SAVINGS, INVESTMENT AND GROWTH IN SOUTH ASIA. Pradeep Agrawal JUNE 2000

Transcription

1 SAVINGS, INVESTMENT AND GROWTH IN SOUTH ASIA Pradeep Agrawal JUNE 2000 Indira Gandhi Institute of Development Research, Gen. A.K. Vaidya Marg, Goregaon (E), Bombay , India pradeep@igidr.ac.in Fax: (22) Phone: (22)

2 SAVINGS, INVESTMENT AND GROWTH IN SOUTH ASIA Pradeep Agrawal 1. INTRODUCTION High savings and investment rates are important in view of their strong and positive association with the GDP growth rate as suggested by endogenous growth theory (see e.g., Romer 1986; Lucas, 1988) 1. The empirical evidence (see, for example, Levine and Renelt, 1992) also indicates that there exists a robust positive correlation between the investment rates and GDP growth. Rapid GDP growth is needed to generate more employment and better living standards. It is also important for reducing poverty given that income distribution has remained largely unchanged in South Asian countries over the last 50 years of independence (Jain and Tendulkar, 1990). Thus, high savings and investment rates appear very desirable for improving national welfare. Therefore, in this paper we study the savings and investment behaviour in five South Asian countries: India, Pakistan, Bangladesh, Nepal and Sri Lanka. Figures 1, 1A and 1B show the savings rates, private savings rates and public savings rates respectively for the South Asian countries for Similarly, Figures 2, 2A and 2B show the investment rate, private investment rates and public investment rates respectively for the South Asian countries for For comparison, the 1 The neoclassical growth theories (see e.g. Solow, 1956) assume that marginal product of capital eventually falls to zero so that equilibrium growth eventually stops, irrespective of the level of the savings or investment rates. But even in the neoclassical model, the long run level of income per capita is higher the higher the savings or investment rate. Further, Srinivasan (1993) shows that if the assumption that marginal product of capital eventually falls to zero is replaced by the assumption that it eventually falls to a certain minimum but positive level, than even the neoclassical model has implications similar to those of the endogenous growth theories. 2

3 average for five East Asian countries (South Korea, Singapore, Malaysia, Thailand and Indonesia) is also shown. It is seen that the savings and investment rates over have been in the range of 7-25 percent for the South Asian countries while they were in the range of percent for the East Asian countries. Furthermore, in the 1960s, the savings and investment rates were comparable in South and East Asia but then East Asia has gradually forged ahead beginning in the 1970s. In terms of the components, regarding private savings (Figs 1A) also, the performance of the East Asian countries is superior, with India, Pakistan and Sri Lanka showing relatively better performance among the South Asian countries. In public savings rates (Fig 1B), while East Asian countries average rates of over 5 percent, Sri Lanka and Pakistan average about zero percent, with India and Bangladesh doing marginally better and averaging about 2 percent. Nepal s public savings rates rival those of East Asia, but this is primarily due to the high levels of foreign aid it receives. In private investment, performance of East Asia and Sri Lanka is best (Fig 2A), followed by that of India. Public investment rates (Fig 2B) are highest in Pakistan, followed by East Asia and India. They are the lowest in Sri Lanka and Bangladesh. To understand the reasons behind the relatively poorer performance of the south Asian countries, we undertake an econometric analysis of the total and private savings and total and private investment ratios in South Asia. There already exists an extensive literature on the subject of savings and investment (see, e.g., Schmidt-Hebbel et al, 1996; Edwards, 1996; Fry, 1980, 95; Bosworth, 1993; Carroll & Weil, 1994, Deaton 1991; Lahiri, 1988, and Giovaninni 1983, 85; Modigliani, 1970; Jorgenson, 1963, 71). The literature provides several stylized facts concerning savings and investment. These are: (1) saving and investment rates and economic growth 3

4 rates are positively correlated; (2) savings rates also depend on demographic factors such as the age dependency ratio; (3) investment depends primarily on the availability of domestic credit; and (4) Growth rates cause (determine) savings rates, rather than the other way round. However, inadequate work has been done on South Asian Countries. We do not really know whether these stylized facts apply for South Asia or do there exist important differences. Furthermore, there are many important policy issues, including understanding the reasons for lagging behind the East Asian countries and finding ways to improve the South Asian Savings and investment rates. Thus, the purpose of this paper is to produce careful and detailed empirical work on the determinants of savings and investment in South Asia and try to understand why South Asian savings and investment rates lag behind the East Asian rates and how they can be improved. We also carry out analysis of causality between savings rate and GNP growth rate as this has significant implications for development policy. If savings drive growth through an automatic translation of savings into capital formation, the main goal of the development policy should be to encourage savings and investment. On the other hand, if growth results less from savings and investment and more from other factors such as technological innovation, human capital and trade policy, then they should be the main targets of the development policy. The plan for the paper is as follows. In Section 2 we econometrically analyze the determinants of total and private investment rates in South Asia using panel data from the five South Asian countries. Similarly, in Section 3, we econometrically analyze the determinants of total and private savings rates in South Asia using panel data. countries. In Section 4, we report on the Granger Causality tests between savings and 4

5 GNP growth rates, to determine whether savings cause growth or vice versa. The main conclusions of the paper are summarized in section THE INVESTMENT BEHAVIOUR In this section, we econometrically analyze the determinants of total and private investment rates (measured by total and private gross fixed investment as a fraction of GDP) using panel data from South Asia. In Section 2A, we briefly discuss the various variables that affect investment and derive a likely functional form for it. Then, we briefly discuss the econometric procedures used in this paper followed by the discussion of the results of our estimation (Section 2B). 2A. The Investment Function Blejer and Khan (1984, p. 383) describe some of the difficulties of estimating neoclassical investment functions for developing countries, such as, the lack of readily available measures of the capital stock or its rate of return. Thus, the investment function estimated here is based on the flexible accelerator model as developed in Fry (1998). The accelerator model has the desired capital stock K* proportional to real GDP, y: K* = α y. (1) This can be expressed in terms of a desired investment rate (INV/Y)*: (INV/Y)* = α GROWTH, (2) where INV denotes gross domestic fixed investment in current prices, Y denotes GDP in current prices and GROWTH is the rate of growth of real GDP, y. A partial adjustment mechanism allows the actual investment rate to adjust to the difference between the desired investment rate and the investment rate in the previous period: 5

6 (INV/Y) t = λ [(INV/Y)* (INV/Y) t-1 ] Or, (INV/Y) t = λ (INV/Y)* + (1 λ)(inv/y) t-1, (3) where λ is the coefficient of adjustment. The flexible accelerator model allows economic conditions to influence the adjustment coefficient λ. Specifically, λ = β0 + (β1 X 1 +β2 X 2 +β3 X 3 + )/[(INV/Y)*-(INV/Y) t-1 ], (4) where X i are the variables (including an intercept term for a constant depreciation rate) that affect λ and βi are their respective coefficients. The explanatory variables used here are GDP growth rate (GROWTH) over the previous year, domestic credit availability as share of GDP (CRTOT/Y), net foreign direct investment inflows as share of GDP (FDI/Y), terms of trade (TOT) an index with base year (1990) value set to 100, real exchange rate (RER), net total foreign borrowing as share of GDP ( TED/Y) and the cost of capital (COC) or real lending rate (RL). A dummy variable for financial liberalization is also included. These variables are explained below. The availability of institutional credit is one of the most important determinants of the investment rate in developing countries (see, Blinder and Stiglitz, 1983; Fry, 1995). The quantity of credit is likely to be important in a credit market where the interest rates are controlled at below market clearing levels and/or directed credit programs exist for selected industrial sectors (the later have continued in most of the countries under consideration even after they have liberalized their financial system over the 1980s). Further, banks specialize in acquiring information on default risk. This information is highly specific to each client. Hence, the market for bank loans is a customer market, in which borrowers and lenders are very imperfect substitutes. A credit squeeze rations out some bank borrowers who may be unable to find loans elsewhere and so be unable to finance their investment projects (Blinder and Stiglitz, 6

7 1983, p. 300). Here, therefore, the investment rate INV/Y is influenced by the ratio of total domestic credit to GDP, CRTOT/Y. The inflows of foreign direct investments constitute a source of funds for investment. The use of this source of funds depends on the various policies towards FDI (such as the maximum share of foreign equity allowed in joint venture, regulations relating to repatriation of profits, various other policies and regulations relating to industry and labour and the overall attractiveness of the country). The foreign direct investment can also promote domestic investment through the backward and forward linkages with the domestic industries. Sri Lanka has attracted significantly more FDI inflows following its economic reforms in Similarly, India has attracted more FDI inflows following it s economic reforms beginning in 1991 (including liberalization of policies towards FDI these include raising the maximum allowed share of foreign equity from 40 percent to 51 in all sectors and as much as 100 percent in selected sectors such as the export oriented and high-technology industries. Nepal and Bangladesh have continued to attract relatively low levels of FDI. Relative to many other regions of the world, such as the dynamic East Asian region, South Asia has attracted much lower levels of FDI (Figure 5). We measure the FDI inflows net of the outflows. The outflows are minor for South Asian countries but have become substantial for some East Asian countries, such as South Korea and Taiwan. Similarly, the foreign borrowing can be used as a source of funds for investment (although public foreign borrowing does have a tendency to go (partially) into meeting urgent government budgetary requirements as well). Thus, we also include as an explanatory variable, the total foreign borrowings (net of interest re- 7

8 payments) each year as a share of GDP. This variable is calculated as the change in the total external debt (available in dollars) over the last year, which is then converted to local currency using average official exchange rate for the year, and then divided by the GDP thus it is denoted as TED/Y. This variable will also allow us to compare the relative effectiveness of foreign borrowing and FDI inflows in promoting investment. An improvement in the terms of trade (unit price of exports divided by the unit price of imports) can increase investment by increasing real income, making capital goods (mostly importable in developing countries) cheaper relative to the domestic goods. In some situations, it could also decrease investment by decreasing the demand for domestic goods compared to importables (see, for example, Cardoso, 1993). Changes in real exchange rate (RER) can also affect investment demand. RER is defined as: RER = E.P f /P (5) where E is the exchange rate (number of domestic currency units per Dollar), P is the domestic price level (the GDP deflator) and P f is the foreign price level (proxied here by the US GDP deflator, given that USA is the most important trading partner of the countries being studied). An increase in RER would increase the price of imported capital and intermediate goods and result in a contraction of investment (Serven and Solimano, 1992; Fry, 1995). Lizardo and Montiel (1989) and van Wijnbergen (1985) develop two sector models that show that the net effect of a real depreciation is ambiguous investment in tradable goods increases while that in the domestic goods declines. 8

9 True domestic costs of capital are difficult to measure in the South Asian countries because of selective credit policies and dis-equilibrium institutional interest rates (Fry 1995). We have used the measure of cost of capital (COC) suggested by Jorgenson (1967): COC = (RL + δ) P k /P (6) where RL is the real interest rate on bank loans, δ is the depreciation rate (set at 15 percent), P k is the price of capital goods (proxied by the implicit price deflator of investment), and P is the general price level of output, proxied by the GDP deflator. This measure excludes the effect of changes in tax policies, for which adequate data was lacking. As an alternative measure, we have also used the real lending rate of banks as a proxy for the cost of capital. The real lending rate (RL) is obtained by subtracting from the nominal lending interest rate of banks, the average of current and next year s inflation rates. Finally, in order to examine whether financial liberalization in the South Asian countries improved investment rate, we have used a financial liberalization dummy (FLIB) which takes the value 1 for all years following financial liberalization and 0 for all years before financial liberalization. A positive coefficient for this variable will suggest that liberalization improved investment rates (over and above its effect on other included variables such as, the credit availability, FDI inflows and the GDP growth rates, etc.). Given the above explanatory variables, equations (3) and (4) suggest estimating a relation for total gross fixed domestic investment of the type: INV/Y = b o + b 1 GROWTH + b 2 CRTOT/Y + b 3 FDI/Y + b 4 TED/Y + b 5 RER + b 6 TOT + b 7 COC + b 8 FLIB (7) 9

10 Similarly, we can also estimate the private gross fixed domestic investment. We use the same explanatory variables as above, but also include the public investment ratio (INVpub/Y) as an additional variable to see if the private investment has a complementary relation with the public investment. Thus we estimate : INVpvt/Y = b o + b 1 GROWTH + b 2 CRTOT/Y + b 3 FDI/Y + b 4 TED/Y + b 5 RER + b 6 TOT + b 7 COC + b 8 FLIB + b 9 INVpub/Y (8) The expected sign is positive for coefficients b 1 to b 4, b 8 and b 9, negative for b 7 and ambiguous for coefficients b 5 and b 6. 2B. Estimation Results The above relation is estimated using pooled time series cross section data for about 25 to 38 annual observations between 1960 and 1998 for the five South Asian countries under consideration. A dynamic version of the linear model was estimated, by including the lagged dependent variable since the explanatory variables can be expected to determine the change in the investment rate rather than its absolute level. We used the fixed effects model, which is usually appropriate when one has relatively small number of countries and large number of observations for each country. In this model, dummy variables were included for all but one country, although their coefficients are not reported for brevity. We employed the method of instrumental variables to estimate the relation since a feedback mechanism can be expected between investment rate and GDP growth rate on the one hand and investment rate and credit to GDP ratio on the other. However, for purposes of comparison and to see the magnitude of the feedback effect, we also report the results of panel estimation using ordinary least square (OLS). Finally, 10

11 heteroscedasticity of error term is a common problem with panel data and therefore we have applied White's heteroscedasticity correction to the standard error of the coefficients (see White 1982). The results of the estimation are shown in Table 1. The co-efficients for terms of trade (TOT), real exchange rate (RER) and the financial liberalization dummy (FLIB) were found to be insignificant and these variables were therefore dropped from the estimated relation. [Insert Table 1] The resulting relation is reported in columns A (Instrumental variable procedure) and column B (OLS procedure). The most important determinant of the investment rate seems to be the ratio of net FDI inflows to GDP (FDI/Y). It has a strongly positive effect with a coefficient of 1.81 (columns A and B). Note that when the lagged dependent variable is included in the regression, as is the case here, the long run co-efficient is obtained by dividing the coefficient of FDI/Y by 1- the coefficient of the lagged dependent variable 2. Using this result, the long run coefficient of FDI/Y is found to be about 5.7 to 5.9. This implies that a 1 percent increase in FDI is associated with over 5 percent increase in domestic investment. This shows that there is a strong complementarity between FDI and the domestic investment, possibly through various backward and forward linkages. Since the FDI inflow as share of GDP over averages about 3,5 percent in East Asia and less than 1 percent in South Asia, most of the difference in the investment to GDP ratios between South Asia and East Asia can be explained by the different levels of 2 To see the logic behind this, note that if Y t = ax t + by t1 then in the long run equilibrium where Y t =Y t-1 = Y and X t = X, we have, Y= ax+by, so that; Y= (a/1-b)x. 11

12 inflows of FDI. The impact of net total foreign borrowing as share of GDP ( TED/Y) is positive but much smaller than that of FDI inflows. This is probably due to the fact that much of public foreign borrowing is also used for balance of payment difficulties (which means it finances consumption of imported goods, rather than investment). Given the additional problems of proper management of projects to ensure interest repayments in the case of foreign borrowing, our results suggest that FDI inflows are superior to foreign borrowing as a way of financing investment. Another important determinant of investment rate is the total domestic credit to GDP ratio, CRTOT/Y. The short-term coefficient is about 0.1 and the long-term coefficient is about 0.3. The better financial development of most East Asian countries also helped the process by making more credit available for undertaking investments. The effect of real GDP growth (GROWTH) and the cost of capital (COC) on investment rate was not found to be significant (in either IV or OLS estimates, though the OLS estimates have the theoretically expected signs. Replacing COC with the real interest rates on bank loans (RL) resulted in very comparable results, with the coefficients of both growth and RL continuing to be insignificant. We have also separately estimated the relations for the private sector s gross domestic fixed investment (INVpvt/Y) as a share of GDP. The explanatory variables were the same as in Equation (7) except that the public investment rate (INVpub/Y) was added as an explanatory variable to account for possible complementarity between public and private investment. The results are shown in Table 1, column C (IV estimates) and Column D (OLD estimates). Generally, the same variables as those for the total investment rate seem to have contributed to 12

13 the private investment rate except that the FDI inflows (which are mostly in the private sector) and the credit ratio have even stronger impact, as might be expected. On the other hand, net foreign borrowing (which in South Asia, is mostly in the public domain and usually finances public sector investment (and consumption)) is less important, as one would expect. The coefficient of public investment ratio, INVpub/Y, is positive though insignificant. 3. DETERMINANTS OF SAVINGS IN SOUTH ASIA In this section, we econometrically analyze the determinants of total and private savings rates (measured by total and private gross national savings (GNS) as a fraction of the Gross National Product (GNP)) using panel data from South Asia. In Section 3A, we briefly discuss the various variables that affect savings rates and derive a likely functional form for it. Then, we briefly discuss the econometric procedures used in this paper followed by the discussion of the results of our estimation (Section 3B). 3A. The Saving Functions: The form of the savings function and the main variables affecting the savings rate has been discussed extensively in the literature. Thus, rather than repeating this discussion here, we go straight to a commonly used specification for the savings function (for a detailed discussion of the functional form and the relevant variables, the interested reader is referred to Fry, 1995; Schmidt-Hebbel et al, 1996; Edwards, 1996; and Bosworth, 1993). We have particularly tried to include variables that might be relevant in the South Asian context (such as the bank offices per capita and the 13

14 financial liberalization dummy). The relation between the total savings rate, growth and other relevant variables is proposed to be as follows: GNS/GNP = a o + a 1 GROWTH + a 2 DEPEND + a 3 FS/GNP + a 4 M2/GDP + a 5 RD + a 6 URBAN + a 7 FLIB. (9) where Growth is the growth arte of real GNP, DEPEND is the age dependency ratio, FS is foreign savings, M2 is broad money, RD is real interest rates, Urban is the proportion of population living in urban areas and FLIB is a dummy variable for post liberalization years. Similarly, we can also estimate the private savings rate. We use the same explanatory variables as above, but also include the public savings rate (GNSpub/GNP) as an additional variable to see if the public savings act as a substitute for the private saving rate and to what extent. Thus, we estimate: GNSpvt/GNP = a o + a 1 GROWTH + a 2 DEPEND + a 3 FS/GNP + a 4 M2/GDP + a 5 RD + a 6 URBAN + a 7 FLIB + a 8 GNSpub/GNP. (10) The expected sign is positive for a 1 and a4 and negative for a 2, a 3 and a 8, while the expected signs of a 5, a 6 and a 7 are ambiguous. A detailed description of the various variables, including how these are measured in our work, is provided below: Savings Rate (GNS/GNP): The savings rate (GNS/GNP) is measured as the ratio of nominal Gross National Savings, S, to nominal Gross National Disposable Income, GNP (Gross National Product). We use gross rather than net savings because of the very arbitrary nature of capital consumption allowances, which also vary across 14

15 countries and thus would create problems of comparability of our estimations across countries. National, rather than domestic savings are used because the former includes the net factor income from abroad while the latter do not. Many South Asian countries have sizeable remittances from workers working abroad, and a good part of such savings are saved and are available for domestic investment. Furthermore, the interest payments on foreign debt are not included in national savings while they are included in domestic savings. Thus the national savings appears to be the more appropriate measure of savings available to finance domestic capital formation, which in turn fuels growth according to the growth theories. Thus, given our emphasis on the relation between the savings and growth rates, gross national savings seems to be the appropriate measure of savings for our purposes. Real Gross National Product (RGNP) and its GROWTH Rate: The endogenous and Harrod-Domar growth models that higher savings rate are associated with higher levels of growth of real income or output. Here, the real (in constant domestic prices) gross national product (RGNP) is used as a measure of real income. The growth rate of RGNP is used as a measure of the growth rate (denoted by GROWTH). As discussed previously, the expected sign of the coefficient of GROWTH is analytically ambiguous, though previous empirical studies have found it to be positive. Demographic Variables (DEPEND): According to the life cycle model, individuals will have negative savings when they are young and have low or zero income, positive savings during their productive years and once again, negative savings when they are old and retired. Thus aggregate savings will be affected by the age distribution of the 15

16 population - if the share of inactive or dependent population is high, the savings ratio will be low. We use the age dependency ratio, the ratio of dependent population (those under 15 years of age and 65 years or older ) to the working age population (aged 15 to 64 years), as a reasonable proxy to capture this effect (even though it is true that not everyone aged years would be working and saving and not everyone under 15 or over 64 would be necessarily dependent or dis-saving). Other proxy measures of this effect, such as the share in population of the labour force or number of employed, suffer from even more serious problems due to the lack of adequate data on those self employed and those working in the informal sectors, especially in rural areas. The expected sign of the coefficients of DEPEND is negative. Urbanization Ratio: This variable measures the fraction of total population living in urban areas. It is included to allow for differential savings rates between urban and rural populations due to such factors as improved access to savings institutions and instruments in urban areas (which should lead to an a positive coefficient for this variable) and different levels of income uncertainty etc.(which could lead to a negative coefficient for this variable since rural areas being much more dependent on agriculture face a higher uncertainty in income which may force them to save a larger proportion of their income). Thus, a priori, the sign of this variable is uncertain. Foreign Savings as Share of GNP (FS/GNP): Greater availability of foreign savings (measured by the current account deficits) may encourage more consumption, especially of importables, and thereby, reduce savings. That is, foreign and national savings are likely to be substitutes. The variable FS/GNP used in the relation to be 16

17 estimated is the negative of the current account balance as a proportion of the nominal gross national product, GNP. Real Interest Rates (RD): The early financial liberalization literature (McKinnon, 1973; Shaw, 1973) argued that raising real interest rates, RD, to market levels would generate greater savings and thus speed up economic growth. However, the actual effect of increases in real interest rates on savings rate is perhaps one of the most discussed controversies in this field, perhaps because the governments can easily manipulate interest rates. Since real interest rate is the Fischerian price of current consumption in terms of the future consumption, analytically, an increase in interest rates will have an ambiguous effect on savings because of a positive substitution effect towards future consumption and a negative income effect due to increased real returns on saved wealth. Olson and Bailey (1981) have argued that the positive substitution effect should dominate the negative income effect. Empirically Fry (1980,1995) has found small but positive interest rate elasticity of savings while Giovannini (1983, 1985) has found savings to be insignificantly related to real interest rates. In view of this controversy in the literature, it is of interest to re-evaluate the interest elasticity of savings. The real interest rates on bank deposits is the relevant rate of real returns for most households and even firms in developing South Asian countries since bank deposits are the principal form of financial assets. We have proxied the interest rate by the weighted average interest rate on all bank deposits in the case of Pakistan and Bangladesh, by unweighted average of all bank time deposits in the case of India and by the average one year bank time deposits in the case of Nepal and Sri Lanka. The choice was determined by the 17

18 availability of data and we have tried to take as broad a measure as the data would allow. At the same time, since the different interest rate series typically exhibit strong correlation among them, we do not believe that using these different measures of interest rates across different countries poses any significant problem for our analysis. The real expected interest rate (RD) is obtained as the difference of the nominal interest rate on bank deposits, ID, and the expected inflation rates over the following year, INF e. The expected inflation rate, INF e, in turn, is proxied by the average of actual inflation rates (calculated using the GDP deflator) for the current year and the following year. Measures of Financial Sector Development (M2/GDP and BOPM ): The savings ratio could depend on the level of financial sector development since that affects the access to appropriate savings instruments. Since bank deposits are the principal financial instruments in developing countries of south Asia, we use the ratio of broad money M2 to GDP (M2/GDP) as a measure of financial development. An alternative measure financial development or of access to banking services is the number of bank offices per million of population (BOPM). This was also used as an alternative to the first measure, M2/GDP. Financial Liberalization Dummy (FLIB): In order to examine whether financial liberalization improved savings rate, we have used a financial liberalization dummy (FLIB) which takes the value 1 for all years following financial liberalization and 0 for all years before financial liberalization (except, in the case of India, the value 1 was also assigned for the pre-bank nationalization years as banks were generally free 18

19 of excessive government regulation and control during this period). A positive coefficient for this variable will suggest that liberalization improved savings (over and above its effect on other included variables such as, the real interest rate and the growth rate of real GNP, etc. Public Sector Savings: This variable is included as an explanatory variable in the estimation of private sector savings since greater public savings imply either greater tax revenue or lower public expenditure, which mean lower disposable income for the private sector and hence lower savings. The main interest here is to see whether the coefficient of public savings is close to one or significantly less than one. In the latter case, an increase in public savings can be used as a possible way to increase aggregate savings. The official data for public savings was available only for India and Sri Lanka. For the remaining countries, the total public revenue minus total current expenditure approximated it. Other variables not included here: The basic life cycle model suggests that the net wealth (and variations in its value due to capital gains and losses) should be a determinant of savings. However, measures of wealth are generally not available for the South Asian countries considered here. While rough measures could possibly be constructed Bosworth (1993, chapter 3), who attempts such an exercise for 12 OECD countries (which have significantly better and longer data series available than is the case for the South Asian countries considered here) found it to be insignificant in all individual country regressions. In view of this, we have not attempted to use this variable here. 19

20 Another variable that is often considered relevant in savings functions is the ratio of expenditure on social security and welfare as a proportion of GNP. However, since social security and welfare system is not well developed in South Asia, this proportion is usually small and good quality data is generally not available. Therefore, this variable was not included in our analysis. 3B. Estimation Results The above relation is estimated using pooled time series cross section data for about 25 to 35 annual observations between 1960 and 1996 for the five South Asian countries under consideration. A dynamic version of the linear model was estimated, by including the lagged dependent variable since the explanatory variables can be expected to determine the change in the savings rate rather than its absolute level. We used the fixed effects model, which is usually appropriate when one has relatively small number of countries and large number of observations for each country. In this model, dummy variables were included for all but one country, although their coefficients are not reported for brevity. We employed the method of instrumental variables to estimate the relation since a feedback mechanism can be expected between savings and growth rates. Finally, heteroscedasticity of error term is a common problem with panel data and therefore we have applied White's heteroscedasticity correction to the standard error of the co-efficients (see White 1982). The results of the estimation are shown in Table 2 (Columns A and B). It is seen from column A that the urbanization ratio (URBAN) and the financial liberalization dummy (FLIB) have insignificant co-efficients. These variables were 20

21 therefore dropped from the estimation. The resulting estimation is shown in column B of Table 2. It is seen that the savings ratio is positively related to income growth (GROWTH), the coefficient being large (about 0.5) and significant at the 5 per cent level. This means that every 1 percent increase in the growth rate of real GNP is associated with about 0.5 percent increase in the savings rate. The coefficient of the age dependency ratio, DEPNED, is found to be negative and significant at the 1 percent level. It is noteworthy that the age dependency ratio remained virtually stagnant in Nepal, Pakistan and Bangladesh (until 1984) at around percent, which are also the countries with lower savings rates. It has declined rather slowly in India from about 80 percent in 1970s to about 65 percent in 1997, while in the case of Sri Lanka and East Asian countries, there has been a sharp decline in dependency ratio to about 50 percent (see Figure 5). Thus the higher GNP growth rates and a much sharper decline in population growth rates leading to lower dependency rates have contributed significantly to larger saving rates in East Asia as compared to South Asia. For example our co-efficients suggest that an additional growth rate of 5percent would contribute an extra 2.5 percent to the savings rate while a 30 percent decline in the dependency ratio would increase the savings rate by another 4 percent. [Insert Table 2] The effect on savings rate of greater availability of foreign savings was found to be significantly negative, as expected. Thus the large trade deficits that most South Asian countries have been accumulating might have lowered their savings rates. By contrast, the large trade surpluses that many East Asian countries have been accumulating since the late 1980s should have helped to raise their savings 21

22 ratios over this period. Among the financial variables, the financial development measure M2/GDP (ratio of broad money to GDP) was also found to be positive and significant. (We also tried the alternative measure of bank offices per capita, BOPM, but found it to be insignificant). However, we believe that the M2/GDP is perhaps a better measure of financial development. Our results suggests that in the context of South Asia, which has had a relatively low to moderate level of financial sector development, low level of financial development may be acting as a barrier to savings by some households (especially in rural areas), perhaps due to lack access to banking institutions. The ratio of M2 to GDP is shown in Figure 4 for South Asia and East Asia. It is seen that Bangladesh and Nepal are also among the least developed on this count too. Sri Lanka is also among the poor performers on this count but makes it up on other fronts, especially the dependency ratio. The coefficient of real interest rates was found to be positive but very small (a 5 percent increase in real interest rates would increase savings by about 0.25 percent) and not significant. Thus our results support the earlier findings of Giovannini (1983, 1985) that interest rates do not have much effect on savings. Our results do not suggest that there is any room for interest rate policies to have any significant effect on savings. We have also separately estimated the relations for the private sector s gross national savings as a share of GNP. The explanatory variables were the same as in Equation (9) except that the share of public savings was added as an explanatory variable to account for possible substitutability between public and private savings. The results are shown in Table 2 (columns C and D). Generally, the same variables 22

23 as those for the total savings rate seem to have contributed to the private savings rates except the age dependency ratio has an even stronger impact, as might be expected. An important result from this estimation is that a 1 percent increase in public savings reduces private savings by about percent. This suggests that a practical way to improve aggregate savings rate is to increase public sector savings (or reduce dis-savings). Figure 3 shows that while the public savings ratio in the early 1990s averaged about 8% in East Asia, it was actually negative for three South Asian countries, India, Pakistan and Sri Lanka. 4. CAUSALITY ANALYSIS: DO HIGH SAVINGS RATES CAUSE HIGH GROWTH? The relationship between the savings rate and economic growth has received considerable attention in the theoretical and empirical literature 3. However, the direction of causality between savings and growth is still far from clear even though the answer to this question is of critical importance for development policy. If savings drive growth through an automatic translation of savings into capital formation, the main goal of development policy should be to encourage savings. While if growth results less from savings and capital accumulation and more from other factors such as technological innovation, human capital and trade policy, then they should be the main targets of development policy. This section explores the relation between savings and growth and the direction of causality between them for the five South Asian countries. The theoretical underpinnings of the relation between savings and growth can be traced to the growth model of Harrod (1939) and Domar (1946). In the simplest version of their model with one factor (capital) and with a fixed capital requirement per 23

24 unit of output (i.e., constant marginal returns to capital), it easily follows that the growth rate of output would be proportional to the investment or savings rate 4. In the model with two factors (capital and labour), a similar requirement of labour per unit of output is added in a full employment economy with labour growing at an exogenous rate. Then the savings and growth rates become endogenous. However, since the labour requirement may not be binding in the context of many developing countries who often have "unlimited supplies of labour" (see Lewis, 1954), the one factor model would be appropriate, so that growth would be proportional to the savings rate, prompting Lewis (1954) to argue that: "The central problem in the theory of economic development is to understand the process by which a community which was previously saving and investing 4 or 5 percent of its national income or less converts itself into an economy where voluntary saving is running at about 12 to 15 percent of national income or more. This is the central problem because the central fact of development is rapid capital accumulation (including knowledge and skills with capital). We cannot explain any "industrial" revolution until we can explain why saving increased relatively to national income." (p. 155). Solow's (1956) model of growth featured a production function with smooth substitution between factors of production in contrast to the fixed proportions structure of Harrod-Domar model. In this model, because of the assumption of decreasing marginal returns to capital, the growth eventually stops, but the economies with higher 3. For a comprehensive review of the literature, see Schmidt-Hebbel et al. (1994). 4. More formally, it easily follows that, (dy/dt)/y = s/k, where s is the savings rate, k is the capital output ratio and Y is the output per period. 24

25 savings rates enjoy a higher steady state income. The more recent endogenous growth models (Romer 1986, Lucas 1988) which return to the Harrod-Domar type assumption (albeit with much better explanations for it) of constant returns to capital, imply that high savings rates and high capital formation lead to high growth rates of output. On the other hand, consumption theories, such as the permanent income and life cycle hypotheses, imply the reverse direction of causality, i.e., they imply that people choose their consumption (and thence also savings) levels depending on current and (expected) future income levels. Modigliani (1970) has argued that the simple version of life-cycle hypothesis implies a positive relation between savings and income growth. He notes that if their were no income and no population growth across generations, the savings of the young would exactly balance the dis-saving of the old and the aggregate savings rate would be zero. Because income growth makes the young richer than the old, the young will be saving more than the old will be dis-saving, resulting in the positive association between savings and growth. However, Carroll and Weil (1994) have argued that, ceteris paribus, an exogenous increase in the aggregate growth will make forward looking consumers feel wealthier and thus consume more and save less - thus implying that the impact of income growth on savings could be negative. On the other hand, if consumption is habit based and changes slowly in response to changing income, a larger fraction of increases in income may be saved resulting in the savings rate increasing with income increases (Carroll, Overland and Weil, 1994). The buffer stock model of savings (Deaton, 1991; Carroll, 1992) also yields a similar relation between savings and growth. 25

26 Thus the theoretical literature is unclear about both the direction of causality between the savings rate and income or growth and about whether the association between savings and growth should be positive or negative. Empirical work on the issue of causality between the savings rate and income or growth remains very sparse. This is probably due to the scarcity of sufficiently long time series data for most developing countries. Only a few studies (Carroll and Weil, 1994; World bank, 1993, Appendix 5.1) 5 have tried to analyze the direction of causality between savings and growth. Both studies have concluded that the direction of Granger causality is primarily from growth to savings. Carroll and Weil (1994) conduct Granger causality tests (in levels and first differences) on the 5 year averages for the savings and growth rates over their pooled time series cross section data sample of 64 countries 6. However, their findings are subject to a number of limitations pertaining to the cross section approach utilized and to the causal interpretation offered. These include the following: (a) the assumption of homogeneous relation between savings and growth across countries, which has to be made in pooled time series analysis, is unlikely to be valid. (b) The cross section nature of their technique cannot allow different countries to exhibit different patterns of causality, yet it is likely that in some countries savings leads growth and in others growth leads savings. (c) The study ignores all variables, except growth, that affect savings. This omission of other relevant variables can vitiate the reliability of Granger causality tests. The World Bank (1993) study does use time series Granger causality testing for several individual countries 5. Also see related work, though with a different approach, of King and Levine (1994) and Young (1995). King and Levine argue against capital fundamentalism - the notion that capital accumulation is the key to economic development while Young argues that the rapid economic growth of East Asian countries has been largely due to capital (and labour) increases. 6. Carroll and Weil (1993) also present some graphical analysis for several East Asian countries by plotting the savings and growth rates over time and using visual analysis to determine causality. However, this is not a satisfactory way to infer causality. 26

27 (mostly from East Asia), but suffers from problem (c). Here we present causality analysis for the five South Asian countries that avoids the above problems. 4A. Determining the Direction of Causality: Preliminary testing for the main variables of interest for our analysis showed that essentially all variables of interest for the South Asian countries are stationary (no unit roots). Thus, the direction of Granger causality between the savings rate and the growth rate can be investigated using the vector auto-regression (VAR) procedure (see for example, Granger, 1969 and Judge et al., 1987, chapter 18). The details of the econometric methods used in this paper are explained below. The vector auto-regression (VAR) procedure): Following Granger (1969), an economic time series Y t is said to be "Granger-caused" by another series X t if the information in the past and present values of X t helps to improve the forecasts of the Y t variable, i.e. if, MSE(Y t Ωt) < MSE(Y t Ωt ) where MSE is the conditional mean square error of the forecast of Y t, Ωt denotes the set of all (relevant) information up to time t, whilst Ωt excludes the information in the past and present X t. The conventional Granger causality test involves specifying a bivariate pth order VAR as follows: p p t = + a iyt i + b jx t j i= 1 j= 1 Y ì + U (11) t 27

28 p 1 p 1 t = ' + CiYt i + d tx t j i= 1 j= 1 X ì + U (12) ' t where µ and µ are constant drifts, U t and U t are error terms, and more generally, the equation may include any number of additional relevant variables. Then, the null hypothesis that X t does not Granger cause Y t amounts to testing, b 1 = b 2 =... = b n = 0. (13) This can be tested by standard methods, such as an F-test. Similarly, the null hypothesis that Y t does not Granger cause X t amounts to testing, c 1 = c 2 =... = c n = 0. (14) 4B. Empirical Results: Direction of Causality between Savings Rate and Growth The results of Granger Causality testing for the five South Asian countries are shown in Table 3. The lag length for the VAR was chosen to minimize the Akaike information criterion (AIC) and the Schwartz Bayesian Criterion (SBC). Nepal and Pakistan, both criterion, both criterion selected the same lag length of 1, while in the remaining cases, the two criterion gave two different lag lengths and we report the results using both the lag lengths. It is seen that in the case of Pakistan and Bangladesh (using the Akaike criterion), the direction of Granger causality is from Savings rate to growth of real GNP (since the null hypothesis that growth does not cause savings rate is not rejected while null hypothesis that savings rate does not cause growth is rejected). In the case of Nepal, the data fails to reject non-causality in either direction. Finally, in the case of India and Sri Lanka (using Akaike Criterion only), the analysis implies causality from growth to savings rate. The same result is also obtained for panel data for all countries 28

29 (probably because we had longer data series for India and Sri Lanka than for other countries, so that their results might have tended to dominate the panel results. [Insert Table 3] To summarize, we found evidence that higher savings rates cause higher growth rates of real GNP in two countries (Bangladesh ad Pakistan) and that higher growth rates Granger cause higher savings rates in two countries (India and Sri Lanka), with the results for Nepal failing to reject non-causality in either direction. Our results for South Asian countries can be interpreted to mean inconclusive results and the possibility of two-way causation overall. Nevertheless, they contrast sharply with the previous empirical findings (mostly based on data from East Asian countries, which have among the highest savings rates in the world) that savings rates do not cause growth but are determined by it (Carroll and Weil, 1994; World bank, 1993, Appendix 5.1). An even more interesting conclusion suggests itself on noting that the countries showing causality from savings to growth have relatively lower savings rates while countries showing causality from growth to savings (including the East Asian countries, as per previous empirical studies) have relatively higher savings rates (Figure 1). Thus, a plausible conclusion from our analysis is that in the case of countries with low savings rates, low levels of savings may well become a significant constraint on growth by restricting the supply of funds available for capital formation and growth. Such a constraint may be absent in countries with relatively high levels of savings. If this hypothesis is correct, than it adds significantly to our understanding of the causality issues. Furthermore, the low savings rate countries, such as Bangladesh and Pakistan can not afford to be complacent, and need to be much more concerned about raising 29