Epidemics and electricity CO₂ emissions: a feedback investigation

Abstract

We examine the short-term and long-term causal effects between epidemics and electricity CO₂ emissions by using panel data from 30 countries over the period of 1990 to 2017. The results show that there is bidirectional relationship between epidemics and electricity CO₂ emissions, especially in OECD and Asian countries.

Introduction

The emergence and spread of coronavirus disease (COVID-19) in 2019 make an epidemic once again become the focus of global attention (Anderson et al. 2020; Velavan and Meyer 2020). There are various causes for the outbreak of epidemics. Despite the pathological explanation, some scholars have paid attention to the relationship between environmental quality and epidemic diseases 30 years ago. Leggett (1990) thinks that the change in air quality is bringing about a secondary effect that cannot be ignored—that is, it provides more favorable living conditions for epidemics. Since then, the discussion on the relationship between epidemics and environment has continued to expand. Saadi (2010) believes that long-term exposure to air pollution will significantly increase the risk of respiratory epidemics, while Thiering and Heinrich (2015) indicate a link between air pollution and type 2 diabetes (T2D). At the same time, epidemics also bring about environmental changes. After the COVID-19 outbreak, the topic of epidemics and environment has seen a trend of further expansion. Craven et al. (2020) find that NO_x pollution in Beijing decreased by 59% during the period of COVID-19’s impact. Chen et al. (2020) also believe that COVID-19 actually improved air quality due to the reduction of travel traffic. Therefore, it seems to us that it is necessary to pay attention to the relationship between epidemics and environmental quality. On the one hand, it can deepen our understanding of epidemics and bring inspiration to prevent and control their spread. On the other hand, environmental governance is an eternal topic. It is necessary to remind people to treat environmental problems more carefully after COVID-19.

We argue in this paper that air pollution accelerates the expansion of the epidemic by creating an appropriate environment for transmission, and the spread of the epidemic also has an impact on the environment by changing people’s way of life and production (Kumar et al. 2020). Most of the recent literature has focused only on cross-sectional analysis and data from a few countries, such as China and the USA, to investigate the impact of air quality on specific epidemics or the short-term impact of COVID-19 on air quality (Hoang and Jones 2020; Chen et al. 2020).

Our research contributes to the literature in two important ways. We believe that the causal analysis of epidemics and environment should be long-term and extensive, which is more conducive to finding their internal relationship, and it is necessary to consider the role of dynamic factors, because if air pollution can lead to epidemics, the history of air pollution is also crucial to explain the formation and spread of current epidemics. Therefore, we have adopted panel data based on 30 countries for 29 years, using panel VECM estimation to confirm the bidirectional equilibrium between epidemics and air quality in the long run, Banerjee and Carrion‐i‐Silvestre (2015) cointegration test and PANICCA results also offer strong evidence that they have a cointegration relationship when endogeneity is considered. Furthermore, three subsamples and CO₂ emissions from different sectors provide deeper evidence that the long-term relationship between epidemics and air quality varies for different regions and economic environments.

In OECD and Asian countries, this bidirectional relationship is stronger, while in Africa, due to the imperfect development of its own electricity and industrial sectors, epidemics do not seem to have a long-term bidirectional relationship with their CO₂ emissions, but the transport sector, as the main source of emissions in Africa, is still affected by epidemics.

Data and methodology

Data description

Using panel data from 30 countries from 1990 to 2018, this paper examines the relationship between epidemics and air quality. In terms of epidemics, we collect three variables, number, death, and effect of epidemics in the 30 sample countries over the years to describe the impact of epidemics. According to the interpretation of EM-DAT, number refers to the number of occurrences of new epidemic cases or abnormal increases in original epidemic cases. Death refers to the number of new deaths when number appears. Effect refers to the estimated loss, expressed in US dollars (‘000), and the figures are shown true to the year of the event. In terms of air quality, we use CO₂ (metric tons per capita emissions). In order to distinguish the sources of CO₂ emissions (Bereitschaft and Debbage 2012; Rahman et al. 2017), we also collect data from the electricity, industry, and transportation sectors. Table 1 shows the definitions of variables and data sources.

Table 1 Variable definitions and data sources

Table 2 shows the descriptive statistics of all variables. We use the logarithmic form for both death and effect. The average value of CO₂ in 1995 was 4.56 increasing to 4.89 in 2005 and 5.34 in 2015. In addition, the CO₂ emissions of all sectors also showed an upward trend, indicating that the air quality is deteriorating. The average number of epidemics also increased from 0.45 in 1995 to 1.41 in 2005, and only declined in 2015. From the standard deviation point of view, CO₂ in 1995 was 5.13 keeping 5.19 in 2005 and 5.29 in 2015, showing a relatively stable performance, and the results of CO₂ emissions in all sectors were also consistent. However, the standard deviation of the number fluctuates greatly, indicating that the outbreak of epidemics may be concentrated in some years.

Table 2 Descriptive statistics

PANICCA panel unit root test

Whether a series is in a stationary or nonstationary state, the PANICCA test (Reese and Westerlund, 2016) can find whether the state is driven by common factors or idiosyncratic components. Such discrimination is of practical significance. For example, when the nonstationarity of the series comes from common factors, each factor will have a stronger contagion effect in all samples, which will make the effect of relevant policy opinions more significant. If the situation is the opposite, then when the nonstationarity of the series comes from an idiosyncratic component, we do not recommend adopting the same strategy for all samples.

Supposing the variable we are going to observe is Y, the data generation process (DGP) is represented by the following common factor model:

$$ {Y}_{i,t}=\beta {\prime}_i{D}_{t,p}+\xi {\prime}_i{G}_t+{\sigma}_{i,t}, $$

(1)

where D_{t, p} presents a vector polynomial trend defined by D_{t, p} = (1, …, t^P)' and (p + 1) × 1; ξ_i represents the corresponding vector of factor loadings; G_t is a r × 1 vector that stands for the common factors; and σ_{i, t} represents a unit-specific error.

Banerjee and Carrion-I-Silvestre cointegration test (BC)

Compared with the Westerlund and Edgerton (2008) test, the BC cointegration test (2015) puts forward more strict conditions for the model, which only allows a horizontal term in the deterministic part of the stochastic process. At the same time, the BC cointegration test also considers possible breaks in the trends of data generation process, which gives it advantages in dealing with structural fracture and cross-section correlation problems. The data generation process is:

$$ {y}_{it}={D}_{it}+x{\prime}_{it}{\delta}_{it}+{u}_{it} $$

(2)

$$ {u}_{it}={F}_t^{\prime }{\pi}_i+{e}_{it} $$

(3)

$$ \left(I-L\right){F}_t=C(L){w}_t $$

(4)

$$ \left(1-{\rho}_iL\right){e}_{it}={H}_i(L){\varepsilon}_{it} $$

(5)

$$ {x}_{it}={\kappa}_i+{x}_{it-1}+{G}_t^{\prime }{\varsigma}_i+\varXi (L){v}_{it} $$

(6)

$$ {G}_t=\varGamma (L){\varpi}_i, $$

(7)

where t = 1,…T stands for the time term; i = 1,…, N represents an individual; and y_it and x_it are the scalar and vector of a stochastics process, respectively.

Panel VECM

According to the BC cointegration test (2015), we realize that there is a co-integration relationship between epidemics and CO₂ emissions. In order to further determine the interaction between them, we use panel-VECM to conduct the Granger causality test. VECM first obtains the estimated residuals ε_it and v_it through (8) and (9):

$$ \mathrm{CO}{2}_{it}={\alpha}_i+{\delta}_it+{\beta}_i{\mathrm{number}}_{\mathrm{it}}+{\varepsilon}_{it} $$

(8)

$$ \kern0.5em {\mathrm{Epidemics}}_{it}={\varphi}_i+{\gamma}_it+{\chi}_i\mathrm{CO}{2}_{it}+{\nu}_{it} $$

(9)

After the residuals ε_it and v_it are obtained, we estimate the Granger causality model by a dynamic error correction:

$$ \varDelta \mathrm{CO}{2}_{it}={\theta}_{1i}+{\lambda}_1{\varepsilon}_{it-1}+{\sum}_k{\theta}_{11k}\varDelta \mathrm{CO}{2}_{it-k}+{\sum}_k{\theta}_{12k}\varDelta {\mathrm{epidemics}}_{it-k}+{u}_{1 it} $$

(10)

$$ \varDelta {\mathrm{Epidemics}}_{it}={\theta}_{2i}+{\lambda}_2{\nu}_{it-1}+{\sum}_k{\theta}_{21k}\varDelta CO{2}_{it-k}+{\sum}_k{\theta}_{22k}\varDelta {\mathrm{epidemics}}_{it-k}+{u}_{2 it} $$

(11)

In Eqs. (10) and (11), θ_12k and θ_21k for all possible k are respectively the standard for judging whether there is a short-term associational relationship between CO₂ emissions and epidemics, under which the null hypotheses are H₀:θ_12k = 0 and H₀:θ_21k = 0. Here, λ is the standard to judge whether there is a long-term relationship between them, under which the null hypotheses are H₀:λ₁ = 0 and H₀:λ₂ = 0.

Results

PANICCA (Reese and Westerlund 2016a, b) unit root test consists of two models. When only constant terms are allowed in the model, P = 0. When both constant terms and trend terms are included in the model, P = 1. Because epidemics and CO₂ emissions are less affected by trends, we use p = 0 model for analysis. We divide the samples into full samples, OECD countries subsamples, Asian countries subsamples and African countries subsamples.¹ Table 3 shows the results of common factors and idiosyncratic factors of all kinds of samples when p = 0. When we pay attention to common factors, the ADF statistics of all variables are significant at the level of 1% except for the CO₂ emissions of Asian subsample countries. This result rejects the null hypothesis of unit root, indicating that there is a common factor of all sample countries that leads to the stability of variables. When we pay attention to idiosyncratic factors, we can find that CO₂ emissions are not significant at the level of 10% in both full samples and the subsamples, indicating that there are idiosyncratic factors in individual countries leading to the nonstationary of CO₂ emissions. We assume that the nonstationary is due to the impact of epidemics.

Table 3 PANICCA test

Table 4 shows the results of model “two breaks” of BC cointegration test. According to the different CO₂ emission sectors, we divide them into four categories: per capita CO₂ emission, CO₂ emission of electricity sector, industrial sector, and transportation sector to observe whether CO₂ emission of different sectors will affect the relationship with epidemics.² It can be seen that when the dependent variable is CO₂ at the statistic of − 12.08, the probability of rejecting the null hypothesis is 53.33% at the 5% significance level; while when the dependent variable is number at the statistic of − 9.51, the probability of rejecting the null hypothesis is 56.67% at the 5% significance level, indicating that there is a cointegration relationship between CO₂ and number, which is also supported by the results of death and effect. Similarly, we can therefore conclude that all sectors of CO₂ emissions have a cointegration relationship with the number, death, and effect of epidemics.

Table 4 Panel cointegration of BC (2015)

According to the results of the two breaks model, we further describe the potential breakpoints of each sample country in the past 29 years. We can find that the first category of breakpoints occurred in 1994–1998, and the second category of breakpoints occurred in 2008–2012. During the first turning point, the India plague broke out in 1994, and the first Norovirus patient appeared in 1995. In 1997, the highly lethal influenza A/H5N1 influenza virus was epidemic, while during the second turning point (Hall et al. 2013; Barberis et al. 2016), the H1N1 virus (2009) originated in Mexico broke out, causing up to 1.4 billion worldwide infections and 151,700–575,400 deaths (Girard et al. 2010). These two turning points are all in the period of global epidemic outbreak, which may be caused by the epidemic outbreak reducing people’s travel and production activities, thus changing CO₂ emissions.

After confirming the cointegration relationship between CO₂ emission and epidemics, Table 5 provides the results of panel VECM causality tests. We want to know not only whether there is short or long term relationship between CO₂ and number, but also be interest in whether this bi-directional relationship extends to all sectors of CO₂ emissions, and whether it is applicable to countries with different regions and economic levels. Therefore, panel A to panel L list all results of possible test in detail, so as to find a deeper relationship between CO₂ emissions and epidemics.

Table 5 Robustness estimation for panel causality

Panel A shows the test results of OECD countries, when the dependent variable is CO₂, the statistical value of number is 0.51, which is not significant at the level of 10%, indicating that there is no causal relationship between number and CO₂ in short term. But in the long run, the statistical value of number is 3.20, which is significant at the level of 1%. In addition, the statistical value of the joint test of CO₂ and number is 3.78, which is significant at the level of 5%, indicating that the outbreak of epidemics has a long-term impact on CO2 emissions. The more the number of outbreaks of epidemic, the lower the air quality will be. This result is inconsistent with the views of Craven et al. (2020) and Chen et al. (2020), but we have reason to believe that the improvement of air quality is only the short-term fluctuation brought by the epidemics, because many industries have been strongly affected under the epidemic, especially catering, tourism, transportation, and service industries, millions of people are suddenly out of work and wondering how to support their families (Kumar et al. 2020). In this case, government will continue to pay attention to the epidemic prevention and control and economic reconstruction work, and relax the supervision of environmental protection, so the future air quality decline is an inevitable trend.

Furthermore, when number is the dependent variable, the statistical value of CO₂ is 9.90, which is significant at the level of 1%, indicating that the change of CO₂ affects number in the short term. This result is consistent with Saadi (2010), indicating that changes in air quality may cause epidemics in the short term. From the results of panel D, panel G, and panel J, it seems that this short-term impact is caused by CO₂ emissions from the electricity sector, because the impact of the industrial and transport sectors on number is not significant. That is to say, in OECD countries, more attention should be paid to the control of CO₂ emissions from the electricity sector so as to prevent future epidemics. In the long run, the statistical value of CO₂ is 19.96, and the statistical value of the joint test of number and CO₂ is 134.67, both of them are significant at the level of 1%. It shows that the long-term pollution emission will also impact the epidemics.

The results of Asian countries are similar to those of OECD countries, except that there is no short-term bidirectional relationship between air quality and epidemics, which shows that in OECD countries with more developed economies, people’s health is more dependent on the environment, and changes in short-term air quality may cause epidemics.

The results of African countries are shown in panel C, panel F, panel I, and panel L respectively. We can find that when the dependent variable is CO₂, except for the sector of transport, the impact of epidemics on other sectors is not significant. Part of the reason for this result may be that the sample size from African countries is too small, but due to the long-term impact of CO₂ on number, we can assume that it is the different levels of economic development that lead to the differences in results. Africa’s transport sector has been significantly affected by epidemics, which means that epidemics have changed at least the way people travel, but because of the underdevelopment of the rest of sectors, the impact is insignificant.

Conclusion

We used panel data from 30 countries over the period of 1990 to 2018 to study the short-term and long-term causal effects between epidemics and air quality. In summary, although the results of OECD, Asian and African countries are slightly different, it can be concluded that there is a long-term bidirectional relationship between epidemics and air quality. No matter in any sector of any region, CO₂ has a long-term impact on number, indicating that continuous improvement of the environment is an important strategy to prevent and reduce the outbreak of epidemics, and the possible improvement of the environment caused by epidemics is not a long-term phenomenon. The more developed economies are more vulnerable in some ways. Although the electricity and industrial sectors in Africa seem to be unaffected, with the development of the economy, it is always difficult to avoid the environmental damage caused by epidemics.