Open Seminar, Eesti Pank 21.05.2015

1. Unemployment Dynamics in the US and Euro
Area: Do Uncertainty Shocks Matter?
Aleksei Netˇsunajev1, Katharina Glass2
1Free University of Berlin, 2University of Hamburg
May 21, 2015
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 1/ 19

2. Motivation
Are there any effects of (economic policy) uncertainty on real
economic activity?
Mainly SVAR literature: Bloom (2004), Baker et al (2013),
Caggiano et al (2014a, b), Benati (2014), etc...;
Single-country studies: uncertainty, independent of the used
methods, is countercyclical and exhibits temporary negative
influence on output and employment;
Few papers are focused on the uncertainty spillovers so far:
Colombo (2013), IMF (2013);
Our interest: local and foreign effects of uncertainty shocks
for the US and EA.
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 2/ 19

3. Motivation
We apply a more sophisticated approach: Bayesian
Markov-switching SVAR and identify shocks via
heteroscedasticity. We label the economic policy uncertainty
shocks based on the approach of Uhlig (2004).
As the measure of uncertainty we use Baker et al (2013)
news-based economic policy uncertainty index (EPU)
The real side of the economic regions is represented by a
measure of unemployment. Caggiano et al (2014 b) explicitly
analyze interaction of uncertainty and unemployment in the
single country case.
We reassure some negative effects of economic policy
uncertainty on unemployment. Local effects of uncertainty
shocks for both US and EA. Some foreign effects: US
uncertainty shocks influence EA unemployment.
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 3/ 19

4. SVAR
Let time evolution of an n × 1 vector yt of endogenous variables be
given by the following SVAR model:
yt = k0 + A1yt−1 + . . . + Apyt−p + B t(st) , (1)
where k0 is an intercept term, B is the instantaneous impact
matrix, A1, . . . , Ap are autoregressive matrices, and t(st) is the
vector of uncorrelated structural innovations that depends on the
hidden state parameter st ∈ {1, . . . , m}. We assume the following
distribution of t(st):
t(st) ∼ Normal( 0, Λ(st) ) ,
where {Λ(s) : s = 1, ..., m} is a family of distinct n × n diagonal
matrices and Λ(1) is normalized to identity matrix.
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 4/ 19

5. SVAR
The model can be written in the reduced form VAR with
time–varying volatility of innovations:
yt = k0 + A1yt−1 + . . . + Apyt−p + ut(st) , (2)
where ut(st) = B t(st) is a vector of reduced form residuals
satisfying:
ut(st) | st ∼ N( 0, Σ(st) ) ,
where Σ(s) = BΛ(s)B is the reduced–form variance–covariance
matrix in each volatility state s ∈ {1, . . . , m}.
The main issue is to recover the matrix B from the family of
variance–covariance matrices {Σ(s)}.
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 5/ 19

6. SVAR
Assumptions: (i) there are two regimes of volatility, i.e m = 2 and
(ii) the matrix B stays the same across the states.
Under these assumptions the following decomposition allows to
back out the structural parameters of the model:
Σ(1) = BB , Σ(2) = B Λ(2)B .
Two regimes MS models allows for a rich patterns in volatility
while regime independent B is standard for structural VAR models.
In the related literature the models of Caggiano et al (2014a, b)
can be interpreted as models with two (extreme) regimes
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 6/ 19

7. SVAR
Orthogonal shocks may be lacking economic interpretation!
Uncertainty shock minimum requirement: a pronounced reaction of
uncertainty variable at least on impact
To find a suitable label for a shock we examine impact effects and
forecast error variance decompositions:
Verify whether there are some shocks that can qualify as
uncertainty shocks i.e. satisfy the minimum requirement.
Do the qualified shocks explain maximum of the forecast error
variance of the uncertainty variable? Yes - labeling is
confirmed! (Uhlig (2004)).
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 7/ 19

8. Inference
We apply Bayesian methods for inference on all relevant model
parameters. We set up Gibbs sampler for reduced–form VAR
parameters with the following steps:
1. Draw unobserved vector ST using algorithm of Chib (1996);
2. Draw elements of Markov transition matrix P;
3. Draw covariance matrices Σ(s), s ∈ {1, 2};
4. Decompose Σ(s), s ∈ {1, 2} into B and Λ(2)
5. Draw VAR parameters
See details on Gibbs sampler in Kulikov, Netˇsunajev (2013).
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 8/ 19

9. Data
We use five dimensional VAR with
yt = [UEA
t , UUS
t , EPUEA
t , EPUUS
t , GEAt] where
UEA
t is the log of number of unemployed people in the EA
(month-on month growth rate)
UUS
t is the log of number of unemployed people in the US
(month-on month growth rate)
EPUEA
t is the log of measure of the economic policy
uncertainty in the EA (demeaned)
EPUUS
t is the log of measure of the economic policy
uncertainty in the US (demeaned)
GEAt is the index of global real economic activity.
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 9/ 19

10. Data
The first two variables are meant to capture the effects of interest
on the local labor market. Number of unemployed people in the
EA is published by the ECB, number of unemployed people in the
US comes from the Bureau of Labor Statistics.
The economic policy uncertainty measures are region specific and
come from http://www.policyuncertainty.com. The
European news-based EPU index encompasses Germany, Spain,
France, Italy. Since these countries account for about 80% to the
total Euro area GDP their uncertainty contribution is a suitable
proxy for the whole Euro area uncertainty.
The last variable in the VAR is the index of global real economic
activity (GEA), which controls for possible global effects. It is
introduced by Kilian (2009) to reflect shifts in the demand for
industrial commodities on global market.
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 10/ 19

11. Data
Data: yt = [UEA
t , UUS
t , EPUEA
t , EPUUS
t , GEAt] . Monthly data for
1997M2 to 2014:M2
Figure: Data
1997 2002 2007 2012
−3
−2
−1
0
1
2
3
4
5
6
Ut
EA
1997 2002 2007 2012
−8
−6
−4
−2
0
2
4
6
8
10
Ut
US
1997 2002 2007 2012
−1
−0.5
0
0.5
1
1.5
EPUt
EA
1997 2002 2007 2012
−1
−0.5
0
0.5
1
1.5
EPUt
US
1997 2002 2007 2012
−60
−40
−20
0
20
40
60
GEAt
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 11/ 19

12. Identifying uncertainty shocks
Most of the studies use zero restrictions/recursive ordering to
identify uncertainty shocks. Real side of the economy reacts to
uncertainty shocks within a lag of at least one month.
Benati (2014): impact and sign restrictions, but non-negligible role
of uncertainty shocks is documented for identification based Uhlig
(2004) approach.
Caggiano (2014b): use quarterly data and allow for possible
impact effects of uncertainty shocks (order the uncertainty
measure first in the VAR).
We want to jointly model several economic regions hence
conventional restriction are more problematic. Identification via
changes in volatility is useful because it does not impose any
distortions on impact effects of shocks and allows to analyze both
US and EA uncertainty shocks identified under the same
conditions.
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 12/ 19

13. Posterior regime probabilities
Figure: Posterior regime probabilities
1998: 02 2000:02 2002:02 2004:02 2006:02 2008:02 2010:02 2012:02 2014:02
0
0.5
1
State 1
1998:02 2000:02 2002:02 2004:02 2006:02 2008:02 2010:02 2012:02 2014:02
0
0.5
1
State 2
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 13/ 19

14. Posterior of B
Figure: Prior-Posterior of B
−5 0 5
0
0.5
1
Ut
EA
−2 0 2
0
1
2
−1 0 1
0
2
4
US uncertainty shock
−1 0 1
0
5
10
EA uncertainty shock
−1 0 1
0
5
10
−4 −2 0 2 4
0
0.2
0.4
Ut
US
−2 0 2 4
0
0.5
1
−2 0 2
0
0.5
1
−2 0 2
0
1
2
−2 0 2
0
1
2
−0.5 0 0.5
0
5
10
EPUt
EA
−1 0 1
0
5
10
−0.5 0 0.5 1
0
5
−0.5 0 0.5
0
5
10
−0.5 0 0.5
0
5
10
−0.5 0 0.5
0
5
10
EPUt
US
−1 0 1
0
5
10
−0.5 0 0.5
0
5
10
−0.5 0 0.5
0
10
20
−0.1 0 0.1
0
20
40
−10 0 10 20 30
0
0.1
0.2
GEAt
−20 0 20
0
0.2
0.4
−10 0 10
0
0.5
−5 0 5
0
0.5
1
−5 0 5
0
0.5
1
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 14/ 19

15. FEVD I
Table: Posterior means of FEVD of the EPU measures to shocks
Variable State Shock
Horizon
0 6 12
EPUEA
t
1
3 0.418 0.459 0.459
4 0.280 0.185 0.158
1,2,5 0.302 0.355 0.382
2
3 0.248 0.335 0.360
4, 0.446 0.365 0.338
1,2,5 0.306 0.300 0.203
EPUUS
t
1
3 0.681 0.607 0.569
4 0.047 0.051 0.050
1,2,5 0.272 0.342 0.381
2
3 0.733 0.664 0.638
4 0.101 0.120 0.126
1,2,5 0.166 0.216 0.236
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 15/ 19

16. Impulse responses
Figure: Impulse responses to uncertainty shocks, solid line - posterior
mean, dashed lines - 68% credible sets
0 12 24 36
−0.4
−0.2
0
0.2
0.4
USuncertainty
shock
U
t
EA
0 12 24 36
−0.5
0
0.5
1
U
t
US
0 12 24 36
0
0.1
0.2
0.3
0.4
EPU
t
EA
0 12 24 36
0
0.1
0.2
0.3
EPU
t
US
0 12 24 36
−6
−4
−2
0
GEA
t
0 12 24 36
−0.2
−0.1
0
0.1
0.2
0.3
EAuncertainty
shock
0 12 24 36
−2
−1
0
1
2
0 12 24 36
0
0.1
0.2
0.3
0 12 24 36
−0.1
−0.05
0
0.05
0.1
0 12 24 36
−2
−1
0
1
2
3
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 16/ 19

17. FEVD II
Table: Posterior means of FEVD of the unemployment to shocks
Variable State Shock
Horizon
0 6 12
UEA
t
1
3,US 0.063 0.075 0.095
4,EA 0.008 0.013 0.016
1,2,5 0.929 0.912 0.889
2
3,US 0.111 0.136 0.177
4,EA 0.055 0.090 0.104
1,2,5 0.833 0.774 0.719
UUS
t
1
3,US 0.099 0.125 0.128
4,EA 0.188 0.151 0.144
1,2,5 0.714 0.724 0.728
2
3,US 0.047 0.069 0.073
4,EA 0.266 0.258 0.257
1,2,5 0.687 0.673 0.670
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 17/ 19

18. Theory
Pure RBC (Gilchrist, Williams 2005): expansionary uncertainty
shocks cause reduction of household wealth, increase in marginal
utility of consumption and labor supply.
Sticky prices and search frictions (Leduc, Liu (2014)):
Under nominal rigidities uncertainty shock has a multiplier effect.
Reduction of aggregate demand lowers relative prices. As the firms
post fewer vacancies, unemployment rate increases and household
income decreases.
In the model with search frictions if uncertainty increases, the
present value of a job match declines and the unemployment rises.
Our result tend to support latter two theoretical standpoints.
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 18/ 19

19. Conclusions
We study the cross effects of uncertainty shocks in the two
biggest economic regions: the US and Euro Area.
The shocks are statistically identified via changes in their
volatility. The economic labeling of ”US uncertainty” and
”EA uncertainty” shocks is succeeded via the maximum
explained variance.
We reassure negative effects of economic policy uncertainty
on unemployment.
The increase in the US economic policy uncertainty tends to
increase the unemployment in the Euro area, but no effect
from the EA uncertainty to US labor market is observed.
Higher economic policy uncertainty in the US also tends to
dampen the global real economic activity.
We document that the influence of uncertainty shocks, both
local and foreign, is stronger in less volatile times.
Aleksei Netˇsunajev, Katharina Glass Eesti Pank 19/ 19