use of dynare software for dsge modelling

ModelData

Bayesian estimationConclusion

Use of Dynare Software for DSGE modelling

Katerina Gawthorpe

Modern Tools for Financial Analysis and Modeling 2018

5. June 2018

Katerina Gawthorpe Use of Dynare Software for DSGE modelling

ModelData


Outline

1 Model

2 Data

3 Bayesian estimation

4 Conclusion


ModelData


Model

Extended version of Aliyev & Bobkova (2014)Characteristics of both models:

External habit formation in the utility functionWage and price rigidityCapital adjustment costs

Main differencesMore detail household heterogeneity

Three HHs groupsIncome effect in the consumption equation

Without non-ricardian HHs specificationImportant for government shock or foreign demand shock

Intermediate-inputsIntroduced into PF, import and exportIntermediate-inputs complement other inputs


ModelData


Data transformation

Data on model-entry side:To define parameters for calibrated modelTo define priors for estimated modelTo define size of shocks or their prior value for conditionalforecasts or as input for shock decomposition.

Output values from model:For model evaluation, compared to dataSometimes subject to transformation to obtain absolute orrelative valuesForecast

The link between data and model might be weaker forcalibrated rather than estimated models as no observationequations are necessary for model simulations.


ModelData


Data transformation

Seasonally adjusted data with X-13 ARIMA are subject totransformation into growth rates and then demeaned toobtain data compatible with observed variables. The finalvalues are multiplied by 100 for easier interpretation inpercentages. (See Pfeifer, 2017)The transformation of data appears as

X mobst =

(log

Xt

Xt−1− X obs

t)∗ 100


ModelData


Bayesian estimation

Bayesian estimation combines prior information about aparameter value with the evolution of data.Mostly used algorithm is the MH-MCMC, this algorithmdecides if a new parameter value given the likelihooddefined by Kalman filter should be accepted.For our model we use two MCMC sequences with 100 000draws. The acceptance ratio is below 0.24.

p(Ω/yt ) =p(yT/Ω)p(Ω)

p(yT )


ModelData


Prior & Posterior probability

0.1 0.2 0.3 0.40

5

10

Share of Gov. cons. on output

0 0.2 0.40

5

Share of Gov. cons. on GR

0 0.2 0.4 0.60

5

AR par. in Gov. cons. shock

1 2 3 40

0.5

1

Government cons. shock2 4 6 8

0

0.5

1Income tax shock

0.2 0.4 0.6 0.80

5

Impact of income tax

0.5 1 1.5 20

2

4TR parameter for inflation

0 0.5 10

5

TR parameter for output


ModelData


Sensitivity analysis

Output variation with Gov. cons. parameter Output variation with income tax parameter

00.6

0.2

0.4

Out

put

0.4 10

0.6

omega yg Time

0.8

0.2 50 0

-0.040.5

-0.03

0.45 25

Out

put -0.02

20

Income tax parameter

0.4 15

Time

-0.01

100.35 50.3 0

0.2 0.80

5

Impact of income tax

X: 0.3333Y: 6.26

Source: Authors’ creation, data from the Ministry of Finance.


ModelData


Conditional forecast

Impact of government consumption shock on output

Output vs government consumption variable in data



ModelData


Conditional forecast

Impact of income tax shock

Dynamic of income tax in data



ModelData


Shock decomposition

2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

-10

-5

0

5

10

15

Government Balance

Monetary policyOther gov. expenditureEU investmentVATIncome taxWagesInvestmentTechnology shockForeign demandLabor demandExchange rateOther



ModelData


Dynare - shortcomings

Difficulty to write a Matlab code into a Dynare scriptPossible to write it at the bottom of the script to create a plotOr to write it in the beginning of the script to call onsteady-state values or parameters

Entering exogenous values for a variableCombining deterministic and stochastic shocksIdentification of shock contributions to variables from ashock-decomposition output and subsequenttransformation of quarterly data to annualBlack-box search for the steady-state of non-linear modelsas well as their subsequent estimation and interpretationgEcon package as an alternative?


ModelData


Thank you for your attention...

Katerina GawthorpeMinisterstvo financí

Letenská 15118 10, Praha 1

email: [email protected]


mailto:[email protected]

use of dynare software for dsge modelling

Documents