The diffusion of financial supervisory governance ideas

ABSTRACT

Who is watching the financial services industry? Since 1980, there have been multiple waves of thought about whether the ministry of finance, the central bank, a specialized regulator or some combination of these should have supervisory authority. These waves have been associated with the convergence of actual practices. How much and through what channels did internationally promoted ideas about supervisory ‘best practice' influence institutional design choices? I use a new dataset of 83 countries and jurisdictions between the 1980s and 2007 to examine the diffusion of supervisory ideas. With this data, I employ Cox Proportional Hazard and Competing Risks Event History Analyses to evaluate the possible causal roles best practice policy ideas might have played. I find that banking crises and certain peer groups can encourage policy convergence on heavily promoted ideas.

KEYWORDS:

• Policy diffusion
• event history analysis
• financial supervision
• crises

ACKNOWLEDGEMENTS

Thank you to the LSE PSPE research seminar, Joseph Chekley, Kristina Gandrud, Charles Goodhart, Simon Hix, Chris Jackson, Jouni Kuha, Eric Neumayer, Guillermo Rosas, Cheryl Schonhardt-Bailey, Vera Troeger and Kevin Young for helpful comments and insights. Earlier versions of this paper were presented at the 2010 ECPR Regulation in the Age of Crisis conference and the 2011 EPSA Annual Conference. Full replication data and code can be found at <http://bit.ly/Qz7KHt>. Please note that the international country risk indicators are made available for replication only. They should not be distributed.

Notes

1. Given the space constraints, I focus on changes to the de jure actors who supervise and look at the period up until the recent crisis. It is admittedly also important to look at de facto governance, regulatory changes and the economic outcomes of supervision choices. Hopefully, future studies will examine the degree to which my conclusions can be generalized to these areas. For recent work examining the economic consequences of financial supervisory governance, see Barth, Caprio Jr. and Levine (2004, 2006), Eichengreen and Dincer (2011), Jordana and Rosas (2011), Masciandaro, Panisini and Quintyn (2011) and Quintyn and Taylor (2003).

2. Financial supervision broadly encompasses banking, securities and insurance. However, for simplicity, this paper focuses on banking and securities both in its discussion and empirical analysis.

3. My use of the term, ‘SEC model', refers not only to the securities regulator, but also the fact that some other body is regulating the deposit banking industry. It describes supervision in both sectors.

4. See Blyth (1997: 236) and Yee (1996) for further details of this critique.

5. The distinction between MoF and CB supervision may be superficial if the CB is not independent. However, I focus on de jure supervision because of the difficulty of measuring the actual supervisory independence for the wide range of countries in my sample. A number of measures have been used for monetary policy independence (famously, Cukierman, Web and Neyapti, 1992), but equivalent measures are not widely available for financial supervision.

6. Information was not widely available on supervisors earlier than this period. Data was gathered by the author using numerous sources detailed in a data appendix available upon request. The author is indebted to Quintyn, Ramirez and Taylor's (2007) work. In many ways, the current sample is an expansion of their sample. An ‘Other' category, which included up to six jurisdictions, was collapsed into the CB/MoF category.

7. The list of sources consulted in the creation of this dataset can be found at <http://bit.ly/Qz7KHt>.

8. It is important to understand the processes behind the creation of these ideas and the reasons why they were promoted. I touch on some of these issues in this paper. However, an in-depth study of these issues is beyond the scope of this paper. For an example of what this research might look like, please see Chwieroth (2010) for an examination of how ideas are developed and come to be promoted by IMF staff.

9. Due to a limited number of CB-only countries and the difficulty of separating CBs from MoFs when the CB is not clearly independent, these two categories are combined throughout the paper.

10. Focusing on official English-language names clearly ignores non-English-language name convergence. Spanish-speaking countries, for example, rarely give official English-language names to their financial supervisors (or have English-language version websites). This would certainly be an interesting area of further study.

11. Coding done by the author.

12. Much of the literature and documents from government and international organizations on financial supervision uses the term, ‘independence' (see Goodhart and Schoenmaker, 1997; Masciandaro, Quintyn and Taylor, 2008). This can be a confusing term since the authors are often referring to making the supervisor independent of a possibly already independent CB. To avoid confusion, I use the term, ‘specialized', instead. See below for a further discussion.

13. Despite the previous moderate SEC model adoption trend, it was so minor that Quintyn, Ramirez and Taylor could argue in 2007 that the attention given to supervisory governance over the past decade was new:

The discussion about independence, accountability, and more broadly, governance of financial sector regulatory and supervisory agencies…is still relatively new…. Previously, the organizational structure of supervision had been widely viewed as a relatively unimportant issue, both in theory and in practice, but this perception changed dramatically about a decade ago (2007: 3).

14. Please note that Goodhart and Schoenmaker (1997) discussed both the potential positive and negative consequences of specialized supervision. However, this piece is often quoted in later research as advocating a unified SR.

15. From an interview conducted by the author in Beijing with Zhixiang Zhang on 11 March 2010.

16. The recommendations' timing closely corresponded to the increasing de jure prevalence of central bank and regulatory independence in other areas (see McNamara, 2002; Jordana and Levi-Faur, 2005).

17. Goodhart and Schoenmaker actually voiced considerable scepticism about the appropriateness of the term, independence, for financial regulation. However, in many later works, particularly by IMF staff writers, their 1997 piece is referenced as being a founding document of the supervisory independence idea (e.g., Quintyn, Ramirez and Taylor, 2007).

18. This is especially true in Northeast Asia. Staff sharing through secondments and agency revolving doors (with both the CB and MoF) was a common theme in interviews conducted by the author with policymakers and experts in China, South Korea and Japan in March 2010.

19. Clearly a number of questions could be explored stemming from this discussion. Primarily, why did the FSA model gain such wide support and usurp the SEC model? This might be a fruitful issue for further study.

20. It is common in diffusion studies to include numerous historical, linguistic and cultural variables. Not only do these usually highly correlated variables tend to produce meaningless coefficients (Schrodt, 2006) and suffer from validity issues (how do you dichotomously code ‘the religion' of a society that is almost evenly split between Christians and Muslims, for example), but exploratory descriptive analysis also indicates that these would not be strong predictors.

21. Initially, the regulatory capture literature (Stigler, 1971) seems a natural place to look for theories concerning financial supervision. Private sector capture was certainly a concern of those proposing supervisory separation from political actors (see Quintyn and Taylor, 2003). However, this doesn't appear to be likely to explain governance reform choices. If regulatory policy was already captured by the financial sector, why would they lobby to have it changed? Financial sector structure variables are included in the models partially to account for potential changes in the power of the sector, which might lead them to have more or less influence over governance choices.

22. Please note that the international country risk indicators are made available for replication only. They should not be distributed.

23. Covariates are omitted for simplicity.

24. We can use a number of PHA diagnostic tests such as residual-based approaches (Box-Steffensmeier and Zorn, 2001; Fine and Gray, 1999) and time interactions (Stata Corp., 2009, 214–5).

25. Only four countries in the entire 83 country sample had unified SRs before 1997. Sweden and Denmark unified multiple SRs in 1991 and 1987, respectively, so only they are included in the model Cox PH model of transitions from Multiple SR to Unified SR. Denmark, like all transitions made in 1987, was not ‘observed' by the model because the year, 1986, was not included due to lack of data availability. Honduras and Nicaragua both had unified regulators well before the beginning of the observation period, so they are not included in the models either.

26. Sometimes referred to also as hazard functions.

27. Cumulative incidence functions are the probability of observing the event of interest and not another event before a certain time, if it hasn't already happened, given certain values of the covariates. Formally: CIF(t|x) = Pr(T ≤ t and event type of interest |x) (modified from Stata Corp., 2009: 532).

28. The specific logarithmic base 10 transformation of the impact of crisis from the first crisis year to some year found by where t_c ⩽t _c0+5. The variable was standardized so that 0 signifies no crisis. Because of this, the crisis variable at t_c0 = −1.78533.

29. Plausibility was determined by examining descriptive statistics and peer organization documents.

30. China, Hong Kong, Japan, South Korea and Taiwan.

31. Denmark, Estonia, Finland, Germany, Iceland, Latvia, Lithuania, Norway, Poland and Sweden.

32. A prime example is found in the communiqué from their 1997 meeting (Council of the Baltic Sea States, 1997).

33. The procedure I used to create the dyadic datasets for finding the spacial effects was from Gilardi and Füglister (2008).

34. Asset diversity for firms with assets of at least US$100 million is calculated by

35. Pakistan and Venezuela, included in Laeven and Levine (2007), were not included in the analysis due to lack of available data on their financial supervisors.

36. Results from models with very highly correlated and insignificant variables are not shown. These are discussed in the table captions.

37. That is, statistically significant at at least the 5 per cent level.

38. To assess the imputation results, I ran the diagnostic test suggested by Honaker, King and Blackwell (2011) and implemented in Amelia II, including comparing observed and imputed variable densities and running models with over-dispersed starting values. These methods did not reveal any major anomalies in the imputed data used for this paper's analyses.

39. The estimated linear time-varying coefficients are made up of two parts, a non-time-varying β and a time-varying β(t). So, the coefficient is β + β(t). Various non-linear functions of time were also tried, but did not substantively change the results.

40. The GDP per capita variable was also negative and significant at between the 5 and 10 per cent significance level, depending on the model specification.

41. From a discussion with Charles Goodhart conducted on 5 October 2010.