Financial Development, Financial Liberalization and Social Capital
Luuk Elkhuizen
Supervisor: prof dr. C.L.M. Hermes
University of Groningen
June 2015
Abstract
The effect of financial liberalization on financial development is controversial in both academic
literature as well as in practice, with the effects of liberalization episodes differing greatly across
countries. The quality of formal institutions has been identified as an important source of this
heterogeneity, as countries with a weak institutional environment generally fail to benefit from
financial liberalization. In this paper, I make use of panel data on 82 countries in the period
1973-2008 and find robust evidence that social capital can also function as a prerequisite to financial
liberalization. Especially in the post Washington-consensus period, countries with a high prevailing
level of social capital could ensure that financial liberalization positively influenced financial
deepening, despite the poor quality of their formal institutions.
Keywords: financial liberalization, financial deepening, financial efficiency, social capital,
generalized trust.
JEL classification: G15; G21; G28; P41; P45
Student Msc. Economics and Msc. Finance, University of Groningen, Faculty of Economics and Business, student
number: S1870203, e-mail address: l.elkhuizen.1@student.rug.nl.
2 I. Introduction
For a long time, the influence of financial development on economic growth has been controversial in academic literature. Where some argued that financial development fostered growth, others considered it to be a result of -rather than a cause of- economic growth. This dichotomy may be best illustrated by the now famous words of Jean Robinson (“Enterprise leads where finance follows”, 1952) and Merton Miller (“the idea that financial markets contribute to economic growth is a proposition too obvious for serious discussion”, 1998).
The above mentioned dichotomy persisted for almost an entire century because it turned out to be difficult to empirically test for the direction of causality.1 This changed in the early 1990s, when empirical methods were developed that allowed for a richer view on the direction of causality between financial development and economic growth (Khan and Semlali 2000; King and Levine 1993). While literature on the financial development and growth nexus is still expanding, there now appears to be a consensus that indeed financial development has a positive influence on economic growth (Beck and Levine, 2000). This consensus renders the factors that influence financial development important. Especially policy makers of countries with less developed financial sectors may benefit from an understanding of the forces that shape their financial sector. Consequently, there has been a spike in research on the determinants of financial development in recent years. Both long-run (e.g. culture and geography) as well as short-run (e.g. macroeconomic policies) determinants of financial development have been discussed in existing literature.
Financial liberalization is one of these short-run forces that has been put forth as a determinant of financial development. The idea that financial liberalization promotes financial development can be traced back to the seminal work of McKinnon (1973) and Shaw (1973), who argued that countries with repressed financial sectors could stimulate financial development by liberalizing their financial sectors (e.g. removing credit controls and interest rate ceilings). They advocated that this would allow the interest rates to reach its competitive market equilibrium, which would subsequently boost savings, investments and ultimately economic growth.2
From the 1970s onwards, policy makers of developing countries throughout the world started liberalizing their financial sectors. This accelerated during the 1990s, after Williamson (1990) introduced what he called the ‘Washington consensus’. Williamson argued that the United States’ government, the United States’
1 Bagehot’s paper (1873) started the discussion with what is considered to be the first paper on financial
development.
2 Section 3 provides a more elaborate description of the ways in which financial liberalization is likely to influence
3
ministry of foreign affairs and the world’s leading international financial institutions (which all happened to be located in Washington), had reached consensus on a set of ten specific policies that developing countries should adopt in order to foster growth. Financial liberalization was one of these principles. In response to this consensus, financial liberalization among developing countries accelerated during the 1990s.
After a couple of decades of financial liberalization, the empirical evidence on the effectiveness of these liberalization policies is mixed. The consensus in academic literature appears to be that, despite heterogeneity across countries, the overall effect of these policies on financial development and growth is positive (Levine 2001; Bekaert Harvey and Lundblad 2005; Huang 2011; Bumann, Hermes and Lensink 2013). However, in some cases financial liberalization seemed to have had adverse effects, triggering financial crises instead of financial development (Demirgüç-Kunt and Detragiache 1998; Klein and Olivei 2008).
In more recent years, researchers have started exploring the underlying sources of this observed heterogeneity. Factors that have already been identified as prerequisites of financial liberalization in the existing literature are bureaucratic efficiency, a strong rule of law, proper contract enforcement, control over corruption and prudential regulation and supervision (Demirgüç-Kunt and Detragiache 1998; Summers 2000; Hermes and Meesters 2015).3
In this paper I contribute to the literature by identifying yet another prerequisite to financial liberalization, namely social capital. While the term social capital can been traced back as far as the 1890s, it became a particular popular concept in the 1990s. This was mostly due to the work of Coleman (1988), who turned social capital into a practical concept. Nowadays, social capital is often referred to as “generalized trust” and essentially is a measure of the extent to which individuals in a country trust each other. The concept has become a popular topic of research in recent years and has already been applied to financial development (Guiso, Sapienza and Zingales, 2000) as well as economic growth (Knack and Keefer 1997; Kouvavas and ten Kate 2013) but it seems to have been overlooked in the specific strand of research on financial liberalization. In this paper, I endeavor filling this gap by conjecturing that financial liberalization success is conditional on the prevailing level of social capital. The intuition behind this proposition rests on the notion that social capital can function as a substitute for formal regulation. While this intuition is not new
4
(Boix and Posner, 1998), to my knowledge it has not yet been applied to the concept of financial liberalization.
This paper continues as follows. First, in section 2 I discuss the concept of financial development. Section 3 addresses financial liberalization and its effect on financial development. In section 4, I discuss the concept of social capital and I hypothesize how it may influence financial liberalization success. Section 5 then continues with the methodology, while section 6 provides a description of the data. The results will be discussed in section 7, after which a robustness analysis will be performed in section 8. Finally, I reflect and conclude in section 9.
II. Financial development
Financial development occurs when financial markets or institutions reduce market imperfections, thereby allowing capital to flow to its most productive use. (Čihák, Demirgüç-Kunt, Feyen, and Levine, 2012). As this is a rather theoretical and strict notion of financial development, a more practical approach entails defining the quality with which financial intermediaries provide their key functions. A commonly accepted classification in the literature of these key functions has been put forth by Levine (1997, 2005). He argues that the financial sector’s key functions are: (1) efficiently allocating capital and producing information, (2) exerting corporate governance, (3) facilitating trading, diversification, and management of risk, (4) pooling and mobilizing savings and finally (5) easing the exchange of goods and services.
5
associated with pooling and mobilizing savings (Levine, 2005). Finally, by providing financial arrangements and providing an easily recognizable and stable medium of exchange, the financial sector can lower transaction costs and ease the exchange of goods and services (King and Plosser 1986; Williamson and Wright 1994; Levine 2005).
Theoretically, financial development thus occurs when the quality with which these five functions are provided, increases. However, the problem is that one cannot simply measure quality. Researchers in the area of financial development thus come up with their own proxies for quality. Traditional research in this field often uses simple one-variable proxies of financial development, for instance the ratio of private credit to GDP (Čihák, Demirgüç-Kunt, Feyen, and Levine, 2012). As Čihák et al. (2012) argue, these proxies often place too much emphasis on the size of the financial sector (i.e. financial deepening) and hence present a too narrow view of financial development. Indeed, a large financial sector may not necessarily be efficient, or vice versa. In collaboration with the Worldbank, Čihák, Demirgüç-Kunt, Feyen, and Levine (2012) established their own dataset that allowed financial development to be measured with multiple variables, along multiple dimensions. In this paper, I will -as is common in recent literature on the determinants of financial development- benefit from this dataset by measuring financial development with multiple variables in the dimensions of depth and efficiency of financial intermediaries.4
III. Financial Liberalization
In the 1950s and 1960s, conventional wisdom stipulated that governments could promote development by protecting -and intervening in- financial markets, a process commonly referred to in literature as financial repression (Andersen and Tarp, 2003). These policies became subject to severe criticism in the early 1970s by McKinnon (1973) and Shaw (1973), who argued that liberalizing financial sectors would spur growth. As mentioned in the introduction, from the 1970s countries throughout the world acted upon this device and gradually started liberalizing their financial sectors. Among developing countries, financial liberalization especially occurred in the post Washington-consensus period (i.e. after 1990), arguably because these countries feared their economies would miss out on the benefits of an increasingly global world economy (Gore, 2000). However, the effects of financial liberalization, and post Washington-consensus liberalization in particular, have differed substantially across countries. In this section, I will discuss both the positive and the negative view on financial liberalization, as suggested in literature. Subsequently, I will close this section
4 The dataset also contains the dimensions ‘financial access’ and financial ‘stability’. Although financial
liberalization and social capital may also have an influence on these dimensions, it does so through different
pathways. Including these dimensions would thus essentially mean telling three stories in one paper. Moreover, these dimension only contain data for a few (recent) years. Hence, I leave it up to further research to explore these
6
with a discussion of the empirical literature on financial liberalization. Moreover, I will discuss the sources of the observed heterogeneity that have been put forth in this branch of literature.
With regard to the positive effects of financial liberalization on financial development, McKinnon (1973) and Shaw (1973) argue that countries that repress their financial sector (e.g. in the form of interest rate ceilings, credit controls and state-owned banks) will have a below-competitive equilibrium real interest rate. This low interest rate has negative effects on the savings- and loan rates in an economy, as incentives to save are low (because the interest rate is low), while demand for loans is excessively high. The low savings rate harms financial deepening. Likewise, the excess demand for credit harms efficient allocation of capital as (since real interest rates are constrained) bankers have no incentive to direct credit towards the most profitable projects. Hence the policy advice that follows from the McKinnon and Shaw models is simple: remove barriers (in particular government policies aimed at controlling the financial sector) that prevent the interest to rise to its competitive equilibrium and allow the market to determine the optimal allocation of credit.
Apart from liberalizing repressed financial sectors (where the government directly manipulates the interest rate), literature has identified more ways in which financial liberalization may occur. These ways include promoting competition (e.g. by reducing entry barriers for national and foreign banks) and liberalizing the capital account. Proponents of these policies argue that by allowing banks to compete against each other, the average interest rate on deposits (loans) will increase (decline). This will promote financial deepening as people will start saving more in response to this higher interest rate. Moreover, competition will cause banks to have an incentive to reduce overhead costs and improve on bank and risk management (Denizer, Dinc and Tarimcilar, 2007). Furthermore, to the extent that promoting competition entails reducing entry barriers for foreign banks, domestic banks may benefit from spillovers, such that they learn new bank- and risk-management techniques and develop new financial instruments and services (Claessens, Demirgüç-Kunt and Huizinga, 2001).
7
capital is not higher (Chinn and Ito, 2006).5 Likewise, capital account liberalization also enables domestic financial institutions to improve their ability to diversify against risk. This allows them to further reduce the costs of loans (Hermes and Lensink, 2008). Finally, capital account liberalization may lead to foreign direct investments with positive spillovers, such as transfers of technology or skills (Baillu 2000).
As already alluded to above, these positive effects of financial liberalization have not come undisputed. First, Stiglitz (2000) argues that the argument that liberalizing repressed financial sectors leads to more efficient credit allocation, is flawed. While under perfect information this may be true, financial markets are characterized by asymmetric information. Stiglitz shows that under asymmetric information, decentralization through the price mechanism (allowing banks to set their interest rates freely) will not necessarily lead to a Pareto-efficient equilibrium.6 In economic terms, his argument rests on the notion that it is unlikely that new (or privatized) banks have access to superior information. As such, the problem of asymmetric information prevails and banks are unable to become more efficient. Second, Boot (2000) has argued that financial liberalization may even aggravate information asymmetries, as borrowers -in the presence of competition between banks- may choose to end long-lasting relationships with their banks. By doing so, the information that the previous bank has gathered on the borrower is no longer of value, and information asymmetries thus increase. Third, increased competition between banks leads to a reduction in franchise value which, in turn, may lead to increased risk taking. As less efficient banks fail to compete by reducing overhead costs, they may adopt a gambling strategy, reduce information gathering and monitoring efforts in order to remain profitable (Hellman, Murdock and Stiglitz 2000; Andersen and Tarp 2003). While in the long run inefficient banks will likely be replaced by more efficient ones (Kaminsky and Schmukler, 2003), at least in the short run, financial liberalization may thus lead to instability instead of efficiency. Fourth, there are several authors who stress that capital inflows following financial liberalization are often speculative of nature and do not lead to long-run investments (Rodrik 1998; Stiglitz 2000). This may lead to sudden capital outflows, potentially followed by banks runs and banking crises (Diamond and Dybvig 1983; Demirgüç-Kunt and Detragiache 1998; Rodrik 1998; Dooley 1998).
3.1 Prerequisites to financial liberalization success
5 I do not measure financial development in terms of stock market development. However, to the extent that
investors invest through banks, portfolio diversification can still lead to an increase in financial deepening.
6 If the new equilibrium is not Pareto-efficient, this means that there it is possible for the government to improve the
8
As is clear from the above, there is a dichotomy in the existing academic literature with regard to the influence of financial liberalization on financial development. This dichotomy is not just restrained to theoretical literature on financial liberalization. Also empirical studies find mixed results of the effectiveness (in terms of triggering financial development and economic growth) of financial liberalization. While the net effect of financial liberalization appears to be positive (Levine 2001; Bekaert Harvey and Lundblad 2005; Huang 2011; Bumann Hermes and Lensink 2013), there is large heterogeneity between countries and between time periods. In light of this heterogeneity, recent empirical literature on financial liberalization has started to identify certain prerequisites to financial liberalization success.
Hermes and Meesters (2015) for example find that the impact of financial liberalization on bank efficiency is conditional on the quality of regulation and supervision of the banking system. More specifically, they find that -in a properly regulated banking sector- reducing credit controls, interest rate controls, entry controls and/or privatization of state banks positively contributes to making banks more efficient (as measured by a reduction in overhead costs). The intuition behind this result is that proper regulation and supervision can make sure that imprudent banking behavior is effectively curbed (Andersen and Tarp 2003), preventing banks in competitive environments (i.e. after liberalizing the financial sector) from taking on more risk than is socially desirable.
Demirgüç-Kunt and Detragiache (1998) find evidence that banking crises are more likely to occur in liberalized countries. On top of that, they find that a weak institutional environment and the absence of proper regulation and supervision makes the occurrence of crises more likely. This result holds even after controlling for macroeconomic stability. In their study, they proxy for institutional quality with measures of the rule of law, level of corruption, law enforcement and bureaucratic efficiency. Demirgüç-Kunt and Detragiache interpret their findings as institutional quality and proper regulation and supervision being prerequisites to financial liberalization. Again, the intuition behind the necessity of regulation and supervision is that with proper regulation and supervision, banking behavior can be effectively controlled. In a similar vein, institutional quality (e.g. by ensuring that contracts are enforceable) can secure proper behavior on the part of the bank’s clients.
9
with Basel Core Principles on banking supervision and the Insurance Core Principles. In particular (1) the ability to demand adjustments to capital, loan loss provisioning and employee compensation; (2) regulatory definitions; and (3) financial reporting and disclosures are considered important regulatory actions that can be taken to promote financial development.
Without proper regulation, supervision and the right institutional environment (henceforth ‘a high quality of formal institutions’), one may be inclined to conclude that it is undesirable to liberalize financial sectors. While this conjecture finds wide support in the existing literature (Caprio and Summers 1993; Stiglitz 1993; Arestis and Demetriades 1999; Gore 2000; Summers 2000; Williamson 2002), this conclusion may be incomplete. That is to say, there may be substitutes to formal institutional quality. Social capital (which I will discuss elaborately in the next section) has been identified in the political economy literature as such a potential substitute (Boix and Posner 1988). As such, it may also influence the relation between financial liberalization and financial development.
IV. Social Capital
Traditional neoclassical growth models have largely focused on physical and human capital as inputs in the process of generating economic growth. In these models, individuals are assumed to be independent and rational agents, who care about maximizing their utility. Sociologists have deemed these assumptions unrealistic, arguing that people are social creatures instead of rational agents. As such, their behavior is formed by the social context. While this view may hold some truth, economists have held great reservations about incorporating cultural aspects in their growth models. By allowing individual behavior to be a result of the social environment -so it was argued- individuals lose their internal “engine of action”, rendering the traditional neoclassical growth models (which are built on the assumption of rationality) unworkable by definition (Coleman, 1988). As a result, the sociological and economic stream of literature have remained separated for a long time.
10
examples.7 First, consider the market for diamonds. It is common practice in these markets that buyers inspect diamonds before making an offer. If sellers can rely on buyers to treat the diamonds with care and not replace one of the diamonds with a less valuable one (and this indeed turns out not to be the case) the sale may proceed quickly. If this is not the case, the seller has to think about ways in which he can govern the buyer, and the sale will thus be inefficient. Second, in environments characterized by information sharing and high social norms, parents might be more inclined to let their children walk to school without their supervision. This might be because social norms make it less likely that the environment is unsafe, or because the parents can implicitly rely on others (e.g. neighbors or other parents) to look after their children and to inform them if necessary. As a result, the parent can devote more time engaging in productive activities. In both cases, a high prevailing level of social capital in a society thus enables individuals to become more productive.
The concept of social capital as a resource has been received positively by the literature. However, questions were raised regarding the endogeneity of the concept. Religion, social networks (Coleman, 1988) income equality (Bjørnskov, 2007) ethnic homogeneity (Delhey and Newton 2005), economic freedom (Berggren and Jordahl, 2006) and the quality of formal institutions (La Porta, Lopez de Silanes, Shleifer and Vishny, 1997a) have been put forth as potential sources of social capital. As a result, the applicability of social capital in empirical models has been questioned. The consensus now appears to be that in the long run social capital indeed is endogenous. However, since it changes very slowly over time, it is safe to assume that at least in the short run, social capital is exogenous (Putman 1993; Algan and Cahuc 2010).8 As a result, there has been a spike in the growth literature that includes Coleman’s (1988) notion of social capital. Nearly all of these studies appear to follow Knack and Keefer (1997) who use data on generalized trust obtained from the World Value Survey to proxy for Coleman’s (1988) concept of social capital (Beugelsdijk and Maseland, 2011) 9
Most of the empirical research focuses on the effects of social capital on economic growth. The general result in this branch of literature is that social capital indeed positively contributes to economic growth (Zak and Knack, 2001). For example, Fukuyama (1995) argues that social capital can reduce transaction costs, as less written contracts are needed. Similarly, la porta et al. (1997a) argue that social capital allows large (government) institutions to function more effectively, as managers can rely more on employees to
7 I do not wish to claim any novelty regarding these examples. Both are mentioned in the paper of Coleman (1988)
too.
8 The data on social capital that I use also changes very little over time, asthe average correlation between two
11
accomplish tasks that are difficult to monitor. Furthermore, Knack and Keefer (1997) argue that innovation is more likely to occur in high trust countries, as entrepreneurs need not devote much time on monitoring possible malfeasance by partners, employees, and suppliers. They also argue that hiring managers in high-trust countries are more likely to base hiring decisions on competency instead of blood ties or personal knowledge. Moreover, they conjecture that social capital promotes greater investment and other economic activity, as investors have greater trust that banks will not expropriate their assets. Finally, Putman (1993) argues that social capital shifts the nature of people’s preferences away from individual concerns towards community-oriented concerns. As a result, in high social capital environments, people are more likely to punish poorly performing governments by not re-electing them. In that sense, social capital thus promotes the performance of government institutions, at least in democracies.
As can be seen above, the proposed intuition behind the effects of social capital on growth often implicitly rests on the notion that social capital is an effective substitute of formal regulation. This substitutability between formal regulation and social capital rests on two pillars. First, by trusting one another two parties can engage in transactions that could otherwise only be conducted if (enforceable) contracts were specified (Knack and Keefer 1997; Fukuyama 1995). Second, substitutability between formal regulation and social capital also requires that both parties are ‘correct’ to trust each other. Boix and Posner (1998) provide an intuition behind this later requirement. They argue that norms and expectations of appropriate behavior induce people to comply to existing rules and regulations, even if enforcement mechanisms are absent. In layman’s terms, by trusting each other people behave in ways not to break this trust.
4.1 Social capital, financial development and financial liberalization
While literature on social capital and growth is extensive, there is surprisingly little research on the influence of social capital on financial development and financial liberalization. In this section, I will discuss a few noteworthy exceptions (Guiso, Sapienza and Zingales 2000; Calderón, Chong and Galindo 2002). Moreover, I will discuss to what extent the above mentioned influences of social capital on economic growth can be applied to the concepts of financial development and financial liberalization. This will lead to the formulation of my hypotheses.
12
a range of household’s characteristics, regional characteristics and the quality of the court system. The proposed intuition behind their results is similar to Knack and Keefer (1997). With regard to the propensity to save, Guiso et al. argue that persons living in high trust areas have less fear that a financial institution expropriates their assets. Hence, they save more. Similarly, financiers in high trust areas provide more loans as they have less fear that their money disappears. Calderon et al. (2002) find similar results in a cross-country setting. They find that countries with a higher level of social capital tend to have larger financial sectors.
The results of both studies imply that a higher level of social capital is likely to lead to a higher level of financial deepening. Similarly, social capital can also influence the efficiency of financial intermediaries. First of all, environments that are characterized by low levels of trust are considered to be more risky by financial intermediaries. Part of this risk is likely to be transferred to borrowers (depositors) in the forms of higher (lower) interest rates (Angbazo 1997; Wong 1997). Secondly, as argued before, financial institutions benefit from social capital in the sense that it facilitates higher productivity, because managers in high trust areas need to spend less time monitoring their employees (La porta et al., 1997a) and can hire personnel based on experience and skills (Knack and Keefer, 1997). Based on the above, I expect that social capital has a direct influence on both financial deepening and financial efficiency. My first hypothesis thus becomes:
Social Capital has a positive effect on financial development (H1)
13 a
I test for a direct effect of social capital on financial development (H1). Second, I test whether social capital influences the relation between financial liberalization and financial development (H2).
Similar to the role played by formal rules and institutions, I hypothesize that social capital can also function as a prerequisite to financial liberalization success. The intuition behind this proposition perhaps is most clear by considering environments where the quality of formal institutions is poor. As both financiers and clients now cannot rely on institutional quality, the way in which they respond to financial liberalization may depend on trust rather than on the quality of formal institutions. For example, individuals may only choose to increase their savings rate if they have enough trust that their funds are being held responsibly. Similarly, on the supply side, banks may only find proper investment opportunities for their increased availability of funds (after the financial sector is liberalized) if the prevailing level of social capital is high enough. Likewise, the extent to which financial liberalization indeed allows banks to become more efficient can also depend on the prevailing level of social capital, as managers in high social capital countries need to spend less time monitoring their employees (La porta et al., 1997a). Finally, another effect of competition among banks is that valuable information between borrowers and lenders may be lost if borrowers decide to switch between banks in response to financial liberalization. To the extent that social capital prevents people from ending long-lasting relationships with their bankers, one would thus expect that financial
Table 1. Flow chart of hypotheses 1 and 2 a
Financial Liberalization
Financial Development Formal
institutions Social capital
H1
14
liberalization promotes efficiency more in high social capital environments. Based on the above, I hypothesize that:
The effect of financial liberalization on financial development is conditional on the prevailing level of social capital (H2)
4.2 The relation between social capital, the Washington-consensus and institutional quality
While the intuition behind hypothesis 2 is most obvious if one considers environments where the institutional quality is poor, the conditional effect of financial liberalization on financial development may -at least theoretically- even be present in countries with a high quality of formal institutions. This would merely imply that in those countries, social capital and the quality of formal institutions are complements. However, the existing literature suggests that social capital is especially important when the quality of formal institutions is low. Guiso et al. (2000) for example find that the effect of social capital on financial deepening is stronger in regions that have weaker law enforcement or a larger fraction of low-educated people, who have trouble understanding the value of law enforcement. Calderon et al. (2002) find a similar result in a cross-country setting. They find that the effect of social capital on financial development is lower when the quality of law enforcement is higher. In light of the results of these two studies, it may be logical to expect that social capital influences financial development more when the quality of formal institutions is low (hypothesis 1). Likewise, one may expect that the conditional effect of financial liberalization on financial development is stronger if the quality of formal institutions is low (hypothesis 2). In other words, my hypotheses may only be confirmed when formal institutions are of poor quality.
15
To summarize, I expect that both my hypotheses are more likely to be accepted in the post Washington-consensus period and when the quality of formal institutions is low. In order to test this conjecture, I will thus distinguish between the pre- and post-Washington-consensus period as well as between countries with low and high institutional quality in my methodology.
Table 2 Financial liberalization over time for the whole sample (a), developing countries (b) and developed countries (c). a
a The sum of financial liberalization is measured by adding up the value of Abiad’s (2010) liberalization index (which can take on values between 0 and 15) for all- (a), all developing- (b) and all developed-countries (c) per 4 year period.
V. Methodology
In order to test the above-mentioned hypotheses, I will adopt the following econometric model:
𝐺𝑟𝑜𝑤𝑡ℎ 𝑜𝑓 𝐹𝐷𝑡,𝑡−4𝑖 = 𝛽1𝑖+ 𝜌1𝐹𝐷𝑡−5𝑖 + 𝜌2𝐹𝑖𝑛𝑙𝑖𝑏𝑡−5𝑖 + 𝜌3𝑆𝐶𝑡−5+ 𝜌4𝑆𝐶 ∗ 𝐹𝑖𝑛𝑙𝑖𝑏𝑡−5𝑖 + 𝜌5𝑋𝑡−5𝑖 + 𝜀𝑡𝑖 (1)
where 𝐹𝐷 refers to financial development measured in terms of financial deepening and financial efficiency, 𝐹𝑖𝑛𝑙𝑖𝑏 refers to financial liberalization, 𝑆𝐶 refers to social capital, 𝑆𝐶 ∗ 𝐹𝑖𝑛𝑙𝑖𝑏 is an interaction term between social capital and financial liberalization and 𝑋 is a vector of control variables.
The model is specified as a growth on levels regression with non-overlapping data, similar to the specification of Chinn and Ito (2006). I sample the data every four years between 1973 and 2008, and use the four-year average growth rate of the level of financial development as the dependent variable and the
16
level conditions at the end of the of the previous period for the explanatory variables.10 One of these explanatory variables is the level of financial development, which is included because there may be a process of convergence in the growth rate of financial development across countries. As such, leaving out the initial level of financial development would create an omitted variable bias. By taking the average growth rate of financial development and taking non-overlapping data, the dependent variable is ensured to be stationary, preventing any results to be driven by a potential spurious relation between the dependent and the regressors.11 Similarly, by regressing financial development growth on lagged level values of the independents, problems associated with simultaneity between financial development and the independent variables are circumvented.
Since a couple of my independent variables are time-invariant, I would ideally use a specification that allows time-invariant variables to be included (e.g. a pooled or random effects specification), as this would mean that both my hypotheses could be tested with the same model. However, the error term of pooled OLS regressions show positive serial correlation. While the errors may be corrected by clustering the error terms, several econometric tests provide evidence that this serial correlation is caused by time-invariant unobserved factors that differ across countries. Moreover, a Hausman test provides evidence that these unobserved factors are correlated with my independents. 12 Using a pooled OLS model or a random effects model in this case would lead to biased and inconsistent estimates. Hence, model 1 is specified as a fixed effects model. This means that 𝜌3 is omitted, such that only my second hypothesis can be tested with this model.
In this model, I am thus primarily interested in the coefficient 𝜌4. Technically, the marginal effect of
financial liberalization on financial development growth can be written as 𝑑𝐹𝐷𝑔𝑟𝑜𝑤𝑡ℎ
𝑑𝐹𝑖𝑛𝑙𝑖𝑏 = 𝜌2+ 𝜌4∗ 𝑆𝐶.
13 Since 𝑆𝐶 is always positive, a positive coefficient 𝜌4 thus indicates that the effect of financial liberalization
on financial development growth is stronger for higher levels of social capital is, which supports hypothesis 2. However, as Brambor, Clark and Golder (2006) point out, this coefficient in front of the interaction term (𝜌4) should not be interpreted in isolation, as the total effect of financial liberalization on financial
development can be insignificant for a range of values of 𝑆𝐶, even if a significant and positive 𝜌4 is found.
Therefore, I will also graph the total marginal effect for financial liberalization for different values of social capital. Moreover, as already alluded to above, I will estimate the model for different subsamples, as it may
10 By selectin 4 year periods, I can include more of the available data compared to 5-year periods. Moreover, this
way 2008 is the last year in the sample, such that the post-crisis years are omitted. 11 Please refer to Appendix B for the results of the unit root tests.
12 Please refer to appendix B for a description and results of the econometric tests used.
13 Similarly, 𝑑𝐹𝐷𝑔𝑟𝑜𝑤𝑡ℎ
17
be the case that hypothesis 2 can only be confirmed in the post Washington-consensus period or when the quality of formal institutions is low. These different subsamples are displayed in table 3.
This distinction in institutional quality is made by using several variables that measure the quality of formal institutions. These variables include the rule of law, voice and accountability, government effectiveness, control over corruption and regulatory quality (World Governance Indicators) and the quality of banking regulation and supervision (Abiad 2010). Rather than controlling for all these variables individually, I perform a principal component analysis (PCA) to retrieve one (principal) component that captures as much of the common variation among these variables as possible (Jolliffe, 2002).14 This principal component functions as my proxy of the quality of formal institutions. Subsequently, I subdivide the sample in a group with a low quality of formal institutions (those with principal component below the mean) and a high quality of formal institutions (principal components above the mean).
In addition to model 1, I also estimate a cross-section model in order to test hypothesis 1. This is a model that is very similar to the model of Calderon et al. (2002), who test for the effect of social capital on financial development in a cross-country setting. The model reads as follows:
𝐹𝐷𝑖 = 𝛽𝑖+ 𝜌2𝐹𝑖𝑛𝑙𝑖𝑏𝑖+ 𝜌3𝑆𝐶𝑖+ 𝜌4𝑋𝑖+ 𝜌5𝑍𝑖 𝜀𝑖
(2)
where the variables have their usual meaning. Only 𝑍 is a new variable in this specification. This is a vector of time-invariant control variables. To compare the effect of social capital on financial development in both the pre Washington-consensus period and the post Washington-consensus period, I will estimate this model
14 In my case, the principal component already explains 80 percent of the variation. Please refer to appendix C for a
description of the principal component analysis.
Table 3. Subsamples for model (1)
All countries Countries with a high quality of formal institutions
Countries with a low quality of formal institutions All years
(1973-2008)
A C E
Post Washington Consensus (1992-2008)
18
for two samples, from 1984 to 1988 and from 2000 to 2004.15 As can be seen in model (2), there is no time dimension, such that I only look at the cross-section. Yet, in an attempt to reduce simultaneity problems, GDP and TRADE are measured as initial values, similar to Calderon et al. (2002). The other variables are measured as averages over the relevant time period.
Compared to model 1, this specification has a few deficiencies. First and foremost, although I control for other variables that can influence financial development, I cannot exclude the possibility that the results are driven by unobserved time-invariant factors. The results should thus be interpreted with caution. Second, by just looking at the cross-section, a substantial amount of observations is lost, which can sometimes make drawing inference on the basis of the results of these models questionable. Third, one could argue that the interaction term (𝑆𝐶 ∗ 𝐹𝑖𝑛𝑙𝑖𝑏) tells us little about whether social capital is of influence in the process of financial liberalization, which would require looking at the within-dimension only. As this interaction term also correlates highly with the 𝑆𝐶 and 𝐹𝑖𝑛𝑙𝑖𝑏 variable (see the data section), this variable is omitted in this model.
Despite these cautionary notes, there are three reasons I want to estimate a cross-sectional model. First, it allows me to estimate the direct effect of social capital on financial development. As social capital is time-invariant, this effect can only be estimated in a cross-country setting. Second, because there is no time dimension in the model, the level of financial development (rather than the growth rate) can be used as the dependent variable, which is a way to test the robustness of the specification of model 1. Third, the model allows for the inclusion of time-invariant controls (in the 𝑍-vector), some of which have been identified in literature as important determinants of financial development.
Again, I estimate the model for different subsamples, to assess whether social capital is especially of influence in the post Washington-consensus period or when the quality of formal institutions is low. The subsamples for model 2 are displayed in table 4. 16
15 The averages are taken over 4-year periods (1985-1988 and 2001-2004), whereas the initial values are taken at
1984 and 2000. The reason I take 4-year periods is that it allows for more variability between countries compared to averaging over longer periods. However, as the choice for 4-year periods is to some extent arbitrary, I will also consider longer time-periods in the robustness analysis.
16 As can be seen from the table, I do not estimate the model for samples including only countries with a high quality
19
5.1 Relevant control variables
Apart from controlling for institutional quality (which I essentially do by considering different subsamples), it is important that the above specified 𝑋- and 𝑍-vectors contain the right control variables.17 To identify these controls, I turn to literature on the determinants of financial development (Huang 2011), as well as to similar studies that aim to measure the effects of financial liberalization or social capital on financial development (Calderon et al. 2002; Chinn and Ito 2006). Huang (2011) uses both a bayesian model averaging approach and a general-to-specific approach to filter out the most important determinants of financial development. He finds that a combination of institutional-, economical-, geographical- and policy-variables are the most important determinants of financial development. When permitted by the specification of my model (if the variables vary over time), these variables are concluded as controls in the 𝑋-vector, otherwise they are included in the 𝑍-vector. 18
First, the vector includes initial GDP (GDP), as a two-way relation between financial development and economic growth is expected; a higher level of initial GDP thus is expected to positively contribute to financial development. Second, it contains the level of trade to GDP (TRADE), as trade openness is expected to positively influence financial development. Third, the inflation rate (INFLATION) is included because high inflation rates may discourage financial intermediation. Moreover, a high average rate of inflation often is a sign of the inflation rate being more volatile, which hampers financial development (Chinn and Ito, 2006). All in all, a negative relation between inflation and financial development is expected. Fourth, the population size (POPULATION) is included, because small countries tend to have much higher rates of liquid liabilities to GDP and private credit to GDP (Huang, 2011). Fifth, the vector contains variables that capture institutional development. These variables are the extent to which the country functions as a
17 One could argue that taking different subsamples alone does not completely control for formal institutional quality,
as the quality of formal regulation may still be of influence in the different subsamples. I address this issue in the robustness analysis.
18 In the description of the variables, I refer to ‘the control-vector’ in general, the descriptive statistics table (table 6) provides an overview of which variables are included in which specific vector (X or Z).
Table 4. Subsamples for model (2)
All countries Countries with a low quality of formal institutions
1984-1988 A C
20
democracy (DEMOC) and a variable that measures the existence of political constraints (POLCON) that prevent policy changes from being implemented. The former variable (DEMOC) is retrieved from the Polity IV database and is often used in the political economy literature. The latter (POLCON) is a less commonly used variable. It is retrieved from a database compiled by Henisz (2002), and is used in the control vector because Huang (2011) identified it as one of the most robust determinants of financial development. For both variables, a higher score on the index (i.e. becoming more a democracy or facing less political constraints) is expected to positively contribute to financial development. Further, Huang (2011) finds that it matters for financial development whether the legal origin of a country is British (LEGAL). This is based on the work of La porta et al. (1997b) who show that a country’s legal origin has predictive power in explaining a country’s law on creditor rights, shareholder rights, and private property rights. A British legal origin is found to promote financial development. Finally, Huang (2011) finds that language fractionalization (LANFRAC) negatively influences financial development, whereas the percentage of people that have a western European language as their first language (EURFRAC) positively influences financial development.
VI. Data
In order to test my hypotheses, I make use of a panel dataset consisting of 82 countries for the time period of 1973 up to and including 2008.19 I make use of data provided by the world bank (Čihák et al., 2012) to proxy for financial development. More specifically, I measure financial deepening with financial system deposits to GDP (DEPGDP), private credit to GDP (PRCGDP), and liquid liabilities to GDP (LLY). Financial efficiency is measured with overhead costs to total assets (OVERHEAD), the net interest margin (INTEREST) and the lending-deposit spread (LEN-DEP), where an increase in these variables should thus be interpreted as an increase in inefficiency.
The main explanatory variables are social capital and financial liberalization. Social capital is measured with data retrieved from the World Values Survey Association (WVS). The WVS is a compilation of national surveys on values and norms, carried out in six time waves (1981-1984, 1990-1993, 1995-1997, 1999-2004, 2005-2009 and 2010-2014. To measure social capital, I use the first five waves, from which I use one specific question. Namely the question: “Generally speaking, would you say that most people can be trusted or that you can't be too careful in dealing with people?”, where respondents (a minimum of thousand per time wave per country) can choose among the options “Most people can be trusted”, “You cannot be too careful”, or “Don’t know”. For those countries that were not included in any of these waves,
21
I made use of the institute of social studies and the economic and social data service, which are organizations that includes the same question in their surveys.20 In order to be able to include the data in this research, I follow a common procedure in the existing literature by excluding the non-respondents and subsequently calculating the proportion of people that answered the question with “Most people can be trusted” (Knack and Keefer 1997; Calderon et al. 2002; Kouvavas and ten Kate 2013). In cases where the same country was included in multiple waves, I subsequently calculated the average over time. I then assume that this average level of trust for each country describes a country’s level of trust in the period from 1973 to 2008. This assumption is based on theory (as social capital is assumed to change very slowly) as well as on the data, as the average correlation between different WVS waves is greater than 0.8 (see appendix A, table A2).
The second main explanatory variable is financial liberalization, which I measure by making use of a recently developed dataset by Abiad (2010). Whereas traditional research often used dichotomous variables that only distinguish liberalized financial sectors from unliberalized sectors, Abiad’s dataset includes 5 different dimensions of financial liberalization.21 According to Abiad, financial liberalization occurs if policy makers can reduce or remove (1) restrictions on international capital flow, (2) credit controls and excessively high reserve requirements, (3) entry barriers, (4) state ownership in the banking sector and (5) interest rate controls. Each country in the dataset is rated (each year) on a scale from 0 to 3 for these five dimensions, where 0 refers to complete repression and 3 refers to a completely liberalized financial sector (concerning that specific dimension). Subsequently, I take the sum of these 5 dimensions, so that I end up with a financial liberalization variable that can take on values in the range of 0 to 15.
An overview of the variables and their respective sources can be found in table 5. Table 6 provides descriptive statistics for those variables. Finally, table 7 is a correlation matrix of all variables that are included in the regressions.22 From table 6, it can be seen that several variables may potentially create a multicollinearity problem if they are included in the same regressions. First, the social capital variable and the financial liberalization variable correlate highly with the interaction term. While this appears problematic, the social capital variable is omitted in model 1. The financial liberalization variable is included in model 1, but correlates less with the interaction term (0.47 instead of 0.94). In model 2 the interaction
20 The economic and social data service allows respondents to rate their answer on a scale from 1 to 9. I rescaled the
answers by taking the proportion of respondents that answered the question with a 1, 2, 3 or 4. In the robustness analysis, I will test whether the omission of the Eurobarometer countries significantly alters the results.
21 Abiad’s dataset actually describes financial reforms, which is similar to -but not the same as- financial
liberalization. His dataset contains 7 dimensions, 2 of which I leave out. Firstly, I leave out a measure of securities market policies, as I will not discuss stock market development in this thesis. Secondly, I leave out a measure of regulation and supervision, as this does not measure financial liberalization. The latter is included in my model, but not as a measure of financial liberalization (see the methodology section).
22
term is omitted, such that multicollinearity between the social capital variable and the financial liberalization variable and the interaction term cannot be a problem in this model.
23 Table 5. Data description and sources
Short Definition Source
Dependents Financial deepening
LLY Liquid liabilities to GDP (%) Global Financial Development Database (GFDD)
DEPGDP Financial system deposits to GDP (%) GFDD
PRCGDP Private credit by deposit money banks and other financial institutions to GDP (%) GFDD
Financial efficiency
OVERHEAD Overhead costs to total assets (%) GFDD
LEN-DEP Lending-deposit spread (%). GFDD
INTEREST Net interest margin (%) GFDD
Independents
SC The average proportion of people within a country that have answered “most people can be trusted” to the following question: Generally speaking, would you say that most people can be trusted or that you can't be too careful in dealing with people?
World Values Survey
FINLIB Measures the existence of credit controls, interest rate controls, entry barrier in the financial sector, state ownership in the banking sector and restrictions on international capital flows, see below.
Abiad (2010)
Credit controls and reserve requirements
Measures whether there are ceilings on credit towards certain sectors, whether there are high reserve requirements and whether there is directed credit towards favored sectors or industries.
Abiad (2010)
Interest rate controls Measures whether the government imposes interest rate controls, either directly or by means of interest rate floors, ceilings or interest rate bands.
Abiad (2010) Entry Barriers Measures whether there are licensing requirements for newly established domestic
financial institutions, restrictions on certain banking practices and entry barriers for foreign banks.
Abiad (2010)
State ownership in the banking sector
Measures the share of banking assets controlled by state-owned banks. Abiad (2010)
Restrictions on international capital flows
Measures whether there are capital account controls and restrictions, transaction taxes and whether multiple exchange rates are used.
Abiad (2010)
Control variables X-vector
GDP Total gross domestic product. GFDD
INFLATION Yearly inflation rates. Inflation rates above 100% and below -100% are excluded. GFDD TRADE The ratio of the sum of exports and imports to GDP GFDD
POPULATION The total size of the population. GFDD
DEMOC An index, ranging from 0 to 20, that measures the extent of democracy, where 0 refers to a full autocracy and 20 refers to a full democracy.
Polity IV Database POLCON Index that estimates the existence of political constraints. It considers various features
of the legislative, executive and judicial branches of government and measures the overall ability of these underlying political structures to support credible policy commitments.
Henisz (2002)
Z-vector
LEGAL Dummy variable that measures whether a country has a British legal origin. La Porta et al. (1997b) LANFRAC Measures the shares of languages spoken as “mother tongues”. Alesina et al. (2003) EURFRAC Measures the share of the population that speaks either German, French, Spanish,
English or Portuguese as their first language.
Hall and Jones (1998)
Additional variables (used in Principal component analysis)
World Governance Indicators These aggregate indicators combine the views of a large number of enterprise, citizen and expert survey respondents to measure a country’s government effectiveness, voice and accountability, control over corruption and regulatory quality.
World Governance Indicators
Banking regulation and supervision
Measures the independence of the banking supervisory agency, whether risk-based capital adequacy ratios based on the Basel standard are adopted and the coverage and conduct of supervisory oversight.
24 Table 6. Descriptive statistics
Variable N Mean SD Median Min Max
Dependents Financial deepening LLY 2470 0.50 0.36 0.42 0.04 2.94 DEPGDP 2446 0.42 0.34 0.33 0.00 2.85 PRCGDP 2468 0.47 0.41 0.30 0.00 2.28 Financial efficiency INTEREST 945 0.05 0.04 0.04 -0.06 0.41 LEN-DEP 1632 0.08 0.08 0.05 0.00 0.92 OVERHEAD 946 0.04 0.03 0.03 0.00 0.30 Independents SC 2819 0.26 0.15 0.22 0.07 0.75 FINLIB 2557 8,18 4,17 8,75 0.00 15.0 Credtit controls 2557 1.62 1.11 1.50 0.00 3.00
Interest rate controls 2557 1.79 1.33 3.00 0.00 3.00
Entry barriers 2557 1.80 1.19 2.00 0.00 3.00
Privatization 2557 1.28 1.19 1.00 0.00 3.00
International capital flows 2557 1.69 1.13 2.00 0.00 3.00
Control variables
X-vector
GDP 2777 2.88e+11 9.84e+11 4.10e+10 6.80e+08 1.40e+13
INFLATION 2560 0.12 0.15 0.08 -0.11 1.00
TRADE 2678 0.66 0.50 0.55 0.06 4.40
POPULATION 2818 5.58e+07 1.57e+08 1.50e+07 1.30e+06 1.30e+09
DEMOC 2818 13.59 6.85 17.00 0.00 20.00 POLCON 2772 0.30 0.21 0.36 0.00 0.72 Z-vector LEGAL 2818 0.31 0.46 0.00 0.00 1.00 LANFRAC 2746 0.34 0.29 0.25 0.00 0.92 EURFRAC 2623 0.33 0.42 0.01 0.00 1.00
Additional variables (used in Principal component analysis)
Rule of law 2818 0.20 1.03 -0.01 -1.43 1.94
Voice and accountability 2818 0.22 0.91 0.01 -1.85 1.62
Government effectiveness 2818 0.35 1.00 -0.02 -1.05 2.14
Control of corruption 2818 0.28 1.11 -0.13 -1.16 2.44
Regulatory quality 2818 0.35 0.92 0.22 -1.74 1.97
25 Table 7. Pair wise correlation matrix a
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] Dependents [1] LLY 1.00 [2] DEPGDP 0.94 1.00 [3] PRCGDP 0.85 0.87 1.00 [4] INTEREST -0.55 -0.55 -0.57 1.00 [5] LEN-DEP -0.38 -0.36 -0.39 0.45 1.00 [6] OVERHEAD -0.56 -0.53 -0.53 0.64 0.53 1.00 Independents [7] SC 0.38 0.34 0.50 -0.53 -0.42 -0.48 1.00 [8[ SC*FINLIB 0.34 0.37 0.52 -0.49 -0.41 -0.44 0.94 1.00 [9] FINLIB 0.11 0.24 0.28 -0.07 -0.22 -0.05 0.19 0.47 1.00 Controls X-vector [10] GDP Log 0.52 0.51 0.60 -0.50 -0.25 -0.48 0.41 0.40 0.14 1.00 [11] INFLATION Log -0.53 -0.50 -0.51 0.41 0.30 0.49 -0.34 -0.35 -0.23 -0.33 1.00 [12] TRADE 0.41 0.44 0.34 -0.16 -0.26 -0.21 0.02 0.08 0.21 0.01 -0.24 1.00 [13] POPULATION Log 0.06 -0.03 -0.03 0.01 0.02 -0.10 -0.14 -0.31 -0.45 0.44 0.09 -0.32 1.00 [14] DEMOC -0.02 0.11 0.20 -0.19 -0.04 0.01 0.27 0.39 0.36 0.31 0.09 -0.08 -0.26 1.00 [15] POLCON 0.15 0.22 0.25 -0.11 0.01 -0.02 0.14 0.24 0.30 0.20 -0.03 -0.04 -0.19 0.52 1.00 Z-vector [16] LEGAL 0.12 0.20 0.21 0.05 -0.15 -0.09 -0.12 -0.09 -0.01 -0.02 -0.05 0.19 0.13 -0.12 -0.10 1.00 [17] LANFRAC -0.08 -0.04 -0.04 0.30 -0.04 0.16 -0.27 -0.23 0.07 -0.26 -0.02 0.09 0.13 -0.27 -0.06 0.47 1.00 [18] EURFFRAC 0.00 0.07 0.12 -0.10 0.27 0.10 -0.07 0.02 0.16 0.28 0.05 -0.23 -0.06 0.36 0.12 -0.12 -0.34 1.00
26 VII. Results
Before discussing the results of the models in detail, it should be noted that in most of the regressions using financial efficiency as the dependent variable, the key variables (social capital, financial liberalization, and the interaction term) are significant. This also does not change if I start looking at different subsamples. No matter which subsample I consider, the key variables in the financial efficiency regressions remain insignificant.23
Partly, this can be due to the fact that there are data limitations for the financial efficiency variables. As can be seen in table 6, the average number of observations for the efficiency variables is less than half of that of the deepening variables. On top of that, there is little variation in the financial efficiency variables. In that respect, it may come as no surprise that few variables are significant in the efficiency regressions. Another explanation for a lack of results in the efficiency regressions is that one could question what the efficiency variables really measure. The expectation that efficiency leads to a reduction in profit margins (especially the lending-deposit spread and the net interest margin essentially measure profitability) implicitly assumes competitive and homogeneous markets. However, banking markets may be far from competitive, such that banks may become more efficient but at the same time are able to retain high profit margins. 2425
Considering the financial deepening regressions for model 1 (see appendix D), it can be seen that none of the key variables are significant if I consider only countries with high institutional quality (subsample C and D). Looking at samples including both countries with high and low institutional quality (subsamples A and B), it is noteworthy that as soon as I consider post Washington-consensus years only, the interaction term is already becoming positive and significant for all the regressions (subsample B, table D2). This can be considered evidence that financial liberalization success (in terms of stimulating financial deepening) was conditional on the prevailing level of social capital in the post Washington-consensus period. Similarly, a couple of the key variables also become significant if I consider all years and countries with poor institutional quality only (subsample E, table D5). This can be interpreted as evidence that social capital is
23 There is also no evidence of an adverse relation between the key variables and financial efficiency. In neither of
the regressions either social capital, financial liberalization or the interaction term is significant with the wrong sign. In subsample F (table E6), the interaction term negatively influences inefficiency (hence positively influences efficiency). However, in light of the various regressions performed, this one variable that behaves as expected can hardly be interpreted as evidence in favor of my hypothesis.
24A better way to measure efficiency would be estimating a cost frontier at the bank level by means of a stochastic
frontier analysis, as performed in Hermes and Meesters (2015). This way, inefficiency is measured as deviations from the statistically derived theoretical best-practice frontier, which controls for omitted factors such as input prices and other market forces.
25Due to the insignificant results for the efficiency regressions, I will only discuss the deepening regressions in the
27
especially of influence when formal institutions are of poor quality. In conclusion, both the time restriction (post Washington-consensus years only) as well as the institutional quality restriction (low institutional quality only) appears to be relevant.
However, when I combine both the post Washington-consensus restriction and the institutional quality restriction (subsample F, table D6), the results become most clear. 26 First, the lagged values of financial development have their predicted sign, indicating that indeed there are diminishing returns in the process of financial deepening. That is, the higher level of financial deepening, the more difficult it becomes to achieve high growth rates of financial deepening. Also the lagged level of GDP has the expected sign, indicating that a higher lagged level of GDP contributes positively to financial deepening. The other control variables are all insignificant. Second, the coefficient for financial liberalization is negative and significant. This perfectly fits in the existing literature, where it is argued that a high quality of formal institutions is a prerequisite to financial liberalization. This effect appears to be picked up by the negative signs for the financial liberalization variable in this sample. Third, the interaction term is positive and significant for all three regressions. This can be interpreted as evidence in favor of hypothesis 2, namely that the effect of financial liberalization is conditional on the prevailing level of social capital. Table D11, D12 and D13 in appendix D illustrate the total marginal effect of financial liberalization on financial development graphically. These tables clearly show that this marginal effect turns from being negative and significant to positive and significant as the level of social capital increases.
As argued above, this may be because social capital and formal institutions are substitutes. Table D14, D15 and D16 in appendix D provide evidence for this conjecture. These tables show, for the post Washington-consensus period, how the interaction effect changes when I move from a sample consisting of countries with very poor institutional quality to countries with very high institutional quality. These tables clearly show that the interaction effect becomes weaker when the quality of formal institutions increases, and that the interaction term is significant and positive for samples with low institutional quality. This can be considered evidence that social capital can take over the role of formal institutions when the latter are of poor quality.
While model 1 thus provides evidence in favor of hypothesis 2, this model cannot be used to test hypothesis 1. Therefore, I now turn to the model 2 results. The results of the cross-section regressions appear to tell a similar story as the results for model 1. Again, few of the key variables are significant if I do not distinguish
26 These results do not change if the democracy index is excluded. Neither the significance nor the signs of any of the
28
between countries with high and low institutional quality. In the 1984-1988 period, social capital is then even significant (at the 10 percent level) with a negative sign in one of the regressions (see table D7).
However, once I look at countries with a low quality of formal institutions (table D9 and D10), the coefficient becomes positive for nearly all regressions, and is positive and significant in half of the cases.27 In none of the cases is it significant with the wrong sign. This indicates that generally, countries that have a higher level of social capital indeed have deeper financial systems. The control variables in all regressions are either significant with the predicted sign or insignificant. The only exception is the democracy variable, which turns out negative and significant in a few instances. This would imply that democracies are not necessarily better in stimulating financial development compared to autocracies. All in all, the model 2 results hint at a weak direct effect of social capital on financial deepening.
Summarizing the results for models 1 and 2, it appears that the effect of financial liberalization on financial deepening indeed is conditional on the prevailing level social capital, which confirms hypothesis 2. This conditionality is most profound in the post Washington-consensus period or when the quality of formal institutions is low. When the quality of formal institutions increases, social capital no longer is of significant influence in determining financial liberalization success. Although the results for model 2 are less strong, it provides evidence in favor of hypothesis 1. Again, it appears that both the post Washington-consensus distinction and the distinction between low and high institutional quality matters.
An intuitive explanation for the fact that these distinctions are so relevant, is that financial liberalization among developing countries (with a low institutional quality) accelerated from the 1990s onwards. These countries acted upon the advice of the Washington-consensus, which stipulated that developing countries could benefit from liberalizing their financial sectors (Gore, 2000). However, as these countries did not have the proper institutional environment, financial liberalization often failed to promote financial development for those countries. This explanation would be in line with other studies on financial liberalization, which have identified institutional quality as an important prerequisite to financial liberalization. However, the results of this study suggest that social capital can be a substitute for formal institutional quality. Consequently, some countries managed to benefit from financial liberalization in the post Washington-consensus period, despite the low quality of their formal institutions.
VIII. Robustness Analysis
