How Stable is the Savings-led Growth Hypothesis in Malaysia? The Bootstrap Simulation and Recursive Causality Tests

Abstract

The purpose of this study is to empirically investigate the vindication of the savings-led growth hypothesis for the Malaysian economy with the long-run TYDL version of the Granger causality test—Toda and Yamamoto (1995) and Dolado and Lütkepohl (1996). This study used the quarterly sample from 1970:Q1 to 2008:Q4. The recursive regression procedure will also incorporate into the TYDL causality test to measure the stability of the savings-led growth hypothesis in the long-run. Our empirical results support that the savings-led growth hypothesis is a long-run phenomenon and stable over time. Therefore, the Malaysian dataset supports the endogenous growth theory.

JEL Classification: C22, E21, O16

Keywords

Causality Malaysia Recursive regression Savings-led growth Stability

1. INTRODUCTION

Are high-savings countries growing faster? Logically, to achieve economic growth, one must invest. To encourage investment, one should save. Therefore, savings may play an important role in the process of economic growth and development. The role of savings in economic growth has long been debated and many empirical studies have been conducted to examine the vindication of the savings-led growth hypothesis. In general, two famous groups of economic thought explained the savings-led growth hypothesis. Harrod (1939) and Domar (1946) were the first group of economists that systematically modelled the role of savings in economic growth. They claimed that the speed of economic growth depends on the ability to save, as a high savings rate will increase the rate of investment and hence trigger economic growth. Solow (1956) added that although savings was an important source for economic growth through its impact on capital formation, it does not affect economic growth in the long-run because of the assumption of diminishing returns to scale and because technological progress is assumed to be exogenous. Therefore, the neoclassical growth economists believed that savings-led growth is just a transitory or short-run phenomenon rather than an endless process. Ironically, the endogenous growth theory proposed by Romer (1986) argued that technological progress is determined endogenously and the production function is subject to increasing returns to scale. Moreover, savings are required to finance technological progress. Therefore, the endogenous or new growth theory articulated that savings-led growth is a long-run phenomenon.

Given the policy relevance and the controversy among growth theorists, many researchers have empirically investigated the validity of the savings-led growth hypothesis in developed and developing economies. For this reason, it is implausible for the present study to comprehensively review all the relevant studies within the ambit of this article. The objective of this study is to review some empirical studies pertaining to the causal relationship between savings and economic growth in the Malaysian economy. A summary of the preceding studies on the subject is reported in Table 1. In general, Table 1 shows that existing research efforts failed to provide clear evidence for the savings-led growth hypothesis. The vindication of the savings-led growth hypothesis is of concern because it is directly related to the formulation and implementation of appropriate growth policy. If savings-led growth is a long-run phenomenon as indicated by the endogenous growth theory, enhanced savings may be a potential long-run stimulus for economic growth policy. Otherwise, alternative growth policies should be implemented to achieve sustainable economic growth in the long-run.

Table 1
Summary of Selected Empirical Studies on the Relationship between Savings and Economic Growth in Malaysia

No. Authors Research Period Econometric Methods
Empirical Results

Cointegration Causality Cointegrated Causality

$S G$ $G S$ $S G$ $S G$

1. World Bank (1993) 1955–88 – Granger (1969) – VAR – ✓

2. Agrawal (2001) 1960–94 – Granger (1969) – VAR – ✓

3. Baharumshah et al. (2003) 1960–97 Johansen and Juselius (1990) Granger (1988) – VECM Yes ✓

4. Baharumshah and Thanoon (2003) 1960–2000 Johansen and Juselius (1990) Toda and Yamamoto (1995) – Augmented VAR Yes ✓

4. Boo and Normee (2004) 1970:Q1– 1999:Q4 Johansen and Juselius (1990) Granger (1988) – VECM Yes ✓

5. Tang (2008a) 1970–2007 Pesaran et al. (2001) Toda and Yamamoto (1995) – Augmented VAR Yes ✓

6. Tang (2009) 1991:Q1– 2006:Q3 – Granger (1969) – VAR – ✓

Geweke et al. (1983) – Modified Sims test ✓

Hsiao (1981) ✓

Toda and Yamamoto (1995) – Augmented VAR ✓

Holmes and Hutton (1990) – Multiple Rank F-test ✓

7. Tang and Chua (2009) 1991:Q1– 2006:Q2 Bierens (1997) Holmes and Hutton (1990) – Multiple Rank F-test Yes ✓

No.	Authors	Research Period	Econometric Methods	Empirical Results
1.	World Bank (1993)	1955–88	–	Granger (1969) – VAR	–			✓
2.	Agrawal (2001)	1960–94	–	Granger (1969) – VAR	–	✓
3.	Baharumshah et al. (2003)	1960–97	Johansen and Juselius (1990)	Granger (1988) – VECM	Yes				✓
4.	Baharumshah and Thanoon (2003)	1960–2000	Johansen and Juselius (1990)	Toda and Yamamoto (1995) – Augmented VAR	Yes			✓
4.	Boo and Normee (2004)	1970:Q1– 1999:Q4	Johansen and Juselius (1990)	Granger (1988) – VECM	Yes		✓
5.	Tang (2008a)	1970–2007	Pesaran et al. (2001)	Toda and Yamamoto (1995) – Augmented VAR	Yes			✓
6.	Tang (2009)	1991:Q1– 2006:Q3	–	Granger (1969) – VAR	–			✓
				Geweke et al. (1983) – Modified Sims test				✓
				Hsiao (1981)				✓
				Toda and Yamamoto (1995) – Augmented VAR				✓
				Holmes and Hutton (1990) – Multiple Rank F-test				✓
7.	Tang and Chua (2009)	1991:Q1– 2006:Q2	Bierens (1997)	Holmes and Hutton (1990) – Multiple Rank F-test	Yes			✓

Notes: (1) ✓ indicates the presence of causality; (2) ‘Yes’ represents the variables are cointegrated; (3) represents uni-directional causality; (4) represents bilateral/bi-directional causality; (5) represents none/neutral causality.

Using annual data from 1955 to 1988, World Bank (1993) attempted to investigate the causal relationship between savings and economic growth within the bi-variate vector autoregression (VAR) framework. The study found that savings and economic growth Granger-cause each other. In another study, Agrawal (2001) performed an empirical analysis on seven Asian economies with the residuals-based cointegration and Granger causality tests. In the case of Malaysia, the study found that savings and its determinants are stationary at different orders. Therefore, the residuals-based cointegration test cannot be used to examine the presence of a long-run equilibrium relationship. For this reason, the above authors assumed that savings and its determinants for Malaysia are not cointegrated. Therefore, the Granger causality test based on the first difference VAR framework was used to examine the causality between savings and economic growth. The authors found uni-directional causality running from savings to economic growth. On the contrary, Baharumshah et al. (2003) and Boo and Normee (2004) used the multivariate Johansen and Juselius (1990) cointegration and Granger causality tests to examine the relationship between savings and its determinants. They found that savings and its determinants for Malaysia are moving together in the long-run; therefore, the Granger causality test within the vector error-correction model (VECM) was used to examine the direction of causality between savings and economic growth.

Contrary to earlier studies and also to the growth theories, Baharumshah et al. (2003) found no causality between savings and economic growth, while Boo and Normee (2004) observed that economic growth Granger-causes savings in Malaysia instead of the opposite. However, Baharumshah and Thanoon (2003) re-investigated the long-run causal relationship between savings and economic growth in Malaysia using the Toda and Yamamoto (1995) causality approach. They found that savings and economic growth in Malaysia is a bi-directional causality in the long-run. Furthermore, Tang (2008a) proposed to incorporate the modified dependency ratio into the savings–growth relationship for Malaysia. In addition to that, the author also used a relatively new cointegration test developed by Pesaran et al. (2001)—the bounds-testing approach to examine the presence of a long-run relationship between savings and its determinants. The TYDL developed by Toda and Yamamoto (1995) and Dolado and Lütkepohl (1996) was used within the augmented-VAR system to verify the causal relationship between savings and economic growth in Malaysia. The authors observed that savings and its determinants are cointegrated and the TYDL Granger causality test implied bilateral causality between savings and economic growth over the sample period of 1970 to 2006. The causality result is consistent with the finding of the World Bank (1993) and Baharumshah and Thanoon (2003), but contradicted earlier studies. Recently, Tang (2009) attempted to investigate whether the causal inference between savings and economic growth in Malaysia is sensitive to the particular causality tests employed to ascertain the causal relationship. To achieve the objective of his study, the author employed five different causality techniques, namely, the Granger (1969), Geweke et al. (1983), Hsiao (1981), Toda and Yamamoto (1995) and Holmes and Hutton (1990) causality approaches. Interestingly, the author found bilateral causality between savings and economic growth regardless of the causality technique employed. The author concluded that causality methods do not influence the causality results. Similarly, using quarterly data from 1991:Q1 to 2006:Q3, Tang and Chua (2009) also detected bilateral causality between savings and economic growth in Malaysia with the Holmes and Hutton (1990) causality test.

Motivated by the paradoxical empirical evidence presented above, this study attempts to re-investigate the existence of the savings-led growth hypothesis for the Malaysian economy from 1970:Q1 to 2008:Q4. It differs from the earlier studies in at least three dimensions. First, none of the Malaysian studies has taken into consideration the effect of structural break(s) in the unit root tests. Perron (1989) pointed out that the conventional unit root tests such as Augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) may be biased when the time series pertain to structural change. Therefore, apart from using the conventional unit root tests, we also apply the Lagrange Multiplier (LM) unit root tests with one and two structural breaks developed by Lee and Strazicich (2003, 2004) to fill the detected gap. This is because the ADF-type structural break(s) unit root tests (e.g., Lumsdaine and Papell, 1997; Zivot and Andrews, 1992) tend to select the incorrect breakpoints (see Lee and Strazicich, 2001). Second, we employ the TYDL Granger causality test within the augmented-VAR system to ascertain the direction of causality in the long-run. Nevertheless, Hacker and Hatemi-J (2006) found that the MWALD test statistics bias rejected the null hypothesis of non-Granger causality. Therefore, a correction has to be done to avoid this bias. Following Hacker and Hatemi-J (2006), we employ the leveraged bootstrap simulation approach to yield robust causality test results. Finally, this study contributes to the empirical literature by testing the persistency or the stability of the savings-led growth hypothesis by incorporating the recursive regression procedure into the TYDL Granger causality test as suggested by Tang (2008b). To the knowledge of the author, no study has analysed the stability or persistency of the savings-led growth hypothesis using the recursive causality procedure. The stability of the savings-led growth hypothesis is of concern because it is an important criterion for modelling appropriate growth policies. If the causal relationship is unstable, the macroeconomic policy initiatives that encourage savings may be less effective in stimulating economic growth in the long-run.

The remainder of this study is organised as follows. In the next section, we provide an overview of savings in Malaysia. In Section 3, we will briefly discuss the econometric techniques used in this study. Section 4 reports the source of data and empirical findings of this study. Finally, the concluding remarks of this study will be presented in Section 5.

2. AN OVERVIEW OF SAVINGS IN MALAYSIA

Malaysia is a small open economy located in the Southeast Asian region. According to Tang (2008a), Malaysia has the second highest savings rate among countries in the ASEAN region. Figure 1 provides the savings rates (i.e., domestic savings, private savings and public savings) as a percentage of GDP in Malaysia from 1970 to 2008. It can be seen that savings rates in Malaysia increase over the analysis period. On average, domestic savings, private savings and public savings are approximately 35 per cent, 19 per cent and 11 per cent of GDP, respectively. In addition, domestic savings reached its peak at 42 per cent of GDP in 1996. However, the savings rate dropped to 31 per cent in 1997 due to the Asian financial crisis that hit the Malaysian economy. The crisis triggered a massive outflow of foreign capital and also caused the depreciation of Malaysian currency. Hence, savings rates in Malaysia dropped drastically from 42 per cent to 31 per cent of GDP.

Based upon the plots, we observe that the behaviour of domestic savings in Malaysia is more likely to be correlated with private savings than public savings, especially before 1990. Specifically, private savings in Malaysia contributes around 15 to 20 per cent of GDP. However, public savings contributed less than 10 per cent of GDP in Malaysia, especially during the 1970s and 1980s. Nonetheless, since the 1990s, the contribution of public savings increased to approximately 15 per cent of GDP in Malaysia. With respect to the increase in public savings, Ang (2009) noted that such an increase is achieved due to cutting of public consumption rather than increase in taxes in Malaysia. In summary, both private and public savings are the main sources of savings in Malaysia and they are equally important in contributing to total domestic savings in the economy.

Figure 1

Source: Author’s calculation.

3. ECONOMETRIC TECHNIQUES

3.1 LM Unit Root Tests with Structural Break

In this section, we discuss the LM unit root test with one and two structural breaks introduced by Lee and Strazicich (2003, 2004). The Model C and Model CC for one and two structural breaks LM unit root tests will be estimated. This is because Sen (2003) noted that Model C and Model CC typically performed better than the other models. Therefore, we estimate the following model to determine the order of integration with one and two structural breaks.

z_{t}^{} W_{t} {\overset{}{S}}_{t 1}_{i 1}^{k}_{i} {\overset{}{S}}_{t i}_{t}

(1)

Here, is the first difference operator and the residuals $_{t}$ are assumed to be spherically distributed and white noise. ${\overset{}{S}}_{t 1} z_{t} {\overset{}{}}_{x} \overset{}{} W_{t}$ , $\overset{}{}$ are the estimated coefficients in the regression of $z_{t}$ on $W_{t}$ . In addition, ${\overset{}{}}_{x} z_{t} \overset{}{} W_{t}$ and $W_{t}$ is a set of exogenous variables. As in the Augmented Dickey-Fuller (ADF) unit root tests, the augmented term ${\overset{}{S}}_{t i}$ is accommodated into the model to remove the autocorrelation problem. For the one-break case (i.e., Model C), $W_{t} {[1, t, D_{1 t}, {DT}_{1 t}]}^{}$ , while for the two breaks case (i.e., Model CC), $W_{t} {[1, t, D_{1 t}, D_{2 t}, {DT}_{1 t}, {DT}_{2 t}]}^{}$ , where:

D_{1 t} \{\begin{matrix} 1 & if t TB 1 \\ 0 & Oth erwise \end{matrix}

{DT}_{1 t} \{\begin{matrix} t TB 1 if t TB 1 \\ 0 Otherwise \end{matrix}

D_{2 t} \{\begin{matrix} 1 & if t TB 2 \\ 0 & Oth erwise \end{matrix}

{DT}_{2 t} \{\begin{matrix} t TB 2 if t TB 2 \\ 0 Otherwise \end{matrix}

TB1 and $TB 2$ are the time periods of the potential breaks and $^{} (_{1},_{2},_{3}) .$ Finally, the potential breakpoints, $TB 1$ and $TB 2$ , are chosen where the $t$ -statistics is minimised. The GAUSS programming codes are employed to perform the LM unit root tests with one and two structural breaks.

3.2 Granger Causality Test

This study uses the Granger causality test proposed by Toda and Yamamoto (1995) and Dolado and Lütkepohl (1996)—TYDL to verify the long-run causality direction between the candidate variables $z_{t} {[\ln Y_{t}, \ln S_{t}, \ln {FCI}_{t}, \ln {FD}_{t}]}^{}$ . Zapata and Rambaldi (1997) explored the performance of Wald and likelihood ratio tests for the Granger non-causality test in the cointegrated systems. The Monte Carlo results indicate that the likelihood ratio and Wald test are very sensitive to the specification of the short-run dynamics. Indeed, Yamada and Toda (1998) conducted a Monte Carlo simulation experiment to investigate the performance of three causality procedures. The simulation results indicate that among the three causality procedures, TYDL is the most stable approach. Furthermore, the Error-Correction Modelling (ECM) and Fully-Modified VAR (FM-VAR) causality approaches tend to have larger size distortions than the TYDL approach. To implement the TYDL Granger causality test, we estimate the following augmented-VAR framework with the $p (k d_{\max})$ lag length.

z_{t} [\begin{matrix} z_{t}^{1} \\ z_{t}^{2} \\ z_{t}^{3} \\ z_{t}^{4} \end{matrix}] [\begin{matrix} b_{1} \\ b_{2} \\ b_{3} \\ b_{4} \end{matrix}] [\begin{matrix} B_{11, 1} & B_{12, 1} & B_{13, 1} & B_{14, 1} \\ B_{21, 1} & B_{22, 1} & B_{23, 1} & B_{24, 1} \\ B_{31, 1} & B_{32, 1} & B_{33, 1} & B_{34, 1} \\ B_{41, 1} & B_{42, 1} & B_{43, 1} & B_{44, 1} \end{matrix}] [\begin{matrix} z_{t 1}^{1} \\ z_{t 1}^{2} \\ z_{t 1}^{3} \\ z_{t 1}^{4} \end{matrix}] [\begin{matrix} B_{11, k} & B_{12, k} & B_{13, k} & B_{14, k} \\ B_{21, k} & B_{22, k} & B_{23, k} & B_{24, k} \\ B_{31, k} & B_{32, k} & B_{33, k} & B_{34, k} \\ B_{41, k} & B_{42, k} & B_{43, k} & B_{44, k} \end{matrix}]

[\begin{matrix} z_{t k}^{1} \\ z_{t k}^{2} \\ z_{t k}^{3} \\ z_{t k}^{4} \end{matrix}] [\begin{matrix} B_{11, p} & B_{12, p} & B_{13, p} & B_{14, p} \\ B_{21, p} & B_{22, p} & B_{23, p} & B_{24, p} \\ B_{31, p} & B_{32, p} & B_{33, p} & B_{34, p} \\ B_{41, p} & B_{42, p} & B_{43, p} & B_{44, p} \end{matrix}] [\begin{matrix} z_{t p}^{1} \\ z_{t p}^{2} \\ z_{t p}^{3} \\ z_{t p}^{4} \end{matrix}] [\begin{matrix} e_{1 t} \\ e_{2 t} \\ e_{3 t} \\ e_{4 t} \end{matrix}]

(2)

Where k is the optimal lag length for the VAR system determined by AIC. Even though Toda and Yamamoto (1995) noted that the maximum order of integration for the economics series are commonly at either I(1) and I(2) process, according to simulation results produced by Dolado and Lütkepohl (1996) the performance of $d_{\max} 1$ is better than any other order of $d_{\max}$ . Thus, the $d_{\max} 1$ is preferable in this study.

From Equation (2), the null hypothesis of $z_{t}^{2}$ does not Granger-cause $z_{t}^{1}$ if $B_{12, k} 0_{k}$ cannot be rejected, if $B_{13, k} 0_{k}$ is rejected, the $z_{t}^{3}$ Granger-cause $z_{t}^{1}$ , and so on. Nevertheless, the following notations for the general form of the VAR system suggested by Lütkepohl (2005) was followed before we define the TYDL version of the Granger causality test—MWLAD:

\begin{matrix} Z : (z_{1},, z_{T}) (4 T) matrix, \\ D : (b, B_{1},, B_{p}) (4 (4 p 1)) matrix, \\ Y_{t} : [\begin{matrix} 1 \\ z_{t} \\ z_{t 1} \\ z_{t p 1} \end{matrix}] ((4 p 1) 1) matrix, \\ Y : (Y_{0},, Y_{T 1}) ((4 p 1) T) matrix, \\ and \\ : (e_{1},, e_{T}), (4 T) matrix \end{matrix}

From the above notation, the augmented-VAR(p) model together with the constant term (b) can be expressed compactly as below:

Z DY

(3)

The estimated $(4 T)$ matrix of the residuals from the unrestricted and restricted regression model (3) can be denoted as $({\overset{}{}}_{UR})$ and $({\overset{}{}}_{R})$ , respectively. Then the variance–covariance matrix of the estimated residuals are generated by $S_{UR} {\overset{}{}}^{}_{UR} {\overset{}{}}_{UR}$ and $S_{R} {\overset{}{}}^{}_{R} {\overset{}{}}_{R}$ . Eventually, the MWALD test statistics for Granger causality can be computed by the following equation:

MWALD (T p) (\frac{S_{R} S_{UR}}{S_{UR}})

(4)

where T is the total number of observation and p is the lag length in the augmented-VAR system. Nevertheless, it is noteworthy to point out here that the additional lag order, that is, $d_{\max}$ in Equation (2), is unrestricted because the inclusion of this additional lag was to ensure that the asymptotic $^{2}$ -distribution critical values can be used when the Granger causality test is conducted with the non-stationary variables (Toda and Yamamoto, 1995).

Next, the residuals-based bootstrapping approach is adopted to compute the critical values for the MWALD test with the empirical distribution. Following Davidson and MacKinnon (2004), 1000 times of bootstrapping will be performed to yield the leveraged bootstrap critical values. As a common interpretation for statistical inferences, the null hypothesis of non-Granger causality is rejected if the computed MWALD test statistic is greater than the bootstrapped critical values.¹

3.3 Recursive Regressions-based TYDL Causality Test

This section will briefly discuss the estimation procedure for the recursive regression-based TYDL causality test. The idea of the stability of the causal relationship was raised by Tang (2008b). His study noted that the causal relationship may be not be stable owing to frequent changes in the global economy and the political environment. In this regard, Tang (2008b) proposed recursive regression-based causality tests to examine the stability of the causal relationship. The major advantage of this recursive causality test is that it can be used to examine the volatility or fluctuation of the causal relationship over time. In other words, we can visually inspect the stability of the causal relationship over the analysis period. The recursive regression-based TYDL causality test can be conducted by first estimating the augmented-VAR system with the initial sub-sample of T observations. Then, a new observation will be added at the end of the sub-sample without deleting the first observation (i.e., T + 1). The causality test will be performed for each sub-sample. For interpretation, the computed MWALD statistics will be normalised by the 10 per cent $^{2}$ critical values. Therefore, the null hypothesis of non-Granger causality can be rejected if the normalised MWALD statistics exceed unity.

4. EMPIRICAL RESULTS

4.1 Source of Data

This study examines the savings-led growth theories for Malaysia within a multivariate framework. It uses the Gandolfo (1981) interpolated quarterly data from 1970:Q1 to 2008:Q4 for real gross domestic product (GDP), real gross domestic savings (GDS), real foreign capital inflows (FCI), and a financial development indicator proxy of the ratio of M2 money to GDP. The interpolated data was used in this study to increase the statistical test power and avoid the size distortion problem (Zhou, 2001). The raw data for interpolation were extracted from the World Bank’s World Development Indicators (WDI) and Bank Negara Malaysia’s Monthly Statistical Bulletin. All data were transformed into the natural logarithm form. The analyses in this study were conducted with GAUSS 9.0 and Eviews 6.0 software.

4.2 Unit Root Tests Results

According to time series econometric literatures, the regression results may be spurious if the variables are non-stationary (Granger and Newbold, 1974; Phillips, 1986). Thus, it is interesting to examine the order of integration for each variable to avoid spurious correlation problems. Although the TYDL version of the Granger causality test does not require pre-testing of the unit root, the unit root test result can shed some light on the order of $d_{\max}$ in the augmented-VAR system.

To ascertain the order of integration, we begin by applying the conventional ADF and PP unit root tests. Interestingly, the conventional unit root test results suggest that the variables under investigation $[\ln Y_{t}, \ln S_{t}, \ln {FCI}_{t}, \ln {FD}_{t}]$ are integrated of order one I(1) process. To save space, these results are not reported here. Since one of the main concerns in this study is the implications of structural break(s) on unit roots, we conducted the LM unit root tests for one and two structural breaks. The LM unit root test results are reported in Table 2. At the 5 per cent significance level, both LM unit root test statistics failed to reject the null hypothesis of a unit root at level for all the variables. Therefore, we concluded that the variables are integrated of order one, I(1) process. This result corroborates Nelson and Plosser’s (1982) assertion that most of the macroeconomic variables are non-stationary at level, but are stationary after first differences. Moreover, the evidence of I(1) process also affirmed that the appropriate order of $d_{\max}$ for the TYDL Granger causality test is one.

Table 2
The Results of the LM Unit Root Tests with Structural Breaks(s)

Panel A: Univariate LM test for unit root with one structural break

lnY_t lnS_t lnFCI_t lnFD_t

TB1 Jun-1981 Dec-1996 Dec-2004 Dec-1978

S_t– ₁ –3.632 –2.853 –4.430 –3.354

Lag length 9 4 12 12

Critical values

1% –5.15 –5.11 –5.11 –5.07

5% –4.45 –4.51 –4.51 –4.47

Panel B: Univariate LM test for unit root with two structural breaks

lnY_t
lnS_t
lnFCI_t
lnFD_t

TB1 Sep-1984 Dec-1980 Dec-1987 Sep-1980

TB2 Dec-1997 Jun-1996 Dec-2001 Sep-1993

$S_{t 1}$ –5.199 –4.673 –5.272 –4.000

Lag length 8 9 12 12

Critical values

1% –6.45 –6.45 –6.42 –6.45

5% –5.67 –5.67 –5.65 –5.67

Panel A: Univariate LM test for unit root with one structural break
TB1	Jun-1981	Dec-1996	Dec-2004	Dec-1978
S_t– ₁	–3.632	–2.853	–4.430	–3.354
Lag length	9	4	12	12
Critical values
1%	–5.15	–5.11	–5.11	–5.07
5%	–4.45	–4.51	–4.51	–4.47
Panel B: Univariate LM test for unit root with two structural breaks
	lnY_t	lnS_t	lnFCI_t	lnFD_t
TB1	Sep-1984	Dec-1980	Dec-1987	Sep-1980
TB2	Dec-1997	Jun-1996	Dec-2001	Sep-1993
$S_{t 1}$	–5.199	–4.673	–5.272	–4.000
Lag length	8	9	12	12
Critical values
1%	–6.45	–6.45	–6.42	–6.45
5%	–5.67	–5.67	–5.65	–5.67

Note: The GAUSS programming codes provided by Dr Junsoo Lee have been used to perform the above LM tests for unit root with one and two structural breaks, respectively.

4.3 TYDL Granger Causality Test Results

In this section, the long-run causal relationships between the candidate variables $[\ln Y_{t}, \ln S_{t}, \ln {FCI}_{t}, \ln {FD}_{t}]$ are examined by using the TYDL Granger causality test. The results of the TYDL Granger causality test statistics together with the leveraged bootstrap critical values are delineated in Table 3. According to the estimation results, we found that at the 10 per cent significance level the causality results could be summarised as follows: (i) causality runs from savings, foreign capital inflows and financial development to economic growth; (ii) causality runs from economic growth and foreign capital inflows to savings; (iii) causality runs from economic growth and savings to foreign capital inflows; (iv) causality runs from economic growth and foreign capital inflows to financial development.

As opposed to the neoclassical growth theory, our empirical results indicate that the savings-led growth hypothesis is a long-run phenomenon. In other words, the Malaysian dataset supports the endogenous growth theory. This is not an unexpected result as Malaysia is the second-highest savings economy among the Association of Southeast Asia Nations (ASEAN), which has attracted large amounts of foreign direct investment that will simultaneously enhance technological progress (see Aghion et al., 2009). Therefore, the discovery of a savings-led growth hypothesis in the long-run is plausible. Indeed, this is consistent with our causality result that savings Granger-causes foreign capital inflows. Towards a knowledge-based economy (K-economy), expenditure on education and health as a proportion of total government development expenditure in Malaysia (i.e., human capital investment) increased steadily during the period from 1970 to 2007.

Table 3
Results of the TYDL Granger Causality Tests

Null Hypothesis Estimated MWALD Tests Bootstrapped Critical Values

1 per cent 5 per cent 10 per cent

$\ln S_{t} \ln Y_{t}$ 20.934*** 19.401 15.383 13.063

$\ln {FCI}_{t} \ln Y_{t}$ 14.983* 27.632 19.097 14.510

$\ln {FD}_{t} \ln Y_{t}$ 83.272* 15.910 10.607 8.486

$\ln Y_{t} \ln S_{t}$ 27.064* 23.118 16.134 13.480

$\ln {FCI}_{t} \ln S_{t}$ 26.454* 20.644 14.684 11.417

$\ln {FD}_{t} \ln S_{t}$ 8.316 20.748 15.724 13.187

$\ln Y_{t} \ln {FCI}_{t}$ 29.865 38.120 26.940 22.602

$\ln S_{t} \ln {FCI}_{t}$ 6.497* 12.969 8.570 6.393

$\ln {FD}_{t} \ln {FCI}_{t}$ 0.044 6.818 3.938 2.667

$\ln Y_{t} \ln {FD}_{t}$ 94.450* 14.909 9.653 7.987

$\ln S_{t} \ln {FD}_{t}$ 12.037 21.132 15.948 12.722

$\ln {FCI}_{t} \ln {FD}_{t}$ 28.539 32.739 24.189 19.594

Null Hypothesis	Estimated MWALD Tests	Bootstrapped Critical Values
$\ln S_{t} \ln Y_{t}$	20.934***	19.401	15.383	13.063
$\ln {FCI}_{t} \ln Y_{t}$	14.983*	27.632	19.097	14.510
$\ln {FD}_{t} \ln Y_{t}$	83.272***	15.910	10.607	8.486
$\ln Y_{t} \ln S_{t}$	27.064***	23.118	16.134	13.480
$\ln {FCI}_{t} \ln S_{t}$	26.454***	20.644	14.684	11.417
$\ln {FD}_{t} \ln S_{t}$	8.316	20.748	15.724	13.187
$\ln Y_{t} \ln {FCI}_{t}$	29.865**	38.120	26.940	22.602
$\ln S_{t} \ln {FCI}_{t}$	6.497*	12.969	8.570	6.393
$\ln {FD}_{t} \ln {FCI}_{t}$	0.044	6.818	3.938	2.667
$\ln Y_{t} \ln {FD}_{t}$	94.450***	14.909	9.653	7.987
$\ln S_{t} \ln {FD}_{t}$	12.037	21.132	15.948	12.722
$\ln {FCI}_{t} \ln {FD}_{t}$	28.539**	32.739	24.189	19.594

Note: ***, ** and * denotes statistical significance at the 1, 5 and 10 per cent levels, respectively.

During the decade 1981–90, the proportion of total expenditure on education increased from 7 per cent to 15 per cent, while that on health rose from 1 per cent to 4 per cent. The proportion of expenditure on education further increased from 13 per cent in 1991 to a peak of 35 per cent in 2002. This implied that the Malaysian government has recognised the importance of human capital investment in stimulating economic growth and development. According to Romer (1986), an increase in the investment in human capital will enhance the labour force and capital productivity and efficiency, because human capital is not subject to diminishing returns. Meanwhile, savings is a prerequisite for human capital investment; hence savings Granger-causes economic growth in the long-run.

Although we have observed strong evidence to support vindication of the savings-led growth hypothesis in the long-run, the causal relationship may not be stable over time, so we also examine the stability of the savings-led growth hypothesis with the recursive regression-based TYDL Granger causality tests. In order to implement the recursive regression-based causality test, we have to decide the initial sample size for estimation. To serve the purpose of long-run causality and also to avoid the size distortion problem, we begin the estimation with a ten-year sample (i.e., 40 observations). For interpretation, the computed MWALD statistics for each sample (i.e., T + 1) will be normalised by the 10 per cent $^{2}$ critical values. Therefore, the null hypothesis that savings does not Granger-causes economic growth is rejected if the normalised MWALD statistics is greater than unity. The plots of the normalised MWALD causality test statistics are presented in Figure 2. From visual inspection, we can see that the normalised MWALD statistics fluctuate widely, but the causal inference for savings-led growth is stable over the analysis period. As a result, our empirical study again rejects the neoclassical growth theory that savings does not affect economic growth in the long-run. Therefore, high-savings economies should grow faster and policy initiatives to enhance savings could be a potential long-run sustainable economic growth strategy for the Malaysian economy.

Figure 2

Source: Author’s calculation.

5 CONCLUDING REMARKS

The aim of this study is to empirically examine the vindication of the savings-led growth hypothesis for Malaysia within a multivariate framework through the TYDL Granger causality test. The sample period for the study was from 1970:Q1 to 2008:Q4. With the TYDL Granger causality evidence, we conclude that the savings-led growth hypothesis for Malaysia is valid in the long-run. Importantly, the recursive regression-based TYDL causality test also affirmed that the savings-led growth hypothesis is valid and stable over the sample period. These findings are in tandem with the new growth theory that increases in savings will affect capital accumulation and thus stimulate economic growth in the long-run.

Although the findings of this study affirmed that savings is necessary, it is not a sufficient condition for economic growth. To effectively promote economic growth, available savings in Malaysia must be mobilised and channelled into productive investment. Since most of the available saving deposits are in financial institutions, we advise decision-makers to promote financial development and liberalisation in order to free available resources and channel them into productive sectors for investment. Moreover, priority should be given to export-oriented and manufacturing industries because they are the best examples of productive industries that gain foreign exchange earning and create more employment opportunities in the country. In doing so, the available savings in Malaysia can be effectively utilised and Malaysia’s long-term economic growth can be sustained.

Footnotes

Acknowledgements

The author would like to thank an anonymous reviewer for his/her valuable comments and suggestion on the earlier draft of this research paper. In addition, the author would like to thank Soo Yean Chua for his valuable comments and suggestions on an earlier draft of this manuscript. The author would also like to express his appreciation to Lee Junsoo and Abdulnasser Hatemi-J for sharing their programming codes to perform the econometric analysis. I am solely responsible for errors (if any), views and conclusions.

Notes

References

Aghion

Philippe

Comin

Diego

Howitt

Peter

Tecu

Isabel

(2009). When does domestic saving matter for economic growth? Harvard Business School Working Paper 09-080, Havard University.

Agrawal

(2001). The relation between saving and growth: Cointegration and causality evidence from Asia. Applied Economics, 33(4), 499–513.

Ang

J. B.

(2009). Financial development and economic growth in Malaysia. London: Routledge.

Baharumshah

A. Z.

Thanoon

M. A.

(2003). Determinants of gross national saving in Malaysia: A macroeconomic analysis 1960–2000. Savings and Development, 27(4), 421–40.

Baharumshah

A. Z.

Thanoon

M. A.

Rashid

(2003). Saving dynamics in the Asian countries. Journal of Asian Economics, 13(6), 827–45.

Bierens

H. J.

(1997). Nonparametric cointegration analysis. Journal of Econometrics, 7(2), 379–404.

Boo

M. C.

Normee

C. S.

(2004). Effect of foreign capital inflow and domestic savings on Malaysia’s economic growth: Time series evidence. In The 16th Malaysian Economic Association Convention, 9 December 2004, Malaysia.

Davidson

MacKinnon

J. G.

(2004). Econometric theory and methods. Oxford: Oxford University Press.

Dolado

J. J.

Lütkepohl

(1996). Making Wald tests work for cointegrated VAR system. Econometric Reviews, 15(4), 369–86.

10.

Domar

E. D.

(1946). Capital expansion, rate of growth, and employment. Econometrica, 14(2), 137–47.

11.

Gandolfo

(1981). Quantitative analysis and econometric estimation of continuous time dynamic. Amsterdam: North-Holland Pub. Co.

12.

Geweke

Meese

Dent

(1983). Comparing alternative tests of causality in temporal systems. Journal of Econometrics, 21(2), 161–94.

13.

Granger

C. W. J.

(1969). Investigating causal relations by econometric models and cross-spectral methods. Econometrica, 37(3), 428–38.

14.

Granger

C. W. J.

(1988). Some recent development in the concept of causality. Journal of Econometrics, 39(1–2), 199–211.

15.

Granger

C. W. J.

Newbold

(1974). Spurious regressions in econometrics. Journal of Econometrics, 2(2), 111–20.

16.

Hacker

R. S.

Hatemi-J

(2006). Tests for causality between integrated variables using asymptotic and bootstrap distributions: Theory and application. Applied Economics, 38(13), 1489–500.

17.

Harrod

R. F.

(1939). An essay in dynamic theory. Economic Journal, 49(193), 14–33.

18.

Holmes

J. M.

Hutton

P. A.

(1990). On the causal relationship between government expenditure and national income. Reviews of Economics Statistics, 72(1), 87–95.

19.

Hsiao

(1981). Autoregressive modeling and money-income causality detection. Journal of Monetary Economics, 7(1), 85–106.

20.

Johansen

Juselius

(1990). Maximum likelihood estimation and inference on cointegration with applications to the demand for money. Oxford Bulletin of Economics and Statistics, 52(2), 169–210.

21.

Lee

Strazicich

M. C.

(2001). Break point estimation and spurious rejections with endogenous unit root tests. Oxford Bulletin of Economics and Statistics, 63(5), 535–58.

22.

Lee

Strazicich

M. C.

(2003). Minimum Lagrange multiplier unit root test with two structural breaks. Review of Economics and Statistics, 85(4), 1082–89.

23.

Lee

Strazicich

M. C.

(2004). Minimum LM unit root test with one structural break. Department of Economics, Appalachian State University, Boone, NC.

24.

Lumsdaine

R. L.

Papell

D. H.

(1997). Multiple trend breaks and the unit-root hypothesis. Review of Economics and Statistics, 79(2), 212–18.

25.

Lütkepohl

(2005). New Introduction to multiple time series analysis. Germany: Springer-Verlag.

26.

Mantalos

(2000). A graphical investigation of the size and power of the Granger causality tests in integrated-cointegrated VAR systems. Studies in Non-linear Dynamics and Econometrics, 4(1), 17–33.

27.

Nelson

C. R.

Plosser

C. I.

(1982). Trends and random walks in macroeconomic time series: Some evidence and implications. Journal of Monetary Economics, 10(2), 139–62.

28.

Perron

(1989). The great crash, the oil price shock and the unit root hypothesis. Econometrica, 57(6), 1361–401.

29.

Pesaran

M. H.

Shin

Smith

R. J.

(2001). Bounds testing approaches to the analysis of the level relationships. Journal of Applied Econometrics, 16(3), 289–326.

30.

Phillips

P. C. B.

(1986). Understanding spurious regression in econometrics. Journal of Econometrics, 33(3), 311–40.

31.

Romer

(1986). Increasing returns and long-run growth. Journal of Political Economy, 94(5), 1002–37.

32.

Sen

(2003). On unit-root tests when the alternative is a trend-break stationary process. Journal of Business and Economic Statistics, 21(1), 174–84.

33.

Solow

(1956). A contribution to the theory of economic growth. Quarterly Journal of Economics, 70(1), 65–94.

34.

Tang

C. F.

(2008a). An empirical modelling on savings behaviour in Malaysia. Labuan Bulletin of International Business and Finance, 6, 57–76.

35.

Tang

C. F.

(2008b). Wagner’s law versus Keynesian hypothesis: New evidence from recursive regression-based causality approaches. ICFAI Journal of Public Finance, 6(4), 29–38.

36.

Tang

C. F.

(2009). Does causality technique matter to savings-growth nexus in Malaysia? Malaysian Management Journal, 13(1 & 2), 1–10.

37.

Tang

C. F.

Chua

S. Y.

(2009). The savings-growth nexus in Malaysia: Evidence from nonparametric analysis. ICFAI Journal of Financial Economics, 7(3 & 4), 1–12.

38.

Toda

H. Y.

Yamamoto

(1995). Statistical inference in vector autoregressions with possibly integrated process. Journal of Econometrics, 66(1–2), 225–50.

39.

World Bank (1993). The East Asian miracle: Economic growth and public policy. New York: Oxford University Press.

40.

Yamada

Toda

H. Y.

(1998). Inference in possibly integrated vector autoregressive models: Some finite sample evidence. Journal of Econometrics, 86(1), 55–95.

41.

Zapata

H. O.

Rambaldi

A. N.

(1997). Monte Carlo evidence on cointegration and causation. Oxford Bulletin of Economics and Statistics, 59(2), 285–98.

42.

Zhou

(2001). The power of cointegration tests versus data frequency and time spans. Southern Economic Journal, 67(4), 906–21.

43.

Zivot

Andrews

D. W. K.

(1992). Further evidence of the greater crash, the oil price shock and the unit-root hypothesis. Journal of Business and Economic Statistics, 10(3), 251–70.