• 沒有找到結果。

銀行監理制度對銀行績效的影響:以101個國家為例(2/2)

N/A
N/A
Protected

Academic year: 2021

Share "銀行監理制度對銀行績效的影響:以101個國家為例(2/2)"

Copied!
62
0
0

加載中.... (立即查看全文)

全文

(1)

行政院國家科學委員會專題研究計畫 成果報告

銀行監理制度對銀行績效的影響:以 101 個國家為例(2/2)

計畫類別: 個別型計畫

計畫編號: NSC94-2416-H-004-007-

執行期間: 94 年 08 月 01 日至 95 年 07 月 31 日

執行單位: 國立政治大學金融系

計畫主持人: 沈中華

報告類型: 完整報告

處理方式: 本計畫可公開查詢

中 華 民 國 95 年 9 月 6 日

(2)

Does the Financial Supervision Structure Affect

Bank Performance?

Chung-Hua Shen

Department of Money and Banking National Chengchi University

Mucha, Taipei, 116 Taiwan ROC TEL.: (02) 29393091–81020 FAX: (02) 29398004 email: chshen@nccu.edu.tw August 30, 2006

(3)

1

Introduction

In recent years the distinction between financial intermediaries in the bank-ing, securities and insurance sectors (hereafter the three sectors) have been blurred in many countries. With respect to financial intermediaries, for example, on November 12, 1999, US Congress passed the Financial Mod-ernization Act,1 which allows bank holding companies to affiliate within a financial holding company (FHC). This new financial conglomerate allows banking, insurance and securities operations to be carried out under the same roof, which up to then had been restricted to some extent. The new financial architecture has not only been confined to the developed countries. Many developing countries also gradually allow their banks to engage in pre-viously restricted non-bank activities. For example, the Taiwan authority passed the financial holding company law in August 2000, allowing the FHC to hold more than 25% shares of the above three financial sectors at the same time. This “blurring of distinction” also occurs among financial products and markets. Many new products share the features of deposits, annuity and securities simultaneously. In the Netherlands, for example, a mortgage combined with a unit-life insurance policy embodies the components of the three financial sectors (Zwet, 2003). In Taiwan, security-linked deposits or insurance-linked deposits are products combining deposit and securities or insurance. This blurring of distinction among financial institutions,

(4)

ucts and markets across the three financial sectors have become the norm rather than the exception.

Allowing a financial conglomerate to engage in the above three activities, or the existing financial institutions to issue hybrid products, challenges the existing supervision system, which is typically organized sector by sector.2 Under the “sectoral supervision” system, however, it is not easy to supervise the new financial conglomerates and products because of the “blurring of re-sponsibility”. For example, ambiguity as regards responsibility arises when a bank is in distress owing to its engagement in securities operations. The ”blurring of responsibility” argument is that none of the supervisors will be responsible for cross-sector behaviour, thereby endangering the finan-cial sector. Moreover, when one finanfinan-cial sector is in distress, the ensuing market turbulence will soon spill over to another sector if the supervisors or regulators do not actively co-operate with each other. This contagion effect becomes even greater owing to global financial integration. Recent Asian financial crises strengthen this belief. Furthermore, banking failures around the world in recent years have been common, large and expensive (See Demirg¨u¸c-Kunt and Detragiache, 1998; IMF, 1998).3 Hence, an effi-cient way to prevent that the “blurring of distinction” transforms into the

2

A majority of countries in the world have adopted sectoral supervision. This issue will be discussed shortly.

3

For example, as mentioned by Kaufman (2000), the loss from currency crises averaged 4.3 percent of the trend GDP and the loss from banking crises was even greater than in the currency crises, averaging 11.6 percent in all banking crises. Also, banking crises last 3.1 years on average, twice as long as currency crises.

(5)

“blurring of responsibility” is called for.

Four possible solutions have been raised to solve the blurring of distinc-tion (Zwet, 2003), namely, strengthening the co-operadistinc-tion between sectors (e.g. the Netherlands),4 adopting functional supervision (e.g., Australia), creating a single regulatory agency (e.g., UK), and asking the central bank to take charge of the three activities (e.g., Singapore). The most persuasive argument against the first approach is the above “blurring of responsibility”. Taylor and Fleming (1999) argue that just because co-ordination is difficult, costly and inefficient when banking, securities, and insurance business are regulated and supervised by different authorities, there is a need to unify them into a signal agency in order to reduce potential financial instability. Hence, the first approach is not widely taken after the recent financial crises The second approach has not been widely adopted probably because the prerequisite of it requires clearly defined responsibilities and objectives, which are often lacking in developing countries.

The recent tide of restructuring financial supervision seems against the fourth approach as some countries recently have removed the central bank’s role of banking supervision.5 Thus, though the majority of central banks still take the responsibility of banking supervision, the number is expected to decrease. The single most common model, however, is the central bank to be

4Note that the Netherlands is going to merge bank and insurance supervisors on April 1, 2004, see Mooij and Prast (2003), in line with the supervisory practice in Australia.

(6)

responsible for banking supervision (Llewellyn, 1999). Tuya and Zamalloa (1994) shows that in the 167countries, “banking supervision is conducted by the central bank in over 60 per cent and that the Western Hemisphere was the only region where the percentage declined to 50 per cent. Nevertheless, in over 80 per cent of Asian, African and Middle Eastern countries, banking supervision is a function of the central bank.”6 Whether the central bank should be given the power of supervision is a long lasting issue. As the main task of the central bank is to conduct monetary policy, the “optional tasks” of banking supervision becomes the center of policy debates (Di Noia and Di Giorgio, 1999). The main objection to allowing central bank to super-vise banks is the “conflict of interest” between monetary policy and banking supervision. The supporters stress the synergy of information between su-pervision and monetary policy.

Among the four approaches, the third one has attracted the most atten-tion probably because the new Labour government in the UK announced plans in 1997 to consolidate financial supervision in a separate and new agency. To integrate existing diversified regulators and supervisors, Tay-lor and Fleming (1999) argue that unified supervision can minimize conflict resolution and improve accountability. They refer to this unified supervi-sory agency as the “integrated supervisors”, in contrast to the conventional ”sectoral supervisors”. Countries, which change from other supervisory

sys-6

The number, however, needs to be interpreted cautiously as they take into account many very small states, see Di Noia and Di Giorgio (1999).

(7)

tems to this unified supervision, include UK, Japan, South Korea (hereafter Korea), Taiwan and Germany.

Though there appears to be a “trend” of adopting the unified supervision system,7 the discussion is not final. Academic discussion of the advantages and disadvantages of each supervisory system still abounds. Advocates of the integration suggest five advantages of adopting the single agency (Bri-ault, 1999). That is, integration creates economies of scale and scope, ef-ficient resource allocation, reduce conflict resolution, appropriate differen-tiation and better accountability. Opponents of the integration argue that a single regulator need not necessarily deliver these advantages. For ex-ample, Goodhart et al. (1998) point out that specialist divisions will exist even within a single agency, thus creating potential problems in communi-cation, information sharing, co-ordination and consistency. Also, Goodhart et al. (1998) and Taylor (1995) argue that the different objectives are better resolved at a political level rather than within a single regulator.

While there are heated discussions regarding the merits of each super-visory system, fewer empirical studies are conducted to verify this crucial policy issue. Most empirical studies focus on the role of the central bank in macro performance when the bank is also responsible for banking super-vision.8 Thus, there is a sizeable conceptual but relatively small empirical

7For only a handful of countries to adopt the unified supervision, it might not be called a “trend” in a strict sense.

8Goodhart and Schoenmaker (1995), Briault, (1997), Di Noia and Di Giorgio (1999), to name but a few.

(8)

literature that bears directly on this important policy issue.9 To the best of our knowledge, Barth, Caprio and Levine’s (2002) study is the first pa-per that empirically pursues a similar issue. They studied 107 countries and showed that countries with multiple bank supervisory authorities had lower capital ratios and high liquidity risk. This correlation is consistent with multiple supervisors engaging in competition in laxity. Also, the struc-ture of banking supervision appeared to have little correlation with banking profitability. They also found that countries whose central banks supervise banks tend to have more nonperforming loans but less liquidity risk.10

The aim of this paper is to study how an institutional choice, i.e., the degree of financial supervisory integration (FSI) affects bank performance. Our paper is similar to Barth et al.’s (2002) pioneer study but there are differences. Firstly, Barth et al. are concerned with the effect of single ver-sus multiple supervisors of banks on bank performance. This paper focuses on the effect of sectoral, partial and unified supervisors of banks, insurers and securities on bank performance. By ”partial supervision” is meant here that one agency supervises two sectors. Thus, a country may have adopted sectoral supervision (three supervisors of the three sectors) but could have either a single supervisor of banks (such as Argentina, Brazil, Thailand and ?) or multiple supervisors of banks (such as the US, and Taiwan before

9This is probably because collection of supervision information and bank data are both laborious and difficult.

10the time of writing this paper, we have discovered another similar paper. Masciandaro (2004) also pursues a similar issue as ours. Check.

(9)

2004). Also, in contrast to their use of a questionnaire to collect informa-tion of bank supervision, our supervisor informainforma-tion is based on Courtis’s (2002) book, which lists supervisors of each sector in 101 countries. We then re-group the countries into three categories: sectoral, partial and unified in-tegration based on the number of sectors that are supervised.11

Next, we also investigate the determinants that affect the choice of su-pervisory system. Are there common features of those countries adopting unified financial supervision? For example, why do some countries choose to adopt sectoral and some move to unified supervision? Anecdotal evi-dence shows that small countries tend to adopt unified supervision becasue of the cost saving. Furthermore, it is often argued that central banks in small countries tend to supervise all sectors, but there is no systematic ev-idence. Thus, is the scale (e.g., size and population) of a country crucial for its choice of system? We investigate these “scale effect” in determining the choice of supervisory system. We plan to answer these questions using a multinomial logit model (MNL). Once the choice of supervision is endoge-nously determined, simply conducting OLS regression to explore the effect of the financial supervision structure on bank performance may be biased. Our model involves both the determinant and performance equations, and hence becomes a simultaneous multinomial logic model (Lee, 1983).

Third, the role of the central bank as banking supervisor is stressed here.

11

Throughout the paper, sectoral and separated supervision are used interchangeably. Also, unified and integrated supervision are used interchangeably.

(10)

Previous studies on this issue mainly focus on how the central bank’s super-visory duties affect monetary policy.12 Di Noia and Di Giorgio (1999), using OECD countries data, find that banks seem to be more profitable when cen-tral banks supervise them. Also, bank incur higher costs and rely more on deposits than on? more sophisticated liabilities as a funding source. Their study, however, is based on basic statistical analysis without resorting to regression analysis. This may neglect the third variable effects from differ-ences across countries in financial system structures as well as differdiffer-ences in the timing and magnitude of their business cycle. This paper uses regression analysis by controlling GDP per capita, rule of law, religion and so on.

Fourth, bank level data is aggreaged into a country level data of each year. We collect bank level data of 101 countries from a data bank of BankScope from 1995 and 2001. While our original goal was to use more sample countries and a more systematic way to link? the responses of bank performance to institutional factors, the availability of data limit the scope of our paper? to some extent. For many small countries, bank level data is not available until the last two years or is only available on a discontinuity basis. We erase the whole row when data is missing. Thus, our data is an unbalanced panel.13

12

For example, Peek, Rosengren and Tootell (1999), who use confidential bank rating data for the U.S., find that information obtained from bank supervision helps the central bank to conduct monetary policy more effectively. Also, see Goodhart and Shoenmaker (1995), Di Noia and Di Giorgioi (1999) and Ioannidou (2002) for discussing the case when monetary policy and banking supervision are given to the same agency.

13

While we do not implement the fixed/random effect of the panel model to remove the country heteroscedasticities, our model contains several country dummy variables, which

(11)

Our cross-country study has the same methodological limitations as other cross-country studies. Firstly, supervisory change is a dynamic process but we can only study for one point of time (see Barth et al. 2002 for the same argument). Also, the information of supervision is employed only up to 2001. This implies that countries whose supervision system has changed after 2001 are not taken into account. This includes China’s new sectoral supervisory system from 2002, Germany’s unified supervision from May 1, 2002 (Sanio, 2003), the Netherlands’s integration of banking and insurance from April 1, 2004 (Mooij and Prast, 2003) and Taiwan’s single new agency to supervise three sectors from July 1, 2004 (Shen, 2003).

Secondly, no operational strategies of banks are being considered, which is subject to the criticism of ignoring banks’ micro factors. In our model, , only supervisory, institutional, governance and macro factors are taken into account to explain bank performance. This loses the important banks’ micro information, such as the strategies, sizes and locations of each bank. In other words, the link between bank performance and supervision could be contaminated by other factors, which can hardly be discovered in our model. This is the limitation of most cross-country studies, and ours is no exception. We hope that the large sample size can mitigate this weakness to some extent

The third limitation? is related to two econometric issues. One is the may mitigate this problem.

(12)

endogeneity problem that still exists because banking sector outcomes may influence the decision of supervisors, creating the simultaneous bias (Barth et al. 2002). While we use the instrumental variable approach to mitigate the bias based on the standard textbook solution, the asymptotic or small sample properties are not well-known in our simultaneous multinomial logit model. The other is the heteroscedasticity problem. The use of bank level data, though expanding our sample size and providing more information, also creates substantial heteroscedasticity in our model. The robust ? is used to reduce heteroscedasticities.

The remainder of the paper is organized as follows. Section 2 provides the background of supervision. In Section 3, a multinomial logit model is built up. Also, the response equation is briefly reviewed. Sections 4 and 5 describe the data and basic statistics and empirical findings, respectively. Concluding remarks are collected in Section 7.

2

Background of Financial Supervision

The choice of financial supervision is being widely discussed probably be-cause of the UK’s establishment of a single statutory body, the Financial Services Authority (FSA) in 1996. The establishment of a single European monetary authority also heats up the question of whether Europe needs cen-tral? European Financial Supervision (Di Noia and Giorgio, 1999). While the UK case drew much attention of the international financial press and

(13)

policymakers worldwide, it is not the first country which has created an in-dependent supervisory agency. The three Nordic countries (Norway, Sweden and Denmark ) have moved to integrate these different supervisory functions into a single agency in the early 1990s, hoping to increase the soundness of the financial sector.

The issue has soon become a global one owing to the re-emergence of financial turbulence, individual bank failures, and systematic crises in Latin America, Asia and Eastern Europe, as well as bank failures and near-failures in developed countries. It is argued that one of the reasons of this financial turmoil is probably that the sectoral supervisors are not used to co-operate with each other. Furthermore, the fact that financial conglomerates are allowed to be established further strengthens this belief. Thus, Japan (in 1998), South Korea (in 1999) and Taiwan (in July 1, 2004) adopted the in-tegrated supervision system or will do so following the Asian financial crisis. Germany also adopted a similar system to create an independent and unified supervisory agency on March 1, 2002. In contrast to the “trend” of adopt-ing unified supervision, Australia, however, adopted functional supervision on 1997 by combining banking, insurance and securities under conduct of business. Banking and insurance supervision in Australia has been inte-grated for the purposes of prudential supervision.14 The Netherlands will

14Whether Australia adopts unified supervision is controversial. While Abrams and Taylor’s (2000) and Llewellyn’s (1999) studies include Australia as having adopted unified supervision, we find its unified supervision includes only securities.

According to Courtis’s (2002) data, securities activities in Australia are supervised by an ”old” supervisory body, i.e., ASIC (Australia Securities and Investment Committee),

(14)

adopt a similar concept as Australia’s on April 1, 2004 (Mooij and Prast, 2003). Contrary to the “tide”, China, on 2002, adopted sectoral supervi-sion by removing the supervisupervi-sion of insurers and securities from the central bank. Different countries seem to have different preferences. The concerted opinion is that not one model of regulatory structure will be appropriate for all countries (Abrams and Taylor, 2000; Briault, 2002) Accordingly, it is interesting to investigate what the determinants of financial supervision are.

Which supervisory system a country should adopt is another question. The natural methodology to answer this question is to study the advantages and disadvantages of each system. Because we have briefly referred to this in the introduction and because these discussions have been fully elaborated by Briault (1999, 2002), Taylor and Fleming (1999) and Abrams and Taylor (2000), we will skip the discussion unless it is necessary.

Another method is to ask what the objectives of supervision are and then to examine whether the existing supervisory systems achieve these goals or not. Typically, three objectives of financial supervision are raised in the lit-erature: systemic stability, financial soundness of individual institutions and consumer protection (Bikker and Lelyveld, 2003; Schoenmaker and Wierts, regardless of which sectors are engaged in the securities business. Simply put, “func-tional supervision” has been adopted in Australia for securities activities. The banking and insurance sectors, however, are supervised by a new agency, APRA (Australia Pru-dential Regulation Authority). Thus, supervisory systems have not yet been completely integrated.

(15)

2002).15 In a sectoral supervision system, each sector is responsible for all three objectives, whereas in a unified supervision system, a single supervi-sor takes charge of the three. Partial supervision, which is the system in between the two, is more complicated. It could be a functional supervision system, which has two supervisory agents. One is for prudential and the other is for conduct-of-business supervision. The Netherlands and Australia seem the only two countries that clearly announce the adoption of the func-tional supervision system. The partial supervision system could also consist of just two supervisory agents, where one supervises two sectors and the other supervises the remaining one. Because it is not possible to distinguish these two types of partial supervision, they are simply termed the partial supervision system here.

Empirical studies to investigate the achievement of these three objec-tives, however, are even more difficult. This is because the operational definitions of the three objectives are not easy to be found in the litera-ture. Barth, Caprio and Levine’s (2002) work could be ranked under? the assessment of the second objective of financial soundness. This is also the methodology used in this paper.

Our study stresses the role of the central bank. Some researchers, how-ever, focus on the effects of monetary policy when central banks are also

15The four statutory objectives in the UK’s supervisory system is to maintain confidence in the financial system, prompt public understanding of the financial system, secure the appropriate degree of protection for consumers and reduce the extent to which it is possible for a financial services firm to be used for a purpose connected with financial crime (see Briault (2002).

(16)

responsible for banking supervision. This double role is found to have a non-pleasant impact on the effectiveness of monetary policy. For exam-ple, Heller (1991), Goodhart and Schoenmaker (1995) and Di Noia and Di Giorgio (1999) find that countries with central banks that have supervisory responsibility experience higher inflation rates. They interpret this as ev-idence supporting the conflict of interest argument. See Ioannidou (2002) and the references therein.

Studies of the double role of a central bank could also be made from the perspective of bank performance. The agency that conducts banking supervision must have clear responsibilities and objectives to be effective. Each such agency should posses operational independence and adequate resources. Di Noia and Di Giorgio (1999) claim that the combination of different responsibilities and objectives in one agency may result in weak banking supervision and negatively affect monetary policy. Their study is based on basic statistical data without regression-type studies. Hence, results may be contaminated by other factors. This paper fills this gap.

3

Econometric Model

3.1 Determinants of Financial Supervision Integration

The study of what determines a country to adopt a particular financial supervision is interesting. A multinomial logit (MNL) model is used to explore the determinants of financial supervisions. MNL is intended for

(17)

use when the dependent variable takes on more than two discrete outcomes with no natural ordering; this is the case when the values assigned to the dependent variable are arbitrary. In the case of our supervision (F SIi =

1, 2, 3) , for example, a country takes the value of 1 if the country adopts sector supervision and 2 if the supervision includes any two sectors, and 3 if it is unified supervision.

Determinant Equation

P rob(F SIij = j)

= F ({[a1j+ a2jSCALEi][a3j + a4jDCi+ a5jLDCi+ a6jCEEi]}(a7j+ a8jCBi)

+a9jBank Activity Restrictions + a10jGovernancei

+a11jSupervisory Power + a12jBank Share Heldi+ εij) (1)

i = 1, 2, . . . , N, j = 1, 2, 3

= F (AjXi+ εi) (2)

where i is the ith country, j is the choice of jth supervision system, aij is

thus an unknown coefficient of ith variable in jth choice, N is the number of countries used in this paper, εij is the error and F is the logistic function.

We could rewrite equation (1) as (2) to reduce notation burden where Aj is

the vector of coefficients aij and vector Xi is the collection of explanatory

variables. The maximum number of countries is 101 but we have another small subset of samples. That is, when variables of Bank Activity Restric-tions are used, our sample countries are reduced to 45, which is the number

(18)

of countries surveyed by Barth et al. (1998). The sample size is further discussed in the next section.

Our dependent variable F SI1, F SI2 and F SI3 are the dummy variables denoting whether a country adopts no integration, partial integration and full integration, respectively, i.e., F SI1 is equal to one if no integration and zero otherwise; F SI2 is equal to one if partial integration and zero otherwise; F SI3 is equal to one if full integration and zero otherwise; For simplicity, we erase the subscript i if no confusion arises.

Our explanatory variables are discussed in sequence.

Central Bank. Whether a central bank should be responsible to su-pervise bank is a controversial issue in both bank supervision and monetary policy (Giddy, 1994; Goodhart and Shoenmaker, 1995). It gives rise to the conflict of interest between the two goals. Term CB is a dummy variable, which is equal to 1 if the central bank is in charge of bank supervision and zero otherwise.

Scale of Country. It is also argued that a small country tend to adopt the unified supervision to have economics of scope. Taylor and Fleming (1999) have suggested that a small transition or developing countries should adopt an integrated agency. They also mention that more detailed consid-eration of the types of governance are needed. Along with this claim, a small country tends to allow central bank to supervise banks. Singapore may be the case, whereas US is the opposite one. The three scale (SCALE)

(19)

variables considered are proxied by GDP per capita (GDPper), population (POPULA) and size of a country (SIZE). One would expect that the smaller the scale of a country, the higher the probability that she will adopt a unified supervision.

The explanatory variables related to scale variables are the development of a country. We consider whether a country is a developed country (DC), a less developed country (LDC) or a Central Eastern European countries (CEE) to capture the features of the development of a country. All least developed country (LLDC) are excluded,16 thus, the sample may not be

dominated by many very small or very poor countries, which have barely financial activities. Terms DC or LDC are the dummy variables and are equal to unity if a country is a developed or a less developing country, respectively and zero, otherwise.

Our model considers the interactions among CB, DC/LDC or SCALE. It is interested to knowing whether a small less developed country, when its central bank being responsible for bank supervision, increases the prob-ability of adopting the unified supervision or not. The concept expressed in mathematical way is

∂F SI

∂CB ∂LDC ∂SCALE = coefficient · SCALE × LDC × CB

In estimation, only one coefficient of SCALE × LDC × CB is obtained. Other arguments can also be tested in our model.

16

The United Nations currently designates 49 countries as the least developing countries, which are not used here.

(20)

Bank Activity Restriction Variables. These variables are the re-striction of banking activity in security, insurance and real estate. As men-tioned in the introduction, one of the argument that the unified supervision should be taken is that the blurring activities of these three industries. Thus, if a country does not allow her banks to engage these activities, she has less intention to adopt the unified supervision. The bank restriction is proxied by three index variables, which are restrictions on banks’ activities on se-curities (Bank-S), insurance (Bank-I) and real estate (Bank-R) activities. They are discrete variables from 1 to 4 denoting unrestricted, permitted, restricted, and prohibited for banks to engage in the above, respectively. Because only 66 countries are surveyed by Barth, Caprio and Levine (2001) adding theses variables into the model downsizes the number of countries used. 17 The higher the number, the more restrictive of the country is. We expect negative coefficients since they decrease the probability of adopting unified supervision.

Governance. Governance is the governance of government (GoodGov), including the broad rule of law (GoodGov) compiled by La Porta, Lopez-de-Silanes, Shleifer and Vishny (1998, LLSV), While the scores in LLSV’s are

17Barth et al. (1998) use only 45 countries to examine the degrees of restrictions on banking activities using cross-country data and find that banks with more diversified powers are less likely to suffer a banking crisis. However, they also mention that this result may be sensitive to other components of the regulatory environment, which is their omitted variables. It may be that countries that allow broader powers to bank may have higher capital requirements. Shen and Chang (2002) use the similar approach and find the rule of law in La Porta et al. (1998) is crucial in affecting the bank performance when banks are restricted to engage in securities.

(21)

based on 1995 survey, academic studies regarding governance still refer to this source probably because governance structures are relatively stable for a country. The six indices used in this paper is rule of law, efficiency of judicial system, corruption, risk of expropriation, risk of contract repudiation and accounting standard. The scores are ranged from 1 to 10 with a higher score indicating a better governance. Barth et al. (2001) also define a Good Government Index similar to ours. Theirs is the sum of law and order tradition, degree of corruption and risk of expropriation by the government. Supervisory Power. The Supervisory Power variable, which con-tains the official supervisory power index (OSPI) and private monitor index (PMI), taken from Barth et al. (2002). The OSPI measures the extent to which official supervisory authorities have the authority take specific actions to prevent and correct problem. It includes three components, i.e., prompt correction action, restructuring power and declaring insolvency power. The PMI measures private-sector monitoring function. It includes four compo-nents, i.e., certified audit required, rated by international rating agencies, explicit deposit insurance scheme and bank accounting transparency. While Barth et al. (2002) have provided eight banking related governance indices, we only take OSPI and PMI because these two measures are the most rele-vant indices to our issue.

Shares Held by Government and Foreigners. Barth, et. al. (1998) use government ownership data from Bankscope and find that greater

(22)

gov-ernment ownership is generally associated with less efficiency and less well-developed financial systems. Thus, the percentage of shares of banks held by government or by foreigners may also have the relation with the supervi-sion system. The use of BankScope, however, has a drawback because the coverage of the banks is not complete. Thus, Barth et. al. (2002) re-collect data from national regulatory agencies, of which the data is used here. This makes data cover all banks and the definition of government owned or foreign owned are consistent across countries.

It is worth noting that not all coefficients are estimatable in the MNL model. If there are N choices in the system, only N − 1 choices’ parameters are estimatable, whereas the remaining one is selected as a benchmark. The explanation of the estimated coefficients are all relatively to this benchmark, which is chosen arbitrarily.18 We select the sectoral supervision ( j = 1) as the benchmark and hence, two sets of coefficients, which are A2 for FSI =

2, and A3 for FSI = 3 are obtained. The MNL is estimated by using the

following logistic function

P r(F SIi = 1) = 1 exp(A2Xi) + exp(A3Xi) (3) P r(F SIi = 2) = exp(A2Xi) exp(A2Xi) + exp(A3Xi) (4) P r(F SIi = 3) = exp(A3Xi) exp(A2Xi) + exp(A3Xi) (5) We can rely on odds ratios for a better understanding of the coefficient signs

18

We must also recall that the possibility of using the estimates in this manner relies on the validity of the independence of irreverent alternatives (IIA) assumption: the inclusion

(23)

and magnitudes. For example, the relative probability of selecting unified supervision with respect to a sector supervision would be

P r(F SI = 3)

P r(F SI = 1) = exp(A3Xi)

which means that we compare probabilities with respect to a sector super-vision decision. The coefficients of A2 are similarly obtained.

3.2 Bank Performance Under Different Supervisions

We investigate the relationship between bank performances and financial integration. The bank performance (BP) measure is based on CAMEL,19 includes tier 1 capital ratio (Tier 1), problem loan/total loan ratio (NPL), problem loan/loan loss reserve (Cover), return on asset (ROA), liquid as-set/total deposit (Liquid). We do not consider return on equity since it produces almost the same results as ROA. We also do not consider the proxy for management. The net interest margin, which is also often used in proxing earning, is also not used here because it is often contaminated with other factors, which are not directly related to profit.20 These bank micro data are taken from BankScope from 1995 to 2001. The number of banks taken from each country will be discussed shortly.

19CAMEL denotes Capital, Asset, Management, Earning and Liquidity. 20

Ho and Saunders (1981) and Saunders and Schumacher (1997) suggest that the ob-served net interest margin contains the three market imperfections, i.e., the implicit in-terest expense, the opportunity cost of required reserve and the default risk premium.

(24)

Performance Equation

BPit = [(b0+ b1CBi) (b2+ b3DCi+ b4LDCi+ b5CEEi)] (b5F SI1i+ b6F SI2i+ b7F SI3i)

+b8Governancei+ b9Financial Structurei+ b10Macroi

+b11Law Origini+ ηit (6)

= (β10+ β11Zi)F SI1i+ (β20+ β21Zi)F SI2i+ (β30+ β31Zi)F SI3i

+β4Wi+ ηit (7)

where the second equality denotes a short hand to re-write the model, i.e., (β10+ β11Zi) = (b0+ b1CBi)(b2+ b3DCi+ b4LDCi) and similar definition for

other transformations. Term Wi is the vector of the remaining explanatory

variables and β4 is the corresponding coefficient vector.

The explanatory variables, which have been used in model (1) are not repeated here. Then new explanatory variables are accounted for as follows. Financial Structure includes bank-based and market-based variables, and the former contains bank asset/GDP (BKasset) and claims to private sector by depository monetary bank/GDP (lending), and the latter contains market capitalization/GDP (MktCap) and traded value/GDP (TradeV/GDP). See Beck, Demirg¨u¸c-Kunt and Levine (2001) for the distinguish of bank-based and market-bank-based economies. They find that (not finish????)

Macro is macro variables includes GDP growth rate and inflation rate. The detailed definitions and sources of variables are explained in the next section.

(25)

There is an econometric issue here. The estimation of (7) yields a biased estimator because the financial supervision is endogenously determined in our system. The corrected method is easier in the simple logit model than in the MNL. Lee (1993) suggests a two-step estimation to correct the biases. The first step is to estimate the determinant equation (1) to get

E(ηi|F SIj = j) = σjρjφ{Φ−1[F (AjXi)]}/F (AjXi), j = 1, 2, 3

= σjρjλ

where φ and Φ are the standard normal density and standard cumulative normal density functions, respectively, σj is the standard deviation of

residu-als of equation (3) ∼ (5), ρj is the correlation between determinant equation

j and performance equation and λ = φ{Φ−1[F (AjXi)]}/F (AjXi), where the

λ is referred to as Miller ratio.

The second step is to add the estimated λj into the performance equation

(??) as

E(BP |F SI = j) = βj0+ βj1Z + σjρjλ1+ β4W j = 1, 2, 3

Putting the three equations

BP = (β10+ β11Z + γ1λ1)F SI1 + (β20+ β21Z + γ2λ2)F SI2

+(β30+ β31Z + γ3λ3)F SI3 + β4W + η (8)

The method is OLS but the standard deviation is obtained by the heteroscedasticity-consistent of Newy-West method.

(26)

Our Miller ratio changes based on the specification in determinant equa-tion we use. Also, the consistency between two models are taken. That is, when bank restricted variables are employed, the same 46 sample countries are used in two equations.

4

Data Description and Basic Statistics

4.1 Two Data Sets of Sample Countries

Table 1 list the mnemonics of variables as well as their definitions and sources. Not all explanatory variables are available for all sample coun-tries. In fact, we have two sets of data based on the sample countries used. The first set uses the 101 sample countries when regressors contain variables such as POPULA, SIZE, LDC, DC and INFLATION because these vari-ables are available for the whole sample countries. It is worth noting that we have only 98 countries when GDP per capita is used because the variable of Bermuda and Cayman Island, and Netherlands Antilles are not available in IFS. We, however, still use the term of the whole sample. Two bank level data of ROA and LIQUIDITY are also available for these 90 countries. This is our benchmark sample.

The second set data is a small subset of the whole countries. When bank restrictions are used, our sample countries are reduced to 66 because this is the countries’ number used by Barth et al. (1998). Thus, whenever Bank-S, Bank-I or Bank-R are used, the subset of sample countries is employed. It

(27)

is worth noting that when both LLSV’s governance and bank restrictions are used, the sample size is reduced further to 41, which is the intersections of two types of variables. Table A2 reports detailed summary of this subset variables.

When dependent variables NPL and COVER are used in performance equation, the number of banks are also reduced. When number of banks in a country reporting the ratios of NPL and COVER less than 10%, we take our the country from our sample. Table A3 reports the average of bank level data over five years.

4.2 Basic Statistics Financial Supervision

Table 2 lists countries’ names of the three supervisory systems of 101 coun-tries. In each system, we further divide them into whether central banks are responsible for banks’ supervision or not. The sectoral supervision includes 56 countries, of which 47 countries’ central banks are also responsible for banks supervision and 9 countries are not. There are 35 countries adopt partial supervision. Among them, 18 countries which have one supervisor for banks and insurances,2114 countries which have one supervisor for banks and securities, and only 3 countries which have one supervisor for insurance and securities. Hence, banks and insurances tend to share one supervisor in the partial supervisory system, whereas securities and insurance have the

21Paraguay is unclear in our case. Its central bank is responsible for regulation but it has an independent supervision for bank supervision.

(28)

least possibilities. Worth noting that though we have three different types of partial supervisions, we do not distinguish them in the following regres-sion analysis due to the losing degree of freedom if we do so. The unified supervisory includes 10 countries but 4 are supervised by their central banks and 6 are supervised by a newly established agencies.

In sum, countries adopting sectoral supervision and their central banks supervise banks outnumber other cases here. Tuya and Zamalloa (1994) reach the similar results as ours, however, their sample contains many small countries (like Sao Tome, Myanmar, Vanuatu etc.). We only consider devel-oped and less develdevel-oped countries and, hence are free from the comments made by Di Noia and Di Giorgio (1999).22 Banks and insurances are easily to be supervised by the same supervisor. The “trend” of newly adopting the unified supervision occurs only in OECD countries with the exception of Taiwan.

The changes of the supervision systems in recent 5 years are more dras-tic than the past three decades. Our classification is based on the informa-tion available before 2001, but makes a note in Table 2 whenever there are changes. This does not affect our estimation because the bank level data also ends before 2001. Also, the impact of institutional change on banks are a gradual process.

Table 3’s panel A further analyzes the countries’ features and also

dis-22

(29)

cusses the number of developed and less developed countries in each system. Averages of the three scale variables, GDP per capita, population and size are also reported. There are 37 LDC adopt sectoral supervision, 25 adopts partial and only 5 adopts unified supervision.23 There are 50 LDC that give the power of bank supervisions to their central banks, but only 8 for DC. Among the 50 LDCs, 36 countries adopt sectoral supervision but only 7 out of 23 DCs adopt it. Out of 12 CEE countries adopt sectoral supervision and 9 of them ask the central bank to take charge of bank supervision. None of CEE countries adopt the unified supervision.

In sum, LDCs have the largest number of adopting sectoral supervision because the large base. CEEs are mostly likely to adopt sectoral supervi-sion with the central bank supervising banks. DCs are equally spread for the three systems but are getting more and more likely to adopt unified supervision.

Table 3’s panel B discusses the scale with the supervisory system. We report the medium, average, standard deviation and number of countries. The medium GDP per capita of sectoral, partial and unified are $3,139, $5,120 and $21,673, respectively. It is found that rich countries tend to adopt the unified supervision. Also, based on the populaton, the mean of three supervisory systems are 78.59, 13.92 and 23.68, respectively. The mean of them are 1294.52, 884.99 and 184.56, respectively based on the size. Thus,

23

If Korea and Singapore are excluded from LDC, then there are only three LDC coun-tries adopting the unified supervision.

(30)

small countries (based on population and size) tend to adopt partial and unified supervision, supporting the “small country effect”. No other patterns can be highlighted because of large variations across countries. Except for GDP per capita in the unified case, standard deviations are overwhelmingly larger than the mean. Thus, no conclusion can really be drawn because of the large heteroscedasticities.

Table 4 presents the five bank variables under three supervisory system. To our surprise, ROA in the first two supervisory systems are larger than the last one, but the standard deviation is large. NPL, on the contrary, is higher in the sectoral supervision, followed by partial and unified supervi-sion. Cover in our definition should be the lower the better. It is almost the same for the three sector but mean is largest for the first sector. Tier 1 is largest for the sectoral supervision.

5

Empirical Results

5.1 Determinant Equation

Tables 5∼8 present the estimation results of using determinant equation. The absolute t-values are in parenthesis. Because FSI = 1 is used as the benchmark, explanations of estimated results under the columns FSI = 2 and FSI =3 are relative to this benchmark. Also, we attempt different spec-ifications, denoted as 1A, 1B and 1C to examine the robustness. Variables such as CB, DC, LDC and CEE are our core variables and are not removed

(31)

when the specification is changed. We refer to the “robust factors” if the variables are insensitive to different specifications and “fragile factors” if re-sults are sensitive to different specifications. Sometimes, we use “marginal” robust factor to describe those condtions which are insensitive to the most, but not all specifications. Similarly, “marginal” fragile factors which are sensitive to the most, but not all specifictions.

Table 5 presents the estimation results with the consideration of the whole sample countries. Model 1A is the simplest model in our specifica-tion, including only five explanatory variables, i.e., CB, DC, LDC, CEE, GDEper and POPULA. Coefficients of CB, which are –1.954 and –1.973 in FSI =2 and FSI =3 equations, respectively, are highly significant. It sug-gests that if a country’s central bank is in charge of banks supervision, it has a high probability to leave away from partial and unified supervision and to adopt sectoral supervision. In short, those countries which adopt sectoral supervision also tend to use CB to supervise banks. Coefficients on DC are insignificantly negative regardless of FSIs, suggesting that the factor of DC has no impacts on the choices of supervisory system. Coefficients on LDC are significantly positive when FSI = 2 but insignificantly positive for FSI = 3, indicating that a LDC tends to adopt partial but not unified supervision. Significantly negative coefficient of CEE in FSI = 2 but not 3 means that a CEE raises the possibility of adopting sectoral supervision but is not the factor for using unified supervision. GDP per capita is significantly positive

(32)

for both FSIs’, which has the implication that the richer countries have less tendency to adopt sectoral supervision.

Model 1B of Table 5 adds an additional variable, SIZE, into the model. The new added variable is insignificantly negative regardless of FSIs and have no effect on the results of the original five variables used in Model 1A. Model 1C replace GDP per capita by POPULA, which shows a significantly negative sign when FSI =2 but insignificantly negative for FSI = 3. Thus, more population decrease the probability of adopting partial supervision but has no effect on unified supervision. In the case of FSI = 3, coefficients of DC, LDC and CEE all change from insignificance to significance. There are two reasons. One is the extra explanation provided by POPULA and the other is the slightly changed sample size. Recall that, in the whole sample, we have 98 countries whenever GDP per capita is considered. Thus, the Model 1C removes GDP per capita and hence there are 101 countries used. Model 1D takes GDPper, POPULA and SIZE into account.

Judging from the results from four specifications in Table 5, CB is over-whelmingly significantly negative and is our robust factor. Furthermore, this evidence does not change even if we use different sample size. Thus, the evidence that a country with her central bank supervising banks strongly prefer sectoral supervision is confirmed. Hence, we do not repeat the discus-sion of this evidence in the following estimated results. The impact of DC is fickle with sign changes in different specification and is only significant in

(33)

Model 1C. It is our marginal fragile factor. LDC, on the contrary, is our marginal robust factor because its effect is not significant only in Model 1B. Population is a robust factor when FSI =2, indicating that higher population prefer sectoral supervision, comparing to the partial case.

Table 6 expands the explanatory variables to consider five interaction terms, i.e., CB×GDPper, CB×DC, CB×LDC, CB×DC ×GDPper and CB×LDC×GDPper. There are three specifications, 2A, 2B and 2C in this table. Because the sample size is the same for three specifications (=98), the LR can be carried out. The log-likelihood values of the three specifi-cations are –68.05, –64.18 and –63.96, respectively, and the latter two are nested in the third one, making the LRs are 8.72 and 0.44. Hence, we reject specification 2A but cannot reject 2B in a statistical sense. In the three specifications, coefficients of CEE are overwhelmingly significantly negative when FSI = 3. Thus, relative to the benchmark FSI, CEE prefer not to choose the unified supervision. This is also our robust factor. Coefficients of DC in specifications 2B and 2C are both significantly positive, suggest-ing taht developed countires tend NOT to adopt sectoral supervision. It is inconsistent with the intuition that rich countries have the capacity and then to adopt the sectoral supervision. LDC has no effects on the choice of supervisory system. GDP per capita, to our surprise, change its signs from positive in Table 5 to negative, though most of them are insignificant. One reason to account for this is probably owing to the interativer tersm added,

(34)

which mitigates the effect of GDP per capita per se. Population remains strongly negative when FSI = 2 and is our maginal robust factor to explain FSI = 2 vs. FSI = 1 because it is also signficant in the next two tables.

The interactive terms also show interesting results. Firstly, coefficients of CB×DC are significantly negative in FSI = 3. Thus, a developed country with her central bank supervises banks are strongly against the adoption of unified supervision but prefer sectoral supervision. This evidence together with the above evidence from DC alone, we could conclude that developed countries tend to adopt partial and unified supervision when their central banks supervise banks; alternatively, they tend to adopt sectoral supervision when their central banks do not supervise banks.

Secondly, coefficients of CB×LDC are significantly positive when FSI =2. This evidence together with insignificant coefficent of LDC, suggesting that a less developed countries with her central bank supervises banks prefer partial supervision. If the central bank of a LDC does not supervise bank, it has no effect on supervisory system. Thirdly, even among the developed countries when their central banks supervise banks, the GDP per capita has negative impact on FSI =3 but positive effect when they are LDC. Thus, increasing GDP per capita increase the probabliyt to adopt unified supervision in DC×CB but decresase it in LDC×CB.

Table 7 presents the estimated results using the subset samples. There are also three specifications, 3A, 3B and 3C in the table. Also, sample sizes

(35)

of each specification are different. Recall that when bank restrictive vari-ables are used, our sample is reduced to 82 and when both bank restrictive variables and OSPI and PMI are jointly used, the sample size is further reduced to 69. Thus, LR test cannot be carried out when sample sizes are different. Specification 3A adds three bank restriction variables, BANK-S, BANK-I and BANK-R but takes out the interaction variables. Specifiction 3B considers both bank restriction and interation variables, making 3A be nested in 3B. The LR test is 18.7 as the log-likelihood value is –58.43 in 3A and is –49.08 in 3B, rejecting the null of zero coefficients on interaction variables. Thus, adding interaction variables are crucial in the determinant equation when in this subset sample. In specification 3B, except for positive coefficients on BANK-S, coefficients of bank restriction variables are mostly insignificant. It indicates that countries whose banks are prohibited to en-gage in securities tend to adopt unified supervision. This result, however, is sensitive to different specifications. Coefficients of interaction variables are found similar to those reported in previous table (Table 7). Specification 3C further adds OSPI and PMI and all bank restriction variables turn out to have no effect on the choice of supervisory system. The effects of interation variables are similar to those reported in the previous table (Table 6)

Table 8 adds GOODGOV into the model and also contains three spec-ifications, 4A, 4B and 4C. Because the sample size is not the same, no LR

(36)

test is conducted.24 This newly added variable is significantly when FSI = 3 regardless of specification. It is not surprising because the countries adopt unified supervision, such as, Denmark, Norway, Sweden, UK have very good governance and Japan and Korea to a lesser degree.25 Also, using the small sample size of only 41 countries, bank restriction variables are signficiant. BANK-S are overwhelmingly significantly postive in FSI = 3, whereas BANK-R are overwhelmingly significantly negative in FSI =2 as well as 3. BANK-I are fragile because it is significiant in FSI = 2 in 4A but in FSI = 3 in 4B.

5.2 Performance Equation

Tables 9 14 present the estimated results using ROA, NPL, Cover, Tier 1 and the Liquid as dependent variables.

6

Conclusion and Policy Suggestions

24

There is a convergence problem when we include all varaibles. Thus, the specifications shown in Table 8 are the ones that converge.

(37)

References

Abrams, R. K. and M. W. Taylor (2000), Issues in the unification of finan-cial sector supervision, working paper, IMF

Abrams, R. K. and M. W. Taylor (2001), Assessing the Case for Uni-fied Sector Supervision, Risk Management and Insurance International Conference, 2001.

Barth, J. R., G. Caprio Jr., and R. Levine (1998), Financial Regulation and Performance: Cross Country Evidence, The World Bank WP. 2037 Barth, J. R., L. G. Dopico, D. E. Nolle and J. W. Wilcox, (2001), An

International Comparison of the Structure of Bank Supervision, The Future of Financial Regulation in Taiwan, Taipei, July 6.

Barth, J. R., L. G. Dopico, D. E. Nolle and J. A. Wilcox (2001), Bank Safety and Soundness and the Structure of Bank Supervision: A Cross-Country Analysis, FMA Annual Meeting, Toronto, Canada, V1.5 Barth, J. R., G. Caprio Jr., and R. Levine (2001), “Banking Systems

Around the Globe: Do Regulation and Ownership Affect Performance and Stability?” in F. S. Mishkin, Editor: Prudential Supervision: What Works and What Doesn’t?” University of Chicago Press

Barth, J. R., G. Capiro, Jr. and R. Levine (2002), Bank Regulation and Supervision: What Works Best? NBER Working Paper 9323

Basel Committee on Banking Supervision (2001), The Relationship Be-tween Banking Supervisors and Banks’ External Auditors, Consulta-tive Paper issued by the Basel Committee on Banking Supervision and the International Federation of Accountants.

Beck, T., Demirg¨u¸c-Kunt, A., R. Levine (2001), ”The Financial Structure Database”. in Demirg¨u¸c-Kunt and Levine, Editor: Financial Struc-ture and Economic Growth: A Cross-Country Comparison of Banks, Markets, and Development, MIT Press, Cambridge, MA, 17-80

(38)

Bikker, J. and I. v. Lelyveld (2003), Economic versus Regulatory Capital for Financial Conglomerates, in Kuppens, Prast and Wesseling (ed.), Banking Supervision at the Crossroads. Edward Elgar, USA

Briaut, C. (1999), The Rationale for a Single National Financial Services Regulator, Financial Services Authority, Occasional Paper No.2, Lon-don, Financial Services Authority.

Briault, C. (2002), Revisiting the Rationale for a Single National Financial Services Regulator, the Financial Services Authority Occasional Paper, series 16.

Carmichael, J. (2001), Experiences With Integrated Regulation, the Finan-cial Regulator, vol. 6.

Courtis, N. (2002), How Countries Supervise their Banks, Insurers and Securities Market, Central Banking Publications, and various issues. De Krivoy, R., Reforming Bank Supervision in Developing Countries. Di Noia, C. and G. Di Giorgio (1999), Should Banking Supervision and

Monetary Policy Tasks Be Ginven to Different Agencies?, The EUI workshop on Macroeconomic Policy After EMU, Florence, April 30th, 1999.

Giddy, I. H. (1994), Who Should Be the Banking Supervisors, The Seminar on Current Legal Issues Affecting Central Banks, IMF.

Goodhart, C. A. E., Hartmann, P., Llewellyn, D. T., Rojas-Suarez, L. and Weisbrod, S. (1998), Financial-Regulation, Routledge, London and New York.

Goodhart, C. A. E. (1995), Some Regulatory Concerns, London School of Economics Financial Markets Group, Special Paper, no.79.

Goodhart, C. A. E. and D. Shoenmaker (1995), Should the Functions of Monetary Policy and Banking Supervision Be Separated? Oxford

(39)

Eco-Ioannidou, V. (2002), Does Monetary Policy Affect the Central Bank’s Role in Bank Supervision? Center and Department of Economics, Tilbury University, no. 2002-54.

Ioannidou, V. P. (2002), Does Monetary Policy Affect the Central Bank’s Role in Bank Supervision? Center, and Department of Economics, Tilburg University.

La Porta, R., F. Lopez-de-Silanes, A. Shleifer and R. W. Vishny (1998), Law and Finance, Journal of Political Economy, 1998, vol.106, no.6. ????

Lee, L. F. (1983), Generalized Econometric Models with Selectivity, Econo-metrica, 51, 507-512.

Lelyveld, I. V. and A. Schilder, (2002), Risk in Financial Conglomerate: Measurement and Supervision, Research Series Supervision No. 49, De Netherlands Bank

Llewellyn, D. T., 1999, Introduction: the Institutional Structure of Reg-ulatory Agencies, in N. Courtis(ed.), How Countries Supervise Their Banks,Insurers,and Securities Markets, London: Central Bank Publi-cations, xi-xix.

Masciandaro, D. (2004), Unification in Financial Sector Supervision: The Trade-Off Between Central Bank and Single Authority, forthcoming in Journal of Financial Regulation and Compliance, 12, 2

Mooij, J and H. Prast (2003), A Brief History of the Institutional Design of Banking Supervision in the Netherlands. in Kuppens, Prast and Wesseling (ed.), Banking Supervision at the Crossroads. Edward Elgar, USA

Roe, M. J. Strong Managers and W. Owners (1994), The Political Roots of American Corporate Finance, Princeton, N. J., Princeton Univ. Press Romano, R. (1997), The Political Dynamics of Derivative Securities

(40)

Sanio, J. (2003), The New Single Regulator in Germany, in Kuppens, Prast and Wesseling (ed.), Banking Supervision at the Crossroads. Edward Elgar, USA

Shen, C. H. (2002), Changes of Financial Supervisions in Taiwan, paper presented in 2nd conference of capital market, Taipei

Shen, C. H. and Y. Y. Chang (2002), Bank Regulations, Bank Performance and Governance: A Global Study, Economic Essay, Academic Sinica, Taiwan. 30, 277-310

Taylor, M. and A. Fleming (1999), Integrated Financial Supervision:Lessons of Northern European Experience, The FPSI Network of the World Bank.

Taylor, M. (1996), Peak Practice: How to Reform the United Kingdom’s Regulatory System, Centre for the Study of Financial Innovation, Lon-don, October.

Taylor, M. (1995), Twin Peaks: A Regulatory Structure for the New Cen-tury, Centre for the Study of Financial Innovation, London.

Taylor, M. and A. Fleming (1999), Integrated Financial Supervision:Lessons from Northern European Experience, World Bank, Policy Research Working Paper, no. 2223.

Tuya, J. and L. Zamalloa (1994), Issues on Placing Banking Supervision in the Central Bank, in T. Bali˜no and C. Cottarelli (eds), Frameworks for Monetary Authority Stability, IMF

Wall, L. D. and R. A. Eisenbeis (2000), Financial Regulatory Structure and the Resolution of Conflicting Goals, Journal of Financial Services Research 17:1, 223-245.

Zwet, A. V. D. (2003), The Blurring of Distinctions between Financial Sectors: Fact or Fiction, De Nederlandsche Bank, Occasional Studies, No. 2

(41)

Table 1: Variable Definitions and Sources Focused Variable

Variable definition Content Source FSI Financial integration 1: no integration, Courtis (2002)

of Supervision 2: partial, 3: full

CB Central bank also 0-1, 1 is yes, 0 is no Courtis (2002) supervises banks

Bank Performance Variables: Dependent Variable

Tier 1 Tier 1 Capital Ratio (%) Bankscope NPL Problem Loan/Total Loan Bankscope CAR Total Capital Ratio (%) Bankscope Cover Problem loan/Loan Loss Reserve Bankscope ROA Bank’s ROA Bankscope Liquid Liquid assets/total deposit IFS

Bank Restrictions

Bank-S Restriction on Bank’s 1: unrestricted 2: permitted Barth et al. (2000) investment activities 3: restricted 4: prohibited

Bank-I Restriction on Bank’s same as above Barth et al. (2000) insurance activities

Bank-R Restriction on Bank’s same as above Barth et al. (2000) real estate activities

Bank Asset Holder

GovAsset Ratio of bank asset Gov’t Asset/Total Barth et al. (2002) held by government bank asset

ForAsset Ratio of bank asset Foreign bank asset/ Barth et al. (2002) held by foreign total bank asset

Institutional Variables

GoodGov Good Government 0-50, sum of eff. of judic. LLSV (1998) Index system, rule of law,

corru-ption, risk of expropriation risk of contract repudiation

OSPI Official supervision 0-16, sum of prompt correc Barth et al. (2001) power index tive action, restructuring

power and declaring insolvency power

PMI Private monitoring 0-10, sum of certified audit Barth et al. (2001) Indexc required, percent of 10

big-gest bank rated, no explicit deposit insurance, bank acc-ounting, discloser of off-sheet, disclosure of risk (continue in next page)

(42)

(continued...)

Bank−Based Variables

BKasset Bank asset/GDP Bankscope Lending Claims of private sector/GDP IFS

Market-Based Variables

MktCap Market capitalization/GDP IFC TradeV Traded value/GDP IFC

D. Macro Variables

Infla Inflation IFS Unemp Unemployment IFS GDPper GDP per capita WDI

Law Origins

Law−UK UK original Law 0-1, 1 is yes, 0 is no LLSV (1998) Law−Fra French original Law 0-1, 1 is yes, 0 is no LLSV (1998) Religion % of Christian 0–100

IFC: Emerging Stock Markets Fact Book, 2000 IFS: International Financial Statistics, 2000, IMF WDI: World Development Indicator, 2000, World Bank

(43)

Table 2: Countries of Financial Integration of Supervision

Sectoral Integration (56)

CB Albania Algeria Argentina Bahamas Bahrain supervise Bangladesh Barbados Botswana Brazil Bulgaria banks (47) China1 Croatia Czech Rep. Egypt Ghana

Greece Hong Kong India Israel Italy Jamaica Jordan Kazakhstan Lithuania Mauritius

Nepal Netherlands2 New Zealand Nigeria Oman

Pakistan Papua New G. Philippines Portugal Romania Russia Slovenia Spain Sri Lanka Taiwan3

Tanzania Thailand Trinidad & Tunisia Ukraine USA Zambia Tobago

CB does not Costa Rica France Germany4 Indonesia Latvia

Supervise (9) Panama Poland Venezuela Turkey Partial Integration (35)

Bank & Insurance (18)

CB Supervise Colombia Ethiopia Gambia Honduras Macao Paraguay ierra Leone Suriname

CB does not Australia5 Austria Canada Cayman El Salvador

supervise Gibraltar Guatemala IceLand Malaysia Peru Bank & Security (14)

CB Supervise Bermuda Cyprus Dominican Rep. Guyana Ireland Saudi Arabia United Arab Emirates

Mexico

CB does not Belgium Finland6 Hungary Ireland7 Luxembourg

supervise Switzerland Insurance & Security (3)

CB Supervise South Africa CB does not

supervise Bolivia Chile Unified Integration (10)

CB Supervise Malta8 Netherlands Antilles8 Singapore Uruguay9 CB does not Japan10 Denmark Norway South Korea10 Sweden

supervise U.K.

Sources are mainly from Courtis (2002) and websites.

This classification of financial supervision is based on his 2001 or earlier survey. 1. China has moved to the sectoral supervisor in 1998.

2. Netherlands adopts functional supervision by merging banking and Insurance Supervision on April 2004. 3. Taiwan joins the unified supervision club on July 1, 2004

4. Germany adopts a modified unified supervision on May, 2002. 5. Australia adopts functional supervisions.

6. See Taylor and Fleming (1999) but Finland’s supervisors have conflict reports. 7. The Central Bank of Ireland takes change the three supervisions on April 2002. 8. These are small countries thus central banks charge all supervisions.

9. Also see Taylor and Fleming (1999).

(44)

Table 3: Basic Statistics I: FSI, Basic Features and Scales FSI and Basic Features of Countries

Sectoral Partial Unified Sum Supervision Supervision Supervision CB Sup. Number 47 16 4

CB Not Sup. Number 9 19 6 101

CD Number 9 9 5 LDC Number 36 25 5

CEE Number 11 1 0 101

CB Sup. and is a DC Number 7 1 0 CB Sup. and is a LDC Number 31 15 4

CB Sup. and is a CEE Number 9 0 0 101

CB Not and is a DC Number 2 8 5 CB Not and is a LDC Number 5 10 1

CB Not and is a LDC Number 2 1 0 101

FSI and Scale of Countries

GDPper median 3139.10 5120.00 21673.60 mean 6931.96 12903.55 21145.17 std. dev. 8682.64 14955.26 14575.37 number 56.00 33.00 9.00 98 Population median 17.01 6.25 6.34 mean 78.59 13.92 23.68 std. dev. 212.45 20.26 40.19 number 56.00 35.00 10.00 100 Size median 268.68 112.09 199.49 mean 1294.52 884.99 184.56 std. dev. 3027.10 2079.67 167.39 number 55 35 10 101 Note:

1. CB Sup. menas that a central bank also supervises banks. 2. CB Not menas that a central bank does not supervise banks.

3. DC, LDC and CEE denote devloped, less developed and Central and Eastern European countries, respectively.

3. GDPper is GDP per capita. 4. Population’s unit is million.

5. Size is the size of a country, whose unit is 1,000 km2. 6. Number is the number of countries belong to this category. 7. std. dev. is the standard deviation.

(45)

Table 4: Basic Statistics II: FSI and Bank Level Data Sectoral Partial Unified Supervision Supervision Supervision

ROA median 0.040 0.031 0.01050 mean 0.254 0.222 0.03580 std. dev. 0.638 0.773 0.07439 number 56 35 10 NPL median 0.099 0.061 0.05000 mean 0.137 0.071 0.09100 std. dev. 0.122 0.055 0.09543 number 42 19 8 Cover median 0.995 0.936 0.75100 mean 4.366 1.167 1.42217 std. dev. 13.373 0.669 1.81099 38 17 6 Tier median 14.274 10.35400 12.293 mean 17.212 15.82424 12.281 std. dev. 10.808 23.53757 4.388 number 40 25 8 Liquid median 0.713 0.64100 0.623 mean 1.406 0.82151 2.916 std. dev. 1.999 0.64126 3.847 number 56 35 10

(46)

Table 5: Determinants of Supervision I: the Whole Sample Size

1A 1B 1C 1D

FSI = 2 FSI = 3 FSI = 2 FSI = 3 FSI = 2 FSI = 3

Constant 0.125 –1.355 0.153 –0.962*** 0.723*** –0.554*** 0.424 –0.819 (0.48) (0.05) (0.37) (3.97) (2.85) (3.48) (1.44) (1.20) CB –1.954*** –1.973*** –1.914*** –2.107*** –1.996*** –2.032*** –1.909*** –2.138*** (4.06) (2.85) (3.86) (3.07) (4.11) (3.76) (4.29) (3.36) DC –0.718 -0.275 –0.600 0.383 0.714 1.485*** –0.109 0.545 (1.13) (0.01) (0.72) (1.02) (1.44) (2.66) (0.14) (0.48) LDC 0.639** 0.279 0.627 0.576 0.974*** 0.957*** 0.848*** 0.731*** (1.98) (0.01) (1.10) (0.83) (2.96) (3.59) (2.76) (2.67) CEE –1.593** –14.999 –1.597* –14.777* –1.661* –17.089*** –1.617* –15.164 (2.03) (0.03) (1.83) (1.79) (1.96) (3.49) (1.94) (1.08) GDPper 0.0001** 0.0001*** 0.0001* 0.0001** 0.000 0.0001** (2.09) (2.58) (1.85) (2.00) (1.51) (2.28) POPULA –0.031*** –0.001 –0.029*** –0.003 (2.90) (0.13) (3.05) (0.23) SIZE –0.0001 –0.001 0.000 –0.002* 0.000 –0.002 (0.42) (1.22) (1.35) (1.66) (1.62) (1.49) Number 98 98 101 98 Log-Likelihood –72.46 –70.81 –70.43 –65.51 absolute t−value in parenthesis.

(47)

Table 6: Determinants of Supervision II: The Whole Sample

2A 2B 2C

FSI = 2 FSI =3 FSI = 2 FSI = 3 FSI = 2 FSI =3 Constant 0.495 –0.929 0.612*** –0.893* 0.754*** –0.607 (0.48) (0.58) ( 2.70) ( 1.90) (2.66) (1.38) CB –12.487*** –2.447*** –12.878*** –1.929*** –12.920*** –2.147*** (4.47) (2.52) (16.53) ( 3.08) (16.04) (3.34) DC 1.651 3.194 1.472* 2.155*** 2.113* 3.270*** (0.76) (1.22) ( 1.89) ( 2.38) (1.79) (3.61) LDC 0.214 –0.469 0.460 –0.493 0.438 –0.588 (0.14) (0.24) ( 0.77) ( 0.55) (1.60) (0.86) CEE –1.102 –17.047*** –0.954 -13.100*** –1.010 –12.997*** (0.72) (3.64) ( 0.89) ( 9.77) (1.08) (4.75) GDPper –0.0001 –0.000 –0.000 –0.0001** (0.43) (0.67) (0.78) (2.13) POPULA –0.021*** –0.008 –0.022** –0.007 ( 2.42) ( 0.73) (2.38) (0.66) SIZE CB × GDPper -0.0001 –0.000 –0.000 –0.000 –0.0001*** –0.000 (0.83) (0.48) ( 4.09) ( 1.52) (6.29) (1.32) CB× DC 5.578 –12.088*** 5.259 –9.876*** 4.516 –11.999*** (1.52) (3.76) ( 1.04) ( 7.69) (1.26) (4.27) CB× LDC 10.560*** 0.486 11.287*** 0.387 11.206*** 0.383 (4.78) (0.58) (46.95) ( 0.49) (11.08) (0.52) CB× DC ×GDPper 0.000 –0.0001*** 0.000*** –0.000*** 0.0001*** –0.0001*** (1.08) (3.80) ( 2.89) ( 6.37) (3.08) (4.31) CB× LDC ×GDPper 0.000 0.0001* 0.000*** 0.000*** 0.0001*** 0.0001*** (1.15) (1.82) ( 4.20) ( 3.83) (3.89) (4.86) Number 98 98 98 Log-Likelihood –68.05 –64.18 –63.96 *, ** and *** denote the significance at 10%, 5% and 1% level, respectively.

(48)

Table 7: Determinants of Supervision III: Small Sample

3A 3B 3C

FSI = 2 FSI =3 FSI = 2 FSI =3 FSI = 2 FSI = 3 Constant 0.838 –0.626*** 1.305* –0.460 –1.399 –3.524 (1.47) (3.24) (1.86) (0.49) (0.95) (1.56) CB –2.466*** –1.956*** –12.863*** –2.317*** –11.832*** –3.214*** (4.51) (2.42) (79.45) (3.39) (6.03) (2.76) DC 0.052 –0.002 2.686 3.529* 0.243 1.283 (0.06) (0.00) (1.56) (1.86) (0.11) (0.51) LDC 1.432** 0.793 0.880 –1.134 –2.046 –3.798 (2.22) (0.98) (1.24) (0.87) (1.26) (1.33) CEE –0.913 –13.075 –0.925 –12.700*** –2.936 —21.091*** (0.85) (1.46) (0.83) (8.56) (1.61) (6.83) GDPper 0.000 0.0001 –0.0001 –0.0001 –0.000 –0.0001 (1.06) (1.51) (0.70) (0.99) (0.72) (1.15) POPULA –0.030 –0.010 –0.031*** –0.003 (2.70) (0.77) (2.35) (0.21) SIZE CB × GDPper –0.0003*** –0.000 –0.0002** 0.000 (7.35) (0.52) (2.03) (0.46) CB× DC 3.898 –10.292*** 1.495 –8.717*** (1.29) (7.89) (0.26) (3.58) CB× LDC 10.628*** –0.097 8.697*** –1.189 (67.49) (0.11) (6.04) (0.87) CB× DC ×GDPper 0.001*** –0.0003*** 0.0005*** –0.0003*** (8.67) (6.27) (3.54) (3.58) CB× LDC ×GDPper 0.000*** 0.0003*** 0.0004*** 0.0003*** (6.43) (3.43) (2.60) (2.50) BANK-S –0.073 0.833 0.273 1.126** 0.659 0.922 (0.19) (1.50) (0.69) (2.04) (1.31) (1.16) BANK-I –0.090 –0.243 –0.090 –0.333 0.316 0.042 (0.30) (0.54) (0.31) (1.06) (0.73) (0.09) BANK-R –0.289 –0.769* 0.267 0.000 –0.333 0.000 (0.96) (1.68) (0.93) (0.00) (0.84) (0.00) OSPI 0.037 –0.115 (0.29) (0.48) PMI 0.504* 0.788* (1.66) (1.75) Number 82 82 82 82 69 69 Log-Likelihood –58.43 –49.08 –37.58

(49)

Table 8: Determinants of Supervision VI: Small Sample

4A 4B 4C

FSI = 2 FSI =3 FSI = 2 FSI =3 FSI = 2 FSI = 3 Constant –0.062 –6.049** –8.873 –22.459*** –0.583 –1.423 (0.02) (1.94) (1.57) (2.45) (0.61) (1.05) CB –3.761*** –2.770** –7.583*** –10.965*** –13.414*** –3.091*** (3.17) (1.95) (2.65) (2.48) (13.44) (3.21 ) DC 0.336 –6.339* –8.295 –24.635*** –0.691 –1.905 (0.10) (1.79) (1.41) (2.40) (0.40) (0.90) LDC –0.426 –5.742 –9.528* –19.913*** –1.055 –1.344 (0.16) (2.04) (1.69) (2.42) (0.99) (0.88) CEE –2.017 15.300*** (1.35) (8.89) GDPper –0.0001 –0.0001 –0.0001 –0.0001 (0.73) (1.16) (0.98) (0.94) Population –0.050*** –0.043* (2.46) (1.80) Size CB × GDPper –0.0001*** -0.00001 (4.72) (0.16) CB× DC 5.509*** -9.270*** (2.254) (5.49) CB× LDC 10.444*** -1.058 (7.49) (0.87) CB× DC ×GDPper 0.0004*** -0.0003*** (4.31) (5.10) CB× LDC ×GDPper 0.0004*** 0.0003*** (5.66) (2.86) BANK-S 2.338** 3.908*** 2.519 5.463** (2.16) (2.83) (1.71) (2.22) BANK-I 1.261** 1.069 2.895 4.396*** (2.15) (1.50) (2.20) (2.35) BANK-R –1.862*** –2.175*** –3.421** –6.567** (2.92) (2.84) (2.21) (2.28) OSPI –0.132* –0.257 –0.009 –0.179 (0.55) (0.80) (0.09) (1.09) PMI 1.913** 1.991** 0.382* 0.553* (2.24) (2.04) (1.68) (1.70) GOODGOV 0.005 0.226** 0.169 0.796** (0.05) (2.02) (0.68) (2.12) Number 46 46 41 41 69 69 Log-Likelihood –25.99 –16.21 –42.79

數據

Table 1: Variable Definitions and Sources Focused Variable
Table 2: Countries of Financial Integration of Supervision
Table 3: Basic Statistics I: FSI, Basic Features and Scales FSI and Basic Features of Countries
Table 4: Basic Statistics II: FSI and Bank Level Data Sectoral Partial Unified Supervision Supervision Supervision
+7

參考文獻

相關文件

External evidence, as discussed above, presents us with two main candidates for translatorship (or authorship 5 ) of the Ekottarik gama: Zhu Fonian, and Sa ghadeva. 6 In

We conclude this section with the following theorem concerning the relation between Galois extension, normal extension and splitting fields..

Enhancing Transition from Early Childhood Phase to Primary Education: Evidence from the research literature. Resources for pre-school educators.Retrieved

More than 90% of the leaders reported that the Panel Chair was also expected to ensure that all teachers followed curriculum guidelines, and to review evidence of teaching

Summarising the whole study, the authors believe that with the evidence, a liberal-arts mathematics course with an emphasis on the culture and history of the discipline can

conglomerates and religious bodies have to consult these high-level stipulations when they settle on their own constitutions. Worldly law developed in this way step by step. The

Based on historical documents and archeological evidence, this thesis provides an analysis of, raises some worth-noting questions on, the development of Western Qin Buddhism

例如:比較P–2%與HIBOR+1%,哪種利息對借款人較有益處,並講解銀行最優惠利 率與香港銀行同業拆息的變動會帶來甚麼風險。..