Financial development, investment, and economic growth

Xu, Zhenhui . Economic Inquiry 38.2 (Apr 2000): 331-344.

In this article, I use a multivariate vector-autoregressive (VAR) approach to examine the effects
of permanent financial development on domestic investment and output in 41 countries between
1960 and 1993. The VAR approach permits the identification of the long-term cumulative effects
of financial development on the domestic variables by allowing for dynamic interactions among
these variables. The results reject the hypothesis that financial development simply follows
economic growth and has very little effect on it. Instead, there is strong evidence that financial
development is important to growth and that investment is an important channel through which
financial development affects growth. (JEL E22, E44, E47, O16, O57)
Keywords: economic growth, financial development, investment, vector-autoregression,
impulse-response
Headnote
ABBREVIATIONS
ARIMA: autoregressive integrated moving average
BIC: Bayesian Information Criterion
OECD: Organization for Economic Cooperation and Development
FD: financial development
IMF: International Monetary Fund
INV: growth rate of real domestic investment
M2: money supply
VAR: vector-autoregression
I. INTRODUCTION
Interest in the relationship between financial development and economic growth dates back to
early this century and has been growing since the 1980s.1 In the literature, three views have
emerged concerning the potential importance of finance in economic growth. The first sees
finance as a critical element of growth (Schumpeter [1911]; Goldsmith [1969]; McKinnon
[1973]; Shaw [1973]; Fry [1978, 1988]; Bencivenga and Smith [1991]; King and Levine [1993a,
1993b]). In this view, services provided by the financial system are essential for growth, and a
repressed financial system, characterized by price distortion, undersaving, negative or unstable
returns on savings and investment, and inefficient allocation of savings among competing users,
impedes growth.2 As the financial system develops, households substitute out of unproductive
tangible assets, raising the total real supply of credit, the quantity and quality of investment, and
thus the rate of economic growth. In addition, financial development can promote technological
innovations and productivity growth (King and Levine [1993a]).
The second view regards finance as a relatively unimportant factor in growth, essentially as the
handmaiden to industry and commerce (Robinson [1952]; Lucas [1988]; Stern [1989]). In this
perspective, the lack of financial development is simply a manifestation of the lack of demand
for financial services. As the real sectors of the economy grow, the demand for various financial
services rises and will thus be met by the financial sector. Based on this view, financial
development simply follows economic growth and has very little effect on it.
Like the first one, the third view ascribes effects to finance but focuses on its potential negative
impacts on growth (Van Wijnbergen [1983]; Buffie [1984]). Economists holding this view
contend that financial development can hinder growth by reducing available credit to domestic
firms. This situation arises from the presence of informal curb markets. As the formal financial
system develops, households are seen to substitute out of curb-market loans, thus reducing the
total real supply of domestic credit. The reduction in the supply of credit can lead to a credit
crunch, thereby lowering investment and slowing production and growth. Further, such a credit
crunch can retard economic growth beyond the short term by lowering the steady-state capital
stock (Wijnbergen [1983]).
From these divergent views, three testable hypotheses emerge with sharply different policy
implications. The first view suggests that government policies should be directed toward
improving the financial system, since financial development has important causal effects on
growth. The second view implies that government policies toward improving the financial
system will have little effect on growth, since financial development results from growth and has
little impact on it. Based on the third view, there is a potential danger of financial development.
Under certain institutional arrangements, government efforts toward financial development can
cost an economy its long-term growth by reducing total real supply of domestic credit.
Therefore, to mitigate the potential negative effects of finance on economic growth, the
government must reform its institutional environment first.
So far, empirical studies of the effects of financial development on economic growth have
produced mixed evidence. For example, in one cross-section study, Jao [1976) found no role for
financial development in explaining per capita real GDP. But in another, Lanyi and Saracoglu
[1983] found a positive relationship. In a more recent cross-section study of 80 countries over the
1960-89 period, King and Levine [1993b] found that various measures of the level of financial
development are strongly and positively correlated with real per capita GDP growth, the rate of
physical capital accumulation, and improved efficiency of physical capital. In another study,
King and Levine [1993a] found that financial development also promotes technological
innovations and productivity growth.
The assumption that all countries have the same economic structure implied in the cross-section
studies may be too strong. To account for different institutional environments and financial
structures in different developmental stages, Jung [1986] performed Granger's [1969] causality
tests using both time-series data for each of 56 countries in his sample and cross-section data
based on the level of economic development. In both instances, Jung found positive effects.3 In a
recent time-series study, however, Thornton [1996] performed Granger's [1969] causality tests
for 22 developing countries and found no clear evidence of positive effects of financial
development on economic growth.
In this article, I use a multivariate vectorautoregression (VAR) framework to identify the effects
of permanent financial development on domestic investment and output for a group of 41
countries. I include proxies for financial development, real GDP, and real domestic investment,
the three variables that are widely used in earlier empirical studies. The multivariate VAR
approach in this article is consistent with the previous literature's focus on causality as it
incorporates the possible existence of short-term links between financial development and other
domestic variables. In addition, it has the following advantages. First, it allows for different
economic and institutional arrangements in each country. Second, it can deal with the
simultaneity problem between financial development and other domestic variables, thus avoiding
the difficult task of determining which variables are truly exogenous, as noted in Sims's [1980]
influential work. Third, it permits us to identify not only the short-term effects but also the long-
term cumulative effects of financial development on other domestic variables by allowing for
interactions among these variables, including both the contemporaneous correlation and the
dynamic feedback.
It is essential to confront the simultaneity problem and to account for dynamic interactions
among financial development and other domestic variables. Since the previous literature has
largely ignored the dynamic interactions between financial development and other domestic
variables, the incorporation of such dynamic interactions is an important element of this article.
Clearly, while financial development can affect domestic investment and GDP, the latter
variables can affect financial development as well. Such dynamic interactions between financial
development and the domestic variables require that the equilibrium values of all variables be
determined simultaneously. In addition, the effects of financial development on GDP and
domestic investment can be enhanced or dampened over time by the feedback effects from the
other variables. Therefore, a model testing the dynamic relationship between financial
development and other domestic variables must properly deal with these two important issues.
The multivariate VAR approach is well-suited for this purpose.4
The results found in this article clearly reject the second hypothesis that financial development
simply follows economic growth and has very little effect on it. Instead, there is strong evidence
that financial development is important to economic growth both in the short term and in the
long term and that investment is an important channel through which financial development
affects GDP growth. Although 14 of the 41 countries display negative long-term cumulative
effects of permanent financial development on the growth of GDP and domestic investment, the
remaining 27 countries show positive long-term cumulative effects. Interestingly enough, the
countries showing negative long-term cumulative effects of permanent financial development on
GDP growth are concentrated in Africa geographically and in the low- and lower-middle income
group economically. Specifically, nine of the 14 countries with negative long-term effects are
low- and lower-middle income African countries (Benin, Congo, Ghana, Mauritania, Morocco,
Niger, Nigeria, Senegal, and Togo), three are small island countries (Costa Rica, Fiji, and
Trinidad and Tobago), and the remaining two are highly government-regulated economies (India
and Saudi Arabia). The results also show that for a group of countries the long-term cumulative
effects of financial development on the growth of GDP and domestic investment are completely
different from the short-term contemporaneous effects. Many countries are able to turn their
short-term negative effects to positive ones in the long-term.
The rest of the article is organized as follows. Section II outlines the variables used in this study,
the sample, and the data sources. The multivariate VAR framework and its specification for each
country are addressed in Section III. Section IV discusses the findings of the article based on the
results of the impulse-response analysis. Section V provides a summary and concluding remarks.
II. THE SAMPLE AND DATA
To analyze the dynamic effects of permanent financial development on the growth of real
domestic investment and real GDP, I selected 41 developing countries.5 These 41 countries have
a broad representation. Geographically, they include seven Asian and Pacific countries, one
Middle-Eastern country, three Caribbean countries, 15 African countries, 11 Latin American
countries, and four European countries. Economically, they include 16 low income countries, 14
lowermiddle income countries, and 11 upper-middle income countries.6 The sample period is
1960-93 for most countries but slightly shorter for some. Annual data are used in order to include
as many countries as possible.
In this multivariate VAR framework, I include three commonly used variables-real GDP, real
domestic investment, and an index of financial development.7 They were derived from the
annual data collected from the World Bank's World Data CD-ROM [1995]. Nominal GDP,
"GDP at market prices (cur. local)," nominal gross domestic investment, "Gross domestic
investment (cur. local)," and GDP deflator, "GDP deflator (1987 = 100, index)," were taken from
World Data to derive real GDP and real domestic investment, respectively.
Although the derivation of the GDP and investment variables is straightforward, the derivation of
the index of financial development requires some explanation. Several indicators have been used
in the literature to measure the level of financial development; each focuses on one aspect.8
Among them, the size of the formal financial sector relative to economic activity is the most
widely used (Goldsmith [1969]; McKinnon [1973); King and Levine [1993a, 1993b], Thornton
[1996]). Underlying this common practice is the belief that the provision of financial services is
positively related to the size of the financial intermediary sector. As the size of the financial
intermediary sector grows, so does the provision of financial services. Hence, the level of
monetization in an economy is a pertinent proxy for measuring the level of financial
development. Following this traditional practice, I use the ratio of liquid liabilities of the formal
financial intermediary sector to GDP as a proxy for the level of financial development.
Liquid liabilities of the financial intermediary sector are measured by the sum of money and
quasi-money (M2) less currency, that is, total bank deposits. Currency is excluded from the
measure, because it is not intermediated through the banking system. "Money supply (broadly
defined) (cur. local)" and "Currency outside banks (cur, local)" were taken from World Data as
M2 and currency, respectively. The difference of the two (M2 minus currency) was then used to
measure liquid liabilities of the financial intermediary sector.
Since M2 and currency are stock variables (measured at the end of the year) and GDP is a flow
variable (measured over the year), adjustment must be made to mitigate the problem of deflating
stock variables by a flow variable. Following King and Levine [1993b, 720], I construct the
index of financial development, that is, total bank deposits in GDP, using the geometric mean of
this year's bank deposits and last year's bank deposits divided by GDP. Table I shows the profile
of the index of financial development for each country in alphabetical order, along with the
sample period. The values under the last column, A, show the ratios of average financial
development between 1987 and 1993 to average financial development between 1961 and 1967.
The use of a seven-year average is to eliminate the random effect that could influence M2,
currency, and GDP in a particular year.
Since the inverse of this index of financial development is a measure of the income velocity of
bank deposits, the values reported in Table I provide information on the behavior of this velocity
in the last three decades for each economy.9 Although the mean values allow for cross-country
comparison of the average income velocity of bank deposits for the period of 1960-93, the values
under column A illustrate the trend of this velocity in each economy. Based on the values under
column A, the 41 countries can be divided into three groups: five economies (Peru, Ghana,
Central Africa, Argentina, and Mexico) show a rising trend of the velocity in the last three
decades, one economy (Malta) displays unchanged trend, and the remaining 35 economies show
a declining trend of the velocity.
III. THE METHODOLOGY
VAR Specifications
As noted earlier, the multivariate VAR approach is well suited for the purpose of this study. Yet
two questions must be answered before we can proceed. The first question is whether we should
difference the time series of real GDP, real domestic investment, and the index of financial
development in this multivariate VAR framework. There are two considerations, one
econometric and the other theoretical, for this question. First, appropriate differencing is
important in time-series analysis because most algorithms used for fitting autoregressive
integrated moving average (ARIMA) models will fail if the time series are nonstationary. In a
VAR model, however, the asymptotic distribution that characterizes the estimates will be the
same whether the model is estimated in levels or in differences (Park and Phillips [1988, 1989];
Sims, Stock, and Watson [1990]). In fact, Fuller [1976] shows in his Theorem 8.5.1 that
differencing does not gain asymptotic efficiency in an autoregression, even if it is appropriate.
But for small samples, the distributions of the estimates may be improved by estimating the VAR
model in differences (Hamilton [1994, 553 and 652]). Second, innovations in financial
development generally appear as periodic episodes. But what is important and interesting to
policy makers and researchers are those policies with long-lasting or permanent effects on
financial development. Innovations in the growth rates can capture such permanent changes in
the level of financial development. First-differencing translates the log of levels into growth rates
and thus allows us to examine the effects of permanent changes in the level of financial
development on other domestic variables. Therefore, I use the first-difference of the log of levels
for each series in the estimation, and I denote RGDP as the growth rate of real GDP, INV as the
growth rate of real domestic investment, and FD as the change in the index of financial
development.
The second question concerns the specification of the VAR model for each country. We must
decide in this VAR framework what deterministic components should be included and what
appropriate orders of lag should be used. Since arbitrarily chosen specifications for a VAR
model will most likely produce unreliable results, a databased model selection criterion is used to
specify the VAR model for each economy in the sample. Among the various model selection
criteria, the one proposed by Schwarz [1978], known as Schwarz's Bayesian Information
Criterion (BIC), is shown to outperform other alternatives (Mills and Prasad [1992]). Therefore,
the specifications of the VAR model for each economy are based on Schwarz's BIC. The BIC
selected the specification of the first-order VAR model, that is, VAR(1), with a constant but no
trend for Argentina, Benin, Bolivia, Chile, Congo, Costa Rica, Ghana, Guatemala, Myanmar,
Niger, Nigeria, Pakistan, El Salvador, and Turkey, with a constant and a trend for Central Africa,
Egypt, Fiji, Greece, Haiti, India, Jamaica, Morocco, Madagascar, Mexico, Peru, Philippines,
Saudi Arabia, Senegal, Togo, Trinidad and Tobago, and Venezuela, and with no constant and
trend for Honduras and Sudan. It chose a VAR(2) model with a constant but no trend for
Paraguay, with a constant and a trend for Burkina Faso, Korea, Mauritania, and Mauritius, and
without a constant and a trend for Thailand, and a VAR(3) model with a constant but no trend for
Portugal and with a constant and a trend for Malta.
Impulse-Response Analysis
The inference of the effects of Fl) on INV and RGDP is based on the results of the impulse-
response analysis, using the VAR specifications reported above. The focus of the analysis is the
long-term and the short-term elasticities of RGDP and INV with respect to FD.
The impulse-response function shows the effects on RGDP and INV of an exogenous once-and-
for-all innovation in FD in the initial period and no innovations to any variables in the future.
Since FD is allowed to be correlated with RGDP and INV within a period, there will be a
contemporaneous effect. The short-term elasticity measures this contemporaneous effect of a one
standard deviation FD shock today on current RGDP and INV. Since it does not include any
feedback from the affected variables, the shortterm elasticity is static in nature.
But there are dynamic effects as well. Since the FD innovation will affect all of the variables in
the system both contemporaneously and with lags, every variable will change after the first
period. In the second period, the autocorrelation of RGDP will affect RGDP; the lagged
relationship between FD and RGDP will matter; and INV, which had been affected by the FD
innovation, will also affect RGDP. The same is true for INV. Because of the dynamic
interactions among the variables, the short-term effects of FD on RGDP and INV can be
enhanced or dampened over a longer horizon. The effects of the dynamic responses will die out
over time but not immediately; they are persistent. By summing up the effects in each period
over a longer horizon, we get the long-term cumulative effects of an FD innovation on RGDP
and INV. In estimations, the maximum periods for the impulseresponse analysis were set at 20
because in most cases the values of the impulseresponse function converge in fewer than ten
periods and, in all the cases, the values converge within 20. Therefore, the long-term elasticity
measures the cumulative effects of a one standard deviation FD shock today on RGDP and INV
over 20 periods. It is normalized by dividing the cumulative impulse responses of RGDP and
INV by the cumulative impulse responses of FD over 20 periods.
It is well known that the results of impulse-response analysis depend on the specific ordering of
the variables under investigation. Since the estimated matrix of variance and covariance is not
diagonal, orthogonalization is necessary before a meaningful impulse-response analysis can be
conducted. The orthogonalization strategy, however, is not unique. I confine the discussion of
orthogonalization strategy to the use of triangular matrices under the Choleski decomposition
method. This method uses all the information in the matrix of contemporaneous correlation
among the estimated residuals. In this study, there are three possible scenarios depending on
whether FD is ranked first, second, or third. These three scenarios completely determine the
range of results in the impulse-response analysis.
Among the three possible scenarios, two seem to be the most plausible a priori. Based on the first
and third views discussed in the introduction, FD should be ranked first. In this case, shocks
from FD affect RGDP and INV contemporaneously, but shocks from RGDP and INV have no
immediate effects on FD. Based on the second view, however, FD should be ranked last, in
which case the opposite is true. Between these two cases, the first one is chosen as the central
case. The short-term and the long-term elasticities of FD on RGDP and on INV for the central
case are reported in Table II.
To establish the robustness of the central case, however, I also computed the values of the
impulse-response function under all of the alternative orthogonalizations. There are a total of six
possible orthogonalizations for each economy. Since the results of the impulse-response analysis
are the same under several orthogonalizations, I need to report the values of the impulse-
response function for only four different cases. The range of results across all possible
orthogonalizations for the long-term elasticities is reported in Table III.
IV. THE EFFECTS OF FINANCIAL DEVELOPMENT ON GDP GROWTH AND
INVESTMENT
In this section, I present the empirical results of the effects of FD on RGDP and on INV, based
on the impulse-response analysis. The estimation of the VAR model for each country in the
sample is performed using RATS (version 4.2).
Financial Development and Long-term GDP Growth
In Table II, while 14 of the 41 countries in the sample exhibit negative long-term cumulative
effects of FD on RGDP, the remaining 27 countries show positive long-term cumulative
elasticities. The range for all elasticities is between - 0.467 (Congo) and 0.543 (El Salvador). The
estimated long-term elasticity for 31 countries ties within 0.20 in absolute value. Only four of the
41 countries display elasticities lower than - 0.20, and six countries show elasticities higher than
0.20. Of the 31 countries, the estimated long-term elasticity is less than 0.10 in absolute value for
21 countries, and less than 0.01 in absolute value for four countries (Costa Rica, Korea, Pakistan,
and Paraguay). Hence, the long-term cumulative effects of FD on RGDP are concentrated in a
relatively narrow range across countries.
The results show some interesting patterns about the long-term cumulative effects of FD on
RGDP. Based on geographic locations, the 14 countries that show negative long-term elasticities
include two Asian and Pacific countries (India and Fiji), one Caribbean country (Trinidad and
Tobago), one Latin American country (Costa Rica), one Middle-Eastern country (Saudi Arabia),
and nine African countries (Benin, Congo, Ghana, Mauritania, Morocco, Niger, Nigeria,
Senegal, and Togo). In other words, about two-thirds of the countries showing negative long-
term effects of FD on RGDP are concentrated in Africa, which also accounts for 60% of the
African countries in the sample.
Based on economic structure and environment, Costa Rica, Fiji, and Trinidad and Tobago are
island economies with small domestic markets. The economic structure and environment in such
economies can largely be determined by their geographic locations. The remaining 11 countries
with negative long-term elasticity, however, were all highly government-regulated economies in
the sample period. For the nine African countries, most adopted interventionist government
policies after their independence. While successful in some cases, the overall outcome has
proved to be a failure. A number of these countries started their economic reforms supported by
the World Bank and the International Monetary Fund (IMF) in the late 1980s or the early 1990s.
For example, after nearly two decades of state-led development, Benin finally started an
economic reform program aimed at changing its socialist-inspired state interventionism in 1989.
In the early 1990s, Congo started its economic reform to improve public sector efficiency. Ghana
has also undertaken wide-ranging financial reforms, including the abolishment of interest rate
controls and sectional credit ceilings. India and Saudi Arabia had a similar economic
environment as the African countries for the sample period. But unlike the African countries,
there has been little government effort in improving the economic structure and environment in
India and Saudi Arabia. According to the World Bank, India still has fundamental structural
problems.10 For instance, state-owned banks in India continue to dominate the banking system
and serve mainly as instruments for financing economic activities selected by the government. In
addition, India's pervasive investment licensing regime made private investment decisions
conditional on cumbersome government approvals and thus discouraged domestic investment.
Based on the level of economic development, seven of the 14 counties with negative long-term
elasticities are low-income countries (Benin, Ghana, India, Mauritania, Niger, Nigeria, and
Togo), five are lowermiddle income countries (Congo, Costa Rica, Fiji, Morocco, and Senegal),
and only two are upper-middle income countries (Saudi Arabia, and Trinidad and Tobago).
Therefore, more than 85% of the countries showing negative long-term elasticities of RGDP with
respect to FD belong to the group of tow- and lower-middle income countries. Alternatively, the
countries with negative longterm elasticities account for 44% of the lowincome countries, for
36% of the lowermiddle income countries, and for 18% of the upper-middle income countries in
the sample.11
From the perspective of the conventional wisdom on financial development and economic
growth, the above results are informative and interesting. In the literature, it is argued that the
effects of financial development on economic growth are related to the stage of economic
development of a country (Patrick [1966]; Jung [1986]). Financial development can induce real
innovation-type investment and economic growth primarily at the early stage of economic
development, and the effect of financial development on economic growth diminishes as
sustained economic growth gets under way (Patrick [1966, 177]).
The results in this article, however, provide no support for this hypothesis. Instead, they clearly
show the contrary-the number of countries showing negative long-term cumulative effects of FD
on RGDP decreases as the level of economic development advances. This suggests that there is a
positive correlation between the level of economic development and the beneficial effects of
financial development on economic growth. Countries at the early stages of economic
development typically face more structural constraints than do their counterparts. Although
financial openness improves domestic financial systems and mobilizes financial resources, these
additional constraints become critical because they can prevent the mobilized financial resources
from being allocated efficiently to investors and firms. Such constraints will be less binding as
these countries become economically more advanced. Therefore, as suggested by the third
hypothesis, along with financial openness, countries at the early stages of economic development
must pay close attention to the reform of their economic environment.
It should also be noted that even for the countries at the same level of economic development,
the effects of financial development on GDP growth can be different from country to country.
Again, such differences are likely to result from differences in institutional, environmental, and
financial structures. Although economic development can make economic institutions more
uniform over time, one is likely to find such crosscountry differences continuing because of
cultural and other social factors. Nevertheless, given the strong evidence that financial
development has long-term effects on GDP growth, government policies toward financial
development have an important impact on long-term economic growth and thus need to be
carefully designed.
Financial Development and the Growth of Domestic Investment
The long-term effects of FD on INV are quite similar to those on RGDP. In Table 11, although
14 of the 41 countries display negative long-term cumulative effects of FD on INV, the
remaining 27 countries show positive long-term elasticities. The range of all long-term
elasticities for INV, however, is wider than that for RGDP. It is between -1.474 (Congo) and
1.612 (El Salvador). The values of the elasticities are greater than 1.00 in absolute value for ten
countries, but less than 0.10 in absolute value for five countries. For the remaining 26 countries,
the values of the elasticities fell in a narrower range between 0.10 and 0.37 in absolute value
except Greece, Haiti, Honduras, Jamaica, and Togo, for which the values of the elasticities are
between 0.540 (Togo) and 0.867 (Jamaica).
Eight of the ten countries with an elasticity greater than 1.00 in absolute value show positive
elasticities. Of the remaining 19 countries with positive long-term elasticities, the values of the
elasticities are between 0.50 and 0.99 for five countries, between 0.25 and 0.50 for four
countries, between 0.10 and 0.25 for seven countries, and less than 0.10 for three countries. For
the 14 countries with negative long-term elasticities, the values of the elasticities are lower than -
1.00 for two countries, between -0.40 and -0.25 for four countries, between -0.25 and -0.10 for
six countries, and greater than -0.10 for two countries.
Twenty-one of the 27 countries displaying positive long-term effects of FD on RGDP also show
positive long-term effects of FD on INV. On the other hand, eight of the 14 countries with
negative long-term effects of FD on RGDP also show negative long-term effects on INV. This
indicates that domestic investment is indeed an important channel through which financial
development affects output. Yet for the remaining 12 countries, the long-term cumulative effects
of FD on INV are different from those on GDP. For five countries (Chile, Korea, Pakistan,
Portugal, and Thailand), the long-term effects of FD are negative on INV but positive on RGDP.
For another seven countries (Mauritania, Morocco, Niger, Nigeria, Saudi Arabia, Togo, and
Trinidad and Tobago), the opposite is true. This implies that in addition to domestic investment
there exist other channels through which financial development can affect GDP, a result that is
consistent with the argument by King and Levine [1993a].
Long-Term versus Short-Term Effects
The short-term elasticities are also reported in Table II next to the long-term elasticities for
comparison. The short-term elasticities show that shocks in FD have positive contemporaneous
effects on RGDP for 12 countries. The values of the short-term elasticities for these 12 countries
range from 0.018 (Honduras and Jamaica) to 0.232 (Portugal). On the other hand, the values of
the short-term elasticities for the 29 countries with negative contemporaneous effects of FD on
RGDP range from -0.471 (Malta) to -0.014 (Guatemala). Bolivia is the only country showing a
near zero short-term negative effect of FD on RGDP. Further, the negative short-term effects in
these countries are all diminished in the long-term, except for Congo, Ghana, India, Nigeria,
Togo, and Trinidad and Tobago. Yet Ghana, India, Togo, and Trinidad and Tobago all
experienced a small increase in negative effects in the long-term. For many countries, the
negative contemporaneous effects of FD on RGDP in the short-term become the positive
cumulative effects in the long-term.
The short-term elasticities also show that shocks in FD have positive contemporaneous effects on
INV for 21 countries with a range from 0.038 (Bolivia) to 1.306 (Venezuela). The range for the
remaining 20 countries with negative short-term elasticities of FD on INV is between -0.788
(Greece) and -0.013 (Congo). Myanmar is the only country in this case to display a near zero
shortterm elasticity. While the negative short-term effects are reinforced in the long-term for five
countries (Chile, Congo, Costa Rica, Ghana, and Malta), 15 of the remaining 20 countries with
negative short-term elasticities of FD on INV show reduced negative effects in the long-term.
An interesting aspect of the results is that many economies are able to turn the negative short-
term effects to the positive longterm effects. For 16 countries, the effect of FD on RGDP is
negative in the short-term but positive in the long-term. Costa Rica is the only exception for
which the opposite is true. As for the effects of financial development on domestic investment,
although the effect of FD on INV is positive in the shortterm but negative in the long-term for
three countries, nine countries are able to turn the short-term negative effects into the long-term
positive effects. These results suggest that financial policies play an important role in economic
growth and that governments must craft these policies with care.
Sensitivity Analysis
The above discussion is based on the orthogonalization strategy that seems the most plausible
and meaningful based on a priori theoretical grounds. Yet as shown by Levine and Renelt
[1992], since many factors are associated with economic growth, the empirical results on the
relationship between one factor and economic growth is not always robust. Therefore, it is
necessary to examine the robustness of the results reported in the previous sections, and a
sensitivity analysis was performed by considering all possible triangular orthogonalizations using
the Choleski decomposition. Table III shows the lower and upper bounds for the long-term
elasticities of RGDP and INV with respect to FD under all possible orthogonalizations, as well as
the number of positive outcomes out of four possibilities.
The values of all RGDP elasticities are from -0.467 (Congo) to 0.725 (Thailand), which is only
slightly wider than that in the central case already considered in the previous section. In 28
countries, the values in the central case coincide or are very close to the upper or lower bounds
of the possible values. This suggests that the actual distribution of results across countries may
be even narrower than the range identified in the central case.
Of the 41 countries in the sample, 32 show absolute robustness and another four show strong
robustness in terms of the qualitative nature of the results. In other words, the sign of the long-
term elasticities of RGDP with respect to FD in the central case is maintained across all of the
orthogonalization outcomes (four in four) for 32 countries and across most of the
orthogonalization outcomes (three in four) for four countries. The 32 countries include 24 of the
27 countries that display positive long-term elasticities in the central case and eight of the 14
countries that display negative long-term elasticities in the central case. Of the four countries that
show strong robustness, Guatemala and Mexico show positive longterm elasticities in the central
case, whereas Benin and Costa Rica show negative longterm elasticities in the central case. For
another four countries (Fiji, Niger, Mauritania, and Portugal), the sign of the long-term
elasticities in the central case is maintained across half of the orthogonalization outcomes (two in
four). Morocco is the only country for which the sign of its negative long-term elasticity can be
maintained only in the central case.
The range of results for the long-term cumulative effects of FD on INV is rather wide, from -
1.853 (Congo) to 2.739 (Mauritius). Yet with the exception of Congo, Mauritania, Mauritius, and
Niger, the values of the long-term elasticities of INV with respect to FD fall in the range of the
central case. There are 12 countries with upperbound elasticities greater than 1.00 and two
countries with lower-bound elasticities lower than -1.00. The values in the central case coincide
or are very close to the upper or lower bounds of the possible values for 24 countries.
For the long-term cumulative effects of FD on INV, 27 of the 41 countries show absolute
robustness; the sign of the long-term elasticities of INV in the central case is maintained across
all of the orthogonalization outcomes (four in four). These 27 countries include 20 of the 27
countries that display positive long-term elasticities in the central case and seven of the 14
countries that display negative long-term elasticity in the central case. Another four countries
show strong robustness across all of the orthogonalization outcomes (three in four). They include
two countries (Burkina Faso and the Central African Republic) that displays positive long-term
elasticity in the central case and two countries (Malta and Nigeria) that display negative long-
term elasticities in the central case. For the remaining ten countries, the sign of the long-term
elasticities in the central case is maintained across half of the orthogonalization outcomes (two in
four) for six countries (Benin, Guatemala, Mexico, Saudi Arabia, Togo, and Trinidad and
Tobago), and Fiji, Korea, Senegal, and Thailand are the countries for which the sign of their
long-term negative elasticities can only be maintained in the central case.
To summarize, the sensitivity analysis indicates that the central results on the longterm effects of
FD on RGDP and on INV are robust for 36 and 31 of the 41 countries in the sample,
respectively. Therefore, about 90% and 80% of the countries in the sample show, respectively,
that the long-term elasticities of RGDP and INV obtained under the central case are the most
plausible results.
V. SUMMARY AND CONCLUDING REMARKS
I use a multivariate VAR approach to investigate the effects of financial development as
measured by total bank deposits in GDP on the growth of domestic investment and GDP for a
sample of 41 countries. This methodology allows for different economic and institutional
arrangement in each country and thus avoids the strong assumption that all countries have similar
economic structures in the cross-section studies. It also deals with the simultaneity problem
among financial development, domestic investment, and output. More important, it permits the
identification of the long-term cumulative effects of permanent financial development on the
growth of domestic investment and GDP by taking into account the dynamic feedback among
financial development, domestic investment, and GDP as well as the short-term
contemporaneous effects.
The results of this study clearly reject the hypothesis that financial development simply follows
economic growth and has very little effect on it. Instead, there is strong evidence that financial
development is important to GDP growth and that domestic investment is an important channel
through which financial development affects economic growth. In 41 countries, positive long-
term effects of permanent financial development on the growth of domestic investment and GDP
are detected for 27 countries in each case. Many countries are able to turn the short-term
negative effects to long-term positive effects. Most important of all, these results are robust.
Therefore, they provide time-series empirical evidence from a broad spectrum of countries that
financial development is important to economic growth.
Footnote
1. For an overview of the recent literature, see Pagano [1993].
2. Price distortion is an inevitable outcome of a repressed financial system. Using a Harris-
Todaro model, Feldman and Gang [1990] show that financial repression can cause lower price
levels through ruralurban migrations. In a recent study, Xu and Feldman (1999] used
cointegration tests to explain why real price levels are systematically lower in developing
countries than in developed countries. They found an evidence of a long-run equilibrium
relationship between financial development and international real price differences for most
developing countries. As a developing country becomes financially more developed, its real
price levels get closer to world real price levels.
Footnote
3. There is no discussion in Jung's [1986] article on how the author dealt with two very important
issues in causality tests: the unit root property of the data and the optimal lags in Granger's
causality tests. Using Japan and Taiwan as a case study for the export-led growth hypothesis, Xu
[1998] shows that causality inferences can be misleading when these two issues are not handled
properly.
Footnote
4. An alternative method can be the dynamic panel model proposed by Arellano and Bond
[1991]. Yet this model is difficult to estimate. Complications in estimating the dynamic panel
model arise from the fact that the lagged dependent variable in such a model is correlated with
the disturbance, even if the disturbance is not itself autocorrelated (Green [1997]).
Footnote
5. Based on the availability of data, 59 countries were originally chosen. Of the 59 countries, 18
are high-income Organization for Economic Cooperation and Development (OECD) countries.
As shown by Levine [1991] and Bencivenga, Smith and Starr [1995], equity markets, along with
the formal banking system, play an important role in economic growth for developed countries,
where ownership of firms is continuously traded but the production process remains undisturbed.
It suggests that the index of financial development used in this article may be inappropriate for
developed countries because the time series of the index can show an inverse U-shaped pattern
(Bordo and Jonung [1987]). While the upward portion of the index is consistent with the measure
of financial development, the downward portion is not. Hence, the 18 OECD countries were
dropped from the sample.
Footnote
6. The classification for geographical location and incomes is based on the World Bank's World
Data [1995].
7. The use of real domestic investment allows us to examine the effects of financial development
on the growth of domestic investment. An alternative to real domestic investment is the share of
domestic investment in GDP. The results of the article, however, are not affected by the choice
of these alternative measures.
8. For a discussion of these financial indicators, see King and Levine [1992].
Footnote
9. Because this index of financial development is a measure of the income velocity of bank
deposits, both financial development and money demand can affect it. To ensure that the changes
in the velocity reported in Table I do not result from changes in money demand because of the
interest effect, I have tried an alternative definition for the index. In particular, I remove the
effects of money demand on the income velocity of bank deposits by performing the following
Ordinary Least Squares (OLS) regression for each country, IFD^ sub 1^ = alpha + Beta^ sub
1^T^ sub 1^ + Beta^ sub 2^(pi^ sub t^) + e^ sub t^ where IFD^ sub t^ is the index of financial
development, T^ sub t^ is the time trend, pi^ sub t^, is the rate of inflation measured by the GDP
deflator, alpha, Beta^ sub 1^, and Beta^ sub2^ are parameters, and e^ sub t^ is the error term.
The conclusion of the article, however, is not much affected by the choice of this alternative
index of financial development.
Footnote
10. See TIDE Country Reports from the World Bank's World Data CD-ROM [1995] for a
discussion of the economic condition for India and other countries.
Footnote
11. To examine how these results fit in the previous literature, 1 used the pooled data to perform
the OLS regression with RGDP as the dependent variable and INV and FD as the independent
variables. The model includes an intercept, a trend, and two lags for each variable. The results
show a significantly positive effect of FD on RGDP based on the pooled data. The effect remains
positive after I introduced intercept and slope dummies into the model. Based on geographic
locations, both the intercept and slope dummies are significantly negative for African countries
and insignificant for other groups. Based on income levels, the intercept dummy is significantly
negative for low-income countries and insignificant for other groups, but the slope dummy is
insignificant for all groups.
References
REFERENCES
References
Arellano, Manuel, and Stephen Bond. "Some Tests of Specification for Panel Data: Monte Carlo
Evidence and An Application to Employment Equations." Review of Economic Studies, 58,
1991, 277-97.
References
Bencivenga, Valerie R., and Bruce D. Smith. "Financial Intermediation and Endogenous
Growth." Review of Economic Studies, 58, 1991, 195-209.
Bencivenga, Valerie R., Bruce D. Smith, and Ross M. Starr. "Transaction Costs, Technological
Choice, and Endogenous Growth." Journal of Economic Theory, 67, 1995, 153-77.
Bordo, Michael D., and Lays Jonung. The Long Run Behavior of the Velocity of Circulation.
New York: Cambridge University Press, 1987.
Buffie, Edward F. "Financial Repression, the New Structuralists, and Stabilization Policy in
Semiindustrialized Economies." Journal of Development Economics, 14, 1984, 305-22.
Feldman, David H., and Ira N. Gang. "Financial Development and the Price of Services."
Economic Development and Cultural Exchange, 38, 1990, 341-52.
Fry, Maxwell J. "Money and Capital or Financial Deepening in Economic Development?"
Journal of Money, Credit and Banking, 10, 1978, 464-75.
. Money, Interest, and Banking in Economic Development. Baltimore: Johns Hopkins University
Press, 1988.
Fuller, Wayne A. Introduction to Statistical Time Series. New York: Wiley, 1976.
Goldsmith, Raymond W. Financial Structure and Development. New Haven, Conn.: Yale
University Press, 1969.
References
Granger, Clive W. J. "Investigating Causal Relations by Econometric Models and Cross Spectral
Methods." Econometrica, 37, 1969, 424-38.
Green, William H. Econometric Analysis. Upper Saddle River, N.J.: Prentice-Hall, 1997.
Hamilton, James D. Time Series Analysis. Princeton, N.J.: Princeton University Press, 1994.
Jao, Y. C. "Financial Deepening and Economic Growth: a Cross-Section Analysis." Malayan
Economic Review, 21, 1976, 47-58.
Jung, Woo S. "Financial Development and Economic Growth: International Evidence,"
Economic Development and Cultural Change, 34, 1986, 333-46.
King, Robert G., and Ross Levine. "Financial Indicators and Growth in a Cross Section of
Countries." World Bank Working Paper, No. 819, 1992.
"Finance, Enterpreneurship, and Growth." Journal of Monetary Economics, 32, 1993a, 513-42.
"Finance and Growth: Schumpeter Might Be
Right." Quarterly Journal of Economics, 108, 1993b, 717-37.
Lanyi, Anthony, and Rusdu Saracoglu. "Interest Rate Policies in Developing Countries." IMF
Occasional Paper, 22, 1983.
Levine, Ross. "Stock Markets, Growth, and Tax Policy." Journal of Finance, 46, 1991, 1445-65.
References
Levine, Ross, and David Renelt. "A Sensitivity Analysis of Cross-Country Growth Regressions."
American Economic Review, 82, 1992, 942-63.
Lucas, Robert E. "On the Mechanics of Economic Development." Journal of Monetary
Economics, 22, 1988, 3-42.
McKinnon, Ronald I. Money and Capital in Economic Development. Washington D.C.:
Brookings Institution, 1973.
Mills, Jeffrey A., and Kislaya Prasad. "A Comparison of Model Selection Criteria." Econometric
Reviews, 11, 1992, 201-33.
Pagano, Marco. "Financial Markets and Growth: An Overview." European Economic Review,
37, 1993, 613-22.
Park, Joon Y., and Peter C. B. Phillips. "Statistical Inference in Regressions with Integrated
Processes. Part 1." Econometric Theory, 4, 1988, 468-97.
"Statistical Inference in Regressions with Integrated Processes. Part 2." Econometric Theory, 5,
1989, 95-131.
Patrick, Hugh. "Financial Development and Economic Growth in Underdeveloped Countries."
Economic Development and Cultural Change, 14, 1966, 174-89.
Robinson, Joan. "The Generalization of the General Theory." In The Rate of Interest and Other
Essays. London: Macmillan, 1952, 69-142.
Schumpeter, Joseph A. The Theory of Economic Deuel
References
opment, Cambridge, Mass.: Harvard University Press, 1911.
Schwarz, Gideon. "Estimating the Dimension of a Model." Annals of Statistics, 6, 1978, 461-64.
Shaw, Edward S. Financial Deepening in Economic De
velopment, New York: Oxford University Press, 1973.
Sims, Christopher A. "Macroeconomics and Reality." Econometrica, 48, 1980, 1-48.
Sims, Christopher A., James H. Stock, and Mark W. Watson. "Inference in Linear Time Series
Models with Some Unit Roots," Econometrica, 58, 1990, 113-44.
Stern, Nicholas. "The Economics of Development: A Survey." Economic Journal, 100, 1989,
597-685. Thornton, John. "Financial Deepening and Economic
Growth in Developing Economies," Applied Economics Letters, 3, 1996, 243-46.
Van Wijnbergen, Sweden "Credit Policy, Inflation and Growth in a Financially Repressed
Economy." Journal of Development Economics, 13, 1983, 45-65.
World Bank. World Data. Washington, D.C.: World Bank, 1995.
Xu, Zhenhui. "Export and Income Growth in Japan and Taiwan." Review of International
Economics, 6, 1998, 220-33.
Xu, Zhenhui, and David. H. Feldman. "Financial Development and Real Price Level
Differences." Review of Development Economics, 3, 1999, 27-43.
AuthorAffiliation
ZHENHUI XU*
AuthorAffiliation
* I thank John Boschen, Robert King, Carl Moody, Alfredo Pereira, Roger Sherman, Charles
Weise, John Whitaker, two anonymous referees, and the coeditor, Dennis Jansen, for helpful
comments and suggestions. All remaining errors are entirely my responsibility.
Xu: Associate Professor, Georgia College and State University, Milledgeville, Ga., Phone 1-912-
445-2592, Fax 1-912-445-5249, E-mail zxu@mail.gcsu.edu
Copyright Western Economic Association Apr 2000