The interrelations between human capital investment and emission trading scheme: A case study in China

Method overview

Figure 2 displays this paper's research framework. HCI is constituted by educational, health, and R&D investments. Educational and health investments crowd out household, government, and enterprise expenditures on commodities, whereas R&D investment decreases government and enterprise expenditures. Since these investments affect the expenditures of the economic entities, they affect economic growth. Educational, health, and R&D investments have direct impacts on emissions, and the aggregation of these impacts is regarded as the direct impact of HCI on emissions in this paper. Human and physical capital investments determine the formation of capital, which, combined with labor and energy, is an input factor of economic growth. Together with emission factor, energy consumption induces carbon emissions.

OPEN IN VIEWER

Figure 2.

The research framework in this paper.

Soaring carbon emissions accelerate global warming which causes catastrophic outcomes; therefore, ETS is implemented to abate emissions. As ETS interrupts market mechanism, it has implementation cost and thus negatively affects GDP growth. In this paper, the ETS is designed based on the Chinese national ETS (CNETS); more details about the designed ETS are presented in Supplementary Materials. Built in 2021, the CNETS is believed to help China achieve the Nationally Determined Contribution target of emission peaking in 2030. ³³ Hence, 2021–2030 is the research period of this paper. ³⁴

The research framework depicted in Figure 2 is established based on a CGE model. The basics of the CGE model are from our previous research^29,35 for instance, the social accounting matrix was displayed in Chen. ²⁹ The CGE model is adopted to quantify the economic interrelations among the divided sectors, and the sector division is presented in Table S1 in Supplementary Materials. According to Table S1, the Chinese economy incorporates nine electricity sectors: one transmission and eight generation sectors. ³⁶ The eight electricity generation sectors are further decomposed into the four sectors generating electricity from nonrenewable energy and four sectors exploiting renewable energy. ³⁷ Transportation needs energy input and thus may have enormous carbon emissions; therefore, the transport sector is separated from the service sector even though it provides transport services. ³³ The CGE model has quantified the relations among the four economic entities (the representative household, government, enterprise, and foreigner) and the relations between the entities and the environment.

To signify the operation of the economic system, the CGE model incorporates the modules displayed in Supplementary Materials. Different from other previous CGE models, this paper's CGE model includes an additional human capital module. In the human capital module, human capital investment ( H C I t ) is defined as the summation of investments in education, ³⁸ health, ³⁹ and R&D, ⁴⁰ as shown in Equation (1). E D I t , H L I t , and R D I t refer to educational, health, and R&D investments. HCI affects GDP growth and carbon emissions, but it is a spending of the representative household, government, and enterprise. Specifically, education and health investments belong to the expenditures of all these three domestic entities, but only the government and enterprise spend their income on R&D investment.

H C I t = E D I t + H L I t + R D I t

(1)

This paper's CGE model has some unique features that need to be listed: Firstly, as technological progress is directly linked to R&D investment, its impact on carbon emissions is endogenously determined in this paper; in contrast, Jia and Lin ⁴¹ used autonomous energy efficiency improvement to measure the impact of technological progress on emissions exogenously. Secondly, with climate damage not quantified, the CGE model does not consider the negative externalities of carbon emissions. A similar setting could be found in the CGE model established by Jia and Lin. ⁴¹ Lastly, like Jia et al., ⁴² the Neoclassical macro closure is adopted for this paper's CGE model. In the business-as-usual (BAU) scenario, full labor employment and full capital usage are assumed; in other words, there is no idle labor or capital in the economy. From the perspective of Neoclassical economics, the market mechanism is capable of allocating resources efficiently. As ETS interrupts the market mechanism, it causes inefficient resource allocation and thus deadweight loss.

Economic impact of human capital investment

Human capital affects economic growth because its investment reshapes the economic system. In this paper, the economic impacts of HCI are decomposed into the impacts of education, health, and R&D investments.

Educational investment contributes to human capital formation because it raises environmental perception and awareness. ⁴³ The Statistical Bulletin on Educational Spending by the Chinese government shows that the total educational expenditures were 5.79 and 6.13 trillion CNY in 2021 and 2022, respectively. To our best knowledge, there are no official projections for educational investment; in this paper, educational investment is calculated based on the China National Academy of Education Sciences ⁴⁴ which expected that the educational expenditure would be 8.5 trillion CNY in 2030. The projected educational investment is assumed to change linearly in 2022–2030 as shown in the second column of Table S2 in Supplementary Materials.

Educational investment is collectively paid by the representative household, government, and enterprise. China Educational Finance Statistical Yearbook 2021 shows that the percentages of household, government, and enterprise spendings on education were 18.44%, 80.91%, and 0.65% in 2020, respectively. Assuming these percentages remain to be fixed over the research period, educational investment ( E D I t ) is calculated in Equation (2). E D I H t , E D I G t , and E D I E t refer to the educational investments by the household, government, and enterprise.

E D I t = E D I H t + E D I G t + E D I E t = ( 18.44 % + 80.91 % + 0.65 % ) × E D I t

(2)

According to China Statistical Yearbook 2021, the health expenditure in China was 7.68 trillion CNY in 2021; it was expected to be 15.3 trillion CNY in 2030. ⁴⁵ If health investment increases linearly in 2021–2030, the projected health investment is shown in the third column of Table S2. China Statistical Yearbook 2021 shows that the percentages of private, governmental, and social health spendings were 27.65%, 30.4%, and 41.94% in 2020, respectively. In this paper, private and social health spendings are assumed to be covered by the household and enterprise. If the percentage spendings are fixed over the research period, health investment ( H L I t ) is defined in Equation (3). H L I H t , H L I G t , and H L I E t stand for the health investments by the household, government, and enterprise. As the aggregation of the three percentages is 99.99%, there is an amplifier 100 % 99.99 % in Equation (3).

H L I t = H L I H t + H L I G t + H L I E t = ( 27.65 % + 30.4 % + 41.94 % ) × 100 % 99.99 % × H L I t

(3)

The Statistical Bulletin on R&D Investment by the Chinese government shows that R&D expenditures were 2.80 and 3.08 trillion CNY in 2021 and 2022, respectively. The R&D investment intensity, defined as R&D investment divided by GDP, was projected to be 2.5–2.6% in 2030. ⁴⁶ In this paper, we have chosen a medium value of 2.55% for the projected intensity in 2030. The Chinese GDP in 2030 was approximately 180.78 trillion CNY from the long-term GDP forecast by OECD. Hence, the projected R&D investment was calculated as the intensity multiplied by the GDP, and the number would be 4.61 trillion CNY in 2030. Assumed to change linearly in 2022–2030, the projected R&D investment is displayed in the fourth column of Table S2. China Statistical Yearbook 2021 shows that the percentages of governmental and enterprise spendings on R&D were 20.34% and 79.66% in 2020, respectively. If the percentage spendings are fixed over the research period, R&D investment ( R D I t ) is defined in Equation (4). R D I G t and R D I E t stand for the R&D investments by the government and enterprise.

R D I t = R D I G t + R D I E t = ( 20.34 % + 79.66 % ) × R D I t

(4)

Equation (5) defines the HCIs by the representative household ( H C I H t ), government ( H C I G t ), and enterprise ( H C I E t ). Household HCI is equal to the summation of household educational and health investments; for the government and enterprise, HCI is the aggregation of educational, health, and R&D investments. The proportion of HCI to income is defined in Equation (6). Y H T t , Y G T t , and Y E T t are household, government, and enterprise income. p e r h c h t , p e r h c g t , and p e r h c e t are the percentages of household, government, and enterprise income spent on HCI; these percentages are assumed to be unaffected by ETS. Nevertheless, ETS influences the incomes of the representative household, government, and enterprise, thereby affecting their investments in human capital.

{ H C I H t = E D I H t + H L I H t H C I G t = E D I G t + H L I G t + R D I G t H C I E t = E D I E t + H L I E t + R D I E t

(5)

{ H C I H t = p e r h c h t × Y H T t H C I G t = p e r h c g t × Y G T t H C I E t = p e r h c e t × Y E T t

(6)

Total capital investment ( T C I t ) is defined as the aggregation of human capital investment ( H C I t ) and physical capital investment ( P C I t ) in Equation (7). HCI contributes to capital investment; therefore, it promotes GDP growth since capital is an input factor of economic output. The embedded assumption of Equation (7) is that human capital is deemed to play the same role as physical capital in economic growth. As human capital affects economic growth, it indirectly impacts emissions, considering that GDP growth is accompanied by anthropogenic emissions from nonrenewable energy consumption.

T C I t = H C I t + P C I t

(7)

Emission impact of human capital investment

The direct impact of human capital on emissions is attributed to the induced innovation of green technology. ¹¹ As this paper has focused on the nexus between HCI and ETS, we do not conspicuously model the role of technological progress in emission abatement. Instead, induced technological change of human capital is embodied in the direct emission impact of HCI, and this impact was extensively studied in previous research. For example, Umar et al. ⁴⁷ analyzed the link between human capital efficiency and carbon emissions, finding a negative relationship between HCI and emissions in eight countries spanning over 10 years. Similarly, Khan et al. ¹ found that human capital played a key role in reducing carbon dioxide emissions in the Belt and Road countries. Nevertheless, how human capital impacts emissions in China exclusively remains to be explored. In this paper, we study the historical relations between HCI (educational, health, and R&D investments) and carbon emissions in China first. Subsequently, we project the direct HCI impact on emissions in the future, assuming that the historical relations continue to exist over the research period.

Owing to its capability of providing reliable results when researched variables are in mixed orders of stationarity, ⁴⁸ autoregressive distributed lag (ARDL) models are employed to study how HCI directly affects carbon emissions in this paper. The prerequisites of conducting ARDL analyses are stationary time series, confirmed by unit root tests, and cointegration relations, confirmed by panel cointegration tests.

According to Bano et al., ¹⁰ carbon emissions can be explained by energy consumption, GDP, and human capital through ARDL models. In this paper, the ARDL models are defined in Equation (8). Owing to the data unavailability, we have only considered the ARDL relations in 1995–2020; all the variables have been transformed into their logarithm forms to measure the percentage impacts of HCI on emissions. C E t denotes carbon emissions, and its value is from World Development Indicators by World Bank. E C t and G D P t stand for energy consumption and economic output; their values are from China Statistical Yearbook. u 1 t , u 2 t , and u 3 t are the residual terms of Equation (8). α 0 , α 1 j − α 4 j , β 0 , β 1 j − β 4 j , γ 0 , and γ 1 j − γ 4 j are the regression coefficients; p 1 − p 4 , q 1 − q 4 , and r 1 − r 4 are the optimal lags of the independent variables to be selected by the information criteria, like Akaike Information Criterion, Schwarz Information Criterion, and Hannan-Quinn Criterion. ⁴⁹

{ C E t = α 0 + ∑ j = 1 p 1 α 1 j C E t − j + ∑ j = 0 p 2 α 2 j E C t − j + ∑ j = 0 p 3 α 3 j G D P t − j + ∑ j = 0 p 4 α 4 j E D I t − j + u 1 t C E t = β 0 + ∑ j = 1 q 1 β 1 j C E t − j + ∑ j = 0 q 2 β 2 j E C t − j + ∑ j = 0 q 3 β 3 j G D P t − j + ∑ j = 0 q 4 β 4 j H L I t − j + u 2 t C E t = γ 0 + ∑ j = 1 r 1 γ 1 j C E t − j + ∑ j = 0 r 2 γ 2 j E C t − j + ∑ j = 0 r 3 γ 3 j G D P t − j + ∑ j = 0 r 4 γ 4 j R D I t − j + u 3 t

(8)

After establishing the ARDL models, we need to confirm the reliability of the model results by performing robust tests to check whether the embedded statistical assumptions are violated. Only when all the statistical assumptions are met will we confirm that the ARDL results are reliable, otherwise remedial measures need to be taken. This paper's robust tests are the Breusch-Pagan-Godfrey test for heteroskedasticity, Breusch-Godfrey LM test for autocorrelation, Jarque-Bera Normality test for normality violation, and Ramsey RESET test for model misspecification from variable omission. The null hypotheses of these tests are no concerning issues of heteroskedasticity, autocorrelation, normality violation, and model misspecification from variable omission, respectively. We have also checked the variance inflation factors (VIFs) to determine whether multicollinearity is worrying. If all the VIFs are less than 10, we conclude that there are no concerning issues of multicollinearity in the ARDL models. The cumulative sum (CUSUM) and CUSUM of square plots are used to check whether the model results are stable. We can confirm the stability of the model results when the residuals lie within the critical lines in the plots.

The short-term impact of HCI on carbon emissions is quantified using error correction (EC) forms of ARDL models, namely error correction models (ECMs), as shown in Equation (9). Δ denotes first-order difference. E C T E D t − 1 , E C T H L t − 1 , and E C T R D t − 1 are the EC terms of Equation (9), and they are equivalent to u 1 t − 1 , u 2 t − 1 , and u 3 t − 1 in Equation (8). a 5 , b 5 , and c 5 are the regression coefficients to denote how fast short-term distortions are corrected to reach long-term equilibria; their expressions are presented in Equation (10). a 1 j − a 4 j , b 1 j − b 4 j , and c 1 j − c 4 j are the coefficients to capture the short-term dynamics unexplained by equilibrium-reverting forces. ⁵⁰ ε 1 t , ε 2 t , and ε 3 t are the residual terms of Equation (9); presumably, they are white noises.

{ Δ C E t = a 0 + ∑ k = 1 p 1 − 1 a 1 k Δ C E t − k + ∑ k = 0 p 2 − 1 a 2 k Δ E C t − k + ∑ k = 0 p 3 − 1 a 3 k Δ G D P t − k + ∑ k = 0 p 4 − 1 a 4 k Δ E D I t − k + a 5 E C T E D t − 1 + ε 1 t Δ C E t = b 0 + ∑ k = 1 q 1 − 1 b 1 k Δ C E t − k + ∑ k = 0 q 2 − 1 b 2 k Δ E C t − k + ∑ k = 0 q 3 − 1 b 3 k Δ G D P t − k + ∑ k = 0 q 4 − 1 b 4 k Δ H L I t − k + b 5 E C T H L t − 1 + ε 2 t Δ C E t = c 0 + ∑ k = 1 r 1 − 1 c 1 k Δ C E t − k + ∑ k = 0 r 2 − 1 c 2 k Δ E C t − k + ∑ k = 0 r 3 − 1 c 3 k Δ G D P t − k + ∑ k = 0 r 4 − 1 c 4 k Δ R D I t − k + c 5 E C T R D t − 1 + ε 3 t

(9)

{ a 5 = ∑ j = 1 p 1 α 1 j − 1 b 5 = ∑ j = 1 q 1 β 1 j − 1 c 5 = ∑ j = 1 r 1 γ 1 j − 1

(10)

In this paper, the short-term HCI impact on carbon emissions, as shown in Equation (9), are assumed to exist over the research period 2021–2030. This assumption is rational as this paper's research period is not long, and thus it is reasonable to assume that the short-term ECM dynamics are extended to the research period.

From Equation (9) and Table S2, we can get the projected impacts of educational, health, and R&D investments on emissions in the BAU scenario. ETS changes HCI and thus affects the HCI impact on emissions. In this paper, HCI is assumed to be proportional to its impact on emissions, as defined in Equation (11). The superscript 0 denotes the BAU scenario. E D I M P t , H L I M P t , and R D I M P t refer to the impacts of educational, health, and R&D investments on emissions.

{ E D I M P t = E D I t E D I t 0 × E D I M P t 0 H L I M P t = H L I t H L I t 0 × H L I M P t 0 R D I M P t = R D I t R D I t 0 × R D I M P t 0

(11)

As HCI is defined as the summation of educational, health, and R&D investments, the HCI impact on emissions is equal to the aggregated impact of educational, health, and R&D investments, as displayed in Equation (12). Noticeably, we do not consider the interactive impacts of educational and health investments or health and R&D investments on emissions as to our best knowledge, few previous researchers have provided such evidence on the interactive impacts. Educational and R&D investments could lay mutual influences on emissions; however, with the statistically insignificant coefficient, the interactive term is excluded in this paper.

H C I M P t = E D I M P t + H L I M P t + R D I M P t

(12)

Scenarios

In this paper, we have designed the following scenarios. The BAU scenario denotes the baseline case without implementing the ETS or considering HCI impact. In the ETI scenario, the designed ETS is implemented without considering HCI impact; conversely, absent from the ETS, the HCA scenario is designed to incorporate HCI impact. In the EHC scenario, we have quantified HCI impact during the ETS implementation.

Autoregressive distributed lag results

To check whether the variables are stationary, we have performed the unit root tests where the inclusions of intercept or time trend are based on the information criteria. The null hypothesis of a unit root test is the existence of a unit root in the tested time series. The results of the Augmented Dickey-Fuller and Phillips-Perron unit root tests are displayed in Table S3 in Supplementary Materials. We cannot reject the null hypotheses of the unit root tests for the variables at the levels and first-order differences, but we can reject the null hypotheses of the tests for the variables at the second-order differences. Hence, the studied variables in Table S3 are not stationary at their levels and first-order differences but stationary at their second-order differences; in other words, these variables are all integrated of order two, denoted by I(2). According to Kripfganz and Schneider, ⁵⁰ when the variables are integrated of order one, the long-term relation embedded in an ECM corresponds to a cointegration relation. As the first-order differences of the variables are integrated of order one, they are introduced in the ARDL models.

Before running the ARDL models, we need to perform the Engle-Granger (EG) cointegration tests to confirm that the cointegration relations are not pseudo. The null hypotheses of EG tests are that the cointegration relations are pseudo. The results of the EG tests are shown in Table S4 in Supplementary Materials. According to Table S4, for the three cointegration relations, the Tau and Z statistics are significant at the 1% level; therefore, the EG tests have verified the cointegration relations.

Based on the information criteria to select the optimal lags, the coefficients of the chosen ARDL models are shown in Table S5 in Supplementary Materials. In Table S5, educational investment does not have a statistically significant impact on carbon emissions. Although health investment does not statistically significantly affect emissions, its second-order lag has a positive and significant impact on emissions. R&D investment has complicated impacts on emissions: its current term statistically significantly decreases emissions at the 5% level, but its first-order lag has a positive and significant impact on emissions. The second-order lag of R&D investment does not statistically significantly affect emissions, but the impact of the third-order lag is negative and statistically significant. Hence, within the domain of HCI, R&D investment plays the predominant role in emission abatement.

The results of the robust tests are displayed in Table S6 in Supplementary Materials. When studying the educational investment impact on emissions, the ARDL model does not have the concerning issues of autocorrelation, normality violation, and model misspecification from variable omission. This is because we have accepted the null hypotheses of the Breusch-Godfrey LM, Jarque-Bera Normality, and Ramsey RESET tests. The statistic of the Breusch-Pagan-Godfrey test is statistically significant at the 10% level, but it is not significant at the 1% and 5% levels; therefore, heteroskedasticity is not a severe problem in the ARDL model. Similarly, the ARDL model for studying the health investment impact does not have the worrying issues of heteroskedasticity, autocorrelation, and normality violation. The issue of variable omission may exist in the model, but it is not a grave issue at the 1% and 5% levels. The ARDL model for studying the R&D investment impact is free from heteroskedasticity, autocorrelation, normality violation, and model misspecification.

Table S7 in Supplementary Materials displays the VIFs of the ARDL models. In the ARDL model for studying the educational investment impact, all the VIFs are less than 10, and thus there is no multicollinearity issue. The ARDL model for studying the health investment impact does not have severe multicollinearity among health investment and its lagged terms, which implies that multicollinearity is not of great concern in this ARDL model. The ARDL model for studying the R&D investment impact is proven to have severe multicollinearity which inflates coefficient variances, but regression coefficients are still unbiased. ⁴⁹ Among R&D investment and its lag terms, only Δ R D I t − 2 has the statistically insignificant coefficient with the p-value much higher than the critical value 5%. Hence, we believe that the coefficient of Δ R D I t − 2 would be still statistically insignificant even without considering the multicollinearity-induced variation inflation.

Figures S1–S3 in Supplementary Materials present the CUSUM and CUSUM of square plots for the ARDL models. In these figures, the residuals all lie within the critical lines; therefore, the residuals are stable. These figures imply that there are no worrying stability issues in the ARDL models.

Table S8 in Supplementary Materials presents the ECMs. In the ECM for studying the educational investment impact, educational investment is excluded, which implies that it does not influence emissions in the short term. Nevertheless, we do consider the economic impacts of educational investment as educational investment increases capital formulation, and it decreases disposable income. In the ECM for studying the health investment impact, the coefficients of health investment and its lagged term are statistically significant; therefore, health investment has short-term impacts on emissions. Similarly, the ECM for studying the R&D investment impact implies that R&D investment is a statistically significant factor that affects emissions in the short term.

As this paper's research period is only 10 years in length, we are particularly interested in the short-term dynamics rather than the long-term equilibrium relation of human capital and emissions. If the short-term dynamics, described in Table S8, exist over the research period 2021–2030, Figure 3 depicts the projected direct impact of HCI on emissions. Health investment was projected to increase emissions since 2024; the R&D investment impact is contrary to the health investment impact. The HCI impact on emissions is defined as the aggregated impact of health and R&D investments. The projected impact of HCI in Figure 3 is inputted into the CGE model in the HCA and EHC scenarios; the results of the CGE model are presented in Section 4.2.

OPEN IN VIEWER

Figure 3.

Direct impact of HCI on emissions. HCI: human capital investment.

Computable general equilibrium results

The projected impacts of HCI on employment and GDP are displayed in Figure S4 in Supplementary Materials and Figure 4. According to these two figures, HCI causes an economic boom characterized by expanding employment and GDP. With the ETS implemented, HCI generates lower economic benefits as the ETS induces deadweight loss and thus economic recession, thereby partially counteracting the positive economic impact of HCI. Nevertheless, the two curves in Figure 4 are convergent in the long term, which implies that the ETS effect diminishes over time. This is because the ETS is designed with the fixed policy content, and thus rational entities may take measures to relieve the negative ETS effect on the economy.

OPEN IN VIEWER

Figure 4.

HCI impact on GDP. HCI: human capital investment.

Figure 5 shows how HCI affects emissions with and without the ETS implementation. HCI increases emissions as it boosts economic growth and thus increases nonrenewable energy consumption. When the ETS is implemented, this HCI impact diminishes because the ETS curbs anthropogenic emissions, and thus it weakens the HCI-induced emission rise.

OPEN IN VIEWER

Figure 5.

HCI impact on carbon emissions. HCI: human capital investment.

In Supplementary Materials, Figures S5–S7 display the ETS effects on educational, health, and R&D investments. According to these figures, the ETS decreases all these investments. Defined as the aggregation of educational, health, and R&D investments, HCI is negatively affected by the ETS, as displayed in Figure 6. Despite the instant fluctuation after the policy implementation, the ETS has a time-decreasing impact on HCI.

OPEN IN VIEWER

Figure 6.

ETS effect on HCI. HCI: human capital investment; ETS: emission trading scheme.

Figure S8 in Supplementary Materials shows how the ETS affects employment, whereas Figure 7 presents the ETS effect on GDP moderated by HCI. Irrespective of considering HCI impact, the ETS generates negative economic outcomes, like employment loss and GDP loss. This is because the ETS curtails carbon emissions by increasing the cost of fossil fuel combustion; therefore, it interrupts the market mechanism and causes inefficient resource allocation. The ETS causes more deadweight loss during the HCI-induced economic boom.

OPEN IN VIEWER

Figure 7.

ETS effect on GDP. ETS: emission trading scheme.

How HCI affects ETS emission abatement is shown in Figure 8. The ETS decreases carbon emissions, and HCI reinforces the emission abatement effect of the ETS. The economic intuition underlying this result is that human capital positively affects emissions, and thus with higher emissions, the ETS causes more emission abatement.

OPEN IN VIEWER

Figure 8.

ETS effect on carbon emissions. ETS: emission trading scheme.