2237 Views
357 Downloads 
The impact of macroprudential policy on financial stability in selected EU countries*
Article  Year: 2022  Pages: 141  170  Volume: 46  Issue: 1 Received: March 5, 2021  Accepted: October 1, 2021  Published online: March 8, 2022

FULL ARTICLE
FIGURES & DATA
REFERENCES
CROSSMARK POLICY
METRICS
LICENCING
PDF

d(CGR)

d(HPGR)

d(DEG)

d(CET)

d(LDR)

d(LR)

d(NDF)

d(CR)

d(INR)

Mean

1.97

0.04

0.59

0.08

1.16

0.12

0.02

0.17

0.17

Median

1.22

0.10

0.24

0.12

0.90

0.07

0.02

0.11

0.15

Maximum

80.41

7.10

18.11

2.03

13.99

1.07

1.32

17.96

3.44

Minimum

34.33

4.90

4.79

2.16

6.65

3.79

1.38

10.47

2.69

Standard deviation

10.80

2.23

2.37

0.54

3.46

0.61

0.44

2.79

0.77

Skewness

4.16

0.45

4.88

0.10

0.99

2.15

0.36

2.88

0.77

Kurtosis

37.63

3.72

35.66

6.91

5.66

14.82

4.98

23.08

8.74

JarqueBera

5,074.44

5.38

4,648.03

61.22

43.92

632.61

17.68

1,746.59

141.32

Probability

0.00

0.07

0.00

0.00

0.00

0.00

0.00

0.00

0.00

Sum

189.22

3.60

56.73

8.08

111.21

11.90

2.19

16.69

16.60

Sum sq. dev.

11,089.82

470.89

531.76

28.07

1,136.47

35.75

18.11

737.41

56.32

Obser.

96

96

96

96

96

96

96

96

96

Notes: “d” denotes the first difference of a variable. For instance, d(CGR) denotes the first difference of CGR.Source: Authors’ calculations.
Response and explanatory variables

ADFFisher Chisquare statistic
(ADFFisher Chisquare probability)

Level (x)

First difference d(x)

CGR

13.2965
(0.3479)

70.8197
(0.0000)

HPGR

74.8063
(0.0000)

88.6962
(0.0000)

DEG

15.3016
(0.2254)

71.0383
(0.0000)

CET

11.9982
(0.4458)

71.0925
(0.0000)

LDR

11.0842
(0.5217)

47.9746
(0.0000)

LR

6.74768
(0.8738)

71.2775
(0.0000)

NDF

10.2671
(0.5925)

35.9455
(0.0003)

CR

5.87756
(0.9221)

54.8494
(0.0000)

INR

4.92658
(0.9604)

88.1111
(0.0000)

Notes: pvalues for the FisherADF panel unit root test are computed using the asymptotic Chisquare distribution and given in brackets. The maximum number of lags was automatically selected with Schwarz Information Criterion. Source: Authors’ calculations.
Model no.

Response
variable

Explanatory
variable/statistics

Crosssection
fixed effects

Period fixed
effects

Crosssection
fixed effects and period fixed effects

Period random
effects

1

DCGR

C

0.678
(0.595)
(0.553)

0.991
(0.991)
(0.369)

0.833
(0.757)
(0.451)

0.844
(0.644)
(0.521)



DCET

4.353
(1.971)
(0.052)*

5.844
(2.479)
(0.016)**

5.533
(2.350)
(0.022)**

5.128
(2.358)
(0.021)**



DINR

2.990
(1.971)
(0.052)*

0.670
(0.390)
(0.698)

0.766
(0.436)
(0.664)

2.298
(1.536)
(0.128)



DLDR

0.541
(1.699)
(0.093)*

0.635
(2.004)
(0.049)**

0.707
(2.138)
(0.036)**

0.523
(1.742)
(0.085)*



DNDF

8.058
(3.322)
(0.001)***

7.599
(3.209)
(0.002)***

8.081
(3.338)
(0.001)***

7.666
(3.344)
(0.001)***



DLR

0.347
(0.151)
(0.880)

1.598
(0.720)
(0.474)

0.883
(0.374)
(0.710)

1.095
(0.518)
(0.606)



DCR

0.058
(0.134)
(0.894)

0.044
(0.107)
(0.915)

0.079
(0.178)
(0.859)

0.017
(0.044)
(0.965)



Rsquared

0.243

0.397

0.442

0.194



S.E.
of regression

9.998

9.503

9.470

9.580



Fstatistic

2.450

2.324

2.102

3.571



Prob.
(Fstatistic)

0.011

0.004

0.008

0.003



Sum
squared resid

8396.389

6682.217

6188.189

8168.947



DurbinWatson
stat

1.567

1.467

1.573

1.454



Redundant
fixed effects
test (F prob.)

0.368

0.085

0.099





Hausman
correlated random effects test
(Chisquare prob.)







0.280



KleinbergenPaap
test





(0.003)

(0.000)



HansenSargan
test





(0.578)

(0.691)

2

DHPGR

C

0.157
(0.599)
(0.551)

0.073
(0.309)
(0.758)

0.142
(0.560)
(0.577)

0.108
(0.329)
(0.743)



DCET

0.482
(0.952)
(0.344)

0.109
(0.215)
(0.831)

0.497
(1.011)
(0.315)

0.200
(0.417)
(0.678)



DINR

0.482
(1.384)
(0.170)

0.558
(1.504)
(0.137)

0.505
(1.515)
(0.133)

0.533
(1.598)
(0.114)



DLDR

0.059
(0.807)
(0.422)

0.012
(0.178)
(0.859)

0.050
(0.724)
(0.471)

0.031
(0.469)
(0.640)



DNDF

0.607
(1.090)
(0.279)

0.344
(0.673)
(0.503)

0.627
(1.183)
(0.240)

0.483
(0.967)
(0.336)



DLR

0.129
(0.245)
(0.807)

0.042
(0.088)
(0.930)

0.232
(0.478)
(0.634)

0.137
(0.295)
(0.769)



DCR

0.024
(0.238)
(0.813)

0.040
(0.443)
(0.659)

0.040
(0.444)
(0.658)

0.039
(0.456)
(0.649)



Rsquared

0.061

0.337

0.057

0.043



S.E.
of regression

2.294

2.053

2.234

2.051



Fstatistic

0.500

1.794

0.890

0.672



Prob.
(Fstatistic)

0.898

0.035

0.506

0.673



Sum
squared resid

441.967

312.018

444.238

374.423



DurbinWatson
stat

2.886

3.026

2.870

2.937



Redundant
fixed effects
test (F prob.)

0.995

0.032

0.113





Hausman
correlated random effects test
(Chisquare prob.)







0.446



KleinbergenPaap
test





(0.009)

(0.000)



HansenSargan
test





(0.487)

(0.542)

3

DDEG

C

0.726
(2.908)
(0.005)***

0.687384
(2.655870)
(0.0097)***

0.712206
(2.834312)
(0.0060)***

0.701930
(2.606203)
(0.0107)**



DCET

0.409
(0.846)
(0.400)

0.613586
(1.103253)
(0.274)

0.647117
(1.203163)
(0.233)

0.393038
(0.785712)
(0.434)



DINR

0.244
(0.734)
(0.465)

0.002
(0.005)
(0.996)

0.060
(0.151)
(0.881)

0.230
(0.675)
(0.502)



DLDR

0.103
(1.476)
(0.144)

0.087
(1.164)
(0.248)

0.125
(1.653)
(0.103)

0.076
(1.097)
(0.276)



DNDF

2.111
(3.968)
(0.000)***

2.034
(3.642)
(0.001)***

2.184
(3.950)
(0.000)***

1.970
(3.682)
(0.000)***



DLR

0.109
(0.216)
(0.829)

0.174
(0.332)
(0.741)

0.181
(0.336)
(0.738)

0.097
(0.197)
(0.844)



DCR

0.004
(0.047)
(0.963

0.066
(0.676)
(0.501)

0.027
(0.269)
(0.789)

0.038
(0.416)
(0.678)



Rsquared

0.241

0.301

0.393

0.155



S.E.
of regression

2.192

2.241

2.163

2.226



Fstatistic

2.422

1.517

1.717

2.716



Prob.
(Fstatistic)

0.011

0.098

0.039

0.018



Sum
squared resid

403.705

371.753

322.844

441.087



DurbinWatson
stat

2.025

1.772

2.024

1.822



Redundant
fixed effects
test (F prob.)

0.077

0.330

0.169




Hausman
correlated random effects test
(Chisquare prob.)







0.570



KleinbergenPaap
test





(0.006)

(0.000)



HansenSargan
test





(0.7285)

(0.815)

Notes: Crosssection random effects and crosssection random effects and period random effects were not possible to estimate, since random effects estimation requires number of cross sections > number of coefs for between estimator for estimate of RE innovation variance. In the table, all regressors and regressands have a »D« in front of their name (e.g., CGR becomes DCGR), since all variables are taken at first difference for stationarity. The tstatistics are given in brackets below the coefficients and the pvalues are in brackets below the tstatistics. Significance levels are denoted as: *** significant at 1%; ** significant at 5%, * significant at 10%. Source: Authors’ calculations. KleibergenPaap test and HansenSargan test were carried out with STATA 12 version. The rest of the analyses were conducted in EViews version 11.
Author(s)

Economies

Methodology

Results

Davis, Liadze and Piggott (2019)

UK, Italy and Germany

National Institute's Global
Econometric Model, NiGEM. NiGEM is a global econometric model, and most
countries in the EU and the OECD, as well as major emerging markets, are
modelled individually. The rest of the world is modelled through a set of
regional blocks so that the model is global in scope. Macroprudential policy
is incorporated in NiGEM. There are three simulations/shocks in the models:
Tightening of loantovalue policy; increase in riskadjusted capital
adequacy target; and historic dynamic simulation for the crisis period.

The loantovalue simulation
predominantly impacts consumption and the housing market, whereas the capital
adequacy simulation has a more significant effect on investment and output.
Both simulations increase bank capital ratios and curb bank lending. The
findings of the study suggest that, overall, the loantovalue tool has a
lower effect than capital adequacy on the probability of a banking crisis occurring
and leads to lower net benefits. The introduction of macroprudential policy
measures before the onset of the crisis leads to an improvement in key
macroeconomic measures and might therefore prevent the crisis from
materializing.

Carreras, Davis and Piggott (2018)

19 OECD countries

Cointegration framework;
VectorErrorCorrection (VECM), Vector Autoregression (VAR), Fully Modified
OLS (FMOLS) and SeeminglyUnrelated (SUR) estimation.
A comparison of results from
cointegration with noncointegrating specifications.

Macroprudential policy instruments
(taxes on financial institutions, capital requirements, loan to value ratios,
debt to income ratio limits, limits on foreign currency lending, limits on
interbank exposures and concentration limits) have a positive impact on
stalling household credit growth and house prices in both shortrun and
longrun. Tools such as limits on debttoincome ratios are more effective
for limiting house price growth, whilst tools such as limits on interbank
exposures are more effective for constraining household credit growth.

Olszak, Roszkowska and Kowalska
(2019)

65 countries

GMM 2step Blundell and Bond
approach and random effects method

Macroprudential policy instruments
indeed curb the procyclical impact of capital ratios on loan growth rate,
whereby this impact is more pronounced in large banks than in smaller banks.
Of the investigated macroprudential instruments, only borrowerbased measures
such as LTV and DTI caps seem to act countercyclically by weakening the positive
impact of capital ratio on bank lending, in particular in crisis periods.

Ma (2020)

A small open economy

A small open economy (SOE) model
with endogenous growth and occasionally binding collateral constraints to analyse
the role of macroprudential policy in the context of a tradeoff between
growth and financial stability.

Macroprudential policy is shown to
substantially strengthen financial stability (it reduces the frequency and
probability of crises) at the cost of a very small negative effect on average
growth and welfare. In two extensions of the model (one with a growth subsidy
and another one with a direct growth externality), the optimal
macroprudential policy proves to have a more pronounced effect on welfare and
growth. Although macroprudential policy slightly curbs average growth, it is
still desirable to use it, since it enhances financial stability and smooths
consumption.

Akinci and OlmsteadRumsey (2018)

57 advanced and emerging economies

Estimation of a dynamic panel data
regression model with country fixed effects using the Generalized Method of
Moments (GMM).

Macroprudential tightening (by
using macroprudential tools such as LTV limits, DSTI limits, other housing
measures, timevarying capital requirements, provision requirements, consumer
loan limits, and credit growth ceilings) dampens bank credit growth, housing
credit growth, and house price appreciation. Macroprudential policies
targeting the housing sector appear to be more effective at constraining
housing credit growth and house price appreciation, in particular in
economies where bank finance is of greater importance. Counterfactual
simulations indicate that, if the countries had not used any macroprudential
policy measures in the period 2011–2013, the bank credit growth, housing
credit growth and house price appreciation would have been substantially
higher.

Meuleman and Vander Vennet (2020)

Listed European banks

Dynamic panel framework

Borroweroriented tools and
exposure limits are found to reduce the individual bank risk component.
Liquidity measures are found to reduce the systemic linkage of banks in
addition to reducing individual bank risk. Credit growth measures and
exposure limits seem to lead to an increase in systemic risk component for
some banks – possibly because some banks, when trying to observe the rules,
take up riskier activities or similar exposures, thus exacerbating
interconnectedness of the banks in the system. Macroprudential policies seem
to be the most effective for distressed banks, that is banks with a high
ratio of nonperforming loans. The results of the study give some indications
for the optimal design of macroprudential measures.

Altunbas, Binici and Gambacorta
(2017)

61 advanced and emerging market
economies

Baseline empirical regression model
adapted from Altunbas, Gambacorta and MarquesIbanez (2014); SGMM estimator

The results of the study are
threefold: Macroprudential policy tools have a substantial effect on bank
risk. Banks with different characteristics do not respond uniformly to
changes in macroprudential policy tools. Small, weakly capitalized banks, and
banks having a high share of wholesale funding respond more strongly to
changes in macroprudential policy tools. Macroprudential policies are more
efficient when employed during a downturn than during a boom.

Cizel et al. (2019)

40 economies

Panel regression; PSM approach to
simulate the effect of a randomized experiment in nonrandom, observed data

The authors investigate whether
the implementation of macroprudential policy leads to a substitution of bank
credit with nonbank credit. On the one hand, it could be claimed that the
substitution effect leads to the propagation of new systemic risks, since
when credit shifts away from banks, households and firms continue to accumulate
debt, thus engendering macroeconomic fragility. On the other hand, it could
be asserted that the substitution effect reduces systemic risks, since
nonbank financial institutions are, by and large, less leveraged and with
lower liquidity risks than the banks, and mostly do not have access to public
safety nets, hence are less prone to the moral hazard problem.

Dumičić (2018)

11 CEE countries

Panel regressions using the OLS
method and crosssection SUR panelcorrected standard errors

The findings demonstrate that in
the CEE countries, macroprudential policies were more effective in weakening
the flow of credit to households than the flow of credit to the nonfinancial
corporate sector prior to the global financial crisis with the onset in 2007.
This is predominantly because the nonfinancial corporate sector also had
access not only to domestic bank credit, but also to nonbank and
crossborder credit. The conclusion of the paper is that some international
cooperation among policymakers is warranted so as to align macroprudential
policies and prevent “regulatory arbitrage”.

Source: Authors’ compilation.
Table 1Descriptive statistics of explanatory and response variables DISPLAY Table
Table 2Unit root test (Fisher ADFtest) DISPLAY Table
Table 3Empirical results DISPLAY Table
Table A1Empirical literature overview DISPLAY Table
* The views and opinions expressed in this paper are solely those of the authors and do not in any way reflect the official policy, position or opinion of the Faculty of Economics and Business, University of Maribor or of Credit Suisse Group AG, Zurich, Switzerland. The authors would like to thank two anonymous reviewers for their valuable remarks and suggestions.
^{1} Given the time span (20152018), the ratio captures a change in (i) the definition of NPLs (due to how EBA pushed for a uniform and conservative definition), and (ii) the way provisions were calculated – until end2017, IAS 39 with the incurred loss concept was valid while from early 2018, banks need to use IFRS 9 with its expected credit loss model.
^{2} In our analysis we actually use a more traditional definition of the leverage ratio (i.e., total assets divided by total equity), but the general idea is the same.


March, 2022 I/2022 