Health effects of ozone and particulate matter pollution in China: a province-level CGE analysis

In this study, we estimate the cost of PM_2.5 and O₃ pollution in China and explore how it differs by province. For the analysis, we extend the China Regional Energy Model—a computable general equilibrium model of the Chinese economy—to explicitly represent the pollution-health linkage within a larger economic system. Our results show that health damage from air pollution in China is substantially large. For each year between 2010 and 2030, China's welfare loss from excess pollution is estimated to be 3.2–5.1% of the baseline level when welfare is measured as the sum of consumption and leisure. The PM_2.5 share of the costs was > 13 times as large as the O₃ share, and premature deaths from chronic exposure to PM_2.5 were the single most important health endpoint, accounting for ≤ 56% of the total costs. Cross-regional heterogeneity is substantial, and populous and wealthy Eastern China is subject to particularly large health damage. When the size of provincial economies is controlled for, however, the dominance of the eastern region is less obvious and several inland provinces (e.g., Henan, Shanxi, and Chongqing) also suffer high pollution-health costs, due to low air quality and fast productivity growth. Finally, broader economic loss from inefficient resource allocation and its cumulative effects, which is often neglected in static analysis, accounts for > 29% of the total costs. Overlooking this cost component will, in particular, lead to substantial underestimation for China's central and western regions, whose economies are growing fast.

Keywords: C68; D58; I18; Q51; Q52; Q53

China has long suffered excess air pollution, and the degree of that pollution has reached a level whose abatement requires immediate public action. Particulate matter (PM) concentrations in China's major cities, for example, exceed the level recommended by the World Health Organization (WHO) by a factor of five or even higher (WHO [79]). Numerous epidemiological studies show that excess air pollution carries severe health risks, posing a serious threat to human welfare. China's recent move toward "rebalancing"—structural transformation aimed at sustainable development—reflects growing recognition of the problem within its policy circles (Nam [43]).

Pollution control in China has been gradually expanded, but further spurring of mitigation efforts seems necessary, given the magnitude and urgency of the excess pollution problem. A robust impact study, examining the socioeconomic costs of neglecting excess air pollution, is perhaps a first step to raising public awareness and thus gaining broader public support. Such a study not only strengthens the need for mitigation actions with scientific evidence but also helps identify the type and magnitude of the actions required. However, the assessment of pollution-caused welfare damage has remained challenging due to the absence of market data that can be used to directly measure the cost of degraded air quality or the benefit of clean air (OECD [50]; World Bank and IHME [77]). A widely accepted approach to valuing air pollution is to treat the human body as a receptor of changed air quality (e.g., Ostro and Chestnut [51]; EPA [14]; Holland et al. [23]; Saikawa et al. [59]; Vennemo et al. [67]; West et al. [70]). Focusing on the pollution-health linkage allows us to incorporate economic transactions (e.g., medical expenses or the lost wages and leisure time required to combat illness) into a larger assessment framework, and thus to integrate environmental externalities into broader economic analyses.

Many studies have been conducted to estimate the adverse economic effects of air pollution in China, focusing on human health. Without exception, they have arrived at very large estimates, ranging from 3 to 5% of China's historic gross domestic product (GDP) (Nielsen and Ho [48]; Zhang et al. [90]). As alarming as these estimates are, they may actually be underestimates, given the studies' methodological limitations (Matus et al. [40]; World Bank and IHME [77]). In brief, the majority of earlier studies draw aggregate damage functions and estimate pollution-health costs by applying these functions to certain air-quality targets in a static manner (e.g., Aunan et al. [1]; Hirschberg et al. [20], O'Connor et al. [49]; World Bank and SEPA [78]). Such aggregate damage-function approaches, however, do not explicitly identify how degraded air quality reshapes the existing supply and demand in the market and thus at best can consider only part of the overall economic impacts. In addition, point estimates based on damage functions also fail to capture the pollution-health effects that remain in the economy beyond a short time period (e.g., labor losses owing to premature deaths).

More recent studies conduct dynamic analysis but treat China as a single entity—this is another critical limitation (e.g., Matus et al. [40]; World Bank and IHME [77]). A coarse spatial unit of analysis is clumsy in reflecting China's subnational heterogeneity in many aspects, such as air quality, urban population, and labor productivity, and can undermine the robustness of the results. A recent city-level study by Nam et al. ([46]), for example, finds that neglecting cross-district variations even within a single city (Hong Kong) in terms of PM levels and socioeconomic conditions can lead to substantial underestimation, compared to when the district is adopted as a spatial unit of analysis. Such noise arising from spatial aggregation is likely even more obvious at the national level, given the magnitude of internal heterogeneity (Li and Nam [32]; Nam [42]; Nam and Reilly [44]; Thompson et al. [65]). Accordingly, subnational details are essential for the analysis of pollution-health effects in China, whose local conditions vary considerably by region.

Little attention to ozone (O3) and the direct application of Western epidemiological evidence to China are other key dimensions subject to improvement. Many studies (e.g., Xie et al. [82]; Zhang et al. [90]) focus solely on PM, but excluding O3 from analysis is a potential source of underestimation, given the solid empirical evidence on O3-associated health damage (Selin et al. [60]). Additionally, using Western epidemiological findings for China, which is common in the literature (e.g., Matus et al. [40]; Xie et al. [82]), can lead to substantial overestimation given the nonlinearity identified in different pollution bands (Burnett et al. [5]; Cohen et al. [12]). In general, the region-neutrality assumption in concentration–response relationships is not supported, and marginal health effects of a unit pollution increase tend to decline with increased ambient concentration levels.

This study is motivated to improve such major methodological limitations and thus the robustness of estimated results. We examine the magnitude of pollution-health costs in China and its cross-provincial variations, with a dynamic and integrated assessment method where an economic simulation model is coupled with an atmospheric chemistry model. Our method explicitly describes the pollution-economy linkage and allows us to trace how economy-wide outputs may change over multiple time periods in response to a given pollution shock. We take the province as the unit of analysis to consider China's subnational heterogeneity, and our analysis focuses on both PM and O3.[1] Of the two popular PM measurement classes for health impact studies, we focus on finer particles with an aerodynamic diameter of ≤ 2.5 μm (PM2.5), rather than those with a diameter of ≤ 10 μm (PM10), given the former's more direct relevance to human health. We have also incorporated recent Chinese epidemiological evidence to avoid potential overestimation resulting from dependence on the Western estimates.

In this study, we explore two main research questions—how large is the socioeconomic cost of excess air pollution in China, and how does it differ by province?—by extending the China Regional Energy Model (CREM) for health effects analysis. In brief, CREM is a recursive-dynamic computable general equilibrium (CGE) model built on 2007 provincial input–output tables and energy data from China and includes 33 regions (30 Chinese provinces [except Tibet] and three global regions) and 16 production sectors (Fig. 1). The model has three agents—households and firms maximize utility and profit, respectively, while the government is modeled as a passive entity in charge of tax collection and redistribution—and conventional closure rules—market clearance, income balance, and zero profit conditions—are applied for an equilibrium.[2] A standard version of CREM is essentially an economic model, lacking representation of the link between air pollution and the economy, and thus does not immediately offer the capacity to explore our research questions. Hence, it must be extended to represent the pollution-economy dynamics, and this extension is our primary methodological contribution. Following conventional approaches, we treat the human body as a receptor of degraded air quality and create a link between air quality and the economy by focusing on pollution-health effects. We call the version of CREM extended for health-effects analysis CREM-HE.

Graph: Fig. 1 CREM's regional and sectoral aggregations

We apply CREM-HE within an integrated assessment framework, in combination with a scientific model of atmospheric chemistry (Fig. 2). Baseline emissions data from the Greenhouse Gas and Air Pollution Interactions and Synergies (GAINS) database (IIASA [26]) for years of our analysis are translated into O3 and PM2.5 concentrations with the Weather Research and Forecasting model coupled to Chemistry (WRF-Chem) (Fig. 3). Then, morbidities and mortalities associated with exposure to pollution are computed using epidemiological relationships, and their effects on the economy are simulated with CREM-HE. A great merit of the CGE model is that it is able to capture not only direct economic costs associated with labor/leisure loss and medical expenditure but also broader economy-wide inefficiency caused by interrupted resource allocation and its cumulative effects over time (Reilly et al. [58]).

Graph: Fig. 2 Analytic framework

Graph: Fig. 3 Baseline air quality layers, 2015: a PM2.5, 24-h mean; b ground-level O3, 8-h daily maximum. Source: Created from WRF/Chem-simulated data

In detail, CREM-HE is modeled to respond to an exogenous air-quality shock in such a way that the economy requires the household sector to produce increased amounts of pollution-health-effects mitigation services, using labor, leisure, and service inputs drawn from other production sectors (Fig. 4a). In this framework, excess pollution imposes both direct (i.e., direct labor and leisure losses from increased morbidities and mortalities) and indirect costs (i.e., efficiency losses from interrupted optimal resource allocation) on the economy. The household production sector, shown in the Mitigation of Pollution Health Effects column under Household Services in the figure, provides final consumers with a pollution-health service whose amount is determined by the increased incidences of morbidity and mortality caused by acute exposure. Production of a pollution-health service requires medical services (drawn from the service sector) and household labor as inputs and thus reduces the amount of service and labor inputs available for other uses. That is, pollution-created demand for health services can be met only at the expense of reduced outputs in other sectors, and its net effects on the economy can be negative.

Graph: Fig. 4 Structure of CREM-HE: a extended social accounting matrix; b simplified production nest. Source: Modified from Matus et al. ([39]), p. 67 and Nam et al. ([45]), p. 5061

Pollution-caused health effects are then modeled as a Hicks-neutral technical change; that is, increased (or decreased) demand for pollution-health services requires proportionally more (or less) of all inputs to deliver the same level of service. The consumption nest shown in Fig. 4b provides an alternative way to look at the extended model. As shown in the figure, the key extensions in the model include (1) leisure as a consumption component and (2) a Household Services sector that is dedicated to mitigating pollution-health effects and has separate production nests for O3 and PM2.5—the two most important air pollutants from a health effects perspective. The substitution elasticity ( ${\sigma }_L$ ) is parameterized to reflect the own-price elasticity of labor supply, which is a widely accepted concept in the economics literature (Babiker et al. [2]). We apply the Leontief production structure to the Household Services sector in relation to other inputs and factors and among different health endpoints. The mortality effects on the economy are modeled as a negative shock to labor productivity, resulting in a loss of work and leisure hours.

Premature deaths from chronic exposure to PM2.5 are treated differently from other health endpoints, as the cardiopulmonary diseases that lead to these mortalities take a fairly long time to develop with only minimal explicit effects during most of the development period. Mortalities in this category can occur to those with normal health conditions and do not require medical services or work/leisure hour loss while they are alive. CREM-HE is modeled to keep track of the lifetime exposure of each age cohort and to remove a fraction of the population groups subject to this mortality category prematurely from the workforce. Premature deaths from chronic exposure affect the economy beyond the specific time period in which they occur, generating dynamic, cumulative effects. For example, assuming 65 to be the average retirement age, if an individual dies at 45 years of age, the economy loses 20-year worth of labor from this individual. CREM-HE tracks these mortalities in each period and adds them forward until the time that each individual no longer remains in the workforce and such mortalities thus would not constitute an explicit economic loss to the economy. The model also conducts a parallel computation for leisure loss, assuming that individuals who left the workforce would no longer earn wage incomes and thus have leisure time alone.

Modeling health effects associated with air quality require the concentration–response (CR) function as key input, and their translation into monetary terms requires a valuation table. In this section, we determine CR functions in China's context with reference to the epidemiological surveys conducted in China and a valuation strategy for each health endpoint. Among conventional air pollutants, we in particular focus on ozone and particular matter, which are known to account for most of the pollution-health costs (Nam et al. [45]; Matus et al. [40]).

The CR function describes how incidence of morbidity or morality responds to a unit pollution level increase. The CR function can be constructed from relative risk (RR) or excess risk (ER) estimates, available in the epidemiological literature. Here, RR is the ratio of incidence rates observed at two different exposure levels and shows the percentage increase in the rate for the experimental group (i.e., those who were exposed to pollution) from the baseline level for the control group (i.e., those who were not exposed). ER conveys similar information, but takes the difference, not the ratio, between the incidence rates where the baseline rate is normalized to one.

For short-term effects estimation associated with acute exposure to pollution, we consider three dimensions—health endpoint $i$ , region $r$ , and time $t$ —and use the CR function shown in Eq. 1 (Nam et al. [46]).

1 $$y_{\mathit{irt}}^{\text{SE}}=\frac{1}{10}\left(1-\frac{1}{{\text{RR}}_i}\right)·F_i·P_{\mathit{rt}}·x_{\mathit{rt}}$$

Graph

In the equation, $y_{}^{\text{SE}}$ is the net incidence of morbidities and premature deaths attributable to acute exposure to pollution level $x$ (measured in µg/m3), and $\text{RR}$ , $F$ , and $P$ refer to relative risk, baseline incidence rate, and total population, respectively. The constant $1/10$ of the right-hand side of the equation is to convert the available RR estimates for a 10 µg/m3 pollution band into a unit value defined on a 1 µg/m3 concentration change.

In the case of long-term effects associated with chronic exposure to PM2.5, the integrated exposure–response (IER) function, shown in Eq. 2, is used to consider a nonlinear epidemiological response with regard to ambient pollution level (Burnett et al. [5]).

2 $$\text{RR}\left(x\right)=\left\{\begin{array}{cc}1\hfill & \text{for}x<x_{\mathit{cf}}\\ 1+\alpha \left(1-e^{-\gamma {\left(x-x_{\mathit{cf}}\right)}^{\delta }}\right)\hfill & \text{for}x\ge x_{\mathit{cf}}\end{array}\right)$$

Graph

Here, $\alpha$ , $\gamma$ , $\delta$ , and $x_{\mathit{cf}}$ are parameters and are as defined in Burnett et al. ([5]). The IER posits that the marginal effect of a unit pollution level increase on mortality rate tends to decline as a background ambient pollution level increases (Cohen et al. [12]).

The CR function for long-term effects having four dimensions—health endpoint $i$ , age cohort $j$ , region $r$ , and time $t$ —is then constructed as shown in Eq. 3 (Nam et al. [46]).

3 $$y_{\mathit{ijrt}}^{\text{LE}}=\left(1-\frac{1}{R{R}_i\left(x_{\mathit{rt}}\right)}\right)·F_i·w_j^F·P_{\mathit{rt}}·D_{\mathit{jrt}}$$

Graph

In the equation, $y_{}^{\text{LE}}$ is the number of premature deaths that occurred due to chronic exposure to pollution level $x$ , and $\text{RR}$ is a nonlinear function of $x$ , as described in Eq. 2. $F$ and $P$ are as defined earlier, and $D_{\mathit{jrt}}$ is the share of age group $j$ in region $r$ at time $t$ . Addition of an age dimension $j$ is to consider age-varying long-term effects, and $F_i$ is adjusted with $j$ -specific weight ( $w_j^F$ ) (Matus et al. [40]). Here, $w_j^F$ is defined as the relative importance of cardiopulmonary diseases (CPL) in all-cause mortality rates for age group $j$ ( $M_j^{\text{CPL}}/M_j$ ), compared with that for entire population group ( $M_{}^{\text{CPL}}/M$ ) (Nam et al. [45]).

In this study, we consider nine health endpoints in total, whose epidemiological relationship with O3 and PM2.5 exposure is well established (WHO [79]). Five of them fall into the acute-exposure category: mortality from acute exposure (AM), hospital admissions associated with cardiovascular (CHA) or respiratory diseases (RHA), emergency room visits (ERV), and outpatient visits (OV). The other four health endpoints are cardiopulmonary diseases, known as the main causes of premature deaths associated with long-term exposure to PM2.5 pollution. This chronic exposure category includes ischemic heart disease (IHD), stroke, chronic obstructive pulmonary disease (COPD), and lung cancer (LC). These diseases take substantial time to develop with chronic exposure, and gradual damage to heart and lung functionalities over a long-term period eventually reduces life expectancy.

A good number of RR estimates are available for acute-exposure endpoints, but are subject to significantly large standard deviations, depending on the time and place each study is conducted. A main reason is the nonlinearity that exists in the epidemiological relationship: marginal health effects of a unit pollution level change tend to decline when absolute ambient pollution levels increase. Accordingly, RR estimates for China, which suffers fairly high O3 and PM levels, tend to be lower than those for Western countries located in much lower pollution bands. Given this fact, we avoid applying Western RR estimates to China, and consider only those estimates from China-conducted studies, displayed in Figs. 5 and 6. Among these Chinese estimates, we eventually pick one for each endpoint by prioritizing those estimates located near the median and subject to lower levels of uncertainty. The RR estimates finally chosen for CREM-HE are listed in Table 1.

Graph: Fig. 5 Excess risk (%) per 10 μg/m3 increase in PM2.5 concentrations: a mortality from acute exposure; b morbidities. Note: *Meta analysis of multiple Chinese studies; ^estimate for COPD; ∨estimate for asthma

Graph: Fig. 6 Excess risk (%) per 10 μg/m3 increase in O3 concentrations: a mortality from acute exposure; b morbidities. Note: *8-h mean; **8-h daily maximum; #1-h mean; ##24-h mean; ^estimate for COPD; ∨estimate for asthma

Relative risk estimates used in this study for short-term exposure endpoints. Source: Created by the authors

aRelative risk is defined on an increment of 10 µg/m3 in pollutant concentrations bConverted from an estimate on TSP, applying the PM2.5/TSP ratio of 1/3 (Burnett et al. [5]; Cao et al. [6]) cMeta analysis of multiple epidemiological studies conducted in Chinese cities, including Beijing, Shanghai, Chongqing, Wuhan, Taiyuan, and Hong Kong

For chronic exposure effects analysis, we adopt IER parameter estimates by Morita et al. ([41]), based on the 2013 Global Burden of Disease study (GBD 2013 Risk Factors Collaborators [15]) database (Table 2). As explained earlier, premature mortalities from chronic exposure take a long time to develop and mostly occur to adult groups. In this regard, we assume that only those at the age of ≥ 30 are subject to the long-term effects, following the literature (e.g., Bickel and Friedrich [4]; Matus et al. [40]; Nam et al. [45]). The incidence of such premature deaths also tends to be an increasing function of age (i.e., cumulative time of being exposed to pollution), and these age-varying effects are considered with an age-specific adjustment term $w_j^F$ in Eq. 3.

IER parameters used for long-term effects analysis. Source: Morita et al. (2014), p. 14661

For all cases, $x_{\mathit{cf}}$ is given as 7.5 μg/m3; 95% confidence intervals are within parentheses

Health damage from air pollution incurs to the economy both market (increased medical expenditure and reduced labor hours) and non-market (leisure loss) impacts. Accordingly, morbidity and mortality cases computed using CR functions are valuated in monetary terms, considering both impact dimensions. For the valuation of health endpoints associated with acute exposure, we apply the valuation table constructed from available market and surveyed data (Table 3). For hospital admission and outpatient visit categories, we adopt the 2005 estimates by Lei and Ho ([31]), which include medical expenditure and associated wage loss computed directly from the P.R.C. Ministry of Health statistics. We then add our estimates for leisure loss to these estimates[3] and adjust them for 2007 by using the GDP deflator. In the absence of available statistics, we assume twice the costs for outpatient visits for the emergency room visit category, referring to World Bank ([76]). China's national estimates are then adjusted for each region, using cross-provincial per capita disposable income ratios (Fig. 7).

Costs associated with short-term exposure health endpoints

aHalf the value of a life year computed from Hoffmann et al. ([21]), with urban–rural population and income ratio adjustments b 2005 estimates by Lei and Ho ([31]) are adjusted for 2007 by using the GDP deflator. Final values include the value of associated leisure loss, referring to Yang ([85]) and Matus ([38]) cTwice the value for an outpatient visit is assumed, based on World Bank and SEPA ([78]) d1999 estimates by Zhou and Hammitt ([91]) are adjusted for 2007 by using GDP deflator

Graph: Fig. 7 Disposable income per capita by province, 2015. Source: Created from National Bureau of Statistics of China ([47])

Valuation of mortality is based on a stated preference approach (Viscusi et al. [68]). Mortalities from acute exposure to excess PM pollution occur mostly to those whose health conditions are already impaired and who are close to death from other causes (Pope et al. [53], [54]). For this reason, labor and leisure loss associated with premature deaths from acute exposure is conventionally valuated at a quarter to a full life year (Holland et al. [22], [23]). In this study, we assume that a mortality from acute exposure incurs to the economy the cost equivalent to half a life year. The value of a life year (VOLY) estimate used for our study is based on the 2006 willingness-to-pay (WTP) survey conducted for three Chinese cities by Hoffmann et al. ([21]). The value of statistical life (VSL) estimate, drawn directly from the surveyed results for China, is 2016 US$614,805, and this estimate is further adjusted to 2007 prices considering inflation, income growth, and urbanization ratios. The adjusted VSL estimate for the sample with mean age 54 is then 2007 US$336,685, and the corresponding VOLY at a discount rate of 3% is 2007 US$21,204.

Premature deaths from chronic exposure are valuated in a different way, as they can occur to those with normal health conditions without involving explicit medical expenses. Primary socioeconomic costs associated with this endpoint are related to the loss of the labor and leisure that would not have occurred if those who died prematurely avoided their chronic exposure. Once premature deaths are computed by region and age group, the net loss of labor and leisure is estimated, considering each region's mean life expectancy. Each mortality case in this category is then valuated using mean wage levels for each region, which are endogenously determined within CREM-HE. A unit leisure loss is evaluated at the same level as a unit labor loss, as lost wage is treated as an opportunity cost of leisure in our model (i.e., each worker has a choice between labor and leisure, and leisure is chosen over labor only when the former is seen as at least as valuable as the latter).

For our analysis, we simulate CREM-HE with four scenarios, each of which assumes different O3 and PM2.5 levels. First, the Reference (business-as-usual) scenario assumes no further pollution abatement efforts beyond the existing ones. Baseline O3 and PM2.5 levels used for this scenario are from WRF-Chem with GAINS-East Asia emissions data (ECLIPSE 5Va scenario reflecting China's legislation during the FYP12 period), and CREM-HE is calibrated to simulate baseline gross domestic product (GDP) levels for each region under this scenario. This calibration setup intends to benchmark an extant market equilibrium which already incorporates the effects of historic ambient air pollution (Nam et al. [45]). A pairwise comparison of a counterfactual scenario case with the Reference simulation outcomes will then help us analyze the net effects of associated policy shocks measured in equivalent variation, which is a standard approach to impact assessment in dynamic CGE studies (Paltsev et al. [52]). Baseline GDP values for past years are set at historic levels drawn from official statistics, and historic provincial growth trends are extrapolated to future years with an assumption of convergence at an annual growth rate of 3% by 2050 (Chen et al. [10]; Zhang et al. [89]).

Second, the Green scenario adopts background air quality close to preindustrial levels (10 μg/m3 for O3 and 2.5 μg/m3 for PM2.5), which could be achieved with minimal anthropogenic pollution. Third, the Policy I scenario assumes that all Chinese regions meet WHO or comparable global pollution control targets (70 μg/m3 for O3 and 10 μg/m3 for PM2.5). Finally, the Policy II scenario assumes that province-specific PM2.5 reduction targets proposed during the 13th Five Year Plan (FYP13) period (2016–2020) are met while O3 levels in the same period are set at the baseline levels in the Reference scenario (see Sect. 4.4 for province-specific mitigation targets).

Excess PM2.5 and O3 pollution is estimated to cause substantial socioeconomic costs to the Chinese economy during the time period analyzed. Throughout this study, we report the costs in two metrics, using an equivalent variation concept. One is consumption loss, which counts net wages lost due to pollution and thus captures market effects alone; the other is welfare loss, which tracks net changes in both wage and leisure given a pollution shock and thus considers non-market impacts, as well.

Absolute welfare costs from excess pollution continue to rise during the period analyzed, due to solid labor productivity growth (i.e., growing unit value of labor lost from pollution) (Fig. 8a). When the Green case is compared with the Reference results—health damage associated with anthropogenic pollution—consumption loss in each year between 2007 and 2030 is estimated to range from US$75 billion to US$263 billion, showing an increasing time trend (Table 4).[4] Estimated welfare loss is greater (US$141 billion to US$466 billion), but shows a similar time trend. When the Policy I case is compared with the Reference case—welfare costs that could be avoided with realistic air-quality control—consumption and welfare loss is US$56 billion to US$199 billion and US$106 billion to US$306 billion, respectively.

Graph: Fig. 8 Pollution-health costs in China, and their decomposition

Estimated costs of anthropogenic air pollution in China. Source: Numbers within parentheses refer to the shares of baseline consumption or welfare levels

Unit: billions of 2007 US$

When the costs are measured in relative terms, however, they show a declining trend over time. In the Green-Reference comparison, the pollution-health costs decline from 5.6% of the baseline consumption level (or 5.6% of the baseline welfare level) in 2007 to 3.4% (or 3.2%) in 2030. Similarly, the costs under the Policy I scenario drop from 4.2% of the baseline consumption level (or 4.2% of the baseline welfare level) in 2007 to 2.6% (or 2.5%) in 2030. These results are closely related to a declining trend in O3 and PM2.5 pollution since 2015, which contrasts continued national economic growth. That is, China's overall economic growth increasingly exceeds the rate at which the pollution-health costs increase.

When the welfare loss is decomposed into direct and indirect components, each category accounts for a comparable share of the total loss (Fig. 8b). Direct loss includes payment for medical services and labor/leisure hours lost to combat illness and contributes to 44–71% of the total welfare loss. Premature deaths from chronic exposure to PM2.5 is the single most important health endpoint sub-group, accounting for 78–80% of the direct loss or 34–56% of the total welfare loss in each year between 2007 and 2030. Morbidities and mortalities associated with acute exposure to O3 and PM2.5 also cause substantial costs, explaining 20–22% of the direct loss. When decomposed by pollutant, PM2.5 dominates O3, in terms of cost share. PM2.5 accounted for > 93% of the direct loss, presenting > 13 times the O3 share.

We also consider indirect loss, arising from cumulative allocative inefficiency and welfare damage. At one level, pollution interrupts an efficient market equilibrium achievable in its absence. Meeting increased demand for health services leads to service-biased resource allocation while reducing inputs available for other sectors and thus curbing their production. At another level, labor loss from pollution reduces regional product, and lower gross income at one time point leads to less consumption and investment in later years, undermining future growth potential. Thus, the effects of welfare loss remain in an economy over a fairly long time period. We call such indirect costs broader economic loss (Matus et al. [40]). The broader economic loss category, which is often neglected in the literature, accounts for 30–56% of the total welfare loss. This result suggests that conventional static analysis failing to capture such dynamic dimensions of the pollution-health damage is likely subject to substantial underestimation.

In absolute terms, welfare loss from pollution is high in the Eastern region (Fig. 9a/c). Guangdong, Jiangsu, Shandong, and Zhejiang, in particular, suffer particularly large welfare damage—e.g., ≥ US$15 billion in 2010 and ≥ US$30 billion in 2030—compared to other Chinese regions, during the period analyzed. Key drivers underlying these results are the absolute size of pollution-exposed urban population and the level of labor productivity, as evinced by the fact that all these provinces are among the most populous and affluent local economies. Air quality also contributes to these results, but not as much as those two drivers do. For example, annual PM2.5 and O3 levels in Guangdong and Zhejiang are substantially below the population-weighted national mean, while those in Jiangsu and Shandong exceed the latter by a wide margin.

Graph: Fig. 9 Provincial distribution of welfare loss in 2010 and 2030: a absolute loss in 2010, Green (vs. Reference); b relative loss in 2010, Green (vs. Reference); c absolute loss in 2030, Policy I (vs. Reference); d relative loss in 2030, Policy I (vs. Reference)

Despite overall air-quality improvement, the absolute size of welfare loss in most regions tends to increase over time due to population and productivity growth, but the pace of its increase varies across regions. Overall, absolute damage grows faster in inland provinces than coastal regions (Fig. 10a). Most notably, the welfare costs in Hubei, Hunan, Sichuan, Shaanxi, and Chongqing are increased by a factor of 3.5–5.3 between 2010 and 2030, compared with a factor of 1.7–2.2 for Guangdong, Jiangsu, Shandong, and Zhejiang. This result is closely related to East-biased pollution regulations (i.e., greater air-quality improvement in coastal mega-city regions) and cross-provincial productivity convergence (i.e., faster productivity growth in inland regions).

Graph: Fig. 10 Pollution-health costs by province in 2010 and 2030: a in absolute terms; b in relative terms. Note: Dashed lines indicate a national mean of province-level welfare damage

Welfare loss in relative terms shows a slightly different picture (Figs. 9b/d and 10b). When the size of each region's economy is controlled for, northern and inland regions subject to higher pollution levels, as well as populous coastal provinces, tend to suffer greater welfare loss than other parts of China. Those northern regions include the capital region (Beijing, Tianjin, and Hebei) and Shandong, showing a welfare damage of 3.3–7.0% of the baseline levels between 2010 and 2030, and those inland regions include Henan, Shanxi, Chongqing, Hubei, Hunan, and Sichuan, presenting a welfare damage of 3.7–5.9% during the same period. In contrast to the absolute welfare loss case, relative welfare damage tends to decline over time with gradually improving air quality—pollution-health costs grow more slowly than provincial GDP does.

When China is split into three regions, the Eastern region is distinguished from its Central and Western counterparts, in terms of the time trend of regional pollution-health costs (Fig. 11). In absolute terms, pollution-health costs in all three regions tend to increase over time, but the Eastern region suffers much higher welfare damage (US$128–269 billion) than the Central (US$39–126 billion) or Western region (US$16–70 billion). This result is understandable, given that the Eastern region consists of wealthy and populous coastal provinces: a greater urban population in this region is exposed to excess pollution, and the region's higher labor productivity translates the health damage of the exposed population into larger labor and leisure loss value. The growth of the welfare damage in the Eastern region slows down over time—more obviously between 2020 and 2030—while it maintains a quasi-linear pattern in the Central and Western regions. The main reason is that compliance with more stringent pollution regulations, which have been imposed on the Eastern region since the FYP12 period, will lead to regional air-quality improvement in greater magnitude, compared with its Central and Western counterparts.

Graph: Fig. 11 Pollution-health costs in China, and their decomposition

Cost decomposition also exhibits striking cross-regional variations. In the Eastern region, the share of the broader economic loss category remains quite stable at 50–61% of the total welfare loss during the period between 2010 and 2030. In the Central and Western regions, however, the direct loss category dominates the total welfare loss in 2010, but its share becomes increasingly overshadowed by the broader economic loss category over time. In these regions, the share of the broader economic loss is only 6–16% in 2010, but increases to 52–61% in 2030. This is primarily due to relative regional growth rates—the Central and Western regions are behind the Eastern region in development in absolute terms, but are projected to continue to grow faster than the latter, presenting a tendency of beta-convergence. Labor and consumption loss at one time point may lead to lower investment and production in the future time periods in a cumulative manner, and the magnitude of such cumulative effects tends to be greater for faster-growing economies.

Among the key parameters used for this study, RR factors are subject to relatively large uncertainty, and can potentially introduce substantial noise into our central estimation results presented in earlier sections. For this reason, we conduct a sensitivity analysis with regard to RR, taking its known 95% confidence interval for each health endpoint, and report the upper and lower bounds of the cost estimates. The high and low RR threshold values tested in this section are drawn directly from the literature and are as listed in Tables 1 and 2.

The pollution-health costs estimation results tend to be sensitive to RR parameters. When the Green scenario is compared with the Reference case, consumption loss associated with O3 and PM2.5 pollution is estimated to range from US$32 billion to US$115 billion (1.5–2.4% of the baseline level) with lower RR thresholds (Table 5) and from US$106 to US$363 billion (4.8–8.2% of the baseline level) with upper RR thresholds (Table 6). Welfare loss also presents a similar pattern, with the cost estimates of US$61 billion to US$205 billion (1.4–2.3% of the baseline level) at their lower bounds and US$198 billion to US$639 billion (4.5–8.0% of the baseline level) at their upper bounds. This result suggests that RR-involved uncertainty can lead to a bias of [− 57%, 41%] from our central estimates in absolute terms. The Policy I case comparison results are similar to the Green case comparison, presenting an estimated consumption (or welfare) damage of US$23 billion to US$84 billion (or of US$44 billion to US$148 billion) with lower bound RRs and that of US$85 billion to US$298 billion (or of US$160 billion to US$530 billion) with upper bound RRs. In this case, the potential bias introduced by RR-embedded uncertainty ranges in [− 59%, 51%].

Estimated pollution-health costs with lower bounds of CR functions. Source: Numbers within parentheses refer to the shares of baseline consumption or welfare levels

Unit: billions of 2007 US$

Estimated pollution-health costs with upper bounds of CR functions. Source: Numbers within parentheses refer to the shares of baseline consumption or welfare levels

Unit: billions of 2007 US$

Despite the sensitivity, our central findings and general conclusions are left intact. First, pollution-health costs in China's context are not trivial, given the size of China's economy and its rapid growth. Even the lower bound cost estimates under the Green and Policy I scenarios reach ≥ 1.4% and ≥ 1.0% of the baseline welfare levels. Second, two key time trends found in our central results—growing absolute pollution-health costs with labor productivity growth and declining relative costs reflecting air-quality improvement over time—still hold in our sensitivity analysis. Finally, error bars apply to each region and each health endpoint in the same direction and with a comparable magnitude; thus, the patterns of cost distributions across regions and among health endpoints are found to be consistent with our main results.

Since 2013, China has imposed strict air-quality control targets on its major urban agglomerations, such as Beijing-Tianjin-Hebei (BTH), Yangtze River Delta (YRD), and Pearl River Delta (PRD), with a goal to push these regions meet the national class 2 ambient air-quality standards (35 µg/m3 in annual means) by 2035 (CAAC [11]; Xu and Stanway [83]). In line with this effort, FYP13 specifies official national level PM reduction targets for the first time in its history, and Chinese local governments have introduced their own mitigation targets in accordance with the national goals. Subnational FYP13 PM reduction targets for 2020 vary by province, ranging up to 36% reduction from the 2015 concentration levels (Fig. 12a). As briefly mentioned in Sect. 4.1, the Policy II scenario is simulated with CREM-HE to estimate potential health benefits which can be attained when these PM mitigation targets are met.

Graph: Fig. 12 FYP13 PM2.5 reduction targets and expected welfare benefit: a provincial reduction targets for 2020 (vs. 2015 concentration level); b avoided health damage with compliance with the FYP13 targets, 2020 (vs. baseline welfare level)

Our results show that compliance with FYP13 PM2.5 mitigation targets will help the Chinese economy avoid substantial health damage from pollution. In 2020 alone, for example, avoided welfare (or consumption) loss thanks to FYP13-stated stricter PM2.5 regulations is estimated to be US$98 billion (or US$50 billion), comparable to 1.2% (or 1.1%) of the baseline welfare (or consumption) levels. When the avoided damage is measured in GDP, it is US$137 billion or 1.2% of the baseline GDP level. At local levels, the avoided damage tends to be positively associated with the relative stringency of the proposed targets, and those with more ambitious reduction targets, such as the BTH region, Shandong, and Henan, demonstrate particularly large benefits from the proposed PM regulations in both absolute and relative terms (Fig. 12b).

In this study, we first introduce an analytic framework for pollution-health effects analysis and then apply the methodology to China at the province level. Our results show that health damage from O3 and PM2.5 pollution in China is still substantially large, although associated welfare loss in relative terms (vs. baseline welfare levels) presents a declining trend with China's increased mitigation effort. For each year between 2010 and 2030, China's welfare (or consumption) loss from excess pollution is estimated to be ≤ 5.1% (or ≤ 5.2%) of the baseline level, suggesting the need for continued pollution control. The PM2.5 share of the costs was > 12 times as large as the O3 share, and premature deaths from chronic exposure to PM2.5 were the single most important health endpoint, accounting for ≤ 56% of the total costs. This result justifies increased stringency of China's ambient PM2.5 control standards. Compliance with FYP13-proposed PM reduction targets, for example, is estimated to help China avoid a welfare damage of US$98 billion in 2020 alone.

Cross-regional heterogeneity is substantial, and populous and industrialized Eastern China is subject to particularly large health damage in absolute terms. When the size of provincial economies is controlled for, however, Eastern dominance is less obvious and several inland provinces (e.g., Henan, Shanxi, and Chongqing) also suffer large welfare loss from O3 and PM2.5 pollution, due to low air quality and fast productivity growth. Broader economic loss from pollution—indirect economic costs arising from inefficient resource allocation and its cumulative effects—accounts for > 29% of the total costs for the Chinese economy, and the share is particularly large in the Central and Western regions (> 50%) during the later periods analyzed. Overlooking this indirect cost component, prevalent in typical static analysis, likely leads to substantial underestimation for fast-growing economies like China—its Central and Western regions, in particular.

Our study can contribute to the literature in several aspects. First, our analytic framework embodies a methodological advancement in bringing environmental pollution into conventional macroeconomic accounts within a top-down modeling structure. Pollution is modeled as an explicit external shock to the economy, and its market (labor and production sectors) and non-market (leisure) effects can be examined in a dynamic setting. Second, improved unit of analysis with reasonable subnational level details is another aspect to be highlighted. CGE application to China often takes a coarse model resolution (e.g., entire country as a single entity) due to data constraints and is thus clumsy in considering subnational heterogeneity. In contrast, CREM-HE is built on a comprehensive subnational dataset and offers capacity for realistic province-level analysis. This is a great merit, given that the magnitude of pollution-health effects is a function of varied local conditions. Also, the model's incorporation of local epidemiological evidence drawn from China-conducted studies represents an improvement over many existing studies adopting Western estimates for China, given the recent empirical evidence on nonlinearity in CR relationships. Finally, our analysis focuses not only on fine particles but also on ozone, which is often excluded from analysis despite its critical impacts on human health.

The significance of the study may go beyond its methodological contribution, given the topic's close relevance to China's ongoing debates on rebalancing or new-type urbanization. Realistic estimates of pollution-health costs at the province level, drawn from this study, could serve as the basis for drafting effective pollution abatement strategies at multiple administrative levels. It is a caveat, however, that this study focuses on the potential benefits from pollution mitigation (or the health damage that could be avoided in the absence of excess air pollution) without considering associated policy-compliance costs. A balanced cost–benefit study would need to attend to the cost dimension, as well, and thus a follow-up study could incorporate such regulatory costs into a dynamic impact assessment framework.

We acknowledge the financial support of the Hong Kong Research Grant Council (ECS Award No.: 27200915) and the National Natural Science Foundation of China (Project No.: 71690244). The authors would like to express our special thanks to Da Zhang and Xiaohan Zhang for their helpful comments and input.