Booking open Available Pune Call Girls Shivane 6297143586 Call Hot Indian Gi...
Prudential policies and systemic risk: the role of interconnections
1. Prudential policies and systemic risk:
the role of interconnections
M. Karamysheva (NRU HSE) and E.Seregina (ING Bank)
Open Seminar of Eesti Pank
February 2020
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 1 / 23
2. Introduction
Prudential tools are employed by policymakers to detect or
prevent the build-up of systemic risk
Interconnections between financial systems give rise to
risk-spillover effects
Uncertain policy impact:
Increase vs decrease of systemic risk
Direct vs indirect implications for banking systems
⇓
Major questions:
Policy effectiveness in systemic risk reduction?
Proportion coming from direct and indirect effects?
Differentiation between country-groups/instrument groups?
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 2 / 23
3. Relevant literature and Contribution
Impact of prudential policies on country-level outcomes: Cerutti et
al.(2016), Cerutti et al.(2015), Akinci et al.(2015) and Nistor and
Ongena (2015)
Propagation of monetary or fiscal policy shocks using the spatial
econometrics approach: Briganti, Favero and Karamysheva
(2018), M. Ozdagli and Weber (2017), Eder and Keiler (2013)
Regulatory arbitrage / risk-shifting across jurisdictions: Buch and
Goldberg (2016) and Reinhardt and Sowerbutts (2015)
⇓
Measures of systemicity reflecting systemic fragility
Decomposition into direct and indirect effects with assesment of
the cross country spillovers
Flexible framework (country-group analysis, bank-group analysis)
Robustness to weighting schemes and country-group analysis
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 3 / 23
4. Preview of results
Importance of prudential policies for systemic risk reduction
Vulnerability to systemic crises (SRISK) reduced through the
network effect by 87%
Reduction of network effect to 80% for the EU-only setup and
policy inefficacy for GIIPS
Lower bound of network impact - 24%
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 4 / 23
5. Roadmap
Major concepts and dataset
Methodology
Baseline estimates
Country-group analysis
Robustness checks
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 5 / 23
6. Systemicity definition systemic fragility
SRISK - capital that a firm is expected to need in a financial crisis
bivariate daily time-series model for bank and market returns
(GARCH, DCC and the distribution of error terms)
system simulated N times, worst scenarios for the market (-40%)
treated as crisis => estimate of bank’s equity loss
treating debt as constant => a measure of how much capital would
the institution need to raise SRISK
we assume no mutual support: negative shortfalls are set to zero
when aggregating across intitutions within a country
CoVaR - increased risk to the system when one firm is at extreme
bivariate daily time-series model for bank and market returns
(assuming joint distribution of returns)
assess the “market model’ - based risk to the system when one
firm is at extreme cCOVAR - original
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 6 / 23
7. Dataset
SRISK (Brownlees and Engle, 2016; Acharya, Pedersen,
Philippon and Richardson, M., 2017 - RFS) - Vlab database
90% percentile for the period of interest
Scaled by market capitalisation for cross-sectional comparability
cCoVaR (Adrian and Brunnermeier, 2016 -AER)
computed at 90% quantile
Prudential policy instruments
cross-country database of policy changes - Cerutti et al., 2016
quarterly data, 64 countries, 2000-2014 time span
focus on financial - based measures (bank level regulation)
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 7 / 23
8. Policy measures
General capital requirements - changes to required capital size
from the Basel Accords
Sector specific capital buffer index - adjustment of risk-weights /
specific bank exposures
Reserve requirements - amount of foreign/local currency reserves
to be held against the liabilities.
Limits on concentrated exposures (i.e. interbank exposure limits)
Examples
Policy measure instances
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 8 / 23
9. Weights matrix - country level interconnections
Securities’ holding data from the IMF coordinated portfolio
investment survey (CPIS)
Baseline weights constructed using the beginning-of-period data
Portfolio investment - most stable part of cross-country operations
Zero diagonal and Asymmetry of investment: investment of i into j
does not need to match investment of j into i
Weights Examples
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 9 / 23
10. Methodology: static specification
yit = αi + ρy∗
it + βxit + γt + uit (1)
y∗
it = ∑
i∗=i
ωi∗jyi∗t (2)
y - country-specific dependent variables
αi - non-spatial fixed effects, γt - time fixed effects
y∗ - vector of foreign variables specific to country i
ωij is i, j - share of country j’s assets in the investment portfolio of
country i
xit - country-specific prudential tools
uit - idiosyncratic shock
Dynamic Specification IV Specification Decomposition Potential Issues Method
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 10 / 23
11. Results: SRISK for the banking system
Variables Static Static - IV Dynamic Dynamic - IV
AutoregressiveTerm µ
- - 0.002 -0.019
- - [0.018] [0.024]
Global Variable ρ
0.986*** 0.927*** 0.887*** 0.879***
[0.009] [0.024] [0.031] [0.043]
Prudential Policy β
-1.425** -1.672*** -1.542*** -1.708***
[0.612] [0.608] [0.619] [0.646]
Country and Year FE YES YES YES YES
Observations 1121 1121 1102 1102
Countries 19 19 19 19
The Table shows the results for SRISK/MCap (logarithmic returns) indicator for
countries’ banking systems
Estimation is done by SUR. Estimation period: 2000Q2 - 2014Q4. *** - p-value
below 0.01, ** - p-value between 0.01 and 0.05, * - p-value below 0.1
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 11 / 23
12. Banking system SRISK - impact decomposition
Banking system Financial system
Variables Static Static - IV Static Static - IV
Total Effect -16.364 -25.773 -6.834 -9.543
Direct Effect -2.174 -2.888 -0.928 -1.352
Incl. instantaneous effect -1.425 -1.672 -0.619 -0.925
Incl. network feedback -0.749 -1.216 -0.309 -0.427
Indirect Effect -14.190 -22.885 -5.906 -8.191
Indirect Effect, % of Total 87% 89% 86% 86%
P 0.014 0.006 0.09 0.024
Simulations done using bootstrap with resampling for 10000
iterations
P stands for the probability for the average effect to be higher than
zero, taking into account the total bootstrap distribution.
Partitioning
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 12 / 23
13. Country groups
Are countries equally responsive to the policy interventions?
⇓
Group Non-EU EU-core EU-GIIPs
Members
Australia,
Canada, the
USA
Austria, Belgium, Denmark, Fin-
land, France, Germany, the Nether-
lands, Norway, Sweden, Switzer-
land and UK
Greece,
Ireland, Italy,
Portugal and
Spain
SRISK/Market Capitalisation by country group
0
1
2
3
4
5
6
non-EU EU
0
1
2
3
4
5
6
7
8
EU-non GIIPS EU-GIIPS
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 13 / 23
14. Country groups - impact decomposition
Are countries equally responsive to the policy interventions?
Is U.S. economy driving the results?
Country-groups World Country -groups Europe only
Variables (1) non-EU (2) EU - core (3) EU - GIIPS (4) EU core (5) EU - GIIPS
Global Variable 0.680*** 1.023*** 0.701*** 0.863*** 0.758***
[0.053] [0.035] [0.128] [0.046] [0.130]
Prudential policy -2.870*** -2.517*** 0.545 -2.820** 0.636
[0.937] [0.859] [1.799] [0.827] [1.668]
Observations 1121 1121 1121 944 944
Countries 19 19 19 16 16
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 14 / 23
15. Country groups - impact decomposition
Are countries equally responsive to the policy interventions?
Is U.S. economy driving the results?
Country-groups World Country-groups Europe only
(1) non-EU (2) EU - core (3) EU - GIIPS (4) EU - core (5) EU - GIIPS
Total Effect -27.81 -26.63 5.01 -20.81 5.17
Direct Effect -6.29 -3.86 0.83 -4.12 1.01
Inst effect -2.87 -2.52 0.55 -2.82 0.64
Network feedback -3.42 -1.35 0.28 -1.30 0.37
Indirect Effect -21.52 -22.77 4.18 -16.69 4.16
from non-European -4.81 -5.95 1.05 - -
from core European -12.88 -12.40 2.47 -11.95 3.21
from GIIPS -3.83 -4.42 0.66 -4.74 0.95
Indirect, % of Total 77% 85% 83% 80% 81%
P 0.002 0.001 0.682 0.000 0.710
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 15 / 23
16. Robustness checks - de-factored series
Are the results driven by the common trends in the financial
markets?
⇓
Potential drivers of the spillover effects
common factors ( strong cross sectional dependence)
true spillovers ( weak cross sectional dependence)
⇓
use year fixed effects
analogue of CAPM-adjustment with residuals from the (quarterly
basis) market model fed into the spatial models instead of the raw
observations (rjt )
rjt = αj + βjrm,t + jt
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 16 / 23
17. Robustness checks - de-factored series
Are the results driven by the common trends in the financial
markets?
Panel A Panel B Panel C Panel D
Variables Baseline IV Residuals Residuals - IV
(1) (2) (3) (4) (5) (6) (7) (8)
Total Effect -8.80 -15.14 -38.08 -25.77 -1.39 -1.58 -1.58 -1.81
Direct Effect -1.04 -2.24 -3.39 -2.89 -0.77 -1.20 -0.90 -1.21
Incl. instantaneous impact -0.63 -1.57 -1.52 -1.67 -0.74 -1.19 -0.88 -1.19
Incl. network feedback -0.41 -0.67 -1.88 -1.22 -0.02 -0.01 -0.02 -0.02
Indirect Effect -7.77 -12.91 -34.68 -22.89 -0.62 -0.38 -0.68 -0.60
Indirect Effect, % of Total 88% 85% 91% 89% 44.9% 23.8% 43% 33.3%
P 0.071 0.010 0.008 0.006 0.085 0.214 0.060 0.133
Year Fixed Effects NO YES NO YES NO YES NO YES
Country Fixed Effects YES YES YES YES YES YES YES YES
Observations 1121 1121 1121 1121 1121 1121 1121 1121
Countries 19 19 19 19 19 19 19 19
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 17 / 23
18. Robustness checks
Are the results driven by the choice of instruments, outcome
variables and/or weights?
⇓
Alternative weighting schemes
Top counterpaty weights
Geography-based weights
Instrument groups
Post crisis and CPIS vs. CDIS
Alternative outcomes
Alternative estimation TSLS, GMM
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 18 / 23
19. Robustness checks - alternative weights
Are the results driven by the choice of weights?
Full sample European sample
Weights (1) Baseline (2) Top 10 (3) >5% (4) Baseline (5) Geographic
Total Effect -25.77 -11.96 -12.60 -10.57 -19.76
Direct Effect -2.89 -2.40 -2.22 -1.91 -2.40
Incl. Instantaneous effect -1.67 -1.95 -1.71 -1.40 -0.99
Incl. network feedback -1.22 -0.45 0.51 0.51 -1.41
Indirect Effect -22.89 -9.56 -10.37 -8.65 -17.36
Indirect Effect, % of Total 89% 79.9% 82.3% 81.9% 87.9%
P 0.006 0.003 0.008 0.024 0.098
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 19 / 23
20. Prudential poicies in use IV
Are the results driven by some group of instruments?
Variables (1) Financial-based (2) No concentration limits (3) No required reserves (4) Centrally set
Total Effect -25.77 -34.54 -20.07 -31.23
Direct Effect -2.89 -3.89 -2.25 -3.58
Incl. instantaneous effect -1.67 -2.67 -1.30 -2.12
Incl. network feedback -1.22 -1.22 -0.95 -1.46
Indirect Effect -22.89 -30.65 -17.82 -27.65
Indirect Effect, % of Total 89% 88.7% 88.8% 88.5%
P 0.006 0.009 0.025 0.013
Countries 19 19 19 19
Baseline results aligned with
including only the measures stemming from the centaral authorities
(i.e. ECB)
excluding the measures targeting the links between banks and
exposures to systemic events (i.e. specific sectors)
omitting required reserves (monetary policy tool)
Point estimates
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 20 / 23
21. Prudential poicies in use IV
Are the results driven by crisis? Post crisis analysis
Period (1) Baseline (2) Post 2009 (3) Post 2009 (4) Post 2009
Weights Baseline Baseline 2009 CPIS 2009 CDIS-CPIS
Total Effect -25.773 -19.321 -12.040 -12.939
Direct Effect -2.888 -2.669 -2.511 -2.528
Incl. instantaneous effect -1.672 -1.802 -2.043 -2.009
Incl. network feedback -1.216 -0.867 0,468 -0.519
Indirect Effect -22.885 -16.651 -9.527 -10.410
Indirect Effect, % of Total 89% 86% 79% 80%
P 0.006 0.041 0.035 0.036
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 21 / 23
22. Prudential poicies in use IV
Are the results driven by alternative outcomes?
Delta - CoVaR Exposure CoVaR
LRMES Asset-based Equity-based Asset-based Equity-based
Total Effect -0.758 -0.306 -5.856 -4.058 -1.595
Direct Effect -0.142 -0. 029 -0.499 -0.702 -0.449
Incl. instantaneous effect -0.111 -0. 199 -0.154 -0.532 -0.397
Incl. network feedback -0.031 0.170 -0.345 -0.170 -0.052
Indirect Effect -0.616 -0.277 -5.406 -3.356 -1.147
Indirect Effect, % of Total 81% 91% 92% 83% 72%
P 0.369 0.416 0.000 0.001 0.010
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 22 / 23
23. Conclusions
Relevance of prudential policies policy for the risk-reduction
More pronounced impact for the overall financial sector
Robustness to the multiple specifications in terms of weights /
dependent variables / country groupings and instruments’ choice
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 23 / 23
24. APPENDIX
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 24 / 23
25. Policy measures - instances
Country 2000-05 2006-10 2011-14 Country 2000-05 2006-10 2011-14
Australia 3 1 2 Italy 1 1 1
Austria 1 0 3 Luxembourg 1 0 4
Belgium 1 0 1 Netherlands 1 1 2
Canada 0 0 2 Norway 2 2 2
Denmark 0 0 2 Portugal 1 1 3
Finland 1 0 1 Spain 1 1 3
France 2 2 4 Sweden 1 0 6
Germany 0 1 2 Switzerland 0 1 4
Greece 1 0 1 United Kingdom 0 1 2
Ireland 1 2 1 United States 0 0 2
79 instances of policy changes
Advanced economies more proactive since the global financial
crisis
Back to Policies
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 25 / 23
26. Policy Examples
General Capital Requirements - 2011Q1 Switzerland 1 Basel
II.5 increse in capital requirements against the market risk
Sector specific CR - real estate - 2000Q1 Norway 1 Risk
weights on loans with loan-to-value higher than 60% raised from
50% to 100%.
Reserve Requirements - 2000Q1 Belgium -1 Effective from 24
January 2000 the ECB increased from 10% to 30% the
standardised deductions for reserve requirements to debt
securities issued with maturity up to 2 years
Concentration Limits - 2000Q2 France 1 Concentration risks are
defined as gross exposure above 10% of bank’s capital or above
300 Millions euros. Weighted concentration risks do not exceed
25% of bank’s capital; Sum of weighted concentration risk shall
not exceed 800% of bank’s capital. Back to Policies Centralised decisions
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 26 / 23
27. Centralized Policy Examples
Sector specific capital requirements
Spain 1
Higher risk weight under Basel II approach for mortgages
that exceed Loan to Value of 95% for residential property
Reserve requirements
Italy -1
ECB increased the standardised deductions to money mar-
ket paper from 10% to 30%
Limits on concentrated exposures
Netherlands 1
...the limits on interbank and intragroup exposures were
tightened; the number of exemptions was limited. Changes
are the result of the introduction of CRD III
Back to Policy measures
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 27 / 23
29. Methodology: dynamic specification
yit = αi + µyit−1 + ρy∗
it + βxi,t + γt + uit (3)
y∗
it = ∑
i∗=i
ωi∗jyi∗t (4)
yi,t - country-specific dependent variables
yi,t−1 - lagged country-specific dependent variables
αi non-spatial fixed effects and γt - time fixed effects
y∗ foreign variables specific to country i
ωij is i, j - share of country j’s assets in the investment portfollio of
country i.
xi,t - country-specific prudential tools
uit - idiosyncratic shocks
Back to Static Specification
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 29 / 23
30. Instrumenting the global variable
Static model considers global variable at time t
Potential problem of reverse causality (if country i’s systemic risk
is significant enough to affect the contemporaneous measures in
related countries)
⇓
Lagged value(s) of a global variable used as an instrument
y∗
it = γ + δy∗
it−1 + i,t (5)
yit = αi + ρy∗
it + βxit + γt + uit (6)
y∗
it = ∑
i∗=i
ωi∗jyi∗t (7)
y∗
it - fitted value from the first stage regression
For robustness substitute (4) with
y∗
it = γ + δL(4)y∗
it−1 + i,t (8)
Back to Static Specification
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 30 / 23
31. Decomposition of impact
yit = αi + ρWyit + βxit + uit (9)
Country-specific effects of prudential policy computed as:
∂∆yi
∂xi
= (I − ρ · W)−1
· 1 · β
Average direct effect: average of the diagonal elements of
(I − ρ · W)−1β
Average total effect: sum across the ith row of (I − ρ · W)−1β -
total impact on country i from the prudential policy shock. There
are n sums like this. The average of this sums is an average total
effect.
Average indirect effect: difference between the average total effect
and average direct effect.
Back to Static Specification
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 31 / 23
32. Method
Lee(2002) - SUR vs Maximum Likelihood estimation
Least squares (SUR) - consistent - if each unit can be influenced
aggregately by a signifficant option of units in the population
Typically held if weights reflect economic characteristics rather than
physical location of units
strong divergence: the row and column sums of W before W is
normalized should diverge to infinity at a rate faster than sqrt(N)
the correlation between two spatial units converges so slowly to
infinity that all units may say to affect each other
does not work with pure spatial autoregression (no X)
Back to Static Specification
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 32 / 23
33. Endogeneity
yit = αi + ρy∗
it + βxit + uit (10)
Matrix of weights
stability of weights
robustness check for neighbours relation
Global variable
lagged global variables
Policy interventions
controls...
as good as exogenous (set by the central institution)
robustness - identifying the most ”exogenous” ones
political instruments?
Back to Static Specification
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 33 / 23
34. Endogeneity of policy interventions
policy intervention might be in fact a response to the worsened
systemic risk levels
β would be biased upwards given the positive correlation of policy
interventions with disturbance terms => policy instruments?
this bias would enforce the policy significance (our β would be more
negative in its absence)
analysis including more emerging countries (more likely to respond
to crisis conditions) results in a positive β
Back to Static Specification
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 34 / 23
35. Spatial partitioning of direct, indirect and total impacts
Order Total (cum.) Direct (cum.) Indirect (cum.)
W0 6.49% 57.89% 0.00%
W1 12.49% 57.89% 6.77%
W5 32.46% 68.24% 27.94%
W10 50.42% 76.69% 47.11%
W15 63.04% 82.62% 60.57%
Overall 100% 100% 100%
Zero-order impact - only direct effect
50% of impact coming up to the 10th order effect
Back to SRISK Decomposition
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 35 / 23
36. Systemic risk indicators - SRISK
SRISK shows the potential shortfall of capital in the crisis event
SRISKi,t = Et−1(CapShortfalli,t | Crisist ) =
E((k(Debti,t + Equityi,t ) − Equityi,t | Crisist ) =
kDebti,t − (1 − k)(1 − LRMESi,t )Equityi,t
Reflects banks’ adjustments through
LRMES - loss given the 40% decline in the market (i.e. tail risk)
Leverage - debt-to-equity ratio (higher leverage - higher risk)
Based on the market data and available at a high frequency
Back to Systemicity
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 36 / 23
37. cCoVaR computation - original approach
Typically compute assets cCoVaR as:
Ri,t = αi + γiMt−1 + i,t (11)
Rm,t = αtotal|i
+ βtotal|i
Ri,t + γtotal|i
Mt−1 +
total|i
m,t (12)
Mt−1 includes the characteristics of financial markets
Predicted values from the quantile regressions used to compute:
VaRi,t (q) = ˆαi,q + ˆγi,qMt−1 (13)
CoVaRi
t (q) = ˆαtotal|i + ˆβtotal|i
VaRi,t (q) + ˆγtotal|i
Mt−1 (14)
The final metrics are obtained as:
cCoVaRi
t (1%) = ˆβtotal|i
(VaRi
t (1%) − VaRi
t (50%)) (15)
cCoVaRi
t (5%) = ˆβtotal|i
(VaRi
t (5%) − VaRi
t (50%)) (16)
Back to Systemicity
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 37 / 23
38. Prudential poicies in use IV
Variables (1) Financial-based (2) No concentration limits (3) No required reserves (4) Centrally set
Global Variable ρ
0.927*** 0.927*** 0.928*** 0.925***
[0.024] [0.024] [0.024] [0.024]
Prudential Policy β
-1.672*** -2.266*** -1.302** -2.117**
[0.608] [0.828] [0.627] [0.835]
Country Fixed Effects YES YES YES YES
Year Fixed Effects YES YES YES YES
Countries 19 19 19 19
Back to Country
M. Karamysheva and E.Seregina Prudential policies and systemic risk: the role of interconnectionsFebruary 2020 38 / 23