Political influences on the costs of banking crises in emerging market economies: testing the U-shaped veto player hypothesis

Abstract

While there has been considerable research on the consequences of financial crises, there has been little empirical research on the possible effects of the role of domestic political institutions that influence a government's ability to implement crisis management policies. This paper investigates the impact of domestic institutions, characterized by a U-shaped veto player framework, on the output costs of banking crises. The analysis extends MacIntyre's qualitative study (2001) of the relationship between veto players and policy risks in the Asian financial crises. For a large sample of emerging market economies, we find support for McIntyre's hypotheses that both too few and too many veto players are associated with greater costs of banking crises.

Keywords:

• veto player
• output cost
• banking crises

Acknowledgements

We have benefited from financial assistance from the Freeman Foundation Program in Asian Political Economy at the Claremont Colleges and the National Science Foundation and the helpful comments of Arthur Denzau, Yi Feng, Jennifer Merolla, and Clas Wihlborg.

Notes

 1. A recent exception with respect to currency crises is Shimpalee and Breuer (2006).

 2. During the Asian financial crises, substantial liquidity support and deposit guarantee policies could not stop bank runs in several of the crisis-hit countries because of lack of credibility of governments in following financial reform plans that had been publicly announced (see Lindgren et al. 1999, 18–21).

 3. A U-shaped relationship has also been formed between the degree of concentration in labour markets and the inflation–unemployment trade-off. In this case, however, intermediate levels of concentration give the worst outcomes, with both highly decentralized competitive markets and highly concentrated corporatist structures giving rise to less inflationary pressure.

 4. Stasavage (2002) also tests for a non-linear impact of checks and balances by entering log(checks), where checks is the variable from the Database of Political Institutions (DPI), into the private investment regression. The positive significance of log(checks) indicates that higher checks lead to higher private investment but at a diminishing rate.

 5. According to Falaschetti's results, intermediate values of veto players when proxied by a political constraint index are between 0.2 and 0.6 (out of a 0–0.8 scale), and those when proxied by the checks variable from DPI is between 4 and 12 (out of a 1–14 scale). The undesirably high and low level of veto powers are values outsides these ranges.

 6. For instance, the rehabilitation criteria (by raising the capital) to resolve the financial problem in 10 of the weakest financial companies at the beginning of 1997 were relaxed since ‘several senior members of government had interests in some of the 10 targeted institutions and used their leverage within the coalition to veto the actual implementation of the tough measures’ (MacIntyre 2001, 98).

 7. This paper does not recognize any expansionary output effect of banking crises. While in the currency crisis literature the currency crisis accompanied by output expansion is possible due to, for example, the J-curve effect. There is, however, no clear theoretical explanation of banking crisis causing the short-run output expansion.

 8. The estimation results using the Ordinary Least Square (OLS) are reported as well.

 9. The ratio of current account to GDP and the ratio of private credit growth to GDP are entered in the regressions with averaged two-year pre-crisis periods while real GDP per capita, real GDP growth rate, and money supply to reserve are entered in the regressions with one lag of the crisis year.

10. The dates of currency crises are from Bordo et al. (2001).

11. On the output costs of twin crises see the analysis and references in Hutchison and Noy (2005).

12. Argentina, Bangladesh, Brazil, Chile, Colombia, Costa Rica, Ecuador, Egypt, Ghana, Hungary, Indonesia, Jordan, Kenya, S. Korea, Malaysia, Mexico, Nigeria, Paraguay, the Philippines, Russia, Singapore, Sri Lanka, South Africa, Thailand, Turkey, Venezuela, and Zimbabwe.

13. The year following the crisis year is included since occasionally the effect of crises to the real economy takes time for financial shocks to transfer to the real sector.

14. See Angkinand (2005) for details of the estimation of output losses associated with crises, and advantages and disadvantages for each estimation methodology.

15. IMF (1998) uses averaged three-year pre-crisis growth rates while Bordo et al. (2001) use five-year average to calculate trend growth rate. According to Mulder and Rocha (2001), however, the use of different pre-crisis periods to calculate trend growth rate does not result in the significant difference in the calculated magnitude of output losses.

16. It has been argued that for many Asian countries the growth rates prior to the 1997 crises would not have been sustainable in any event. Such issues are an important area for future research.

17. For GROWTHLOSS, the Mexican crisis in 1981 is an outlier observation and excluded in all regressions.

18. The correlation between LEVELLOSS and GROWTHLOSS is 0.43.

19. In column (2), Table 3, when the veto player variable is proxied by polconv, only the linear term of polconv is significant at the 10% level with a p-value of 0.093 while the squared term of polconv has a p-value of 0.126.

20. For example, the predicted change in the magnitude of output losses in GDP level for a change in the number of checks from 1 to 2 is −61.212 + (2 × 9.032) × 1.5 = −34.12.

21. Results indicate that the veto player variables have more significant impact on GROWTHLOSS than LEVELLOSS. However, these could be partly due to the estimations of output losses. As noted by Hoggarth, Reis, and Saporta (2002) and Mulder and Rocha (2001), GROWTHLOSS may underestimate the severity of crises (see the Data section).

22. Polconv of the Philippines reduces from 0 in 1981 to 0.7 in 1998, where 0 out of 0–1 scale indicates no political constraints.

23. The statistics for identifying outliers or observations with large discrepancy are Studentized Residuals and for identifying influential observations are DFBeta, which measures the change in individual coefficient caused by dropping a single observation. Cook's Distance (Cook's D) is statistics for detecting outlying observation that has influence on the coefficients since it is composed of a discrepancy and a leverage term.