Financial structure and economic growth

Abstract

Recent empirical work on financial structure and economic growth analyzes multi-country dataset in panel and/or cross-section frameworks and concludes that financial structure is irrelevant. We highlight their shortcomings and re-examine this issue utilizing a time series and a Dynamic Heterogeneous Panel methods. Our sample consists of fourteen countries. Tests reveal that cross-country data cannot be pooled. Financial structure significantly explains output levels in most countries. The results are rigorously scrutinized through bootstrap exercises and they are robust to extensive sensitivity tests. We also test for several hypotheses about the prospective role of financial structure and financial development on economic growth.

Introduction

The debate on the relative merits of bank-based versus market-based financial systems has a long history of over a century (Gerschenkron, 1962, Allen and Gale, 2000, Levine, 2002). Nonetheless, there is hardly any consensus at the theoretical level. Competing theoretical models posit the superiority of one type of financial system over the other or they simply relegate financial structure as irrelevant. On the one hand, Gerschenkron (1962), Diamond (1984), Stiglitz (1985), Boyd and Prescott (1986), Bencivenga and Smith (1991), Bhide (1993), Stulz (2002), to name but a few, argue that the bank-based system is superior to the market-based one. On the other hand, Levine (1997), Boyd and Smith (1998), Holmstrom and Tirole (1993), Jansen and Murphy (1990), Boot and Thakor (1997), Wenger and Kaserer (1998), among others, suggest the opposite. Still, Merton and Bodie (1995) and Levine (1997) maintain that it is neither the banks nor the markets; instead, it is the provision of overall financial services that is crucial in promoting growth. Similarly, Huybens and Smith (1999) underline the complementarities between banks and markets in the provision of financial services. The theoretical debate on financial structure culminates into four distinct views: the bank-based, the market-based, the financial services and the law and finance. We briefly discuss them in Section 2.

A large body of empirical literature has attempted to evaluate this debate. Early studies focus on UK and the US as market-based systems versus Japan and Germany as bank-based systems (e.g. Goldsmith, 1969, Hoshi et al., 1991, Allen and Gale, 2000, Mork and Nakkamura, 1999, Weinstein and Yafeh, 1998, Arestis et al., 2001). They rigorously compare and contrast the country-specific financial structure, that is, an assortment of financial markets, instruments and intermediaries in operation, and conclude that financial structure is important for economic growth. However, Goldsmith (1969), highlighting their shortcomings, argues that these four industrialized countries have resembling real per capita income levels and they historically share similar growth rates. Consequently, it is hard to attribute their analogous growth rates to alternative forms of either the bank-based or the market-based financial system. Similarly, Beck and Levine (2002) and Levine (2002) assert that although UK, US, Germany and Japan did experience periods of divergent growth rates, nonetheless, “it is very difficult to draw broad conclusions about bank-based and market-based financial systems from only four countries” (Beck and Levine, 2002, p. 148). They argue that the empirical assessment of the role of financial structure should be based on broad dataset that encompasses wide-ranging national experiences.

Indeed, Beck and Levine (2002), Levine (2002) and an edited volume by Demirguc-Kunt and Levine (2001), among others, analyze multi-country dataset at firm-, industry- and aggregate-levels, employing different econometric methods under the panel and/or pure cross-country setup. They are some of the most influential work in recent years that directly investigate the role of financial structure in industrial expansion, economic growth and the sources of growth. Their broad conclusions are that the overall level of financial development (the financial services view) and legal system efficiency (the law and finance view) are important; however, financial structure (either the bank-based or the market-based view) is irrelevant. The overall financial development exerts significant and economically large effect on economic growth but there is no cross-country empirical support for financial structure. These results are shown to be robust to various sensitivity tests, namely, data, specifications and econometric methods.

These studies are immensely important because they offer broad insights on the issue. They are not without concerns however. Demirguc-Kunt and Levine (2001) concede that the result of economic performance being impervious to financial structure does not necessarily imply that institutional structure is of no consequence to growth. Instead, it may simply indicate that either there is not one optimal institutional structure, which fits everywhere and at all times, or the indicators used in the literature may not satisfactorily capture the roles of banks and markets. Likewise, Levine and Zervos (1996, p. 325) state that panel regressions mask important cross-country differences and suffer from ‘measurement, statistical, and conceptual’ problems. Quah (1993) shows the difficulties associated with the lack of balanced growth paths across countries when pooling data (see also, Caselli et al., 1996). Pesaran et al. (2000) point out the issue of parameter heterogeneity across panel units (countries) and show that unless this (heterogeneity) is addressed panel estimates become biased and inconsistent. Luintel and Khan (2004) show lack of correspondence between panel and country-specific estimates, hence, the generalizations based on panel results, i.e., the very ‘broad conclusions’, may proffer incorrect inferences for several countries, industries or firms of the panel.

If the problems of information asymmetries, moral hazards and adverse selection were not acute in financial markets and financial institutions were operationally efficient (a feasible scenario if financial institutions are of sufficiently high quality) then either form of financial system (market-based or bank-based) should, in principal, provide just about the same financial services for augmenting growth. Financial structure, in this scenario, would be irrelevant. However, the reality is far from it. Countries exhibit different ‘states of the world’, namely, they have different production structures, levels of banking, financial and capital market development. These structural makeups tend to be rigid requiring significant amount of time and effort for any change. Thus, different ‘states of the world’ may require different financial arrangements to cater for the diverse financial needs.

All the above concerns imply that panel estimates may be misleading at country, industry and firm level; consequently, their policy relevance may be limited. The neglected parameter heterogeneity across firms, industries and countries may bias the estimates. Further, the panel regressions, which do not address the cross-country heterogeneity, may dilute the likely effects that emanate from different ‘states of the world’. Our aim is to address these issues. We conduct country-by-country time-series analyses for 14 countries by utilizing the World Bank dataset on financial development and financial structure, which is recently updated to 2005 (Beck et al., 2000). The number of countries is dictated by the availability of sufficiently long-time series (see Section 5). Our sample consists of low- and middle-income countries with varied growth experiences.

This paper complements the existing empirical literature by way of new results based on rigorous country-by-country time-series analyses. We also apply a Dynamic Heterogeneous Panel Estimator so that we can (i) compare our results with the existing empirical literature, and (ii) perform tests of equivalence between the time-series and panel estimates. Our approach does not suffer from the ‘measurement and conceptual errors’, arising from different definitions and accounting and measurement practices across countries; nor does it mask ‘important cross-country difference’, issues raised by Levine and Zervos (1996). We also address the issue of cross-country parameter heterogeneity by explicitly testing for the poolability of cross-country data, a concern raised by Pesaran et al. (2000). The equivalence between the panel and the country-specific parameters for our sample countries is also examined, a concern raised by Pesaran et al. (op. cit.) and Luintel and Khan (op. cit.).

Our basic specification augments the Cobb–Douglas production function by measures of financial structure and financial development. The long-run relationship between real per capita GDP, per capita physical capital stock, and measures of financial development and financial structure is estimated through co-integration tests. We apply the Fully Modified OLS (FMOLS) of Phillips and Hansen (1990) for our time-series analysis. FMOLS is shown to perform better in small samples. The Dynamic Heterogeneous Panel Estimator proposed by Pedroni, 1999, Pedroni, 2001 is used for panel estimates. To our knowledge, this is the first ever study of this kind, which (i) evaluates the debate both in time-series and panel frameworks and tests whether the two sets of results (time series versus the panel) are equivalent; (ii) tests for the cross-country data poolability; (iii) sheds new light on a number of relevant hypotheses regarding the roles of financial development and financial structure when countries develop both financially and economically; and (iv) scrutinizes, through extensive bootstrap exercises, whether the asymptotic approximations are valid for the finite sample estimates and their distributions. We also conduct extensive sensitivity tests vis-à-vis data (measures of financial development and financial structure) and specifications.

Our results are quite revealing. First, for the majority of sample countries, financial structure appears significant in explaining economic growth. Second, we find significant heterogeneity in cross-country parameters and adjustment dynamics; tests show that data cannot be pooled for the countries included in our sample, which reinforces the use of time-series approach. Third, tests also reveal that the panel estimates (parameters) do not correspond to country-specific estimates. Fourth, our bootstrap results provide a new and interesting insight. Asymptotic approximations tend to remain valid for the finite sample results so long as the empirical models do not utilize impulse dummies and/or interacted regressors. However, when empirical models use impulse dummy and/or interacted covariate, the distributions of empirical tests statistics do not appear symmetric. This suggests that those empirical studies, which utilize impulse dummies and/or interacted regressors, should base their inferences on suitably computed (by way of bootstrap) finite sample critical values. Our results are robust to various sensitivity tests. Overall, our findings imply that the complete absence of cross-country support for financial structure, reported by panel or cross-section studies, may be because they do not sufficiently account for the cross-country heterogeneity.

The rest of the paper is organized as follows. In the section that follows we briefly discuss the theoretical arguments; this is followed by a discussion of the existing empirical evidence in Section 3. Section 4 outlines our model specifications and the econometric methods employed. Section 5 discusses the dataset; Section 6 discusses the results pertaining to cross-country heterogeneity; Section 7 presents the main empirical results, and Section 8 discusses sensitivity tests. Finally Section 9 summarizes and concludes.