Modeling by the general equilibrium approach

Stochastic Dynamics
Derivation, log-linearization and simulations

Grace Kiala* and Jean-Paul K. Tsasa†

LAREQ ONE PAGER 11-001 – FEBRUARY 01, 2016

[Link]

Summary

Modern macroeconomics is built on microeconomic foundations. Economic agents are

supposed rational, and the cyclical fluctuations of GDP, consumption, investment or the
decisions to work are perceived as optimal responses of economic agents, i.e. households and
firms, to stochastic exogenous shocks. In this paper, we illustrate the resolution of a model
dynamic stochastic general equilibrium model of the RBC type. This model is generally considered as the
starting point of modern analysis on the economic cycle.

Keywords: RBC model, economic cycle, log-linearization, stochastic simulations.

Abstract

Modern macroeconomics is built on the microeconomic foundations. A representative household maximizes its
intertemporal utility and representative firm maximizes his profit. Cyclical fluctuations of the GDP,
Consumption, investment, or worked hours are considered the result of the optimal responses of economic.
agents to stochastic shocks on global productivity. In this paper, we illustrate how a dynamic stochastic
general equilibrium model, especially the RBC model, can be solved. In the modern literature, the RBC model
is often considered the starting point for most studies on the business cycles.

RBC model, business cycle, log-linearization, stochastic simulations.

JEL classification: E13, E32, E37.

*
Research candidate at the Laboratory of Quantitative Economics Analysis-Research. Protestant University in
Congo. Address: Intersection of Av. Libération and Blvd Triomphal, P.O. Box 16.626 Kinshasa I, DRC. Email:
[Link]@[Link]
†
Researcher at the Laboratory of Analysis-Research in Quantitative Economics. Email:jeanpaultsasa@[Link]
 G. Kiala and J-P K. Tsasa. Modeling using the DSGE approach
[2016, Lareq One Pager, Vol. 11, no. 1, 1-14]

I. Introduction

If we had to choose two papers not to be ignored in the modern analysis of the economic cycle,
We will choose Lucas's paper (1977) and that of Kydland and Prescott (1982). In history
economically, the first paper is seen as a reference article in defining what one
understands 'economic cycle' in the modern sense. Indeed, unlike Burns and Mitchell (1946)
who consider the cycle as a succession of four phases that inevitably evolve from one
towards the other: expansion, crisis, recession and recovery, Lucas (1977) redefines the cycle as a set
of regularities1characterizing the overall evolution of the economic system. Moreover, in the second
paper, Kydland and Prescott (1982), drawing on the new conception of the cycle advocated by Robert
E. Lucas aims to provide a framework for analysis that allows for the reproduction of the main properties.
cyclical patterns of the American economy. The main contribution of the paper by Kydland and Prescott (1982)
is to have succeeded in imposing, both on classical economists and on those of a different allegiance
Keynesian, dynamic stochastic general equilibrium models as a frame of reference for
the analysis of economic fluctuations.

In this paper, we will demonstrate in more or less detail how the models work.
dynamic stochastic general equilibrium (DSGE). The DSGE model is a model
macroeconomic systems that are built on microeconomic foundations. To focus on
On the methodology and strategy for solving, we consider a standard version of this class.
of models. Indeed, we assume that households and firms are price takers in the
markets for goods, capital, and labor. Households are assumed to maximize their utility, and the
strong their profit. We disregard the public sector (government) and the central bank.
Prices and wages are assumed to be flexible and markets are competitive. Finally, there is no
financial intermediary, no information asymmetry. The dynamics in this economy is
governed by an exogenous stochastic technological process. Therefore, there are no monetary shocks
neither budgetary rigidities nor real or nominal rigidities in our analysis model.

The rest of the paper is organized as follows. In the second section, we proceed to the derivation of
In our analytical framework, and in the third section, we discuss the results of the simulations of the log model.
calibrated linear. The fourth section concludes.

II. Model

Consider a dynamic stochastic general equilibrium model of the RBC type. For simplicity,
we assume that capital does not depreciate, labor is normalized to 1, and utility is
logarithmic. The household maximizes its utility function, and the firm maximizes its profit. The market
Being competitive, each factor of production is compensated by its marginal product. By the theorem
From Euler, we can easily transition from the market problem to that of the social planner. At
In light of our assumptions and based on two theorems of welfare, it follows that the market solution
and that of the coincident planner. Thus, in the following lines, we will solve our model
following the planner's approach. For an illustration of the resolution following the market approach,
see Lokota and Tsasa (2014). The problem of the representative agent is such that:

1
Notably the co-movements of the main macroeconomic variables, their persistences, and then their
deviations from the trend.

2
 G. Kiala and J-P K. Tsasa. Modeling using the DSGE approach
Lareq One Pager

max% '(% log

', (1)
'.%

subject to :

%− %(2= %− %, 2

%=
2 (6
%
6
%(2 , (3)

log% = f log y%(2+ log %, (4)

where % is the operator of mathematical expectation, % is the production function, % is the capital, % the

consumption % the technological shocklog % with a mean equal to zero and a variance <. There is
three possible approaches to solving: one approach based on dynamic Lagrangian and another
approach based on solving the Bellman equation. We will adopt the approach based on the
solution of the Bellman equation. For the approach using dynamic Lagrangian, the concerned party can
refer to Lokota and Tsasa (2014). For the mathematical foundations of the use of the equation
on Bellman in the resolution of modern macroeconomic models, see Emone et al. (2014).

The first step in solving a dynamic stochastic general equilibrium model is to

derive the equilibrium system. The latter is obtained by executing the first-order conditions. The
the representative agent problem being well defined, the first-order conditions are sufficient for
characterize the solution. The Bellman equation corresponding to this problem is written as:

%(2, % maximum log %+ % %, %B2

? @ A@

subject to:

2 (6 6
%= % %(2− %+ %

log% = F log G%(2 + log %,

where %(2and % are state variables and, % and % control variables. By replacing the variable %

by constraint and by executing the first-order conditions with respect to %, we find:

%:
1
− + %2 %, %B2= 0, 5
%

where :

%, %B2
2 %, %B2= (6)
%

3
 G. Kiala and J-P K. Tsasa. Modeling using the DSGE approach
[2016, Lareq One Pager, Vol. 11, no. 1, 1-14]

The envelope condition is such that:

2 (6 6(2
1
2 %, %B2= % %(2 +1
%
2(6 (7)
% 1
= +1 ,
%(2 %

with :
2(6
%
%= + 1, (8)
%(2

where % is the gross interest rate. By shifting by one period, we find:

2(6
%B2 1
2 %, %B2= +1 (9)
% %B2

By substituting (9) into (5), we obtain:

2(6 2(6
1 %B2 1 %B2 %
= % +1 1= % +1 10
% % % %B2

The equation (10) is a condition of optimality, describing the intertemporal Euler equation. Indeed, this
equation expresses the intertemporal arbitrage determining the optimal level of the
consumption during the period and the period + 1. According to this equation, it follows that a reduction
of a unit of consumption during the period decreases the usefulness of1% . From then on, the unity of the

consumption that is spared during the period can on average (expectation) be converted to

%B2 %
2(6+ 1units for the period + 1 and thus increase the usefulness of % %B2 %
2(6+ 1 %B2, where
the expression %B2 %
2(6+ 1is nothing other than the gross interest rate.

The second step is to characterize the stationary equilibrium system. After executing the
first-order conditions, we now know that the entire set of equilibrium equations of the model
as follows:
2(6
%B2 %
1= % +1 (11)
% B2

2 (6 6
%= % %(2− %+ %(2 (12)

log% = logarithm%(2+ log % (13)

Otherwise, the balance system can also be expressed as follows:

%
1= % % (14)
%B2

2(6
%B2
%= +1 15
%

4
 G. Kiala and J-P K. Tsasa. Modeling through the DSGE approach
Lareq One Pager

2 (6 6
%= % %(2− %+ %(2 (16)

log% = f log g%+ log % (17)

In the stationary state, the equilibrium system becomes:

1= (18)

2(6
= +1 (19)

= 2(6 6− + (20)

log = log + log (21)

After a basic algebraic calculation, the stationary system becomes:

1
= (22)

=1 (23)

=1 (24)

2
2(6
= (25)
-1

= 6
(26)

After rearrangement, we finally find that:

1
= (27)

1 (28)

=1 (29)

2
2(6
= 30
1−

6
2(6
= 31
1−

where now, each variable depends on the parameters.

5
 G. Kiala and J-P K. Tsasa. Modeling using the DSGE approach
[2016, Lareq One Pager, Vol. 11, no. 1, 1-14]

Let's proceed to the log-linearization of the equilibrium system. Let % the deviation in logarithm of the
variable % in relation to its value in steady state :

% = log G% − log ⅏. (32)

It comes that:
%
M
@= ⟹ %=
M
.
@ (33)

The first-order Taylor approximation implies that:

%≈ 1+ % . (34)

By applying this principle to the equilibrium system, we find:

⎯ Equation 14:
% %
1= % % 0= % % -1
%B2 %B2

Log-linearization:
?
@
0isapproximatelyequalto @ O- 1
% ?
@PQ

0≈ R@ ?@ (?@PQ -1
%

0≈ % 1+ % 1+ % − %B2- 1

Since1 and % %− %B2≈ 0he comes that :

0≈ %1+ % 1+ʂ % − %B2- 1

0≈ % %+ %− %B2 (35)

⎯ Equation 15:
2(6
%B2
%= +1
%

Log-linearization:
2(6
S@PQ
0≈ R− @ -1
A@

0≈ 1+ % − 2(6
1 +1 − bag %B2 − %- 1

Since:

= 2(6
+ 1,

he comes that:

6
 G. Kiala and J-P K. Tsasa. Modeling using the DSGE approach
[2016, Lareq One Pager, Vol. 11, no. 1, 1-14]

Item 1−
0≈ + %− 2(6 %B2− % − 2(6
+1
.R

Bouquet 1−
0≈ %− 2(6 %B2− %

1−Ᵽ
0≈ %− 2(6 %B2− %

Since:

2(6
=L-1

he comes that:
1
%≈ 1− 1− %B2− %

Knowing that:
1
= ,

we finally find:

%≈ 1− 1− %B2− % (361)

⎯ Equation 16:
2 (6 6
%= % %(2− %+ %

Log-linearization:

2(6 6
T
@≈ U@ V@WQ V@ V@WQ
− +

1+ 61
%≈ +1 − A% + K%(2− 1+ % -1+ %

Since = 6, he comes that:

%≈ 1− A% + K%(2 − %− %(2
(37)

⎯ Equation 17:
log% = F log G%(2+ log %

Log-linearization:

% + log ≈ %(2+ log + % + log

0≈ %− %(2− % (38)

7
 G. Kiala and J-P K. Tsasa. Modeling using the DSGE approach
Lareq One Pager

After the log-linearization, the set of equations of the equilibrium system is written as:

0≈ % %+ %− %B2 (39)

%≈ 1− 1− %B2− % (40)

%≈ 1− A% +K%(2 − %− %(2 (41)

0≈ %− %(2− % 42

This system of log-linear equations thus characterizes our economy. It is therefore the model that we
Let's use this to conduct stochastic simulations in the following section.

III. Simulations

III.1. Calibration

The numerical exercise of simulations requires defining, above all, the values to assign to the parameters.
of the model, i.e. the parameterization or calibration. Our model distinguishes three classes of parameters.
The first class includes any parameter that describes the behavior or preferences of the household.
In the case of our model, there is only one. It is the factor " (BETA). This parameter
defines the degree of consumer patience. It ranges between 0 and 1:0< < 1. The closer it is to
the unit, the more patient the household is. We set it at 0.99. The second class groups all parameters
allowing to characterize the behavior of the firm. Our model being very basic, we do not have
a single parameter to describe the company's technology: " (ALPHA). Indeed, the parameter
represents the share of capital in production. Its value, in the literature, is generally set at
0.33 based on the empirical evidence of the American economy. But it is worth noting that its value
can vary significantly from one economy to another.

Table 1: Calibration

Preference

BETA 0.99

Technology

ALPHA 0.33

Stochastic process

RHO 0.98 / 0.80

SIGMA 0.712

Finally, the third class defines the parameters of the stochastic process. The productivity process
is supposed to be an autoregressive representation of order 1, i.e. an AR(1). The shock can be
permanent (BETA = 1) or transitory (0 < BETA < 1). In our model, we assume that the shock

8
 G. Kiala and J-P K. Tsasa. Modeling using the DSGE approach
[2016, Lareq One Pager, Vol. 11, no. 1, 1-14]

technological is transitional and highly persistent. The strong persistence is reflected through the value
taken by the autocorrelation coefficient which is close to one. Then, for the simulation exercise,
we will assume a decrease in the persistence of technological innovations to 0.8. Finally, the gap-
The type of innovations is set at 0.712. This value allows for matching the model's predictions to
regularities observed on data.

III.2. Programming on Dynare

We will now program our model on the Dynare software. The Dynare platform allows for
simulate and estimate rational expectations models. It was originally developed by Juillard
(1996). Dynare is free software and works either under MatLab or under Octave.

Table 2: Dynare code of the log-linear model

close all;
clc;

% Declaration Endogenous Variables

var c_hat k_hat a_hat r_hat;

% Declaration Variable exogenous

varexo u;

% Declaration Parameters
parameters ALPHA BETA RHO SIGMA;

Calibration
ALPHA = 0.33; % Share of capital in the production function
BETA = 0.99; % Discount factor
RHO = 0.98; % Autoregressive coefficient of the stochastic process
%RHO = 0.8;
SIGMA = 0.712; % Ecart-type des innovations

% Declaration Module
% 4 endogenous variables & 4 equations
model(linear);
#r_ss=(1/BETA);
#k_ss=(ALPHA/(r_ss-1))^(1/(1-ALPHA));
#c_ss=k_ss^ALPHA;

% Inter-temporal Euler
0 = r_hat + c_hat - c_hat(+1);

Capital remuneration
(1-ALPHA)*(1-BETA)*(a_hat(+1)-k_hat);

Equilibrium: c = y - i
c_hat = (1 - ALPHA) * a_hat + ALPHA * k_hat(-1) - (k_ss / c_ss) * (k_hat - k_hat(-1));

Stochastic process
0 = a_hat - RHO * a_hat(-1) - u;
end;

% Initialization
steady_state_model;
c_hat=0;

9
 G. Kiala and J-P K. Tsasa. Modeling using the DSGE approach
Lareq One Pager

k_hat=0;
r_hat=0;
a_hat=0;
end;

% Shocks
shocks;
var u = SIGMA^2;
end;

Steady state
steady
check;

Simulations
stoch_simul(hp_filter = 1600, order = 1, irf=100, periods=250);

III.3. Simulation with a more persistent shock

First, let us consider that the autocorrelation coefficient of the technological process takes the value
of 0.98. In this case, the results provided by Dynare are summarized as follows.

Table 3: STEADY-STATE RESULTS

c_hat 0
k_hat 0
a_hat 0
r_hat 0

EIGENVALUES

Modulus Real Imaginary

0.98 0.98 0
0.9899 0.9899 0
1.02 1.02 0
Inf Inf 0

Table 5: COVARIANCE MATRIX OF EXOGENOUS SHOCKS

Variables u
u 0.506944

Table 6: POLICY AND TRANSITION FUNCTIONS

c_hat k_hat a hat

r_hat
k_hat(-1) 0.658873 0.989933 0
-0.006633
hat(-1) 0.171948 0.014835 0.980000
0.006335
u 0.175457 0.015138 1.000000
0.006465

10
 G. Kiala and J-P K. Tsasa. Modeling using the DSGE approach
[2016, Lareq One Pager, Vol. 11, no. 1, 1-14]

Table 7: MOMENTS OF SIMULATED VARIABLES (HP filter, lambda = 1600)

VARIABLE MEAN Standard DeviationVARIANCE

c_hat 0.143255 0.146586 0.021488 0.011206 -0.207604
k_hat 0.707805 0.042677 0.001821 0.216230 -0.936339
a_hat -1.982328 0.839823 0.705303 -0.039813 -0.072359
r_hat -0.017758 0.005471 0.000030 -0.052554 -0.037784

Table 8: CORRELATION OF SIMULATED VARIABLES (HP filter, lambda = 1600)

VARIABLE c_hat
c_hat 1.0000 0.3607 0.9816 0.9705
k_hat 0.3607 1.0000 0.1762 0.1253
a_hat 0.9816 0.1762 1.0000 0.9987
r_hat 0.9705 0.1253 0.9987 1.0000

Table 9: AUTOCORRELATION OF SIMULATED VARIABLES (HP filter, lambda = 1600)

VARIABLE 1 2 3 4 5
c_hat 0.6808 0.3832 0.1577 0.0669 -0.0067
k_hat 0.9487 0.8443 0.7072 0.5575 0.4016
a_hat 0.6676 0.3600 0.1293 0.0415 -0.0279
r_hat 0.6676 0.3599 0.1290

Figure 1: IMPULSE RESPONSE FUNCTIONS

c_hat k_hat
0.25 0.4

0.2
0.3

0.15
0.2
0.1

0.1
0.05

0 0
20 40 60 80 100 20 40 60 80 100

a_hat x 10−3 r_hat

0.8 6

0.6 4

0.4 2

0.2 0

0 -2
20 40 60 80 100 20 40 60 80 100

The details on the interpretation of these different results have been largely presented in the series of
technical sheets dedicated to modeling using the dynamic general equilibrium approach
stochastic, see the following linkThe provided text is a URL and does not contain translatable content..

11
 G. Kiala and J-P K. Tsasa. Modeling using the DSGE approach
[2016, Lareq One Pager, Vol. 11, no. 1, 1-14]

III.4. Simulation with a less persistent shock

Now, let's suppose that the autocorrelation coefficient of the technological process takes the value of
0.80, that is to say becomes less persistent. In this case, the results printed by Dynare are summarized
as follows.

Tableau 10: STEADY-STATE RESULTS

c_hat 0
k_hat 0
a_hat 0
r_hat 0

Table 11: EIGENVALUES

Modulus Real Imaginary

0.8 0.8 0
0.9899 0.9899 0
1.02 1.02 0
Inf Inf 0

Table 12: MATRIX OF COVARIANCE OF EXOGENOUS SHOCKS

Variables u
u 0.506944

Table 13: POLICY AND TRANSITION FUNCTIONS

c_hat k_hat a_hat
r_hat
k_hat(-1) 0.658873 0.989933 0
-0.006633
a_hat(-1) 0.030093 0.015485 0.800000
0.004184
u 0.037616 0.019357 1.000000
0.005230

Table 14: MOMENTS OF SIMULATED VARIABLES (HP filter, lambda = 1600)

VARIABLE MEAN STD. DEV. VARIANCE

c_hat -0.088807 0.037316 0.001392 0.073830 -0.411655
k_hat -0.119007 0.041472 0.001720 0.223206 -0.684757
a_hat -0.388095 0.794739 0.631610 -0.130020 0.267893
r_hat -0.001283 0.004206 0.000018 -0.134828 0.270717

Table 15: CORRELATION OF SIMULATED VARIABLES (HP filter, lambda = 1600)

VARIABLE c_hat
c_hat 1.0000 0.8582 0.6932 0.6454
k_hat 0.8582 1.0000 0.2251 0.1619
a hat 0.6932 0.2251 1.0000 0.9979
r_hat 0.6454 0.1619 0.9979 1.0000

12
 G. Kiala and J-P K. Tsasa. Modeling using the DSGE approach
[2016, Lareq One Pager, Vol. 11, no. 1, 1-14]

Table 16: AUTOCORRELATION OF SIMULATED VARIABLES (HP filter, lambda = 1600)

VARIABLE 1 2 3 4 5
c_hat 0.7729 0.5262 0.3097 0.1817 0.0673
k_hat 0.9196 0.7579 0.5632 0.3740 0.1919
a_hat 0.5763 0.2178 -0.0217 -0.0781 -0.1129
r_hat 0.5758 0.2172 -0.0217 -0.0786 -0.1139

Figure 2: IMPULSIVE RESPONSE FUNCTIONS

c_hat k_hat
0.05 0.06

0.04 0.05

0.04
0.03
0.03
0.02
0.02
0.01 0.01

0 0
20 40 60 80 100 20 40 60 80 100

a_hat −3
x 10 r_hat
0.8 4

3
0.6

2
0.4
1

0.2
0

0 -1
20 40 60 80 100 20 40 60 80 100

All in all, it shows that by lowering the value of the parameter RHO, that is to say by moving from a shock
more persistent technological response to a less persistent technological shock, the response functions
impulsive for consumption and capital take on a bell-shaped curve
shaped). Besides this exercise, stochastic dynamic general equilibrium models are also
very capable of generating artificial data to test whether the model's predictions can
to be empirically relevant.

Conclusion
The objective of this paper was to show how general equilibrium models work.
stochastic dynamics of the RBC type. We proceeded to derive the equilibrium system, by
solving the social planner's problem, and then we proceeded to the derivation of the system
of log-linear equilibrium. Finally, we showed how the parameters of this class of models

13
 G. Kiala and J-P K. Tsasa. Modeling using the DSGE approach
[2016, Lareq One Pager, Vol. 11, no. 1, 1-14]

can easily be calibrated, before programming the Dynare code and conducting the simulations
stochastic.

Considering that these models are the starting point of modern macroeconomic analysis.
Important on the cycle, it will be interesting to see to what extent we can further engage with it.
serve to address questions related to the cyclical dynamics in developing countries.
For example, in the model we presented, we abstracted from the public sector and
from the central bank. This simplification, particularly the neutrality of the effects of policies
Economic models based on real variables are not realistic. Consequently, this model cannot provide us with anything.

to say how public spending affects household consumption or output,

or also on the impact of monetary policy on real variables.

Thus, building on the continuation of this paper, we will attempt to enrich our model in two
on one hand, we will increase it in order to model the public sector and thus address
notably the question of the macroeconomic impact of public spending, and on the other hand, we will
take into account the existence of a central bank in order to study the effect of monetary policy on
real variables. Another extension to explore is also the consideration of the sector
outside, because for several reasons, it may seem less relevant to think within a framework
autarkic.

References

• BURNS, Arthur F. and Wesley C. Mitchell, 1946, Measuring Business Cycles, NBER Book Series
Studies in Business Cycles, 590p.

• EMONE, Roger, Moïse MBIKAYI and Jean-Paul TSASA, 2014, 'From the Riemann Integral to the Integral
of Lebesgue. Construction, Properties, Limits, and Application in Dynamic Macroeconomics
Pager Laréq, vol. 9, no. 001, 1 – 46.

• JUILLARD Michel, 1996, “Dynare: A program for the resolution and simulation of dynamic models
with forward variables through the use of a relaxation algorithm,” CEPREMAP, Orange Coverage,
9602.

• KYDLAND, Finn E. and Edward C. PRESCOTT, 1982, "Time to Build and Aggregate Fluctuations"
Econometrica, Vol. 50, No. 6: 1345 – 1370.

• LOKOTA, Michel-Ange and Jean-Paul TSASA, 2014, 'Estimation of DSGE Models on Dynare. Study
Case I: Real Business Cycle (RBC) Model, Technical Sheet Laréq, Series Alpha–II, 1, 1-11.

• Understanding business cycles

Public Policy, Elsevier, vol. 5, No 1, pp. 7-29.

14