Small actions make a big difference engilsh dkbee.com

홈

Climate Risk Attention and Nonlinear Stock Market Responses: Evidence from an Emerging Market

 

Climate Risk Attention and Nonlinear Stock Market Responses: Evidence from an Emerging Market

Highlights

  • •
    Constructs four climate attention indices from Chinese corporate disclosures.
  • •
    Reveals nonlinear (inverted U-shaped) effects on stock market reactions.
  • •
    Identifies heterogeneous impacts by ownership, region, carbon intensity, and climate policy phase.
  • •
    Applies Huber robust regressions to handle outliers and nonlinearity.
  • •
    Offers micro-level evidence for climate disclosure and sustainable policy design.

Abstract

Climate change is increasingly shaping financial markets, particularly in developing economies with evolving institutional frameworks and disclosure practices. This study constructs four firm-level climate attention indices—Aggregate, Physical Risk, Transition Risk, and Opportunity—based on over 117,000 Chinese-language earnings calls and broker reports. Using a keyword discovery method enhanced by natural language processing (NLP), we quantify climate-related disclosure intensity across A-share listed firms. Huber robust regressions within an event-study framework reveal a significant inverted U-shaped relationship between climate attention and cumulative abnormal returns (CARs), especially over longer event windows. Among the four dimensions, transition risk attention elicits the strongest and most persistent market responses. Moreover, the effects vary systematically by ownership type, carbon intensity, policy regime, and geographic region. These findings provide novel micro-level evidence on how investors in emerging markets process climate disclosures, offering implications for disclosure regulation, sustainable finance, and capital market reforms in low- and middle-income countries (LMICs).

Keywords

Climate Attention
Nonlinear Stock Returns
Emerging Markets
China
Textual Analysis
Developing Countries
Sustainable Finance

1. Introduction

Climate change poses systemic risks to firms and financial markets, especially in LMICs, where institutional capacity, disclosure norms, and investor sophistication remain underdeveloped (Bartram et al., 2022; Zhai et al., 2024). Among physical risks, transition risks, and opportunity, transition risks—linked to regulatory uncertainty, technological disruption, and shifting market expectations—have become particularly salient in investment decisions and asset pricing (Sautner et al., 2023; Zhu et al., 2023). In emerging economies such as China, fragmented policy frameworks and the dominance of textual disclosures further complicate the interpretation of climate-related information (Ouyang et al., 2025; Wu et al., 2022).
This study investigates how firm-level climate disclosures affect market responses in such emerging contexts. Prior research applies NLP techniques to extract climate attention indices from earnings calls and regulatory filings (Ilhan et al., 2023; Kolbel et al., 2022). In China, climate-related language has been shown to influence investor sentiment and crash risk (Q. Li et al., 2024; Lin & Wu, 2023). Recent work also highlights different market responses to transition versus physical risk narratives (Bua et al., 2024; Chen et al., 2024).
However, most existing studies rely on linear assumptions. A growing body of evidence points to nonlinear investor responses to climate disclosures, including threshold effects, saturation, or overreaction (An et al., 2022; Faccini et al., 2023; Xu et al., 2025). Such dynamics are particularly relevant for transition risk, where regulatory ambiguity and greenwashing suspicion may induce diverging interpretations (Pankratz et al., 2023; Wang et al., 2024). In China, Chen et al. (2023) document non-monotonic volatility responses to climate policy uncertainty. Investor heterogeneity across ownership types, carbon exposure, and geographic location further adds complexity (Y. Li et al., 2024; Ma et al., 2025).
These observations motivate two innovations: (i) constructing semantically precise textual attention indices to capture risk-type distinctions, and (ii) employing flexible econometric tools to uncover nonlinear market reactions. This study has three core objectives:
First, we construct four firm-level climate attention indices—Aggregate, Physical Risk, Transition Risk, and Climate Opportunity—based on over 117,000 Chinese-language earnings calls and broker reports. These indices are generated through supervised keyword discovery and domain-specific semantic filtering. Second, we apply Huber robust regressions to assess the (nonlinear) effects of these indices on CARs over multiple event windows. Third, we explore heterogeneity across ownership structure, industry carbon intensity, regional policy exposure, and national climate policy regimes.
We find that transition risk attention yields the strongest and most persistent effects, and that the relationship between climate attention and returns follows a robust inverted-U shape—particularly over medium and long horizons.
Our study provides novel micro-level evidence from LMICs, showing how investors in emerging markets respond to complex climate disclosures. The findings offer guidance for improving disclosure standards and regulatory frameworks in carbon-intensive economies.

2. Data and Methodology

We compile over 117,000 Chinese-language earnings calls and broker reports from CNINFO (2013–2023), covering the full universe of A-share listed firms. This period reflects standardized disclosure formats and intensifying climate discourse following major national policy transitions. Firm-level financials and return data are obtained from the Wind database, with the CSI 300 index used as the market benchmark.
We construct four firm-level climate attention indices: Aggregate (CI), Physical Risk (CIphy), Transition Risk (CItran), and Opportunity (CIopp). Following Sautner et al. (2023), we develop a keyword discovery method tailored to Chinese texts by combining authoritative glossaries (e.g., IPCC) with Word2Vec embedding and expert validation. Climate attention is measured as the standardized frequency of climate-related characters, rather than words, to ensure consistency in Chinese-language processing. For detailed keyword sets and classification procedures, see Supplementary Appendix C.1–C.3.
To ensure semantic distinctiveness among sub-indices, we perform topic-specific filtering and iterative manual refinement. Overlapping terms are removed to enhance conceptual clarity and index discriminability (see Appendix C.3). For benchmarking against public indices—such as the Germanwatch Climate Risk Index and the Policy Uncertainty Climate Risk Index—see Appendix C.4.
We compute CARs across six event windows, ranging from (0,1) to (0,240) around earnings announcements. We estimate the nonlinear relationships using Huber robust regressions, which are resilient to outliers and heteroskedasticity. The empirical model is specified as:(1)where  denotes the cumulative abnormal return of firm i within the event window (τ1,τ2). CIit represents the climate attention indicator—referring to one of the four dimensions above. The squared term  captures potential nonlinear marginal effects. Xit includes firm-level controls: size (SIZE), return on assets (ROA), leverage (LEV), book-to-market ratio (BM), unexpected earnings (UE), and price volatility. μi and λt denote industry and year fixed effects, respectively.
To examine heterogeneity in investor responses, we perform subsample analyses by ownership structure, industry-level carbon intensity, geographic location, and policy regime (pre- vs. post–China’s carbon neutrality pledge). Detailed variable definitions and construction procedures are documented in Appendices A–F.

3. Results

3.1. Baseline Regression and Nonlinear Effects

We begin by evaluating the impact of firm-level climate attention on stock market performance. Descriptive trends and subgroup comparisons are available in Appendices G and H.
Table 1 presents regression estimates across six event windows. In the shortest window, neither the linear nor the squared term of the CI is statistically significant, suggesting delayed investor response to climate-related disclosures. As the window lengthens, the linear coefficient becomes increasingly negative, while the squared term turns significantly positive—indicating a robust inverted U-shaped pattern.

Table 1. Regression Results on the Nonlinear Effects of Aggregate Climate Attention (CI) on Cumulative Abnormal Returns (CAR).

Empty Cell(1)(2)(3)(4)(5)(6)
Empty CellCAR(0, 1)CAR(0, 5)CAR(0, 20)CAR(0, 60)CAR(0, 120)CAR(0, 240)
CI0.0017-0.0111*-0.0314⁎⁎⁎-0.0378⁎⁎⁎-0.0300⁎⁎⁎-0.0296⁎⁎⁎
Empty Cell(0.2990)(-1.9307)(-5.7019)(-8.4586)(-7.1592)(-7.0240)
CI2-0.0065-0.00080.0117⁎⁎0.0147⁎⁎⁎0.0157⁎⁎⁎0.0136⁎⁎⁎
Empty Cell(-1.2163)(-0.1469)(2.3006)(3.5801)(4.0760)(3.5036)
BM0.00370.00180.00090.0150⁎⁎⁎0.0236⁎⁎⁎0.0343⁎⁎⁎
Empty Cell(0.6461)(0.3130)(0.1683)(3.3333)(5.6247)(8.0936)
LEV-0.0020-0.00110.0003-0.0064⁎⁎⁎-0.0169⁎⁎⁎-0.0277⁎⁎⁎
Empty Cell(-0.3486)(-0.1956)(0.0514)(-1.4567)(-4.1092)(-6.6755)
ROA0.00390.0094⁎⁎⁎0.0076⁎⁎⁎0.01340.0227⁎⁎⁎0.0310⁎⁎⁎
Empty Cell(1.3921)(3.3390)(2.8074)(6.1309)(11.0821)(15.0453)
SIZE-0.00120.00510.00420.0220⁎⁎⁎0.0205⁎⁎⁎0.0326⁎⁎⁎
Empty Cell(-0.3238)(1.3758)(1.1676)(7.6198)(7.5735)(11.9220)
UE0.0075⁎⁎⁎0.0060⁎⁎0.00310.0154⁎⁎⁎0.0186⁎⁎⁎0.0278⁎⁎⁎
Empty Cell(3.0144)(2.4056)(1.2960)(8.0020)(10.3399)(15.3064)
Volatility-0.0133⁎⁎⁎-0.0187⁎⁎⁎-0.0193⁎⁎⁎0.00280.0057⁎⁎⁎0.0254⁎⁎⁎
Empty Cell(-5.0160)(-7.0393)(-7.6042)(1.3792)(2.9369)(13.0899)
Constant-0.0822⁎⁎⁎-0.0647⁎⁎⁎-0.2255⁎⁎⁎-0.6440⁎⁎⁎-0.5805⁎⁎⁎-0.4629⁎⁎⁎
Empty Cell(-4.4902)(-3.5322)(-12.8557)(-45.3375)(-43.6182)(-34.4926)
Time effectIncludedIncludedIncludedIncludedIncludedIncluded
Industry effectIncludedIncludedIncludedIncludedIncludedIncluded
N117068117068117068117068117068117068
Note: This table presents regression results of Aggregate Climate Attention (CI) and its squared term (CI2) on cumulative abnormal returns (CAR) across different event windows. CAR(τ1,τ2) denotes the cumulative abnormal return over the event window (τ1,τ2). Control variables include the book-to-market ratio (BM), return on assets (ROA), firm size (SIZE), leverage (LEV), unexpected earnings (UE), and stock return volatility. All regressions control for industry and year fixed effects and are estimated using Huber robust regression to address heteroskedasticity and outliers. t-values are calculated based on robust standard errors. Significance levels are indicated as * p<0.10 **p<0.05, and ⁎⁎⁎ p<0.01.
Moderate levels of climate attention may be interpreted as superficial or symbolic, potentially triggering negative valuation effects. In contrast, more intensive disclosures are perceived as strategic and credible, thereby restoring investor confidence. This nonlinear response reflects an adaptive learning process in how investors interpret climate-related signals over time.
To decompose the aggregate effect, Table 2 reports regression results for the three sub-indices. CItran exhibits the strongest and most consistent nonlinear pattern: its linear term becomes significantly negative even in early windows, while the squared term turns significantly positive over medium- and long-term periods. These results suggest that early-stage transition risk disclosures may initially raise investor concerns but are ultimately rewarded when perceived as credible adaptation strategies.

Table 2. Nonlinear Effects of Climate Risk Sub-Dimension Attention on Cumulative Abnormal Returns (CAR).

Empty Cell(1)(2)(3)(4)(5)(6)
Empty CellCAR(0, 1)CAR(0, 5)CAR(0, 20)CAR(0, 60)CAR(0, 120)CAR(0, 240)
CIopp-0.00240.0023-0.0109⁎⁎-0.0086⁎⁎0.0070*0.0107⁎⁎⁎
Empty Cell(-0.4335)(0.4200)(-2.0501)(-2.0012)(1.7332)(2.6295)
CIphy0.0030-0.0029-0.0190⁎⁎⁎-0.0190⁎⁎⁎-0.0204⁎⁎⁎-0.0143⁎⁎⁎
Empty Cell(0.6267)(-0.6074)(-4.1742)(-5.1456)(-5.9075)(-4.0950)
CItran-0.0014-0.0182⁎⁎⁎-0.0276⁎⁎⁎-0.0361⁎⁎⁎-0.0474⁎⁎⁎-0.0507⁎⁎⁎
Empty Cell(-0.2814)(-3.5968)(-5.7129)(-9.2221)(-12.9423)(-13.7225)
(CIopp)20.0002-0.00490.0088*0.00400.0042-0.0027
Empty Cell(0.0415)(-0.9490)(1.7957)(1.0047)(1.1239)(-0.7213)
(CIphy)2-0.0045-0.00210.0114⁎⁎0.0088⁎⁎0.0142⁎⁎⁎0.0085⁎⁎
Empty Cell(-0.9513)(-0.4369)(2.5487)(2.4211)(4.1831)(2.4773)
(CItran)2-0.00270.00550.00680.0147⁎⁎⁎0.0188⁎⁎⁎0.0216⁎⁎⁎
Empty Cell(-0.5885)(1.1972)(1.5335)(4.0927)(5.5883)(6.3704)
BM0.00380.00230.00180.0159⁎⁎⁎0.0252⁎⁎⁎0.0361⁎⁎⁎
Empty Cell(0.6539)(0.3896)(0.3191)(3.5400)(6.0104)(8.5103)
LEV-0.0020-0.0015-0.0003-0.0071-0.0181⁎⁎⁎-0.0291⁎⁎⁎
Empty Cell(-0.3543)(-0.2562)(-0.0620)(-1.6163)(-4.4113)(-7.0223)
ROA0.00390.0094⁎⁎⁎0.0075⁎⁎⁎0.0135⁎⁎⁎0.0226⁎⁎⁎0.0310⁎⁎⁎
Empty Cell(1.3853)(3.3405)(2.7787)(6.1684)(11.0699)(15.0608)
SIZE-0.00130.00430.00260.0205⁎⁎⁎0.0180⁎⁎⁎0.0305⁎⁎⁎
Empty Cell(-0.3575)(1.1475)(0.7205)(7.0723)(6.6408)(11.1139)
UE0.0075⁎⁎⁎0.0060⁎⁎0.00330.0155⁎⁎⁎0.0188⁎⁎⁎0.0279⁎⁎⁎
Empty Cell(3.0154)(2.4220)(1.3777)(8.0597)(10.4218)(15.3445)
Volatility-0.0133⁎⁎⁎-0.0189⁎⁎⁎-0.0196⁎⁎⁎0.00250.0050⁎⁎0.0246⁎⁎⁎
Empty Cell(-5.0199)(-7.1374)(-7.7343)(1.1939)(2.5744)(12.6508)
Constant-0.0829⁎⁎⁎-0.0621⁎⁎⁎-0.2214⁎⁎⁎-0.6389⁎⁎⁎-0.5730⁎⁎⁎-0.4554⁎⁎⁎
Empty Cell(-4.5208)(-3.3847)(-12.5995)(-44.8737)(-43.0077)(-33.8873)
Time effectIncludedIncludedIncludedIncludedIncludedIncluded
Industry effectIncludedIncludedIncludedIncludedIncludedIncluded
N117068117068117068117068117068117068
Note: This table presents regression results of three sub-indicators of climate risk attention and their squared terms on cumulative abnormal returns (CAR) across different event windows. CIphy denotes Physical Climate Risk Attention, CItran denotes Transition Climate Risk Attention, and CIopp denotes Climate Opportunity Attention. Squared terms capture potential nonlinear marginal effects. Control variables include the book-to-market ratio (BM), return on assets (ROA), firm size (SIZE), leverage (LEV), unexpected earnings (UE), and stock return volatility (Volatility). All regressions include industry and year fixed effects and are estimated using the Huber robust method.*, **, and ⁎⁎⁎ indicate significance at the 10%, 5%, and 1% levels, respectively.
These findings are consistent with recent studies such as Bua et al. (2024) and Lin & Wu (2023), which also document significant market responses to transition risk narratives. However, by explicitly incorporating nonlinear terms, our framework extends prior work by uncovering threshold and saturation effects—dynamics that are typically missed in linear models. This refinement improves both the interpretability and empirical robustness of our estimates.
The CIphy shows a similar but weaker nonlinear effect, emerging primarily over longer time horizons. By contrast, CIopp exhibit limited and statistically insignificant effects, suggesting that markets may underreact to climate opportunity narratives. We assess robustness using quantile regressions, propensity score matching (PSM), and 30 iterations of dictionary perturbation, where 20% of keywords are randomly removed in each round. As shown in Appendix I, the inverted U-shape persists across estimation methods, subsamples, and index variants—particularly for transition risk.
Overall, the results confirm that firm-level climate attention significantly and nonlinearly influences stock performance. Among the sub-indices, transition risk attention emerges as the most influential driver of market reactions. These findings underscore the growing relevance of climate-related communication in shaping investor behavior across different temporal horizons.

3.2. Heterogeneity in Nonlinear Market Response

We assess the heterogeneity of market responses to climate attention across four dimensions: ownership structure, policy regime, industry carbon intensity, and geographic region. This subsection focuses on ownership effects, while other dimensions are discussed in Appendices J–L.
We conduct subgroup regressions for state-owned enterprises (SOEs) and non-state-owned enterprises (non-SOEs) and visualize the differential nonlinear effects in Figure 1. Complete estimation outputs are reported in Appendix M.
Figure 1
  1. Download: Download high-res image (489KB)
  2. Download: Download full-size image

Figure 1. Annual Evolution of Corporate Climate Risk Attention Indicators

Note: Panel (a) shows Aggregate Climate Attention (CI); panel (b) presents Climate Opportunity Attention (CIopp); panel (c) displays Physical Climate Risk Attention (CIphy); and panel (d) illustrates Transition Climate Risk Attention (CItran). All indicators are standardized annual averages, reflecting the evolution of disclosure intensity and structural changes in corporate climate-related information between 2013 and 2023.
For non-SOEs, the linear term of CI is significantly negative across multiple windows, while the squared term turns significantly positive over medium- to long-term horizons, forming a robust inverted U-shaped pattern. These results suggest that moderate disclosure intensity may be perceived as insufficient or symbolic, whereas more intensive disclosure is interpreted as a signal of credible climate strategy.
In contrast, SOEs exhibit limited short-term responsiveness: both the linear and squared terms are statistically insignificant in early windows, with nonlinear patterns emerging only over longer horizons. This lagged effect may stem from institutional buffers—such as policy alignment or preferential capital access—that temporarily shield SOEs from market pressure.
Sub-index results further confirm this divergence. For non-SOEs, CItran shows a consistently negative linear term across most windows, while its squared term becomes significantly positive in medium-term windows—highlighting investor focus on firms’ responses to transition risks. CIphy follows a similar pattern with delayed effects, whereas CIopp remains largely insignificant, suggesting that investors discount opportunity-oriented narratives in this group.
For SOEs, only CItran yields consistent significance, with its linear term remaining negative and its squared term turning positive over longer horizons. In contrast, CIphy and CIopp do not produce statistically significant results, implying limited investor differentiation regarding physical and opportunity risks in state-owned firms.
Ownership structure shapes investor responses to climate disclosures, reflecting institutional and strategic differences between SOEs and non-SOEs. Non-SOEs face quicker and more nonlinear reactions under greater scrutiny, while SOEs appear buffered by policy alignment.

4. Conclusion and Policy Implications

This study examines how firm-level climate disclosures in textual form influence stock returns in China’s policy-driven, carbon-intensive market. Based on over 117,000 Chinese-language earnings calls and broker reports, we develop four climate attention indices—Aggregate, Physical Risk, Transition Risk, and Climate Opportunity—using a localized, supervised keyword approach.
The analysis reveals a robust inverted-U-shaped relationship between climate attention and abnormal stock returns, particularly over medium to long horizons. Among the four indices, transition risk attention generates the strongest and most persistent effects, while opportunity-related narratives appear systematically underreacted to by the market.
Heterogeneous effects are also evident. Non-SOEs experience more immediate and pronounced market responses, likely due to greater investor scrutiny, whereas SOEs show delayed effects, reflecting potential policy alignment. The responses are further amplified in high-emission sectors, southern provinces, and after China’s carbon neutrality pledge.
By explicitly modeling nonlinear effects, this study extends prior linear approaches (e.g., Chen et al., 2024; Ma et al., 2025) by uncovering threshold responses and marginal saturation—patterns especially relevant in LMICs with fragmented institutional environments.
Beyond academic contributions, the findings yield actionable implications. Economically, the nonlinear pattern emphasizes the need for balanced and credible disclosures. Managerially, firms should calibrate the timing and framing of climate communication. Practically, the four-index framework enables investors and regulators to better differentiate climate-related risks and opportunities, thereby supporting informed and sustainable decision-making.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
NextGen Digital... Welcome to WhatsApp chat
Howdy! How can we help you today?
Type here...