ))1−σ − 1
max E0 βt
ct ,lt ,kt+1 1−σ
t=0

subject to the resource constraint:

ct + kt+1 = zt F (kt , lt ) + (1 − δ)kt
log zt+1 = ρ log zt + t+1
Using recursive formulation and eliminating consumption:
((zF (k, l) + (1 − δ)k − k 0 )v(1 − l))1−σ − 1

V (k, z) = max +
k 0 ,l 1−σ

0
βE(V (k 0 , e ρ log z+ ))

for all (k, z). k 0 and 0 are the values of capital and  next period.
Irasema Alonso The Real Business Cycle Model February 2025 12 / 48
Solution

This problem delivers two decision rules:

One for capital:
k 0 = g (k, z)
and one for labor:
l = h(k, z)

In general, it is not possible to find explicit forms for g and h.

One needs to use numerical analysis to approximate these
functions.

Irasema Alonso The Real Business Cycle Model February 2025 13 / 48

In a very special case of the model, it is possible to obtain a full
analytical (as opposed to numerical) characterization of the
equilibrium.
If
u(c, 1 − l) = (1 − φ) log c + φ log(1 − l),
F (z, k, l) = zk α l 1−α ,
and δ = 1, then it can be verified that all the equilibrium
conditions are satisfied when labor supply is constant over time
and consumption is a constant fraction of output.
In this case,
h(k, z) = l̄ and
g (k, z) = szk α (l̄)1−α
for some constants l̄ and s.

Irasema Alonso The Real Business Cycle Model February 2025 14 / 48

To verify that this is a solution (and solve for the values of l̄ and s
as a function of primitives), just substitute into the functional
Euler equations (see below)—k and z will drop out.
This special case is not interesting from a business-cycle
perspective:
Since physical capital depreciates at a rate on the order of 10% per
year, and full depreciation occurs after many, many years, a period
in the model would be very long.
Business cycles models require the period length to be much
shorter, a quarter or a year.
Labor supply is counterfactually constant: labor supply doesn’t
respond to shocks.

Irasema Alonso The Real Business Cycle Model February 2025 15 / 48

Solving the model numerically

To solve the model numerically for a more reasonable parameter

calibration (where δ is around 0.1), we will use perturbation
methods.
Perturbation methods are Taylor expansions of the key equations
and their dynamic variables around a steady state.
Linearization is the simplest case.

Irasema Alonso The Real Business Cycle Model February 2025 16 / 48

The functional equations

First-order condition with respect to capital, k 0 :

(zk α (h(k, z))1−α + (1 − δ)k − g (k, z))−σ v(1 − h(k, z))1−σ =

n
βE (z 0 g (k, z)α (h(g (k, z), z 0 ))1−α + (1 − δ)g (k, z) − g (g (k, z), z 0 ))−σ
o
v(1 − h(g (k, z), z 0 ))1−σ (αz 0 g (k, z)α−1 h(g (k, z), z 0 )1−α + 1 − δ)

for all (k, z).

First-order condition with respect to labor, l:

(zk α (h(k, z))1−α + (1 − δ)k − g (k, z))v 0 (1 − h(k, z)) =

v(1 − h(k, z))(1 − α)zk α (h(k, z))−α

for all (k, z).

Irasema Alonso The Real Business Cycle Model February 2025 17 / 48

Linearizing the policy functions

In these functional equations, there are two unknown functions:

the saving function g and the labor function h.
To approximate these functions to a linear one, there are two
steps:
First, solve for the steady state.
This delivers k̄ and l̄.
The steady-state value for z is solved from its law of motion with
 = 0, so log z = 0, i.e., z = 1.
Second, to obtain gk , gz , hk , and hz , take derivatives of the two
functional equations, one with respect to k, and one with respect
to z, thus delivering four equations.
Then evaluate at steady state (where we now know k̄ and l̄) to
solve for the slopes of the decision rules at the steady state.
For gk , we obtain a second-order polynomial equation.

Irasema Alonso The Real Business Cycle Model February 2025 18 / 48

Policy functions – linearization

Recall Taylor’s approximation:

g (k, z) ≈ g (k̄, z̄) + gk (k̄, z̄)(k − k̄) + gz (k̄, z̄)(z − z̄)

Together, the two steps give the levels and the slopes of the two
unknown functions, and thus, the two first-order Taylor
approximations:

k 0 ≈ k̄ + gk (k − k̄) + gz (z − 1)

and
l ≈ l̄ + hk (k − k̄) + hz (z − 1).

Irasema Alonso The Real Business Cycle Model February 2025 19 / 48

Policy functions – loglinearization

Now we write variables as deviations from the steady state which

is convenient when we do IRFs.
Denote
k − k̄
k̃ =
k̄
Then,
k − k̄ = k̃ k̄ and k 0 − k̄ = k̃ 0 k̄
We had:
k 0 ≈ k̄ + gk (k − k̄) + gz (z − 1)
Then,
k̃ 0 k̄ ≈ gk k̃ k̄ + gz z̃
Simplifying
gz
k̃ 0 ≈ gk k̃ + z̃
k̄
Irasema Alonso The Real Business Cycle Model February 2025 20 / 48
Policy functions – loglinearization

Our policy functions are now:

gz
k̃ 0 ≈ gk k̃ + z̃
k̄

k̄ hz
l̃ ≈ hk k̃ + z̃
l̄ l̄

Irasema Alonso The Real Business Cycle Model February 2025 21 / 48

The slopes

The consumption Euler equation, derivative with respect to k:

will have gk , gk2 , and hk .
The labor-leisure equation, derivative with respect to k: will have
gk , and hk .
These two equations can be solved jointly for gk , and hk ; here,
one equation has a square and there will be two solutions, one
with |gk | > 1 that can be discarded and one with |gk | < 1.
Why do we discard |gk | > 1?
gz
k̃ 0 = gk k̃ + z̃
k̄

Then the third and fourth equations can be solved for gz , and hz .

Irasema Alonso The Real Business Cycle Model February 2025 22 / 48

Transmission mechanism: a numerically solved
case

The economy is at the steady state when zt = 1, yt = ȳ, ct = c̄,

kt = k̄, and so on for all t.
We can study local dynamics: the response around the steady
state of the different macroeconomic variables from a (small)
one-time higher-than-average productivity shock (zt > 1).

Irasema Alonso The Real Business Cycle Model February 2025 23 / 48

Transmission mechanism, cont’d

How do technology shocks translate into output movements in

this model?
A positive technology shock in period t represents a
higher-than-average total factor productivity.
Higher productivity rises wages, so labor supply in period t
increases as workers find work more profitable than leisure.
Period-t output rises through the direct effect of higher
productivity and the indirect effect of higher labor input.
The return to capital increases as well, but the capital stock in
period t is predetermined.
The boost in period-t output has dynamic consequences.
Part of the increase in output is consumed, while the remainder
is saved and invested.

Irasema Alonso The Real Business Cycle Model February 2025 24 / 48

The less quickly the productivity shock dies out, the more
profitable will it be to save and invest.
An investment response to a current shock is higher than if
technology growth were uncorrelated over time.
This raises the capital stock in period t + 1, while technology is
still above trend due to autocorrelation.
These dynamic effects constitute the model’s “propagation
mechanism”, where an “impulse” of a temporary technology
shock shapes the path of future macroeconomic variables.

Irasema Alonso The Real Business Cycle Model February 2025 25 / 48

Impulse-response functions

The following impulse-response functions are obtained by

hitting the economy with a 1% positive shock in z in period t
(t = 0 on the graphs), and shutting down shocks for the
subsequent periods.
The utility function is given by

u(c, 1 − l) = (1 − φ) log c + φ log(1 − l)

The values of the parameters are φ = 0.7, α = 0.32, δ = 0.025,

ρ = 0.95, and β = 0.9.
The vertical axis in the figures below represents deviations from
the steady state.

Irasema Alonso The Real Business Cycle Model February 2025 26 / 48

Technology parameter z 1.01

1.008

1.006

1.004

1.002

1
0 20 40 60 80 100
Time
Irasema Alonso The Real Business Cycle Model February 2025 27 / 48
 1.01 2

Capital: % Change from SS val

Technology parameter z
1.008
1.5
1.006
1
1.004
0.5
1.002

1 0
0 20 40 60 80 100 0 20 40 60 80 100
Time Time

Investment: % Change from SS value Output: % Change from SS value

Consumption: % Change from SS value Labor: % Change from SS value

0.8 2

0.6
1.5
0.4
1
0.2
0.5
0

-0.2 0
0 20 40 60 80 100 0 20 40 60 80 100
Time Time

1.5 15

10
1

0.5
0

0 -5
0 20 40 60 80 100 0 20 40 60 80 100
Time Time
Irasema Alonso The Real Business Cycle Model February 2025 28 / 48
Calibration

Calibration is a way of picking values for the parameters of the

model.
It requires that values for parameters be picked from sources
independent of the phenomenon under study.
Admissible sources of parameter values are:
Household data on consumption, hours worked, and other
microeconomic evidence, for individual preference parameters.
Long-run trend data for the factor shares in production (α in the
Cobb-Douglas case).

Irasema Alonso The Real Business Cycle Model February 2025 29 / 48

Utility

We will assume that preferences are given by the utility function:

(cl θ )1−σ − 1
u(c, l) =
1−σ

The economy is populated by a number of identical households.

The size of the household’s population grows at rate η.

Irasema Alonso The Real Business Cycle Model February 2025 30 / 48

Social planner’s problem

The social planner’s formulation of the utility maximization

problem is:
∞
X
max β t (1 + η)t u(ct , lt )
ct ,lt
t=0

The central planner faces an aggregate resource constraint:

Ct + Xt = A(1 + γ)t(1−α) Ktα Nt1−α ,

where Ct (consumption), Xt (investment), Kt (capital), Nt (labor)

denote aggregate variables.
The production technology is subject to a labor-augmenting
(deterministic) process with growth rate γ.

Irasema Alonso The Real Business Cycle Model February 2025 31 / 48

Pt is the population size at t (that grows at rate η).
Divide the resource constraint by this population size:
 α  1−α
Ct Xt Kt Nt
+ = A(1 + γ)t(1−α)
Pt Pt Pt Pt

ct + xt = A(1 + γ)t(1−α) ktα nt1−α ,

where small-size letters denote per-capita variables.
Individuals’ time endowment is limited:

lt + nt = 1.

Irasema Alonso The Real Business Cycle Model February 2025 32 / 48

The accumulation equation for capital is the usual one:

Kt+1 = (1 − δ)Kt + Xt .

Divide by population at t to obtain per capita terms:

Kt+1 Kt Xt
(1 + η) = (1 − δ) +
Pt (1 + η) Pt Pt

(1 + η)kt+1 = (1 − δ)kt + xt .

Thus, we rewrite the resource constraint as:

ct + (1 + η)kt+1 = A(1 + γ)t(1−α) ktα nt1−α + (1 − δ)kt .

Irasema Alonso The Real Business Cycle Model February 2025 33 / 48

Stationary model
Using our knowledge that in this model y, c, k, x grow at rate γ,
we define the de-trended variables:
ct xt kt
c̃t = t
, x̃t = t
, k̃t = ,
(1 + γ) (1 + γ) (1 + γ)t

The resource constraint:

ct + (1 + η)kt+1 = A(1 + γ)t(1−α) ktα nt1−α + (1 − δ)kt
in terms of the transformed variables becomes:
c̃t (1 + γ)t + (1 + η)k̃t+1 (1 + γ)t+1 =
A(1 + γ)t(1−α) (k̃t (1 + γ)t )α nt1−α + (1 − δ)k̃t (1 + γ)t
Dividing by (1 + γ)t :
c̃t + (1 + η)(1 + γ)k̃t+1 = A k̃tα nt1−α + (1 − δ)k̃t
Irasema Alonso The Real Business Cycle Model February 2025 34 / 48
We specify σ = 1 in the utility function, which leads to a
logarithmic utility function:

u(c, l) = log c + θ log l

Notice that
log c = log c̃ + log(1 + γ)t
but the term (1 + γ)t does not depend on choice variables, and
hence it is irrelevant when we take first-order conditions with
respect to capital next period and labor.

Irasema Alonso The Real Business Cycle Model February 2025 35 / 48

We ignore this term and thus the transformed problem is:
∞
X
max β t (1 + η)t (log c̃t + θ log lt )
c̃t ,lt ,k̃t+1 t=0

s.t.

c̃t + (1 + γ)(1 + η)k̃t+1 = A k̃tα (1 − lt )1−α + (1 − δ)k̃t .

We need to pick values for the parameters involved. We begin

with the ones that are immediately available:
α = 0.4 (capital share of output – constant)
γ = 0.02 (average long run growth rate)
η = 0.01 (average long run population growth rate)

Irasema Alonso The Real Business Cycle Model February 2025 36 / 48

Finding δ
δ can be found in the following way. In the steady state of the
transformed model, we have that
k̃t+1 = k̃t = k̃ ∗ .

Recall the capital accumulation equation:

(1 + γ)(1 + η)k̃ ∗ = (1 − δ)k̃ ∗ + x̃ ∗ .

Dividing both sides by k̃ ∗ yields:

x̃ ∗ X∗
(1 + γ)(1 + η) = 1 − δ + =1−δ+
k̃ ∗ K∗

In this equation, γ, η, and the ratio of investment to the capital

∗
stock, KX ∗ , are known.
Irasema Alonso The Real Business Cycle Model February 2025 37 / 48
Finding δ

From the data

X US
= 0.076
K US
and since
X∗
(1 + γ)(1 + η) = 1 − δ +
K∗
hence δ can be solved for: δ = 0.0458.

Irasema Alonso The Real Business Cycle Model February 2025 38 / 48

Finding β
Now we look at the parameters in the utility function: β and θ.
For these we need to take first-order conditions of the problem.
We differentiate with respect to next period’s capital k̃t+1 and,
assuming a balanced growth path (c̃t = c̃t+1 ), we obtain:
1 + γ = β(αA k̃tα−1 (1 − lt )1−α + 1 − δ)

We can observe that

ỹt Yt
αA k̃tα−1 (1 − lt )1−α = α =α ,
k̃t Kt
Yt
where Kt is available from actual data:

YtUS
≈ 0.3012.
KtUS
With this, β can be solved for. The result is β = 0.94912.
Irasema Alonso The Real Business Cycle Model February 2025 39 / 48
Finding θ
For θ, we need the remaining first order conditions. The Euler
equation determining labor supply is:
1 1
θ = (1 − α)A k̃tα (1 − lt )−α
lt c̃t

Multiplying LHS and RHS by 1 − lt :

1 − lt ỹt Yt
θ = (1 − α) = (1 − α)
lt c̃t Ct

Yt
Let us first look at Ct . We have that:
Yt Yt Kt
=
Ct Kt Ct
and
Xt Ct Yt Ct Yt Xt
Xt + Ct = Yt ⇒ + = ⇒ = − .
Kt Kt Kt Kt Kt Kt
Irasema Alonso The Real Business Cycle Model February 2025 40 / 48
Since we know the values of KYtt and KXtt from actual data, we can
find CYtt .
Ct Yt
= 0.2252 and = 1.3375
Kt Ct
Next is to find a reasonable estimate of lt .
We use knowledge from microeconomic studies. Out of the total
24 hours daily endowment, 8 hours are used for sleeping and 8
for work.
Then, we can use lt ≈ 2/3. Using this, we can solve for θ, which
yields θ = 1.605.

ỹt Yt
0.5θ = (1 − α) = 0.6 = 0.6 × 1.3375 = 0.8025.
c̃t Ct

Irasema Alonso The Real Business Cycle Model February 2025 41 / 48

Technology parameters: ρ and σ 2
Solow growth accounting measures technological change as:

gz ≈ gy − αgk − (1 − α)gl .

We can measure all these growth rates (gy , gk , and gl ) in the data.
Using the Solow residual, gz , we construct the raw series for the
productivity shock zt in levels.
Then we take the logarithm of this series and HP-filter it. That
gives a series of percentage deviations from the trend, z̃.
We use this series to estimate the AR(1) process for z̃.

z̃t+1 = ρz̃t + t+1 .

We match ρ to its first-order autocorrelation and the variance of

, σ 2 , to the variance of the error terms in the estimated equation.
This delivers ρ = 0.9 and σ = 0.7 percent.
Irasema Alonso The Real Business Cycle Model February 2025 42 / 48
Simulate the model and analyze the outcome

Once all of the parameter values have been assigned, the model
is fully specified and can be solved numerically.
A random number generator is used to simulate a realization of
the stochastic shock.
Given the functions g and h, output and other variables of
interest can be simulated by selecting an initial capital stock.
This gives rise to a time series in each of the variables.
These series are the researcher’s “data set”.
Sample second moments of the variables are computed and
compared to actual data.

Irasema Alonso The Real Business Cycle Model February 2025 43 / 48

Quantitative evaluation, continued

Some key statistics of the de-trended data and the associated

model output are summarized in the table below.
The model was simulated 100 times, each with a sample length
of 150 time periods.
The numbers in the table are averages across simulations.

Irasema Alonso The Real Business Cycle Model February 2025 44 / 48

Some statistics from the model (data)

Irasema Alonso The Real Business Cycle Model February 2025 45 / 48

Results

The table shows that the variation in output in the model is

somewhat lower than in the U.S. data, but the discrepancy is not
large.
The model predicts that consumption varies much less and
investment much more than output. This is a result of
consumption smoothing.
The variation in hours in the model is less than half of that in the
data.
This discrepancy is largely due to the fluctuation in the number
of employed workers in the data; the simple model here focuses
on variations in the number of hours worked per worker (the
intensive margin) and abstracts from variations in labor-force
participation and unemployment (the extensive margin).

Irasema Alonso The Real Business Cycle Model February 2025 46 / 48

The table also shows strikingly high correlations between output
and other macroeconomic variables.
The correlations between consumption and output and between
investment and output are quite similar in the model and in the
data.
As for the correlation between hours and output, the model does
not generate a co-movement as high as that observed in the U.S.
economy.
This is another consequence of the rudimentary modeling of
labor supply in this simple model.

Irasema Alonso The Real Business Cycle Model February 2025 47 / 48

The RBC model I presented is quite special but it is the core of
modern macro models.
The model only has technology shocks – more advanced models
have many other shocks, such as monetary shocks, fiscal policy
shocks, or “demand” shocks.
In addition, the model I presented has markets that work well;
nowadays the main models also introduce various frictions, i.e.,
deficiencies in various parts of the markets (like the credit
market and the labor market).

Irasema Alonso The Real Business Cycle Model February 2025 48 / 48