DETERMINANTS OF DEPOSITINSURANCE ADOPTION AND DESIGN

1
DETERMINANTS OF DEPOSIT-INSURANCE ADOPTION AND
DESIGN
February, 2005

• Asli Demirgüç-Kunt, World Bank
• Edward J. Kane, Boston College
• Luc Laeven, World Bank

2

• Most countries do not have an explicit
  deposit-insurance scheme (EDIS)
• A countrys level of economic and financial
  development tells partbut only partof the
  story.
• Most of the poorest and least financially
  developed countries have no EDIS at all and most
  high-income countries do.
• Table 1 partitions 181 sample countries for which
  we have per capita income data into quartiles and
  shows that the propensity to adopt an EDIS rises
  with income

3
Table 1 Distribution of Countries with and
without explicit deposit insurance by income
quartile at yearend 2003The data come from the
World Bank Deposit Insurance Database (2004),
compiled from the International Association of
Deposit Insurers (IADI) and national sources. The
total number of countries included is 181.
Blanket guarantees are coded as EDIS.
4
(No Transcript)
5

• We use logit regressions and hazard models to
  investigate the extent to which other economic,
  political, and institutional variables help to
  explain whether and when a country decides to
  install an EDIS.
• Treating the logit adoption model as a Heckman
  selection equation, we go on to estimate whether
  an how country characteristics affect the
  effectiveness of the risk-shifting controls that
  the scheme embodies.

6
Structural Supply and Demand Functions For an
EDIS Cannot Readily Be Identified

• Per capita GDP affects both demand for coverage
  and the tax capacity needed to supply EDI in a
  credible fashion
• Crises and follow-on threats of systemic runs
  increase both the demand for EDI by bank
  stakeholders and incumbents willingness to
  supply it
• Political power-sharing increases voice of
  sectors seeking EDI subsidies and the benefits to
  politicians of supplying these subsidies

7

• In some cases, a presumption that an EDIS
  represents a hallmark of regulatory best practice
  may have outweighed the contrary influence of
  internal economic, institutional, and political
  counterforces.
• To test this hypothesis, our model inserts into
  models featuring domestic determinants of
  regulatory decisions a proxy for emulation,
  reflecting the spread of EDI systems.
• It is important to see both that emulation is
  an accelerating trend variable and that we are
  assigning an interpretation for the trend
• Without a potentially testable interpretation,
  any trend is an empty concept

8

• We also test the complementary hypothesis that
  outside (especially IMF) crisis-management advice
  might direct countries that experience a systemic
  crisis to erect an EDIS as either a myopic way of
  containing crises or an overly hopeful way of
  formally winding down crisis-generated blanket
  guarantees.

9
EMULATION
Crisis-Influenced Adoption of EDIS
10
Hypothesis-Testing Strategy

• As potential economic determinants, we include
  macroeconomic conditions, financial-crisis
  events, fiscal costs incurred in crises, and
  inefficiencies presumed to be associated with
  state-owned banks.
• As potential institutional and political
  determinants, we include various features of the
  countrys private and public contracting
  environments
• Transparency (T)
• Deterrency (D)
• Bonding (B)
• Accountability (A)

11
MTMDMBNA Thought Experiment When Would
Government DI Produce No Benefits at All?

• MAXIMAL TRANSPARENCY (MT)
• MAXIMAL DETERRENCY (MD)
• MAXIMAL BONDING (MB)
• NO ACCOUNTABILITY (NA)

12

• Table 2 names the design features our dataset
  covers and the country characteristics that the
  regression experiments employ. The unit of
  observation is a country-year. The table reports
  summary statistics on all variables.

13

• Table 3. Explicit deposit insurance systems at
  yearend 2003
• This table lists the countries that adopted
  explicit deposit insurance systems by yearend
  2003. The data come from the World Bank Deposit
  Insurance Database (2004). GDP and bank deposits
  per capita are from International Financial
  Statistics (IFS). The following non-adopting
  countries are included in our sample
  Afghanistan, Angola, Armenia, Australia,
  Azerbaijan, Barbados, Belize, Benin, Bhutan,
  Bolivia, Botswana, Brunei, Burkina Faso, Burundi,
  Cambodia, Cameroon, Cape Verde, Central African
  Republic, Chad, China, Comoro Islands, Costa
  Rica, Cote d'Ivoire, Cuba, Djibouti, Egypt,
  Equatorial Guinea, Eritrea, Ethiopia, Fiji,
  Gabon, Gambia, Georgia, Ghana, Grenada, Guinea,
  Guinea-Bissau, Guyana, Haiti, Hong Kong (China),
  Iran, Iraq, Israel, Kiribati, Kyrgyz Republic,
  Laos, Lesotho, Liberia, Libya, Madagascar,
  Malawi, Maldives, Mali, Mauritania, Mauritius,
  Moldovad, Mongolia, Morocco, Mozambique, Myanmar,
  Namibia, Nepal, New Zealand, Niger, Pakistan,
  Panama, Papua New Guinea, Qatar, Republic of
  Congo, Rwanda, Saudi Arabia, Senegal, Seychelles,
  Sierra Leone, Singapore, Solomon Islands,
  Somalia, South Africa, St. Lucia, Sudan,
  Suriname, Swaziland, Syria, Tajikistan, Togo,
  Tunisia, United Arab Emirates, Uruguay,
  Uzbekistan, Vanuatu, W. Samoa, Yemen, Zaire,
  Zambia. The total number of countries covered is
  181.

14
Evidence from Logit Models of the Adoption
Decision

• Parsimonious models restricted to economic
  determinants
• GDP per capita shows the strongest influence
• Including the emulation proxy wipes out the other
  economic variables, though we carry a few
  insignificant variables in subsequent rounds of
  testing as a robustness check.
• Expanded Models
• Crisis experience proves significant
• Government ownership/privatization does not
• Most variables representing political
  power-sharing and social capital prove
  significant, but ICRG measures of corruption and
  law and order do not.
• Effect of emulation is stronger and effect of per
  capita GDP is weaker when we exclude countries
  with small populations or introduce continent
  dummies.

15

• Robustness tests employ three kinds of
  statistical models and alternative indices of
  political and cultural influences.
• Qualitative conclusions about the separate
  effects of emulation and other types of
  determinants prove robust.

16
(No Transcript)
17
Evidence from Weibull Hazard Models

• As represented by the evolutionary parameter,
  emulation is even more significant than per
  capita GDP.
• Crisis experience and political power-sharing
  proxy are also significant

18
Hazard models of deposit-insurance adoption

• This table estimates the hazard rate of adopting
  explicit deposit insurance over the period
  1934-2003. The model considers the adoption of
  deposit insurance as a transforming event. The
  endogenous variable is the number of years
  between 1934 and the adoption date. The
  coefficients reported are the logarithms of the
  underlying relative-hazard coefficients. The
  number of transformations is the number of
  countries that adopted deposit insurance during
  the observation period.

19
EDIS Design Features Shown Empirically to Inhibit
Risk-Shifting

• Enforceable Coverage Limitations
• Coinsurance
• Risk-rated Premiums
• Private Involvement in System Management
• Compulsory Membership
• Funding of Contingent Liabilities Entails Clear
  Right to Levy Ex Post Assessments on Insured
  Institutions

20
Evidence from Heckman Two-Step Models of Design
Features

• Heckman Lambda significantly and favorably
  influences the adoption of
• Four individual features the ratio of coverage
  to GDP per capita compulsory membership
  coinsurance absence of permanent funding.
• A multifeature moral-hazard exposure composite
  extracted via principal-component analysis.
• Models for private participation in EDIS
  management, private funding, and coverage for
  foreign or interbank deposits show a mixed and
  marginal influence for per capita GDP and the
  two crisis variables.

21

• GDP per capita, emulation, and crisis dummy show
  a positive (i.e., perverse) and significant
  influence on design. Post-crisis adoption is
  positive (i.e., also perverse) in the four cases
  where it is significant.

22
Table 10 Heckman two-step selection model for
adoption of deposit-insurance design features