You are on page 1of 23

Available online at www.sciencedirect.

com

Journal of Policy Modeling 32 (2010) 231–253

Does domestic saving cause economic growth?


A time-series evidence from India
Tarlok Singh ∗
Department of Accounting, Finance and Economics, Griffith Business School, Griffith University,
Nathan Campus, 170 Kessels Road, Brisbane, Queensland 4111, Australia
Received 1 November 2008; received in revised form 1 April 2009; accepted 1 August 2009
Available online 24 September 2009

Abstract
This study examines the long-run effects of domestic saving on income and tests the null of non-causality
between saving and growth in India. The optimal single-equation and the maximum-likelihood system
estimates of the model consistently support the predictions of the neoclassical exogenous and the post-
neoclassical endogenous models of economic growth, and suggest the significant long-run effects of saving
on income. The innovation accounting shows the bidirectional causality between saving and growth. The
stylized evidence for the steady-state effects of saving on income suggests the need to accelerate domestic
saving to finance capital accumulation and foster higher income and growth. Most of the saving comes from
the surplus household sector, and the deficit private corporate and public sectors draw on household saving
to meet their investment requirements and finance the resource gaps. A two-pronged approach with the
incentive-based measures to induce the motivation to save and the productivity-based measures to increase
income and strengthen the capacity to save, would be useful to generate higher saving and reinforce the
acceleration of income and growth.
© 2009 Society for Policy Modeling. Published by Elsevier Inc. All rights reserved.

JEL classification: C50; C51; O40; O41; O49

Keywords: Cointegration; FMOLS; DOLS; NLLS; Bounds test

1. Introduction

The theoretical and empirical literature examining the relationship between saving and eco-
nomic growth comprises three distinct strands of studies: one examining the sources of economic

∗ Tel.: +61 7 37357796; fax: +61 7 37353719.


E-mail address: Tarlok.Singh@griffith.edu.au.

0161-8938/$ – see front matter © 2009 Society for Policy Modeling. Published by Elsevier Inc. All rights reserved.
doi:10.1016/j.jpolmod.2009.08.008
232 T. Singh / Journal of Policy Modeling 32 (2010) 231–253

growth and predicting the positive effects of exogenous saving on income and growth, second
analysing the determinants of saving and suggesting the positive effects of exogenous income
and growth on saving, and third resolving the identification problem and testing the null of non-
causality between saving and growth. The Harrod–Domar model of growth (Domar, 1946; Harrod,
1939) predicts saving as the principal determinant of growth and suggests that the growth depends
on the marginal propensity to save and capital-output ratio. The neoclassical theory of economic
growth (Solow, 1956; Swan, 1956), however, postulates that saving has only the level and not the
growth effects in the long-run. The exogenously increased saving leads to the higher levels of
capital stock and output per capita in the steady-state, and to a higher rate of growth temporarily in
the transition to steady-state.1 The post-neoclassical endogenous theory of economic growth that
came into vogue since the mid-1980s predicts that the increase in saving generates a permanently
higher rate of growth through its positive effects on investment and capital accumulation (Barro
& Sala-i-Martin, 1995; Lucas, 1988; Mankiw, Romer, & Weil, 1992; Romer, 1986, 1987; Rebelo,
1991). The representative-agent Ramsey–Cass–Koopmans optimal model of growth (Cass, 1965;
Koopmans, 1965; Ramsey, 1928) suggests that the increase in saving leads to the increase in
income through the accumulation of capital. The rate of saving in this model is, however, not
exogenous and is determined endogenously by the optimizing households and firms interacting
on competitive markets subject to intertemporal budget constraints. The strand of studies exam-
ining the determinants of saving show the positive effects of income and growth on saving, and
point towards the endogeneity of saving (Attanasio, Picci, & Scorcu, 2000; Carroll, Overland, &
Weil, 2000; Carroll & Weil, 1994; Deaton & Paxson, 2000; Loayza, Schmidt-Hebbel, & Serven,
2000; Rodrik, 2000; Schmidt-Hebbel, Servén, & Solimano, 1996).
The issue of endogeneity and non-causality of saving remains unresolved in the time-series
models and least addressed in the cross-section and panel data models. The studies examining
the sources of economic growth commonly estimate the cross-section and panel data models, and
these models do not seem adequate to control for endogeneity and account for non-causality. The
panel data models combine normally a small time-series with a cross-section of countries and
obtain a large number of observations. Since it is the asymptotic time-span, rather than the number
of observations, that is crucial for determining the steady-state relationship, the evidence obtained
from these models estimated on a shorter time-span becomes weak and dubious. The time-series
models that draw on a longer time-span are instead more robust to examine the steady-state
relationship and test the null of non-causality. In the time-series strand, however, most studies, to
date, have drawn conclusions from the estimates of a single estimator and such reliance could lead
to biased assessment in terms of both statistical inference and magnitude of long-run parameters.
This study examines the long-run equilibrium and short-run dynamic effects of domestic saving
on income and tests the null of non-causality between saving and growth in India. The contribution
of the study that merits attention is that it uses both optimal single-equation and the maximum-
likelihood system estimators to estimate the model and examine the robustness of results to the
choice of estimator. These estimators resolve the problems of spurious regression and endogeneity
and provide efficient parameter estimates. The scheme of the study is as follows. Section 2
specifies the model. Section 3 presents the empirical results. Section 4 presents a discussion
on the saving and growth accelerations. Section 5 sums up the conclusions emerging from the
study.

1 For a critical survey of literature on the relationship between saving and economic growth in the neoclassical theory,

see Cesaratto (1999).


T. Singh / Journal of Policy Modeling 32 (2010) 231–253 233

2. The model

Consider the Cobb–Douglas technology of the profit-maximizing firms in a discrete-time rep-


resented by Y(t) = A(t)[K(t)∝ L(t)(1−∝) ]; where α ∈ [0,1], Y(t) is output, L(t) labour, K(t) capital
stock and A(t) the labour-augmenting technological progress. Under the assumption of constant
returns to scale (both capital and labour), but diminishing returns to a factor (capital or labour),
the technology predicts that the accumulation of capital does not have any long-run effects on
growth, {∂Y/∂K} =LimK→∞ {αY/K} = 0. By reverting the neo-classical assumption of diminish-
ing returns and instead assuming constant returns to capital with α = 1, the technology can be
transformed into AK model, Y = AK (Barro & Sala-i-Martin, 1995; Rebelo, 1991; Romer, 2006).2
The conventional measure of L disappears from the model and it is merged and embodied as
‘human capital’ in K. The AK model postulates that the accumulation of capital have positive
long-run effects on growth, {∂Y /∂K} = A > 0. The saving, S(t), affects the steady-state level of
income and growth through its effects on investment and capital accumulation. The capital stock,
K(t) = I(t) + (1 − δ)K(t − 1); δ ∈ [0,1], in period t − 1 is augmented with gross investment, I(t), in
period t and an increase in S(t) = I(t) leads to a higher accumulation of capital and a higher level of
output. A reduced-form bi-variate model for the long-run effects of saving on income can, thus,
be specified as
lnGDP(t) = α + β lnGDS(t) + ε(t); t = 1, 2, . . . , T (1)
Model (1) is estimated on annual data from 1950–1951 to 2001–2002. The GDP is measured in
terms of the real gross domestic product at factor cost at 1993–1994 prices and the GDS in terms
of the real gross domestic saving (GDS). The real GDS is obtained by deflating nominal GDS
by the gross domestic capital formation deflator (base: 1993–1994 = 100). All the data are taken
from various issues of the Reserve Bank of India publication: Handbook of Statistics on the Indian
Economy, and the Central Statistical Organisation publication: National Accounts Statistics.

3. Empirical results

3.1. Unit root tests

The unit root tests are first performed to examine the univariate time-series properties of the
model series. The augmented Dickey–Fuller (ADF) model (Dickey & Fuller, 1981),
k
Y (t) = ␮ + ␥Y (t − 1) + ϑT + δ(i)Y (t − i) + e(t) (2)
i=1
does not reject the null of a unit root in the log-levels of both income (lnGDP) and saving (lnGDS).
The Phillips–Perron (PP) test (Phillips & Perron, 1988) suggests that the model estimated with
constant does not reject the null for both lnGDP and lnGDS. The PP test for model with constant
and trend rejects the null for lnGDS, but not for lnGDP (Table 1). Both ADF and PP tests reject the
null of a unit root in the first-difference series. The KPSS test (Kwiatkowski, Phillips, Schmidt,
& Shin, 1992) rejects the null of no unit root in lnGDP and lnGDS in both the models estimated
with constant as well as with constant and trend. It is well-documented that the ADF and PP

2 The share of capital α = 1 seems to be a strong assumption, as the capital normally have a smaller share of around

one-third of output. This assumption of α = 1, however, could be rationalized considering that the capital K comprises
both physical and human capital.
234 T. Singh / Journal of Policy Modeling 32 (2010) 231–253

Table 1
Unit root tests.
Series Conventional tests GLS-based optimal tests
ADF PP KPSS DF-GLS PT DF-GLSu QT

Level series

Model I: drift and no trend


lnGDP(t) 2.36 (1) 2.36 1.133* 0.86 (4) 35.33* (4) 0.15 (4) 9.10* (4)
lnGDS(t) 0.50 (3) 0.80 1.131* 0.88 (4) 60.89* (4) −0.35 (4) 28.81* (4)
Model II: drift and trend
lnGDP(t) −0.24 (1) −0.56 0.27* −0.62 (3) 26.74* (3) −0.85 (3) 13.80* (3)
lnGDS(t) −1.48 (5) −3.04** 0.19** −1.62 (5) 8.63* (5) −1.74 (5) 4.24* (5)

First-difference series

Model I: drift and no trend


lnGDP(t) −5.58* (1) −8.52* 0.63** −4.88* (1) 1.17 (1) −5.63* (1) 1.79 (1)
lnGDS(t) −6.23* (2) −7.46* 0.15 −6.29* (2) 0.03 (2) −6.30* (2) 0.06 (2)
Model II: drift and trend
lnGDP(t) −6.48* (1) −9.42* 0.11 −4.69* (4) 5.60* (4) −4.69* (4) 3.19* (4)
lnGDS(t) −4.15* (2) −7.60* 0.06 −7.15* (1) 1.72 (1) −7.15* (1) 0.94 (1)

Notes: (1) * and ** denote the statistical significance and the rejection of null at 1% and 5% levels, respectively; (2)
Figures in parentheses are the number of lags. The truncation lags in ADF test are selected using the AIC. The PP test is
performed using the spectral estimation lag windows of lw = 1 and lw = 4. Similarly, the KPSS test is performed using
the lag windows of lw = 1 and lw = 4 for the residual variance of the Newey and West estimator (Newey & West, 1987).
The results obtained under both the lag windows provided similar evidence for the null and are, therefore, reported only
for one of the lag windows (lw = 4) for both PP and KPSS tests.

tests have low power in small samples, while the KPSS test has a tendency to over-reject the
null. The asymptotically powerful DF-GLS, PT, DF-GLSu and QT tests (Elliott, 1999; Elliott,
Rothenberg, & Stock, 1996) are performed to further test the null of a unit root. These tests require
the estimation of GLS-transformed model,
k
yd (t) = φyd (t − 1) + ψ(i)yd (t − i) + e(t) (3)
i=1

The de-trended yd (t) in (3) is obtained as yd (t) = y(t) − β̂0 − β̂1 t and the parameters
(β̂0 , β̂1 ) are estimated by regressing ȳ = [y1 , (1 − ᾱL)y2 , . . . , (1 − ᾱL)yT ] on z̄ = [z1 , (1 −
ᾱL)z2 , . . . , (1 − ᾱL)zT ]; where ᾱ = 1 + c̄T −1 , L is the lag operator and T the number of obser-
vations for y(t). The asymptotic power function is derived for the test against a sequence of local
alternatives of α to one. In the local-to-unity representation for ᾱ, the parameter c̄ is fixed at −7 in
the model with constant mean [z(t) = (1) ] and at −13.5 in the model with linear trend [z(t) = (1,t) ].
The lag-length k in (3) is determined using the Modified Information Criterion (MAIC) of Ng
and Perron (2001),
  2(τ (k) + k)
T
MAIC(k) = ln σ̂k2 + (4)
T − kmax
 −1 2 T d 2 −1 T
The τT (k) = σ̂k2 φ̂ t=kmax +1 (y )t−1 and σ̂k = (T − kmax )
2 2
t=kmax +1 êtk in (4).
Schwert (1989) and Ng and Perron (2001) use the criterion kmax = int{12(T/100)1/4 } to determine
the maximal lag-length; where {x} is the integer value. The study sets the maximal lag-length at
T. Singh / Journal of Policy Modeling 32 (2010) 231–253 235

kmax = int{12(T/100)1/4 } = 10 and estimates each autoregression using the same T − kmax number
of effective observations, a la Ng and Perron (2001). The value of k that minimizes MAIC in (4)
for the GLS model (3) is selected as the optimal lag-length for the DF-GLS, PT, DF-GLSu and
QT tests. These tests provide mixed evidence regarding the order of integration of lnGDP and
lnGDS (Table 1). While the evidence across tests remains mixed, most tests generally suggest the
I(1) properties of the model series.

3.2. Test for cointegration

3.2.1. Standard OLSEG estimates


The studies using the OLS-based two-step cointegration estimator of Engle and Granger (1987)
(OLSEG) perform unit root tests on the residuals of a static model, Y(t) = α + βX(t) + ε(t), and
test H0 : ε(t) ∼ I(1) (no cointegration among I(1) variables) against H1 : ε(t) ∼ I(0) (cointegration
among I(1) variables). The study performs unit root tests on the residuals of model (1) and tests
the mutually-reinforcing nulls of no cointegration (ADF and PP tests) and cointegration (KPSS
test) between saving and income. The ADF test on ε(t) is performed by estimating ε(t) =
μ + γε(t − 1) + ki=1 ζ(i)ε(t − i) + ν(t) and using an autoregressive lag of k = 1. The PP and
KPSS tests are each performed using the lag windows of lw = 1 and lw = 4. The results obtained
from both the lag windows provided similar evidence for the nulls in PP and KPSS tests and are,
therefore, reported only for one of the lag windows (lw = 4). The OLS estimates of model (1)
and the unit root test statistics for the residuals are as follows.
lnGDP (t) = 4.2576 + 0.7615 lnGDS (t); R̄2 = 0.9851; DW = 0.69
(28.72) (58.00) (5)
ADF = −3.15710[−3.34]; PP = −3.31256[−3.34]; KPSS = 0.347[0.463]
The figures in parentheses are t-ratios and those in brackets are 5% critical values for the null of a
unit toot for the ADF and PP tests (Davidson & MacKinnon, 1993) and no unit toot for the KPSS
test (Kwiatkowski et al., 1992). The ADF and PP statistics are marginally lower than the critical
values and do not reject the null of a unit root in the residuals of model (5).3 In contrast, the KPSS
test does not reject the contrary null of no unit root. These residual-based tests provide mixed
evidence; the ADF and PP tests suggest the absence, while the KPSS test supports the presence
of cointegrating relationship between saving and income.
The residual process of a cointegrating regression contains useful information on the equilib-
rium relationship between the model series. Xiao (1999) and Xiao and Phillips (2002) suggest
that the cumulative sum (CUSUM) of recursive residuals test of Brown, Durbin, and Evans (1975)
can be applied to the residuals of a regression to directly test the null of cointegration. If the given
Y(t) and X(t) sequences are cointegrated, then the residuals of a cointegrating regression should be
stable with long-run movements within the critical bounds. The study performs the CUSUM test
of Brown et al. (1975) on the OLS recursive residuals of model (5) and examines the equilibrium
relationship between saving and income.4 Such CUSUM test represents a graphical rendition of

3 The use of 5% critical values from Phillips and Ouliaris (1990) also provided the similar evidence for the null for the

ADF and PP tests.


4 The CUSUM and CUSUM of squares (CUSUMSQ) tests have been commonly used to test the stability (constancy)

of model parameters, and examine the structural break in a regression function. Ploberger and Krämer (1992) replace
recursive residuals by OLS residuals and provide a OLS residual-based rendition of the CUSUM test of Brown et al.
(1975). They show that the CUSUM test for the constancy over time of the coefficients of a linear regression model, which
236 T. Singh / Journal of Policy Modeling 32 (2010) 231–253

Fig. 1. CUSUM of OLS recursive residuals.

the residual-based cointegration tests and it ‘band tests’ the null of stability (cointegration among
I(1) variables) against instability (no cointegration among I(1) variables) of regression residuals.
The plot of the CUSUM of OLS recursive residuals suggests that these residuals do not drift
beyond 5% upper and lower critical bounds (Fig. 1). The CUSUM test, thus, shows the long-run
stability of recursive residuals and suggests the cointegrating relationship between saving and
income.

3.2.2. Optimal DOLS, FMOLS and NLLS estimates


The neoclassical model of economic growth does not control for the endogeneity of saving.
The strand of studies examining the determinants of saving instead suggests the endogeneity of
saving (Attanasio et al., 2000; Carroll et al., 2000; Carroll & Weil, 1994; Deaton & Paxson, 2000;
Loayza et al., 2000; Rodrik, 2000; Schmidt-Hebbel et al., 1996). Sarno and Taylor (1998) argue
that some degree of consensus is emerging that most commonly suggested determinant of the
rate of saving is the rate of growth of output. In the presence of endogeniety of regressors and
serial correlation of errors, the standard OLS estimates become biased and inefficient. The ‘super-
consistency’ property of OLS indeed allows to omit I(0) regressors from the cointegrating model
and asymptotically ignore the problems of endogeniety and serial correlation.5 In small samples,
however, the OLS estimates remain biased and have inferential problems for the significance
of long-run parameters. The bias is often substantial (Banerjee, Dolado, Galbraith, & Hendry,
1993; Inder, 1993) and the t statistics of the cointegrating coefficients are not generally valid
for statistical inference. The use of instrumental variable (IV) estimator to resolve endogeneity
could be beset with problems, if the instruments do not satisfy the orthogonality conditions and
are autocorrelated and I(d). These limitations of standard OLS motivate the need to introduce an
explicit AR(1) specification for X(t), along with the stochastic model for the relationship between
Y(t) and X(t). The triangular representation of the cointegrated system of Phillips (1991) with I(1)

is usually based on recursive residuals, can also be applied to OLS residuals. Hao and Inder (1996) derive asymptotic
distribution of the OLS-based CUSUM test to test structural break in cointegrated regression models. The CUSUM test for
the null of parameter constancy and that for the null of cointegration have the same behaviour under the null hypothesis,
but are different under the alternative (Xiao and Phillips, 2002). For a review and discussion on the tests for model stability,
see Hansen (1992).
5 The OLS estimators of a regression are ‘super-consistent’ when the model series are cointegrated. Instead of approach-

ing their true values at a rate proportional to n−1/2 , the OLS estimates will approach them at a rate proportional to n−1
(Davidson and MacKinnon, 1993). The ‘super-consistency’ properity allows to asymptotically ignore the problems of
endogeniety bias and serial correlation. In small samples, however, the OLS estimates become biased, and such bias leads
to inferential problems for the significance of the parameters of long-run model.
T. Singh / Journal of Policy Modeling 32 (2010) 231–253 237

series of Y(t) and X(t),


Y (t) = α + βX(t) + μ(t) (6)
X(t) = η(t) (7)
and I(0) series of μ(t) suggests that the OLS estimator of β is consistent, but not generally fully
efficient. The asymptotic distribution of such estimator depends on various nuisance parame-
ters engendered by serial correlation in μ(t) and by correlation between μ(t) and innovation
term for X(t) in (7). The μ(t) and η(t) are cross-correlated not only contemporaneously,
but also at various leads and lags. Phillips (1991) suggests using the following representation
for μ(t),

k
μ(t) = δ(j)η(t − j) + ξ(t) (8)
j= −k

The ξ(t) in (8) is not correlated with η(t − j), ∀j ∈ [−k,k]. The cointegrating model (6) can be
augmented with the leads and lags of X(t) to resolve the problem of cross-correlation between
μ(t) and η(t) (Phillips & Loretan, 1991; Saikkonen, 1991; Stock & Watson, 1993). By substituting
(7) into (8) and then substituting the resulting equation into (6), the leads and lags cointegration
estimator can be expressed as

k
Y (t) = α + βX(t) + δ(j)X(t − j) + ξ(t) (9)
j=−k

Since ξ(t) is not correlated with η(t) in (8), it will also be uncorrelated with X(t) in (9).
The X(t) asymptotically eliminates the effect of endogeniety of X(t) on the distribution of
OLS estimator of β. If ξ(t) is independently and identically distributed, then the standard dis-
tribution theory can be used to perform inference on the OLS parameter estimates. While the
leads and lags of X(t) resolve the problem of endogeniety of X(t), they do not necessarily
eliminate all serial correlation and heteroskedasticity in ξ(t). Stock and Watson (1989, 1993)
suggest using the generalised least squares (GLS) to estimate (9). The GLS estimates of stan-
dard errors and variance–covariance matrix could be used to construct the asymptotically valid
Chi-squared hypothesis tests on β. Phillips and Loretan (1991) argue that due to persistence in
the effects of innovations arising from unit roots in the system, the lags of X(t) are generally
not an adequate proxy for the past history of μ(t), and they suggest using a parametric correc-
tion in (9) to account for potential serial correlation in ξ(t). The requisite information set for
valid conditioning is better modelled by using lagged equilibria than by using lagged differ-
ences of the dependent variable, and they recommend augmenting (9) with the lagged levels of
[Y(t) − α − βX(t)],

k 
k
Y (t)=β0 +βX(t)+ δ(j)X(t − j)+ φ(j)[Y (t − j)−α − βX(t − j)]+ζ(t) (10)
j=−k j=1
The ξ(t) is serially uncorrelated and (10) can be estimated using the non-linear least
squares (NLLS) estimator. The NLLS estimator of β is asymptotically efficient and the
estimates of variance–covariance matrix have the standard limiting distribution. The NLLS
variance–covariance matrix can be used to perform hypothesis test on β in a standard manner. Both
DOLS (9) and NLLS (10) estimators are unbiased and asymptotically efficient in the presence of
238 T. Singh / Journal of Policy Modeling 32 (2010) 231–253

endogeniety of regressors and serial correlation of errors. An alternative to modelling dynamic


processes and I(0) terms is to use the fully modified OLS (FMOLS) estimator of Phillips and
Hansen (1990). The FMOLS makes a non-parametric correction to OLS estimates. The modified
t-statistics from FMOLS are asymptotically normal and the hypotheses tests can be performed in
a standard manner.6
The lag, contemporaneous and lead structures of the dynamic regressors used in DOLS and
NLLS are determined using the model selection criteria. Both AIC and SIC suggested the dynamic
structure of k{−2, 0, +2} for DOLS and k{−1, 0, +1} for NLLS. In the model estimated with
k{−1, 0, +1}, the NLLS estimates were, however, marked by the problem of serial correlation.
The model using NLLS is, therefore, estimated using a dynamic structure of k{−2, 0, +2}. The
model is also estimated using a higher structure of k{−3, 0, +3} to examine the sensitivity of
results; both the structures are supported by Sawa criterion (Sawa, 1978).7 The NLLS estimation
is carried out using the Gauss–Newton algorithm, and the standard errors are adjusted using the
heteroscedasticity and autocorrelation consistent (HAC) estimator of Newey and West (1987).
The FMOLS estimates are obtained using the lag windows of lw = 1 and lw = 4. The results
obtained from DOLS, FMOLS and NLLS consistently suggest the significant long-run effects
of domestic saving on income. The long-run coefficient on lnGDS(t) carries the theoretically
predicted positive sign and is statistically significant at 1% level in all the models (Table 2).

3.2.2.1. New CUSUM and MOSUM Tests. The new CUSUM and MOSUM tests of Xiao (1999)
and Xiao and Phillips (2002) represent the optimal variants of the conventional CUSUM test
of Brown et al. (1975). Xiao and Phillips (2002) use the fully modified OLS to resolve serial
correlation and endogeneity, and they construct the cumulative sum (CSn ) and moving sum (MSn )
test statistics to directly test the null of cointegration against the alternative of no cointegration.

1  k
CSn = max 2 n
û+
t (11)
k=1,...,n ω̂u,x
t=1

1 
k+[nh]
MSn = max 2 n
û+
t (12)
k=1,...,n−[nh] ω̂u,x
t=k

The CSn and MSn measure the magnitude of variation in the optimal residuals relative to the
magnitude of estimated long-run variance of ut in (6) conditional on ηt in (7). In (11) and (12),
û+ 2
t denotes the optimal residuals of potential equilibrium relationship, ω̂u,x is a semi-parametric
kernel estimator of ␻2u,x , and 0 < h < 1 is a bandwidth parameter for the moving window. The ω̂u,x
2

measures the long-run variance of ut conditional on ηt , and it depends on the bandwidth parameter.
The upper tail critical values for CSn statistics are provided in Xiao and Phillips (2002), and the
critical values for MSn statistics depend on the width of the moving window. The study uses
the optimal residuals from DOLS, FMOLS and NLLS estimators and constructs the CSn and
MSn statistics (Table 2). Since the CSn and MSn use optimal residuals, these tests overcome the
problems of endogeneity of regressors and serial correlation of errors and provide unbiased and

6 The FMOLS estimator provides valid t statistics for the long-run coefficients of a static regression in levels and resolves

the problem of statistical inference.


7 Sawa (1978) argues that AIC tends to choose the model with a higher lag order than the true model, but the bias is

small or negligible when k < [T/10].


Table 2
Optimal single-equation estimates for the long-run effects of domestic saving on income.
Regressor Dependent variable: ln GDP(t)
DOLS FMOLS NLLS

T. Singh / Journal of Policy Modeling 32 (2010) 231–253


k{−2, 0, +2} k{−3, 0, +3} [lw = 1] [lw = 4] k{−2, 0, +2} k{−3, 0, +3}

Constant 3.9191* (7.98) 3.8915* (7.13) 4.1970* (21.92) 4.2147* (16.69) 0.2262 (1.01) 0.2097 (0.90)
lnGDS(t) 0.7892* (18.66) 0.7915* (16.78) 0.7668* (45.32) 0.7652* (34.28) 0.8867* (9.16) 0.9267* (6.51)
␸ – – – – 0.9282* (16.99) 0.9263* (17.39)

CUSUM and MOSUM tests for the null of cointergration

CSn
0.9529 0.8173 1.2421 1.2889 0.5869 0.5356

Bandwidth parameter MSn

h = 0.1 0.4679 0.4543 1.2689 1.2691 0.5231 0.5052


h = 0.2 0.7931 0.7360 1.9282 1.9316 0.7132 0.7132
h = 0.3 0.8140 0.7850 2.3130 2.3208 0.8711 0.7312
h = 0.4 1.0050 0.9977 2.2821 2.2957 0.9107 0.8089

Notes: (1) Figures in parentheses are t-ratios; (2) * denotes the statistical significance at 1% level; (3) The critical values for the CSn statistic are 2.326 (1%) and 1.842 (5%)
(Xiao & Phillips, 2002; Table 1; p. 49); (4) DOLS estimates are corrected for AR(1) serial correlation using the Cochrane-Orcutt iterative procedure; (5) The coefficients of
the lag, contemporaneous and lead first-difference dynamic regressors of k{−2, 0, +21} and k{−3, 0, +3} used in DOLS and NLLS estimators do not carry any economically
meaningful interpretation and are, therefore, not reported to conserve space.

239
240 T. Singh / Journal of Policy Modeling 32 (2010) 231–253

efficient estimates as compared to the conventional CUSUM test based on standard OLS recursive
residuals. The constructed CSn statistics are less than the 5% critical value and do not reject the null
of cointegration for all the models. These results reinforce the equilibrium relationship between
saving and income.8

3.2.3. Bounds test


The bounds testing estimator of Pesaran, Shin, and Smith (2001) is invariant to the integration
properties of the model series. The estimator requires the estimation of conditional error-correction
model (ECM) based on the autoregressive distributed lag (ARDL) model,
k
 lnGDP(t) = γ0 + δ1 lnGDP(t − 1) + δ2 lnGDS(t − 1) + θ(i) lnGDP(t − i)
i=1


k
+ π(i)Δ ln GDS(t − i) + ε(t) (13)
i=1

Under the condition lnGDP = lnGDS = 0, the reduced-form solution of (13) yields the
long-run model for lnGDP(t) as
lnGDP(t) = ω0 + ω1 lnGDS(t) (14)
where ω0 = −[γ 0 /δ1 ] and ω1 = −[δ2 /δ1 ].
/ δ2 =
The joint F-test is performed to test H0 : δ1 = δ2 = 0 against H1 : δ1 = / 0, and the t-test is
used to test H0 : δ1 = 0 against H0 : δ1 =
/ 0 in (13). The asymptotic distributions of both the test
statistics are non-standard and Pesaran et al. (2001) provide the upper [if regressors are I(1)] and
lower [if regressors are I(0)] asymptotic critical value bounds for the F and t statistics to ‘bound
test’ the null of no cointegration among the I(d) (0 ≤ d ≤ 1) variables.9 Model (13) is estimated
using k = 5 and the standard errors are adjusted using the HAC estimator (Newey & West, 1987).
The estimates of ARDL-ECM (with t-values in parentheses) are as follows.
 lnGDP(t) = −1.5645 + 0.3082 lnGDP(t − 1) − 0.1998 lnGDS(t − 1)
(−7.16) (6.49) (−5.78)

5 
5
− 2.9132 lnGDP(t − i) + 0.4662 lnGDS(t − i) (15)
(−7.00) (3.10)
i=1 i=1
R̄2 = 0.41; DW = 1.81; Ljung-Box Q = 8.99; F = 37.57 : H0 : δ1 = δ2 = 0
The F-statistics rejects H0 : δ1 = δ2 = 0 and the t-statistics rejects H0 : δ1 = 0 in (15); both the
statistics suggest the cointegrating relationship between domestic saving and income. Using the
reduced-form solution (14) and normalizing the coefficients of lagged level regressors (including
constant term) on the coefficient of lnGDP(t − 1), the long-run rendition of (15) is derived as
lnGDP(t) = 5.0762 + 0.6483 lnGDS(t) (16)

8 Since cumulative sum in CS is replaced by moving sum in MS , the critical values for MS should intuitively be in
n n n
the vicinity of the critical value for CSn . By this analogy, the MSn statistics for the DOLS and NLLS estimators do not
reject the null of cointegration between income and saving. The MSn statistics for the FMOLS are, however, higher for
the bandwidth parameter of h = 0.2, h = 0.3 and h = 0.4, and these statistics seem to reject the null.
9 If the F and t statistics exceed the upper bound critical value, then the null of no cointegration is rejected. If the F and

t statistics fall below the lower bound critical value, then the null of no cointegration is not rejected, and if these statistics
fall within the upper and lower critical bounds, then the statistical inference becomes inconclusive.
T. Singh / Journal of Policy Modeling 32 (2010) 231–253 241

The rejection of H0 : δ1 = δ2 = 0 in (15) implies the rejection of H0 : ω1 = 0 in (16). This suggests


that the implied long-run coefficient on lnGDS(t) in (16) is statistically significant.

3.2.4. ML system estimates


The maximum-likelihood (ML) system estimator of Johansen (1991) takes a system-based
account of endogeneity and provides a useful alternative to single-equation approach. For the data
generated by a kth order vector autoregressive (VAR) process with non-zero drift and Gaussian
errors, the ML estimator estimates the VAR model represented by


k−1
X(t) = Γ (i)X(t − i) + Π̃X(t − 1) + μ + ε(t) (17)
i=1
k−1 k
where Γ (i) = − I − i=1 Π(i) ; Π̃ = − I − i=1 Π(i) ; Π = αβ .
The X = [lnGDP lnGDS] is a p × 1 vector of p number of I(1) variables, μ is a vector of
constants
 and ε(t) is a
p-dimensional vector of disturbances with zero mean and covariance matrix
[i.e. ε(t) ∼ iid(0, )]. The model selection criteria and the lag-length selection tests did not
provide any unequivocal indication for the truncation of lag structure. The SIC suggested k = 1,
Sims (1980) LR test k = 3, and the LM test for first-order serial correlation suggested the absence
of serial correlation at k = 2. Since lag k = 1 (suggested by SIC) seems too small to sap serial
correlation, the model is estimated using lag k = 2 (as suggested by LM test for serial correlation)
and lag k = 3 (as suggested by LR test); the use of both the lag structures is intended to examine
the sensitivity of results. The asymptotic distribution of the rank test and the hypotheses tests on
β in small samples may not be good approximations to the actual finite sample distributions. The
small sample correction suggested by Johansen (2000, 2002) is carried out to adjust the λ-trace
for small sample and account for the possible bias that may arise from the length of sample span
on the test statistics. The asymptotic λ-trace and the λ-trace adjusted for small sample suggest two
cointegrating vectors in both the models estimated with k = 2 and k = 3. The long-run coefficients
of the first cointegrating vector normalised on lnGDP carry the theoretically predicted positive
signs, and LR test rejects the null of zero-restrictions on these coefficients (Table 3).10
The test of parameter constancy, based on recursive estimation, is performed to test the temporal
stability of the cointegrating vector (Hansen & Johansen, 1993, 1999). The test involves selecting
a base sample of X−k+1 , . . . , XT0 and then recursively estimating the eigenvalues by increasing
samples X−k+1 , . . ., Xt for t = T0 + 1, . . ., X. The recursive estimation is performed in two different
ways; one by reestimating the full system and all the parameters of model (17) including short-
run dynamics for each sub-sample (called X-form), and second by reestimating only the long-
run parameters α and β and concentrating out the short-run dynamics (called R1-form) prior to
performing recursive estimation. The study performs these recursive estimations using the sample
starting 1953 for the model with k = 2 and 1954 for the model with k = 3, and then increasing
these samples until 2002. The maximum test statistics for both X-form and R1-form recursive
estimations are plotted in Fig. 2. These test statistics are scaled by 5% critical value such that the
value of X(t) and R1(t) larger than unity would reject the null of constancy of β vector. In both the
models estimated with k = 2 and k = 3, the maximum test statistics for X(t) and R1(t) remain less

10 The second cointegrating vector normalised on lnGDP was estimated to be [1, −0.894] for Model I (k = 2) and [1,

−1.164] for Model II (k = 3).


242 T. Singh / Journal of Policy Modeling 32 (2010) 231–253

Table 3
Johansen’s ML estimates for the long-run relationship between domestic saving and income.
Model I: X = [lnGDP, lnGDS]; [VAR lag k = 2]

Null Eigenvalues λ-trace λ-trace@ (small sample) 95% critical value ␭-max 95% critical value

r≤0 0.244 18.593 17.580 15.41 13.95 14.07


r≤1 0.089 4.639 4.277 3.841 4.64 3.76

Cointegrating vector normalised on lnGDP LR Test of weak exogeneity

lnGDP lnGDS lnGDP lnGDS 95% χ2

1 −0.732 (8.257) 3.256 7.943* 3.841

Model II: X = [lnGDP, lnGDS]; [VAR lag k = 3]

Null Eigenvalues λ-trace λ-trace@ (small sample) 95% critical value λ-max 95% critical value

r≤0 0.215 17.359 16.342 15.408 11.84 14.07


r≤1 0.107 5.518 2.957 3.841 5.52 3.76

Cointegrating vector normalised on lnGDP LR Test of weak exogeneity


lnGDP lnGDS lnGDP lnGDS 95% χ2

1 −0.683 (3.458) 5.904* 1.590 3.841

Notes: (1) The r denotes the number of cointegrating vectors; (2) @ denotes λ-trace corrected for small sample; (3) Figures
in parenthesis represent the LR test statistics for the exclusion restriction on lnGDS.

Fig. 2. Recursive test of beta constancy.


T. Singh / Journal of Policy Modeling 32 (2010) 231–253 243

than unity and do not reject the null of constancy of β vector. These tests suggest the temporal
stability of the estimated parameters.

3.2.5. Short-run dynamics


The estimates of cointegrating model show only the long-run equilibrium relationship between
saving and income, and do not provide any information on the short-run dynamics. The well-known
Granger Representation Theorem (Engle & Granger, 1987) suggests that if a set of variables are
I(1) and are cointegrated, then there exists a valid error-correction representation of the time series.
The error-correction model (ECM) is estimated to examine the short-run dynamics and test the
null of Granger non-causality between saving and growth.

k
 lnGDP(t) =  + β(i)  lnGDP(t − i)
i=1


k
+ γ(i) lnGDS(t − i) + α1 z(t − 1) + ν1 (t) (18)
i=1


k
 lnGDS(t) = τ + ζ(i)Δ ln GDS(t − i)
i=1


k
+ ς(i)Δ ln GDP(t − i) + α2 z(t − 1) + υ2 (t) (19)
i=1

The lagged error-correction term z(t − 1) in (18) and (19) is obtained from the first cointegrating
vector of the model estimated with k = 2 (Model I, Table 3). The z(t − 1) mimics the linear
combination of the lagged
 level regressors used in ARDL-ECM (13). The saving  causes economic
growth, if the null of ki=1 γ (i) = 0 or α1 = 0 in (18) is rejected, but the null of ki=1 ς(i) = 0 or

␣2 = 0 in (19) is not rejected. The economic growth causes saving, if the null of ki=1 ς(i) = 0 or

α2 = 0 in (19) is rejected, but the null of ki=1 γ(i) = 0 or α1 = 0 in (18) is not rejected. The saving

and growth are characterised by bidirectional causality, if the null of ki=1 γ(i) = 0 or α1 = 0 in

(18) and ki=1 ς(i) = 0 or α2 = 0 in (19) are rejected. The standard errors of estimated (18) and
(19) are adjusted using the HAC estimator (Newey & West, 1987). The coefficients of lagged
distributed regressors are insignificant, while those of autoregressive regressors are significant in
most cases (Table 4). The coefficients of z(t − 1) are significant at 1% level in the model with
lnGDP(t) as the regressand, and at 5% (k = 2) and 10% (k = 3) levels in the model with lnGDS(t)
as the regressand. The coefficients of z(t − 1), however, carry the counterintuitive positive signs
in all the models, and this evidence is difficult to reconcile with the long-run convergence towards
steady-state equilibria suggested by the cointegrating model. The ECM provides a somewhat
weak and fragile evidence for the short-run effects of saving on growth.
The impulse response and variance decomposition analyses are carried out to trace the time-
profile and map the response trajectories of lnGDP [lnGDS] to the shocks to the innovations of
lnGDS [lnGDP] and further examine the direction of causality between saving and growth.11

11 The impulse response and variance decomposition analyses decompose the determinants of endogenous variable into

the innovations identified with a specific variable. If the innovations are characterised by contemporaneous correlation,
244 T. Singh / Journal of Policy Modeling 32 (2010) 231–253

Table 4
Error-correction model and the non-causality between saving and growth.
Regressor Dependent variable

lnGDP(t) lnGDS(t)
k=2 k=3 k=2 k=3

Constant −0.8493* (−4.96) −0.8288* (−4.10) −1.1758** (−2.14) −1.0539*** (−1.75)


k
lnGDP(t − i) −0.8507* (−3.69) −0.8403** (−2.48) 0.4348 (0.76) 0.5674 (0.54)
ki=1

i=1
lnGDS(t − i) 0.1060** (2.41) −0.0089 (−0.06) −0.2971*** (−1.70) −0.6530 (−1.66)
z(t − 1)@ 0.1797* (5.19) 0.1768* (4.34) 0.2396** (2.21) 0.2186*** (1.82)

R̄2 0.1441 0.1683 0.1544 0.1769


DW 1.99 2.06 2.17 2.01
LB-Q 26.69 18.46 3.82 5.68

Notes: (1) Figures in parentheses are t-ratios, and LB-Q is the Ljung–Box Portmanteau statistics; (2) *, ** and *** denote
the statistical significance at 1%, 5% and 10% levels, respectively; (3) @ The lagged error-correction term z(t − 1) is
obtained from the first cointegrating vector of the model estimated with k = 2 (Model I, Table 3).

Fig. 3. Impulse response functions and the standard error bands.

The analysis is carried out for a time horizon of 10 years. The Monte Carlo simulations using
10,000 draws are performed to compute the standard error bands around the impulse response
functions. The estimates suggest that all the impulse response functions move within the narrow
standard error bands and converge to zero quite fast (Fig. 3). The decomposition of variance
shows the quantitative effects of shocks given to the innovations of model series (Table 5).12 Both

then the results can be sensitive to the Cholesky ordering scheme of the system variables. The plots of the impulse
response functions and the quantitative analysis in terms of the decomposition of variance showed consistent results
across alternative ordering schemes. The results are, therefore, reported only for one ordering scheme given by [DlnGDP,
DlnGDS].
12 The variance decomposition shows k-step ahead forecast error variance of each variable explained by its own inno-

vations and the innovations of other variables in the system. The forecast error variance is the sum total of the variances
and covariances of all the innovation series.
T. Singh / Journal of Policy Modeling 32 (2010) 231–253 245

Table 5
Decomposition of variance of model series [VAR lag k = 2].
Period DlnGDP DlnGDS

S.E. DlnGDP DlnGDS S.E. DlnGDP DlnGDS

1 0.03 100.00 0.00 0.07 29.66 70.35


2 0.03 97.98 2.02 0.07 31.73 68.27
3 0.03 97.24 2.76 0.07 29.87 70.14
4 0.03 97.14 2.86 0.08 31.30 68.70
5 0.03 96.89 3.11 0.08 31.35 68.66
10 0.03 96.86 3.14 0.08 31.62 68.38

Notes: (1) S.E. denotes the standard error; (2) Cholesky ordering: [DlnGDP, DlnGDS].

impulse response functions and decompositions of variance suggest that lnGDP [DlnGDP]
and lnGDS [DlnGDS] show large responses to the shocks to their own innovations. Besides,
the response of DlnGDS to the innovations of DlnGDP is relatively larger than the response of
DlnGDP to the shocks to the innovations of DlnGDS. These results suggest bidirectional causality
between saving and growth, with a relatively stronger evidence for the feedback effects of growth
on saving. The higher saving is both succeeded (saving causes growth) and preceded (growth
causes saving) by higher growth.

4. Saving and growth accelerations: some analytics

The stylized evidence supporting the steady-state effects of domestic saving on income has
important implications for the formulation of macroeconomic policies. It suggests the need to
accelerate domestic saving to foster higher income and growth and generate the virtuous circles
of high saving and high growth.13 The acceleration of saving becomes particularly essential, as a
predominant proportion of investment is financed by domestic saving. The heavy reliance of invest-
ment on domestic saving reinforces the ‘Lucas Puzzle’ (Lucas, 1990) on the lack of capital flows
from the developed countries to the developing countries.14 The saving is a provision for future
consumption for the individual households and a source of capital accumulation for the economy at
large. A myriad of factors underpin the motivation to save, such as income and wealth, demograph-
ics, unobservable tastes and preferences, intergenerational altruism, borrowing and liquidity con-
straints, financial infrastructure, uncertainty in labour and capital incomes, time-inconsistencies
in household preferences and saving plans, rates of returns, budgetary and fiscal policies, and
the economic incentives. The saving is positively associated with the precautionary motive in the
Keynesian model and the consumption-smoothing motive in the Life-Cycle (Modigliani, 1988;
Modigliani & Ando, 1957; Modigliani & Brumberg, 1954), permanent-income (Friedman, 1957)

13 The acceleration of saving is also useful to reduce the trade and current account deficits. Mankiw (2006) argues that

a smaller federal budget deficit would mean more national saving, less reliance on foreign capital flows, and a smaller
trade deficit. Feldstein (2007) asserts that the increase in saving (or, more accurately, in saving relative to investment) is
a necessary condition for reducing the trade deficit, but it is not sufficient.
14 Lucas (1990) argues that the poor countries and regions with scarce capital and implied higher marginal product of

capital are not able to attract capital that would facilitate their convergence towards the rich and frontier countries or
regions. Several factors seem catalytic to the lack of capital inflows from the rest of the world to developing countries:
lack of well-developed financial markets, ‘lemons problem’ (Akerlof, 1970), market imperfections, exchange rate risks
and the inadequate infrastructure in the developing countries.
246 T. Singh / Journal of Policy Modeling 32 (2010) 231–253

Fig. 4. Rates and sectoral composition of gross domestic saving.

and buffer-stock (Carroll, 1997; Deaton, 1991; Deaton & Paxson, 2000) models. The acceleration
of saving hinges on the accelerations of private saving or public saving or a combination of both.

4.1. Private saving

The private saving comprises the saving of the household and corporate business sectors.15
The household sector in India has consistently been a net saver, and it makes a predominant
contribution to domestic saving (Fig. 4(4.1)). The saving involves a trade-off between present and
future consumption, and is a corollary of the intertemporal consumption-smoothing decisions of
households. The uncertainty in income is predicted to have positive effects on saving through
precautionary effect. Several factors could be catalytic to net household saving, such as self-
imposed debt-aversion, borrowing and liquidity constraints, and the precautionary saving induced
by uncertainties in incomes. The inter- and intra-annual incomes of the rural households have been
commonly prone to exogenous weather shocks in agriculture. The influence of these shocks has
tended to subside with the diminishing dependence on agriculture, as embedded in the process
of structural transformation of the economy; these shocks, however, still remain an influential
source of non-linear disruptions in rural incomes and dynamic inconsistencies in the saving plans
of households. The insurance market is generally sluggish in the urban and virtually non-existent
in the rural areas. The unknown probability distributions of future incomes juxtaposed with an
inadequate insurance accentuate the need for precautionary saving or buffer-stock of assets for the
risk-averse households with binding liquidity and borrowing constraints. The removal of credit
constraints and the increased access to credit market have ambiguous effects on saving: these are

15 The household sector in India comprises a wide spectrum of all individuals, consumers, non-profit institutions, non-

government non-corporate enterprises of farm business and non-farm business like sole proprietorships and partnerships.
The household sector holds its saving in both physical and financial assets. The household saving in financial assets is
measured on a net basis, as gross financial assets minus financial liabilities. The saving in physical assets represents the
household capital formation. The private corporate business sector consists of all the non-government non-financial public
and private limited joint stock companies, and the cooperative non-credit societies.
T. Singh / Journal of Policy Modeling 32 (2010) 231–253 247

likely to reduce saving through the increase in borrowing as much as induce saving through the
stimulus to investment and growth. The inflation is another factor that has ambiguous effects on
saving. The inflation erodes the real value of wealth held in financial assets and reduces the private
saving, while the uncertainty arising from inflation may entice precautionary saving and offset the
decline in saving. The neo-Ricardian equivalence rendition of inflation-tax would suggest that the
inflation arising from seigniorage or supply-side shocks induces an equivalent increase in saving
to pay for future inflation-tax and perfectly offset the inflation-eroded decline in saving.
The rate of interest represents the nominal return on financial assets and that of inflation
measures the return on physical assets. The rise in interest and/or reduction in inflation increases
the relative returns on financial assets through the Fisher effect and possibly leads to the reshuffling
of the saving portfolios. A representative household with two-asset (physical and financial)
 saving
portfolio would prefer
 a higher acquisition of financial assets when r̃ = ĩ − π̃ >
 0, physical

assets when r̃ = ĩ − π̃ < 0, and would have a portfolio equilibrium when r̃ = ĩ − π̃ = 0;
where r̃ is the relative (real) rate of return, ĩ is the nominal rate of return on financial assets
and π̃ is the rate of return on physical assets proxied in terms of the rate of inflation on these
assets (household capital formation deflator). The tax incentives raise the net (after tax) returns
and induce the acquisitions of tax-favoured financial assets. The compositional dynamics of the
two-asset saving portfolio of households suggest that the share of saving in physical assets has
decelerated from a high of three-fourth in the 1950s to around a half of the household saving in
the early 2000s, while that of saving in financial assets has commensurately accelerated from a
low of one-fourth to around a half of the household saving over the same period; the horizontal
grid in Fig. 4(4.3) shows the 50% mark. While the relative rates of return (interest rate and
tax incentives) incite the reshuffling and rebalancing of existing portfolios, these have ambiguous
effects on the net accumulations of saving. The effects of returns on saving depend on the opposing
income and substitution effects. The rates of returns augment saving only when the substitution
effect is stronger than the income effect and when the contributions are financed through the
abstinence from consumption and increases in labour supply, rather than through the reshufflings
of the existing portfolios towards the high-return instruments.16 The observed behaviour of the
saving portfolio of households suggests that the substitution effect seems to dominate the income
effect, as the dynamic reshuffling from physical to financial assets has been accompanied by the
increases in saving in that the rates of saving in physical and financial assets have tended to show
increases over time (Fig. 4(4.4)). The magnitude of the effect depends on the sensitivity of saving
with regard to the rates of return. The tax levies on consumption (such as sales tax on goods
and services) and the tax concessions and exemptions on investment incomes (such as interest
on bonds and deposits, and dividends on stocks) lead to lower current consumption and higher
private saving. The tax exemptions on the incomes accruing from saving instruments encouraged
the households saving in financial assets.

16 The rate of interest measures the opportunity cost of choosing between present and future consumption. A rise in

interest rate raises the opportunity cost of present consumption and could induce an increase in saving (fall in present
consumption) through the substitution effect. The rise in interest rate also increases the income and could stimulate
the increase in both present and future consumptions and result in a decline in saving; assuming that both present and
future consumptions are normal goods. This suggests that the eventual effects of an increase in interest rate on present
consumption and saving are ambiguous; the substitution effect increases the saving (reduces the present consumption),
while the income effect reduces the saving (raises the present consumption). The net effect of the increase in real interest
rate on saving depends on the relative magnitudes of negative income and positive substitution effects. The substitution
effect is, however, generally deemed to be stronger than the income effect and hence a rise in rate of interest is generally
expected to encourage the saving.
248 T. Singh / Journal of Policy Modeling 32 (2010) 231–253

The development of financial infrastructure is another factor that has been catalytic to the
increased preferences of households for financial assets. The financial sector facilitates the diver-
sion of saving into productive investment. The institutional approach to the development of
financial sector and the geographical spread of banking facilities in the 1960s and the 1970s
contributed to the promotion of rural and urban financial intermediation and to the mobilization
of spatially scattered household saving. The household sector channels a large proportion of its
saving into investment indirectly through financial intermediaries, rather than directly through the
financial markets. The distribution of the financial saving portfolio of households has remained
consistently skewed in favour of the safe and risk-free financial assets comprising deposits, cur-
rency, government securities, small saving and social security funds (life fund and provident fund).
The saving portfolios of rural households mainly remain concentrated on bank deposits, and some
of the non-bank financial instruments are still not very popular with the rural households. The
compulsory saving instruments, such as provident fund, remain restricted to the organised sector,
and do not encompass the mobilizations from the self-employed households in the unorganised
sector. Given the rural-dominated demographic structure, the increase in financial literacy and
geographical proximity of non-bank financial instruments would help rural households make
informed asset choices and hold the optimally diversified saving portfolios. The financial devel-
opment commonly carries along the risks of financial instability and the likelihoods of non-linear
disruptions to financial institutions and markets. The wide-ranging economic reforms undertaken
since the 1990s (July 1991) pro-actively evolved several measures to strengthen the financial
sector and enhance its resilience to the exogenous shocks.17
The positive effects of a policy parameter based on partial equilibrium analysis could have
crowding-out offsets in a general equilibrium setting. The partial equilibrium implications of the
incentive-based policy measures would need to be examined in juxtaposition with the general
equilibrium effects so as to generate the positive net outcomes. The positive effects of interest on
saving that may arise from the likely dominance of substitution effect over income effect could
be crowded out by its negative effects on investment and implied output. Similarly, the reduction
in inflation raises the real rate of interest, which possibly encourages saving, but discourages
investment. The interest rate is vulnerable to several endogenous and exogenous macroeconomic
shocks and, given its ambiguous effects on saving and unambiguous negative effects on investment,
it does not seem to be a tenable tool for inducing long-term saving and investment. The tax
incentives are useful to smooth the intertemporal distortions in the saving portfolios and foster
the financialization of saving. However, if the tax incentives induce only a realignment of private
saving towards the tax-favoured assets, and do not make any net addition to private saving, then
these incentives would reduce the aggregate saving through the reductions in tax revenues and
public saving. The economic incentives also tend to become susceptible to diminishing returns
in terms of saving generations, if these are not accompanied by the accelerations of income
and growth. The incentive-based measures to induce the motivation to save (reduce the present
consumption) need to be supported by the productivity-based measures to accelerate income
and growth, and strengthen the capacity to save. The economic growth in a labour-abundant and
capital-scant economy like India hinges heavily on the deepening of capital and the improvements

17 The emphasis has been laid on strong prudential supervision system to reduce excessive risk-taking by banks and

provide sound risk management mechanism, more transparency in business operations to remove or reduce information
asymmetries, greater competition to encourage efficiency, and strong capital-adequacy norms to ensure adequate capital
base. The institution of deposit insurance would be useful to provide safety nets to the depositors and encourage the
intermediation of deposits.
T. Singh / Journal of Policy Modeling 32 (2010) 231–253 249

in productivity. Since the increased efficient use of capital is economically equivalent to the
increased volume of capital, the acceleration of productivity would help reduce the strain on
saving and capital accumulation. The increase in productivity would also contribute to saving
through the increase in aggregate output.18
The saving of the corporate sector remained low and insufficient to finance its investment.
The rate of private corporate saving has been virtually stagnant and it displayed only sluggish
improvements (Fig. 4(4.2)). The main sources of funds for the corporate firms are the internal sav-
ing, institutional borrowings and the marketable debt and equity securities. The indirect finance
(bank and non-bank loans) is generally a more important source of external finance for businesses
than the direct finance (bonds and stocks) even in the developed economies, such as Canada,
Germany, Japan, and the United States (Mishkin, 2006). The reliance on bank-dominated indirect
finance becomes essentially inevitable in the developing economies with underdeveloped finan-
cial markets. The low collateral is predicted to raise the probability of default and impede the
ability of firms to raise required funds directly from the debt and equity markets, even with high
risk premium imbedded in the yield on corporate securities; net worth and internal saving serve
as the collaterals for the firms. The household saving in the form of the holdings of corporate
stocks has been abysmally low and such portfolio choices reflect the equity premium puzzle in
the financial markets.19 The lackadaisical participation of households in the stock market seems
to arise from several factors, such as the lack of information, aversion to risk, ‘lemons problem’
(Akerlof, 1970), and the inaccessibility to rural areas. The increase in corporate saving is required
for financing the investment requirements and strengthening the collaterals of the corporate firms.
The direct incentives such as tax benefits and interest subsidies for private investment could
be made covariant with the creation of capital stock, increases in expenditure on research and
development (R&D), improvements in productivity and the achievement of a benchmark growth.
The indirect incentives through the development of public infrastructure and creation of con-
ducive macroeconomic environment would help encourage market-driven and distortion-neutral
allocation of resources.

4.2. Public saving

The increase in public saving is more a direct measure to accelerate the domestic saving.
The effects of budget deficits on saving and the normatives of fiscal policy, however, remain
unresolved.20 The neoclassical paradigm postulates that the budget deficits raise the total life-
time consumption of the finitely-lived and farsighted individuals by shifting the taxes to future
generations, and lead to the decline in saving and rise in interest rates. The Keynesian paradigm
with consumer myopia contrarily predicts that the budget deficits have positive second order
effects on the saving and capital accumulation (Bernheim, 1989). The consumers are unable
to envision the future tax liabilities arising from the current deficits. The deficits stimulate the
aggregate demand and real output, and lead to the increase in both consumption and saving. The
Barro-Ricardian equivalence theorem with consumer foresight suggests that the budget deficits

18 For a review and discussion on the effects of productivity on output and economic growth, see Taylor (2008).
19 The equity premium puzzle would predict that the household should hold a predominant proportion of its wealth in
stocks, rather than bonds.
20 The government sector comprises the central government administration and its departmental undertakings; state gov-

ernments, union territories and their departmental undertakings; government non-departmental non-financial undertakings;
and the local authorities and their undertakings.
250 T. Singh / Journal of Policy Modeling 32 (2010) 231–253

and public debts have no effects on consumption, saving and interest rates (Barro, 1974, 1989).
The forward-looking optimizing agents anticipate the rise in future tax liabilities and equivalently
increase their saving to pay for future taxes in response to the rise in budget deficits and public
debts. The Say’s Law rendition of the neo-Ricardian equivalence would predict that the supply
of bonds to increase government borrowings and finance budget deficits creates its own demand
through the equivalent increase in private saving and, as a result, the prices of bonds do not fall and
implied rates of interest do not rise. The empirical support for the neo-Ricardian debt-neutrality
theorem is unresolved. The deviation from the neo-Ricardian equivalence implies that the ‘supply
of bonds does not create its own demand’ and the increases in deficits and the declines in public
saving are not perfectly offset by the increases in private saving. The imperfect offsets between
public and private saving underline the need to reduce the budget deficits (increase public saving)
to avoid the accumulations of public debts (and implied increases in interest rates) and refrain
from the plausible resorts to seigniorage.
The need to reduce the budget deficits is well-recognised in the IMF-supported structural
adjustment programs prescribed for the (developing) countries seeking IMF financial assistance
and undertaking economic reforms. The reduction in deficits and increase in public saving require
either the increase in tax revenues or the reduction in government spending or an optimal com-
bination of both. The response of tax revenues to taxes depends on whether the economy is
located on the upward (rise in tax raises tax revenue) or downward (rise in tax reduces tax rev-
enue) sloping segment of the Laffer curve or alternatively on the upward or backward sloping
segment of the labour-supply curve. The explicit tax-levies are welfare-superior to the implicit
inflation-tax levied, through seigniorage, on all the income strata regardless of the capacity to pay.
The developmental expenditures, such as the expenditures on R&D, human capital, provision of
public good and infrastructure, are essentially crucial and, as a result, the reduction in low-return
and non-developmental expenditures becomes an inevitable corollary. The profits of the public
enterprises also contribute to public revenue and saving. The potentials of these profits, however,
remain limited given the public utility characteristics and the welfare-orientation of most of the
enterprises. The contributions of public enterprises to public saving could be increased through
the improvements in the productivity of these enterprises. The major problem in the developing
economies is that it is difficult to raise adequate tax revenues (due to low incomes) to finance
higher developmental expenditures and expansionary fiscal policies, and this leads to the pub-
lic borrowing, which increases interest rates, and eventually to the seigniorage, which produces
inflation. The inflow of foreign saving relaxes the domestic saving constraints on investment and
enables the intertemporal smoothing of consumption.21 The foreign direct investment facilitates
the transfer of technology and could be encouraged to complement domestic investment and
accelerate the economic growth.

5. Conclusions

This study has examined the long-run effects of domestic saving on income and tested the
null of non-causality between saving and growth in India. The optimal single-equation and the
maximum-likelihood system estimates of the model consistently support the predictions of the

21 The foreign borrowing also leads to the deterioration of balance of payments and the rise in external debts. The

reversal of international capital flows could lead to the instability of exchange rate, erosion of foreign exchange reserves,
and plausibly the onset of financial crises.
T. Singh / Journal of Policy Modeling 32 (2010) 231–253 251

neoclassical exogenous and the post-neoclassical endogenous models of economic growth, and
suggest the significant long-run effects of saving on income. The innovation accounting shows
the bidirectional causality between saving and growth. The stylized evidence for the steady-state
effects of saving on income suggests the need to accelerate domestic saving to finance capital
accumulation and foster higher income and growth. Most of the saving comes from the surplus
household sector, and the deficit private corporate and public sectors draw on household saving
to meet their investment requirements and finance the resource gaps. A two-pronged approach
with the incentive-based measures to induce the motivation to save and the productivity-based
measures to increase income and strengthen the capacity to save, would be useful to generate higher
saving, and reinforce the acceleration of income and growth. The development of more efficient
financial infrastructure to enable the optimal diversification of saving portfolios, provision of
higher tax incentives on investment incomes to raise net rates of returns on financial instruments,
improvements in fiscal balance to increase public saving, and the increase in productivity to
accelerate growth and strengthen the capacity to save, would be useful to accelerate domestic
saving. The direct incentives such as tax benefits and investment subsidies for corporate investment
could be made covariant with the creation of capital stock, increases in expenditure on R&D,
improvements in productivity and the achievement of a benchmark growth. The indirect incentives
through the development of public infrastructure and the creation of conducive macroeconomic
environment would help encourage market-driven and distortion-neutral allocation of resources.
The foreign direct investment facilitates the transfer of technology and could be encouraged to
complement domestic investment and accelerate the economic growth.

Acknowledgements

I am grateful to the Editor and the anonymous Referees of the journal for very useful comments
and suggestions. I am, however, solely responsible for any error and omissions that may remain
in the paper.

References

Akerlof, G. A. (1970). The market for “Lemons”: Quality uncertainty and the market mechanism. The Quarterly Journal
of Economics, 84(August (3)), 488–500.
Attanasio, O. P., Picci, L., & Scorcu, A. E. (2000). Saving, growth, and investment: A macroeconomic analysis using a
panel of countries. The Review of Economics and Statistics, 82(May (2)), 182–211.
Banerjee, A., Dolado, J. J., Galbraith, J. W., & Hendry, D. F. (1993). Co-integration, error-correction, and the econometric
analysis of non-stationary data, advanced texts in econometrics. Oxford, UK: Oxford University Press.
Barro, R. J. (1974). Are government bonds net wealth? The Journal of Political Economy, 82(November–December (6)),
1095–1117.
Barro, R. J. (1989). The Ricardian approach to budget deficits. The Journal of Economic Perspectives, 3(Spring (2)),
37–54.
Barro, R. J., & Sala-i-Martin, X. (1995). Economic growth. New York: McGraw-Hill.
Bernheim, B. D. (1989). A neoclassical perspective on budget deficits. The Journal of Economic Perspectives, 3(Spring
(2)), 55–72.
Brown, R. L., Durbin, J., & Evans, J. M. (1975). Techniques for testing the constancy of regression relationships over
time. Journal of the Royal Statistical Society, Series B, 37(2), 149–192.
Carroll, C. D. (1997). Buffer-stock saving and the life cycle/permanent income hypothesis. The Quarterly Journal of
Economics, 112(February (1)), 1–55.
Carroll, C. D., Overland, J., & Weil, D. N. (2000). Saving and growth with habit formation. The American Economic
Review, 90(June (3)), 341–355.
252 T. Singh / Journal of Policy Modeling 32 (2010) 231–253

Carroll, C. D., & Weil, D. N. (1994). Saving and growth: A reinterpretation. In Carnegie-Rochester conference series on
public policy, vol. 40 June, (pp. 133–192).
Cass, D. (1965). Optimum growth in an aggregative model of capital accumulation. The Review of Economic Studies,
32(July (91)), 233–240.
Cesaratto, S. (1999). Savings and economic growth in neoclassical theory: Critical survey. Cambridge Journal of Eco-
nomics, 23(November (6)), 771–793.
Davidson, R., & MacKinnon, J. G. (1993). Estimation and inference in econometrics. New York: Oxford University Press.
Deaton, A. (1991). Saving and liquidity constraints. Econometrica, 59(September (5)), 1221–1248.
Deaton, A., & Paxson, C. (2000). Saving and growth among individuals and households. The Review of Economics and
Statistics, 82(May (2)), 212–225.
Dickey, D. A., & Fuller, W. A. (1981). Likelihood ratio statistics for autoregressive time series with a unit root. Econo-
metrica, 49(July (4)), 1057–1072.
Domar, E. D. (1946). Capital expansion, rate of growth, and employment. Econometrica, 14(April (2)), 137–147.
Elliott, G. (1999). Efficient tests for a unit root when the initial observation is drawn from its unconditional distribution.
International Economic Review, 40(August (3)), 767–784.
Elliott, G., Rothenberg, T. J., & Stock, J. H. (1996). Efficient tests for an autoregressive unit root. Econometrica, 64(July
(4)), 813–836.
Engle, R. F., & Granger, C. W. J. (1987). Cointegration and error correction: Representation, estimation, and testing.
Econometrica, 55(March (2)), 251–276.
Feldstein, M. (2007). Why is the dollar so high? Journal of Policy Modeling, 29(September–October (5)),
661–667.
Friedman, M. (1957). A theory of the consumption function. Princeton, NJ: Princeton University Press.
Hansen, B. E. (1992). Testing for parameter instability in linear models. Journal of Policy Modeling, 14(August (4)),
517–533.
Hansen, H., & Johansen, S. (1993). Recursive estimation in cointegrated VAR-models. In Manuscript. Institute of
Mathematical Sciences, University of Copenhagen.
Hansen, H., & Johansen, S. (1999). Some tests for parameter constancy in cointegrated VAR-models. The Econometrics
Journal, 2(December (2)), 306–333.
Hao, K., & Inder, B. (1996). Diagnostic test for structural change in cointegrated regression models. Economics Letters,
50(February (2)), 179–187.
Harrod, R. F. (1939). An essay in dynamic theory. The Economic Journal, 49(March (193)), 14–33.
Inder, B. (1993). Estimating long-run relationships in economics: A comparison of different approaches. Journal of
Econometrics, 57(May–June (1–3)), 53–68.
Johansen, S. (1991). Estimation and hypothesis testing of cointegration vectors in Gaussian vector autoregressive models.
Econometrica, 59(November (6)), 1551–1580.
Johansen, S. (2000). A Bartlett correction factor for tests on the cointegrating relations. Econometric Theory, 16(October
(5)), 740–778.
Johansen, S. (2002). A small sample correction for the test of cointegrating rank in the vector autoregressive model.
Econometrica, 70(September (5)), 1929–1961.
Koopmans, T. C. (1965). On the concept of optimal economic growth. In The econometric approach to development
planning. Amsterdam: North Holland.
Kwiatkowski, D., Phillips, P. C. B., Schmidt, P., & Shin, Y. (1992). Testing the null hypothesis of stationarity against
the alternative of a unit root: How sure are we that economic time series have a unit root? Journal of Econometrics,
54(October–December (1–3)), 159–178.
Loayza, N., Schmidt-Hebbel, K., & Serven, L. (2000). What drives private saving across the world? The Review of
Economics and Statistics, 82(May (2)), 165–181.
Lucas, R. E., Jr. (1988). On the mechanics of economic development. Journal of Monetary Economics, 22(July (1)), 3–42.
Lucas, R. E., Jr. (1990). Why does not capital flow from rich to poor countries? In The American economic review: Papers
and proceedings, vol. 80, no. 2 May, (pp. 92–96).
Mankiw, N. G. (2006). Reflections on the trade deficit and fiscal policy. Journal of Policy Modeling, 28(September (6)),
679–682.
Mankiw, N. G., Romer, D., & Weil, D. N. (1992). A contribution to the empirics of economic growth. The Quarterly
Journal of Economics, 107(May (2)), 407–437.
Mishkin, F. S. (2006). The economics of money, banking, and financial markets (8th ed.). New York: Addison Wesley.
Modigliani, F. (1988). The role of intergenerational transfers and life cycle saving in the accumulation of wealth. The
Journal of Economic Perspectives, 2(Spring (2)), 15–40.
T. Singh / Journal of Policy Modeling 32 (2010) 231–253 253

Modigliani, F., & Ando, A. (1957). Tests of the life cycle hypothesis of savings: Comments and suggestions. Bulletin of
the Oxford University Institute of Economics & Statistics, 99–124.
Modigliani, F., & Brumberg, R. (1954). Utility analysis and the consumption function: An interpretation of cross-section
data. In K. K. Kurihara (Ed.), Post-Keynesian economics (pp. 388–436). New Brunswick: Rutgers University Press.
Newey, W. K., & West, K. D. (1987). A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent
covariance matrix. Econometrica, 55(May (3)), 703–708.
Ng, S., & Perron, P. (2001). Lag length selection and the construction of unit root tests with good size and power.
Econometrica, 69(November (6)), 1519–1554.
Pesaran, M. H., Shin, Y., & Smith, R. J. (2001). Bounds testing approaches to the analysis of level relationships. Journal
of Applied Econometrics, 16(May/June (3)), 289–326.
Phillips, P. C. B. (1991). Optimal inference in cointegrated systems. Econometrica, 59(March (2)), 283–306.
Phillips, P. C. B., & Hansen, B. E. (1990). Statistical inference in instrumental variables regression with I(1) processes.
The Review of Economic Studies, 57(January (189)), 99–125.
Phillips, P. C. B., & Loretan, M. (1991). Estimating long-run economic equilibria. The Review of Economic Studies,
58(May (195)), 407–436.
Phillips, P. C. B., & Ouliaris, S. (1990). Asymptotic properties of residual based tests for cointegration. Econometrica,
58(January (1)), 165–193.
Phillips, P. C. B., & Perron, P. (1988). Testing for a unit root in time series regression. Biometrika, 75(June (2)), 335–346.
Ploberger, W., & Krämer, W. (1992). The CUSUM test with OLS residuals. Econometrica, 60(March (2)), 271–285.
Ramsey, F. P. (1928). A mathematical theory of saving. The Economic Journal, 38(December (152)), 543–559.
Rebelo, S. (1991). Long-run policy analysis and long-run growth. Journal of Political Economy, 99(June (3)), 500–521.
Rodrik, D. (2000). Saving transitions. The World Bank Economic Review, 14(3), 481–507.
Romer, P. M. (1986). Increasing returns and long-run growth. Journal of Political Economy, 94(October (5)), 1002–1037.
Romer, P. M. (1987). Growth based on increasing returns due to specialization. The American Economic Review, 77(May
(2)), 56–62.
Romer, D. (2006). Advanced macroeconomics (3rd ed.). New York: McGraw-Hill.
Saikkonen, P. (1991). Asymptotically efficient estimation of cointegration regressions. Econometric Theory, 7(March
(1)), 1–21.
Sarno, L., & Taylor, M. P. (1998). Savings-investment correlations: Transitory versus permanent. The Manchester School,
66(Suppl.), 17–38.
Sawa, T. (1978). Information criteria for discriminating among alternative regression models. Econometrica, 46(November
(6)), 1273–1291.
Schmidt-Hebbel, K., Servén, L., & Solimano, A. (1996). Saving and investment: Paradigms, puzzles, policies. The World
Bank Research Observer, 11(February (1)), 87–117.
Schwert, G. W. (1989). Tests for unit roots: A Monte Carlo investigation. Journal of Business & Economic Statistics,
7(April (2)), 5–17.
Sims, C. A. (1980). Macroeconomics and reality. Econometrica, 48(January (1)), 1–48.
Solow, R. M. (1956). A contribution to the theory of economic growth. The Quarterly Journal of Economics, 70(February
(1)), 65–94.
Stock, J. H., & Watson, M. W. (1989). Interpreting the evidence on money-income causality. Journal of Econometrics,
40(January (1)), 161–181.
Stock, J. H., & Watson, M. W. (1993). A simple estimator of cointegrating vectors in higher order integrated systems.
Econometrica, 61(July (4)), 783–820.
Swan, T. W. (1956). Economic growth and capital accumulation. The Economic Record, 32(November), 334–361.
Taylor, J. B. (2008). A review of the productivity resurgence. Journal of Policy Modeling, 30(July–August (4)), 619–626.
Xiao, Z. (1999). A residual based test for the null hypothesis of cointegration. Economics Letters, 64(August (2)), 133–141.
Xiao, Z., & Phillips, P. C. B. (2002). A CUSUM test for cointegration using regression residuals. Journal of Econometrics,
108(May (1)), 43–61.

You might also like