The Causal Nexus among Energy Consumption, Environmental Degradation, Financial Development and Health Outcome: Empirical Study for Pakistan

Abstract

Pakistan is facing an energy crisis and is also severely affected by climate change. Moreover, Pakistan is not doing well as far as health outcome indicators are concerned. The causal nexus among energy, environment, and health outcomes is well-established in literature. Besides, financial development also grabs the attention of health outcome literature as financial development can play a significant role in improving health outcomes. Thus, this study was conducted to test the causal nexus among energy consumption, environmental degradation, financial development, and health outcomes in the case of Pakistan. This study proxies health outcomes with life expectancy and infant mortality. Time series data have been analyzed through different econometric techniques, such as unit root tests, cointegration techniques, causality techniques, and cointegration regressions. Moreover, this study not just discovers the causal direction among variables but also determines the strength of causality through variance decomposition. Results of the study confirm that all variables of the study are cointegrated in the long run. The causality analysis reveals that unidirectional causality is running from energy consumption and environmental degradation to health outcomes, whereas bidirectional causality is found between financial development and health outcomes in the long run. Besides, this study also determines the effect of energy, environmental degradation, and financial development in the health outcome model and finds that energy and financial development can help Pakistan to improve health outcomes. Policy implications are recommended for Pakistan.

1. Introduction

Sound and persistent economic growth is essential for developing countries to address socio-economic problems such as food security, poverty, and a lack of basic health facilities. However, a sufficient energy supply is necessary for economic growth and development as energy is considered the backbone of economic development. The use and adaptation of modern energy will ensure economic development as energy enhances welfare as it positively affect and improves sectors such as agriculture, education, health, manufacturing, commerce, and trade. However, most developing countries are fulfilling their energy needs with fossil fuels to uplift living standards, to generate job opportunities, and for sound economic growth. Masses of people of most of developing countries are fulfilling their outdoor and indoor energy requirements with fossil fuels and solid fuels, while compromising their health as these fuels are producing toxic gases that have adverse impacts on human health [1,2,3]. As energy requirements intensify, this is challenging health, welfare, and economic problems. Henceforth, developing countries have to maintain sustainable economic growth to tackle health and environmental challenges [4]. In a study they carried out, Youssef et al. [5] argued that energy consumption and health outcomes have a strong relationship. They emphasized that access to energy produces positive health effects as it reduces absenteeism in schools and, moreover, enhancing energy efficiency will ensure an increase in life expectancy and result in a reduction of infant mortality.

In literature, several channels have been identified for how environmental degradation affects health outcomes. For instance, air pollution is causing asthma, lung cancer, and heart diseases [6,7], and is also responsible for premature mortality [8]. Besides, environmental degradation leads to waterborne diseases such as dengue fever and malaria [9,10,11] and, as environmental degradation adversely affects human health, this consequently leads to low labor productivity [12] as adverse effects of environmental degradation enhance economic costs in the form of efficiency loss and treatment [13]. On the other hand, availability of health facilities leads to an increase in public health expenditure and less funds are thereafter left for education and environmental issues; health expenditures may enhance environmental degradation through crowding-out effects [12], whereas Apergis et al. [6] argued that increases in health expenditures degrade environmental quality because of a rise in energy demand.

Some research studies [14,15,16,17] reached the conclusion that environmental degradation is responsible for the variability of the ecosystem, thus increasing the probability of droughts and floods. Henceforth, environmental degradation might bring adverse variations in food supply and water quality, consequently resulting in high mortality, especially among the elderly and infants [12]. Air pollution badly affects the longevity of elders [15,16], whereas Majeed and Ozturk [17] concluded that infant mortality is higher in those countries where environmental degradation is higher. In a similar fashion, Rahman, Rana, and Khanam [18] examined not only the causal relationship between life expectancy and environmental degradation but also examined the short-run and long-run effects of environmental degradation on life expectancy. They found unidirectional causality from environmental degradation to life expectancy. Results of their study reveal that environmental degradation has negative and significant effects on life expectancy in both time spans, i.e., short-run and long-run.

Financial development is associated with human welfare as economic theory explains divergent forecasts and explains that financial development affects human welfare as it may reduce poverty and vulnerability to shocks, and enhance human capital [19]. Financial development improves human capital in the same manner as it nurtures physical capital [20,21]. In their study, Giovannini, Iacopetta, and Minneti [22] described that access to credit, savings facilitations, risk management, low transactions costs, and sharing of information are different channels that explain how financial development can affect human capital positively. Whereas, Claessens and Feijen [23], Hakeem and Oluitan [21], and Bhatta [24] opined that financial development improves health outcomes through several channels, namely access to credit, risk management, income generation, and improving education. Likewise, Pagano [25], and Jeanneney and Kpodar [26] emphasized that at the household level, financial development improves health outcomes because of risk management, access to credit, and income generation. Besides, financial development also enhances access to credit for firms and government; thus, it leads to higher expenditures on health care to improve health quality as higher expenditures ensure care and invest in medical personnel and therefore help in curing disease [27]. Furthermore, financial development enhances tax revenues for governments as it helps in reducing the size and share of the informal sector in the economy. On the other hand, Friedman’s revenue-spend hypothesis postulates that an increase in revenue leads to an increase in government spending; thus, it raises a government’s capacity to investment more in human capital and also enables a government to subsidize human capital investments [28].

Pakistan is facing an energy crisis and is also severely affected by climate change. Moreover, Pakistan is not doing well as far as health outcome indicators are concerned [29]. Figure 1 presents the life expectancy in selected South Asian countries. The life expectancy in Pakistan was better than in Bangladesh and India in 1972, whereas in Sri Lanka life expectancy was higher in 1972 and still higher in 2019 compared to in other South Asian countries [30]. Bangladesh experienced an increase in life expectancy from 46 to 69 during 1972–2019, whereas India’s life expectancy increased from 49 to 67 in the same period. As far as Pakistan is concerned, its life expectancy was higher than Bangladesh’s and India’s in 1972, but Pakistan did not perform well compared to other countries in the region and has a lower life expectancy compared to Bangladesh, India, and Sri Lanka. Figure 2 presents the infant mortality rates in selected South Asian countries. Bangladesh’s infant mortality rate was higher than India’s, Pakistan’s, and Sri Lanka’s. However, during 1972–2019, the infant mortality rate dropped for Bangladesh by almost by six times as it was 147 and 25 in 1972 and 2019, respectively. During the same period, India also experienced a decrease in its infant mortality rate by almost five times. The infant mortality rate for Sri Lanka also decreased as it was 51 in 1972 and it dropped to just 6 in 2019. As far as its infant mortality rate is concerned, Sri Lanka is performing better than any other South Asian country, whereas Pakistan is in a worse condition. In 1972, among the selected South Asian countries, Pakistan was third; however, it has not done well and in 2019, the infant mortality rate was 55, which is quite high compared to Bangladesh, India, and Sri Lanka.

There are several good reasons to test the causal nexus among energy consumption, environmental degradation, financial development, and health outcomes in the context of Pakistan. First, Pakistan has been facing an energy crisis since the early 2000s and it adopted different energy policies to tackle the crisis. Second, Pakistan is among the top 10 climate-change affected countries. Third, it is evident from Figure 1 and Figure 2 that Pakistan has a lower life expectancy and higher infant mortality than other South Asian countries (Bangladesh, India, and Sri Lanka). Thus, these facts provide a good background to conduct a research study on the causal nexus among energy consumption, environmental degradation, financial development, and health outcomes in Pakistan. This a unique study for Pakistan as far as this causal nexus is concerned, as researchers did not trace any other research paper for Pakistan that tested the causal link of energy consumption, environmental degradation, and financial development with health outcomes. Additionally, this study also examines the effect of energy, environmental degradation, and financial development on health outcomes.

This research study is arranged in such a manner that the next section describes the literature review, while the third section presents research material and methods. Results and discussion are presented in the fourth and fifth sections, respectively. The last section concludes the study and also the policy implications are discussed in this section.

2. Literature Review

Empirical studies tested the causal relationship between energy and health outcomes [5,31,32]. The empirical study [5] investigated the causal relationship between energy consumption and health outcomes, i.e., life expectancy and infant mortality for 16 selected African countries over the period from 1971 to 2010. The results of this study documented a strong nexus between energy consumption and health outcomes. Moreover, the results of this study showed that there is a unidirectional causality from energy consumption to life expectancy, as well as to infant mortality in Algeria, Benin, Congo, Egypt, Senegal, South Africa, and Morocco, whereas bidirectional causality is found between energy and life expectancy and also between energy consumption and infant mortality in the case of Nigeria [5]. Likewise, Rahman and Alam [31] examined the causal relationship between energy consumption and health outcomes in 10 emerging countries and concluded unidirectional causality from energy to health outcomes. A study [32] that examined the causal nexus among energy consumption, economic growth, and health outcomes in the context of sub-Saharan African countries did not confirm any causal relationship between energy and health outcomes; however, it concluded that energy consumption and economic growth jointly influenced the Granger Causality in health outcomes. On basis of literature discussed here, the first hypothesis of the study is developed that is as follows:

Hypothesis 1 (H1).

Is there causal nexus between energy consumption and health outcomes?

In a similar fashion, empirical studies analyzed the causal relationship between environment degradation and health outcomes. For instance, Balan [33] investigated the causal nexus between environment degradation and health outcomes. This study considered life expectancy and health expenditures as proxies for health outcomes, as it argued that life expectancy is a vital factor in the health outcomes and economic development of a country. This study established a long-run relationship between environment degradation and health outcomes and concluded there was a bidirectional causality between environment degradation and health outcomes. Likewise, Sinha [34] also confirmed bidirectional causality between environment degradation and health outcomes, where Sinha used infant mortality and child mortality as proxies for health outcomes. Besides, Mariani et al. [35] hypothesized a bidirectional causality between environment degradation and health outcomes, whereas Rjoub et al. [36] concluded that there is a positive co-movement between environment degradation and health outcomes and that environmental degradation influences health outcomes. So, following the studies [33,34,35,36] among others, second hypothesis of the study is as follows:

Hypothesis 2 (H2).

Is there unidirectional causality from environmental degradation to health outcomes?

Claessens and Feijen [23] carried out a study to examine the effect of financial development on health outcomes and reached the conclusion that a strong positive nexus is present between financial development and life expectancy. Bhatta [24] also determined the impact of financial development on health outcomes in a comparative study for low-income Asian and high-income OECD countries. In both panels of countries, his study results documented a positive and significant effect of financial development on health outcomes, as financial development is positively related to life expectancy and negatively related to infant mortality. However, his study revealed that the magnitude of the impact of financial development was higher for low-income countries than for high-income countries. Thus, he concluded that financial development can play a vital role in health outcomes in developing countries. In a similar fashion, Rewilak [37] also reached the same conclusion that financial development can play a significant role in health outcomes, as he found that financial development negatively impacts infant mortality. The causal relationship between financial development and health outcomes is supported in different empirical studies [38,39]. A study conducted by Prasad et al. [40] concluded that financial development positively influences socio-economic indicators, including health outcomes. On basis of the empirical studies discussed here, the third hypothesis of the study is developed that is as follows:

Hypothesis 3 (H3).

Is there bidirectional causality between financial development and health outcomes?

3. Materials and Methods

This study was designed to examine the nexus among energy consumption, environmental degradation, financial development, and health outcomes. In addition to this, the current study also examines the effect of energy consumption, environmental degradation, and financial development on health outcomes in Pakistan. For the determinants of health outcomes, this study is based on Grossman’s health demand model [41]. Grossman’s health demand model assumes that health is a capital good. Every individual is born with some initial health inventory that depreciates with the passage of time and decreases in the health inventory results in mortality. However, medical care [42] and better environmental quality can improve health [17]. Henceforth, the health outcome (HOC) function can be expressed as:

$$HOC=f \left(X\right)$$

(1)

where X is the vector of variables that determines HOC. As discussed earlier, this study measures health outcomes through life expectancy and infant mortality. From the macroeconomic perspective, different determinants of HOC are determined in the literature, such as financial factors [43], energy factors [44], and environmental factors [43]. Thus, the HOC function is then developed for this study as follows:

$$HOC=f \left(EC, ED, FD\right)$$

(2)

where EC, ED, and FD denote energy consumption, environmental degradation, and financial development, respectively. As this study uses two proxies, life expectancy and infant mortality, for health outcomes, this study will henceforth estimate the following two models as shown in Equations (3) and (4) as follows:

$$LXP=f \left(EC, ED, FD\right)$$

(3)

$$IMR=f \left(EC, ED, FD\right)$$

(4)

where LXP and IMR denote life expectancy and infant mortality, respectively. In order to estimate the models, a double-log model is applied and the estimated models are presented in Equations (5) and (6) as follows:

$$lnLX{P}_t={\alpha }_0+{\gamma }_{1}lnE{C}_t+{\gamma }_{2}lnE{D}_t+{\gamma }_{3}lnF{D}_t+{\epsilon }_t$$

(5)

$$lnIM{R}_t={\beta }_0+{\delta }_{1}lnE{C}_t+{\delta }_{2}lnE{D}_t+{\delta }_{3}lnF{D}_t+{\epsilon }_t$$

(6)

where ${\alpha }_0$, ${\beta }_0$, and ${\epsilon }_t$ are intercepts and error terms in the respective models, and ${\gamma }_i$ and ${\delta }_i$ represent coefficients of the respective variables in the models, and $t$ denotes the time period of the study. The reason for taking the natural log of variables of this study is, first, to marginalize the effect of explanatory variables on dependent variables so that coefficients can be interpreted in percentages. Second, considering the natural log can help to address the problem of heteroscadasticity, if present.

The first step in analyzing the time series data is to check for a unit root problem as, in general, most of the time series data contain unit root problems, which depict that the mean, variance, and co-variance of the data change over time. This implies that when a series carrying a unit root is regressed it yields misleading results. To remove the problem of the unit root in time series data, the augmented Dickey–Fuller (ADF) test developed by Dickey and Fuller [45] is used, for which the general model is given below:

$$∆Y_t={\alpha }_1+{\displaystyle \sum }_{j=1}^n\varnothing ∆Y_{t-j}+e_t$$

(7)

The hypotheses given below are tested through the statistical probability value of “$\varnothing$”, the hypotheses for the ADF test are given as follows:

Ho: the data is non-stationary and there is unit root,

Ha: the data is stationary and there is no unit root.

The problem associated with the conventional unit root testing, including ADF, is that they do not take the structural breaks into account. Time structural break is assumed as the exogenous phenomenon and the intensity and power to reject the hypothesis of unit root decreases when the alternative of stationarity is true and when the time structural break is ignored [46]. Zivot and Andrews (ZA) [47] proposed a unit root test with a single structural break in which they assumed that the exact period of the structural break is unknown. They proposed three models to test the unit root, which are depicted in Equations (8)–(10) as below:

$$∆Y_t=c+\alpha y_{t-1}+\beta t+\gamma D{U}_t+{\displaystyle \sum }_{j=1}^k{d}_j∆y_{t-j}+{\mu }_t$$

(8)

$$∆Y_t=c+\alpha y_{t-1}+\beta t+\theta D{T}_t+{\displaystyle \sum }_{j=1}^k{d}_j∆y_{t-j}+{\mu }_t$$

(9)

$$∆Y_t=c+\alpha y_{t-1}+\beta t+\theta D{T}_t+\gamma D{U}_t+{\displaystyle \sum }_{j=1}^k{d}_j∆y_{t-j}+{\mu }_t$$

(10)

where $D{U}_t$ shows the dummy variable for a mean shift occurring at each possible break date and $D{T}_t$ depicts the trend shift variable. Equation (8) depicts the first model for the ZA structural break unit root test; this model allows a one-time change in the level of the series. Equation (9) shows the second model, which permits a one-time change in the slope of the trend function, and Equation (10) combines both one-time changes in level with the series and slope of the trend. The hypotheses for the ZA’s unit root test are given below:

Ho: $\left(\alpha =0\right)$ series has a unit root with a drift excluding structural break,

Ha: $\left(\alpha <0\right)$ series is trend-stationary with a one-time break.

The combined movement of the variables in the long run is generally referred to as cointegration. Several time series approaches are used to find the cointegration of variables in which Johansen’s cointegration test [48] is used to discover the cointegrating vectors. According to this cointegration test, for $k$ number of variables it is more likely that there will be $k-1$ cointegrating vectors. These cointegration vectors refer to the long-run equilibrium relationship of the variables. Cointegration is a characteristic of two series, where both share a common stochastic drift. Stochastic drift is the change in average value of the random or stochastic process. The advantage of the Johansen cointegration test is that it can presume more than a one-time series cointegration of variables. The general model for this test is given as follows:

$$∆y_t=\Pi y_{t-1}+{\displaystyle \sum }_{i=1}^{p-1}{\Gamma }_i∆y_{t-i}+\beta x_t+{\epsilon }_t$$

(11)

Although the Johansen cointegration can determined in a long-run relationship, if a structural break is present then one needs to have a cointegration technique that can take care of the structural break. The autoregressive distributive lag (ARDL) model [49] is such a technique that can account for structural breaks. It is widely considered for the testing of cointegration, and the main advantage of the ARDL model is that it can take variables for different levels of integration $\left\{\right(Y~\left(0\right) and X~\left(I\right)\}$. The main characteristic of the ARDL model is that it derives both the long- and short-run relationships of the variables. The general model specified of the ARDL is displayed in Equation (12) as follows:

$$\begin{array}{cc}∆Y_t={\beta }_0& +{\displaystyle {\displaystyle \sum }_{i=1}^m}{\beta }_{1i}∆Y_{t-i}+{\displaystyle {\displaystyle \sum }_{j=0}^n}{\beta }_{2j}∆X_{1t-j}+{\displaystyle {\displaystyle \sum }_{k=0}^q}{\beta }_{3k}{X}_{2t-k}+{\displaystyle {\displaystyle \sum }_{p=0}^o}{\beta }_{4p}∆X_{3t-p}\dots \dots \dots \dots +{\beta }_{if}{\displaystyle {\displaystyle \sum }_{f=0}^e}∆X_{it-f}+{\phi }_1{Y}_{t-1}\hfill \\ & +{\phi }_2{X}_{1t-1}+{\phi }_3{X}_{2t-1}+{\phi }_4{X}_{3t-1}\dots \dots \dots \dots \dots .+{\phi }_8{X}_{it-1}+\pi EC{T}_{t-1}+D_t+u_t\hfill \end{array}$$

(12)

where ${\beta }_0$ is the intercept term; all the${\beta }_i$ depicting the short-run coefficients and all the ${\phi }_s$ are displaying the long-run coefficients. The ECT term describes the adjustment effect and the significance of its coefficient $\pi$ will show the dynamic stability of the overall model. $D_t$ is an exogenous-treated variable and represents a dummy for the unknown structural break.

Once the order of integration and cointegration is determined, one can move on to test the causal relationship among the variables of the study. It is quite apparent that the cointegration refers to the causality between two time series variables, but it still does not specify the direction of causality. Engle and Granger [50] stated that the presence of cointegration refers to a unidirectional or bidirectional causal relationship between variables. Furthermore, Engle and Granger presented an error correction mechanism that can be estimated by a standard estimation procedure. The general model for the vector error correction model (VECM) is given as follows:

$$∆Y_t={\alpha }_1+p_{1}ecm{1}_{t-1}+{\displaystyle \sum }_{i=0}^n{\beta }_i∆Y_{t-i}+{\displaystyle \sum }_{i=0}^n{\delta }_i∆X_{t-i}+{\displaystyle \sum }_{i=0}^n{\gamma }_i∆Z_{t-i}+{\epsilon }_{1t}$$

(13)

$$∆X_t={\alpha }_2+p_{2}ecm{2}_{t-1}+{\displaystyle \sum }_{i=0}^n{\beta }_i∆Y_{t-i}+{\displaystyle \sum }_{i=0}^n{\delta }_i∆X_{t-i}+{\displaystyle \sum }_{i=0}^n{\gamma }_i∆Z_{t-i}+{\epsilon }_{1t}$$

(14)

where ${\beta }_i, {\delta }_i$, and ${\gamma }_i$ depict the short-run coefficients, p is the order of lag, and ecm1 and ecm2 show the error correction terms. Furthermore, $ecm{1}_{t-1}$ is the lagged value of the residual, which is derived from the regression of Y over X, and $ecm{2}_{t-1}$ is the lagged value of the residual of the X over Y regression. Long- and short-run causality can be captured through the VECM procedure.

Although the determination of causal direction among variables can be detected through VECM if variables of the model are cointegrated [49], Shahbaz et al. [51] argued that causality through VECM failed to control the comparative strength of the effects of causality beyond a selected period. Thus, this study will also adopt an innovative accounting approach (IAA). The IAA technique of causal relationship not only accounts for the integration of a series, but also takes care of the endogeneity problems associated with cointegrated variables. The most notable advantage of the IAA technique is that it is analyzes the causal relationship ahead of the selected sample time period. In simple words, this IAA technique forecasts the causal relationship results. The two main components known in literature are forecast error variance decomposition and the impulse response function of the IAA technique. As discussed earlier, we are now interested in the magnitude of the causal relationship; thus, this study restricts itself to the forecast variance error and variance decomposition developed by [52]. It is based on the Cholesky technique and it analyzes the effects of shocks on the adjustment time path of variables over time. Variance decomposition shows the response of the endogenous variable when a shock is given to itself and other variables in the system over periods of time [53] and, moreover, the variance decomposition function identifies the time path of the effects of shocks in the system of the vector autoregressive (VAR) model. Enders [54] postulated that each variable in a system of equations depends on the shock of standard deviation and forecast error variance and this expresses a causal relationship among variables. The general VAR model for these analyses is given below:

$$Y_t=A_1{Y}_{t-1}+\dots \dots +A_p{Y}_{t-p}+u_t$$

(15)

where $Y_t=Y_{1t}\dots \dots .Y_{kt}$ is a prime vector of $K$ number of variables, the $A_i$ depicts the $\left(K\times K\right)$ parameter matrices, $p$ shows the lag terms that are included, and $u_t$ is assumed to be a zero mean error process that is considered to be white noise $\left(E(u_t\right)=0$).

This is a time series study for Pakistan and the time period covers the period from 1965 to 2019. Data have been collected from the World Bank’s online database [29] and British Petroleum’s online database [55].

4. Results

Table 1. Description of Variables.

Variable	Description	Source
Life Expectancy (LXP)	Average number of years a newborn is expected to live	[30]
Infant Mortality (IMR)	No. of deaths to children under age 12 months per 1000 live births	[30]
Energy Consumption (EC)	Energy consumption (Million tons oil equivalent)	[55]
Environmental Degradation (ED)	Proxy with carbon dioxide emission (Million tons of carbon dioxide)	[55]
Financial Development (FD)	Proxy with domestic credit to private sector (% of GDP)	[30]

portrays the description of variables and also shows the source from where data on specific variable is taken. Descriptive statistics of variables are provided in

Table 2. Descriptive Statistics.

	LXP	IMR	FD	ED	EC
Mean	60.01	133.22	22.88	79.09	34.38
Maximum	67.27	212.70	29.79	198.33	83.95
Minimum	49.42	67.20	15.31	17.63	7.04
Jarque–Bera	3.17	3.49	2.37	5.02	4.81
Probability	0.20	0.18	0.31	0.08	0.09

. Life expectancy during the study period has a mean value of 60 (years) with a minimum and maximum value of 49 and 67 (years), respectively. The infant mortality rate has registered a maximum value of 212.70 and a minimum of 67.20, while its mean is 133.22. This is measured as the number of infant deaths per 1000 live births. Financial development, which is a proxy for private domestic credit, measures as a percentage of GDP, has minimum and maximum values of 15.31 and 29.79, respectively, and a mean value of 22.88. Carbon emissions are measured in millions of tons. It is carrying a mean value of 79.09, whereas its minimum and maximum values are 17.63 and 198.33, respectively. Energy consumption is carrying a mean value of 34.38 and its minimum and maximum values are 7.04 and 83.95, respectively. The Jarque–Bera statistic shows that all variables are normally distributed.

Before estimation of the study’s models, data have to be checked as to whether the data meet the assumptions of regression. One of the assumptions of regression is that data have to be normally distributed and there should be no trend in data over time, otherwise regression results will not be reliable. So, it is advisable to check time series data for trends (unit root problems) and if there is evidence of trends in data it has to be eliminated before applying regression. Thus, in this study, we applied ADF to determine whether data trend over time. If it is found that data trend over time, then we have to look for methods to de-trend time series data. The reason is that trended data does not have a constant mean and variance over time. However, if data trend at a level and if someone applies a differencing method, then data can be de-trended and time series data can be stationary. Results of the ADF unit root tests are provided in

Table 3. ADF Unit Root Test Results.

Variable	Level	First Difference	Conclusion
lnLXP	−2.72	−7.42 *	I (1)
lnIMR	0.25	−3.62 **	I (1)
lnEC	−1.02	−5.45 *	I (1)
lnED	−2.01	−4.41 *	I (1)
lnFD	−2.34	−6.05 *	I (1)

Note: * and ** shows significate at 1 % and 5 % level respectively.

. Results of the ADF unit root tests indicate that all variables of the study suffer from a unit root problem when variables are considered at level. However, all variables are free from unit root problems after taking their first difference. In other words, all variables become stationary at the first difference. Thus, it can be concluded from unit root test that the variables of the study are integrated of order one. To account for unknown structural breaks in series, the ZA unit root test results are presented in

Table 4. ZA Unit Root Test Results.

Variable	Level	Break Period	First Difference	Break Period	Conclusion
lnLXP	−0.11	1980	−5.17 *	2004	I (1)
lnIMR	−3.42	2008	−6.79 *	2001	I (1)
lnEC	−1.45	1994	−5.91 *	2012	I (1)
lnED	−3.11	1986	−4.61 **	1979	I (1)
lnFD	−3.58	2009	−6.35 *	2009	I (1)

Note: * and ** shows significate at 1 % and 5 % level respectively.

. Results of the ZA unit root test also confirm the results of an ADF test, along with the structural break periods for each series.

Results of the Johansen cointegration test for health outcomes, when proxied with life expectancy, are provided in

Table 5. Results of the Johansen Cointegration Test for Health Outcomes (Life Expectancy).

Hyp. No. of CE (s)	Trace Stat.	Critical Value	Max-Eigen Stat.	Critical Value
None *	88.83	40.18	69.56	24.16
At most 1	19.27	24.28	13.35	17.80
At most 2	5.92	12.32	5.92	11.22
At most 3	0.01	4.13	0.01	4.13

Note: * shows one cointegration vector at 5% level of significance.

. Results of statistics, trace, and maximum Eigen statistics show that one cointegration vector is present among the variables, so it can be concluded that there is a long-run relationship among energy consumption, environmental degradation, financial development, and life expectancy in Pakistan. Similarly,

Table 6. Results of the Johansen Cointegration Test for Health Outcomes (Infant Mortality).

Hyp. No. of CE (s)	Trace Stat.	Critical Value	Max-Eigen Stat.	Critical Value
None *	63.24	40.18	29.92	24.16
At most 1	23.31421	24.28	23.07	17.80
At most 2	10.24	12.32	6.01	11.22
At most 3	4.11	4.13	4.23	4.13

Note: * shows one cointegration vector at 5% level of significance.

presents results of the Johansen cointegration test for health outcomes when proxied with infant mortality. The results of the Johansen cointegration once again show that one cointegration vector is present among the variables; thus, we conclude that there is a long-run relationship among energy consumption, environmental degradation, financial development, and infant mortality. From the results of the Johansen cointegration it is deduced that energy consumption, environmental degradation, and financial development are determinants of health outcomes in Pakistan.

The results of the Johansen cointegration test are validated through an ARDL bounds test as this test can account for a structural break.

Table 7. Results of ARDL Bounds Test for Health Outcomes.

Life Expectancy Model	F-stat.	43.70 *
Infant Mortality Model	F-Stat.	39.69 *
Critical Values	Lower Bounds	Upper Bounds
1%	4.29	5.61
5%	3.23	4.35
10%	2.72	3.77

Note: * shows significance at 1% level.

depicts the results of an ARDL bounds test for health outcomes, i.e., life expectancy and infant mortality. These results validated the results of the Johansen cointegration that variables of both models are cointegrated as the estimated F-statistic value is greater than the upper bounds at a 1% level of significance. So, energy consumption, environmental degradation, and financial development are determinants of health outcomes in the long run.

It is confirmed by the Johansen cointegration test that variables of the models of the study are cointegrated, so VECM is applied for long-run and short-run causality. Results of causality are provided in

Table 8. Causality Results for Health Outcomes (Life Expectancy).

Variables	Short Run Causality				Long Run Causality
	ΔlnLXP	ΔlnEC	ΔlnED	ΔlnFD	ECT
ΔlnLXP	-	80.46 *	79.90 *	81.13 *	−8.44 *
ΔlnEC	0.48	-	4.37 *	2.61 ***	1.07
ΔlnED	0.72	4.64 *	-	1.64	−0.42
ΔlnFD	2.46 ***	3.03 ***	2.25	-	−1.72 ***

Note: *, ** and *** denote significance at 1%, 5%, and 10% level, respectively.

for the life expectancy model. Long-run causality results indicate that there is unidirectional causality from energy consumption and carbon emissions to life expectancy, while there is bidirectional causality between life expectancy and financial development. It is evident from the short-run causality results that there is bidirectional causality between life expectancy and financial development, between energy consumption and carbon emissions, and between energy consumption and financial development. This study finds that unidirectional causality runs from energy consumption to financial development.

Causality results obtained through VECM are provided in

Table 9. Causality Results for Health Outcomes (Infant Mortality).

Variables	Short-Run Causality				Long-Run Causality
	ΔlnIMR	ΔlnEC	ΔlnED	ΔlnFD	ECT
ΔlnIMR	-	25.80 *	25.44 *	25.82 *	−4.27 *
ΔlnEC	2.30	-	3.72 **	1.79	0.47
ΔlnED	0.86	4.39 **	-	2.48	1.47
ΔlnFD	4.86 **	3.14 **	3.21 **	---	−2.64 ***

Note: *, ** and *** denote significance at 1%, 5%, and 10% level, respectively.

for the infant mortality model. In the long run, unidirectional causality is found from energy consumption to infant mortality. Likewise, unidirectional causality is witnessed from carbon emissions to infant mortality. Bidirectional causality is found between infant mortality and financial development. Short-run causality results indicate that bidirectional causality exists between energy consumption and carbon emissions, whereas unidirectional causality runs from energy consumption and carbon emissions to infant mortality. Bidirectional causality is found between infant mortality and financial development in the short run.

Table 10. Results of Variance Decomposition for Health Outcomes (Life Expectancy).

Variance Decomposition of lnLXPt
Period	lnLXP	lnEC	lnED	lnFD
1	100.00	0.000	0.00	0.00
5	57.09	23.97	15.14	3.80
10	25.933	55.34	14.54	4.18
20	9.72	80.39	6.98	2.90
Variance decomposition of lnECt
	lnLXP	lnEC	lnED	lnFD
1	1.84	98.16	0.00	0.00
5	1.62	97.30	1.015	0.06
10	1.35	96.99	1.56	0.10
20	1.09	97.26	1.45	0.20
Variance decomposition of lnCEt
Period	lnLXP	lnEC	lnED	lnFD
1	0.45	79.65	19.90	0.00
5	0.99	93.36	5.47	0.17
10	1.04	94.96	3.80	0.20
20	0.88	95.59	3.29	0.25
Variance decomposition of lnFDt
Period	lnLXP	lnEC	lnED	lnFD
1	19.19	0.10	2.13	78.57
5	17.81	13.76	25.91	42.52
10	16.48	14.61	33.04	35.88
20	16.15	15.07	33.70	35.08

presents results of variance decomposition for life expectancy Model (1). The variance decomposition method provides a proportional influence in one variable because of innovative shock in other variables and, moreover, over different time periods beyond the selected time period [52]. Thus, this method has an advantage over other causality approaches as it not only shows direction but also the magnitude of causality. This research study considers a 20-year forecasting period. The results of the variance decomposition method show that the 100 percent variation in life expectancy is due to its own shocks in the first projecting period, whereas innovative shocks in energy consumption, carbon emissions, and financial development explain about 24%, more than 15%, and almost 4%, respectively, in the fifth projecting period. However, in the 20th projecting period, one standard shock in energy consumption, carbon emissions, and financial development explains more than 80%, 7%, and almost 3%, respectively, in life expectancy. These results of variance decomposition describe that most variations in life expectancy are occurring because of innovations in energy consumption in the long run. Variation in energy consumption is due to its own shocks, irrespective of the projecting period as its own shock explains more than 97% of the variation. This result is like the VECM causality results, as the study did not find long-run causality from any variable to energy consumption. In the first projecting period, one standard shock in energy consumption is responsible for a 20% variation in carbon emissions, whereas its own shock explains almost 80%. However, as the projecting time period extends, the variation caused by its own shock is above 95%. Life expectancy explains a 19% variation in financial development and the financial development variation from own shock is about 78% in the first projecting period. However, as the projecting time extends variation in financial development from its own shock decreases, whereas life expectancy, energy consumption and carbon emissions explain 16%, 15%, and almost 34%, respectively, in the 20th projecting period. This finding confirms the unidirectional causality established through VECM from life expectancy, energy consumption, and carbon emissions to financial development in the long run.

The results of the variance decomposition method, presented in

Table 11. Results of Variance Decomposition for Health Outcomes (Infant Mortality).

Variance decomposition of lnIMRt
Period	lnIMR	lnEC	lnED	lnFD
1	100.00	0.00	0.00	0.00
5	60.11	4.88	23.07	11.94
10	40.35	16.02	31.38	12.26
20	23.93	46.27	19.51	10.28
Variance decomposition of lnECt
	lnIMR	lnEC	lnED	lnFD
1	0.418384	99.58162	0.000000	0.00
5	0.126672	95.16241	3.990379	0.72
10	0.096449	87.02407	11.25110	1.63
20	0.084977	78.07225	18.88456	2.96
Variance decomposition of lnCEt
Period	lnMR	lnEC	lnED	lnFD
1	3.08	75.49	21.43	0.00
5	1.44	91.01	7.16	0.39
10	0.71	90.70	7.49	1.09
20	0.44	81.19	16.09	2.27
Variance decomposition of lnFDt
Period	lnIMR	lnEC	lnED	lnFD
1	0.52	0.003	0.00	99.47
5	1.08	5.85	0.611715	92.46
10	2.09	14.71	5.812591	77.38
20	3.73	15.88	18.24979	62.13

, for the infant mortality model show that the 100% variation in infant mortality is due to its own shocks in the first projecting period, whereas innovative shocks in energy consumption, carbon emissions, and financial development explain about 5%, more than 23%, and almost 12%, respectively, in the fifth projecting period. This scenario changes when the projecting time is extended, and one standard shock in energy consumption, carbon emissions, and financial development explain more than 46%, 20%, and 10%, respectively, in infant mortality in the 20th projecting period. These results of variance decomposition describe that most variations in infant mortality are occurring because of innovations in energy consumption in the long run. This was also true for life expectancy as most variation was occurring because of energy consumption. The variation in energy consumption that is due to its own shocks is above 87% up to the 10th projecting period. However, in the long run, carbon emissions explain the variation in energy consumption of about 19%, although its own shock explains the above-78% variation in the 20th projecting period. This result is like the VECM causality results as the study did not find long-run causality from any variable to energy consumption. In the first projecting period, one standard shock in energy consumption is responsible for 20% variation in carbon emissions, whereas its own shock explains almost 80%. However, as the projecting time period extends, the variation caused by its own shock is above 95%. Life expectancy explains the 19% variation in financial development and financial development variation from own shock is about 78% in the first projecting period. However, as the time extends variation in financial development from its own shock decreases, whereas infant mortality, energy consumption and carbon emissions explain almost 4 percent, 16 percent and above 18 percent respectively in 20th projecting period.

Long-run and short-run estimates of health outcome models based on the Johansen cointegration are presented in

Table 12. Estimates for Health Outcome Models.

	Life Expectancy Model	Infant Mortality Model
Long-run estimates
Regressors	Coefficient	Coefficient
lnEC	2.05 *	−1.18 *
lnED	−1.77 *	0.75 **
lnFD	0.65 *	−0.52 *
Short-run estimates
ΔlnEC	0.05 **	−0.02 **
ΔlnED	−0.06 *	0.06
ΔlnFD	0.01	0.04
ECT (−1)	−0.12 **	−0.04 **
Diagonastic tests
Test	Test Stat.	Test Stat.
Heteroscedasticity	1.73	1.80
Autocorrelation	0.46	0.89
Ramsey RESET	0.03	0.23
JB Normality	2.96	4.69

Note: * and ** denote significance at 1% and 5% level respectively.

. Columns 2 and 3 in Table 12 portray results for the life expectancy and infant mortality models, respectively. Results indicate that energy consumption has positive and significant effects on life expectancy in both time spans, i.e., long-run and short-run. Environmental degradation has negative and significant effects on life expectancy in the long run, as well as in the short run. Moreover, results indicate that financial development has positive and significant effects on life expectancy, but has an insignificant positive effect on life expectancy in the short run. Similarly, results indicate that energy consumption and financial development are negatively associated with infant mortality, whereas environmental degradation has a positive and significant effect on infant mortality in the long run. The error correction term (ECT) for the life expectancy model carries a negative and significant coefficient with a magnitude of 0.12. This result indicates that the model for life expectancy is dynamically stable and it will come to equilibrium by 12% if disequilibria happens to any external shock. Likewise, the ECT for the infant mortality model is also significant and carries a negative sign and its magnitude is 0.04. This magnitude of ECT indicates that the model of infant mortality is also stable; however, it will adjust to any external shock by only 4% per annum. The results of the study are reliable as different diagnostic tests are carried out for both models of the study. The JB and Ramsey RESET test statistics confirm that the error term is normally distributed and the model is stable as long as the functional form of the model is correct. The model is also free from autocorrelation and heteroscedasticity problems. Figure 3 and Figure 4 show results of CUSUM and CUSUM of squares for the life expectancy model. Likewise, Figure 5 and Figure 6 show results of CUSUM and CUSUM of squares for the infant mortality model. The stability of both models of the study is also confirmed by the recursive stability check as CUSUM and CUSUM of squares lines are within the limits. Besides, for the robustness of the long-run estimates, this study applied dynamic ordinary least squares (DOLS) [56]. This cointegration regression may be applied when variables are integrated in the first order and cointegration is confirmed as in the case of this study. The results of DOLS are consistent and efficient, as these results are free from autocorrelation and simultaneous bias. The reason is that this cointegration regression considers leads and lags of the explanatory variables and, moreover, this regression eliminates the feedback effect in the cointegration system.

Table 13. Long-Run Estimates Based on DOLS for Health Outcome Models.

Regressors	Life Expectancy Model	Infant Mortality Model
	Coefficient	Coefficient
lnEC	0.32 *	−0.23 *
lnED	−0.23 *	0.15 *
lnFD	0.05 *	0.38 *

Note: * denote significance at 1% level.

displays the long-run estimates of DOLS and the long-run results are the same as Table 12. All explanatory variables are significantly associated with health outcomes and these results show and confirm that energy consumption and financial development improves health outcomes, whereas environmental degradation worsens health outcomes. This study was done for a developing country, so these findings may be helpful for other developing countries in that energy consumption, environmental degradation, and financial development are significant factors of health outcomes.

5. Discussion

The long-run cointegration results indicate that all study variables are in the long-run relationship; thus, it can be concluded that energy consumption, environmental degradation, and financial development are determinants of health outcomes in Pakistan. The causality results indicate that unidirectional causality runs from energy consumption and environmental degradation to health outcomes. Henceforth, this result shows that the first two hypotheses of the study are achieved. Similarly, long-run causality results indicate that financial development and health outcomes Granger Causality each other; thus, the third hypothesis of the study is also achieved. The finding of this study that energy is a Granger Causality health outcome in the long run in the case of Pakistan resembles the findings of [5,31] and denies the claim of [32] that there is no causal relationship between energy and health outcomes. This study finds unidirectional causality from environmental degradation to health outcomes in Pakistan, irrespective of health outcome proxies, i.e., life expectancy and infant mortality. This finding resembles the findings of studies [5,18], whereas bidirectional causality is concluded between environmental degradation and health outcomes [31,33,34,35]. The results of bidirectional causality between financial development and health outcomes in the case of Pakistan support the conclusion of [23] that a strong causal relation is present between these financial developments and health outcomes. Studies [38,39] also found bidirectional causality between financial development and health outcomes. The causality results of the study between energy and financial development indicate a one-way causation from energy to financial development. This finding resembles the findings of [57], as this study concludes that energy consumption initiates the Granger Causality on financial development in the long run and not vice versa. Likewise, the study [58] also confirmed unidirectional causality from energy consumption to financial development. However, the finding of this study is opposite to what Islam [59] claimed, which is that energy consumption causes financial development and not vice versa.

The estimates of the health outcome models of this study indicate that energy and financial development have positive and significant effects on health outcomes in the case of Pakistan. Energy has a positive impact on health outcomes, as the results of the study show that energy is positively related with life expectancy and negatively related with infant mortality. These findings are in line with [25,26,32,44]. The positive association of energy with health outcomes postulates that the government of Pakistan has to take steps to tackle the energy crisis in the country, especially through renewable energy projects, as the share of non-renewable is very high. If one does not consider electricity from hydro in Pakistan, the share of renewable is below 1% and it has been witnessed that in last couple of decades the share of hydro in electricity is continuously falling, while the share from thermal is on the rise. In such a scenario, it is not possible to fulfill energy requirements with sustainable development in Pakistan. The government must take measures and initiate projects for renewable energy. As far as financial development is concerned, only 16% of adults have accounts in banks, whereas 54% of the adult population is not receiving any kind of formal or informal financial services [60]. Pakistan needs to devise a financial inclusion strategy along with green finance schemes. Likewise, a financial strategy for health care and services is also essential for Pakistan to encourage health services. The results show that environmental degradation is hurting life expectancy and leads to infant mortality as environmental degradation has negative and positive effects on life expectancy and infant mortality, respectively. Hence, environmental degradation is damaging health outcomes in Pakistan. This finding implies that Pakistan can improve health outcomes by restraining environmental degradation. This restraining of environmental degradation can be done through various policy measures, such as steps to encourage plantation, reserve forests, and lessen deforestation. The government of Pakistan already initiated the billion tree tsunami program and the good point is that the government is also taking steps not just to increase plantation, but is also encouraging and supporting plantation according to local areas so forestation and poverty alleviation go side by side. Another important policy implication is that the government of Pakistan should make sure to cover the energy requirements of those areas where forest is under great stress because of tree-cutting, done especially for domestic needs.

6. Conclusions

Pakistan has been facing an energy crisis since the early 2000s and has adopted different energy policies to tackle the energy crisis and at the same time is among the top 10 climate-change affected countries. Additionally, health outcome indicators are not satisfactory in Pakistan as life expectancy is lower and infant mortality is higher than in Bangladesh, India, and Sri Lanka. Besides, the financial sector is also not up to the mark as 54% of Pakistanis are not receiving any kind of financial services. These facts about Pakistan led this study to test the causal nexus among energy consumption, environmental degradation, financial development, and health outcomes in the case of Pakistan. This study proxies health outcomes with life expectancy and infant mortality. Time series data have been analyzed over the period from 1965 to 2019 through different econometric techniques. Moreover, this study not just discovers the causal directions among variables, but also determines the strength of causality through variance decomposition. Results of the study confirm that all variables of the study are cointegrated in the long run. The causality analysis reveals that unidirectional causality is running from energy consumption and environmental degradation to health outcomes, whereas bidirectional causality is found between financial development and health outcomes in the long run. Besides, this study also determines the effect of energy, environmental degradation, and financial development in the health outcome model and finds that energy and financial development can help Pakistan to improve health outcomes, whereas environmental degradation is hurting health outcomes.

The policy implications of the study are that the government of Pakistan has to take steps to tackle the energy crisis in the country, especially through renewable energy projects. These policy measures will not only ensure energy security, but also help Pakistan to address climate-change adversities. Likewise, the restraining of environmental degradation can be done through various policy measures, such as steps to encourage plantation, preserve forests, and lessen deforestation. The government of Pakistan already initiated the billion tree tsunami program and the good point is that the government also taking steps not just to increase planation but also to encourage and support fruitful trees so forestation and poverty alleviation go side by side. Another important policy implication is that the government of Pakistan should make sure to cover the energy requirements of those areas where forest is under great stress because of tree cutting done especially for domestic needs. Pakistan needs to devise a financial inclusion strategy along with green finance schemes. Likewise, a financial strategy for health care and services is also essential for Pakistan to encourage health services. As Pakistan is a developing country, the implications of this study can be generalized for other developing countries. Developing countries have to ensure energy security as this will not only help them in economic development but can also improve their health outcomes. However, developing countries have to ensure their energy security with renewable energy resources as most developing countries rely on fossil fuel and environmental degradation can hurt health outcomes. Likewise, developing countries may strengthen their financial sectors, especially through digitalization, as it is a smart way to enhance financial inclusion and this will also help developing countries to improve on health indicators. In future studies, especially in the context of developing countries, researchers may consider other indicators for environmental degradation that have more local than global impacts, as most developing countries are not much affected by global warming and their share of global carbon emissions is also negligible. Besides, researchers may also include the role of the public in health outcomes in future studies for developing countries.