Finance and Growth in Mauritius-The Case of a Small Open Economy

Causality in Finance and Growth: The Case of a Small Open Economy VINAY PRASANDJEET NUNDLALL International Business School Brandeis University Waltham MA 02452-9110 USA ABSTRACT This paper investigates causality between economic growth and financial development in Mauritius over the period 1968 through to 2004 Using Engle and Granger error correction methodology with annual data, we find that financial intermediation has been contributing to growth in Mauritius since independence However, the equity market has not had any impact on the economy during its relatively shorter life span A channel of growth from financial intermediation to the construction sector is identified The study also finds that exports also have had a significant impact on growth, lending support to the export led growth strategy adopted by the authorities Introduction The UNDP 2003 Human Development Index ranks Mauritius sixty second overall and third behind the Seychelles and Libya among African countries Based upon GDP per capita, Mauritius ranks third among African countries, behind the Seychelles and Republic of South Africa Mauritius is a small, densely populated island of 1.2 million inhabitants living in an area of 1,860 square kilometers (720 square miles) The island does not have any natural mineral resources and has relied heavily on its monocrop sugar sector for exports during most its life as an independent nation Situated about 1,000 km (620 miles) off the eastern coasts of Africa in the Tropic of Capricorn, it is a victim of the vagaries of the Indian Ocean’s tropical climate However, its volcanic origin has endowed it with beautiful sandy beaches and a calm blue lagoon which has made it a popular holiday resort for European and South African tourists and made tourism an important sector of the economy Economic history teaches us that Mauritius was never destined to achieve economic success because, as Meade reports in 1961, the island was a crucible waiting to explode due to ethnic tension During the 1960’s, the economy relied solely on sugar for exports, a sector that was prone to trade shocks (and climactic conditions), while at the same time experiencing unbridled population growth Meade actually predicted that the then British colony would be caught in the Malthusian trap, and that the scramble for jobs would create tension between the ‘underdogs’ who were descendants of Afican slaves and Indian indentured labourers, and the wealthy Franco-Mauritian ‘top dogs’ However, Mauritius never fell in the Malthusian trap and if anything, achieved the opposite by developing an export processing zone, gradually diversifying away from sugar to textiles, tourism and financial services, and perhaps pertinently, upholding a stable economic and political environment after independence in 1968 The same, sadly, cannot be said for most Sub-Saharan African countries post independence Per capita income rose from US $1,000 in the early eighties to more than US $4,000 in 2004 The annual growth rate has been about 5% over the past two decades which has boosted the ranking of the country to the top of middle-income category economies In this study, we investigate some of the determinants of this growth, with special emphasis on the finance sector Barro’s (1991) seminal paper on economic growth has led to a spurt of creativity in the empirical growth literature Sala-I-Martin’s (1997) curiously titled “I just ran two million regressions” points to the direction which research has taken in the field; a medley of economic and socio-political variables have been tried in growth regressions However, the majority of studies are cross-sectional in nature, with the main determinants of growth identified as initial income level, investment rate, secondary school enrollment rate and the rate of population growth [Levine and Renelt (1992)] Unfortunately, there are not many case studies of countries using time series approaches This paper uses Mauritian annual data from 1968 to 2004 to estimate a growth regression applying time series techniques The purpose of the study is to identify factors that have contributed to growth over the past 35 years in this small island economy with particular emphasis on the role of financial intermediation While we control for capital investment, human capital and exports (since export led growth was a strategy explicitly adopted by the authorities), we find evidence of a positive contribution by the financial sector in facilitating growth The results also confirm that Mauritius has experienced export led growth Whilst the role of banks (financial intermediaries) has been significant in assisting economic growth both in the long term and in the short term, the stock market is, on the other hand, not important in defining growth at this stage of the countries development The paper is organized as follows; Section reviews from the existing literature the role of financial intermediation in an economy, Section contains a description of the data and the methodology used, Section presents the results and a discussion of their implications and finally, Section concludes Section – Financial development and economic growth Starting from the pioneering work by King and Levine (1993), Levine (1997), Levine and Zervos (1998) and Levine, Loayza and Beck (1998), many studies have investigated and uncovered a positive contribution of financial development on growth The seed of this idea actually goes as far back as Alexander Hamilton (1781, in Levine et al (2000)) who argued that “banks were the happiest engines that ever were invented” for spurring economic growth Other early records are from Bagehot (1873, in Levine et al (2000)) and Schumpeter (1911), who postulated that technological innovations, an important factor for growth, rely on external funds to come to fruition If the economy has a financial system, then banks can fund productive investments, and give innovators access to funding which enables them to undertake projects An illustration of this example at work is the Industrial Revolution in England Since England already had a functioning financial system, backed by an established and credible legal system, the country progressed by channeling funds into its industries during those crucial years Schumpeter explains how banks can choose which firms or entrepreneurs get to use society’s savings, hence positively influencing the path of economic development by tweaking the allocation of savings On the other hand, Bencivenga and Smith (1991) warn that higher returns from more efficient allocation of funds could depress savings rate and hence hamper growth Lucas (1988) further counters by saying that economists have badly over-stressed the contribution of the financial system Robinson (1952) too is skeptical of its influence on the economy, concluding that banks respond passively to economic growth Going way back in history, opponents to the banking system have been found among leading people of the nation - President John Adams (1819, in Levine et al (2000)) asserted that “banks harm the morality, tranquility, and even wealth” of nations Patrick (1966) and Goldsmith (1969) are among the earliest of modern writers who find a positive correlation between financial development and growth However, Patrick cautions that there is only proof of correlation and not causality Patrick actually sets up two relationships: causality can be supply-leading or demand-following Supply-leading means that development of financial institutions services induces investment and growth Demand-following says that the financial sector responds to increasing demand for their services from a growing real economy In addition, Patrick also hypothesizes there are stages of development that will experience the different causal relationships That is, causality between finance and growth changes over time as the economy develops During the early stages, financial development spurs growth and innovation as it reallocates funds from savers to modern sectors of the economy and encourages entrepreneurs to put their ideas into practice At higher development levels, the supply-leading force of financial development gradually weakens demand-following Financial development responds increasingly to output growth, so we have McKinnon (1973) and Shaw (1973) specifically address the supply-leading hypothesis and recommend governments to liberalize their financial sector in order to spur growth More recent studies like Jung (1986) delve into the time series aspect of the problem Using bivariate causality tests to detect temporal patterns in causality, Jung does not find support of Patrick’s hypothesis Xu (2000) finds a negative relationship between bank-based financial development and growth in 14 middle and low income countries (mostly African), but finds significant positive long run effects of financial development on growth in 27 other countries Wachtel and Rousseau (2000) show that banks and stock market development both explain growth Arestis, Demetriades and Luintel (2000) use quarterly data from five OECD countries and find that banks and stock markets both cause growth, but that the effect of banks is larger This paper develops an error correction model and finds that while financial intermediation as proxied by bank lending to the private sector is important for economic growth, the stock market is not significant in explaining growth in a small developing economy However, since the Stock Exchange of Mauritius was only established in 1989, we have only 16 years of observations for carrying out tests on the stock market’s importance The result, even if not surprising due to the smallness of exchange, cannot be generalized because of the length of the time series Section – Data and Methodology We analyze the effect of stock market and bank development on growth in Mauritius using annual data from 1968 to 2004 - quarterly data of economic variables are not available 1968 marks the year of independence from British rule, and also the year when most socio-economic data collection started Data for this study has been extracted from the Central Statistical Office (CSO), and The International Financial Statistics (IFS) webpage of the IMF In what follows, we describe the indicators of stock market development and bank development We use three measures of stock market development; market capitalization to GDP ratio, turnover ratio and value of shares traded ratio Market capitalization ratio is an indication of size and it is the value of all listed shares divided by GDP Total value traded to GDP is an indicator for activity or liquidity and is defined as total shares traded on the exchange divided by GDP The efficiency indicator we use is turnover ratio, which is the value of total shares traded divided by market capitalization It measures the activity of a stock market relative to its size because it is important to distinguish between a small stock market that is active (has high turnover ratio) and a large market that is less liquid (and has a low turnover ratio) In theory, one should be careful in using the market capitalization indicator as, if markets are efficient, market capitalization already reflects the discounted future value of the economy Hence, if causation is from economic growth to stock market, it is the opposite that will be revealed Measuring bank development is more straightforward We use activity which is claims on the private sector made by deposit money banks divided by GDP This measure excludes loans issued to public enterprises and government, thus isolating loans given only to the private sector (which includes corporations, various enterprises and households) A measure of liquidity, or financial depth in our study, is currency plus demand and interest-bearing liabilities of banks and other intermediaries divided by GDP Financial depth is also a measure for the overall size of the financial sector The measure for capital investment is gross domestic fixed capital formation from the national accounts Since statistics for labour force is only available as from 1976 onwards, we use population as a proxy for labour The correlation between population and labour force is 0.99 over the available sample Further, normalizing by population gives us a more interesting measure; GDP per capita as opposed to GDP per labour A measure of human capital is gross secondary enrolment, which is available for the country More pertinent measures, such as the level of education attained by members of the workforce, are unfortunately not available for the sample we are looking at Exports are measured by exports of goods and services, as a share of GDP Since tourism, an exported service, is very important for Mauritius, we adopt exports and services as opposed to exports of goods All variables are measured in MRU, and deflated by CPI (base year 1992) We start with the following aggregate production function: Y F  K , L Yt ( At Lt )1    K t H t (1) Where Y is real GDP, L is population, K is physical capital, H is human capital, and A is the level of technological efficiency and economic efficiency Economic efficiency includes economic and institutional variables exports X and financial intermediation F Normalizing with respect to L and taking logs, we have ln y    1 ln K   ln H   ln X   ln F (2) The model to be estimated is therefore: ln y t   1 ln K t   ln H t   ln X t   ln Ft   t (3) All the coefficients are expected to have a positive sign and be significant Controlling for K, H and X, F is expected to be positive and significant In order to construct the error correction mechanism (ECM), we first need to test whether the series in the model are all stationary and integrated of the same order If they are all integrated of the same order d (if they are I(d)), we check whether they all share a common stochastic trend - that is whether they are co-integrated Following Engle and Granger (1987) breakthrough theory of co-integration, suppose two time series xt and yt are related via the following relations: y t  xt u t u t c1  1u t    1,t (4a) y t  x t v t (4b) v t  c   v t    ,t where δ ≠ η hence restricting both parameters from being equal to zero at the same time ≤ ρ1 ≤ ≤ ρ2 ≤ c1 and c2 are intercept terms ε1,t and ε2,t are standard white noise error processes, mutually independent at all lags Equations (4a) and (4b) represent two distinct linear combinations of xt and yt that can be described by AR(1) models The interpretation of the two models however depend upon the values that ρ1 and ρ2 take We have three relevant cases which will each imply a different interpretation of (4a) and (4b) In the first case, where ρ1 = ρ2 =1, any linear combination of xt and yt is a random walk Therefore both xt and yt are non-stationary processes Both series have a stochastic trend, and they not share this trend as no linear combination of xt and yt is itself stationary In the second case, both ≤ ρi ≤ 1(for i = 1, 2) Then any linear combination such as (4a) and (4b) above is a stationary AR(1) process, and xt and yt are individually stationary variables The third and most interesting case is when ρ1 = and ≤ ρ2 ≥ (or vice-versa) There is then one linear combination of xt and yt which is a stationary AR(1) process, while the other combination is a random walk Further, it means that even though individually xt and yt are I(1) time series, there is one combination of these two which is stationary In the language of Engle and Granger (1987), these two time series are cointegrated Cointegration implies that these series have a common stochastic trend – in other words, they move together in unison, and any divergence between these two series is only transitory Testing for cointegration is then quite straightforward We first test that xt and yt are I(1) This is done by applying the Augmented Dickey-Fuller (ADF) test on each process: k 1 y t   t  y t     i  y t  i   t (5) i 1 The null hypothesis is ρ = 0, that is there is a unit root However, the proper test statistic to use in the ADF is not the t-statistic, but the τ-statistic The number of differenced lags to be used is also important as one should care about the degrees of freedom (especially in a small sample like the one we have here) In this study, one lag happens to be sufficient So once xt and yt are found to be I(1), a linear combination of the two processes is run (consistent with causality) and the residuals saved Suppose we run y t ˆ  ˆxt  uˆ t (6a) Then uˆ t  y t  ˆ  ˆxt (6b) ut could be I(1) However, in special circumstances where ut is I(0), (ie it is stationary and rarely drifts away from zero) then the constant δ is such that the ‘bulk of the long run components of xt and yt cancel out’ xt and yt and are said to be cointegrated with a cointegrating vector [1 -δ]’ Generally, if the variables are I(d) and the errors are I(b), where b < d, then we have cointegration Equation (6a) is called the cointegrating equation Formally, the auxiliary test regression for cointegration is k 1uˆ t   uˆ t     i 1uˆ t  i   t (7) i 1 So if ut is I(0), then it can be used in the dynamic regression below in what is known as the Granger Representation Theorem: k l i 1 i 1 1 y t    1 (uˆ t  )    2,i  y t  i    3,i  xt  i   t (8) β1 reflects the speed of adjustment towards equilibrium Equation (8) is the Error Correction Model (ECM), where generally, there is Granger causality if either β1 is significant, or the β2’s and the β3’s are significant The number of lags k and l to be included will be determined by the Akaike Information Criterion (AIC) Fortunately, for this sample, one lag in each differenced variable gives the most significant results and hence there is minimum loss of degrees of freedom Section – Results 3.1 Financial Intermediation and Growth We first present the results for financial intermediaries (or banks) The ADF tests for the levels of the variables and the differenced variables are given in Table below: Table 1: ADF tests for levels and differences in variables Levels First Differences ( ) Variable Type Rho Tau Pr < Tau Rho Tau Pr < Tau lnGDP Zero Mean 0.26 2.72 0.9978 -10.86 -2.23** 0.0266 Single Mean -1.33 -1.29 0.6233 -21.27 -3.16** 0.0310 Trend -10.48 -2.26 0.4407 -22.90 -3.27* 0.0885 Zero Mean 0.37 1.23 0.9417 -9.59 -2.12** 0.0345 Single Mean -4.14 -2.17 0.2218 -12.25 -2.39 0.1513 lnK Levels Variable lnH lnEXP lnACT lnDEPTH First Differences ( ) Type Rho Tau Pr < Tau Rho Tau Pr < Tau Trend -10.33 -2.40 0.3707 -14.12 -2.58 0.2891 Zero Mean -1.01 -2.81 0.0062 -8.39 -1.97** 0.0482 Single Mean -0.56 -0.54 0.8716 -21.52 -3.13** 0.0332 Trend -8.33 -1.97 0.5973 -21.70 -3.08 0.1268 Zero Mean -0.67 -0.68 0.4151 -26.08 3.47*** 0.0010 Single Mean -10.67 -2.32 0.1710 -26.21 -3.41** 0.0171 Trend -14.17 -2.38 0.3808 -26.83 -3.41* 0.0670 Zero Mean -0.88 -1.82 0.0655 -46.94 4.70*** |t| Intercept 6.241*** 0.3055 20.43 r Eigenvalue Trace 5% Critical Value 0 0.6287 72.9656 68.68 1 0.4475 38.2862 47.21 2 0.2709 17.5226 29.38 3 0.1656 6.4656 15.34 4 0.0037 0.1294 3.84 Tables 7a and 7b show the results from the Johansen maximum likelihood test on the cointegrating regression in levels (Equation (5)) Table 7b: Johansen Maximum Eigenvalue test for Cointegration Rank Cointegration Rank Test Using Maximum Eigenvalue H0: Rank=r H1: Rank=r+1 Eigenvalue Maximum 5% Critical Value 0.6287 34.6794 33.46 0.4475 20.7636 27.07 0.2709 11.0571 20.97 0.1656 6.3362 14.07 0.0037 0.1294 3.76 The tests use a VAR(2) model with intercept and no linear trend Both the trace and l-max tests consistently reveal that we have a unique root in our model With only one cointegrating vector, we can plausibly estimate a single equation instead of a system of equation The single equation estimated the VECM gives a cointegrating vector [1 - 5.65 - 0.35 - 0.62 - 0.26 - 0.34]’ or, in equation form: ln GDPt 5.65  0.35 ln K t  0.62 ln H t  0.26 ln EXPt  0.34 ln ACTt Note that since the estimation methods for the VECM and the single equation ECM are different, the coefficients are not directly comparable Coefficients for the Single Equation estimation from the VECM are given in Table below α the coefficient of EC(t-1), the error correction term, is -0.6451 A test for weak exogeneity, which is a test for the null hypothesis that α = 0, is rejected at the 5% significance level (Chi-Squared value of 6.08, P value of 0.0137) α is negative and significant which implies that the long run relationship is an equilibrium relationship The test for weak exogeneity also implies that financial activity causes GDP The number of lags in log differences of the dependent variables used in the VECM is three The results indicate that in the short run, banking activity has quite a big impact after one year (significant at 10% significance level) as the coefficient (elasticity) is 0.22 The slightly lesser impact is discerned after two years, as the elasticity drops to 0.17 but the significance is again marginal After three years, the impact is substantially higher at 0.34, and highly significant Table 8: VECM(3) estimations Dependent variable is ΔGDP(t) Parameter Estimate EC(t-1) **-0.6451 Standard Error t Value Pr > |t| 0.0137 ΔGDP(t-1) *0.5188 0.2761 1.88 0.0847 ΔK(t-1) -0.1906 0.1400 -1.36 0.1981 ΔH(t-1) -0.6664 0.5298 -1.26 0.2324 ΔEXP(t-1) 0.0132 0.1123 0.12 0.9087 ΔACT(t-1) *0.2209 0.1151 1.92 0.0791 ΔGDP(t-2) *0.5891 0.3226 1.83 0.0928 ΔK(t-2) *-0.2541 0.1391 -1.83 0.0927 ΔH(t-2) 0.0517 0.6315 0.08 0.9361 ΔEXP(t-2) **0.2628 0.1090 2.41 0.0328 ΔACT(t-2) *0.1749 0.0939 1.86 0.0869 ΔGDP(t-3) 0.2295 0.2757 0.83 0.4214 ΔK(t-3) -0.0925 0.0782 -1.18 0.2600 ΔH(t-3) -0.2351 0.3260 -0.72 0.4847 ΔEXP(t-3) **0.3596 0.1514 2.38 0.0350 ΔACT(t-3) ***0.3436 0.1032 3.33 0.0060 *significant at 10%, ** significant at 5%, ***Significant at 1% Section 3.3 – Identifying some Channels In this subsection, we attempt to identify some channels through which financial intermediation could lead to growth Data on the number of non-residential building permits issued is available from 19762004 Data is also available for the total floor area for each year In what follows we posit that if loans from banks are used to invest in these construction projects (whether it is a factory, warehouse, or administrative offices), we should be able to detect the impact in a regression However, since Foreign Direct Investment (FDI) has also been flowing in the country, especially in the manufacturing sector and tourism sector, it would be interesting to include FDI in the model as well Results are reported in Table below Table 9: Effect of Financial Intermediation on New Buildings Dependent variable is ln(Mean Area) Variable Parameter Estimate Standard Error t Value Pr > |t| Intercept ***2.6421 0.9348 2.83 0.0101 lnAREA(t-1) **0.4156 0.1972 2.11 0.0472 Activity *1.2022 0.6874 1.75 0.0949 FDI 2.4113 5.1471 0.47 0.6443 DEXP 0.0663 0.6084 0.11 0.9143 Tradelib **0.4497 0.2068 2.17 0.0413 F-Value ***19.500 R2 0.8228 Adj-R2 0.7806 Durbin-Watson D 2.2610 Number of Obs 27 |t| Intercept -0.0082 0.02015 -0.41 0.6861 lagdcons ***0.2965 0.08811 3.37 0.0028 growth ***0.6116 0.06527 9.37