Financial Development and Source of Growth
New Evidence1

This paper examines how financial development affects the sources of growth—productivity and investment—using a sample of 145 countries for the period 1960-2011. We employ a range of econometric approaches, focusing on the CCA and MENA countries. The analysis looks beyond financial depth to capture the access, efficiency, stability, and openness dimensions of financial development. Yet even in this broad interpretation, financial development does not appear to be a magic bullet for economic growth. We cannot confirm earlier findings of an unambiguously positive relationship between financial development, investment, and productivity. The relationship is more complex. The influence of the different dimensions of financial development on the sources of growth varies across income levels and regions.

Abstract

This paper examines how financial development affects the sources of growth—productivity and investment—using a sample of 145 countries for the period 1960-2011. We employ a range of econometric approaches, focusing on the CCA and MENA countries. The analysis looks beyond financial depth to capture the access, efficiency, stability, and openness dimensions of financial development. Yet even in this broad interpretation, financial development does not appear to be a magic bullet for economic growth. We cannot confirm earlier findings of an unambiguously positive relationship between financial development, investment, and productivity. The relationship is more complex. The influence of the different dimensions of financial development on the sources of growth varies across income levels and regions.

I. Introduction and Motivation

We provide new evidence that financial development is not a magic bullet for economic development. Instead, the relationship between finance, productivity, and investment is complex, and there is no unambiguously positive relationship where “finance” boosts productivity and investment everywhere and at all times. This result seems intuitive once we acknowledge that financial development has many dimensions. Consistent with Čihák et al. (2012), we distinguish between four dimensions: (i) financial depth; (ii) financial efficiency; (iii) financial stability and openness; (iv) and access to financial services. Financial stability and stock market efficiency are the only dimensions that appear related to productivity and capital accumulation across time and countries. The effect of other dimensions depends on country characteristics, such as income level and region. For example, financial openness can boost productivity in low-income countries, but the positive effect vanishes in advanced countries. Effects can also vary over time, with financial depth, for example, contributing positively to growth in the pre-2005 period.

We also examine whether financial development exerts particular effects in the CCA and MENA region. But we find only mild evidence for differential regional effects. CCA and MENA oil importing countries may stand to gain extra benefits from increasing the efficiency of their banking sector efficiency, measured by interest rate spreads and overhead costs. This results likely reflects the generally weak level of competition in these markets (see Rocha, 2011, for example), translating into high prices for borrowers and high operational costs.

Our paper complements the existing literature in three respects. First, we extend the data to include the recent global financial crisis and a large number of countries. Specifically, the sample contains unbalanced panel data for 145 countries from 1960 to 2011. We focus in particular on the CCA and MENA countries to see if financial development in these countries differs from other regions. Second, we use 16 different indicators of the four main dimensions of financial development—depth, efficiency, stability/openness and access—to capture all aspects of the financial sector. Our results about financial development are therefore more general than earlier contributions that rely merely on some financial indicator measuring countries’ financial depth. Third, we focus on how financial development affects the sources of growth, productivity and capital accumulation, and not growth itself.

The relationship between financial development and economic growth on the theoretical level has always been controversial. Robinson (1952) and Lucas (1988) believe that financial intermediaries develop in response to demand from the real sector. On the other hand, Schumpeter (1912), Gurley and Shaw (1955), Goldsmith (1969), Greenwood and Jovanovich (1990), and others see finance as an important contributor to growth by improving resource allocation through the provision of ex-ante information on investment projects, promoting saving through risk diversification, and easing exchange through the reduction of transaction costs.

The early consensus of the empirical literature on the finance and growth nexus has, by and large, supported the positive relationship between development and growth using cross-country, time-series, and panel data, as well as industry- and firm-level studies (see Levine 2005 for a literature review). More recent evidence, however, points to a more complex relationship, which depends on the level of financial and economic development, as well as the quality of institutions. Applying a threshold regression model, Deidda and Fattouh (2002) argue there is no significant relationship between financial development and growth in low-income countries, whereas the relationship is positive and strongly significant in high-income countries. Rioja and Valev (2004a) add that this relationship varies according to the level of financial development, finding a positive and significant effect of financial development on growth only with medium and high levels of financial development. Rioja and Valev (2004b) find that finance affects growth through capital accumulation in low-income countries and through productivity growth in middle- and high-income countries.

Recent papers by Cechetti and Karroubi (2012) and Arcand and others (2012) have revisited the finance-growth nexus, showing that the level of financial development is good for economic growth only up to a point between 90 percent and 100 percent of GDP, turning negative for high-income countries. This result is consistent with “the vanishing effect of financial development” (Law and Singh, 2014). These studies suggest that the positive effect of finance on economic growth may be more nuanced, but they do not reject the prevailing consensus that finance is good for growth.

The global financial crisis has been a turning point. Using recent data up to 2010, Rousseau and Watchel (2011) and Beck and others (2012b) find a much lower effect of finance on growth than previous studies. In fact, Rousseau and Watchel (2011) find that the finance-growth relationship disappeared during the period between 1990 and 2004. They attribute the vanishing effect to financial crises related to rapid and excessive financial deepening. Arcand and others (2012) suggest that the vanishing relationship between finance and growth could be attributed to “the fact that many countries have reached the point at which financial deepening starts having a negative effect on growth.” Beck and others (2012a) explain that the vanishing effect could also be related to the increase in the share of household loans to the detriment of company loans: they find that enterprise credit is positively associated with economic growth, whereas household credit is not. By extending the sample to include the global financial crises, though on a relatively small sample of 46 countries, Bezemer and others (2014) find a high ratio of bank credit to GDP has a negative effect on growth. They suggest that this negative relationship between finance and growth is due to a shift in the share of credit away from nonfinancial institutions.

This paper is organized as follows: section 2 describes the dataset, the empirical model, and the econometric methodology; section 3 discusses the empirical findings; and section 4 concludes with some plausible factors that might explain the vanishing effect of the finance-growth nexus.

II. Data

This study utilizes available data on financial development and the sources of growth for a large number of countries between 1960 and 2011. Our dataset is limited only by source data availability. Accordingly, the number of observations across both country and time dimensions varies in each model. In line with prior work, we employ multi-year, non-overlapping averages of the available data when possible, 2 which isolates and removes business cycle effects, focusing on the relationship between each financial indicator and the sources of economic growth. This section describes the measures of sources of growth, financial development, and control variables.

A. Sources of growth

We add to the literature by decomposing economic growth using standard growth-accounting practices into total factor productivity (TFP) growth and capital stock accumulation, both of which are extracted from the Penn World Tables. This dataset offers a comprehensive global database with estimates for capital stock and TFP since 1950. The methodology used accounts for heterogeneity in labor income over time and constructs the capital stock based on decomposed assets (higher weights to fixed assets), allowing for accurate and comparable estimates of TFP for a wide array of countries over a long-time period.3

B. Financial development indicators

We examine a wide range of financial indicators to capture the four main dimensions of financial sector development—depth, efficiency, stability/openness, and access. We therefore extend the analytical approach of much of the existing literature that focuses on credit and monetary aggregates. Relying primarily on the Global Financial Development Database, which includes a wealth of financial sector indicators, we consider the effects of depth, efficiency, stability, openness, and access on the two dominant sources of growth (productivity and capital accumulation). Due to a potentially non-linear relationship between economic growth and control variables, we transform all variables into natural logarithm forms. 4 Table 1 defines each included indicator by dimension, while Figures 1 and 2 show basic relationships between the financial variables and the sources of growth

Figure 1.1
Figure 1.1

Productivity Growth and Financial Sector Indicators

Five Year Non-Overlapping Averages

Citation: IMF Working Papers 2017, 143; 10.5089/9781484302637.001.A001

Figure 1.2
Figure 1.2

Productivity Growth and Financial Sector Indicators

Three Year Non-Overlapping Averages

Citation: IMF Working Papers 2017, 143; 10.5089/9781484302637.001.A001

Figure 1.3
Figure 1.3

Productivity Growth and Financial Sector Indicators

Annual Observations

Citation: IMF Working Papers 2017, 143; 10.5089/9781484302637.001.A001

Table 1:

Data Description

article image
article image

C. Control variables

To assess the strength of an independent relationship between growth and financial development, we introduce control variables as suggested by the finance-growth literature. The logarithm of initial real GDP per capita is introduced to control for economic convergence. Average years of schooling are included to control for the level of human development. We also use the trade-to-GDP ratio, the ratio of government consumption to GDP, and oil prices to control for openness and the role of the state in the economy.

D. Empirical Methodology

The basic specification used in this paper follows the general regression model used in other studies (Levine 2005) and can be summarized as:

git=αyi,t1+βFi,t+γXi,t+ηi+λt+ϵi,t(1)

where git represents growth in either total factor productivity or the capital stock. yi,t−1 represents initial real GDP per capita and serves as a measure of the tendency for growth rates to converge across countries over time. The nexus of interest is the impact of Fi,t, the financial dimension, on each respective component of economic growth. Xi,t represents the vector of macroeconomic control variables and includes initial GDP per capita, the trade to GDP ratio, average years of schooling, and the government consumption to GDP. ηi is an unobserved country-specific effect, λt is a time-specific effect, ϵi,t is the time-varying error term, and i and t represent country and time period, respectively.

Rewriting Eq.1 using the first difference as suggested by Arellano and Bond (1991), we obtain the following equation:

Δgit=αΔyi,t1+βΔFi,t+ΔγXi,t+Δλt+Δϵi,t(2)

Although this differentiation eliminates the country specific effect, it introduces a new bias because of the correlation of the new error term with the lagged new dependent variable.

Arellano and Bond (1991) propose that the lagged levels of the dependent variables be used as instruments in the regression equation in differences.

To reintroduce the cross-section dimension of the regression and to address the issue of the persistence of the lagged dependent variables as weak instruments in the GMM difference regression, we use the system GMM proposed by Arellano and Bond (1997). The new estimates consist of the stacked regression in differences and levels where the lagged levels are used as instruments in the difference regression and the difference as instruments in the level regression.

The consistency of the system GMM is tested using the tests proposed by Arellano and Bond (1997). The first is a Hansen test of over-identifying restrictions, which tests the validity of the instruments. The second test examines whether the differenced error term is second-order serially correlated. Failure to reject both tests lends support to our estimator.

To ensure that the Arellano-Bond test detects the desirable serial correlation properties in the residuals of the differenced equation, the capital accumulation regressions contain two lags of the dependent variable. In the TFP regressions, one lag of the dependent variable is sufficient. We restrict the number of instruments to less than the number of included countries to guard against a proliferation of instruments, which can bias the GMM estimates.

We supplement the baseline specification above to investigate heterogeneity and non-linearity within the sample. Descriptions of the five included specifications follow, each of which was estimated for each financial indicator and both sources of growth:

git=αyi,t1+βFi,t+δFi,txINT+γXi,t+ηi+λt+ϵi,t(3)

To capture non-linearities, we interact the financial variables with one of five alternatives: (1) income level; (2) inflation regime; (3) quality of institutions; (4) level of financial development; and (5) regional dummies. The dummy variables used in specifications (i-iv) were created by splitting the sample equally into three ranked subgroups. Regional subgroups include the Caucasus and Central Asia (CCA), Middle East and North Africa Oil Exporters (MENAPOE), and the Middle East and North Africa Oil Importers (MENAPOI).

III - Empirical Results

We find evidence to support a nuanced view of the importance of financial development and economic growth. Our results display no unambiguously positive relationship between “finance” and the sources of growth. In some cases, excessive financial development may have detrimental effects on growth. We do find evidence, however, that the dimensions of financial stability and efficiency are linked to growth. Nonperforming loans and stock market turnover emerge as the only two indicators of financial development to exhibit a general relationship with productivity growth and capital accumulation which is both robust and economically significant.

A. Unconditional correlations

Data plots (Figures 1 and 2) display simple scatter plots between each of the financial indicators and either TFP growth or capital accumulation. The data included for each financial dimension— access, depth, efficiency, openness, and stability—are represented in the logarithmic forms as discussed in the data section above. As shown in the panels, there is little evidence of a strong overarching relationship between any of our financial indicators and the sources of growth. This is echoed in unconditional correlation coefficients, presented in Table 3. Even when the effect is significant, it is small and fails to consider the important contributions from other relevant macroeconomic, institutional, regional, or developmental characteristics.

Figure 2.1
Figure 2.1

Capital Stock Growth and Financial Sector Indicators

Five Year Non-Overlapping Averages

Citation: IMF Working Papers 2017, 143; 10.5089/9781484302637.001.A001

Figure 2.2
Figure 2.2

Capital Stock Growth and Financial Sector Indicators

Three Year Non-Overlapping Averages

Citation: IMF Working Papers 2017, 143; 10.5089/9781484302637.001.A001

Figure 2.3
Figure 2.3

Capital Stock Growth and Financial Sector Indicators

Annual Observations

Citation: IMF Working Papers 2017, 143; 10.5089/9781484302637.001.A001

Table 2:

Summary Statistics of Raw Financial Variables

article image
Note: Summary statistics are of untransformed annual observations.
Table 3:

Unconditional Correlations

article image
Note: Financial variables are included in log forms.* denotes p<0.05, **denotes p<0.01, ***denotes p<0.001

B. Dynamic panel regressions

Baseline regressions

Results for the productivity and capital accumulation regressions are reported in Tables 4-19. For each dependent variable (productivity growth or capital accumulation), column 1 displays the parsimonious baseline specification, while columns 2-6 include additional considerations for income differentiation (column 2), inflation regimes (column 3), institutional quality (column 4), financial depth (column 5), and regional characteristics (column 6). Consistent with other studies in the literature, measures of initial income levels, openness, government size, and human capital are included as control variables (Beck and others 2000c).

Table 4:

Private Credit

article image
Dummy variables for time periods were also included in each model. Results for these are not reported here.t-statistics in parentheses based on cluster-robust standard errors* p<0.1 ** p<0.05 ***p<0.01
Table 5:

Liquid Liabilities

article image
Dummy variables for time periods were also included in each model. Results for these are not reported here.t-statistics in parentheses based on cluster-robust standard errors* p<0.1 ** p<0.05 *** p<0.01
Table 6:

Stock Market Capitalization

article image
Dummy variables for time periods were also included in each model. Results for these are not reported here.t-statistics in parentheses based on cluster-robust standard errors* p<0.1 ** p<0.05 *** p<0.01
Table 7:

Interest Rate Spread

article image
Dummy variables for time periods were also included in each model. Results for these are not reported here.t-statistics in parentheses based on cluster-robust standard errors* p<0.1 ** p<0.05 *** p<0.01