Impacts of Carbon Pricing in Reducing the Carbon Intensity of China's GDP

In contributing to global climate change mitigation efforts as agreed in Paris in 2015, China has set a target of reducing the carbon dioxide intensity of gross domestic product by 60-65 percent in 2030 compared with 2005 levels. Using a dynamic computable general equilibrium model of China, this study analyzes the economic and greenhouse gas impacts of meeting those targets through carbon pricing. The study finds that the trajectory of carbon prices to achieve the target depends on several factors, including how the carbon price changes over time and how carbon revenue is recycled to the economy. The study finds that carbon pricing that starts at a lower rate and gradually rises until it achieves the intensity target would be more efficient than a carbon price that remains constant over time. Using carbon revenue to cut existing distortionary taxes reduces the impact on the growth of gross domestic product relative to lump-sum redistribution. Recycling carbon revenue through subsidies to renewables and other low-carbon energy sources also can meet the targets, but the impact on the growth of gross domestic product is larger than with the other policies considered.

In contributing to global climate change mitigation efforts as agreed in Paris in 2015, China has set a target of reducing the carbon dioxide intensity of gross domestic product by 60-65 percent in 2030 compared with 2005 levels. Using a dynamic computable general equilibrium model of China, this study analyzes the economic and greenhouse gas impacts of meeting those targets through carbon pricing. The study finds that the trajectory of carbon prices to achieve the target depends on several factors, including how the carbon price changes over time and how carbon revenue is recycled to the economy. The study finds that carbon pricing that starts at a lower rate and gradually rises until it achieves the intensity target would be more efficient than a carbon price that remains constant over time. Using carbon revenue to cut existing distortionary taxes reduces the impact on the growth of gross domestic product relative to lump-sum redistribution. Recycling carbon revenue through subsidies to renewables and other low-carbon energy sources also can meet the targets, but the impact on the growth of gross domestic product is larger than with the other policies considered. -2-

Introduction
In its Nationally Determined Contribution (NDC) coming out of the Paris Agreements reached at the 21 st meeting of the Conference of Parties to the United Nations Framework Convention on Climate Change (UNFCCC), the Chinese government has set a goal to cut CO2 emissions per unit of GDP by 60 to 65 percent by 2030, compared to 2005 levels. It also has reiterated previously announced goals that carbon emissions would peak, and that the share of non-fossil fuels would increase to 20% of total energy consumption, by 2030. Earlier in 2016 the government announced the 13 th Five-year Plan, with the goals of reducing energy per unit GDP by 18% between 2015 and 2020 and to reduce CO2 per unit GDP by 20%. The longer term targets imply that China would need additional low carbon infrastructure investments, including further deployment of renewable energy, as well as further improvements in energy efficiency.
This study addresses the following "what-if" question: what would be the economic consequences of China seeking to meet its NDC target with carbon pricing? In practice, China can be expected to use a range of policy instruments to that end. However, an examination of economic consequences using a hypothetical least-cost instrument, a dynamic carbon price, provides a useful context for evaluating more complicated policy portfolios. Moreover, an examination of carbon pricing is consistent with the expressed desire of the government to rely more on market mechanisms.
The analysis is carried out using a multi-sector growth model of China, including a household model component that allows us to discuss the impacts of policies on different demographic groups. While there exist a large number of studies analyzing impacts of a hypothetical carbon tax in China, studies analyzing the potential impacts of carbon pricing to meet China's NDC do not exist. To the knowledge of authors, this is the first analysis that examins implications of meeting China's NDC through a carbon pricing mechanism.
We first use the model to develop a base case up to 2030. The base case includes existing plans to expand the renewable energy and nuclear energy included in China's 12 th and 13 th Five year plans out to 2020. The base case does not include carbon pricing. The government has announced an intention to extend current experimental carbon trading programs to a national system beginning in 2017. However, no details of the level of cap and coverage of sources have been given, and so the planned national emission trading scheme is not included in the study. In the base case, the carbon emissions are rising continuously during the 2012-2030 period, though -3-they peak around 2030 through various GHG mitigation measures already in the base case. The base case is then compared to scenarios with carbon pricing as well as a policy for additional investment in renewables for electricity generation. Carbon pricing is implemented in the model through surcharges on fossil fuel prices based on their carbon contents.
The study considers the following policy formulations. First, "lower" carbon pricing and "higher" carbon pricing trajectories are implemented to achieve, respectively, 60% or 65% reduction of carbon intensity of GDP from 2005 levels. For the 65% target, we compare a fixed carbon surcharge on fossil fuels over time to a trajectory where the surcharge starts at a low level but gradually rises.
As has been emphasized by many others in the literature, the method of recycling the revenues generated from carbon pricing is another important policy component, as it has an important impact on the net economic cost of such a pricing policy. We consider two approaches to recycle the carbon revenue to the economy-a lump sum transfer to households, and a cut in the VAT and capital income tax. In addition, the study considers a policy that subsidizes renewable energy with revenue generated through the carbon pricing.
We find that the carbon surcharge on fossil fuels aimed at achieving a 60% reduction in CO2 intensity of GDP would rise from 1.6 yuan /ton of CO2 in 2016 to 26 yuan/ton of CO2 in 2030 (all units measured in 2010 yuan). This carbon price will reduce the absolute level of 2030 emissions by 3.3% with GDP only 0.11% lower than the base case, if carbon revenue is recycled to cut existing taxes. In this scenario, 2030 energy use falls by 2.6% and electricity use by 1.5%.
With the carbon pricing policy to achieve a 65% reduction in CO2 intensity of GDP, the carbon surcharge on fossil fuels rises to 157 yuan/ton of CO2 in 2030, generating a 13% reduction in energy use, a 16% reduction in CO2 and a 0.74% lower GDP, all relative to the base case..
If instead of cutting existing taxes, the carbon revenues are recycled by giving them to households as a lump sum rebate, then GDP is 1.2% lower than the base case in 2030, instead of 0.74%. In this case, the same carbon price slows aggregate output growth more, and hence also slows growth energy consumption and emissions more. This reinforces the lesson emphasized by many others -recycling revenues from carbon pricing by reducing existing tax wedges is a useful opportunity for softening adverse economic impacts.
The paper is organized as follows. Section 2 briefly highlights the CGE methodology developed for this study (detailed description of the model is presented in the Appendix A), -4-followed by results of the model for the base case scenario in Section 3. Section 4 presents results of carbon pricing scenaruios to meet China's NDC, followed by sensitivity analysis in which the baseline is altered in Section 5. Section 6 concludes the paper.

The CGE model
The CGE model used for this study is a dynamic recursive growth model where the main agents are the household, producers, government and the rest of the world. Household savings, enterprise retained earnings and government-funded investments are the main sources of investment; unlike most developed economies the government role in China is much larger.
Detailed description of the model is presented in Appendix A, here we summarize its key features.
One of the key features of this model that distinguishes it from large number of CGE models available for China for climate policy analysis is that it allows for heterogeneity among households, whereas most existing CGE models have a single representative household. In our model, households are distinguished by size, presence of children, age of head and region. The income elasticity is different for different consumption items and thus projects a different structure of consumption in the future when incomes rise. Such a function allows us to distinguish the impact of policies on different households via the consumption channel. 3 Labor is supplied inelastically by households.
The private household savings rate is set exogenously and total national private savings is made up of household savings and retained earnings of enterprises. These savings, together with allocations from the central plan, finance national investment. They also finance the government deficit and the current account surplus. Investment in period t increases the stock of capital that is used for production in future periods. The plan part of the capital stock is assumed immobile in any given period, while the market part responds to relative returns. Over time, plan capital is depreciated and the total stock becomes mobile across sectors.
The government imposes taxes on value added, sales and imports. On the expenditure side, it buys commodities, makes transfers to households, pays for plan investment, makes interest payments on the public debt and provides various subsidies. Expenditures on interest and 3 See Equations A20-A24 for the household demand function in Appendix A.
-5-transfers are exogenous, and the exogenous deficit target is met by making government spending on goods endogenous.
Finally, the rest of the world supplies imports and demands exports. Domestically produced goods are imperfect substitutes for imports. World relative prices and the current account balance are set exogenously in this one-country model, and an endogenous variable for terms of trade clears the current account equation. The world price of commodity i relative to j is assumed to be at base year ratios throughout the projection period with the exception of world oil prices, where projections from the U.S. Energy Information Administration 2013 Annual Energy Outlook are used (and treated as exogenous).
On the supply side, 33 industries are distinguished, each producing output that is given by a nested series of constant-returns-to-scale CES functions. 4 Primary factors include capital, labor and land. Pure TFP growth and biased technical change are allowed; in particular, energy input per unit output can decline faster than other input-output ratios.
Since the electricity sector is the main contributor of CO2 emissions in China, in the version of the model used for this study, the electricity sector is disaggregated into one transmission subsector and 9 distinct generation sub-sectors -coal, gas, nuclear, hydro, other, wind, solar, coal-CCS and gas-CCS. The tier structure of electricity production is given in Figure   1. At the top node total output is the aggregate of Generation and Transmission, and Generation is in turn an aggregate of Baseload and Intermittent Renewables. 5 The Baseload aggregate is composed of the output from coal, gas, nuclear, hydro, miscellaneous other minor sources, and the potential technologies, coal with carbon storage and sequestration (CCS) and gas-CCS. The Intermittent Renewables are wind and solar. The elasticities of substitution among these generation sources are presented in Table A2 in Appendix A and also noted in Figure 3; we assume a high degree of substitution among the baseload sources, and we follow other studies in assuming an elasticity of one between Baseload and Renewables.
This structure reflects the reality that average generation costs are different and yet different generation sources co-exist; there are considerations beyond average costs that determine the share of various sources in a highly regulated sector. We should make a technical note regarding this formulation: the imperfect substitution of kWh's means that the quantity 4 The production structure is given in Figure A1 and equations A4-A12 in Appendix A. 5 This aggregation of Baseload and Renewables is similar to that in C-GEM ). The output from each generation source is also expressed as a nest of CES functions; for nuclear, hydro, wind and solar there is a "Resource" input that represents non-produced inputs such as suitable rivers, and land with wind and sun. In any given period, these resources are given by an upward sloping supply curve to represent the short-run costs of developing such regions. This is similar to the representation in Vennemo et al. (2013) and Sue Wing et al. (2014) as discussed in the Appendix A.
The electricity generation and distribution system in China is dominated by a few large enterprises and tightly regulated by the NDRC. The prices are set by the NDRC after negotiation among the generators, distribution monopolies and major users. The dispatch order (which generation units are used at any moment) is determined for the most part by a "fairness" principle and a loose aim to promote renewables; it is not set according to least-cost dispatch, and it does not reflect bid prices for electricity in a competitive wholesale generation market. For a carbon price to work in the electric power sector, there must thus be reforms that would allow price signals to matter -that the net cost of fossil fuels reflects the carbon price, and that dispatch is sensitive to the prices asked by the generators. Our policy simulation here is predicated on the assumption that such reforms will take place.
There are 33 markets for the commodities; that is, there are 33 endogenously determined prices that equate supply with demand for the domestic commodities identified in the model. The total supply consists of domestically produced goods and imported varieties; these are assumed to be imperfect substitutes. There are three markets for the factors of production -land, capital and labor -and three prices to clear them.
The base case simulation is determined by the projections of the exogenous variables and the initial stocks of debt, capital and labor force. 6 Given the initial stocks, we solve for the three factor prices and the 33 commodity prices that clear the markets in the first period. This gives us all the quantities for the first period, including investment that augments the next period stock.    Electricity use is projected to grow at 4.4% during 2015-20 and at 3.0% during 2020-30.
The change in the structure of generation sources is given in Table 2 and graphed in Figure 3; we see that electricity use grows faster than coal due to the rise of renewables and nuclear energycoal-fired electricity grows at 2.7% during 2015-20, nuclear at 12%, wind at 11%. Solar starts from a very tiny base but rises rapidly, at 40% per year. In the 2020-30 period, renewable growth decelerates sharply but is still faster than coal, with the exception of hydro growth which falls to 0.4% per year.
As a result of the shift towards cleaner fuels driven by the assumed improvements in energy efficiency and the endogenous changes in prices in the base case, CO2 emissions,

Simulations of Carbon Pricing Scenarios
In our base case the CO2 intensity falls by 58.7% in 2030 relative to 2015. We consider carbon price trajectories that can reduce CO2 intensity by 60 or 65% by 2030, as in the NDC.
To that end, we consider flat rates that apply for all years, and prices that start low and rise every year so as to achieve the same cumulative emission reductions. We also consider two alternative approaches for recycling the carbon revenue to the economy: (i) recycle the carbon revenue by cutting existing taxes and (ii) rebating it lump-sum to households.
These policy scenarios are defined in Table 3. In the R1CUT scenario, we find through trial-and-error that a carbon price starting at 1.6 yuan/ton of CO2 in 2015, 10 and rising gradually to 26 yuan in 2030, hits the 60% reduction goal. 11 In this scenario, the carbon revenues collected 9 Our energy projection is based on IEA (2014). In the "Current Policies" scenario of IEA, China's primary energy use is projected to grow at 2.9% during 2012-20 and 1.9% for 2020-30, and project coal use to grow at 2. 1% (2012-20) and at 1.3% (2020-30). 10 All yuan units are measured in 2010 yuan, the base year of the model. 11 We should digress here to clarify what is meant by a statement like "a carbon price of ¥10 per ton in each year of the projection." In an economy with no TFP growth and little capital accumulation, absolute (and thus relative) commodity prices will differ little from the base year. In this case there is little ambiguity about what ¥10 in each year means. With rapid growth in output and incomes per capita, the price of commodities relative to the price of labor is -12-are offset by cuts in the VAT and capital income tax. In the R2CUT scenario, through trial-anderror we find that a carbon price starting at ¥10/ton of CO2 in 2015 and rising to ¥157 in 2030, hits the 65% reduction goal,. In the F2CUT scenario, we keep the carbon price flat at ¥82 level throughout the study horizon in order to achieve the same cumulative reductions in CO2 emissions relative to the base case over the 2015-2030 period (-9.8%) as the rising carbon pricing policy R2CUT. 12 To examine the effects of different revenue recycling options we have Policy R2lump which has a rising carbon price and the same cumulative CO2 reductions as R2cut, but here the carbon revenues are returned lump sum to households instead of cutting VATs. The carbon price paths under various scenarios are illustrated in Figure 4. Finally, in the "rising renewable subsidies" (RRsub) scenario we introduce subsidies for renewable power in the electricity sector so that the share of wind, solar and hydro reaches those specified in the IEA (2014). These subsidies are financed by a fossil fuel tax in proportion to caron content of fuels and it rises over time as the renewable share rises.

Figure 4. Carbon price (Yuan/tCO 2 )
falling rapidly, and relative prices are changing due to different capital-labor ratios (and possibly different industry TFP growth) . In our accounting system we measure things in terms of the GDP basket, and since relative prices change in our economy, "¥10 per ton" means a different basket of goods in each period. A flat ¥10/ton results in a changing ad valorem rate on the coal price. 12 We also considered an analogous policy F1CUT for comparison with R1CUT, but the results were not particularly useful and so we omit it here. -13-

High versus low carbon prices
The NDC announced by the government set a goal for reducing the CO2:GDP intensity in 2030 by 60-65% compared to 2005 levels. Here we compare the economic impacts of achieving the more stringent target, 65%, with the costs of reducing the intensity by only 60%, recalling that in the base case the reduction was 58.7%. The macroeconomic impacts of R2CUT versus R1CUT are given in Table 4. We present the results for the first year of the carbon pricing, 2015, and the final year. The impact of CO2 intensity over time is plotted in Figure 5 while the impact on GDP, energy use and CO2 emissions for the high carbon price case are plotted in Figure 6.
The impact on prices and output of each of the 33 industries in 2030 are plotted in Figures 7 and 8.

Figure 5. CO 2 intensity in R1CUT and R2CUT scenarios
We first describe the impact of the lower carbon price case (R1). The initial impact on GDP and energy use in 2015 of the small carbon price is correspondingly small, about 0.2% reduction in energy use. By 2030, when the carbon price reached 26¥/ton CO2, energy use has fallen 2.6%, and CO2 emissions has fallen 3.3%. The carbon price has discouraged the use of -15-fossil fuels, and raised the price of electricity, leading to a rise in the prices of energy intensive goods relative to the base case as shown in Figure 7 for 2030. Coal price at the mine-mouth rises by 6.2% while crude oil price rises by 1.2% inclusive of the tax, leading to a 1.4% rise in electricity prices.  Figure 6. Impacts of carbon pricing to achieve 65% reduction in CO 2 intensity (Scenario

Figure 7. Impacts of carbon pricing on commodity prices in 2030
-18-These changes in relative prices, and the reduction in capital stocks, lead a fall in industry output relative to the base case, as shown in Figure 8. By 2030, the consumption of coal falls in relative terms by 4.1%, oil by 0.7% and electricity by 1.5% (Table 4). The output of energy intensive goods, including primary metals and cement (building materials sector), fall more than the output of consumer goods and services industries. These price distortions reduce GDP, leading to lower investment every year. By 2030 GDP is lower y 0.11%, aggregate consumption is lower by 0.11%, investment by 0.13%, and total energy by 2.6%. As a result of these reductions, CO2 emissions fall by 3.3% in 2030 in Policy R1CUT.
The composition of electricity generation change at the same time that total electricity output falls by 1.5%; coal generation falls by 2.7% in 2030 while nuclear and hydro rise by about 1% (bottom section of Table 4). The price of intermittent renewables (wind and solar) are essentially unchanged and their output remains unchanged leading to a rise in their share contribution.
When we raise taxes much higher in the R2CUT scenario, the effects are amplified. The tax by 2030 is 157¥/ton, and the price of coal rises by 37% instead of 6%.The impact of lower investment and capital stocks due to these price shocks is a 0.23% reduction in 2020 GDP, and a 0.74% reduction in 2030 (compared to the 0.11% cut in GDP under the low tax policy). The tax on CO2 raises revenues equal to 5.8 percent of total revenues and enables a corresponding cut in the VAT and capital taxes. The lower GDP and higher energy prices lead to a 13% reduction in energy use compared to 2.6% under policy R1. Carbon emissions are 16% lower in order to reach the 65% intensity target. Within the electricity group, coal generation falls by 14% while nuclear and hydro rises by more than 5%. This is due to the rise in coal price leading to a rise in the relative price, P(coal electricity)/P(hydro), of 5.6%. Solar power output rises by only 0.3% given that we assumed that the elasticity between intermittent renewables and baseload is only 1, that is, most of the substitution is from coal to nuclear and hydro and not to intermittent sources. -19-

Figure 8. Impacts of carbon pricing on sectoral outputs in 2030
Flat versus rising carbon prices In this comparison we illustrate the benefits of phasing in a carbon price gradually rather than imposing a flat rate over time. We compare two paths of carbon prices that deliver the same cumulative carbon emission reduction by 2030 -R2CUT versus F2CUT (carbon prices are -20-plotted in Figure 4). The comparison of impacts is shown in Table 5 for the first and final years.
Note that the reduction in cumulative CO2 emissions over 2015-30 is the same, 9.8%, and that the first year the tax is 82¥/ton in F2CUT compared to 9.8¥ in the rising tax case.
With the higher tax rate in the first year of F2CUT, the larger distortion lead to a 0.26% fall in GDP relative to base case, compared to 0.03% in R2CUT. There is a bigger impact on investment that cumulates to a 0.56% reduction in the 2030 capital stock relative to base case, compared to a 0.51% relative reduction in the rising tax scenario.
The other variables such as energy consumption, electricity composition and CO2 emissions follow the same pattern -much bigger changes in year 1 and a more modest change in 2030, but cumulative losses that are bigger. The cumulative reduction in CO2 is the same but the reduction relative to base case in year 2030 is only 9.5% in the flat tax case compared to 16% in the rising tax case. In that year the changes relative to base case in coal generated electricity are -8.2% versus -14.2%, and in hydro they are +2.7% versus +5.2%.
To summarize, the flat tax path that delivers the same cumulative CO2 reduction as the rising tax trajectory imposes greater GDP losses. A richer model that takes explicit account of adjustment costs would generate even bigger differences between a gradually rising tax and a flat tax profile.

Options for recycling carbon tax revenue
To give an illustration of the importance of how recycling carbon tax revenues is done, we compare the previous R2cut scenario where we cut existing VAT and capital income tax to a case where the revenues are given back to households as a lump sum transfer. The results for this R2LUMP case are shown in Table 6 for 2030 together with R2cut. These two paths of rising carbon tax rates are set so that the cumulative change in CO2 emissions over 2015-2030 are the same -9.8%. It turns out that the relatively weaker economic performance in the lump-sum case generates lower emissions, and so the same cumulative reduction requires a lower carbon price than in R2CUT. This difference means that the change in carbon intensity in 2030 is not exactly equal to the 65.0% reduction in the R2cut case, but very close.
-21- The carbon tax rate in 2030 is ¥149 in the R2lump case but slightly higher at ¥157 in the tax cut case. The lack of an offsetting cut means that the VAT's in the lump sum case remain high and generate greater distortions. The net effect for many goods is a higher price in R2lump; for example, the price of primary metals rise by 5.7% versus 4.6% in R2CUT, for food manufacturing it is 2.1% versus 0.9%, and for construction it is 2.9% versus 1.8%. These greater price distortions contribute to a smaller GDP. The cut in capital taxes in R2CUT also allows -22-enterprises to retain more earnings and invest more, and thus the capital stock by 2030 falls relative to base case by only 0.51% compared to a 1.3% reduction in R2lump. As a result of this lower capital stock and higher value-added tax rates in the lump sum case, GDP in 2030 is 1.2% below base case, compared to only 0.74% lower in the tax cut case.
With lower aggregate output and lower carbon taxes prices in the lump sum case, the reduction in energy consumption is very similar, 13.2% versus 13.1%. The changes in CO2 emissions are corresponding similar, -16%, and thus the fall in CO2:GDP intensity relative to base case is slightly smaller in the R2lump case. Hydro generation is +4.7% in R2lump versus +5.2% in R2CUT. The change in intermittent renewables is different; in the lump-sum case the reduction in total energy demand leads to a 0.22% relative fall in wind power, while it rose by 0.12% in the tax cut case.
In sum, we have similar reductions in emissions at the cost of somewhat greater reduction in GDP growth (relative to base case) with the lump-sum transfer case than the tax-cut case. The slower GDP growth is compounded over time. Note that our gains in efficiency by reducing existing value-added taxes and capital taxes are calculated for a recursive model; in a dynamic model with foresight the impact of cuts in capital taxation would be even bigger, that is, the slowdown in GDP with the tax cuts would be even more modest than we have calculated. -23-

Subsidies for renewables and other low-carbon sources of electricity
In the policy RRsub we subsidize the cost of nuclear, hydro, wind, and solar using the revenues from a fossil fuel tax in proportion to their carbon contents. The subsidies are chosen so that the share of each of these sources in generation is equal to the projection by IEA (2014) under their "New Policies Scenario (NPS)". We focus on the kilowatt-hours generated rather than the capacity of the various sources which are also projected separately in IEA (2014). Table 7 gives the shares of electricity generation (by kWh) projected in IEA (2014) under their two scenarios -"Current Policies" and "New Policies" for two selected years. We first compute the change in shares for each year of the projection period between the two policies, e.g. under NPS the coal share in 2030 falls by 8.8 percentage points, and the wind share rises by 2.2 points. Although our base case shares not identical with the IEA's Current Policies projections in each year, they are set to be very close. We set the annual renewable targets in our policy scenarios as our base case shares plus the differences between the IEA scenarios, as illustrated in Table 7 for 2020 and 2030. In the RRsub policy we set subsidies separately for nuclear, hydro, wind, solar and other to hit the higher shares under the New Policies Scenario.
The results of using the carbon tax to subsidize renewables and other low-carbon energy sources are given in Table 8, together with those from R1cut and R2cut for comparison. In RRsub, the subsidies for renewables start at low rates but become quite substantial in the later years as given in the last column of Table 7. The share for wind has to rise from 5.7% in the base case to 7.9%, which requires a subsidy of 48%, while the share of hydro has to rise from 12.8% to 14.7%. As noted in Section 2, these renewables require natural resource inputs that have upward sloping supply curves and thus require higher returns to meet higher resource demands. The value of output of the electricity sector was ¥3240 billion in 2010, with a value added of 810 billion equaling 2.1% of GDP. These high subsidies thus require a sizable carbon tax to finance them. By 2030 the CO2 tax needed is ¥58/ton, which is in between the ¥26 carbon price -25-in R1cut and ¥157 in R2cut. These subsidies result in a 2.2% fall in the price of electricity compared to the base case and thus the 1.9% reduction in the consumption of electricity is only slightly bigger than the 1.5% reduction in the R1cut scenario with its much lower carbon price.
That is, the carbon prices that raised the price of electricity in R1cut encouraged conservation and generated a 3.3% cut in total CO2 emissions at a modest cost of reducing GDP by 0.11% in 2030. In this Renewable subsidy scenario, the subsidies discourage electricity saving and mutes the total CO2 impact of generating less coal electricity.
Total CO2 emissions in 2030 fall by 7.9% relative to base case, given the 58 yuan carbon price. The cumulative 2015-2030 relative cut in emissions of 2.1% is only slightly bigger than the 1.9% cut in R1cut, with the 26 yuan carbon price. The large carbon price and distortions of the electricity price generate a relative GDP loss of 0.53% in 2030 compared to only 0.11% loss in R1cut. The GDP loss in R2cut is bigger at 0.74% but that achieved a cumulative CO2 reduction of 9.8% compared to 2.1% for RRsub. That is, the ambitious renewable targets require large distortions in the later years, which generate CO2 reductions as a byproduct of a greater slowdown in GDP growth.
With the subsidies, coal generation in 2030 falls to 58.4% of total kWh compared to 64.6% in the base case and 63.8% in R1CUT, and hydro rises to a 14.5% share compared to 13.0% in R1CUT. These changes result in a larger reduction coal use in 2030, 10% versus 4.1% in R1cut, and CO2 emissions fall by 7.9% compared with 3.3% in R1CUT.

Sensitivity Analysis on Non-Fossil Fuels Growth Assumptions
In the base case the projection of resource inputs and capital inputs into the electricity generation functions is set according to the projections under the "Current Policies" scenario (CPS) for China in IEA (2014). As shown in Table 2, this scenario assumes a rapid growth of nuclear, wind and solar generation. The rapid expansion of relatively expensive generation resources (i.e., nuclear, wind, solar) implies higher subsidies and ultimately higher costs to the society.
-26- In order to get an idea of the economic impacts of this push away from coal and towards non-fossil fuels in the IEA Current Policies Scenario, we conducted an alternative set of simulations with a "no coal reduction" (NCR) scenario and a "CPS targets" scenario. In the NCR case we allow the coal share to fall from the first year 2010 to 2016 as in the base case, but then maintain close to that share out to 2030. The share of hydro in the base case is falling due to the projected difficulty in finding more hydro resources, and we maintain this in NCR. Note that even with the falling share the absolute production of hydro power is rising over time. The -27-remainder of the electricity demand is met by the other sources (gas, nuclear, wind, solar, others) that rise modestly in their share contribution to offset the falling hydro contribution.
In the CPS case, we do not change the resource and capital supplies exogenously as in the base case, but let them remain at the NCR paths and use a system of taxes and subsidies to hit the higher targets for renewables and nuclear. We impose taxes on coal and subsidies for the rest in order to attract more resources and capital into the non-fossils and discourage coal generation.
We require that the tax revenue exactly equal the subsidies so that there is no net (new) transfer to the government.
The comparison of these two scenarios is given in Table 9. In the NCR case, total coal consumption in 2020 is 4,147 million tons compared to 4,083 in the base case. The coal power is cheaper and thus the electricity demand in NCR is slightly higher, 6,949 billion kWh in 2020 compared to 6,886 in the base case. The impact of achieving the CPS targets via taxes and subsidies is substantial even in 2020; coal use is 3.5% lower than in NCR leading to a 2.1% reduction in total energy consumption. The price of average electricity is 1.3% higher leading to a 2.7% reduction in 2020.
By 2030 the impact is magnified by the cumulative GDP losses and reduction in capital stock. Coal use in 2030 is 5.5% lower in the CPS case while total energy consumption in 3.5% lower. Aggregate GDP is 0.29% lower due to these distortions with reductions in both consumption and investment. The generation mix shifts from 70.6% coal and 4.9% nuclear to 65.3% coal and 7.6% nuclear. The wind contribution rises from 4.7% to 5.7%.
While our specifications of the cost functions for the renewables and nuclear are simple, it allows a representation of the costs of finding suitable sites for such generation methods, and also possibly higher costs of transmission. The simulated costs are comparable to the renewable subsidies case shown in Table 8 given that the same cost function elasticities are used. There the cost of reducing CO2 emissions by 7.9% in 2030 is a 0.53% relative cut in GDP, here the 4.0% reduction in CO2 reduces GDP relative to base case by 0.29%. -28-

Conclusion
As part of the Paris Agreement, China has set for itself targets for reducing energy intensity by 60-65% in 2030 compared to 2005 levels. This study examines the economic consequences if the targets were met through a carbon pricing mechanism, where all fossil fuels are taxed in proportion to their carbon contents. The study uses a recursive dynamic computable general equilibrium model of the Chinese economy for the purpose this analysis.
The study finds that the carbon pricing policy would lead to a such a 65% reduction in intensity at a modest cumulative cost to GDP, which would be 0.7% lower than the base case by 2030, when revenues generated from the carbon pricing are recycled to the economy to cut existing value-added taxes. The goal of reducing the intensity by 60% is projected to cut GDP by only 0.1% relative to the base case by 2030. The carbon price not only reduces energy demand but also causes a shift away from more energy intensive industries to less energy intensive ones; it also causes a larger-scale deployment of renewable energy in electricity and heat generation.
A policy focused on promoting renewables in electricity generation using subsidies financed by a fossil fuel tax in proportion to their carbon contents would achieve the carbon reduction at a somewhat greater relative loss in GDP growth than a carbon pricing case. Putting more of the burden of adjustment on the power sector, when there are costs to a rapid ramp-up of hydro, nuclear, wind and solar power, results in a greater distortion of the price of electricity and a more difficult adjustment. A more careful study of the costs of resources going into these renewables is needed, including land, and suitable rivers.
As noted, this study does not attempt to model China's actual plan to achieve its NDC, which includes both market and non-market measures. Instead, it focuses on the cost of achieving the NDC through a market mechanism, carbon pricing. Comparison of the economic consequences of the carbon pricing mechanism considered here with that of government's actual plan of implementing the NDC is a natural extension of this study. Moreover, this study focuses only on CO2 emissions from fossil fuel combustion, and does not include other GHG emissions from fossil fuel combustion and also GHG emissions from industrial processes and land use change. This is a limitation of the study to be addressed in future work. -31-

Appendix A: Economic-Environmental Model of China
In this appendix we describe the model for China in some detail, beginning with the modeling of each of the main economic agents. Then in section A.2 we describe the data and parameters underlying the model. A previous version of this model of the Chinese economy is used in Nielsen and  and here we describe the updates to it. This is a multi-sector model of economic growth where the main drivers of growth are population, total factor productivity growth and changes in the quality of labor and capital. It has a dynamic recursive structure, i.e.
where investment is determined by fixed savings rate as in the Solow model. Consumption demand is driven by a translog household model that distinguishes demand by different demographic groups. The electricity sector is composed of 7 different generation technologies.

A.1 Structure of the Model
We discuss the five main actors in the economy in turn -producers, households, capital owners, government and foreigners. The electricity sector is described in greater detail explaining how the different generation technologies are allocated. For easy reference Table A1 lists variables which are referred to with some frequency. In general, a bar above a symbol indicates that it is a plan parameter or variable while a tilde indicates a market variable. Symbols without markings are total quantities or average prices. To reduce unnecessary notation, we include the time subscript, t, only when necessary to note the time dependence.

A.1.1. Production
Each of the 33 industries is assumed to produce its output using a constant returns to scale technology. Except for electricity, for each sector j the output at time t, jt QI , is expressed as: where KD j , LD j , TD j , and A ij are capital, labor, land, and intermediate inputs, respectively. 13 In sectors for which both plan and market allocation exists, output is made up of two components, 13 QI j denotes the quantity of industry j's output. This is to distinguish it from, QC j , the quantity of commodity j. Each industry may produce more than one commodity and each commodity may be produced by more than one industry. In the language of the input output tables, we make use of both the Use and Make (or Supply) matrices.
-32-the plan quota output ( j QI ) and the output sold on the market (~j QI ). The plan quota output is sold at the state-set price ( j PI ) while the output in excess of the quota is sold at the market price ( PI j ). The PI and QI names are chosen to reflect that these are domestic industry variables, as opposed to commodities (PC) or total supply (PS), the sum of domestic output and imports.
A more detailed discussion of how this plan-market formulation is different from standard market economy models is given in Garbaccio, Ho, and Jorgenson (1999). In summary, if the constraints are not binding, then the "two-tier plan/market" economy operates at the margin as a market economy with lump sum transfers between agents. The capital stock in each industry consist of two parts -the fixed capital, K j , that is inherited from the initial period, and the market portion, ~j KD , that is rented at the market rate. The before-tax return to the owners of fixed capital in sector j is: For each industry, given the capital stock K j and prices, the first order conditions from maximizing equation A2, subject to equation A1, determine the market and total input demands.
We represent the production structure with the cost dual, expressing the output price as a function of input prices and an index of technology. The 3 primary factors and 33 intermediate inputs for each industry are determined by a nested series of constant elasticity of substitution (CES) functions taken from the GTAP model (version 7). The nest structure is given in Figure A1 and applies to all industries except electricity which is treated separately below.
-33- For the top tier, the value equation and cost function are, respectively: where Mj  is the weight for all non-energy inputs into industry j, and 1 / QI jt  is the elasticity of substitution between the two inputs. jt g is the index of the level of technology where a rising value indicates positive TFP growth and falling output prices. We assume that the change follows an Coal Oil-mine Gas-mine Refining Electricity Gas This implies technical change that is rapid initially, but gradually declines toward zero.
The primal function corresponding to the above dual cost is: The input demands derived from the CES cost function are: The weights for the CES functions are explained in Rutherford (2003) and Klump, McAdam and Willman (2011); these are calibrated using the base year values: The corresponding value, price and input demand equations for the primary factor-energy basket (VE) and the value-added basket (VA) are: The parameters of the value-added-energy and value-added nodes are calibrated to base year values in the following way: The energy basket equations give the demands for the 6 types of energy: The non-energy basket is a Cobb-Douglas function and the corresponding equations are: We set the energy share  Ej to fall gradually over the next 40 years while the labor coefficient,  Lj , rises correspondingly. The composition of the aggregate energy input Ej (i.e. the  kj E coefficients) are also allowed to change over time.
The price to buyers of industry output includes the indirect tax on output, the externality ad-valorem tax, the externality unit tax: A carbon tax on coal, e.g., is represented by

Industries versus Commodities
The model distinguishes industries from commodities as in the official Use and Make input-output tables. Each industry may make a few commodities and each commodity may be made by a few industries; e.g. the Refining industry produces Refining commodity and Chemical commodity, and the Chemical commodity comes from Refining, Chemical, Primary Metal and other industries. The quantity of domestic commodity is denoted QC and its price PC; the sum of column i in the Make matrix gives the value of commodity i, and the sum of row j is the industry output value. The relation between commodity and industry output and prices are written as: m is the row share and c ji m is the column share.
A.1.2. The electricity sector In version 17 of the China Model we disaggregate the electricity sector into different generation technologies unlike the previous versions that represent the output of electricity using the common production function (A.4b) above. The production and input structure of this sector is illustrated in Figure A2; The superscript tt denotes that this is a price inclusive of indirect business taxes that are levied at the level of the generating sectors and transmission. This is explained in (A21b) below. The quantity and value equations are: At the Generation node we assume that this consists of Base load sources and Renewables (intermittent) in a way similar to the C-GEM model , this simplifies to: (1 ) Elec, Refine, Gas Intermed ..

-41-
The quantities of Base load output and Renewable output are: , the elasticity of substitution between "peak load" (which includes wind and solar) and "base load" sources is also 1.
For the intermittent renewable aggregate we only identify two types in this model: wind and solar (the others are part of the miscellaneous "other" in the base load tier). We assume that wind and solar are close, but not perfect, substitutes with an elasticity R E  =4. This is the elasticity chosen in Sue Wing et al. (2011, p 31). The equations for the renewable tier are: The price to the purchasers of such electricity is The input demand structure for coal generation is given in Figure A2; at the top tier coal power is produced by a low elasticity CES function of value-added-energy (VE) and non-energy intermediates (M). 16 Productivity growth in this sector is represented by the BL ct g term in the price function (A22). The VE bundle is a CES function of value-added (VA) and energy (E) with VE  =0.5, while the VA node has an elasticity VA  of 1.0 between capital and labor. The Energy aggregate is a function of coal and noncoal energy which is a small item that includes electricity and refined petroleum products (lubricants and vehicle fuels) as described in the data appendix. We set the elasticity between them to a low value ( E  -43-=0.2). The non-coal quantity is an aggregate of only electricity and refined oil since the other energy inputs (oil mining, gas) are zero.
The input structure for gas generation is similar to the one for coal for the top tiers for output price, price of VE, price of VA and price of M. The equations for the top tiers, in terms of the producer prices, are: For the energy tier, the equations for coal and noncoal (NC) inputs in sector c are:   . This model explicitly recognizes the small amount of energy used besides the main fuel source unlike the other models mentioned so far; in the case of coal, the non-coal inputs include refined petroleum, electricity and gas. The substitution between the main fuel and the small non-coal energy bundle is set at 0.25, this is similar in spirit to the value of energy-capital substitution used in Phoenix for electricity generation. The substitution among the components of noncoal energy is set at 0.5, the general elasticity for energy inputs in GTAP and EPPA.
The input structure for gas-fired power plants is similar to that for coal, except that in the bottom tier for energy, gas inputs (GS) are aggregated with non-gas (NG; electricity and refined petroleum). The value equations for the gas and gas_ccs nodes are: The values of the substitution elasticities follow that of coal and are given in Table A2.
We also project fossil fuel technologies that will be cost competitive in the future with a high carbon price: coal integrated gasification with carbon capture (IGCC) and natural gas with carbon capture (NGCC). The IGCC tier structure is given in Figure A2; the NGCC structure is identical. It is the same as the coal tier structure except that the energy node is replaced by a "fuel and sequestration" node. This structure follows Sue Wing et al. (2011, p33) where the fuel and sequestration technology in combined in a Leontief function, and the sequestration input is a fixed factor resource with an upward sloping supply curve.
The sequestration resource supply is given by: The modeling of nuclear power supply is always treated specially in the models cited given its unusual nature. While there is no obvious constraint like the availability of rivers for hydro power, it is recognized that actual construction of nuclear plants has been difficult with substantial opposition by those worried about safety.  uses an upward sloping supply function due to "the political suitability of different locations." Sue Wing et al. (2014) also use a "fixed factor" in their specification of nuclear power, but justify it by saying that a "supply curve is used to parameterize the mining and milling of resources to produce the fuel rods." EPPA-4 also has a fixed factor in nuclear generation but interprets it as a stock of knowledge that builds over time with cumulative output.
We follow the logic of Vennemo at al., and the Phoenix model, and require a non-reproducible resource input for nuclear power. In the second tier (see Figure A2), the VR bundle is an aggregate of resource (R NUCL ) and value-added-energy (VE) with an elasticity of VR  =0.4, following the EPPA-4 -48-value (Phoenix uses a top elasticity of 0.5). We follow Vennemo et al. in using a nuclear resource supply elasticity of 2.5.
The specification of the hydro power production function is another complicated matter. Rivers suitable for hydro power is an obvious input, and one might think of the stock of such water resources and the cost of using the marginal river. If the next unused water source requires a more costly structure (or implementation costs including relocation of people) than existing dams, then one may represent this as a production function with a lower TFP factor, or as a paying more for a fixed quantity of effective water resource R. For simplicity we have chosen the latter approach and set up the hydro function like that of nuclear, with a rising supply curve for the water resource input, R HYDR . In the same spirit, our wind and solar output functions follow those of nuclear and hydro with specific Resource inputs. This is illustrated in Figure A2(b). The parameterization of the resource input shares follows that given in EPPA 4 (Paltsev et al. 2005 Table 11); the resource share for wind and solar is set at 0.05 and we assume the resource share for hydro is also 0.05 of total gross output.
The cost functions for the renewables -nuclear, hydro, wind, solar and "other" -are the same The following equations are for the sub-aggregates -output price , (1 )     the wind resource grows in a logistic manner, while in Phoenix the supply curve is a constant elasticity one. EPPA-4 introduces a "Fixed Factor" for wind and solar and interprets it as knowledge that grows with cumulated output, and sets an elasticity between the Fixed Factor and value-added (

VR BLb
 ) at 0.6. Paltsev et al. (2005) states that the "Choice of the substitution elasticity creates an implicit supply elasticity of wind in terms of the share of electricity supplied by the technology." We interpret of the resource variable for hydro, wind and solar as a limited supply of sites suitable for such technologies, and set the Nuclear, Hydro, Wind and SolarPV resources to grow in the base case at the rates projected in IEA (2014). In any period t, in the policy case, the supply curve for Wind or SolarPV resources is given by: where r t R is the projected base case resource availability. This means that opening an additional river, or wind farm, or solar farm, will require paying a price for the resource that is higher than the base case price. We follow  in using an elasticity of 2.5, which means that this price is only slightly higher than base case resource price. It may be reasonable to not assume an identical function for all possible values of the base case resources availability, that is, when there is a high utilization of these water or wind sites, the marginal cost might increase substantially. That is, imposing a lower elasticity in the future years when there is a high penetration of such renewable sources.
The final element of the electricity sector is transmission and distribution. There is little data on this; even the 500-sector IO table for the U.S. has just one sector for total electric utilities. In the data appendix we describe how the data for Electric Utilities is separated into two sets for Generation and Transmission. The resulting input vector for Transmission includes almost all 33 commodities identified in the model. The tier structure for it is given in Figure A2(b) and is similar to the general structure for all other regular sectors given in Figure A1. The equations for the top tier of the Transmission sector, in terms of the producer price, are:   One may express the output index as a product of a quality index and the total kWh:

Note on aggregation and units of measurement
, The quality, or composition, index EL l  represents the impact of changes in shares of components that have different relative prices. For example, the price per kWh of coal is much lower than that of solar and if the share of solar kWh rises, then the quality index rises. This is analogous to the relation between effective labor input, quality of labor and hours worked; the index of labor input is the product of labor quality and total hours, and a rising labor quality indicates that hours from more highly paid workers is rising as a share of total hours worked by all workers. One may think that the term "quality" of electricity may be misleading since kilowatt hours as perfectly homogenous and substitutable, it however, represents an economically meaningful distinction between kWh and value of kWh. The electrons that are identical from the user point of view are distinguished by the method of production -clean versus polluting, steady versus intermittent, near versus far. The EL l  index represents changes in the costs of production as the composition of methods change; EG t Q is an index of the (marketed) economic resources that went into producing , kWh TOT t Q kWh of electricity. The non-marketed resources such as clean air, or clean water, are accounted for separately.

A.1.3. Households
Private consumption in this model is driven by an aggregate demand function that is derived by aggregating over different household types. Each household derives utility from the consumption of commodities, is assumed to supply labor inelastically, and owns a share of the capital stock. It also receives income transfers from the government and foreigners, and receives interest on its holdings of public debt. Aggregate private income is the sum over all households, and this income, after taxes and the payment of various non-tax fees (FEE), is written as: YL denotes aggregate labor income from supplying LS units of effective labor, less income taxes: The relationship between labor demand and supply is given in equation A63 below. Aggregate supply LS is a function of the working age population, average annual hours, and an index of labor quality:

DIV denotes dividends from the households' share of capital income and is explained below in A75. G_I and G_transfer represent interest and transfers from the government, and
R_transfer is transfers from the rest-of-the-world. LUMP T is the lump sum tax that is used in policy simulations.
Household income is allocated between consumption (VCC t ) and savings. In this model we use a simple Solow growth model formulation with an exogenous savings rate ( s t ) to determine private savings ( S t p ): Total consumption expenditures are allocated to the 33 commodities identified in the model. We do this with a demand function estimated over household consumption survey data.
This consumption data is at purchaser's prices and follows the expenditure classification; these have to be linked later to the IO classifications and the factory-gate prices of the IO system. We arrange the demand system in a tier structure shown in Table 1. At the top tier total expenditures is allocated to Food, Consumer Goods, Housing and Services. In the sub-tiers these four bundles are allocated to 27 items.
-54- bundles. Each household type has its own distinct utility function and A k is a vector of demographic dummy -55-variables to indicate the size of the household, the presence of children, the age of the head, and the region.
The budget constraint for household k is: denote the share of expenditure allocated to bundle i. Applying Roy's Identity we get the demand share vector: The aggregate demand is obtained by summing over all household types. Let n k be the number of households of type k; the aggregate share vector is then: The above equations (A52) and (A53) are estimated simultaneously, with (A52) estimated over one year of cross-sectional consumer expenditure data, and (A53) estimated using time series national prices and aggregate consumption expenditures.
To use the estimated equation (A53) in the model that include projections into the future we make some modifications. Firstly, the consumer survey data does not include some items that are in the National Accounts such as imputed rentals for owner-occupied housing and FISIM. We make some adjustments to the p  's to scale the shares to match the consumption in the Input-Output table for our base year 2010. We project the distribution and demographic terms to account for the aging impact and thus re-write the share demand system as: The prices of the consumption commodities are linked to the prices of the IO commodities via the same share matrix: The total value of consumption of commodity i is the sum of the urban and rural components: The value of national consumption in equation (A48) is the sum over all the commodities:

A.1.4. Government and Taxes
In the model, the government has two major roles. First, it sets plan prices and output quotas and allocates investment funds. Second, it imposes taxes, purchases commodities, and redistributes resources. Public revenue comes from direct taxes on capital, labor, value-added taxes, indirect taxes on output, tariffs on imports, the externality tax, and other non-tax receipts: where D j is the depreciation allowance and X i and M i are the exports and imports of good i.
Externality taxes, such as those on SO 2 or CO 2 , may be on the value of output, or on the quantities.
We allow for both, the total revenue from the externality tax on output is: In one application of the model described in Ho and Nielsen (2007, Chapter 10), the externality tax rate is set proportional to the marginal air pollution damages (MD O ) from output j: When we consider a tax on fossil fuels based on the carbon content, the externality tax per unit of fuel j is: where j c is the carbon content per unit of fuel of type j.
The nontax payments to the government are set as a fixed share of household income: Total government expenditure is the sum of commodity purchases and other payments: Government purchases of specific commodities are allocated as shares of the total value of government expenditures, VGG. For good i: We construct a price index for government purchases as log log The real quantity of government purchases is then: Transfers are set equal to a fixed rate of the population multiplied by the wage rate: The difference between revenue and expenditure is the deficit, G , which is covered by increases in the public debt, both domestic ( B ) and foreign ( B G* ): The deficit and interest payments are set exogenously and equation A70 is satisfied by making the level of total nominal government expenditure on goods, VGG, endogenous in the base case. In simulating policy cases we would often set the real government expenditures in the policy case equal to those in the base case. In this counterfactual we would use some endogenous tax variable to meet equation A70.

A.1.5. Capital, Investment, and the Financial System
We model the structure of investment in a fairly simple manner. In the Chinese economy, some state-owned enterprises receive investment funds directly from the state budget and are allocated credit on favorable terms through the state-owned banking system. Non-state enterprises get a negligible share of state investment funds and must borrow at competitive interest rates.
There is also a small but growing stock market that provides an alternative channel for private savings. We abstract from these features and define the capital stock in each sector j as the sum of two parts, which we call plan and market capital: The plan portion evolves with plan investment and depreciation: In three sectors, agriculture, crude petroleum and gas mining, "land" is a factor of production. We have assumed that agricultural land and oil fields are supplied inelastically, abstracting from the complex property rights issues regarding land in China. After taxes, income derived from plan capital, market capital, and land is either kept as retained earnings by the enterprises, distributed as dividends, or paid to foreign owners: -62-

A.1.6. The Foreign Sector
Trade flows are modeled using the method followed in most single-country models.
Imports are considered to be imperfect substitutes for domestic commodities and exports face a downward sloping demand curve. We write the total domestic supply of commodity i as a CES function of the domestic ( i DC ) and imported good ( M i ): where DC is the quantity of domestically produced goods that are sold domestically. The elasticity . The cost dual corresponding to the above primal function is: and the value of total domestic supply is: The purchaser's price for domestic goods, i PD , is related to the commodity supply price i PC and is discussed in the export section below. PSi is the price of the basket of commodity i to domestic purchasers. The price of imports to buyers is the foreign price plus tariffs (less export subsidies), multiplied by a world relative price, e: Domestically produced commodities (QC) are allocated to the domestic market and exports according to a constant elasticity of transformation (CET) function: The ratio of exports to domestically sold goods depends on the domestic price (PD) relative to world prices adjusted for export subsidies ( s it e ): -63- The value identity is: The weights and constant terms are set using base year values: , , , The share parameters x it  are projected exogenously to take into account the rising role of exports during 1980-2010 and a falling role in the future. The price PC is given in equation (A14) above, and is also an implicit dual function of (A85), PC=f(PX,PD).
The current account balance is equal to exports minus imports (valued at world prices before tariffs), less net factor payments, plus transfers: Like the government deficits, the current account balances are set exogenously and accumulate into stocks of net foreign debt, both private ( B t * ) and public ( B t G* ):

A.1.7. Markets
The economy is in equilibrium in period t when the market prices clear the markets for the 33 commodities and the three factors. The supply of domestically produced commodity i must satisfy the total of intermediate and final demands: For the labor market, we assume that labor is perfectly mobile across sectors so there is one average market wage which balances supply and demand. As is standard in models of this  In this model without foresight, investment equals savings. There is no market where the supply of savings is equated to the demand for investment. The sum of savings by households, businesses (as retained earnings), and the government is equal to the total value of investment plus the budget deficit and net foreign investment: The budget deficit and current account balance are fixed exogenously in each period. The world relative price (e) adjusts to hold the current account balance at its exogenously determined level.
The model is a constant returns-to-scale model and is homogenous in prices, that is, doubling all prices leaves the economy unchanged. We are free to choose a price normalization.

A.1.8 Welfare Other accounting identities
The household welfare function ( The welfare function depends on the average level of consumption as well as inequality of consumption (efficiency and equity). Here we compute only the average levels to give the efficiency measure which is given by: V is the household utility in (A50), and 0 ( , ) k m p A is the household equivalent to the reference household which is aged 18-34, male, elementary school, two members and in the East. The equivalence scale is explained in Jorgenson and Slesnick (1987) and is given by: The money measure of welfare is given by a social expenditure function (Jorgenson and Slesnick 1987, eq. 5.15): The money measure of the change in welfare due to a policy (from 0 W to 1 W ) is a function of the policy case measured at base case prices ( To construct real, constant yuan, GDP we need to first define real consumption, investment, etc. These are expressed as the divisia aggregate of the 33 commodities that make up each component, for example, real personal consumption expenditures is: Real GDP is then a divisia index of these components: , , , ; ,...) div div div div div rGDP divisia CC II GG X PCC M  

A.1.9 Energy, emissions and environmental accounting
To account for atmospheric environmental damages we consider a range of criteria pollutants: particulate matter (PM 25 and PM 10 ), sulphur dioxide, nitrogen oxides, VOCs, ammonia. We also account for greenhouse gas emissions, in particular carbon dioxide. The PM concentration is due to primary PM emissions as well as secondary particles such as sulfates and nitrates which are formed from sulfur dioxide and NOx respectively. The emissions inventory is described in Clearer Skies, Chapters 4-6. To illustrate the calculations we describe here a simplified account of energy flows and primary PM, SO 2 and NOx emissions.
We begin by describing the energy variables. Very often a simple indicator of total primary energy production and consumption is produced by summing the energy equivalents of the fossil fuels and primary electricity and heat. This may not be a very useful indicator given that a joule of energy from burning coal is very different in the ease of use from a joule from gasoline or a joule of electricity; a difference that is reflected in the prices per joule. Nevertheless, for comparison with well-known series we compute the standard coal equivalent (sce) of these primary sources of energy.
First, recall that we distinguish between industry output (QI) and commodity output (QC). QC ft is the constant yuan quantity of commodity produced (billions of 2010 yuan). QP f , the total quantity of coal, crude, and gas produced (whether combusted or not) in year t is given by the commodity output (QC) multiplied by the fuel conversion coefficient, We compute energy consumption in two ways. The first way simply uses the total output of fuels (production based account); the second way sums over the industry consumption of energy that is calibrated to the official estimates in the base year (consumption based account). First, E PROD , the total sce of energy produced domestically, is: where f e is the energy content of a unit of fuel f (e.g. tons of sce per ton of oil) and the PRI superscript denotes primary electricity from renewables and nuclear.( In this calculation we ignore the tiny amount of heat from natural sources.) We set α PRIelec , the share of electricity produced from primary sources, exogenously by considering the projected generation of renewables and nuclear power. Then QP PRIelec , the quantity of primary electricity produced from renewables and nuclear, is:

QP QP
  E EXP , the total sce of energy exported, on net, is: where X f is the value of exports of fuel f (in billion yuan) and M f is the value of imports of fuel f (in billion yuan). Exports of electricity are not counted in this measure since it is a secondary energy, the pollution due to the generation of electricity for exports is located in the country and they are not exported.
E CONS , the total energy consumed in China (in tons sce) is then given by production less net exports, less changes in inventory (  Second, in consumption-based accounting, we also calculate national energy consumption by adding over each industry, using industry specific information about the consumption of coal, coke, liquid fuels, etc. We first define consumption coefficients (  Table 7-9 "Consumption of Energy by Sector") and dividing by the value of energy purchases given in the Input-Output table.
To disambiguate, the fuel conversion coefficient ( f mean  ), presented in equation (A101), is computed using the production data at the aggregate level: the total quantity of the commodity output divided by the total value of the commodity output. In contrast, the consumption coefficient ( That is, the value of refined liquids is the product of: 1) the share of liquids in total refined petroleum products; 2) the share of non-coal products in the Refining & Coal Products sector that industry j purchases;  We now move on from calculating energy consumption to calculating energy combustion. The above consumption coefficients refer to the purchases of the different fuels. Some of these oil and gas inputs are not combusted but converted to other products such as fertilizer or bitumen. In the Refining sector part of the crude input is combusted but most are converted to liquid fuels or other petroleum products; the uncombusted portion is represented by the "refining loss" coefficient, In the Gas Products (Utilities) industry, gas is purchased from the Natural Gas Mining sector and sold to consumers, that is, there is assumed to be no combustion in this industry. In the Chemicals sector, raw gas is purchased from the Gas Mining sector and part of it is converted to plastics and other products. The   where C elect , C coal , C refining , C oil , C gas , and C gasprod denote the constant yuan value of Consumption by households of those fuels (in billion yuan). I f is the value (in billion yuan) of purchases of fuel f by the Investor (these are essentially business inventories in the Investment column of the input-output accounts).
The un-combusted portions in this version are the other petroleum products ("nonliqref") and part of the gas use by the Chemicals industry. We denote the un-combusted fuel use by FU: [In this version we have not separated out the combusted portion of "other petroleum products (nonliqref)". A more detailed accounting would have refined products divided to "refined liquids", "other combustible refined products", and "noncombustible refined products". The equations would be:

Emissions
The national emissions of carbon dioxide may be computed from the production accounts by adding over the emissions from all fossil fuels f. This is given by the quantity of fuel consumed ( The elasticities are summarized in Table A2 A

.4 Parameters, exogenous variables and data sources
The key input into the model is the Social Accounting Matrix (SAM) for 2010. This traces the flow of commodities and payments among the producers, household, government and rest of the world. The SAM is assembled from the 2007 benchmark input output table. 22 A summary of this SAM is given in Figure A4, the actual matrix used is disaggregated to the 33 sectors and commodities. From this we derive the labor and capital incomes, the tax revenues for each type of tax, the expenditures on specific commodities by the household, government and foreign sectors, and government payments of all types in equation A76.
These payments are combined with employment and capital input data to give the compensation rates for labor and capital for each sector. The estimates for employment and capital stocks by sector are taken from a productivity study of China (Cao, Ho, Jorgenson, Ren, Sun and Yue 2009) that supplements the official data with labor force surveys. The various tax and subsidy rates are not statutory rates but are implied average rates derived by dividing revenues by the related denominator -value of industy output, capital income, total value added, and imports.
The exogenous variables in the model include total population, working age population, saving rates, dividend payout rates, government taxes and deficits, world prices for traded goods, current account deficits, rate of productivity growth, rate of improvement in capital and labor quality, and work force participation. These variables may, of course, be endogenous (i.e. they interact among each other) but we ignore this and specify them independently. Our assumptions for these exogenous drivers are summarized in Table A3.
The assumption that affects the growth rate the most is the household savings rate, s t . Our assumption is to have s t beginning at the observed 41.2% for 2010 and gradually falling to 22% in 2020 and 15% in 2050. National private savings is household savings plus the retained earnings of enterprises. 21 电力工业统计资料汇编. 22 The 2010 input-output table is given in NBS (2014). The benchmark IO table for 2007 is derived from detailed enterprise data, the 2010 IO table is extrapolated by the NBS using simpler aggregated data. -80- The share of retained earnings is assumed to fall, and dividend payouts to rise to reflect the diminishing role of state enterprises in the economy. The dividend rate, i.e. the share not used for retained earnings, was 38.9% in 2010 and we project it to rise to 58% by 2020. It should be pointed out that national savings and investment in the Chinese data includes capital such as roads and other public infrastructure, items that are excluded from the "gross fixed private investment" item in most other countries National Accounts.
In the labor supply expression eqn. (A47) we have the product of the working-age population, annual average hours and quality. In  we discussed various population growth scenarios including the different two-child policies. Projections by age groups are taken from projections made by the Population Division of the Department of Economic and Social Affairs of the United Nations Secretariat 23 and in the central population scenario we adjust them to incorporate a looser child policy. The results are plotted in Figure A5.
The composition of the work force changes over time with a bigger portion of educated workers, bigger or smaller portion of more experienced workers, and an older average age. This quality of labor input index, L t q , in estimated in Cao et al. (2009) to have grown at 0.9% per year for the period 1983-2000.
Given the expectation of continued higher educational attainment in the future we assume that China's aggregate labor quality continue to rise, but at a diminishing rate. By 2040 the quality index is assumed to grow at only 0.2% per year. For comparison, the U.S. labor quality growth peaked at 0.5% during the 1960s, and fell to 0.3% per year during 1995-2000 (Jorgenson, Ho and Stiroh, 2005, Table 6.5).
Total labor hours depend also on the participation rate and annual working hours. There is no comprehensive data on the number of hours worked and based on comparisons to other countries we project it to rise due to improvements in the functioning of the labor market --lower underemployment, seasonal unemployment and other labor market frictions. We assume that hours worked per capita rises at 0.2% per year initially but slowing down over time. The results are plotted in Figure A6.
We allow for improvements in future capital "quality," or composition, as represented by the I t  coefficient in (A73). Cao et al. (2009) note how the composition of the capital stock in China has shifted towards assets with shorter life, i.e. towards a smaller share of structures and a larger share of equipment such as computers. They explain how assets which have shorter useful lives generate higher annual capital services per dollar of capital stock, and hence is of a higher quality in the terminology of Jorgenson, Ho and Stiroh (2005). While the ratio of equipment to structures has moved in different directions over the past 30 years, we believe it will return to a more typical development trend of rising equipment ratios. We project that capital quality rises by 1.5% per year initially, then gradually decelerating. For land, the supply of land for agriculture, oil mining and gas mining is simply set fixed for all periods equal to the base year value.
Tax rates are set equal to those for 2010 derived from the SAM. These are summarized in Table   A4. For the government deficit, G  , we set it at the base year 1.69% of GDP initially, declining steadily towards zero in the long run. These deficits are cumulated into the stocks of domestic and foreign debt,  (Table A4).
The current account balance was in a huge surplus in the mid-2000s but has since declined. There is no consensus about the future evolution of this variable, for simplicity, after setting it as a share of GDP at the observed sample period values, we set it to decline rapidly to zero. This CA t deficit is also the assumed rate of borrowing from the world. Import prices, PM i* , are assumed fixed at the base year value for every period with one important exception. World oil price forecasts are taken from the U.S. Energy Information Administration and shown in Figure A7. 24 The model also requires projections of the export share; while this has been rising rapidly in the past, it fell with during the 2008 Global Financial Crisis and never recovered. Given the aim to rebalance the economy away from exports and investment towards consumption, we simply project a constant value for the share parameter.
The base year data for 2010 was constructed in 2013 since then, the macro variables for 2011-13 is now available; these include the GDP, investment and current account surplus. The current account surplus has fallen, and the unusually high share of investment in GDP has risen even more after 2010. We take these into account in setting the savings rate and current account balance as share of GDP for these years.

Parameters
The rate of productivity growth is another factor that has a large effect on the base case growth rate of the economy but has little impact on the difference between cases. Total Factor Productivity growth at the industry level in the 1982-2000 period show a very wide range of performance as estimated by Cao et al. (2009), ranging from -10% to 5% per year. The Domar-weighted productivity growth for all industries was 2.7% for 1982-2000. To keep the base case as simple as possible we ignore this wide range of observed TFP growth, and in our projections of sector productivity terms in eqn. (A3) we initially set all the j  's to the same value, 0.018. These are then adjusted to match actual GDP growth rates in the initial years for which we have actual data.
The value share parameters of the production functions ( Kj The import and export elasticities are set to the values in GTAP v4. The base share of exports and imports are taken from the SAM. -85-   To provide a comparison to these estimates of average feed-in tariffs, we also report in Wind power is expected to be expanded rapidly under the current government plans and we provide more information in Table B2(c). The first two columns give the levelized costs estimated in IEA (2010) for various wind turbine capacities under the 5% discount option. These range from 0.32 to 0.55 yuan/kWh. The last two columns give the size distribution of the wind turbines in 2012. 31 The most common type is the 50MW turbine. Using the IEA estimated costs, the average wind cost for that distribution of sizes is 0.42 yuan/kWh. This is also substantially lower than the average feed-in tariffs which depend on the region where the turbine is located.
That is, the tariff takes into consideration the wind conditions and not just the cost of operating the turbine.
For technologies that are not currently used but might be considered in the future, especially if there is a carbon emission policy, we report the estimated costs of generation with carbon capture and sequestration (CCS) in part (d) of Table B2. This gives the estimated cost of a plant with CCS relative to a reference plant without CCS. The IEA (2010) projections give estimated costs for 2015, 2020 and 2030 and we report only the 2030 estimates to give readers an idea of the magnitudes involved. For coal with Chemical absorption with a 85% CO2 capture rate, the cost ratio is projected to be 1.49, while the gas combined cycle plant with CCS has a lower cost ratio of 1.23.
With these estimates of the average feed-in tariffs of the various technologies in Table   B2, and the output data in Table B1, we estimate the revenues. These output values are computed in order to be reconciled with the values in the Social Accounting Matrix. The values derived from the prices in Table B2 are given in the first column of Table B3. The value of coal power is 78.9% of the total 1,595 billion yuan. Since the price of hydro power is the lowest, the value of hydropower is only 12.4% of total output even though it is 16.2% of the TWh. The other sources of power were less than 3% of the total in 2010.
Turning back to Table B1, we see that the gross output of Electricity and Heat in the input-output table is 3,236 billion. If we take the total of 1595 billion as the generators' revenues, then the residual for transmission would be 1641 billion yuan (Table B3b), which is larger than the generating sector. As we noted above, in 2010, the average feed-in tariff was 0.38 yuan/kWh 31 This data is given in the 2014 Report on Wind Power Market.
-96-and the average transmission price was 0.20 giving a total end-user price of 0.58 yuan/kWh. That is, the average price for the transmission is quite a bit smaller than the average price paid to the generators.
The national accounts are most likely more comprehensive and take into account aspects of the electric power sector that is not included in these prices. To reconcile these two sets of data, we take a simple approach and scale the feed-in tariffs in Table B2 upwards so that the ratio of generator revenue to total generator plus transmission revenue is 0.38:0.58 as shown in the last two columns of Table B3b. This rescaled total generator revenue of 2121 billion yuan is then applied to the individual technologies and the revised output values are given in the third column of Table B3.

Input-output of electricity sector
With these outputs and prices of each generation technology, we can now disaggregate the Electricity & Heat sector of the Use matrix into the 7 technologies and electric power distribution. The Use column gives the inputs into this sector and is given in the first column of Table B4. We distribute this column to the 7 generation types and distribution in two steps; first divide the Use column into a Generation column and a Distribution column; then divide the Generation column into 7 technologies. We proceeded in the following manner.
-97-First, set the target for total generation to 2,121 billion and for distribution to 1,115 as calculated in Table B3b  -99-electricity to fossil fuels versus non-fuel goods. That is, we assume that these margins only apply to inputs 1 through 21 out of the 33 commodities in the Model.
Up to this point we have allocated the fuel inputs (commodity numbers 2,3,4), electricity (22) and a portion of the trade (27) and transportation (25) margins to Generation and to Distribution. Of the 2,121 billion yuan target total for Generation, 1,080 remains unallocated, and for Distribution 957 remain unallocated. Each of the remaining intermediate inputs (1,5,6,…33), and each of the value added items, is allocated to Generation and to Distribution in proportion to these unallocated amounts. In this way, the sum of all inputs for Generation equals the target 2,121 billion yuan. The results of this exercise are given in the second and third columns in Table B4.
The second step disaggregates the total Generation column to the 7 technologies, where the output targets are given in Table B3 in the "rescaled values" column. First, we allocate the entire coal mining input of Generation to "Coal generation", oil mining to "Others", and gas mining and gas utilities to "Gas generation". See Table B5. Then for each technology we compute the unallocated total as the output target minus these allocated fuels. Each of the nonfuel rows in the Generation column (including the value added rows) is then allocated to the 7 technologies in proportion to these unallocated totals.
The results of this disaggregation are given in Table B5. Coal generation is 79% of total generation output but since fuel input is such a big factor here, its share of capital value-added in total Generation is only 72%. For hydro, the situation is reversed, it has 12.4% of Generation output value but 19.0% of value added.
The above simple disaggregation procedure preserves the Use column of the electricity & heat sector, that is, the sum of generation and distribution output is the electricity & heat output, and for each commodity input, the sum across the 7 generation technologies and distribution is the value in that row of the Use column. This means that the rest of the SAM for the other sectors is undisturbed.
-100- In order to compare this disaggregation with the other sources of data, we compute the costs shares implied by this input-output data for coal and hydro (the predominant sources of electricity in 2010), and report them in Table B6. The cost shares are given for fuels, other nonfuel intermediate inputs, labor and capital costs in order to compare them with the categories in the IEA (2010) levelized cost calculations. In the IO tables, the finance row contains a large entry which we interpret to include interest margins on loans. When we compute the "capital cost share," we include both the value added row and the finance row of the Use column in order to be closer to the accounting concepts in IEA (2010). Similarly, the fuel cost share is the sum of the coal row (in factory gate prices) and the trade and transportation margins in order to be consistent with the accounting concepts. The IEA estimated costs for selected technologies were given in Table B2c, and we also report in Table B6 the cost shares averaged over all the coal technologies given in IEA (2010). The average over all the hydro technologies is also given. We can see that these cost shares are very different; for coal generation, the IEA (2010) projections allocates 76.8% to fuel and 17.7% to capital compared with the input-output table allocation of 44.6% for fuel, 14.8% to capital and 35.3% to non-fuel intermediates. These differences may be due to: (a) distinction between average and marginal fuel costs (some coal input is allocated at the controlled price which is much lower than the market price, the IEA calculations are based on the market price); (b) differences in the cost of capital assumptions (the IEA likely assumes a long run cost of capital in a deregulated capital market, the actual profits in 2010 that is embodied in the National Accounts may be much lower than such a long-run rate of return given that the feed-in tariffs are controlled by the government); -102-(c) size and vintage of the coal units (the IEA estimates are for new large plants to be constructed now whereas the current stock includes many older and smaller units, the maintenance costs may be quite different for the older units, the old plants may be allowed less generous tariffs); (d) different coverage of enterprises included in the electricity and heat sector (the National Accounts may include enterprises that are not simple generators or distributors).
Given these distinctions we regard the 2010 SAM values as the correct ones to represent the base year cost structures, and use the IEA (2010) cost shares to represent future cost structures. That is, in our projections of the share parameter in the cost functions (or production functions), we will begin with the estimates in Table B4 and B5 and then gradually trend them towards the cost shares in the IEA rows of Table B6.

B.2 Projections in the base case
The exogenous drivers of economic growth are this model are the demographics (population and working-age population), saving rates, and total factor productivity growth as described in Appendix A. The future structure of the economy is affected by projected consumer preferences, biases in technical change and world commodity prices; production parameters are based on US input-output tables and consumption parameters are estimated using consumer expenditure surveys. This appendix focuses on the projection of the electricity sector which is treated in a distinct manner compared to the other production industries.
As explained in the Appendix A, we regard the investment in the electricity sector as being externally determined by the Plan. There is the detailed 5-year plans and also long-term targets; in the case of electricity generation there are targets for nuclear capacity, wind and solar capacities. IEA (2014) contains projections of China's energy use, including projections of various electricity generation technologies, based on their reading of these plans and their projections of economic growth.
The IEA (2014 , Table 1.1) uses the GDP projections from the IMF and these are given in Table B7. These are very close to our GDP projections (somewhat slower than , with about 7% in the current decade and slowing to 5% during 2020-30. -103-IEA (2014) projects three energy scenarios: "Current Policies" scenario is based on "policies that were enacted as of mid-2014", "New Policies" is based on "the continuation of existing policies and measures as well as the implementation (albeit cautiously) of policy proposals, even if they are yet to be formally adopted", while the "450 Scenario" is intended to illustrate "what it would take to achieve an energy trajectory consistent with limiting the longterm increase in average global temperature to 2°C." 33 We take the "Current Policies" projections of the generation capacities and power output to guide our base case projection of the composition of electricity output. The IEA (2014) report also gives the estimates of actual capacities in 2012. The Chinese data reported in the China Energy Yearbook (and China Statistical Yearbook) that are given in Table B1 are slightly different in that they only include sources that are connected to the grid. We first rebase the 2012 IEA figures to these official data for 2012, and adjust their projections. These rebased projections out to 2040 are plotted in Figure B1 and summarized in Table B7.
Total electricity output is projected to rise from 4,986 TWh in 2012 to 6,930 in 2020, and to 10,333 in 2030. These translate to growth rates of 5.2% per year (2012-20), 3.2% (2020-30) and 1.9% (2030-40). By 2040, with a projected population of 1,466 million, the annual percapita production will be 8.50 MWh. For comparison, the U.S. net generation in 2010 was 13.3 MWh per capita and Japan's consumption was 8.34 MWh per capita. 34 33 IEA (2014) page 33. 34 Electricity output is given in  In this Current Policies scenario (CPS) the coal share of total output falls from quickly 77.1% in 2010 to 67.3% in 2020 and then decline slowly to 64.1% in 2030, and remain at about 64% thereafter. In terms of growth rates, coal generation grows at 3.5% during 2012-20 compared to 5.2% for total electricity output. Wind and solar together rise from 1.2% of total output in 2010 to 5.5% in 2020, and then continue to rise to 7.3% in 2030.
In the New Policies scenario (NPS) a more aggressive conservation path is assumed, with an even bigger shift to renewables. Figure