THE IMPACT OF LIFE EXPECTANCY ON ECONOMIC GROWTH: PANEL COINTEGRATION AND CAUSALITY ANALYSES FOR OECD COUNTRIES Eyyup Ecevit Erciyes University, Kayseri, Turkey Email: [email protected]

Abstract This paper examines the impact of life expectancy at birth on economic growth for 21 OECD countries using panel data for the period of 1970-2010 in the context of panel cointegration and causality tests. Three different model specifications are considered for this purpose. We use the variable of life expectancy as an indicator of health and also employ real per capita gross domestic product (constant 2000 $) as a criterion of economic growth. The other independent variables are real exports, real fixed capital and energy use per capita. Based on the findings of the empirical analyses, some important findings can be suggested. Firstly, the results of the LLC, IPS and Breitung panel unit root tests indicate that all of the series are I(1). Secondly, by using Kao and Maddala-Wu cointegration tests, it is showed that there is a long-run relationship between the variables. Thirdly, based on panel OLS, Pedroni DOLS and FMOLS techniques, the estimated coefficient for life expectancy is found positive and statistically significant. Fourthly, the results of panel Granger causality tests based on panel VAR models indicate that there is a unidirectional causality running from life expectancy to real per capita GDP. Thus, this paper provides some important evidence that life expectancy is a fundamental determinant of economic growth in the OECD countries for the period considered for all of three models. Additionally, some important policy implications are drawn from the findings of this study. Key Words: Health, Life Expectancy, Economic Growth, Panel Cointegration, Panel Causality.

technological progress. Moreover, in these models health and determinants of health were not taken into consideration.

1.0 INTRODUCTION Examining the link between health outcomes and economic growth at the macroeconomic level has been getting interested in theorists and policy makers. Why healthier people might be richer than others? Do improvements in health help to boost economic growth? Can health explain cross-country differences in levels and growth rates of income? The answer is that health may lead to income growth through its effect on human capital accumulation provided that people have sufficient food and satisfactory educational opportunities.

Nowadays, it is known that investment in human capital, innovation and knowledge are significant contributors to economic growth. Two key components of human capital are education and health. Human capital theory, which is primarily developed by Schultz (1961), Becker (1962), Denison (1962) and Mincer (1974), is about the role of human capital in the production process and about the incentives to invest in skills, including in the forms of schooling and training.

Economic growth theories or models should be considered to evaluate the link between health and economic growth realistically. The modern explanation of economic growth began with the classical economists, notably Smith (1776) and Ricardo (1817), argued that industrial growth fueled by the expansion of domestic and international markets was the driving force behind economic growth (Barro, 1996). Harrod (1939) and Domar (1946), who are the pioneers of Keynesian economics, argued that economic growth depends on policies to increase investment, by increasing saving, and using that investment more efficiently through technological advances. The Harrod-Domar model was extended by the neo-classical economists (Solow, 1956; Swan, 1956) by including productivity growth. But neoclassical growth models could not explain the source of

The first generation endogenous growth models, such as Romer (1986, 1990), Grossman and Helpman (1991) and Aghion and Howitt (1992), focused on education, R&D and innovation rather than health. Mushkin (1962), who first emphasized the importance of health, pointed out that health constitutes an important form of investment unlike other forms of human capital formation like education. Nelson and Phelps (1966) argued that a higher stock of health could stimulate growth by facilitating technological innovation. Accordingly, productivity growth should be positively correlated with the level of health, in particular with the

1

average level of life expectancy in a country. Thereafter, Grossman (1972) constructed a model of the demand for the commodity “good health”. According to Grossman, health can be viewed as a durable capital stock that produces an output of healthy time. This triggered the developing of new theoretical models (Mankiw et al., 1992, Barro, 1996, and Barro and Sala-i-Martin, 2004) incorporating health as a determinant of economic growth.

Martin (2004) using a panel dataset of some countries indicated that the life expectancy has a positive and statistically significant effect on the rate of economic growth with a coefficient of 0.042, which implies an annual rate of increase of per capita real GDP about 4.2%. Acemoglu and Johnson (2007) following a Lucas approach regressed income growth on the increase in life expectancy between the period of 1940 and 1980. They showed that life expectancy has no significant effect on income growth over the period. Lorentzen, McMillan and Wacziarg (2008) adopting a Nelson-Phelps model argued that the effect of life expectancy on income per capita is non-monotonic. They found that life expectancy after the onset of the demographic transition has a positive effect on wealth.

Weil (2005) examines the relationship between health investment and economic growth by considering both macroeconomic and microeconomic foundations. According to Weil, studies examining the link between health and economic performance have generally investigated health inputs or health outcomes. Health inputs are the physical factors that influence the individual’s health such as nutrition, exposure to pathogens, and the availability of medical care. Health outcomes include life expectancy, the ability to work hard, and cognitive functioning. For the purpose, in explaining income differences among countries, life expectancy can be one of the key health outcomes.

Barro and Lee (2010) which examine low-income and high-income countries and use cross-country OLS found that life expectancy has a weak or negative effect on economic growth. Aghion et al. (2010) reconsidered the relationship between health and growth in the context of endogenous growth theory. They used a unified framework which includes Lucas and Nelson-Phelps approaches. The results of cross-country regressions over the period 19602000 showed that a higher initial level and a higher rate of improvement in life expectancy were significantly positive impact on per capita GDP growth. Finally, Hansen (2012) investigating the non-linearity dimension and using 10yearly observations from 1940 to 1980 presented a nonmonotonic relationship between life expectancy and per capita GDP.

With respect to the empirical literature, Knowles and Owen (1995, 1997); Hamoudi and Sachs (1999); Bloom et al. (2001); Devlin and Hansen (2001); McDonald and Roberts (2002) and Li and Liang (2010) emphasized the importance of health in a country’s economic growth. Knowles and Owen (1995, 1997) indicated that there is a significant statistical relationship between health and growth. Hamoudi and Sachs (1999) demonstrated an endogenous relationship between the variables. Bloom et al. (2001) extending the Solow model with human capital found that there is a significant relationship between health capital and economic growth. Li and Liang (2010) showed that the impact of the stock of health and education on economic growth is statistically significant and that the statistical impact of health on growth is stronger than that of education.

These findings have very important implications for the academic researches as well as for the policy debate because they challenge the view that improving health is beneficial for economic development. Nowadays, panel data analyses on the relationship between life expectancy and economic growth estimate long-run parameters using different panel estimation methods as in Table 1. It reports a selection of the empirical studies that include health as a determinant of economic growth and the findings of them. It also provides eight panel data and three cross-country studies investigating relationship between life expectancy and economic growth. It can be said that the results are more complex and the question whether the improvements in life expectancy cause increases in per capita income is the subject of a lively debate. But these studies do not take the issues of panel unit root, cointegration and causality into consideration. From this aspect, our paper becomes somewhat different from this empirical literature.

Recently, an important literature group dealing with the impact of life expectancy as an indicator of health on economic growth has been developing. For example; Barro (1997) has found that 10% increase in life expectancy lead to a four-tenth percent increase in economic growth. For every 10% increase in life expectancy you can expect almost half a percent in economic growth. Bloom, Canning, and Sevilla (2004) indicated that the panel results, which come from regressing residual, productivity provided that a one-year increase in life expectancy raises output by 4 percent. Barro and Sala-i-

2

Table 1: Overview of Selected Empirical Studies (Dependent variable: Life expectancy) Study

Barro (1996)

Caselli, Esquivel and Lefort (1996)

Country

Period

Sub-Saharan Africa

1965-75

Latin America

1975-85

East Asia

1985-90

1965-85 Developed and Developing Countries 1965-87

Sub-Saharan Africa Sachs and Warner (1997)

Fast growing 1965-90 developing economies, Other developing countries

Other Independent Variables Male secondary and higher schooling, log(GDP) X male schooling, log fertility rate, government consumption ratio, rule of law index, terms of trade change, democracy index, democracy index squared, inflation rate, continental dummies Male education, female education, revolutions, assassinations, terms of trade, investment ratio, government consumption ratio, log (1+black market premium),

78 Asian and nonAsian Countries

Bloom, Canning and Malaney (1999)

East Asia

Log GDP per capita, log GDP per worker, located in the tropics, access to ports (landlocked), quality of institutions, openness, log years of 1965-90 secondary schooling, growth in total population, growth in working age population, log of coastal population density, log of inland population density.

Bloom, Canning and Sevilla (2004)

104 countries

Andres, GuerraTurrubiates and Oca-Hernandez (2005)

Three-stage OLS

Panel data OLS, GMM

OLS, Instrumental Variables

Life expectancy has a positive effect on growth. Life expectancy has a negative and insignificant effect on economic growth.

Differences in life expectancy help account for Africa’s slow growth compared to other developing countries.

Life expectancy has a positive effect on economic growth.

Panel data OLS, Instrumental Variables

Cross-country 1970-80 Capital, labor, schooling

Life expectancy has a positive effect on economic growth.

Panel data

1960-90 Capital, labor, schooling, experience, Panel data technological catch-up, percentage of Ten year Non-linear twointervals land area in the tropics, governance stage OLS

104 countries

Conclusion

Panel data

Openness, average national saving ratio 1970-1990, Central government saving, Institutional quality index, Cross-country Natural resource abundance, Average OLS annual inflation, Growth of neighboring economies, Index of ethnolinguistic fractionalization

Log years of secondary schooling, natural resource abundance, openness, quality of institutions, 1965-90 access to ports dummy, average government savings rate, tropics dummy, ratio of coastline to land area

Bloom and Williamson (1998)

Econometric Technique(s)

OLS,

1980-90 GLS

Increases in life expectancy have a large effect on economic growth in East Asia.

Life expectancy has a weak effect on economic growth.

There is no significant impact of life expectancy on economic growth.

Panel data Two-stage OLS, Acemoğlu and Johnson (2007)

120 countries

1940-80 Log population, log total births, log GDP per working age population, log 1960-00 years of schooling

Instrumental Variables,

Life expectancy has a negative effect on economic growth.

GMM

Aghion, Howitt and Martin (2010)

Sub-Saharan Africa, Developed and

1960-00

Initial log GDP per capita, average infant mortality, Initial adult

3

Panel data

Life expectancy has a positive impact on economic growth.

developing countries

mortality 1960, initial infant mortality 1960

OLS, Instrumental Variables SGMM

Barro and Lee (2010)

Male upper-level schooling, log total fertility rate, government Low-income countries 1965-75 consumption ratio, rule of law, 1975-85 High-income democracy, openness ratio, change in countries terms of trade, investment ratio, 1985-95 inflation rate

Hassan and Cooray (2012)

84 countries

Capital, labor, male primary enrolment, female primary enrolment, 1960-09 male secondary enrolment, female secondary enrolment

Cross-country OLS

Life expectancy has a weak or negative effect on economic growth

Panel data OLS, SGMM,

Male life expectancy has a positive effect on economic growth. Female has a negative effect.

LSDVC

Sources: Author 2. The Data and Model Specifications The data of our study setting are derived from World Bank’s World Development Indicators, 2012 in an attempt to test the cointegration and causality relationship between life expectancy and economic growth in a panel data. I use the variable of life expectancy as an indicator of health and also employ real per capita GDP as a criterion of economic growth. The other dependent variables are real exports, real fixed capital and energy use per capita. The study uses annual data and covers the period of 1970 to 2010 for 21 OECD countries.1 The logarithms of the variables are used in the empirical analyses.

The main objective of this study is to test the effect of health on economic growth. It is preferred the variable of life expectancy as a health indicator like the other empirical studies. In this paper, a further panel data analysis in the context of cointegration and causality relationship is provided for 21 OECD countries spanning over 1970-2010. The panel unit root tests proposed by Levin et al. (2002), Im et al. (2003) and Breitung (2000) are implemented. More precisely, panel cointegration tests developed by Kao (1999) and Maddala and Wu (1999) are used. Nowadays, Pedroni (2000) DOLS and FMOLS techniques are among the most popular and common estimators in panel data analyses. I employ the panel OLS, Pedroni DOLS and FMOLS estimators to investigate long-run parameters in three different model specifications. Finally, panel Granger causality test is used to analyze the causal relationship between life expectancy and real per capita GDP.

All the variables and their descriptive statistics are reported in Table 2. The cross-sectional dimension is 21 units and the time dimension is 41 years. In total, there are 861 observations for each variable. For all variables, it seems that there are no strong outliers. On the other hand, Figure 1 shows the series in natural logarithm over the period.

This paper provides some important findings. The results show that (i) life expectancy has a positive and statistically significant effect on real per capita GDP in the long run, (ii) there is a Granger causality running from life expectancy to real per capita GDP. Thus, these findings provide an empirical evidence supporting the hypothesis of life expectancy oriented economic growth in OECD countries in three cases.

The following linear panel data specifications are used in the empirical analyses to evaluate how life expectancy has affected real per capita GDP. Panel cointegration and causality tests are conducted for three cases because the variables are cointegrated only under these cases: Model 1. lnpercapita GDP=f(lnlife, lnexports) (1)

The rest of the paper is organized as follows. Section 2 presents the data and model specifications. Section 3 provides the econometric methodology and findings. Section 4 submits the conclusion and policy implications.

1

The countries considered in this study are Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Hungary, Iceland, Italy, Japan, Luxemburg, Netherland, Norway, Portugal, Spain, Sweden, United Kingdom and United States.

4

1 - 70 1 - 95 2 - 79 2 - 04 3 - 88 4 - 72 4 - 97 5 - 81 5 - 06 6 - 90 7 - 74 7 - 99 8 - 83 8 - 08 9 - 92 10 - 76 10 - 01 11 - 85 11 - 10 12 - 94 13 - 78 13 - 03 14 - 87 15 - 71 15 - 96 16 - 80 16 - 05 17 - 89 18 - 73 18 - 98 19 - 82 19 - 07 20 - 91 21 - 75 21 - 00

11

9

7

30

lnexport

28

26

24

22

20

5 1 - 70 1 - 95 2 - 79 2 - 04 3 - 88 4 - 72 4 - 97 5 - 81 5 - 06 6 - 90 7 - 74 7 - 99 8 - 83 8 - 08 9 - 92 10 - 76 10 - 01 11 - 85 11 - 10 12 - 94 13 - 78 13 - 03 14 - 87 15 - 71 15 - 96 16 - 80 16 - 05 17 - 89 18 - 73 18 - 98 19 - 82 19 - 07 20 - 91 21 - 75 21 - 00

1 - 70 1 - 95 2 - 79 2 - 04 3 - 88 4 - 72 4 - 97 5 - 81 5 - 06 6 - 90 7 - 74 7 - 99 8 - 83 8 - 08 9 - 92 10 - 76 10 - 01 11 - 85 11 - 10 12 - 94 13 - 78 13 - 03 14 - 87 15 - 71 15 - 96 16 - 80 16 - 05 17 - 89 18 - 73 18 - 98 19 - 82 19 - 07 20 - 91 21 - 75 21 - 00

GDP=f(lnlife,lnexports,lncapital)

lnpercapita GDP

30

lncapital 16

28 14

26 12

24 10

22 8

20 6

1 - 70 1 - 95 2 - 79 2 - 04 3 - 88 4 - 72 4 - 97 5 - 81 5 - 06 6 - 90 7 - 74 7 - 99 8 - 83 8 - 08 9 - 92 10 - 76 10 - 01 11 - 85 11 - 10 12 - 94 13 - 78 13 - 03 14 - 87 15 - 71 15 - 96 16 - 80 16 - 05 17 - 89 18 - 73 18 - 98 19 - 82 19 - 07 20 - 91 21 - 75 21 - 00

1 - 70 1 - 95 2 - 79 2 - 04 3 - 88 4 - 72 4 - 97 5 - 81 5 - 06 6 - 90 7 - 74 7 - 99 8 - 83 8 - 08 9 - 92 10 - 76 10 - 01 11 - 85 11 - 10 12 - 94 13 - 78 13 - 03 14 - 87 15 - 71 15 - 96 16 - 80 16 - 05 17 - 89 18 - 73 18 - 98 19 - 82 19 - 07 20 - 91 21 - 75 21 - 00

Model2.lnpercapita (2) Model3.lnpercapitaGDP=f(lnlife, (3) lnenergy,

Table 2: Descriptive Statistics

Balanced panel: N=21, T=41, Observations=861 Variables Descriptions Mean Median Std. Dev. Minimum Maximum

lnlife Life expectancy at birth 4.329 4.333 0.041 4.201 4.418

lnexports Exports (constant 2000 $) 24.873 25.016 1.445 20.586 28.056

lncapital Fixed capital (constant 2000 $) 24.642 24.485 1.674 20.176 28.441

lnpercapitaGDP Per capita GDP (constant 2000 $) 9.774 9.853 0.530 7.786 10.940

lnenergy Energy use (kg of oil equivalent per capita) 10.941 10.832 1.587 6.762 14.664

Figure 1: The Plots of the Series (1970-2010)

4.45

lnlife

10 4.40

4.35

4.30

8 4.25

4.20

4.15

lnenergy

lncapital)

Im et al. (2003) exhibit a power comparison of the LLC and IPS tests and argue that the IPS test is more powerful than the LLC test. Although the null hypothesis is the same in the two tests, the alternative hypothesis is different. Both the Levin et al. (2002) and Im et al. (2003) tests suffer from a dramatic loss of power when individual specific trends are included, which is due to the bias correction. However, the Breitung (2000) panel unit root test does not rely on bias correction factors. Monte Carlo experiments showed that the Breitung (2000) test yields substantially higher power and smallest size distortions compared to Levin et al. (2002) and Im et al. (2003).

3.0 Econometric Methods and Findings The main aim of the study is to analyze the impact of life expectancy on economic growth in the context of cointegration and causality relationship. The cointegration and causality tests will be performed in four steps. Firstly, I test for the order of integration in the variables and implement the panel unit root tests proposed by Levin et al. (2002), Im et al. (2003) and Breitung (2000).2 Secondly, conditional on finding that these variables are integrated of order one I test for panel cointegration using the approaches suggested by Kao (1999) and Maddala and Wu (1999). Thirdly, I estimate the long-run parameters using panel OLS, Pedroni DOLS and FMOLS estimators. Finally, I test for Granger causality between life expectancy and real per capita GDP.

Panel unit root test results obtaining from the model with constant and trend terms are illustrated in Table 3. It is seen that the panel variables are not stationary at level. When the tests are applied to the first-differenced variables, however, all of the variables are found to be stationary. From these panel unit root tests I conclude that all variables are integrated at order of I(1).

3.1. Panel Unit Root Tests In panel cointegration, an essential step is to examine whether the variables contain a panel unit root. Non-stationary panels have become extremely popular and have attracted much attention in both theoretical and empirical research over the last decade. A number of panel unit root tests have been proposed in the literature which include Levin et al. (2002), Im et al. (2003), Breitung (2000), Maddala and Wu (1999), Choi (2001) and Hadri (2000). Technically, all first generation tests share the null hypothesis of stationarity but diverge on the alternative one, which can be homogeneous or heterogeneous. In this study I adopt three different methods, namely those of Levin et al. (2002) (LLC), Im et al. (2003) (IPS) and Breitung (2000). These tests don’t consider crosssectional dependency. Levin et al. (2002) generalized the individual unit root test to panels with heterogeneous serially correlated errors, fixed effects and individual deterministic trends. One of the drawbacks of the Levin et al. (2002) is that, it requires a homogeneous autoregressive root under the alternative hypothesis. However, Im et al. (2003) proposed a panel unit root test that allows for a heterogeneous autoregressive coefficient under the alternative hypothesis.

2

The former two are widely used panel unit root analysis in the literature on panel cointegration. The null hypothesis of the tests is a unit root in the panel. However, while IPS (2003) assumes that the cross-sectional units have individual unit root process, LLC (2002) and Breitung (2000) assume that the cross-sectional units share a common unit root process.

6

Table 3: Panel Unit Root Test Results Im, Pesaran, and Shin

Prob.

Levin, Lin, and Chu

Prob.

Breitung

Prob.

ln percapita GDP

0.612

0.729

2.071

0.980

4.443

1.000

lnlife

-0.786

0.215

-0.377

0.352

1.173

0.879

lnexports

0.807

0.790

1.184

0.881

1.572

0.942

lncapital

-2.518*

0.005

-0.857

0.195

0.622

0.733

lnenergy

-1.524

0.063

-1.575

0.057

3.245

0.999

ln percapita GDP

-14.571*

0.000

-17.224*

0.000

-9.598*

0.000

lnlife

-31.224*

0.000

-30.190*

0.000

-14.960*

0.000

lnexports

-18.894*

0.000

-21.705*

0.000

-7.225*

0.000

lncapital

-

-

-16.873*

0.000

-9.705*

0.000

lnenergy

-20.937*

0.000

-19.977*

0.000

-11.380*

0.000

Variables Series in levels

Series in differences

first

Note: * denote significance at the 1% level.  is the first difference operator. Newey-West bandwidth selection with Bartlett kernel was used for the LLC test. Schwarz Bayesian Criterion was used to determine the optimal lag length.

Augmented Dickey-Fuller (ADF) type test. DF-type test can be computed from the estimated residuals as:

3.2. Panel Cointegration Tests The cointegration methodology as applied to time series data was first introduced in the 1980s as in Engle and Granger (1987), Johansen (1988, 1991), Johansen and Juselius (1990), and others. The econometric methodology behind testing and estimation in cointegrated panels has only recently been developed. Important contributions in this field have been made by for instance Pedroni (1999), McKoskey and Kao (1998), Kao (1999), Maddala and Wu (1999), Breitung (2005), Westerlund (2007) and Larsson and Lyhagen (2007).

uˆit  uˆit 1  vit

For the ADF test, I run the following ADF regression: 

uˆ it  uˆ it 1    j uˆ it  j  vit

(6)

j 1

In analyzing of cointegration and causality, secondly, I test for panel cointegration using the approaches suggested by Kao (1999) and Maddala and Wu (1999). Consider the following system of cointegrated regressions:

yit   i   i xit  uit

(5)

Where the residuals

uˆ it are obtained from equation

(4). The following specification of null and alternative hypotheses is used.

(4)

Where i=1,. . . , N, t=1, . . . , T., i are individual constant terms,  is the slope parameter, it uit are stationary disturbance terms, and finally, by construction, it yit and it xit are integrated processes of order one for all i. Kao (1999) derives two types of panel cointegration tests. The first is a Dickey-Fuller (DF) type test and the second is an

H 0 :   1, for all i and H A :   1, for all i Kao (1999) proposes four DF-type statistics. The first two DF statistics are based on assuming strict exogeneity of the regressors with respect to the errors in the equation, while the remaining two allow for endogeneity of the

7

regressors. In addition, Kao (1999) proposes an ADF test statistic. Finally the DF statistics, which allow for endogeneity, and the ADF statistic involve deriving some nuisance parameters from the long run conditional variances Ω. The asymptotic distributions of all tests converge to a standard normal distribution N (0, 1) as T → ∞ and N → ∞. ADF test statistic is as follows: Where,

(   1)  

t ADF

N

i 1

(e Q e )  ' i

1 2

i i

(7)

Sv

Kao cointegration tests results are seen from the Table 4. For three models, the findings indicate existence of long run relationship between variables.

  t ADF  6 N  v /( 2 0v ) ADF       02v /( 2 v2 )  3 v2 /(10 02v )

Table 4: Kao Panel Cointegration Test Results Models

ADF

Prob.

Model 1

-5.032*

0.000

Model 2

-5.234*

0.000

Model 3 -6.758* 0.000 Note: * denote significance at the 1% level. Schwarz Bayesian Criterion was used to determine the optimal lag length. The second panel cointegration test applied is the Johansen-Fisher type panel cointegration test developed by Maddala and Wu (1999). They use Fisher’s result to propose an alternative approach to test for cointegration in panel data by combining tests from individual cross-sections to obtain a test statistic for the full panel. The Johansen-Fisher panel cointegration test is panel version of the individual Johansen cointegration test. The Johansen-Fisher panel cointegration test is based on the aggregates of the p-values of the individual Johansen maximum eigenvalues and trace statistic. If pi is the p-value from an individual cointegration test for cross-section i, under the null hypothesis for the panel

n

2  2 log( pi )   2n

(8)

i 1

Where 2 (2n) is a chi-square distribution with 2N degrees of freedom. The 2 value is based on MacKinnon‐Haug‐Michelis (1999) p-values for Johansen’s cointegration trace test and maximum eigenvalue test. In the Johansen type panel cointegration tests results heavily depends on the number of lags of the VAR system. The results for Johansen-Fisher based panel cointegration test are presented in Table 5. The findings strongly support an evidence of cointegration relationships between variables in all models.

Table 5: Johansen-Fisher based Panel Cointegration Test Results Models Model 1

Model 2

Model 3

Hypothesis

Trace test

Prob.

Max. Eigen value test

Prob.

None

115.70*

0.000

107.70*

0.000

At most 1

44.47

0.368

40.21

0.549

At most 2

28.23

0.948

28.23

0.948

0.000

99.20

*

0.000

*

None

165.80

At most 1

91.70*

0.000

68.06*

0.006

At most 2

50.09

0.183

46.36

0.297

At most 3

28.92

0.937

28.92

0.937

None

204.9

*

0.000

143.20

At most 1

95.97*

0.000

71.77*

0.002

At most 2

50.60

0.170

41.60

0.488

At most 3

36.01

0.730

36.01

0.730

8

*

0.000

Note: * indicates significant at 1% level. 3.3. Panel Cointegration Estimation Once the cointegration relationship is established, the next step is to estimate the long-run parameters. In order to estimate panel cointegration parameters, various methods have been proposed, namely panel pooled OLS, panel dynamic OLS (DOLS), and panel fully modified OLS (FMOLS). The study employs all of three estimation method. A standard panel OLS estimator for the coefficient

yit   i   i xit  uit xit  xit 1  eit

 t 1 ( xit  xi )  T

Where

1 N

2

T

  (x i 1

t 1

it

 xi )( yit  yi ) (9)

 i   i0  i  i' where

homogeneous across i members of the panel

ˆi, FMOLS  N

ˆi,OLS is

1

N

 i 1

1

(11)

Lˆ 21i xit , Lˆ 22i

ˆ0 ˆi  ˆ 21i   21i

Lˆ ˆ0  21i ˆ 22i   22i Lˆ





22i

and Li is a lower triangular decomposition of Ωi defined as follows:

11i  '21  i     21i  22i  Pedroni (2001) also considers a group-mean DOLS estimator which is non-parametric approach. This is done modifying equation (10) to include lead and lag dynamics:

The test statistics derived from the group-mean estimators are constructed to test the null hypothesis, the

the

Where

The FMOLS and DOLS methodologies are proposed by Kao and Chiang (2000) to estimate the longrun cointegration vector, for non-stationary panels. These estimators correct the standard pooled OLS for serial correlation and endogeneity of regressors that are normally present in long-run relationship. Pedroni (2000) has suggested the group-mean FMOLS and DOLS estimators. According to Pedroni (2000) group mean tests are preferred over the pooled tests since they allow greater flexibility under alternative hypotheses.

against

is

T  T    ( xit  xi )2    ( xit  xi ) yit*  Tˆi   t 1   t 1 

yit*  ( yit  yi ) 

i

 i0

contemporaneous covariance and is a weighted sum of autocovariances. The Panel FMOLS estimator which is parametric approach for the coefficient β is defined as follows:

unbiased.

all

and

= L iL’i (Li is a lower triangular decomposition of Ωi). In this case, the variables are said to be cointegrated for each member of the panel, with cointegrating vector β. The terms αi allow the cointegrating relationship to include member specific fixed effect. The covariance matrix can also be decomposed as

 i in

x i and y i refer to the individual means for

for

I(1)it

 i  (uit  eit )  I(0) with long run covariance matrix Ωi

each i member of the cross section. According to Pedroni (2000) this estimator is asymptotically biased and its distribution is dependent on nuisance parameters associated with the dynamics underlying the processes determining x and y. Only if x is strictly exogenous and the dynamics are

H 0 : i  0

z it  ( yit , xit ) 

Where

equation (4) is given by:

N ˆi ,OLS     i 1

(10)

alternative

H 1 :  i   0 , so that the values for βi are not constrained to be the same under the alternative hypothesis. Consider the following co-integrated system for a panel of i =1, 2,…., N members,

yit   i   i xit 

Ki



j  Ki

ik

xi ,t k  eit

and the estimated coefficient β is given by:

9

(12)

ˆ

i , DOLS

where

1  1 N  T   T    N    z it z it    z it yit*  i 1  t 1   t 1  1 

zit  ( xit  xi , xit k ,......, xit k )

I provide pooled OLS, group-mean DOLS and FMOLS results of the cointegration slopes for three different models. Based on these findings, the estimated coefficient for life expectancy is positive and statistically significant in three cases. This means that life expectancy has a positive effect on real per capita GDP. The results indicate that life expectancy is a fundamental determinant of economic growth in OECD countries (Table 6).

(13)

is

the

2(K+1)x1 vector of regressors.

Table 6: Panel OLS, DOLS and FMOLS Estimates Models

Variables

OLS

tstatistics

DOLS

tstatistics

FMOLS

tstatistics

Model 1

lnlife

2.277*

11.251

2.314*

9.971

1.69*

5.02

lnexports

0.260*

22.710

0.193*

14.819

0.23*

12.92

lnlife

1.805

*

10.207

2.155

*

lnexports

0.189*

14.615

0.106*

lncapital

0.194

*

lnlife

4.321*

Model 2

Model 3

**

lnenergy

0.030

lncapital

0.274* *

Note: and

**

9.014

1.37

15.379

0.16*

12.49

13.464

0.196

*

16.363

0.21

*

13.66

36.006

3.947*

25.908

3.41*

19.88

p

k 1

k 1

*

2.047

0.275

21.704

0.193*

*

8.537

0.27

17.797

0.22*

4.95

7.99 15.64

denote significance at 1% and 5% respectively.

3.4. Panel Causality Test Kao and Maddala-Wu tests don’t indicate the direction of causality when the variables are co-integrated. Panel Granger causality which is based on panel VAR model can be used to assess the long-run causal relationship and the direction of causality between life expectancy and real per capita GDP. For the panel Granger causality test, the following model is estimated: p

*

Wald F and 2 tests are applied to the hypothesis above. The appropriate lag lengths are selected using AIC for variables. After defining the appropriate lag lengths, the long-run causality is investigated for three models. The results of panel Granger causality tests are presented in Table 7. According to the results, there is Granger causality running from life expectancy to real per capita GDP, which is significant in 1 %. Thus, these findings suggest that life expectancy affects economic growth in three cases.

yit  1i  11ik yit k  12ik xit k  u1it (14) H 0 : 12ik  0 H 1 : 12ik  0

(15)

Table 7: Panel Granger Causality Test Results Models Model 1

Hypotheses

Wald F-test

Prob.

Wald χ2-test

Prob.

Causality

Δlnlife doesn’t Granger cause ΔlnpercapitaGDP

4.228*

0.002

16.915*

0.002

Yes

Δlnexports doesn’t Granger cause ΔlnpercapitaGDP

2.170**

0.070

8.683**

0.060

Yes

10

Model 2

Model 3

Δlnlife doesn’t Granger cause ΔlnpercapitaGDP

4.038*

0.003

16.154*

0.002

Yes

Δlnexports doesn’t Granger cause ΔlnpercapitaGDP

0.990

0.411

3.962

0.411

No

Δlncapital doesn’t Granger cause ΔlnpercapitaGDP

3.403*

0.009

13.615*

0.008

Yes

Δlnlife doesn’t Granger cause ΔlnpercapitaGDP

4.120*

0.002

16.481*

0.002

Yes

Δlncapital doesn’t Granger cause ΔlnpercapitaGDP

4.464*

0.001

17.858*

0.001

Yes

Δlnenergy doesn’t Granger cause 1.613 0.169 6.452 0.167 No ΔlnpercapitaGDP Note: * and ** denote significance at 1% and 10% respectively. Akaike Information Criterion was used to determine the optimal lag length. significant in three cases. In other words, life expectancy has a positive effect on real per capita GDP. These findings are consistent with the results of Bloom et al. (1999), Bhargava et al. (2001), Canning and Sevilla (2002) and Aghion et al. (2010). I also employ panel Granger causality test based on panel VAR model. According to the findings, in all models there is a Granger causality running from life expectancy to real per capita GDP, which is significant at 1%. Thus, these estimations and causality findings suggest that life expectancy is a fundamental determinant of economic growth for OECD countries over the period in three cases.

4.0 CONCLUSION Growth and development economists have commonly accepted that the role of human capital is one of the main determinants of sustained economic growth. New endogenous growth theorists such as Grossman (1972), Mankiw et al. (1992), Barro (1996) and Barro and Sala-iMartin (2004) have modeled health as human capital. These economists agree that health raises the level of human capital and has a positive effect on productivity and economic growth. But determining the cointegration and causal relationship between health and economic growth is not simple owing to the features of the variables used.

Nearly all of the OECD countries have experienced large gains in life expectancy over the past five decades. Across OECD countries, on average, life expectancy for the whole population reached 79.5 years in 2009. Gains in life expectancy in OECD countries in recent decades can be attributed to a number of factors, including rising in living standards, improving in lifestyle and better education opportunities, as well as greater access to quality health services. For this reason, the policies improving life expectancy of population for sustainable economic growth should take into consideration these factors. However, another empirical study to investigate the main determinants of life expectancy in OECD countries can be performed.

This paper has examined the role of life expectancy in economic growth in 21 OECD countries over the period 1970-2010 by using cointegration and causality methods. I adopt three different panel unit root tests, namely those of Levin et al. (2002) (LLC), Im et al. (2003) (IPS) and Breitung (2000). The results of panel unit root tests indicate that all variables are integrated at order of I(1). Secondly, I test for panel cointegration using the approaches suggested by Kao (1999) and Maddala and Wu (1999). Panel cointegration tests strongly support the existence of cointegration relationships between the variables in three specifications. In order to estimate panel cointegration parameters, I employ different estimators, namely panel OLS, Pedroni DOLS, and Pedroni FMOLS. The results indicate that the coefficient of life expectancy is positive and statistically

11

14. Bloom, D.E., D. CANNING, and J. Sevilla, 2001. The Effect of Health on Economic Growth: Theory and Evidence, NBER Working Paper No. 8587. 15. Bloom, D.E., D. CANNING, and J. SEVİLLA, 2004. The Effect of Health on Economic Growth: A Production Function Approach, World Development, 32 (1), 1-13. 16. Breitung, J., 2000. The Local Power of Some Unit Root Tests for Panel Data. In: Baltagi, B.H. (Ed.), Nonstationary Panels, Panel Cointegration and Dynamic Panels, Advances in Econometrics, JAI Press, Amsterdam, 161-78. 17. Breitung, J., 2005. A Parametric Approach to the Estimation of Cointegration Vectors in Panel Data, Econometric Reviews, 24, 151-73. 18. Caselli, F., G. ESQUIVEL and F. LEFORT, 1996. Reopening the Convergence Debate: A New Look at Cross-Country Growth Empirics, Journal of Economic Growth, 1 (3), 363-89. 19. Choi, I., 2001. Unit Root Tests for Panel Data, Journal of International Money and Finance, 20, 249-72. 20. Denison, E.W., 1962. Education, Economic Growth and Gaps in Information, The Journal of Political Economy, 70 (5), 124-28. 21. Devlin, N. and P. HANSEN, 2001. Health Care Spending and Economic Output: Granger Causality, Applied Economics Letters, 8, 561-64. 22. Domar, E. D., 1946. Capital Expansion, Rate of Growth and Employment, Econometrica, 14, 13747. 23. Engle, R. F. and C. W. J. GRANGER, 1987. Cointegration and Error-correction: Representation, Estimation and Testing, Econometrica, 55, 251-76. 24. Grossman, M., 1972. On the Concept of Health Capital and the Demand For Health, Journal of Political Economy, 80, 223-55. 25. Grossman, G. M. and E. HELPMAN, 1991. Innovation and Growth in the Global Economy, Cambridge, MA: MIT Press. 26. Hadri, K., 2000. Testing for Stationarity in Heterogeneous Panel Data, Econometrics Journal, 3, 148-61. 27. Hamoudi, A. A. and J. SACHS, 1999. “Economic Consequences of Health Status: A Review of the Evidence”, CID Working Papers Series No. 30.

REFERENCES 1. Acemoğlu, D. and S. JOHNSON, 2007, Disease and Development: The Effect of Life Expectancy on Economic Growth, Journal of Political Economy, 115 (6), 925-85. 2. Aghion, P. and P. HOWITT, 1992, A Model of Growth through Creative Destruction, Econometrica, March, 60, 323-51. 3. Aghion, P., P. HOWITT and F. MURTIN, 2010, The Relationship between Health and Growth: When Lucas Meets Nelson-Phelps, NBER Working Paper, www.development.wne.uw .edu.pl/uploads/ Courses/gtac_aghionetal_2010.pdf. 4. Andres, Aguayo-Rico, I. A. GUERRATURRUBIATES and R. D. OCA-HERNANDEZ, 2005. Empirical Evidence of the Impact of Health on Economic Growth, Issues in Political Economy, Vol. 14, www.org.elon.edu/ipe/aguayorico%20final.pdf. 5. Barro, R. J. and X.S. MARTİN, 2004. Economic Growth, MIT Press: Cambridge, MA. 6. Barro, R. J., 1996. Three Models of Health and Economic Growth, Unpublished Manuscript, Harvard University, Cambridge, MA. 7. Barro, R. J., 1997. Determinants of Economic Growth: A Cross-country Empirical Study, MIT Press, Cambridge, MA. 8. Barro, R.J. and J. W. LEE, 2010. A New Data Set of Educational Attainment in the World, 19502010, NBER Working Paper No. 15902. 9. Becker, G.S., 1962. Investing in Human Capital: A Theoretical Analysis, Journal of Political Economy, 70 (2), 9-49. 10. Bhargava, A., D. JAMISON, L. LAU, and C. MURRAY, 2001. Modeling the Effects of Health on Economic Growth, Journal of Health Economics, 718, 1-18. 11. Bloom, D.E. and J. G. WILLIAMSON, 1998. Demographic Transitions and Economic Miracles in Emerging Asia, World Bank Economic Review, 12 (3), 419-55. 12. Bloom, D., D. CANNING and P. MALANEY, 1999. Demographic Change and Economic Growth in Asia, CID Working Paper, 15, 1-60. 13. Bloom, D.E., D. CANNING and P. MALANEY, 2000. Demographic Change and Economic Growth in Asia, Population and Development Review, 26, 257-90.

12

28. Hansen, C. W., 2012. The Relation between Wealth and Health: Evidence from a World Panel of Countries, Economic Letters, Vol. 115, pp. 175-76. 29. Harrod, R.F., 1939. “An Essay in Dynamic Theory”, Economic Journal, 49 (193), 14-33. 30. Hassan, G. and A. COORAY, 2012. The Effect of Female and Male Health on Economic Growth: Cross-country Evidence within a Production Function Framework, University of Wollongong, Economics Working Paper Series 2012:07. 31. Im, K.S., M. H. PESARAN and Y. SHİN, 2003. Testing for Unit Roots in Heterogeneous Panels, Journal of Econometrics, 115, 53-74. 32. Johansen, S., 1988. Statistical Analysis of Cointegration Vectors, Journal of Economic Dynamics and Control, 12, 231-54. 33. Johansen, S., 1991. Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian Vector Autoregressive Models, Econometrica, 59, 1551-80. 34. Johansen, S. and K. JUSELIUS, 1990. Maximum Likelihood Estimation and Inference on Cointegration with Application to the Demand for Money, Oxford Bulletin of Economics and Statistics, 52, 169-210. 35. Kao, C., 1999. Spurious Regression and Residualbased Tests for Cointegration in Panel Data. Journal of Econometrics, 90, 1-44. 36. Kao, C. and M. H. CHIANG, 2000. On the Estimation and Inference of a Cointegrated Regression in Panel Data, in: Baltagi B. (ed.), Nonstationary Panels, Panel Cointegration, and Dynamic Panels, Advances in Econometrics, Amsterdam: JAI Press, 161-78. 37. Knowles, S. and D. P. OWEN, 1995. Health Capital and Cross-country Variation in per capita in the Mankiw-Romer-Weil Model, Economics Letters, Vol. 48, No. 223, pp. 99-106. 38. Knowles, S. and P. D. OWEN, 1997. Education and Health in an Effective-labour Empirical Growth Model, Economic Record, 73, 314-28. 39. Larsson, R. and J. LYHAGEN, 2007. “Inference in Panel Cointegration Models with Long Panels”, Journal of Business and Economics Statistics, 25, 473-83. 40. Levin, A., C. F. LIN, C. and C. J. CHU, 2002. Unit Root Tests in Panel Data: Asymptotic and Finitesample Properties, Journal of Econometrics, 108, 1-24.

41. Li, H. and H. LIANG, 2010. Health, Education and Economic Growth in East Asia, Journal of Chinese Economic and Foreign Trade Studies, 3 (2), 11031. 42. Lorentzen, P., J. MCMILLAN and P. WACZIARG, 2008. Death and Development, Journal of Economic Growth, 13 (2), 81-124. 43. Mackinnon, J. G., A. A. HAUG and L. MICHELIS, 1999. Numerical Distribution Functions of Likelihood Ratio Tests for Co-integration, Applied Econometrics, 14, 563-77. 44. Maddala, G.S. and S. WU, 1999. A Comparative Study of Unit Root Tests with Panel Data and a New Simple Test, Oxford Bulletin of Economics and Statistics, Vol. 61, pp. 631-52. 45. Mccoskey, S. and C. KAO, 1998. A Residual-based Test of the Null of Cointegration in Panel Data, Econometric Review, 17, 57-84. 46. McDonald, S. and Roberts, J. (2002). Growth and Multiple Forms of Human Capital in an Augmented Solow Model: A Panel Data Investigation. Economics Letters, 74, 271-76. 47. Mankiw, N.G., D. ROMER and D. N. WEIL, 1992. A Contribution to the Empirics of Economic Growth, Quarterly Journal of Economics, 107, 40737. 48. Mincer, J., 1974. Schooling, Experience and Earnings. National Bureau of Economic Research, New York. 49. Mushkin, S. J., 1962. Health as an Investment, Journal of Political Economy, 70, 129-57. 50. Nelson, R. and E. PHELPS, 1966. Investment in Humans, Technological Diffusion, and Economic Growth, American Economic Review, Papers and Proceedings, Vol. 51(2), 69-75. 51. Pedroni, P., 1999. Critical Values for Cointegration Tests in Heterogeneous Panels with Multiple Regressors, Oxford Bulletin of Economics and Statistics, 61, 653-70. 52. Pedroni, P., 2000. Fully Modified OLS for Heterogeneous Cointegrated Panels. In: Baltagi, B.H. (Ed.), Nonstationary Panels, Panel Cointegration and Dynamic Panels, Advances in Econometrics, 15, 93-130. 53. Romer, P. M., 1986. Increasing Returns and LongRun Growth, Journal of Political Economy, 94, 1002-37. 54. Romer, P.M., 1990. Endogenous Technological Change. Journal of Political Economy, 98, 71-102.

13

55. Sachs, J. and A. WARNER, 1997. Sources of Slow Growth in African Economies. Journal of African Economics, Vol. 6, No. 3, pp. 335-76. 56. Schultz, T.W., 1961. Investment in Human Capital. The American Economic Review, 51(1), 1-17. 57. Solow, R., 1956. A Contribution to the Theory of Economic Growth, Quarterly Journal of Economics, 70, 65-94. 58. Swan, T.W., 1956. Economic Growth and Capital Accumulation, Economic Record, 32 (2), 334-61. 59. WeiL, D. N., 2005. “Accounting for The Effect of Health on Economic Growth”, NBER Working Paper 11455, 1-58. 60. Westerlund, J., 2007. Testing for Error Correction in Panel Data, Oxford Bulletin of Economics and Statistics, Vol. 69, No. 6, 709-48.

14