Financial development and economic growth nexus: a time-series evidence from India

Abstract

This study examines the relationship between financial development and economic growth in India for the period 1951–52 to 1995–96. The long-run equilibrium and short-run dynamic models are estimated using financial interrelations ratio and new issue ratio as the measures of financial development, a la Goldsmith (1969). The Johansen (1991) estimator rejects the null of zero cointegrating vector and shows the presence of long-run equilibrium relationship between financial development and economic growth. The error correction model, impulse response and variance decomposition analyses (Sims, 1980), and the Toda and Yamamoto (1995) estimator show the presence of bidirectional Granger-causality between financial development and economic growth. The presence of bidirectional Granger-causality suggested by these estimators points towards the possible problem of endogeneity and simultaneity bias in the growth models that examine the contemporaneous effect of financial development on economic growth. The economic reforms that started since July 1991 emphasized on the liberalization and development of financial sector to supplement the efforts aimed at achieving high economic growth in India.

Acknowledgements

I am grateful to the anonymous Referee and the Editor of the journal for very useful comments and suggestions. However, I am solely responsible for any error and omission that may remain in this article.

Notes

¹ Patrick (1966) was the first to recognize that the positive relationship between financial development and economic growth is insufficient to establish a causal relationship between these variables. This has been subsequently reinforced by McKinnon (1988) who argues that although a higher rate of financial growth is positively associated with successful real growth, the Patrick (1966) problem remains unsolved: what is the cause and what is the effect? Is finance a leading sector in economic development, or does it simply follow growth in real output which is generated elsewhere?

² A wellknown limitation of the crosssection approach is that it is premised on the implicit assumption of similar structural characteristics across countries. The structural differences in terms of production technologies and institutional patterns – social, economic and political across countries are assumed to have either no effect on growth or have only random effects with zero mean. Such a restrictive assumption makes it difficult to control for the unobserved country specific structural differences and account for the nonrandom distribution of their effects.

³ A major advantage of dynamic models is that these models resolve the problem of nonstationarity and ‘spurious regression’. Granger and Newbold (1974) pioneered the work on the problem of spurious regression. They consider the series generated by random walks without drift, i.e. the series generated by process specified by y(t) = y(t − 1) + u(t). Based on Monte Carlo experiment, they observe that if x(t) and y(t) sequences are independent random walks, then in the regression model specified by , t statistics rejects the null of much more than it should. In case of a larger sample size, it tends to reject the null more frequently. In spurious regression, the R ² is high, t-statistics are significant (but without any economic interpretation), estimates are not consistent and the conventional tests of statistical inference do not hold.

⁴ Statistically, the ratio of high money supply to high GDP in the developed countries would be similar to the ratio of low money supply to low GDP in the developing countries.

⁵ Murinde and Eng (1994) provide a discussion on the financial interrelations ratio (FIR); however, they do not use this ratio in the estimation of the model.

⁶ The term will be consistent with first-difference stationary VAR, only if it is also stationary and this will happen if the elements of Y are cointegrated. The long-run relationship between ΔY(t) and can exist, only if contains one or more stationary linear combinations of variables in vector Y. The p × p matrix of coefficients contains information regarding the number of cointegrating vectors r and hence the long-run relationships among the variables in data vector Y. The provides the cointegrating vectors for the linear combination of stationary variables.

⁷ Toda and Phillips (1993) suggest that the Johansen ECM estimator is more useful for testing the null of noncausality; it involves first determining the rank of matrix and then estimating the ECM and testing the null of noncausality.

⁸ The data on the flow of funds accounts of the Indian economy required to construct the financial interrelations ratio FIR and new issue ratio NIR were available only until 1995–96.

⁹ The total debt (credit) includes both primary securities issued by nonfinancial institutions as well as secondary securities issued by financial institutions. Alternatively stated, the sum of primary issues of nonfinancial institutions and secondary issues of financial institutions represents the total financial claims or total debt (credit) and FIR is the ratio of such total debt (credit) to net physical asset formation.

¹⁰ The financial interrelations ratio and the new issue ratio ; where TI stands for total issues which include both primary and secondary issues; PI: primary issues; and NDCF: net domestic capital formation at current prices (1993–94 series, National Accounts Statistics).

¹¹ The results for the (i) Dickey–Fuller unit root tests, (ii) model selection criteria (Akaike's information criterion, AIC and Schwarz Criterion, (SC) and the (iii) Sims (1980) likelihoods ratio (LR) test are not reported to conserve space.

¹² Sawa (1978) argues that AIC tends to choose the model with a higher lag order than the true model, but the bias is small or negligible when .

¹³ The univariate test statistics are based on the estimated residuals of each of the VAR equation and include three central moments of estimated residuals (standardised deviation, skewness and kurtosis), LM test (Engle,1982) for autoregressive conditional heteroskedasticity (ARCH) and the modified version of Shenton–Bowman test for normality of individual residual series (Shenton and Bowman,1977; Doornik and Hansen,1994). The multivariate test statistics are based on estimated auto- and cross- correlations of the residuals of overall VAR system and include LM test for first and fourth order autocorrelation (Godfrey,1988) and the χ ² test for normality. The χ ² test for normality is based on the multivariate version of the univariate Shenton–Bowman test. The results of these tests are not reported to conserve space.

¹⁴ Robinson (1952) and Lucas (1988) are sceptical about the positive role of financial development and intermediation in the process of economic growth.

¹⁵ One approach to modelling short-run dynamics is to use differenced variables in a regression (or VAR) model. However, such a model containing only the differenced variables can be considered sufficient to capture the short-run dynamics, only if there is no long-run relationship among the set of level variables of the model. Hendry and Mizon (1978) and Davidson et al . (1978) argue that it is not possible to draw any inference about the long-run steady state solution from a model containing only the differenced variables. Campbell (1987) argues that if an economic theory imposes cointegration on a set of nonstationary variables, then the simple first-differencing of all the variables does not lead to a well-behaved system for statistical modelling. Granger (1988) shows that in a cointegrated I(1) process, the simple dynamic model suffers from a misspecification problem and the standard causality tests are not valid. He argues that the exclusion of error correction term from the dynamic model results in model misspecification and yields spurious results. Park and Phillips (1989) and Sims et al. (1990) show that if the variables are integrated or cointegrated, then the conventional asymptotic theory is generally not applicable to hypothesis testing in level VARs.

¹⁶ Patrick (1966) postulates the relationship between financial development and economic growth in terms of the supply-leading and demand-following hypotheses. In supply-leading hypothesis, the causality runs from financial development to economic growth and this hypothesis suggests that economic growth generates the supply of financial institutions and financial services prior to the demand for these institutions and services. In demand-following hypothesis, the causality runs from economic growth to financial development, and this hypothesis suggests that the growth in demand for financial institutions and services is conditioned by the growth in real output.

¹⁷ For a discussion on the issue of weak exogeneity, strong exogeneity and superexogeneity, see Charemza and Deadman (1992).

¹⁸ The ECM with ΔlnNIR(t) as the dependent variable (Model II) is estimated using eight lags.

¹⁹ The results of the alternative ECMs estimated using the instrumental variable (IV) estimator are not reported to conserve space.

²⁰ In the presence of long-run cointegrating relationship among the model variables as suggested by the Johansen estimator, it does not seem appropriate to use dynamic VAR model to perform impulse response and variance decomposition analysis.

²¹ The forecast error variance is the sum total of the variances and covariances of all innovation series.

²² For a discussion on the effects of other factors, such as exports, on the growth and productivity in India, see Singh (2003).

²³ In a regression model specified by , the contemporaneous effect of Y(t) on X(t) violates the classical assumption of . When the error term is correlated with the exogenous variable and , then the ordinary least squares (OLS) estimates of the parameters, α and β, become biased and inconsistent.

²⁴ The main features of the pre-1990s financial system in India included the stipulation of high cash reserve (CRR) and statutory liquidity (SLR) ratio, prevalence of administered interest rates, dominance of public sector banks, lack of competition. The CRR, which remained around 3% of demand and time liabilities of banks in the 1960s, increased to 5%–7% in the 1970s, 6%–9% in the mid-1980s and eventually to its statutory upper limit of 15% in the early 1991. Since the beginning 1990s, the CRR of banks has been gradually reduced to 11% in August 1998 and further to 4.75% in November 2002. The SLR, which remained around 25% in the 1960s, showed steep increases and it reached the zenith of 38.5% of net demand and time liabilities of banks in 1991. In the reform phase, the SLR has been brought down to 25% of net demand and time liabilities of the banks.