Bootstrap ARDL on Health Expenditure, CO2 Emissions, and GDP Growth Relationship for 18 OECD Countries

Using annual time-series data over the period 1975–2017, the researcher applied the bootstrap autoregressive-distributed lag (ARDL) cointegration model developed by McNown et al. (1) to examine whether there is a long run relationship among health expenditure, CO2 emissions, and gross domestic product (GDP) per capita in 18 Organization for Economic Cooperation and Development (OECD) countries. We find cointegration exists in Netherlands when real GDP per capita serves as a dependent variable, in New Zealand when health expenditure is the dependent variable, and in the United States when CO2 emissions are dependent variables. The main results show evidence of a short run relationship between the three variables. The empirical results support that there is a bidirectional causality between health expenditure and GDP growth for Germany and the United States, between CO2 emissions and GDP growth for Canada, Germany, and the United States, and between health expenditure and CO2 emissions for New Zealand and Norway. The results also indicate that there are unidirectional causality in other countries.


INTRODUCTION
Climate change caused by increased greenhouse gas emissions not only causes temperature rise but also affects global precipitation. Climate change has become an indisputable fact that poses a serious threat to the sustainable development of human survival, society, economy, and the environment (2). There is an important link between high concentrations of greenhouse gases and climate change that there is the impact of climate change on public health. In recent decades, however, the relationship among economic growth, environment deterioration, and health expenditure has received increasing attention in the literature. Economic growth, environmental degradation, and health care expenditure vary between these relationships-complex and important. Therefore, the negative externalities of low environmental quality due to the impact on human health expenditure are ignored (3).
Health is one of the most important factors to decide the quality of human capital. There are many factors that can affect the health status of the population, such as environmental health, socioeconomic status, economic development, and the environment quality. As environmental quality deteriorates, the deterioration of global environmental quality poses a serious challenge to healthy living. Particles from the burning of fossil fuels, sulfur dioxide, and CO 2 emissions are a major contributor to global climate change, and this has been a topic of policy makers and a focus of researchers in many different fields of research (4).
Most of the literature in the past examined the pairwise relationship CO 2 emissions, gross domestic product (GDP), and health expenditure. Over the past decade, the first empirical study is based on Grossman and Krueger (5), and a number of researchers have followed the Environmental Kuznets Curve (EKC) between economic growth and environmental quality (6)(7)(8)(9)(10). The second pairwise of empirical study shows that many researches have focused on the relationship between health expenditure and GDP (11)(12)(13). The majority of previous economists have studied to estimate the health care demand about the size of income elasticity and allocate health care resources. Finally, the third pairwise of empirical study was on the relationship between CO 2 emissions and health expenditure (3,14,15). There are many researchers focusing on oneway causation from CO 2 emissions to health expenditure and finding a positive relationship. However, only a limited number of empirical studies have concentrated on two-way causation between CO 2 emissions and health expenditure. CO 2 emissions from Organization for Economic Cooperation and Development (OECD) countries account for half of global O 2 emissions, yet CO 2 emissions in OECD countries have been reduced since 2001. During this period, the OECD has made huge efforts to reduce CO 2 emission intensity. CO 2 emissions per unit of GDP decreased from 4.9 kg/USD in 2001 to 2.7 kg/USD in 2015. The share of CO 2 emissions in global CO 2 emissions also fell from 54% in 2001 to 38% in 2015 (13). Therefore, it is important to examine CO 2 emissions for health expenditures and economic development in OECD countries.
This article aims to examine the health, CO 2 emissions, and GDP growth nexus for 18 OECD countries by applying a bootstrap autoregressive-distributed lag (ARDL) test proposed by McNown et al. (1). This new method helps identify the occurrence of degenerate cases in ARDL tests. We, for the first time (compared to previous literature in OECD countries), used bootstrap models estimated by the ARDL test for long-and short-run relationships.
The main contribution of our research work to the existing studies is helpful for policy makers of OECD countries in particular and will provide some imperative understanding to other developed countries that include the 18 OECD countries. We used bootstrap ARDL approach for the long-run relationship and Granger causality tests to find directions of the links among the three variables. Therefore, this study aims at filling the gap in the literature by analyzing the health expenditure, CO 2 emissions, and GDP growth using the newly proposed econometric methodology to test.
The study will be separated into five sections. Section Literature presents a review of the literature. The data and methods used are described in section Materials and Methods. Section Results presents the results. Section Discussion has discussions. Section Conclusion concludes the research.

LITERATURE
The purpose of this study was to demonstrate the interaction between CO 2 emissions, health spending, and GDP growth in OECD countries. The pairwise correlation variables between the primary studies will be discussed.

CO 2 Emissions and GDP Growth
Most of the literature in the past concentrate on the relationship between environmental pollution and economic growth. First, the previous study was done to test the validity of the EKC hypothesis, which assumes that the relationship between economic growth and the environment is in the form of an inverted U. Selden and Song (9) used a nation's panel of data and the results confirm the inverted U. This hypothesis, however, was first proposed by Grossman and Krueger (5), where environmental degradation increases proportionally to income, then reaches a certain threshold when income reaches a stable point and then begins to decrease. Holtz-Eakin and Selden (16) examined marginal propensity to emit (MPE) and found that a diminishing MPE carbon dioxide as GDP per capita rises. Some authors (17)(18)(19)(20) have reported the existence of the EKC hypothesis. Not all results support the ECK hypothesis. Cole et al. (21) have found evidence against the EKC hypothesis.
Using the cointegration technique of ARDL bound test and Graager causality, Halicioglu (22) examined the feedback hypothesis between CO 2 emissions and economic expansion for Turkey. Saboori et al. (23) utilized ARDL bound approach to study the causal relationship between GDP growth and CO 2 emissions for Malaysia and found unidirectional causality from GDP growth to CO 2 emissions. Saboori and Sulaiman (24) determined a significant relationship between CO 2 emissions and economic growth that supported the EKC hypothesis in Singapore and Thailand for the long run. Jebli et al. (25) found that two-way causalities existed in the long run between CO 2 emissions and economic growth for 25 OECD countries. Cai et al. used the application of the newly introduced bootstrap ARDL bound test with structural breaks to examine the relationship among clean energy consumption, economic growth, and CO 2 emissions. They found the existence of a long-term and shortterm relationship between economic growth and CO 2 emissions for the G7 nations (6).

Health Expenditures and GDP Growth
The second part focuses on health expenditures and GDP growth. Much of the previous research has focused on estimating the impact of health care income elasticity on scale, except for distribution of health care resources.
Gerdtham and Löthgren test the cointegration between health expenditure and GDP in the 25 OECD nations for the period 1960-1997. The outcomes revealed that health expenditure and GDP are cointegrated for the 12 OECD nations (26). Baltagi and Moscone (27) also considered OECD countries using a panel of data to reconsider the long-run relationship between health care expenditure and income for the period 1971-2004. Their findings suggest that health care, a necessity, is in elastic demand.
Wang applied the quantile panel-type analysis to determine the causality between health expenditure and economic growth from 1986 to 2007. The findings indicated that the estimation of quantile regression of the effect of health expenditure growth on economic growth is different (28). Albulescu et al. utilized health care spending for six EU member countries for the time period 1972-2013. The findings indicated that no significant convergence occurred with regard to the ratio of health care expenditure to GDP (29). Erçelik and Murthy and Okunade used ARDL to estimate the relationship between health expenditure and economic growth (30,31).

Health Expenditures and CO 2 Emissions
The third research strand on the issue that has limited literature regarding the relationship between CO 2 emissions and health expenditure discusses the relationship between environmental pollution (carbon emissions) and health expenditure. Only a few studies have focused on the relationship among health expenditures, CO 2 emissions, and economic growth. Chaabouni et al., Chaabouni and Saidi, and Chaabouni and Abdnnadher provide evidence of causality among health expenditures, CO 2 emissions, and economic growth (33)(34)(35). Zaidi and Saidi showed a one-way causal relationship from health expenditure to GDP and a two-way causal relationship between health expenditure and CO 2 emissions in sub-Saharan African countries (36). Ghorashi and Rad applied dynamic simultaneous equation model to investigate health expenditures, CO 2 emissions, and economic growth in Iran. The empirical findings supported the presence of two-way causality flowing among CO 2 emissions and economic growth and one-way causality from health expenditure to economic growth (37). Wang (28) used the ARDL model for the time period 1995-2017 to examine the dynamic linkages among CO 2 emissions, health expenditures, and economic growth in Pakistan. The results of their empirical study show a long-run relationship among the three variables and a two-way relationship of causality between health expenditure and CO 2 emissions and between health expenditure and economic growth, as well as a one-way causal relationship from CO 2 emissions to health expenditure.

MATERIALS AND METHODS
This study covers 18 OECD countries between 1975 and 2017. The 18 OECD countries selected for this study are Australia, Austria, Belgium, Canada, Denmark, Finland, Germany, Ireland, Japan, Netherland, New Zealand, Norway, Portugal, Spain, Sweden, Turkey, the United Kingdom, and the United States. Real GDP per capita (GDP) are obtained from the World Development Indicators, World Bank. Per capita real health expenditure (HE) in U.S. PPP dollars older came from OECD Statistics. CO 2 emissions (CO 2 ) data are retrieved from The BP Statistical Review of World Energy (available at: http://www.bp. com/). All variables are transformed into natural logarithm form.

Research Framework
The ARDL model is as follows: i and j are the indicators of the lag period, i = 1, 2..., k; j = 0, 1..., k; h = 0, 1..., p. t represent time t = 1, 2,..., T. The y t in the equation is the explanatory variable, x t and z t are the explanatory variables, and the variable S t,j is a dummy variable. The parameters α i , β I , and γ i are the coefficients of the lag of y i , x i , and γ i . The error term is µ t , and Equation (2) can be rewritten and expanded into the following equation: i=0 γ i and other parameters are the function values of the original parameters in Equation (1).
McNown et al. showed the bootstrap method to the ARDL tests of cointegration and proposed a cointegration that requires rejecting the F-test (denoted as (28). They showed that two degenerate situations can arise. Situation 1 occurs when F ′ and t are significant but F ′′ is not significant. In this situation, the joint significance of the error correction terms is due solely to the lag of the dependent variable. Thus, the explanatory variables are not part of the long-run cointegrating relation. The other situation, degenerate situation 2, occurs when F ′ and F ′′ are significant but t is not significant. McNown et al. added a test on the lagged independent variables (F-test denoted as F ′′ ) to complement the existing F ′ and t-tests for cointegration as proposed by Pesaran et al. (1,38).
McNown et al. showed that the bootstrap ARDL test has better size and power properties than the asymptotic test in the ARDL bound test. After testing the long-term relationships, we concluded that short-run causal relationship will be determined by standard Granger causality tests if there is no cointegration relationship among y, x and z. We used the Granger causality test for x and z, which should include the lagged difference on x or z, and we tested a 2,j = 0 or a 3,j = 0 in Equation (2). However, if there is cointegration between the dependent variable and the independent variable, they will form a fixed linear combination. In this case, the short-term relationship test should include the hysteresis difference of x or z and the hysteresis level of x or z, that is, test a 2,j and b 2 or a 3,j and b 3 .

RESULTS
The ARDL bound test allows the variables with different orders of integration [I(0) or I(1), not for I(2)] (29). The Augmented Dickey-Fuller (ADF), Kwiatkowski-Phillips-Schmidt-Shin (KPSS), and Zivot-Andrews (ZA) unit root tests were employed to test the integration level (39)(40)(41). We found that all series have a unit root at 5% level of significanec, but all variables are stationary in the first differences when AD, KPSS, and ZA tests are used. To conserve space, the unit root test results are not shown here. Table 1 provides the results of the Bootstrap ARDL tests. This study uses the two F-test (F c ′ and F c ′′ ) and one T-test (t c ) as critical values for determining the long-run forcing variable found in McNown et al. We believe that only several estimated equations are able to reject the null hypothesis of the t-test and both F-tests simultaneously. We could only find cointegrated evidence for Netherlands when GDP serves as the dependent variable, for New Zealand when health expenditure is the dependent variable, and for the United States when CO 2 emission is the dependent variable. Degenerated case 2 also occurs for Finland, where both the F ′ -test and F ′′ -test are statistically significant but not the t-test. Besides, we believe that no cointegration can be revealed by bootstrap ARDL test among the GDP, health expenditures, and CO 2 emissions for Australia, Austria, Belgium, Canada, Denmark, Germany, Ireland, Japan, Norway, Portugal, Spain, Sweden, Turkey, and the United Kingdom. Although not many countries show a longrun relationship between GDP, health expenditures, and CO 2 emissions, however, some causality patterns are identified in the short run.

Short-Run Granger Causality Test
Based on the results from ARDL cointegration test, we test the causality among the variables for 18 OECD countries by standard Granger causality test to indicate the casual relationship among health expenditure, GDP growth, and CO 2 emissions. From Table 2, we can see that a feedback exists between health expenditure and GDP, between health expenditure and CO 2 emissions, and between CO 2 emissions and GDP.
We find a two-way Granger causality running between health expenditure and GDP for Germany and the United States. This finding supports the past empirical studies (42)(43)(44). The results from Granger causality tests indicate the existence of bidirectional causality between health expenditure and CO 2 emissions for New Zealand and Norway. Studies look at the causal relationship between health expenditures and GDP growth that, in most of these countries, have bidirectional causality (35). We report the causal link between CO 2 emissions and GDP growth for Canada, Germany, and the United States by employing a bootstrap ARDL approach. In the short running analysis, many studies also find evidence to support the twoway Granger causality between CO 2 emissions and GDP growth (28,37,45,46).
There was unidirectional causality running from GDP to health expenditures for Belgium, Denmark, Finland, New Zealand, Sweden, Portugal, and Spain. The shortrun relationship Granger causality was found from health expenditure to CO 2 emissions for Austria, Belgium, Sweden, and the United Kingdom, and from CO 2 emissions to health expenditures in Germany, Ireland, Japan, Netherlands, and the United States, and these countries exhibit only unidirectional causality. In the short run, it is unidirectional causality. Granger causality tests indicate that CO 2 emissions Granger causes GDP in Finland and Netherlands, and GDP Granger causes CO 2 emissions in Australia, Denmark, Japan, Spain, Sweden, and the United Kingdom.

DISCUSSION
The overall results revealed that there is no long-run relationships in the 18 OECD countries. There is no long-run relationship among health expenditures, CO 2 emissions, and GDP growth that we support Atay and Ergun that we get the same result in Turkey (40). Wang et al. found a long-run relationship among CO 2 emissions, health expenditure, and GDP in Pakistan (47). The researchers examined health expenditure, environmental pollution (CO 2 emissions; nitrous oxide emissions), and GDP growth and indicated the long-run relationship among them in the sub-Saharan African countries (36). We only report a long-run relationship among health expenditure, CO 2 emissions, and GDP growth in Netherlands, New Zealand, and the United States. There is important available that examined the link between health expenditure, CO2 emissions, and GDP growth in Netherlands, New Zealand and the United States.
Furthermore, we also estimated the Granger causality based on the ARDL model. The short-run relationship among the three variables, the coefficient value for CO 2 emissions, reports a positive and significant effect with health expenditures in Ireland, Netherlands, United States, New Zealand, and Norway that means CO 2 emissions increase health expenditures. The important policy is that the government actively limits carbon emissions because environmental pollution will increase health expenditures. The coefficient for GDP growth is showing the positive and significant effect between GDP and health expenditures, which shows that in the short run, the increasing level of GDP increases health expenditures. GDP growth sacrifices health factors because economic development may The parentheses is optimal lag order based on Akaike Information Criterion (AIC). D84 means a dummy variable for the year 1984; other years are 0. F ′ is the F-statistics for the coefficients of HE t−1 , GDP t−1 and CO 2 t−1 ; t_dep is the t-statistics for the dependent variable, and F ′′ is the F-statistics for the independent variable. F C ′ , t C and F C ′′ are the critical values at the 10% significance level, generated from the bootstrap program. Bold value represents Cointegration results.  The asterisks ***, **, and * denote the significance at the 1, 5, and 10% levels, respectively. Additionally, the parentheses are p-value and sign for the coefficients. The case of non-cointegration and its causality test involved only lagged differenced variables. Bold value represents Granger causality results.
cause environmental pollution, and low-quality human capital forms more health losses. However, there do exist negative relations with short-run relationship running from GDP growth to health expenditures for a number of countries, including Germany and the United States. This may be the reason of the negative link between GDP and health expenditure for the case of Germany and the United States because human capital and institutional quality are in good condition. When growth increases, people's investment in physical exercise reduces the expenditure for health. There are negative relations with short-run relationship from health expenditures to GDP growth for Germany but positive relations for the United States. Increasing health expenditures is associated with low productivity of human being because health creates disturbance for the economic growth (48). Health expenditures cannot lead to social security, efficient resource allocation, and better economies of scale in Germany (37). The human capital stock is positive and statistically significantly affects GDP growth in the United States (34). CO 2 emissions have a positive and significant relation to GDP growth in Finland, Netherlands, Canada, and Germany. These countries have a high intensity of CO 2 emissions due to the use of fossil fuels and rapid GDP growth. However, energy consumption and CO 2 emissions do not increase the GDP growth of the United States but have a negative effect.
This study has found that health expenditure has a significant and positive effect on CO 2 emissions at the level of 5% in Austria and Norway. In addition, CO 2 emissions are negatively and significantly affected by health expenditures in Belgium, Sweden, the United Kingdom, and New Zealand. There are two effects on health expenditures that affect carbon emissions. The effect of health expenditures is related to population growth, thus as long as population increases, the rise in energy consumption results in greater environment pollution (33). On the other hand, health expenditures increase people's awareness of pollution and reduce carbon emissions. The first effect is significantly larger than the second effect in Austria and Norway and vice versa in Belgium, the United Kingdom, and New Zealand. In addition, GDP has a positive and significant effect on CO2 emissions in Canada, Denmark, Japan, and Spain but a negative effect on CO 2 emissions in the United Kingdom, Australia, Germany, and the United States.

CONCLUSIONS
This study contributes to the literature in the dynamic relationship among CO 2 emissions, health care expenditure, and GDP growth for the 18 OECD countries over the period of 1975-2017. We investigate the bidirectional causal relationship between health expenditure and GDP and between CO 2 emissions and GDP and between health expenditure and CO 2 emissions for the three ARDL models. The results show that the bootstrap ARDL analysis is useful in evaluating the effects of different socioeconomic scenarios in the context of different positions for health expenditure and environment policy in OECD countries.
Our study results showed that there is cointegration existing in Netherlands when real GDP per capita serves as the dependent variable, in New Zealand when health expenditure is the dependent variable, and in the United States when CO 2 emission is the dependent variable. The short-run relationship exists in a unidirectional causality relationship from CO 2 emissions to health expenditure for Germany, Ireland, Japan, Netherlands, and the United States and from health expenditure to CO 2 emissions for Austria, Belgium, Sweden, and the United Kingdom. There is a bidirectional causality relationship between health expenditure and CO 2 emissions in New Zealand and Norway. This empirical result supports previous researches (33,34,47). There is a unidirectional causality relationship from GDP to health expenditure for Belgium, Denmark, Finland, New Zealand, Sweden, Portugal, and Spain. There is a bidirectional causality relationship between GDP and health expenditure for Germany and the United States (42)(43)(44). It is unidirectional causality from CO 2 emissions to GDP for Finland and Netherlands, and GDP Granger causes CO 2 emissions for Australia, Belgium, Denmark, Japan, Spain, and the United Kingdom. Our study results showed the bidirectional causality relationship between Canada, Germany, and the United States (28,37).
The most important policy of these OECD countries is to limit CO 2 emissions. In addition, these countries should adopt measures and policies to protect the quality of the environment to reduce the occurrence of health diseases. Environmental quality as a contributing factor to increased health spending has been proven. Our results demonstrate a significant increase in health spending on CO 2 emissions in Ireland, Netherlands, the United States, New Zealand, and Norway. In some countries, health factors are sacrificed because of GDP growth because economic development may cause environmental pollution and low-quality human capital will cause more health losses. For some economies, however, including Germany and the United States, Granger causality does have a negative correlation from GDP to health spending. This may be the reason for the negative relations between GDP and health expenditure in Germany and the United States, as institutional quality and human capital are in good condition. When growth increases, people's investment in physical exercise reduces health spending. Therefore, environmental degradation is becoming a contemporary issue, and we urgently need to reexamine health and environmental policies so that GDP growth should not increase at the expense of public health or the environment. Due to the consumption of fossil fuels by OECD countries, industrial and household natural resources contribute significantly to CO 2 emissions. It seems that policy makers should study the need for investment to promote technology transfer to protect environment quality.

DATA AVAILABILITY STATEMENT
The datasets generated for this study are available on request to the corresponding author.