The green spark of partial reform: Evidence from China's electricity market experiment

Abstract

Can incremental market reforms stimulate clean energy transitions in coal-dependent systems? China's electricity spot market pilot in Guangdong, which is a regulated experiment in partial liberalization, reveals how even limited market exposure can spark systemic transformation. Leveraging quarter-hourly, unit-level generation data, and a novel causal inference approach, we demonstrate that the implementation of market mechanisms decreased coal-fired power output by 6.6% while increasing renewable utilization and system flexibility reserves. The “green spark” originated from two mechanisms: (i) exposing cost structure differences between inflexible coal units and near-zero-marginal-cost renewables and (ii) reducing renewable curtailment through market-driven dispatch. However, the reform's decarbonization potential was partially offset by an increasing dependence on gas-fired and small coal units for flexibility. While partial marketization can drive energy transitions, legislative safeguards are needed to address flexibility gaps. These insights guide developing economies in accelerating renewable adoption through incremental institutional innovation.

Significance statement.

Policymakers debate whether full electricity market liberalization is essential for clean energy transitions. Our study of China's pioneering partial spot market reform reveals that even limited market exposure can accelerate systemic shifts toward renewables. Using real-world data, we show this “green spark” reduced coal power by 6.6% while boosting renewable grid access—primarily by exposing cost advantages of renewables over coal and reducing wasted clean energy. The study demonstrates that institutional factors and operational mechanisms also contribute to curtailment rates. Partial marketization can ignite transitions in coal-heavy economies, but maximizing climate benefits requires policies that incentivize clean flexibility. These insights offer a roadmap for emerging nations pursuing incremental power reforms.

Introduction

Coal's continued dominance in electricity generation remains a major barrier to global decarbonization. China is the world's largest coal consumer, and half of its coal consumption is attributed to the power sector (1). In 2020, the total installed capacity of coal-fired power plants in China reached 1,095 GW, emitting significant air pollutants and generating a carbon footprint of 2,438.58 Mt, threatening public health and jeopardizing global climate targets (2–9). For decades, strict command-and-control governance of China's power sector hid market signals (10, 11), which raised economic, environmental, and social costs due to inefficiency in the power sector (12, 13). In recent years, China has gradually reformed its electricity market to improve energy efficiency and reduce dependence on coal (10). However, the extent to which these market reforms can drive systemic change in a coal-dependent system remains unclear.

The market-oriented reforms in China's power sector are cautious and gradual. The Chinese government has a very low tolerance for energy supply disruptions or sharp increases in end-user energy prices (13). Unlike the US approach of establishing spot markets before medium- to long-term markets, China first introduced medium- and long-term electricity trading mechanisms (14). Only in 2018 did it begin pilot spot markets in eight provinces. When Guangdong launched its continuous spot market operation (including day-ahead and real-time markets) in November 2021, about 21% of electricity was settled through spot trading. In this market, only electricity output from coal- and gas-fired generators was traded. At the same time, government-guided prices still set the feed-in tariffs for wind and solar generators. Therefore, Guangdong's spot market reform in 2021 was incremental and partial. Nevertheless, the introduction of the spot market represents a key structural change as it provides a pricing anchor for electricity, including medium- to long-term contracts. Studies found that China's power market reforms have lowered thermal power plants’ generation costs (11, 12, 15, 16). While existing studies focus on how deregulation and marketization affect thermal power, an essential but unanswered question remains: how partial marketization reshapes the entire power system's energy mix.

The literature demonstrates the potential of marketization in the power sector to enhance energy efficiency. Theoretically, transitioning from a lagging cost-of-service mechanism to a market-based pricing system would enable a more efficient allocation of electricity shares among units, thereby improving market allocation efficiency and encouraging power plants to reduce costs (17). Researchers have found that introducing market mechanisms to determine production can increase electricity generation efficiency in the United States (14, 18, 19). Cicala discovered that deregulation can also mitigate regulatory capture and lower fuel costs for coal-fired power plants (20). Using cross-country data, Nepal et al. documented a general pattern that market-related reforms improve macro energy efficiency in transition countries (21). Research on China's electricity market reform also indicates that marketization improves grid dispatch efficiency and lowers wholesale electricity prices (12, 20, 21). However, these studies primarily consist of ex ante evaluations focused on coal-fired generators (22, 23). The effects of market reforms on renewable energy development and the underlying economic mechanisms are still not well understood.

In this study, we evaluate whether and how introducing a spot market leads to systemic changes in the energy structure using quarter-hourly unit-level data from Guangdong Province from May 2019 to April 2022. First, we establish a causal framework that allows us to assess the impact of market structure changes on the installed capacity of wind power. Next, we analyze changes across the entire electricity system by assessing the effects of the spot market on all types of generation units, including thermal, nuclear, variable renewable energy (VRE), hydropower, biomass, and pumped storage (PS), even though only thermal power is traded in the market. Finally, we clarify the mechanisms through which the establishment of the new market affects nonthermal power generation from both extensive and intensive margins.

Results

Spot market reform boosts wind power development

We first document that introducing a spot market for thermal power significantly accelerated the growth of wind power capacity. Among various renewable energy technologies, we first focus primarily on wind power, as solar power in Guangdong is immaterial, and hydropower projects typically require years to complete (Fig. 1A). Notably, we observe a rapid increase in wind power capacity connected to the grid following the introduction of the spot market (Fig. 1C). From November 2021 to April 2022, the average monthly growth rate of wind power capacity in Guangdong was 29.7%, compared with an average growth of only 5.8% in the 6 months prior to the spot market reform. However, attributing this growth to the spot market reform within the empirical setting is challenging due to confounding factors such as energy demand (grid load) and thermal generation costs (coal prices) during the same period (Fig. 1D).

Figure 1.

The energy structure and the spot market reform in Guangdong province. A) The energy structure in Guangdong province in 2022. B) Monthly average capacity factor of power-generating units in Guangdong province in 2022. C) The installed wind capacity in Guangdong province before and after the partial spot market reform. D) The population-adjusted temperature exposure, grid load, and imported coal price in Guangdong from May 2021 to April 2022. E) The market price of electricity in Guangdong before and after the reform.

Figure 2 illustrates the impact of introducing a spot market on wind power capacity using a period-month difference-in-differences (DID) strategy that allows us to regard the reform as a quasi-experiment (24). As shown in Fig. 2A, following the introduction of the spot market, the installed capacity in Guangdong Province experienced rapid growth compared with ex ante periods (before April 2021). Furthermore, after controlling for potential confounding factors and fixed effects, Guangdong Province had an average monthly capacity of 44 MW connected to the grid per wind farm compared with the counterfactual (Table S1 and Fig. 2B). This result is unlikely to be explained by ex ante trends in wind power development (Fig. 2B).

Figure 2.

Spot market reform increases installed wind power capacity. A) Trend of wind power capacity in two periods. B) Event study results for the impact of introducing spot market on installed wind power capacity. The points illustrate the estimated results of model 2, while the error bars represent the 95% robust CIs for the point estimates.

Partial marketization activates systemic energy transition

Figure 3 summarizes the impact of introducing a thermal power spot market on the output of all power-generating units connected to the grid, categorized by technology type. In Fig. 3A, marketization, even if only partial, significantly reduced the output of coal-fired units by 6.6% (15.9 MW per unit) while increasing the output of renewable energy units. Compared with the counterfactual without a spot market, the outputs of wind, hydropower, and solar power units increased by 227.5, 161.6, and 137.9%, respectively. Natural factors (proxied by wind speed in this study) can also drive the increase in wind power output and displace coal-fired units’ share (Fig. 3B), as the wind energy resource is naturally a function of the fluctuating wind speed (25, 26). In comparison, the effect of the spot market is more pronounced. For wind power, the impact of the spot market reform is equivalent to an increase in wind speed of 6.9 SDs (16.8 m/s) from the average (4.5 m/s). Additionally, we observe an increase in the output of biomass and PS units, which provide essential flexibility for integrating intermittent VRE into the grid (27). There is no significant change in the output of gas-fired units before and after the spot market reform, likely due to the combined effects of the displacement of new energy generation and the need for flexibility. In addition, we are aware that the impact of nuclear output may require more caution due to the limited number of nuclear units in the data, which results in larger estimation errors.

Figure 3.

The impact of spot market reform and natural driver on generator output power. All estimates in (A) and (B) are derived from a single regression using model 3. The diamonds and error bars represent point estimates and the 90% CIs of model 3 coefficients. The P-values for these point estimates are indicated above the error bars. The values on the right of each plot show the results of converting the point estimates based on the ex ante mean into percentages.

Figure S1 reports the dynamic effects of introducing a spot market on the output of different types of generation units. We do not find significant ex ante differences, which provided suggesting evidence to support the parallel trends assumption required for interpreting the impact as causal (24, 28). Observing the dynamic trends, we find significant complementarity in the output of VRE units (wind and solar) and flexibility provision units (gas, hydro, biomass, and PS). Furthermore, our results remain robust after a series of robustness checks (Fig. S2 and Table S2).

Figure S3 illustrates the heterogeneous effects of spot market reform on the output of thermal power units. We found that the reform primarily reduced the output of larger units (L, over 300 MW; Fig. S3A). The spot market reform decreased the output of high-emission coal units (high NO) by 15.5% and low-emission coal units (low NO) by 9.1% (Fig. S3E). The reform significantly reduced the output of state-owned enterprise (SOE) coal units, while there was no significant impact on the output of non-SOE units (Fig. S3G). We do not observe this difference in the natural factor-driven displacement effects (Fig. S3H). Theoretically, this could be a natural outcome of introducing market mechanisms, as non-SOE units might have lower costs, thereby displacing SOE market share (29). However, no evidence supports significant differences in capacity scale and thermal efficiency between non-SOE and SOE units (Fig. S4). These results indicate that coal-fired power units operated by SOEs bear the majority of the costs of the reforms. This ownership heterogeneity is linked to the additional social obligations and higher overheads shouldered by SOEs (23). In the Chinese context, it is common for the government to internalize the costs of reform by leveraging SOEs’ market positions (13).

Induced carbon emissions from increased demand for flexibility partially offset the reform's decarbonization benefits. Units with a smaller capacity can start and stop more flexibly and at lower cost, thereby enhancing the system's ability to integrate a growing share of VRE into the grid (30). Our results indicate that after the spot market reform, the output of smaller gas-fired units (<300 MW) significantly increased by 19.7%. Counterfactual analysis further reveals that market reforms have reduced SO₂ emissions by 15.9%, ash by 15.7%, NO_x by 13.4%, and CO₂ by 6.7% per year (Fig. S5). However, carbon emissions from gas and smaller coal units have offset some of the carbon reduction effects to provide flexibility. Prioritizing market incentives for investments in clean flexibility provision capacity should be a key focus in the next phase of reforms (27).

Driving mechanisms at the extensive and intensive margins

We then elucidate two mechanisms through which introducing a spot market for thermal power activates systemic changes. First, the spot market highlights the economic viability of renewable energy, incentivizing the capacity growth of VRE and flexibility provision units. The spot market provides an effective price discovery mechanism, revealing the cost structure of thermal power (Fig. 1D and E). Although the spot market prices were higher than the regulated price prior to the reform, most of the time, the fluctuating coal prices have made coal power plants less profitable (Fig. 1E) (31). Compared with the counterfactual scenario without a spot market, the profit margin of coal units significantly decreased by 44.1% (Fig. 4A). Electricity markets and power plant operators require sufficient time to adjust to the impacts of the new price mechanism. We capture that the spot market reached a new equilibrium point after 2 months of adjustment (Fig. 4A). Due to the increase in output and capacity factor, the profit margin of wind power units rose by 24.4% 6 months after the reform compared with the counterfactual. Allowing a portion of wind power generation to be dispatched through the spot market will make wind power and flexible power investments even more profitable (Figs. 4B and S6).

Figure 4.

Counterfactual analysis of profitability of coal-fired and wind power units. A) The cumulative changes in profit and revenue for coal-fired power units after introducing the spot market compared with the counterfactual scenario without a spot market. B) The impact of introducing the spot market on revenue for wind power plants under varying spans of market dispatch scenarios. The shaded area represents the 90% CI for the estimated trajectories.

Second, the spot market has increased existing wind power units’ efficiency and output. An important background to consider is that, before the operation of the spot market, the government had published the policy and established a principle urging that renewable energy (wind and solar) should be prioritized for access to the grid (32). However, this priority principle does not imply that wind power units were fully operational (33). The reasons are 2-fold. On the one hand, under the regulated dispatching system, efficiently executing the priority principle is technologically and politically challenging (33, 34). On the other hand, when the total capacity of wind power is relatively small, the flexibility costs of integrating wind energy into the grid are high (27, 34). As a result, even with the priority dispatching principle, the wind curtailment rate in Guangdong was relatively high before 2022 (Fig. S7 and Text S1: Estimates of wind power curtailment rate). We find that as the scale of wind power units increased and more flexibility entered the system, the curtailment rate of wind power units decreased rapidly (Fig. 5 and Table S3). This reduction illustrates how spot markets operationalize renewable priority through not only real-time pricing signals but also economies of scale and integration of flexibility.

Figure 5.

The impact of spot market reform on wind power curtailment rate. The plot presents the regression results of applying the event study model to wind power units’ monthly curtailment rate. The circles and error bars illustrate the point estimates and their 95% CIs of the relative changes in curtailment rates (in percentage) compared with the counterfactual at different months relative to the start of the spot market.

Discussion

Our research establishes an empirical framework for evaluating systemic electricity supply-side transformations induced by partial spot market reforms. Unlike existing studies that rely on ex ante simulations or power dispatch models, our approach integrates real-time, high-frequency operational data and employs quasi-experimental methods to document the effects of market reforms in a genuine institutional context (15, 29, 35). Our empirical findings reveal how Guangdong's market experiment acted as a “green spark,” accelerating systemic shifts toward renewables through two mechanisms. First, market mechanisms have revealed the cost structure of thermal power generation, highlighting the differences in profit margins between coal-fired and renewable energy. This has incentivized coal phase-out and stimulated investment in renewables. Second, marketization has increased the capacity of VRE and flexibility provision capacity, which strengthened priority dispatch for variable renewables and reduced curtailment.

Our results highlight the necessity of incorporating market mechanisms into the design of a clean transition pathway. The key to triggering systemic change through market reforms lies in the linkage between electricity and coal prices, and expanding the coverage of the spot market can further compress the marginal returns of thermal power generation. However, our findings also demonstrate that the spot market reform significantly impairs thermal power plants, especially coal-fired generating units. Affected units experience immediate reductions in profit margins and incur greater losses due to heightened asset-stranding risks (36). Further analysis indicates that leveraging the state-owned coal-fired power units is an essential strategy for absorbing the impact. Yet this approach requires state-owned coal plants to dominate the power grid. Therefore, economies with greater private-sector participation in coal-fired power generation face more severe challenges of balancing short-term economic shocks and long-term energy transition when implementing spot market reforms.

We also show that institutional factors drive up VRE curtailment rates and constrain renewable development. Without costly upgrades such as grid connectivity improvements, partial marketization can still reduce curtailment by reshaping dispatch priorities through price signals. Guangdong's spot market reform offers a template for broader marketization in China and other developing economies. Our results emphasize the essential role of flexibility, aligning with China's post-2024 efforts, such as capacity pricing policies and building regional ancillary service markets.

Some limitations warrant consideration. While Guangdong's coal-dependent structure mirrors that of many transitioning economies, its unique institutional features set it apart from other nations and regions. We document that the partial reform imposes negative impacts on coal-fired power units engaging in trading. This prompts a hypothesis: could comparably adverse effects occur when market pricing mechanisms incorporate VRE units? Therefore, cross-regional and multiscenario validations are important for the generalizability of our findings, when data permits. Furthermore, due to the unavailability of more specific transaction-level data on the spot market, we do not capture the strategic bidding dynamics among spot market participants. Nevertheless, our empirical framework provides a methodical foundation for exploring these dimensions in future work. Despite these limitations, this paper delivers an important policy insight: partial marketization can stimulate energy transitions, but sustaining the green spark requires complementary policies to manage flexibility trade-offs during the decarbonization process.

Materials and methods

Empirical framework of estimating the impact of spot market reform on installed wind power capacity

The primary dataset for this study contains quarter-hourly grid load and unit-level output for all technology types of power-generating units in Guangdong Province, China, spanning May 2019 to April 2022 (Fig. S8). This dataset was obtained from the Guangdong Power Exchange Center (GPEC). For wind and solar generation clusters, output data are aggregated to the farm level. We augmented this dataset by merging it with Global Energy Monitor records and conducting manual web searches, adding information on locations, affiliated companies, and ownership structures. The dataset only includes the power-generating units that were registered with the GPEC. Consequently, our analysis does not cover distributed PV and smaller-scale wind and hydropower units. Under the prevailing policy, these units can neither connect to the grid nor participate in spot market trading. Table S4 compares the installed capacity of the units in our study at the end of 2021 with the installed capacity reported in the statistical yearbook. Our research data include 93% of thermal power capacity. The combined output of all units in our data (plus electricity imported from other provinces) covered an average of 94% of Guangdong's grid load. Due to a significant amount of distributed photovoltaic generation not being integrated into the grid as per policy, the coverage is lower. The data encompass 41% of the province's wind power installed capacity, primarily from offshore wind clusters (Fig. S8). Therefore, our analysis primarily focuses on wind power.

Guangdong is the province with the highest gross domestic product (GDP) and the most active socioeconomic activities in China. In 2024, its GDP exceeded $2 trillion, supported by a resident population of ∼128 million. By 2021, Guangdong's total installed power capacity reached 158 GW, equivalent to 2.2 times that of California. The province's average grid load reached 88 GW, triple Spain's grid load and 1.5 times Germany's. Despite substantial policy support for a clean energy transition, thermal power remains Guangdong's dominant electricity source (∼65% of installed capacity). Accelerating clean energy deployment while reducing thermal power's air pollution and carbon emissions continues to be central to Guangdong's sustainable development objectives.

We begin with a monthly model to analyze the effect of spot market reform on installed wind power capacity. Following (24), we estimate the reform's causal impact using this regression specification:

$$\mathrm{WindCa}{\mathrm{p}}_{iym}={\beta }^{\mathrm{DID}}\mathrm{DI}{\mathrm{D}}_{pm}+\mathit{C}+{\pi }_m+{\lambda }_{pi}+{\epsilon }_{iym}$$ (1)

where $\mathrm{WindCa}{\mathrm{p}}_{iym}$ represents the installed capacity of wind farm i that connects to the grid in year y and month m. $\mathrm{DI}{\mathrm{D}}_{pm}$ is a dummy variable that equals 1 only for the months from November 2021 to April 2022; 0 otherwise. $\mathit{C}$ is a set of control variables, including the monthly average of temperature $(\mathrm{MeanTemperatur}{\mathrm{e}}_{ym})$, daily imported electricity $(\mathrm{MeanGridImpor}{\mathrm{t}}_{ym})$, coal prices $(\mathrm{MeanCoalPric}{\mathrm{e}}_{ym})$, and wind speed $(\mathrm{MeanWindSpee}{\mathrm{d}}_{iym})$. We obtain hourly 2 m temperature and 100 m wind speed data for Guangdong at 0.25° × 0.25° resolution from the ERA5 reanalysis dataset. To better capture the temperature's effect on electricity demand and grid load, we aggregate temperatures at the city level and compute province-level exposure using 2018 population weights. For wind speeds, we use values from grid cells containing wind farm center coordinates. The coal price refers to the imported thermal coal price at Guangzhou Port.

${\pi }_m$ and ${\lambda }_{pi}$ denote month and field-period-fixed effects, respectively. ${\epsilon }_{iym}$ is the random error term. To achieve optimal estimation windows, we restructure the timeline into three “periods,” each consisting of 12 months (period 1: May 2019 to April 2020; period 2: May 2020 to April 2021; period 3: May 2021 to April 2022; Fig. 2A). By controlling for month and period-fixed effects, we establish an identification strategy analogous to a DIDs approach. Specifically, we treat period 3 as the treatment group and periods 1 and 2 as the control group. The key underlying identification assumption is that, in the absence of spot market reform, the wind farm's counterfactual trend of installed capacity in period 3 would be similar to that in periods 1 and 2 (the parallel trends assumption, PTA). Including field-period interaction fixed effect ${\lambda }_{pi}$ ensures that comparisons are made solely within the same wind farm across different periods. Thus, coefficient ${\beta }^{\mathrm{DID}}$ captures the causal effect of spot market reform, representing the policy-induced change in installed wind power capacity per wind farm per month.

Our empirical framework helps minimize confounding factors and avoids violating the stable unit treatment value assumption (SUTVA). SUTVA requires unambiguous treatment assignments and no spillover effects between units. Since provincial grid dispatch operates centrally, comparing different generators within the same period would violate SUTVA. However, to establish causality, we also need the PTA to be satisfied. We further provide empirical evidence for supporting the PTA using the following dynamic model:

$$\begin{array}{rl}\mathrm{WindCa}{\mathrm{p}}_{iym}& =\sum _{\begin{array}{c}j=-6\\ j\ne -1\end{array}}^5{\beta }_j\mathrm{Trea}{\mathrm{t}}_p\mathrm{Mont}{\mathrm{h}}_j+\mathit{C}+{\pi }_m+{\lambda }_{pi}\\ & +{\epsilon }_{iym},\end{array}$$ (2)

where $\mathrm{Trea}{\mathrm{t}}_p$ is a binary indicator equal to 1 for period 3 and 0 otherwise. $\mathrm{Mont}{\mathrm{h}}_j$ denotes a dummy variable that takes the value 1 when the current month is exactly j months from November (the reference month) and 0 otherwise. To prevent multicollinearity, we omit $\mathrm{Mont}{\mathrm{h}}_{-1}$.

Estimates of spot market effects on power generation unit output

We apply the same identification strategy used in monthly analysis to quarter-hourly data to estimate spot market impacts on power-generating units’ output. The regression specifications are defined as follows:

$$\mathrm{Outpu}{\mathrm{t}}_{kiymdhq}=\mathit{M}+\mathit{N}+\mathbf{\Omega }+e_{kiymdhq}$$ (3)

and

$$\{\begin{array}{c}M=\sum _{k=1}^8{\alpha }_k(\mathrm{Fue}{\mathrm{l}}_k\times \mathrm{Trea}{\mathrm{t}}_p\times \mathrm{Refor}{\mathrm{m}}_m)\\ N=\sum _{k=1}^8{\beta }_k(\mathrm{Fue}{\mathrm{l}}_k\times \mathrm{Natur}{\mathrm{e}}_{iymdh})\\ \mathrm{\Omega }=\theta {\mathit{X}}^{\mathbf{\prime }}+{\eta }_k+{\pi }_m+{\lambda }_{pi}+{\mu }_h\end{array}$$ (4)

where $\mathrm{Outpu}{\mathrm{t}}_{kiymdhq}$ represents the output of unit i with technology type (or fuel type) k (including coal, gas, nuclear, wind, hydro, solar, biomass, and PS) in year y, month m, day d, hour h, and quarter q. $\mathrm{Fue}{\mathrm{l}}_k$ is a binary indicator for the technology type of the unit. $\mathrm{Trea}{\mathrm{t}}_p$ has the same definition as in model 2. $\mathrm{Refor}{\mathrm{m}}_m$ is a dummy variable that equals 1 if month m falls within the postreform period (November, December, and from January to April). In model 3, we allow the reform to have heterogeneous effects on units of various technology types. To achieve this, we control for month $({\pi }_m)$, unit period $({\lambda }_{pi})$, and type $({\eta }_k)$ fixed effects. Additionally, we omit the constant term in our estimation to avoid multicollinearity issues. The Hour $({\mu }_h)$ fixed effect is also included to control periodic output patterns among different hours in a day. The control variables $\mathit{X}$ include workday $(\mathrm{Workda}{\mathrm{y}}_{ymd})$, temperature $(\mathrm{Temperatur}{\mathrm{e}}_{ymdh})$, imported electricity $(\mathrm{GridImpor}{\mathrm{t}}_{ymdhq})$, and coal price $(\mathrm{CoalPric}{\mathrm{e}}_{ymd})$. Thus, coefficient ${\alpha }_k$ can be interpreted as the average treatment effect of the spot market reform on units’ output in each technology type $(k)$. To compare the reform's effects with changes in output caused by natural factors, we also introduce the term $\mathrm{Natur}{\mathrm{e}}_{iymdh}$ (proxied by hourly wind speed) at the location of each unit. More precisely, the economic interpretation of ${\alpha }_k$ and ${\beta }_k$ can be expressed as follows:

$$\{\begin{array}{c}{\alpha }_k=\mathbb{E}[\mathrm{\Delta }{\mathbb{E}}_k^{pre-post}(1)-\mathrm{\Delta }{\mathbb{E}}_k^{pre-post}(0)]\\ {\beta }_k=\mathbb{E}\left[\frac{\mathrm{\partial }}{\mathrm{\partial }\mathrm{Natur}{\mathrm{e}}_{dh}}\mathbb{E}(\mathrm{Output}|k)|M+\mathrm{\Omega }\right]\end{array}$$ (5)

and

$$\begin{array}{rl}\mathrm{\Delta }{\mathbb{E}}_k^{\mathrm{pre}\text{-}\mathrm{post}}(s)=& \mathbb{E}[\mathbb{E}(\mathrm{Output}|k,\mathrm{Reform}=1,\mathrm{Treat}=s)\\ & -\mathbb{E}(\mathrm{Output}|k,\mathrm{Reform}=0,\mathrm{Treat}=s)|\mathit{N}+\mathbf{\Omega }],\end{array}$$ (6)

where s = 0, 1. Here, α and β represent the treatment effects expressed as absolute changes in output. To further interpret the economic significance, we convert these point estimates into percentages by dividing by the units’ ex ante mean output. We do not use a logarithmic transformation to avoid the unit sensitivity issue (37). Using this approach, the percentage estimates should be interpreted as the relative change in output power of type k units resulting from the reform, compared with the ex ante mean. Similar to model 2, we use the following specification for testing the dynamic effects of the reform:

$$\begin{array}{rl}\mathrm{Outpu}{\mathrm{t}}_{kiymdhq}=& \sum _{k=1}^8\sum _{\begin{array}{c}j=-6\\ j\ne -1\end{array}}^5{\alpha }_{kj}^{\mathrm{\prime }}\times (\mathrm{Fue}{\mathrm{l}}_k\times \mathrm{Trea}{\mathrm{t}}_p\times \mathrm{Mont}{\mathrm{h}}_j)\\ & +\mathit{N}+\mathbf{\Omega }+e_{kiymdhq}.\end{array}$$ (7)

Counterfactual analysis

We conduct a counterfactual analysis based on empirical estimation results to assess the impact of the spot market on profit margins and emissions of units. Based on model 6, we have:

$${\mathrm{Output}}_{kiymdh}^{counterfactual}=\mathrm{Outpu}{\mathrm{t}}_{kiymdh}\times (1+\widehat{{\alpha }_k}{)}^{-1}.$$ (8)

We calculate the actual and counterfactual revenue of the power-generating units based on:

$$\{\begin{array}{c}{\mathrm{Revenue}}_{kymdh}^{real}=\sum _i\mathrm{Outpu}{\mathrm{t}}_{kiymdh}\times \mathrm{EPric}{\mathrm{e}}_{ymdh}\\ {\mathrm{Revenue}}_{kymdh}^{counterfactual}=\sum _i{\mathrm{Output}}_{kiymdh}^{counterfactual}\times 453,\end{array}$$ (9)

where $\mathrm{EPric}{\mathrm{e}}_{ymdh}$ is the hourly spot market electricity price. Similarly, we estimate the actual and counterfactual trends in profit and emissions for units with different technologies (Text S2). Finally, we use the ratio of the cumulative values of the actual and counterfactual scenarios to reflect the average impact over 6 months following the spot market reform for different technology types $(k)$ and emissions $(j)$:

$$\{\begin{array}{c}\text{\%}\mathrm{\Delta }\mathrm{Revenu}{\mathrm{e}}_k=100\text{\%}\times \frac{{\int }_{t_1}^{t_2}[{\mathrm{Revenue}}_{kymdh}^{real}-{\mathrm{Revenue}}_{kymdh}^{counterfactual}]\mathrm{d}t}{{\int }_{t_1}^{t_2}{\mathrm{Revenue}}_{kymdh}^{counterfactual}\mathrm{d}t}\\ \text{\%}\mathrm{\Delta }\mathrm{Profi}{\mathrm{t}}_{\mathrm{coal}}=100\text{\%}\times \frac{{\int }_{t_1}^{t_2}[{\mathrm{Profit}}_{ymdh}^{real}-{\mathrm{Profit}}_{ymdh}^{counterfactual}]\mathrm{d}t}{{\int }_{t_1}^{t_2}{\mathrm{Profit}}_{ymdh}^{\mathrm{counterfactual}}\mathrm{d}t}\\ \text{\%}\mathrm{\Delta }\mathrm{Emissio}{\mathrm{n}}_j=100\text{\%}\times \frac{{\int }_{t_1}^{t_2}[{\mathrm{Emission}}_{jkymdh}^{real}-{\mathrm{Emission}}_{jkymdh}^{counterfactural}]\mathrm{d}t}{{\int }_{t_1}^{t_2}{\mathrm{Emission}}_{jkymdh}^{counterfactural}\mathrm{d}t}\end{array}$$ (10)

Supplementary material

Supplementary material is available at PNAS Nexus online.

Competing interest

The authors declare no competing interests.

Funding

The authors acknowledge funding support from the Research Grants Council of Hong Kong (N_CityU146/23), the Science and Technology Innovation Commission of Shenzhen Municipality (JCYJ20240813153136047), the Key Policy Project of China Meterological Administration (2026ZCYJZDIAN03), and the Fundamental and Interdisciplinary Disciplines Breakthrough Plan of the Ministry of Education of China (JYB2025XDXM905).

Author contributions

Xinya Hao (Data curation, Formal analysis, Investigation, Methodology, Visualization, Writing—original draft, Writing—review & editing), Ao Sun (Data curation, Formal analysis), Feng Song (Conceptualization, Investigation, Supervision, Validation), Huanhuan Zheng (Conceptualization, Investigation, Writing—review & editing), and Lin Zhang (Conceptualization, Funding acquisition, Investigation, Project administration, Resources, Supervision, Writing—original draft, Writing—review & editing)

Data availability

All the codes used for the paper have been stored at https://osf.io/wyj57/overview?view_only=319323e981534ce491c718e1044fb13c. The unit-level quarter-hourly data and grid load data used in this study were provided by the Guangdong Power Dispatch Control Center (No. 75, Meihua Road, Tianhe District, Guangzhou, Guangdong, China). Due to the terms of a nondisclosure agreement regarding sensitive grid information, these data are confidential and cannot be made publicly available in the replication package. The ERA5 reanalysis dataset can be accessed for free through the Climate Data Store (CDS) API (https://cds.climate.copernicus.eu). The imported coal price data were obtained from the Wind Financial Terminal, specifically the “Steam Coal Price Index CCI7 (USD): Import 3800” and the “Thermal Coal Price Index (Imported, 3800 kcal/kg, USD).” These data are available through the Wind database (https://www.wind.com.cn) via institutional subscription. The authors also include the imported coal price data in the replication package, which is available at ./data/fig1d.csv. Meteorological data were retrieved from the publicly available ERA5 reanalysis dataset. Data acquisition was performed via the Climate Data Store (CDS) API, and the specific implementation scripts are documented in the replication package (./codes/ERA5/_cds_download.ipynb and ./codes/ERA5/_cds_download20192020.ipynb). The Stata codes for further calculations are also included in the replication package (./codes/ERA5/_clean_era5.do). We use Stata as the primary software for data analysis. Python scripts are employed to download data from the CDS, and MATLAB is used for processing meteorological data. The codes and related materials are available in the online Supplementary material.