Savings transition in Asia: Unity in diversity

This paper examines the national savings behaviour in the process of economic growth through a comparative analysis of countries in developing Asia from a historical perspective. Developing Asia provides an ideal laboratory for the study with considerable differences in the savings behaviour among countries and over time within individual countries, notwithstanding the ‘model saver’ image based on the average savings rate. The empirical analysis distinguishes between private and government savings rates, with speci�c emphasis on the former. The �ndings are consistent with the view of ‘virtuous circle’ between growth and savings, with growth initiating the savings transition. No evidence to suggest that a prior phase of promoting savings through speci�c policy initiatives is needed to initiate the process of growth and structural transformation. The private savings rate is associated positively with per capita gross domestic product, export orientation, and foreign resource in�ows and negatively with the young dependency ratio of the population and domestic credit availability.


Introduction
The literature on national savings in the process of economic development and structural change has evolved around two separate but interrelated issues: what is the relationship between the savings rate and growth, and why do savings rates differ across countries and over time in a given country?The debate on the rst issue has been virtually settled, even though there has been some controversy about why a given savings rate is associated with different growth rates and in what way the causality runs.In the formative stage of development thinking during the early post-war years, the Harrod-Domar model, which held sway as the workhorse of development policy, linked growth directly and almost exclusively to the savings rate (Meier 1984).Given the perceived structural constraints on domestic savings mobilization in developing countries, foreign savings (foreign capital in ows) was considered a key prerequisite for economic take-off.The supremacy of savings (and hence investment) in the growth process was, however, questioned by the neoclassical growth model (Solow 1956) that received increased attention in the policy debate from the late 1960s.It postulated that an increase in savings rates generates higher growth only in the transition between steady states, and long-term growth depends solely on technological progress.From about the late 1980s, the new endogenous growth models have, however, provided theoretical support for the view that investment results in a permanent increase in growth rates.New multi-country empirical growth studies spawned by these theoretical advances have supported the notion that the rate of investment is the single-most important determinant of intercountry differences in growth rates (Levine and Renelt 1992; Sala-i-Martin 1997; Bond et al. 2010).
In contrast to the emerging consensus on the investment-(savings) growth nexus, the issue of why some countries save more than others remains an unresolved issue.What is the process by which a community that was previously saving a low percentage of national income dramatically increases its savings?Do countries need to start with speci c savings proportion policies to initiate the growth process or by harnessing foreign capital in ows?Alternatively, would initiation of the growth process through economy-wide marketoriented reforms generate a 'virtuous cycle' of growth-induced savings, resulting in a further increase in savings to generate even higher savings and growth?How do demographic dynamics in uence the savings trajectory in the process of growth and structural change?
The purpose of this paper is to contribute to this debate through a comparative analysis of savings behaviour in countries in developing Asia from a historical perspective.The focus on Asia is motivated by two reasons.First, the experiences of these countries as 'model savers' gure prominently in the contemporary policy debate on the role of domestic savings in economic development and how to bridge the domestic investment-savings gap that constrains the growth process in most developing countries.[1] Second, notwithstanding their 'model saver' image, which is based mainly on the experience of high-performing East Asian economies, there are considerable differences in the savings behaviour among countries and over time within individual countries in the region.Therefore, the region provides an ideal laboratory to study the determinants of savings in the process of economic growth and structural transformation.
The paper aims to add to the existing knowledge of the savings behaviour of countries in developing Asia in several ways.First, for the rst time in the study of comparative savings behaviour in the region, [2] the analysis distinguishes between private and the government savings rates, with speci c emphasis on the former.The speci c focus on private savings is important from the policy point of view because public savings is mostly driven by unobservable political factors.Second, export orientation in the development process is explicitly included in the savings function as a conditioning variable in examining the relationship between the savings rate and per capita income growth.Third, bene ting from recent improvement in the national data reporting systems, we use an annual balance panel data set for the period 1980-2019 with a wider regional coverage, encompassing countries in Northeast Asia, Southeast Asia, and South Asia.Finally, we use improved econometric techniques.The panel data Auto-Regressive Distribution Lag (ARDL) methodology used in estimating the savings function may offer a solution to the problem of bias caused by unobserved heterogeneity among countries, a common problem in estimation with cross-sectional data, while minimizing endogeneity bias in the savings-growth nexus.
The paper is structured in four sections.Section 2 presents a comparative analytical narrative of savings performance in Asia in the global context, with emphasis on policy regime shifts.Section 3 undertakes an econometric analysis of the determinants of the savings rate.The key ndings are summarised in the concluding section.
2 Savings Behaviour In Developing Asia: A Historical Perspective

Overall patterns
The savings rates in Asian countries were not unusually large in the early post-war years.Rosenstein-Rodan (1961), in a pioneering study undertaken to inform the policy debate on international development aid, estimated the average gross savings rate of Asian countries at 7.0 per cent compared to 9.4 per cent in Latin America and only one percentage higher than that in Africa (5.9 per cent).Interestingly, at the individual country level, Burma (Myanmar) and India had a higher similar savings rate of 8.5 per cent compared to Taiwan (8.0 per cent) and South Korea (6.5 per cent) (Rosenstein-Rodan 1961, Table 3-A).
The patterns began to change from about the late 1960s.By the early 1970s, the average Asian savings rate exceeded that of Latin America and was more than double the average rate recorded in Sub-Saharan Africa.During the ensuing years, the Asian rate and those of the other major regions and the overall world savings rate has widened.Overall, the Asian savings rates have also been much more stable (Figure 1 and Table 1).When the observed patterns of savings rates are placed in the context of the trajectory of policy reforms in these countries, [3] we can see a clear relationship between the timing and nature of market-oriented policy reforms and savings transitions.Korea, Taiwan, Hong Kong, and Singapore [4] were the earliest reformers in the region.The main Southeast Asian countries followed suit about a decade later.The dramatic savings transition in China began following the country's gradual shift from 'plan to market' in the late 1970s.India, which had lost its early lead in savings ranking in the devaluing world for over four decades, has begun to catch up following the liberalization reforms initiated in the early 1990s.Sri Lanka recorded a signi cant increase in the savings rate following the liberalization reforms in the later 1970s, but the trend has begun to reverse in recent years, underpinned by a notable reversal of reforms.Within Southeast Asia, a comparison of the savings rates for the past three decades with those during 1965-79 points to the impact of policy regime shifts on savings.In Indonesia, the domestic savings rate has recorded a notable increase following reforms that began earlier in that decade.Vietnam began to replicate the early experiences of Korea and Taiwan as the reform process gathered momentum in the early 2000s.
At the formative stage of the emergence of development economics as a separate discipline, Sir Arthur Lewis (1954) made the following highly cited observation on the role of the savings transition in the process of economic development: 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 per cent of its national income or less, converts itself into an economy where voluntary saving is running at about 12 to 15 per cent of national income or more.This is the central problem because the central fact of economic development is rapid capital accumulation (including knowledge and skills with 'capital').We cannot explain any 'industrial' revolution (as the economic historians pretend to do) until we can explain why saving increased relatively to national income (Lewis 1954, p. 155).
When we take the data reported in this section at face value and assume a capital consumption allowance of 10 per cent (following Srinivasan 1994), the countries in Northeast Asia had already passed the Lewisian threshold by the early 1980s, all major Southeast Asian countries other than Indonesia and the Philippines by the early 1990s, and India in the early 2000s.

Private and government savings
The data on gross national savings disaggregated by private and public (government) savings are summarized in Table 2  [4] From about the early 1980s, Singapore has recorded by far the highest savings rate in the region (and perhaps in the world).This 'exceptional saver' status of Singapore is partly the result of a unique government policy that required all workers to make very large annual contributions to a pension fund (which can be used ahead of retirement for a variety of purposes other than current consumption).

Savings Rate Determination: Empirical Analysis
We proceed in this section to undertake an econometric analysis of the determinants of savings behaviour using a panel data set for 12 Asian countries 5 with data that are available for the entire period 1981-2019.We focus speci cally on private savings because preliminary analysis suggested that public savings is mostly driven by unobservable political factors.Data are not available for most of the countries under study for further disaggregating private savings into household savings and corporate (business) savings in the empirical analysis.Apart from this data constraint, the focus on aggregate private savings is justi ed by the 'consideration that corporate savings, just like personal (or household) saving, will tend to result, at least in the long run, in an increase in private net worth by way of its net effect on the market value of corporate equity' (Modigliani 1966, pp 184-5).Provided the shareholders look through the corporate veil to corporate earnings and take into account corporate retained earnings (pro ts that are not distributed as dividends) in their lifetime savings/consumption decisions, the appropriate dependent variable for savings analysis is aggregate private savings; no separate treatment of household and corporate savings is needed to understand the saving behaviour of the economy.This view is no doubt an approximation (Gersovitz 1988;Poterba 1991).Shareholders may be myopic and fail to devote the necessary resources to monitor corporate performance.Moreover, various factors encountered by companies such as liquidity constraints, tax policies, and other kinds of capital market imperfections could hamper the shareholders' ability to pierce the corporate veil and thus limit the extent to which personal savings behaviour counterbalances corporate savings (retained earnings) or vice versa. 6

The model
There is no single model that is capable of dealing with every dimension of savings behaviour.Our modelling strategy is to use the life cycle model (LCM) as the foundation and draw on the subsequent development in the related literature to formulate the nal empirical model used in this paper.The attractiveness of the LCM for our analysis lies in both its elegant formulation of the impact of income growth and demographic dynamics, which are central to understanding the savings transition in the process of growth and structural change.The LMC also has the exibility for incorporating other relevant theoretical considerations to form an integrated analytical framework, without changing the basic structure of the model.
The LCM is speci cally designed to explain the savings behaviour of a representative individual (the agent) based on the simplifying assumptions of perfect capital markets and perfect foresight of the individual about the income generation process during his lifetime.
The agent chooses to maximize utility derived from lifetime resources by allocating them between current and future consumption: as income tends to uctuate systematically over the course of the agent's life, savings behaviour is determined by his stage in the life cycle (Modigliani 1966(Modigliani , 1986)). 7When the model is extended to the national level, the rate of growth of per capita income and the rate of population growth are the key determinants of the savings rate.To the extent that the economy is growing, workers are saving on a larger scale compared to retirees, resulting in an increase in the measured aggregate savings.At the same time, an increase in the population growth rate increases the working-age population (savers) relative to the number of retirees (dissavers).Thus, even if all the individuals in two given economies have the same savings pro le over their life cycles, the aggregate savings rate can be different depending on the rate of population growth.
The other variables suggested by the LCM as relevant for allocating lifetime resources between current and future consumption (and hence savings behaviour) include the real interest rate, social security payments (and other government transfers to households by the government), and wealth.These three variables have the potential to impact the savings rate by conditioning the impact of economic growth and population dynamics on the households' ability to make an intertemporal transfer of resource.
The real interest rate has two countervailing effects on savings depending on whether the person is a net borrower or a net lender.In the former case, a higher interest rate increases the present price of consumption relative to the future price and thus provides an incentive to increase savings (the substitution effect).By contrast, in the latter case, an increase in the interest rate raises lifetime income and thus tends to increase consumption and decrease savings (the income effect).Social security payments could have a negative impact on personal savings as individuals substitute these expected government transfers for personal savings otherwise accumulated for retirement (Feldstein 1974(Feldstein , 1996;;Modigliani and Cao 2004).As in the case of social security payments, wealth can have a negative effect on private savings because the ability to draw on accumulated wealth to maintain consumption levels diminishes the need to save for retirement (Deaton 1992).
The empirical implementation of the core model described above for analysing the savings behaviour in developing countries raises a number of issues.First, the LCM postulates that the savings rate is related to the growth of per capita income, not the current level of per capita income as postulated by the standard Keynesian theory of consultation (absolute income hypothesis).This postulate stems from the assumption that individuals are forward-looking and, therefore, base their savings decisions on lifetime income rather than current income.The relevance of LCM for analysing the savings behaviour of a country depends on the existence of a signi cantly large core of households that are able to carry over resources to provide for old age at a standard of living commensurate with that of preretirement.In fact, the LCM was formulated for developed market economies for which this assumption holds fairly well.However, in developing countries the portion of the population in the bottom rungs of the income distribution may nd it impossible or too burdensome to set aside resources now in order to provide for later consumption.For these reasons, Modigliani (1993, p 276) has admitted that 'conceivably for a su ciently low value of per capita income, … the saving-income ratio for given growth would … tend to rise with income'.We, therefore, include both the growth and level of household disposable income (GY and YD, respectively) as explanatory variables.
Second, there is a sizeable body of empirical evidence that the degree of export orientation of the development strategy plays an important role in explaining intercountry differences in growth and the savings rate (Maizels 1971;Weisskopf 1972;Papanek 1973;Chow and Papanek 1981;Michaeley 1977;Balassa 1989).Export orientation leads to better growth performance than policies favouring import substitution by facilitating resource allocation according to comparative advantage, allowing for greater capacity utilization, permitting exploitation of scale and greater technological improvement in response to competition from abroad, and contributing to increasing employment.To the extent that the propensity to save is associated with marginal rates of growth exceeds that associated with the average rates as postulated by the LCH, the rates of savings would be higher under export orientation.Moreover, as already discussed, a labour surplus economy growth through greater export orientation in a labour surplus economy has the potential to tilt income distribution in valour of the capitalist (entrepreneurs) whose propensity to save might be higher.Foreign direct investment, attracted by high returns in export-oriented production, would also add to domestic savings.We therefore include export orientation (EOR) on its own as well as interactive integration with economic growth (EOR*GY) as explanatory variables in the model.The EOR variable would capture the direct effect of export orientation on the savings rate, while EOR*GY is expected to capture the conditional (savings-enhancing) effect of EOR on the savings impact of a given growth rate (GY).The coe cients of both variables are expected to be positive.
Third, the original formulation of the LCM postulates that an increase in the population growth rate increases the aggregate savings rate by increasing the number of active workers (savers) relative to the number of retirees (dissavers).However, in reality, an increase in the population growth is naturally associated with not only an increase in the labour force but also a change in the relative shares of the young and elderly dependents in the population.Moreover, a given change in the degree of 'childhood dependency' may not have the same impact on savings compared to a similar change in 'aged dependency' (Masson 1988; Deaton and Paxon 1999; Kelley and Schmidt 1996; Curtis et al. 2017).Economies of scale in family consumption enable large families to provide a child with the same welfare with a less-than-proportionate increase in expenditure compared to a small family.Furthermore, when they decide to have large families, parents may choose to decrease their consumption or increase savings in advance of a birth by increasing their work time (or effort).In the cultural tradition of Asian countries, the young generation is supposed to take care of the older members of the family, while the elders will bequeath the house and other assets to their children.Under such a system, a child is an effective substitute for life cycle savings, and investing in children's education is considered a source of old age support.Households with young children are therefore likely to save to nance their education (Ge et al. 2018;Curtis et al. 2017).These considerations make a strong case for the inclusion of the aged dependency ratio (ADEP) and young dependency ratio (YDEP) as separate variables in place of population growth as explanatory variables in the savings ction in order to capture the impact of population dynamics on savings.Since the aged and young dependents generally consume without generating income, normally the coe cients of both variables are negative.However, the magnitudes of the two coe cients can be different because of the differential impact of the socio-economic factors discussed here.
Fourth, the hypothesized link between income growth and the savings rate is based on the stringent assumption of perfect capital markets that enable households to borrow freely against future income in order to smooth consumption over their lifetime.If the households are liquidity constrained-they are unable to borrow freely against future income-the consumption behaviour might be linked to current income rather than to lifetime income (Liu and Woo 1994).Thus, the borrowing constraint, in addition to forcing households to maintain consumption at current income levels, can in fact convert a negative saver into a positive saver by forcing them to save more at present in order to undertake lumpy (indivisible) expenditure plans in the future (Gersovitz 1988).We therefore include in the savings equation a control variable to represent the availability of institutional credit (CPR).
Fifth, the accumulation-based explanation of savings behaviour of the LCM is based on the implicit assumption of certainty of future income streams in the mind of the individual.This assumption presumably holds reasonably well for households in developed countries who save in large part for future consumption (accumulate wealth).However, income prospects are much more uncertain for most households in developing countries.Saving is, therefore, not only about accumulation for future consumption but also about consumption smoothing in the face of volatile incomes.In other words, a precautionary motive rooted in economic uncertainty can be an important driver of savings behaviour (consideration behind saving).Guided by previous studies (Corbo and Schmidt-Hebbel 1991; Deaton 1989; Loayza et al. 2000), we include the rate of in ation (INF) to capture precautionary savings effects of macroeconomic uncertainty.In ation can have a positive effect on saving, as uncertainty about future real incomes in an in ationary environment may encourage saving for maintaining future consumption levels.However, it can also have negative effects on saving by increasing the uncertainty about future value of accumulated savings.The direction of the impact of INF is, therefore, indeterminate a priori.
Sixth, the impact of foreign resource in ows ('foreign savings') on domestic savings remains a debatable issue (Papanek 1972(Papanek , 1973;;Weisskopf 1972; Reinhart and Talvi 1998; Obstfeld 1999; Adams and Klobodu 2018).Foreign savings can act as a substitute for domestic savings if the agents draw on foreign savings to smooth current expenditure over time.However, there is room for developedoriented governments to harness foreign resources to meet the gap between domestic investment and domestic savings without an adverse effect on domestic savings.Also, foreign resource in ows are not a homogenous phenomenon.Foreign direct investment, which directly contributes to the domestic production capacity of the economy unlike concessionary foreign aid and other forms of capital in ows, has the potential to help promote domestic savings.
Finally, the scal policy stance of the country can affect private savings behaviour.There are two possible channels with opposing effects.First, the Ricardian equivalence proposition (a la Barro 1974) postulates that issuing bonds to nance government dis-saving (budget de cit) results in an equal increase in private savings because the private sector saves in anticipation of a future increase in taxes to service the bonds.While the government can choose the level of its own savings directly, a change in this variable need not imply a one-for-one change in national savings if the private agents respond in such a way as to offset the government action, at least to some extent.Second, government savings behaviour can be indicative of the soundness of macroeconomic management, including a lower rate of in ation, prudential exchange rate policies, and capable monetary management.Stable economies, in turn, lower the risk for investors and therefore lower the cost of capital for long-term investment and encourage savings (and investment) by the private sector.Moreover, when governments shift funds from consumption to particular types of investments, such as infrastructure that the private sector is unlikely to undertake, the return to, and the volume of, private savings may increase.To test the net effects on the private savings rate of these countervailing effects of the scal policy stance, we include a government budget balance (BBL) measured as a percentage of gross national income as an explanatory variable in the model.
Based on the above discussion, the empirical savings function in a panel data setting is speci ed as follows:

Data source and variable measurement
The estimation of the savings function (Eq. 1) is undertaken using an annual unbalanced panel data set of 12 Asian countries, covering the period 1981-2019. 8Data on the savings rate for all countries other than Taiwan are compiled from the Key Indicators of Asia and the Paci c (KIAP) database of the Asian Development Bank.The KIAP database is a direct compilation of data from the o cial records of individual ADB member countries.In the national accounts of these countries, the data on national savings are estimated indirectly, subtracting net resource in ows ('foreign savings') (after allowing for changes in the holding of foreign exchange reserves) from aggregated domestic investment.Data on national savings disaggregated into private savings and public savings are available from the national data systems only for India and South Korea. 9For the other countries, we derived private savings by deducting government savings from total national savings.Government savings is derived as the difference between government revenue and government recurrent expenditure using data from the same data source.Any data series that is derived as a 'residual' from two other national account aggregates naturally incorporates possible estimation errors of the latter two magnitudes.Therefore, the use of the savings date used in the econometric analysis is based on the assumption that, in the data series for each country, the estimation errors remain consistent over the period under study. 10  The data on the other variables are collected or compiled from several sources.The deposit interest rate of India is compiled from the reserve bank of India.The data on deposit interest rates of other countries 11

Econometric procedure
We began the estimation process by examining the time series properties of the panel data using the CIPS test (Pesaran 2007). 14The results reported in Table 4 indicate that the saving series, SR, and all other explanatory variables except GY are non-stationary (I(1)).
Based on this result, we conducted test to examine the existence of a long-run relationship among the variables (Pedroni 1999(Pedroni , 2004).
The results indicate that two of the four test statistics relating to cointegration of the 'within dimension' of the data panel and all three relating to cointegration of the 'between dimension' of the data panel are statistically signi cant (Table 5).These results provide su cient grounds to use the panel-data ARDL estimator to estimate the savings function.Cointegration along the 'within dimension' of the data panel.( 4) Cointegration along the 'between dimension' of the data panel.
Equation ( 1) can be rewritten in ARDL form as follows: 2 where is the saving rate; is a vector of explanatory variables; and is scalars or the coe cient of the lagged dependent variable.
When reparametrized in error correction form, Eq. ( 2) takes the following form: 3 where is the speed of adjustment coe cient (expected that ), representing the speed of adjustment of imports to a shock to move back to the long-run equilibrium; is the vector of long-run coe cients; is the error correction term; and and are the short-run coe cients.
The error correction formulation of the ARDL model (Eq.3) permits us to examine short-and long-run dynamics and the speed of adjustment of the model to equilibrium.It is 'robust to integration and cointegration properties of the regressors, and for su cient lagorders, could be immune to the endogeneity problem, at least as far as the long-run properties of model are concerned' (Pesaran 2015, p 726).When, observations for each individual countries in the data panel are available for a long enough period are available to systematically test su cient lad order (as in our case) the possible endogeneity bias could be asymptotically negligible due to the super consistency property resulting from the parametrization of the model in levels and divergences.
The Akaike information criterion (AIC) is used to decide the choice of lags for each country group per variable, and then the most common lag for each variable is chosen to represent the lags for the model.We use two alternative estimators to explore the potential heterogeneity of parameters among the countries within the data panel.We utilize two alternative estimators: the Dynamic Fixed Effects (DFE) estimator and the Mean Group (MG) estimator (Pesaran 2015).The DFE estimator allows the intercepts to differ freely across groups, while all other coe cients and error variances are constrained to be the same.Although this estimator could be biased when applied to dynamic models, the size of the bias tends to zero as the time dimension increases (Nickell 1981).The MG estimator allows coe cients to differ freely across groups, by rst estimating one equation per group (a country in our case) and taking the average across groups (countries).The most appropriate estimator is selected using the test proposed by Hausman (1978).

Results
The savings function (Eq. 1) is estimated separately for the 12 countries and the countries other than China to see the possible sensitivity of the results of China's dominance in the overall savings performance in Asia.The results are reported in Table 6.
Per capita real private income (YD) was dropped by the ARDL estimator because of its high collinearity with the growth rate of GY.In alternative estimates that excluded GY, the coe cient of YD was not statistically signi cant even though it had the expected positive sign.15 Dropping YD for the nal estimates was supported by the standard variable deletion F test.The social security payment (SSP) is dropped from the reported equations because data were not available for three countries (China, Indonesia, and India) and data for some years are missing for other countries.In the equation estimated for the other nine countries, the coe cient had the expected negative sign but was not statically signi cant, and its inclusion had no notable impact on the estimated coe cients of the other variables.
In both equations, the coe cient of the ECM term is highly statistically signi cant with the expected negative sign, suggesting the appropriateness of the ARDL speci cation of the model.In the equation for all countries, both the short-run and steady-state (long-run) coe cients of GY are statistically signi cant at the one per cent level.The results suggest that a one per cent increase in the growth rate of per capita private income is associated with a 0.25 parentage point increase in the private savings rate in the short run and a 1.27 percentage point increase in the long run.The results are remarkably resilient to the inclusion or exclusion of China from the country coverage.
Relating to the interpretation of this result, an important issue is the possible endogeneity of GY in the model (Deaton 1989;Gersovitz 1988).However, as noted, the panel ARDL estimator has the advantage of minimizing possible endogeneity of the right-hand variables by reparametrizing the model in levels and differences.To supplement the results, we performed the Granger causality test for the relationship between GY and PSR using the methodology of Dumitrescu and Hurlin (2012).The results suggest that the causation runs for growth to the savings rate-not the other way around (Table 7).Thus, our results are consistent with the 'virtuous circle' growthsavings nexus in which the initial spurt of savings comes from growth.It is important to note that this inference is also consistent with the discussion in Section 2 on the relationship between the timing of market-oriented policy reforms and the time pro le of savings patterns, both among and within the Asian countries.There is strong evidence that export orientation (EOR) is signi cantly associated with the intercountry difference in the savings rate.A one percentage point increase in the degree of export orientation is associated with a 0.03 per cent increase in the savings rate in the long run.Moreover, the coe cient of EOR*GY indicates that export orientation adds 0.09 percentage points to the association between the per capita income growth rate and the savings rate.
Relating to the implications of demographic dynamics on the savings rate, the coe cient on (YDEP) is statistically signi cant with the expected negative sign, suggesting that a one percentage point increase in the share of young dependents in the population contributes to a 0.17 percentage point decline in the savings rate.However, interestingly, there is no statistically signi cant evidence to support the standard LCH that aged dependency (ADEP) contributes to dampening savings propensity.This result is not consistent with the available evidence for development countries (Leff 1969;Modigliani 1966;Horioka 1996;Bloom et al. 2007).There are a number of possible reasons why ageing of the population in developing countries may not necessarily lead to lower savings rates.First, given the prevalence of informal sector employment and the limited coverage of retirement bene t schemes even in formal employment, the line of demarcation between the age and formal retirement remains blurred in the Asian context.Second, the rise in life expectancy as an integral facet of economic growth could have a signi cant effect on savings behaviour in old age.This effect has the potential to be especially pronounced in developing Asia, particularly in the East Asian high-performing countries because mortality transition has been very rapid (Kinugasa and Masson 2007).Third, households in developing countries generally tend to be larger than in advanced countries, and resources are shared between members actively engaged in the labour force dependents.Finally, relating to the third point, bequeath motive can be a much more potent determinant of savings behaviour in developing countries (Deaton 1989;Gersovitz 1988).
There is strong evidence that foreign capital in ows (FS) are complementary to private savings, in contrast to the ndings of some previous studies that foreign capital in ows tend to crowd out domestic savings.This nding is consistent with the evidence that foreign capital in ows to Asian countries have mostly taken the form of direct foreign investment (which directly contributes to the production capacity of the countries) rather than foreign aid (which mostly takes the form of budgetary supports) (Athukorala and Rajapatirana 2003).Our result is also consistent with the inference of Reinhart and Talvi (1998) that, unlike in Latin America, capital in ows were complementary to domestic savings in Asia because the development strategy in these countries speci cally focussed on using these resource in ows for investment.
The bank credit variable (CRP) has a signi cant negative effect on private savings, as expected.This result is consistent with the hypothesis that, in the presence of easy access to bank credit, there is no compelling reason for people to save more at present in order to undertake lumpy (indivisible) expenditure plans in the future (Gersovitz 1988).The coe cient of the proxy variable for wealth (WL) has the expected positive negative sign but is not statistically different from zero.
The coe cient of the real interest rate variable (RID) is not statistically signi cant, and its magnitude is barely different from zero.It seems that the income effect of the real interest rate counterbalances its substitution effect in the Asian context.The coe cient of the budget balance to GNI ratio (BBL) does not support for the Ricardian equivalence hypothesis.On the contrary, the result provides some weak support for the view that scal discipline helps promotion savings.
We estimated the savings equation (Eq. 1) for the total national savings rate (NSR) for comparison.The results are reported in Table 8.
Both the long-run and short-run coe cients of GY are highly signi cant as in the private savings equations, but their magnitudes are slightly smaller.The coe cients of the other variables except EOR&GY are broadly similar but only in terms of the coe cient signs and statistical signi cance.The negative and statistically signi cant coe cient of EOR*GY perhaps captures the scal costs (tax incentives and other expenditures) involved, which counterbalance the direct positive effect of export orientation on national savings under the export-oriented development strategy.The magnitude of the coe cient of YDEP in the national savings equation (-0.089) amounts to almost half of that of the private savings equation (-0.169).[12] The results are remarkably resilient to the use of the GDP de ator as an alternative price de ator.[13] In some savings studies of Japan and Taiwan, the age 20 (instead of 14) has been used as a more appropriate lower cut-off age for separating young and adult populations because of the heavy emphasis placed in these countries on upper-secondary education (e.g., Horioka 1991Horioka , 1996;;Athukorala and Tsai 2004).To see the sensitivity of results to the particular de nition adopted, we re-estimated the savings function using alternative measures of YDEP and ADEP using age 20 as the lower bound.The results (available from the corresponding author on request) are closely comparable, but the results reported here are statistically superior in terms of the overall t of the savings function, its other statistical properties, and the signi cance of the coe cients of ADEP and YDEP.
[14] The CIPS test is a modi ed version of the t-bar test proposed by Im et al. (2003)-hence the acronym 'CIPS'.It is now considered more powerful than the test proposed by Maddala and Wu (1999) as it relaxes the assumption of independence among cross-sections and allows for the possible correlation between cross-sections.
[15] The alternative estimates are available on request.
4 Concluding Remarks savings rates in Asia were broadly comparable to those in the rest of the developing world in the early post-war years.The patterns began to change from around the late 1960s.During the ensuing year, the difference between the average Asian savings rate and those of the other major regions and the overall world savings rate has widened.By the late 2010s, the average Asian saving rate was abbot 37 per cent compared to the global average of about 28 per cent.
The regional average hides substantial sub-regional and individual country differences in savings behaviour in Asia.Countries in Northeast Asia top the savings rate ranking followed by Southeast Asia.Savings rates in countries in South Asia, though much lower compared to Southeast Asia, are higher compared to the other regions.Within Northeast Asia, the high savings rates of Taiwan and South Korea began to decline around the late 1990s, but the spectacular increase in savings in China has more than counterbalanced this decline.China now accounts for over two-thirds of total national savings (in value) in the region.Within Southeast Asia, the savings rate of Singapore has continued to increase in contrast to the recent decline in the savings rates of the other three Asian 'tigers' (South Korea, Taiwan, and Hong Kong).In Southeast Asia, a comparison of the savings rates for the past three decades with those during 1965-79 points to the impact policy regime shifts on savings.Notwithstanding these differences, a unifying theme of overall savings behaviour in Asia is that policy regime shifts in favour of an outward-oriented development strategy have underpinned the time patterns of savings behaviour.
The trends and patterns of saving in Asia have been predominantly driven by the private sector: the governments directly accounted for only about 15 per cent of total national savings in the region.Overall, the countries' relative performance in terms of their evolving investment-savings gap seems to mirror differences relating to the timing and depth of the market-oriented reforms noted above.
The ndings of the econometric analysis are consistent with the idea of the 'virtuous circle' between savings and growth, with growth initiating the savings transition.There is no evidence from the Asian experience to suggest that the prior phase of promoting savings through a speci c policy initiative to promote domestic savings or lling the investment-savings gap is needed to initiate the process of growth and structural transformation.There is strong evidence that export orientation contributes to higher private savings both by its direct contribution and by compounding the impact of the rate of income growth on the savings rate.Foreign capital in ows are complementary to domestic savings.
As regards the nexus of demographic transition and domestic savings, only the change in the young dependency ratio seems to have signi cant impact on national savings behaviour.The pattern of the aged dependency ratio dampening national savings in developed countries is not revealed by the data in Asia presumably because, given the prevalence of informal sector employment and the limited coverage of retirement bene t schemes even in formal employment, the line of demarcation between working age and formal retirement remains blurred in these countries.
Finally, it is important to emphasize that the econometric evidence reported in the paper simply re ects the average macroeconomic pattern of the savings behaviour of the 12 countries covered in the analysis.Obviously there can be notable exceptions to the depicted average pattern.Also, the results are subject to the well-known limitations of savings data derived as a residual from the related macroeconomic variables (the 'tyranny of residual').Presumably, the magnitude of the measurement error varies among countries and even over time in given countries.Because of these reasons, the inferences made in this paper need to be treated only as a point of

Figure 1 World
Figure 1

Table 1 :
Gross domestic investment rate (%) Source: Complied from the World Bank, World Development Indicators database.
13r13Asian countries for which data are available for the past four decades.It is important to note that these data are not strictly comparable with those reported in Table1.The gross national savings rate additionally captures remittances by migrant workers.The savings rates reported here are, therefore, larger for countries that receive a signi cant in ow of migrant worker remittances (in particular in Sri Lanka it + 3 YDEP it + 4 ADEP it , + 5 EOR it + 6 EOR*GY it + 7 CRP it + 8 RID ti + 9 INF ti + 10 SSP it + 11 WL it + 12 FS + where PSR is a private savings rate; and are country-and time-speci c effects;.The explanatory variables are de ned below with the expected signs of given in brackets:GY (+) The rate of growth of per capita private income YD (+) Per capita real private income YDEP (-) Young dependency measured as the ratio of the population aged 15 and under to the working-age population (aged 16-64) ADEP (-) Aged dependency measured as the ratio of the population aged 65 and older to the working-age population (aged 16-64) BBL (?) Government budget balance as a percentage of gross national income and money stocks are retrieved from the International Monetary Fund.All other data series (except for Taiwan) are extracted from the World Bank World Development Indicator database.All data series for Taiwan are compiled from various issues of the Taiwan Statistical Data Book (Council for Economic Planning and Development, Taipei).Nominal private income is converted into real terms using the consumer price index (CPI = 2010).12Thereal national currency value of private income (YD) is converted into US dollars using the average annual exchange rate of national currency per USD exchange rate for 2010.The young-age dependency ratio (YDEP) is the percentage of the population aged 15 and under relative to the working population aged 15-64.Similarly, the old-age dependency ratio (ADEP) is constructed by dividing the population aged 65 and older by the working population aged 15-64.
13Private wealth is proxied by the money stock (broad money), including money in circulation plus checkable deposits in banks, savings deposits, money market mutual funds, and so on.Social security payments are measured by the government transfer payment, including subsidies, grants, and other social bene ts.The real interest rate (RID) is measured as ln[(1 + NID )/(1 + INF)], where NID is the average time deposit rate in commercial banks and INF is the current rate of in ation calculated from the CPI.Data series YD is used in natural logarithms.All other variables (except all dummy variables) are in percentage form.

Table 4
Note: the null hypothesis is 'the data series is I(1)'; the time trend is included in the test equations, and maximum selected lag length is 2; the Chi-square stats are reported in the table where ** denotes statistical signi cance at the 5 per cent level.

Table 5
Cointegration tests1) all test statistics are distributed N(0,1) under a null of no cointegration and diverge to negative in nity (save for panel v); data are time-demeaned, and a time trend has been included; Akaike information criterion (AIC) is used to decide the optimal lags, and ***, **, and * indicate being statistically signi cant at 0.01, 0.05, and 0.1 levels, respectively.(2)v is a non-parametric variance ratio statistic, ρ is is the Phillips and Perron rho-statistic, t is Phillips and Perron t-statistic, and ADF is the augmented Dickey-Fuller t-statistic.(3)

Table 6
Determinants of private savings (PSR)1 Heteroscedasticity corrected standard errors are in parentheses; ***, **, and * denote statistically signi cant at 0.01, 0.05, and 0.1 levels, respectively; ARDL indicates the appropriate lag length of each variable detrmined in the estimation process.

Table 8
Ang and Sen (2011))(2003)onalsavings (NSR)1) heteroscedasticity corrected standard errors are in parentheses; ***, **, and * denote statistically signi cant at the 0.01, 0.05, and 0.1 levels, respectively; ARDL(.)indicates the lag length of each variable used in the original ARDL equation from which the longterm coe cients are derived.The impact of exogenous political factors on government savings seems to understate the impact of changes in the age pro le of the population on national behaviour.In sum, this comparison alerts the risk of making inferences about the impact of demographic dynamics on savings behaviour using aggregate national savings data.Finally, how do our multi-country results for Asia compare with the ndings of the few available individual country studies of private savings in the region (Sun and Liang 1982; Athukorala and Tsai 2003; Athukorala and Sen 2004; Modigliani and Cao 2004; Park and Rhee 2005; Ang 2008; Ang and Sen 2011; Jongwanich 2010; Curtis et al. 2015; Ge et al. 2018)?There are vast differences among thesestudies in terms of the speci cation of the savings ction, re ecting the nature of data availability and presumably re ecting the researchers' own preferences dictated by methodological reasons.The only explanatory variable commonly used in all studies is the per capita income growth rate.The results for this variable support a positive association between per capita income growth and private savings rates, with the magnitude of the o ce varying in the range of 0.03 to 1.7 per cent.OnlyAthukorala and Tsai (2003)forTaiwan  and Jongwanich (2010)for Thailand have included young and aged dependency ratios separately for testing the impact of the demographic transition on the private savings rate.The results in both studies suggest that both young and aged dependency have a negative impact on the private savings rate and the impact of the former is greater in magnitude compared to that of the latter.However, household survey-based studies ofPark and Rhee (2005)for South Korea andCurtis et al. (2015)for China failed to detect a signi cant impact of population aging on the savings rate.These mixed results seem consistent with the failure of our savings function estimates to detect a negative relationship between population ageing and the savings rate.Consistent with our results,Ang (2008)for Malaysia,Ang and Sen (2011)for Malaysia and India, and Jongwanich (2010) for Thailand nd that the availability of access to bank credit is negatively associated with the savings rate.