Transcript Fiscal multipliers

FISCAL MULTIPLIERS
Arbresh MAMUDI, State University of Tetovo,
Geoff PUGH, Staffordshire University Business School
Theoretical approaches
Far from consensus
• The range of reported multipliers varies from negative to higher than one
• Empirical evidence on the size of the multiplier cannot distinguish
between the competing theories
•
Range
0<k<1
k >1
k <0
RBC
• fully competitive labor and goods
markets
• ‘Ricardian consumers’
•
complementarity of public and private
consumption
• distortional effects of taxation
• ‘wage pressure’ effect of public
employment
NK-DSGE
Inflation targeting monetary authority
• Monopolistic competition
• Sticky prices and wages
Post crisis studies:
• ‘non- Ricardian consumers’
• zero lower bound interest rate
Post crisis studies:
• risk premium on interest rates for high
government debt
The dataset
• 65 empirical studies, 914 observations estimated by
– single equation approaches (SEE) or
– Vector autoregression (VAR) models;
• Primary data for structural characteristics of the countries:
o The indebtedness of the economy, (central government debt/ GDP)
o Monetary policy reaction, (short term money market rates)
o
The degree of openness, (imports of goods and services/GDP)
o Financial development, (domestic credit to private sector/ GDP)
• NOVELTY- augment MRA with primary data on labour market variables
• Why? Labour market characteristics important in both leading theories:
o
o
o
o
Employment protection legislation indicator, EPL-(index scaled 0-5)
Trade union density, TUD-(ratio)
Benefit replacement rates, BRR-(ratio)
Coordination of wage bargaining, COOR-(index scaled 1 to 5)
Moderator variables: coding the
literature
•
•
•
•
•
•
•
•
•
•
•
Type of model class
Type of fiscal impulse
Direction of the impulse
The way fiscal shocks are financed
The duration of the shock
Type of the country
Type of data
Horizon of estimation
Type of fiscal multiplier
Controlling for country specific characteristics
Controlling for the quality of the study
MRA methodology
ki= k0 +
–
–
–
–
–
–
𝑚
𝑚=1 𝛼 mZim
+ β1(
1
)+
N i
ei
ki is the multiplier value of observation i;
k0 is the “underlying” or “reference” multiplier value to be estimated;
Zim are m characteristics (“moderator variables”) of observation i;
αm are m parameters to be estimated ( effects of Zim on ki);
ej is the meta-regression disturbance term;
1
(
) is a proxy for publication bias (N =sample size for observation i)
N
• Standardization is not necessary; multiplier is dimensionless
• Multiple estimates per study used;
– each estimate is weighted by the inverse of number of estimates in a
given study
• standard errors adjusted for data clustering,
– using each study in our dataset as a distinct cluster
20
10
0
sqrsamsiz
30
40
Publication bias: Funnel plot & FAT-PET
-5
0
5
k
10
• ‘Funnel plot’ –ambiguous:
• slightly skewed to the right, but weight to the left
• around a mean that is positive,
• Cf. ‘FAT’:
Ho:β1=0 ; no systematic variation of effect size with sample size
• Ho rejected, β1=-2.37, (p-value=0.01) , negative coefficient indicates
positive relationship between size of multiplier and sample size
• No ‘classical’ reason for publication search
• Competing theories with different predicted multipliers
15
Multivariate MRA: 2 models
• Baseline model:
– All MVs
– 2 dummies for Japan studies
• All Japan studies
• Japan studies after 1990
• Preferred model:
– Exclude DV controlling for the financial crisis
•
A few observations controlling for financial crisis during the
sample period
– Cures substantial multicollinearity effects
•
e.g. with Japan dummies
– Cures diagnostic failure with respect to linearity
Table 4: Total sample results- different specification (WLS and cluster-robust SEs)
Variables
Description
k0
Constant
Model class
var
VAR models
Fiscal impulse
see
cons
SEE models
Public consumption
invest
Public Investments
milita
Military
expenditure
Tax shocks
Group
tax
pubemp
notspec
•
•
Public employment
expenditure
General government
expenditure
Baseline model
(1)a
1.198**
(0.434)
0.887***
(0.232)
Preferred model
(2)b
1.010**
(0.395)
0.873**
(0.233)
0.609***
(0.145)
0.562***
(0.200)
-0.276
(0.405)
-0.474***
(0.141)
-0.017
(0.421)
0.639***
(0.145)
0.584***
(0.197)
-0.280
(0.406)
-0.487***
(0.142)
-0.008
(0.417)
Diagnostic test:
• Preferred model is well specified with respect to linear functional form ( 1% level) and normality
• But may suffer from heteroscedasticity; so model is estimated with cluster robust standard errors
Multivariate results :
• Multipliers from the VAR model are significantly higher than estimates from SEE
• Public investment and public consumption produce higher multiplier values
• Tax shocks have lower impact compared to unspecific/general public spending
Table 4: Total sample results- different specification (WLS and cluster-robust SEs)
Group
Variables
Description
Baseline model
(1)a
-0.247
(0.194)
-0.925***
(0.331)
Preferred model
(2)b
-0.266
(0.194)
-0.902***
(0.330)
Direction of
impulse
Duration of
impulse
incr
Positive fiscal shock
tempor
Temporary shock
Type of economy
transit
Transition countries
0.661**
(0.268)
0.673**
(0.263)
Data
characteristics
quart
Quarterly data
horiz
Horizon after shock
-0.575**
(0.239)
0.018*
(0.010)
-0.535**
(0.241)
0.018*
(0.009)
• A longer horizon of measurement yields significantly higher multipliers
• Studies using quarterly data report significantly lower multipliers compared to
studies using annual data
• Multipliers from a temporary shock are lower than multipliers from a permanent
shock
• Fiscal policy in transition countries appears to be more effective than in advanced
economies although the results are not stable across different specifications
Table 4: Total sample results- different specification (WLS and cluster-robust SEs)
Group
Variables
indebt
open
er
MV for
augmented
models
employ
lmi
recc
exp
fincrisis
Description
Baseline model
(1)a
Indebtedness of the
-0.159
country
(0.102)
Openness
-0.470***
(0.162)
Exchange rate
0.251*
(0.153)
Employment
0.063
(0.116)
Labour market
0.595***
institutions
(0.125)
Multipliers estimated
-0.198
assuming reccesion
(0.332)
Multipliers estimated
-0.406**
assuming expansion
(0.209)
Multipliers estimated
0.362
assuming financial crisis
(0.331)
Preferred model
(2)b
-0.171
(0.104)
-0.499***
(0.163)
0.279*
(0.133)
0.049
(0.116)
0.579***
(0.126)
-0.273
(0.293)
-0.484**
(0.212)
• Primary studies controlling for the degree of openness, type of exchange rate and labour
market characteristics, yield significantly different estimates compared to conventional
studies
• The multipliers estimated for expansion periods are smaller
Table 4: Total sample results- different specification (WLS and cluster-robust SEs)
Group
Primary data
Variables
Description
epl
tud
Employment protection
rate
Trade union density
brr
Benefit replacement rate
coor
Wage coordination
ir
Interest rate
impgdp
Import/GDP
credgdp
Domestic credit/GDP
debtgdp
Debt/GDP
Baseline model
(1)a
0.029
(0.067)
-0.001
(0.004)
0.566
(0.481)
0.058
(0.072)
-0.002
(0.024)
-0.013**
(0.005)
0.000
(0.002)
0.002
(0.002)
Preferred model
(2)b
0.043
(0.066)
-0.000
(0.004)
0.788*
(0.440)
0.059
(0.077)
-0.002
(0.023)
-0.013**
(0.005)
0.001
(0.002)
0.003
(0.002)
• Primary (contextual) variables:
• All models:
• Openness channel is an important determinant of the multiplier
• A difference between economies of 30pp implies a difference in multiplier of 0.39
• Preferred model:
• The replacement ratio also affects the value of the multiplier
• A difference between economies of 0.10 implies a difference in multiplier of 0.08
‘True’ multiplier
Study characteristics (other factors held constant)
Study estimated by SEE
(incr=0; tempor=0; transit=0; quart=0; horiz=mean; primary data=mean)
Study estimated by VAR
(incr=0; tempor=0; transit=0; quart=0; horiz=mean; primary data=mean)
Fiscal impulse is CONSUMPTION
(incr=1; tempor=0; transit=0; quart=0; horiz=mean; primary data=mean)
Fiscal impulse is INVESTMENT
(incr=0; tempor=0; transit=0; quart=0; horiz=mean; primary data=mean)
Fiscal impulse is MILITARY SPENDING
(incr=0; tempor=0; transit=0; quart=0; horiz=mean; primary data=mean)
Fiscal impulse is TAX SHOCK
(incr=0; tempor=0; transit=0; quart=0; horiz=mean; primary data=mean)
Fiscal impulse is PUBLIC EMPLOYMENT
(incr=0; tempor=0; transit=0; quart=0; horiz=mean; primary data=mean)
Study estimated by SEE, fiscal impulse is NOTSPE,
(incr=1; tempor=1; exp=1; quart=1; japan=1)
Study estimated by SEE, fiscal impulse is TAX,
(incr=1; tempor=1; exp=1; quart=1; japan=1)
Study estimated by SEE, fiscal impulse is TAX,
(incr=1; tempor=1; transit=0; quart=1; horiz=mean; primary data=mean)
Study estimated by VAR, fiscal impulse is TAX,
(incr=1; tempor=1; transit=0; quart=1; horiz=mean; primary data=mean)
Study estimated by VAR, fiscal impulse is TAX,
(incr=0; tempor=1; exp=1; quart=1; horiz=mean; primary data=mean)
Combined
effect
t-stat
p-value
CI
1.549
3.67
0.001
0.70;2.39
2.423
4.75
0.000
1.4;3.44
3.041
6.09
0.000
2.03;4.04
2.995
5.83
0.000
1.96;4.02
2.294
3.81
0.000
1.08;3.5
1.934
4.17
0.000
1;2.86
2.508
3.98
0.000
1.24;3.76
-1.504
-2.64
0.011
-2.64;-0.36
-1.957
-3.38
0.001
-3.11;-0.79
-0.627
-2.41
0.02
-1.15;-0.10
0.25
2.01
0.049
0.001;0.49
-0.003
-0.02
0.459
-0.34;0.76
Main findings
• The heterogeneity of the reported multipliers
arises from many study characteristics
• Structural characteristics:
– Openness channel - very large
– Replacement ratio - smaller but still substantial
• There is no true multiplier
– the multiplier is time and state dependent
Thank you!