ФЕДЕРАЛЬНОЕ ГОСУДАРСТВЕННОЕ АВТОНОМНОЕ ОБРАЗОВАТЕЛЬНОЕ

УЧРЕЖДЕНИЕ ВЫСШЕГО ОБРАЗОВАНИЯ

«НАЦИОНАЛЬНЫЙ ИССЛЕДОВАТЕЛЬСКИЙ УНИВЕРСИТЕТ
«ВЫСШАЯ ШКОЛА ЭКОНОМИКИ»
Международный институт экономики и финансов

Умурзаков Содик Илхамидинович

Байесовская оценка модели DSGE
(Bayesian Estimation of a DSGE Model)
Курсовая работа
по направлению подготовки 38.03.01 «Экономика»
образовательная программа «Финансовая экономика»

Научный руководитель
доцент, PhD
У. С. Пейрис

Москва 2020
 Abstract

This paper is about estimation of the structural parameters of a DSGE

model - standard New Keynesian model using Bayesian approach. There
are set of respective shocks as total factor productivity, monetary policy
and government spending shocks and endogenous variables used to sim-
ulate the model with the real data. To estimate the model’s structural
parameters we apply to the US economy data and we detail the con-
struction of the extended time interval from 1q2003 to 4q2013. Mostly,
the deviation of GDP of the US economy is caused by the government
spending shock. Other two variables: inflation and aggregate interest
rate deviation are caused by the monetary policy and total factor pro-
ductivity shocks.

Key words: DSGE model, Bayesian Estimation, TFP shock, Business

Cycle, New Keynesian model
Contents
1 Introduction 2

2 Related literature 3

3 The Model 5
3.1 Firm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.2 Household . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.3 Monetary and Fiscal Policies of the Government . . . . . . . . . . . . . . . 6
3.4 Exogenous Processes Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.5 Equilibrium Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.6 Steady State Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.7 Log-linearization of New Keynesian Model . . . . . . . . . . . . . . . . . . 9
3.8 Assessment of the Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

4 Bayesian Estimation 11
4.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.2 Parameterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.3 Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

5 Discussion 13
5.1 Historical Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
5.2 Error Variance Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.3 Impulse Response Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5.4 Bayesian Time Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

6 Conclusion 22

7 References 23

1
1 Introduction
The Dynamic, Stochastic and General Equilibrium, usually it is called as DSGE, models
are widely used to study the economic decisions by academic researchers and Central
Banks. The academic researchers use them to conduct an empirical analysis in monetary
economics and financial stability, but mostly models are used by the Central Banks to
analyse the current situation of the economy, to find the answer to the issue of changing
in monetary and fiscal policies, and to create forecasts for key macroeconomic indicators.
By the DSGE models we mean Real Business Cycle (RBC) model, is the first kind of
DSGE models that, is used to replicate the stylized facts (Kydland and Prescott, 1982;
King, Plosser, and Rebelo, 1988) and New Keynesian models (Rotemberg and Woodford,
1997) that characterize the nominal price and wage rigidities, and play a key role for
Central Banks to change the interest rate level in response to high/low inflation and gross
domestic product (GDP) fluctuations (e.g. Schorfheide et al. (2016)). Overall, the com-
mon characteristic of DSGE models is that the preferences and technologies assumptions
of the economic agents are derived by solving the maximization problems and can solve
it in the recursive way. But there are some challenges that could be occur when economic
agents face with exogenous stochastic processes as total factor productivity, monetary
shock which is about setting the interest rate level by the central bank, government shock
(fiscal authority’s taxation strategy).
Now in our term paper we consider a DSGE model in case of small-scale New
Keynesian model where we discuss the optimal price and wage rigidities. Further, we
set up the model with agents and appropriate budget constraints for household, firm and
government.
The main issue that we are going to consider is to simulate above mentioned shocks
and get deviations of such endogenous variables as GDP, consumption, inflation and
investment. These variables depend on the numerical parameters of the models for further
implications of the economic agents’ preferences and productivity technologies in case of
uncertainty that occurs by above mentioned shocks.
In section 2 we consider literature by analysing the last update researches in our field.
Then, in section 3 we set up the model with frameworks and properties of equilibrium
conditions. In the last section we conduct the quantitative analysis and interpret the
main findings of the study.

2
2 Related literature

To date there are many fundamental researches on estimation of DSGE models and var-
ious kind of approaches have been used to replicate the model with the real-world data.
The model estimated for closed and open economy Adolfson (2007) where considered es-
timation of the different shocks and financial frictions to explain the dynamic direction of
an open economy by modifying the closed economy monetary shocks model of Chtistiano
et al. (2005). In his work he found that support for the nominal and real financial fric-
tions that are embedded in the model and sticky prices in the domestic import and export
sectors, sticky wages, investment adjustment costs and habit persistence in consumption.
Now before going into the technical parts of the model a few alternative methods
of estimating the standard New Keynesian model. There are some classical methods of
estimation if the small-scale New Keynesian models that characterize the nominal price
and wage rigidities, and play a key role for central banks to change the interest rate level in
response to high/low inflation and gross domestic product (GDP) fluctuations (Rotemberg
and Woodford, 1997). For first time the Bayesian estimation approach was used to assess
a DSGE model 20 years ago by Schorfheide (2000), DeJong, Ingram, and Whiteman
(2000), and Otrok (2001). For various approach of estimating DSGE models, one can see
Schorfheide et al. (2016). However, the limitation is that the author considered just only
for New Keynesian model, but not for RBC model. The approach consists of assessing the
accuracy of the Bayesian technical part which are applied in the DSGE models research
practice last two decades. Moreover, the author explored recent mathematical tools which
enhance the certainty of Monte Carlo (MC) approximations of posterior distributions
combined with a DSGE model.
Last years, the academic researchers and Central Banks actively use the Bayesian
approach to estimate the parameters of economic models by the difference shocks and
with different aspects of the economy (e.g. Andreev et al. (2020)), in the original paper
they have considered small open economy with New Keynesian model with commercial
banks and firms that have the probabilities going to be bankrupt because of unsecured
debt. The case was analysed for emerging markets particularly for Russian economy in
commodity price and financial stability. Main findings state that the impact of price shock
to GDP fluctuations could be from the formation of financial frictions in the mediation
of internal borrowing.
For Russian economy with this Bayesian techniques, one can see Malakhovskaya
and Minabutdinov (2014). They come into the conclusion that the huge influence of
fluctuations of GDP also depends mostly on internal market commodity shocks not only

3
from exporting oil products to abroad. But there are some limitations of the paper, first,
they do not separate consumption into public and private, therefore it’s not included
government purchases which are flexible when prices on oil products increase or decrease.
Another one, the model consider the stationary case and it can’t take into account different
permanent shocks in the economy.
Another example from European experience is Kliem and Uhlig (2016), that con-
sidered with DSGE model with asset prices. They applied the Bayesian estimation for
two models, first, the model consists of manner information about consumption, leisure
without hours of worked, wage rigidities and investment adjustment costs. In the paper
was used Sharpe ratio to produce technical model with asset-pricing and business-cycle
suggestions.
In the next section we try to setup small open economy with New Keynesian model
with several economic agents.

4
3 The Model

3.1 Firm
We start setting-up the model with New Keynesian small economy that is based on study
An and Schorfheide (2007). There are several economic agents as firms that produce final
goods and firms who produce intermediate goods, households which supply labor to earn
and consume final goods, the Central Bank, set the level of interest rate and Government
care about the fiscal policy. At the beginning we want to solve maximization problems
for all mentioned above agents and then describe the law of motion of the exogenous
processes, and finally summarize the equilibrium conditions.
Perfectly competitive firms can implement a perpetuity of intermediate goods:
1
Z 1  1−u
1−u
Yt = Yt (j) dj (1)
0

where j ∈ [0, 1]
Firms take prices of input Pt (j) and prices of output Pt as given and then maximize
profits:
1
Z 1  1−u Z 1 
1−u
π t = Pt Yt (j) dj − Pt (i)Yt (j)dj (2)
0 0
R1
where Pt Yt - revenue from sale of the output and 0
Pt (i)Yt (j)dj is the input costs
Demand function for intermediate good j is follows:
  −1
Pt (j) u
Yt (j) = Yt (3)
Pt

Where 1/u is the demand elasticity for each intermediate good. Then the zero-profit
conditions imply:
u
Z 1  u−1
1−u
Pt = Pt (j) u dj (4)
0

Now we can derive the expression of the Intermediate good j that is produced by a
monopolist producer in accordance with:

Yt (j) = At Lt (j) (5)

where At - total factor productivity shock that is the same for all firms, Lt (j) - labor
force of j firm. In order to set the simple model, we will not add capital as a factor of
production and workers is hired in a perfectly competitive market with the real wage
which is account as Wt .

5
 Here we can see the nominal price stickiness by adjustment costs (AC) of quadratic price
 2
φ Pt (j)
ACt (j) = − π Yt (j) (6)
2 Pt−1 (j)
where φ manage the rigidity of price and π - steady state inflation rate combined with
the final goods.
Firm j chooses its labor input Lt (j) and takes the price Pt (j) in order to maximize
the future profits’ present value in a recursive way:

∞  !
X Pt+s (j)
max = Et β s Q t+s Yt+s (j) − Wt+s Lt+s (j) − ACt+s (j) (7)
s=0
t Pt+s

where Q t+s - time value of the one consumption good in period t + s to the household,
t

which is iterated as exogenous factor by firms.

3.2 Household
Household maximizes consumption and leisure

∞
!
(Ct+s /At+s )1−τ − 1
   
X Mt+s
max = Et βs + ψM ln − ψH Ht+s (8)
s=0
1−τ Pt+s

where β - discount factor 1/τ - intertemporal elasticity of substitution, Ct - consumption

and Mt /Pt - real money balance. ψM and ψH - determine steady state of hours worked
and money balances, set ψH =1.
Then budget constraint:

Ct + Bt + Mt + Tt = Wt Ht + Rt−1 Bt−1 + Mt−1 + Dt + N Ct (9)

We can say that the households supply elastic labor forces to the producers and
receive the real wage Wt for hours of worked. Bt - bond which can be traded by household
and get annually interest rate Rt . Dt - profit from the activities of firms which is get to
household and pay tax Tt - lump-sum taxes. N Ct - net cash flow from trading a set of
securities.

3.3 Monetary and Fiscal Policies of the Government

Here we derive for monetary and fiscal policies of the government by using notes from
Schorfheide (2016). The central bank can adjust money supply in order to achieve desired
interest rate for the credits and can set the monetary policy rule:

6
 ρR R,t
Rt = Rt∗ 1−ρR Rt−1 e (10)

Here, R,t - a shock from monetary policy. Rt∗ - targeting interest rate:

 π φ1  Y φ2
t t
Rt∗ = rπ ∗
(11)
π ∗ Yt∗
where r - real interest rate steady state, π - annual inflation rate which is equal to
π = Pt /Pt−1 and π ∗ - targeted inflation rate. Finally, Yt∗ - level of output.
Fiscal authority consumes fraction of aggregate output: Gt =t Yt .
Government budget constraint:

Mt + Tt + Bt = Gt + Mt1 + Rt−1 Bt−1 (12)

Here, Gt - government spending, Rt−1 Bt−1 - level of annual payments to bond holder,
Mt1 - money that should be produced by the government, Tt - lump-sum taxes to cover
the share of costs in government purchases.

3.4 Exogenous Processes Analysis

Technology
ln At = ln γ + ln At1 + ln zt (13)

ln zt = ρ ln zt1 + z,t (14)

Here, both processes are about the total factor productivity, and ln zt - serially
correlated with error and defines as the exogenous fluctuations of growth rate of the
technology, therefore we have to consider one more process for this to get z,t which is
assumed to be not serially correlated one. Where γ - annually technology growth rate.
Government purchasing is defined by gt = 1/(1t ) - the fraction of the government
purchasing in GDP; assume

ln gt = (1 − ρg ) ln g + ρg ln gt−1 + g,t (15)

Here, z,t , g,t , R,t - are identically independently distributed (iid) and normally dis-
tributed with mean zero and standard deviations are σz , σg , σR . Monetary policy shock
R,t which is assumed to be not serially correlated.

3.5 Equilibrium Conditions

Consider the symmetric equilibrium in which all intermediate goods producing firms make
identical choices; omit j subscript. Finally, we can get market clearing conditions as:

7
 Yt = Ct + Gt + ACt , Ht = Lt . (16)

Where Yt - aggregate output which is equal to consumption, government purchasing and

adjustment costs, hours of worked and labor input are equal to each other, then the
market is clear.
Complete markets:

Qt+s/t = (Ct+s /Ct )−τ (At /At+s )1−τ (17)

Consumption Euler equation and New Keynesian Phillips curve:

 −τ !
Ct+1 /At+1 At Rt
1 = βEt (18)
Ct /At At+1 πt+1

Here, we get the Euler equation for consumption which is about the FOC via gov-
ernment bond (Bt ). In general equilibrium, the household is equal to the marginal utility
of consuming a currency today with the discounted marginal utility by investing the dollar
which get gained interest rate at Rt and by consuming it at the t + 1 period.

    −ρ
1 π Ct+1 /At+1
1 = φ(πt − π) 1− πt + − φβEt (
2v 2v Ct /At

  ρ 
Yt+1 /At+1 1 Ct
(πt+1 − π) πt+1 + 1− (19)
Yt /At v At
multline Here, this equation features the profit maximization problem of the firms that
produce the intermediate goods. From this maximization problem we get the optimal
value of the wage for hours of worked.
In the absence of nominal rigidities (φ = 0) aggregate output is given by
1
Yt∗ = (1) τ At gt (20)

where, it is the target level of production that is used in the monetary policy rule

3.6 Steady State Analysis

Set R,t = g,t = z,t = 0 at all the time. Because technology shock ln At expands in
accordance with a wiener process with drift ln γ. Let’s assume that the consumption and
production should be detrended for satisfy a steady state to exist. Then let be

ct = Ct /At , (21)

8
 yt = Yt /At , (22)

yt∗ = Yt /At (23)

Steady state is given by:

γ
π = π∗, r = , R = rπ ∗ (24)
β
1
c = (1 − v) τ , y = g, c = y ∗ (25)

Here, we can see the different situation r - real interest rate is the ratio of annually
technology growth rate to discount factor. R - aggregate interest rate is the multiplying
the real interest rate with the target inflation rate. In steady state π - the gross inflation
rate is equal to the target inflation rate. Government spending is equal to the output. The
consumption is the level of output which could be realized in case of nominal rigidities
absence. Consumption is depends on the distortion that come into the realization by the
actions of intermediate goods producers.

 
ˆ + τ cˆt + R̂t − zt+1
1 = βEt exp(−τ ct+1 ˆ − πt+1
ˆ ) (26)

φπ 2 g
exp(cˆt − yˆt ) = exp(−gˆt ) − (exp(π̂t ) − 1)2 (27)
2

R̂t = ρR Rˆt−1 + (1 + ρR )ψ1 π̂t + (1 − ρR )ψ2 (yˆt − gˆt ) + R,t (28)

gˆt = ρg gt−1
ˆ + g,t (29)

zˆt = ρz zt−1
ˆ + z,t (30)

In the equilibrium consumption law of motion, interest rates, output, inflation and
other key variables have to amuse the expectational difference equations

3.7 Log-linearization of New Keynesian Model

After Log-linearization and simple technical manipulation of above mentioned equations
in the steady state we can derive consumption with Euler equation:
1 
ŷt = Et [ŷt+1 ] − R̂t − Et [π̂t+1 ] − Et [ẑt+1 ] + ĝt − Et [ĝt+1 ] (31)
τ
By iterating Euler equation a head at t + 1 we get the real returns on holding the bonds
by forwarding the output as a function of expected future sum of the real returns. Here,

9
prices are sticky, then the central bank make an adjustments of the nominal interest rate
in order to impact on GDP growth rate.
Here, we can see the representation of New Keynesian Phillips curve:

π̂t = βEt [π̂t+1 ] + κ(ŷt − ĝt ) (32)

where
1−v
κ=τ (33)
vπ 2 φ
The New Keynesian Phillips curve links inflation to real activity and Iterated to
forward implies that inflation is the expected discounted function that is sum of the
future production gap which means yˆt - gˆt .
Monetary policy rule:

R̂t = ρR R̂t−1 + (1 + ρR )φ1 π̂t + (1 + ρR )φ2 (ŷt − ĝt ) + R,t (34)

yˆt , π̂t and R̂t combined with law of the exogenous shocks as R,t , and a linear system
of expectations hat defines gˆt , zˆt then get the evolution of

xt = [yˆt , π̂t , R̂t , R,t , gˆt , zˆt ]0 (35)

Then solve the law of motion of xt

3.8 Assessment of the Model

We estimate the model by Bayesian Estimation methods for observable variables as Per
Capita Real Output Growth (used the level of real gross domestic product), Annualized
Inflation rate (used the CPI price level) and Federal Funds Rate (used the effective federal
funds rate). The data source is from the official web-site of the Federal Reserve of St.
Louis. And we detail the structure of the durable sample selection from 1q2003 until
4q2013. Below we can see three observables’ equations. NK model is:

Y GRt = γ (Q) + 100(ŷt − ŷt−1 + ẑt ) (36)

IN F Lt = π (A) + 400π̂t (37)

IN Tt = π (A) + r(A) + 4γ (Q) + 400R̂t (38)
where

γ (Q) 1 π (A)
γ =1+ ,β = , π = 1 + (39)
100 1 + r(A) /400 400
More generically:

yt = ψ(θ) + ψ1 (θ) + ψ2 (θ)st + ut (40)

10
4 Bayesian Estimation

4.1 Data

Here, we try to estimate after linearization step of the New Keynesian model with several
observations via Bayesian approach. To set up a model we conduct with the standard NK
model with three shocks as TFP, government spending shock and monetary policy shock
and the model is conducted with Schorfeide (2016). To estimate the model structural
parameters with real-world data we apply to the US economy. The data source is from
the official web-site of the Federal Reserve of St. Louis. And we detail the structure of
the durable sample selection from 1q2003 until 4q2013

4.2 Parameterization

We set structural parameters according to the Schorfheide (2016) suggestions due to

we have used the approach of the calibration of the parameters from this literature for
estimations. We set the parameter values for prior. The values are depicted below in the
table.

Parameters Value Description

τ 2.83 Risk aversion
κ 0.78 Phillips Curve
ψ1 1.80 Monetary policy 1
ψ2 0.63 Monetary policy 2
r(A) 0.42 Annual interest rate
π (A) 0.42 Annual inflation rate
(Q)
γ 0.52 Quarterly growth rate
ρr 0.77 MP shock persistence
ρg 0.98 government shock persistence
ρz 0.88 TFP shock persistence
σr 0.22 SD of monetary shock
σg 0.71 SD of government shock
σz 0.31 SD of TFP

Table 1: Parameter calibrations and Ratios.

11
4.3 Estimation Results
In the first stage in the stating of a prior distribution, it would be better to decompose the
components of θ into three different predicaments as in the study of Schorfheide (2016).
First, named by θ(ss) , includes set of parameters that influence the steady state of the
standard NK model as θ(ss) = [r(A) ; π (A) ; γ (Q) ]. These set of parameters demonstrate the
steady state real interest rate, inflation rate, and annual GDP growth rate. The second
set of parameters demonstrates the law of motion of the shocks as mentioned above follow
θ(exo) = [ρg ; ρz ; σg ; σz ; σR ]. And the third set of parameters that varificate the endogenous
variables propagation mechanisms that is no any influence of the model’s steady state:
θ(endo) = [τ ; κ; φ1 ; φ2 ; φR ]. Below table demonstrates above mentioned set of parameters
with prior distributions
Moreover, we separated results into several outcomes, estimation results, posterior
estimation, Markov chain and Monte Carlo (MCMC) Inefficiency factors per block and
MCMC chains (follow the link below to get access to the code). In consequently, the
acceptance ratio is selected on in the interval of 0.2 and 0.4 in the literature which can
be performed by adjusting the jump scale Robert and Casella (2010). In Dynare, the
default jump scale is 0.35. As we already know that acceptance ratios in the interval
of 0 to 1 are not good for further explanation. A small acceptance ratio does not give
an information that the MCMC is not good in a low density area. We can say that the
current state of the MCMC is the posterior distribution mode and If the scale-jumps given
by the prior distribution are huge enough it is because of all the priors will be rejected.
To sum up, if the MCMC chains are huge enough, then the acceptance ratios have to be
similar across chain. In case of our estimation, the acceptance ratio reduces to 0.297282
in the first MCMC Chain and 0.297787 in the second MCMC Chain , which lie between
the range between 0.2 and 0.4 suggested by the literature. By the following the link
technical computation in Dynare (Matlab) can be found https://www.dropbox.com/sh/
ti9de5itxlkq5kl/AACOoiyJcX-qrSxXE_nXiQtKa?dl=0

12
 Prior Distribution and Posterior Distribution
Name Domain Dist. Mean Std. Mean Mode Std.
Steady State Related Parameters θ(ss)
(A) +
r R Gamma 0.500 0.5000 0.4120 0.3679 0.2420
π (A) R+ Gamma 7.000 2.000 3.6304 3.6341 0.5929
γ (Q) R Normal 0.400 0.200 0.6212 0.6284 0.1396
Endogenous Propagation Parameters θ(endo)
+
τ R Gamma 2.000 0.5000 2.2772 2.1534 0.4890
κ [0,1] Uniform 0.500 0.2887 0.8827 1.0000 0.0900
ψ1 R+ Gamma 1.500 0.250 1.8404 1.8764 0.1867
+
ψ2 R Gamma 0.500 0.250 0.5093 0.3852 0.2529
ρR [0,1] Uniform 0.500 0.2887 0.8252 0.8205 0.0249
Exogenous Shock Parameters θ(exo)
ρg [0,1) Uniform 0.500 0.2887 0.9717 0.9782 0.0210
ρz [0,1) Uniform 0.500 0.2887 0.9773 0.9835 0.0144
R 1 Invg 0.200 0.2000 0.1632 0.1590 0.0169
g 1 Invg 0.500 0.5000 0.5942 0.5718 0.0517
z 1 Invg 0.250 0.2500 0.1055 0.0975 0.0154

Table 2: Estimation results for posterior distribution.

5 Discussion

5.1 Historical Decomposition

Historical Decomposition of the endogenous variables. Here, the black line demonstrates
that at the specified value of parameter the standard deviation of the smoothed value of
the endogenous variable from its steady state. Figure 1.1 is about the historical decom-
position of output. The colors depict the fraction of the respective smoothed shocks in
the fluctuation of output from its steady state. As we see the share of government shock
and technology shock is more attractive than monetary policy shock. In a nutshell, our
best prediction is which shocks lead to our best prediction for output. And, the initial
values are also the part of the fluctuation from steady state that are not explained by
the respective smoothed shocks, but by the unknown value of the state variable. Usually,
these types of the initial values eliminated quickly.
In figure 1.2 we see the respective smoothed shocks in the deviations of the interest
rate. Here, monetary shock and technology shock correspond the contribution is more

13
than other shock in the deviation of the smoothed interest rate.
Finally, in figure 1.3 the monetary shock calls more deviation for smoothed inflation
and technology shock as well. It means that inflation is well matched with monetary and
technology shocks.
Other papers on US economy we can see Schorfheide (2016) proves that the gov-
ernment spending and TFP shocks play key role in the deviation of the output. And,
in the paper of Bernanke et al. (1999) states that in period of Great recession decrease
of the GDP is caused by the oil export price shocks. Moreover, from emerging markets
we can see the study of Russia in Andreev at al. (2020) paper with different types of
shocks especially oil price shock for endogenous and exogenous financial frictions models.
In this study we can see that GDP for both endogenous and exogenous financial frictions
is matched well with TFP. And, from Malakhovskaya and Minabutdinov (2014) we can
conclude for Russian case the GDP decreased in 2009 due to the pool decrease of oil
export shock and not effective monetary policy of the central bank of Russia.
1.5

1
eps_z

0.5

0 eps_g

-0.5

eps_r

-1

-1.5

Initial values

-2
0 10 20 30 40 50 60 70 80

Figure 1.1.Output
6 4

4 2
eps_z eps_z

2 0

eps_g eps_g

0 -2

-2 -4
eps_r eps_r

-4 -6

Initial values Initial values

-6 -8
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80

Figure 1.2.Interest rate Figure 1.3.Inflation

5.2 Error Variance Decomposition

Below we can see in horizontal the three different shocks and in vertical our endogenous
and observable variables. Table 2 depicts the deviation in percentage of each variable
that is featured by different exogenous shocks. There are totally eight observable and

14
non-observable variable that are used in practice to estimate the model and we can see the
output variates with government shock well matched. TFP shock is contributed also huge
that is equal to 75.97 percent. Total factor productivity shock contribute 100 percent into
the aggregate interest rate. Government purchasing is explained quite well by government
shock which is 100 percent. We can come into the conclusion that observable variables
are explained well by these three different shocks. But unfortunately, for all three main
endogenous variables we have the same value without changing in model variables and
observable variables except GDP, but the value of GDP also looks not acceptable.
Concerning other studies in case of Russia we see that study of Andreev et al. (2020)
states that in endogenous and exogenous financial frictions for GDP deviations are caused
by the oil price shock and mostly by TFP shock. But in our study we see that the share of
GDP in model and observables are dramatically increase it shows 1.56 percent in model
and 55.30 percent in observables. But the GDP is featured very good with government
spending shock in model

Variables epsz epsg epsr Tot. lin.

contr.
y 1.56 98.28 0.60 100.45
infl 75.97 0.00 13.63 89.60
R 100.91 0.00 4.33 105.23
g 0.00 101.01 0.00 101.01
z 101.01 0.00 0.00 101.01
Robs 100.91 0.00 4.33 105.23
inflobs 75.97 0.00 13.63 89.60
yobs 55.30 52.09 2.93 110.32

Table 3: Error variance decomposition simulating at a time (in percent)

5.3 Impulse Response Functions

Below the figures depict the Bayesian IRFs for prior and posterior distributions. The
grey color demonstrates the prior distribution density function, while the black color
line demonstrates the density function of the posterior distribution. The Green line in
the center depict the mode of the posterior distribution. We say that in case when
the posterior’s shape is the same as the prior, then there is could be two reasons, first,
it happens usually, the parameter is weakly determined and data set not give enough
information to update the prior or, second, prior distribution has good demonstration of

15
the information in the data set (Canova, 2007). In our case there are some features of the
parameters and standard errors of the exogenous shocks. For instance, τ is distributed
by the same value for priors and posteriors, it is because of the shortage of information
etc. Moreover, Bayesian estimation associates the likelihood functions with the prior
distributions in order to create the posterior distributions
SE_eps_r SE_eps_g SE_eps_z
30
8
20
6 20
4
10 10
2
0 0 0
0.2 0.4 0.6 0.8 0.5 1 1.5 2 0.2 0.4 0.6 0.8 1

ra pia gammaq
3
2
0.6
2
0.4
1
1
0.2

0 0 0
0 1 2 3 0 5 10 0 0.5 1 1.5
tau kappa psi1
0.8 6 2
0.6
4
0.4 1
2
0.2
0 0 0
0 2 4 6 0 0.5 1 1 2 3

Figure 2. Priors and Posteriors distributions

psi2 rhor rhog

15 20
1.5
10
1
10
0.5 5

0 0 0
0 1 2 3 0 0.5 1 0 0.5 1

rhoz
30

20

10

0
0 0.5 1

Figure 3.Priors and Posteriors distributions

Here, the Bayesian IFRs to monetary, TFP and to government shocks are depicted
in the figures below. The gray color part shows the highest posterior density intervals. In
figure 6 It means that is the government spending shock is negatively influence to the GDP
then we can see this value is decreasing until negative values. On the other hand, when
the GDP increases if the government spending shock influence positively, respectively. We
can see others shocks direction for example for monetary shock that all three variables
are move in one direction in figure 4, but for the total factor production shock in figure 5

16
we see for GDP it moves in one direction but for other variables these values change for
negative and positive sides.
y infl R
0 0
0.08
-0.05
-0.1 0.06
-0.1 0.04
-0.2 0.02
-0.15
0
10 20 30 40 10 20 30 40 10 20 30 40

Figure 4.Bayesian IRF to monetary shock

y infl R
0.2
0.15 0.3
0.15
0.1 0.2
0.1
0.05 0.1
0.05

10 20 30 40 10 20 30 40 10 20 30 40

Figure 5.Bayesian IRF to TFP shock

y
0.6

0.4

0.2

10 20 30 40
Figure 6. Bayesian IRF to government shock

Below we see the Monte Carlo Markov Chain (MCMC) univariate diagnostics
(Brooks and Gelman, 1998)
In the first column with the Interval figures we san see the Brooks and Gelman (1998)
convergence diagnostics for the interval with 80 percent. The blue color demonstrates
the 80 percent quantile range which is based on the joined draws from other orders,
however the red color demonstrates the mean quantile range which is based on the draws
of the private orders. In the second and in the third column an estimation of the similar
data for the second and third central moments are demonstrated by which are named
m2 and m3. In other words, the squared and cubed absolute or relative fluctuations
depend from the joined and the within-sample mean. If the sequences have coincided
then the two lines have harden horizontally and have to be close to each other. Other
demonstrated figures are based on the parameter draws increasing. In figure 8 we see

17
MCMC for exogenous processes where it generates interval, m2 and m3 columns. The
model and data are matched very well, due to we see the 105 MCMC chains have to good
multivariate convergence. In figure 9-10 we results for the other structural parameters

SE_eps_r (Interval) 10 -4SE_eps_r (m2) 10 -6SE_eps_r (m3)

4 12
0.05
3.5
10
0.045
3
8
0.04 2.5
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
10 5 10 5 10 5
SE_eps_g (Interval) SE_eps_g (m2)
10 -3 SE_eps_g (m3)
10 -4
0.13 2.6 2.5

0.125 2.4 2
2.2
0.12 1.5
2
0.115 1.8 1
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
5 5 5
10 10 10
SE_eps_z (Interval) 10 SE_eps_z
-4 (m2) 10 SE_eps_z
-6 (m3)
0.045 3 10

8
0.04 2.5

6
0.035 2
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
10 5 10 5 10 5

Figure 7. Monte Carlo Markov Chain (MCMC) univariate diagnostics (Brooks and Gelman, 1998)

ra (Interval) ra (m2) ra (m3)

0.7 0.07 0.03

0.06
0.6 0.02
0.05

0.5 0.04 0.01

2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
10 5 10 5 10 5
pia (Interval) pia (m2) pia (m3)
1.8 0.4 0.4

1.6 0.35 0.35

0.3
1.4 0.3
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
10 5 10 5 10 5
gammaq (Interval) gammaq (m2) 10 -3gammaq (m3)
0.45 0.024 6

0.4 0.022 5

0.35 0.02 4
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
10 5 10 5 10 5

Figure 8. MCMC univariate diagnostics (Brooks and Gelman, 1998)

18
 tau (Interval) tau (m2) tau (m3)
1.4 0.3 0.3

1.2 0.2
0.2

1 0.1
0.1
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
10 5 10 5 10 5
kappa (Interval) kappa (m2) 10 -3 kappa (m3)
0.012
0.28
2
0.011
0.26 1.8
0.01 1.6
0.24 1.4
0.009
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
10 5 10 5 10 5
psi1 (Interval) psi1 (m2) 10 -3 psi1 (m3)
0.5 0.04
12
0.035
0.45 10
0.03
8
0.4 0.025
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
5 5 5
10 10 10

Figure 9. MCMC univariate diagnostics (Brooks and Gelman, 1998)

Now, in this part we see Multivariate convergence diagnostic in figure 10.

Interval
6.5

5.5
1 2 3 4 5 6 7 8 9 10
10 5
m2
7

4
1 2 3 4 5 6 7 8 9 10
10 5
m3
30

20

10
1 2 3 4 5 6 7 8 9 10
10 5

Figure 10. Multivariate convergence diagnostic

19
 psi2 (Interval) psi2 (m2) psi2 (m3)
0.7 0.08 0.04

0.65 0.07
0.03
0.6 0.06

0.55 0.05 0.02

2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
10 5 10 5 10 5
rhor (Interval) 10 -4 rhor (m2) 10 -5 rhor (m3)
0.07 8
3
0.065
6
2
0.06

0.055 4 1
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
5 5 5
10 10 10
rhog (Interval) 10 -4 rhog (m2) 10 -5 rhog (m3)
0.06 5
1.5
4
0.05
1
3
0.04 0.5
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
5 5 5
10 10 10

Figure 11. MCMC univariate diagnostics (Brooks and Gelman, 1998)

rhoz (Interval)
0.04

0.035

0.03
1 2 3 4 5 6 7 8 9 10
5
10
-4 rhoz (m2)
10
2.5

1.5
1 2 3 4 5 6 7 8 9 10
10 5
-6 rhoz (m3)
10
8

4
1 2 3 4 5 6 7 8 9 10
10 5

Figure 12. Multivariate convergence diagnostic

5.4 Bayesian Time Series

Now, the black color line shows the mean estimation of the one step ahead forecast of the
endogenous variables as GDP, inflation, aggregate interest rate. From the Kalman filter

20
we can derive for the best guess of these endogenous variables in the period one ahead
concerning on given data up to the current situations. The green color lines demonstrate
the tenth of the one step ahead prediction distribution.
eps_z eps_g eps_r
1 0.5
0.2

0 0
0
-0.2 -1

-0.4 -2 -0.5
20 40 60 80 20 40 60 80 20 40 60 80

Figure 13. Smoothed Shocks to monetary shock

y infl R
5 1 2

0
0 0
-1

-5
-2 -2
20 40 60 80 20 40 60 80 20 40 60 80

Figure 14.Smoothed Variables to Interest rate

y infl R
5 1 2

0
0 0
-1

-5 -2 -2
20 40 60 80 20 40 60 80 20 40 60 80

Figure 15.Updated Variables to Interest rate

21
6 Conclusion
The structural parameters of DSGE models can be estimated via various kind of methods.
One of them we have considered in this paper. From our measurement we can conclude
that respective shocks play key role in deviation of endogenous variables especially ob-
servable variables. For the US economy we come into the conclusion that the government
shock leads the production to high deviation, because government can purchase or ex-
port final goods abroad then we get positive effect this shock, if increase the government
spending and collect taxes from population then we can wait the negative effect from this.
In real life, the GDP of the US economy is deviated mostly due to the TFP shock.
According to the International Monetary Fund for 2012 US GDP was accounted approx-
imately 22.1 % from industrial production and 76.8 % from services, respectively. But
from the first quarter of 2020 the GDP goes down because of exogenous shocks as pan-
demic shock and oil price shock that slow down, but now the government is supporting
to boost the economic growth. On the other hand, the government spending also effects
to the GDP when there are some adjustments in fiscal policy.
Finally, the Central Banks and other economic agents have to take into account
that if the exogenous shocks call the deviation of GDP then we can conclude that in our
situation it could be when government rapidly increases or decreases the spending in public
sector. Aggregate interest rate and inflation are adjusted by total factor productivity
and monetary shock when the Central Bank change the targeting interest rate. The
high inflation also calls to increase the interest rate to stablize the price of primary and
secondary goods in the economy.

22
7 References

References
[1] Almeida, Vanda, Bayesian Estimation of a DSGE Model for the Portuguese Econ-
omy. Bank of Portugal Working Papers Series No. 14/2009, 2009 Available at SSRN:
https://ssrn.com/abstract=1535285 or http://dx.doi.org/10.2139/ssrn.1535285

[2] Andrieu, C., A. Doucet, and R. Holenstein Particle Markov Chain Monte Carlo Meth-
ods. Journal of the Royal Statistical Society Series B, 72(3), 269–342, 2010

[3] Andreasen, M. M. NonLinear DSGE Models and the Central Difference Kalman Filter.
Journal of Applied Econometrics, 28(6), 929–955, 2013

[4] Bernanke, Ben, Mark Gertler and Simon Gilchrist, The financial accelerator in a
quantitative business cycle framework, Handbook of Macroeconomics, Elsevier (1999)

[5] Edward Herbst and Frank Schorfheide Bayesian Estimation of DSGE Models. Read-
ing, Princeton University Press, 2016

[6] Goodhart, C.A.E., M.U. Peiris and D.P. Tsomocos Debt, recovery rates and the greek
dilemma. Journal of Financial Stability 36, 265 – 278., 2018

[7] Kliem, Martin and Uhlig, Harald, Bayesian Estimation of a DSGE Model with As-
set Prices. Bundesbank Discussion Paper No. 37/2013, 2013 Available at SSRN:
https://ssrn.com/abstract=2796932

[8] Malin Adolfson and Stefan Laséen and Jesper Lindé and Mattias Villani Bayesian
estimation of an open economy DSGE model with incomplete pass-through. Journal of
International Economics, ISSN: 0022-1996, 2007

[9] Peiris, M. Udara and Dimitrios P. Tsomocos International monetary equilibrium with
default. Journal of Mathematical Economics 56, 47 – 57., 2015

[10] Christiano, Lawrence J., Martin Eichenbaum, and Charles L. Evans. Nominal Rigidi-
ties and the Dynamic Effects of a Shock to Monetary Policy Journal of Political Econ-
omy 113, no. 1 (2005): 1-45.

[11] Kliem, M. and H. Uhlig Bayesian estimation of a dynamic stochastic general equlib-
rium model with asset prices, Quantitative Economics, Vol. 7(1), 257-287, (2016)

23