Climate Risk Perception and Firms’ Energy Productivity: Evidence from China

Abstract

Whether firms translate climate risk perception into energy-related operational productivity remains unclear. Panel data on non-financial Chinese firms (2012–2023) are used to examine the association between climate risk perception (CRP) and energy productivity (EE). Firm-level CRP is constructed from management discussion and analysis (MD&A) sections using a term frequency–inverse document frequency (TF–IDF)-weighted, Word2Vec-expanded climate-risk lexicon. Energy productivity (EE) is measured as the natural logarithm of operating revenue per total energy consumption unit converted into tons of coal equivalent, capturing the economic value generated per energy input unit. Two-way fixed-effects models with firm-level clustered standard errors show a positive CRP–EE association. Digital transformation, proxied by an annual report text-based index across five digital technology domains, partially mediates this association, which is stronger when analyst coverage is higher and weaker when financing constraints are more severe. The results are robust to an alternative CRP proxy based on raw keyword frequency, dynamic specifications, and an instrumental-variable approach exploiting province-year extreme-weather exposure (share of days meeting extreme temperature or precipitation thresholds), using leave-one-province-out aggregation as the instrument and systematic heterogeneity across state ownership, pollution intensity, and high-tech status. This study extends CRP research from disclosure-oriented to energy-productivity outcomes, and highlights how digital capabilities, information scrutiny, and financial friction shape climate-aware energy productivity improvements.

1. Introduction

Climate risk has shifted from a distant externality to first-order uncertainty, shaping firms’ cash flows, costs, and investment opportunities. Asset-pricing evidence suggests that markets increasingly treat climate-related exposure as economically material. Carbon-intensive firms face a priced carbon risk premium, and investors demand compensation for downside tail risks associated with climate transition and regulation. Bolton and Kacperczyk [1] show that carbon risk is priced in the cross-section of stock returns, while Ilhan, Sautner, and Vilkov [2] highlight the relevance of carbon tail risk. Consistent with this shift, Krueger, Sautner, and Starks [3] report that institutional investors view climate risk as financially important, reinforcing the perspective that corporate responses to climate risk are no longer discretionary add-ons but increasingly central to strategy.

However, an important gap remains regarding the connection between climate risk and firms’ operational choices. A large body of empirical literature examines climate risk through market pricing, disclosure behavior, or innovation outcomes, leaving the operational channel comparatively underexplored. Recent research has begun to quantify firms’ climate exposure and managerial attention using corporate text and link these measures to real decisions. For example, Sautner et al. [4] develop a firm-level climate change exposure measure from earnings conference-call language and show that climate exposure predicts meaningful corporate policies and outcomes. In the Chinese context, Tian, Chen, and Dai [5] construct a text-based climate risk perception measure from management discussion and analysis (MD&A) disclosures and find that higher perceived climate risk is associated with stronger corporate green innovation. These advances support the feasibility and relevance of text-based climate measures; however, they do not directly answer a more operational question: when managers perceive climate risk more acutely, do they improve energy-related operational productivity, that is, the economic value generated per unit of energy input, or do climate narratives remain only loosely connected to operational upgrading?

We address this question by focusing on corporate energy productivity (EE) as a concrete manifestation of strategic adaptation. Energy productivity is not only an environmental metric but also an operations- and profitability-relevant outcome that captures the economic value generated per unit of energy input, reflecting process discipline, energy management, and technology adoption. Prior research has also examined energy efficiency and productivity at the firm level from the perspectives of operational management, technological upgrading, and competitive performance. Empirical studies document a positive relationship between energy efficiency and firm productivity across countries and industries [6,7]. From a capability perspective, technological innovation and knowledge spillovers have been shown to improve energy efficiency outcomes and reduce energy intensity [8]. At a broader level, energy efficiency improvements are also associated with stronger economic performance and growth [9]. This literature suggests that energy productivity is closely linked to firms’ operational capabilities and competitive outcomes, making it a meaningful lens through which to examine the operational implications of climate risk perception. Moreover, firms may leave substantial energy-related productivity potential unrealized because of managerial frictions and imperfect internal incentives, implying that attention and capabilities can help close the “energy–management gap” [10]. Building on this insight, we argue that heightened climate risk perception can act as an attentional shock that increases the salience of transition risks and energy-related cost exposures, thereby strengthening incentives to increase the economic value generated per unit of energy input, tighten operating processes, and adopt practices that improve energy-related operational productivity [1,2]. This logic yields a testable prediction at the firm-year level: climate risk perception is positively associated with energy productivity.

We further examine how this association is realized. Digital transformation provides a plausible capability channel through which climate-related attention is converted into operational improvement. Digital technologies can support energy productivity by enabling real-time monitoring, predictive maintenance, process optimization, and better coordination across production systems. These applications can reduce energy waste and improve output reliability and quality, thereby increasing the economic value generated per unit of energy input [11]. Firm digitalization can complement environmental upgrading by reducing information frictions and improving the effectiveness of green investments. Fang and Li [12] show that corporate digitalization facilitates green innovation among Chinese listed firms, which is consistent with the notion that digital capability strengthens a firm’s capacity to respond substantively to environmental challenges. Accordingly, we expect digital transformation to partly mediate the climate risk perception–energy productivity association. Managers’ heightened awareness of climate risk is more likely to translate into measurable energy productivity gains when accompanied by capability-building and data-driven process upgrading.

We also consider two boundary conditions related to information and financing frictions. A tight information environment can discipline managerial action and increase the cost of inaction. Analysts play a central monitoring and information intermediary role. Han, Kong, and Liu [13] show that analysts can gain informational advantages through direct corporate visits, consistent with deeper scrutiny and stronger external oversight. If analyst coverage strengthens monitoring and reduces information frictions, it should amplify the extent to which climate risk perception translates into operational improvements. By contrast, financing constraints can prevent firms from undertaking energy-productivity-enhancing investments that require upfront capital, even when managers recognize climate risks. We use the Whited–Wu index [14] as a proxy for financing constraints, and expect such constraints to attenuate the positive link between climate risk perception and energy productivity.

We empirically test these arguments using Chinese non-financial A-share listed firms from 2012 to 2023. We construct firm-level climate risk perceptions from MD&A text using a TF–IDF-weighted climate-risk lexicon [15] expanded via Word2Vec, and measure energy productivity as the log ratio of operating revenue to total energy consumption converted into standard coal equivalent. We mitigate concerns that unobserved firm traits jointly drive climate-related language and operating outcomes by adopting stringent fixed-effects specifications, dynamic formulations, and an instrumental variable approach that leverages province-year extreme weather exposure based on meteorological thresholds. We also examine heterogeneity across state ownership, pollution intensity, and high-tech status. This study contributes to the literature by extending text-based climate risk research from disclosure-oriented outcomes to energy-related operational productivity, clarifying the role of digital capability as a channel of strategic adaptation, and documenting how information scrutiny and financing frictions shape the translation of climate risk perception into energy productivity improvements.

2. Hypothesis Development

2.1. Climate Risk Perception and Energy Productivity

We conceptualize climate risk perception (CRP) as managerial attention to climate-related uncertainty embedded in a firm’s forward-looking narratives. The attention-based view implies that what decision-makers attend to shapes resource allocation and operational priorities [16].

Linking attention to real outcomes requires evidence that climate risk is not only “talk” but is also associated with concrete firm actions. Text-based climate exposure measures derived from corporate disclosures have been shown to predict substantive corporate policies and outcomes, consistent with the firm-level economic and operational relevance of climate risk [4]. Furthermore, climate-policy shocks generate real effects across firms, and financial frictions shape how strongly these effects propagate, highlighting a transmission channel that is operational and not merely based on disclosure [17].

Our dependent variable, EE, captures firms’ energy-related operational performance using a revenue-based measure: operating revenue per unit of total energy consumption (in tons of coal equivalent, tce). Accordingly, EE should be interpreted as energy productivity (i.e., the economic value generated per unit of energy input) rather than as a purely physical energy-use efficiency measure. The numerator is revenue; therefore, this proxy may also reflect pricing and product-mix variations that can affect measured productivity outcomes. Thus, it should be understood as an energy-related economic performance indicator rather than a purely engineering efficiency metric [18]. Nevertheless, EE remains a meaningful operational endpoint that is typically shaped by managerial routines, monitoring, and process control rather than purely symbolic responses. Prior evidence documents an “energy–management gap,” in which cost-effective improvements in energy-related performance are left unrealized owing to organizational and managerial frictions [10]. Recent firm-level studies further show that improvements in energy efficiency are closely associated with higher productivity, stronger competitiveness, and better economic performance across industries and countries [6,7,8,9]. If CRP elevates the internal salience of transition risk and energy cost exposure, it should be associated with managerial actions that increase the revenue generated per unit of energy input (energy productivity), although this revenue-based measure may partially reflect pricing and product mix variations.

H1. 

Climate risk perception is positively associated with corporate energy productivity.

2.2. Moderator: Analyst Coverage

Analyst coverage is a central element of the external information environment. Higher coverage is associated with stronger external discipline and reduced managerial opportunism, which is consistent with a monitoring channel [19]. Analysts also acquire information through deeper engagement, such as corporate site visits, which can enhance scrutiny and improve information efficiency [13]. In our setting, the monitoring implication is straightforward. When analyst scrutiny is higher, managerial attention revealed by CRP is more likely to be evaluated against observable operating outcomes, thereby strengthening the association between CRP and EE. We recognize a competing force in which analyst pressure can discourage long-horizon innovation investment and induce managerial myopia [20]. However, because energy productivity can be improved through operational tightening and energy management actions with relatively near-term cost and risk benefits, we expect the monitoring channel to dominate in shaping the CRP–EE relationship.

H2. 

Analyst coverage strengthens the positive association between climate risk perception and corporate energy productivity.

2.3. Moderator: Financing Constraints

Energy productivity upgrading often requires upfront investments (e.g., equipment, process redesign, and measurement systems), implying that financial frictions can limit implementation, even when climate-related attention is present. We proxy financing constraints using the Whited–Wu (WW) index, which is a standard measure of external finance constraints in corporate finance [14].

Evidence further supports an “investment feasibility” mechanism for energy-related upgrading. In China, the Green Credit Guidelines improve firms’ energy-related performance, and financing conditions are integral to the policy’s effectiveness, suggesting that capital availability conditions firms’ abilities to undertake upgrades and operational changes that improve energy use and management [21]. More broadly, evidence from climate policy indicates that real effects operate through financial constraints and spillovers, highlighting that constrained firms have tighter implementation capacities when responding to climate-related regulatory pressures [17]. Accordingly, financing constraints should weaken the extent to which CRP is associated with higher energy productivity.

H3. 

Financing constraints weaken the positive association between climate risk perception and corporate energy productivity.

2.4. Mediator: Digital Transformation

We focus on digital transformation (DT) as a capability channel to make the CRP → EE link operationally “mechanistic.” Digital technologies have been linked to productivity and operational performance at the firm level [11]. In our setting, DT is expected to support higher energy productivity because digital capability enhances measurement, coordination, and process control. These improvements can reduce energy waste and downtime and improve output reliability and quality, thereby increasing the economic value generated per unit of energy input. Consistent with this channel, domain-specific evidence from Chinese listed firms indicates that corporate digital technology adoption can reduce energy intensity and improve energy-related operational performance [22], potentially translating into higher revenue per unit of energy input.

For the “CRP → DT” leg, we avoid over-claiming: external evidence indicates that climate-risk pressure/exposure can motivate firms’ digital upgrading [23]. Consistent with this view, we posit that a more highly perceived climate risk is associated with stronger DT, which supports energy productivity improvement in turn. EE can also be improved through non-digital routes; therefore, DT is expected to be a partial mediating mechanism.

H4. 

Digital transformation partially mediates the positive association between climate risk perception and corporate energy productivity.

3. Variable Description and Methodology

3.1. Data Sources

We construct a firm-year panel of Chinese non-financial A-share listed companies from 2012 to 2023. MD&A texts are obtained from the CSMAR annual report text repository, from which we extract the MD&A sections of reports to build text-based measures. Firm-level accounting and governance variables are sourced from CSMAR. Energy consumption data (e.g., coal, oil products, natural gas, and electricity) are collected from firms’ annual report disclosures, as compiled by CSMAR, and converted into their standard coal equivalent (tce) using national conversion coefficients (GB/T 2589-2020) [24]. The meteorological data used to design the instrumental variable are obtained from the China National Meteorological Information Center and aggregated at the province-by-year level. Following standard practices in Chinese capital market studies, we exclude financial firms based on CSMAR industry classification, special treatment (*ST/ST) firms, and observations with missing values required for baseline regressions. Continuous variables are winsorized at the 1st and 99th percentiles unless otherwise stated.

3.2. Variable Definitions

3.2.1. Dependent Variable: EE

Our dependent variable is EE, measured as the natural logarithm of the revenue generated per unit of energy consumption in standard coal equivalents:

$$E{E}_{i,t}=\ln \left({\displaystyle \frac{Operating Revenu{e}_{i,t}}{Energy Consumptio{n}_{i,t}^{tce}}}\right)$$

EE captures an operational outcome closely tied to the production discipline and resource utilization. Evidence of an “energy-management gap” motivates EE as an outcome sensitive to managerial routines and organizational friction [10]. Physical energy quantities are converted to tce and summed before forming the ratio to ensure comparability across energy types. The conversion of different energy sources into standard coal equivalent and the detailed construction of the EE variable are described in Appendix B.

3.2.2. Independent Variable: Climate Risk Perception (CRP)

We measure climate risk perception (CRP) using firms’ MD&A texts, following disclosure-based textual measurements in finance [4,25]. We train Word2Vec on the full MD&A corpus to expand a seed lexicon for the “climate risk” semantic domain. We then construct CRP as a TF–IDF-weighted intensity measure based on the final lexicon (113 terms, expanded from 37 seed terms), scaled by document length, and standardized within each year (yearly z-score). As a robustness check, we compute an alternative CRP measure using unweighted term frequencies. The seed dictionary and expanded climate-risk terms are reported in Appendix A.

3.2.3. Moderating Variables

Analyst coverage (AnaRep). Analyst coverage is measured as the annual number of analyst research reports issued by a firm, transformed to $\ln \left(1+\mathrm{reports}\right)$. Analyst coverage is a canonical proxy for the external information environment and monitoring intensity in capital markets [19].

Financing constraints (WW). The Whited–Wu (WW) index, a standard firm-level measure widely used in empirical corporate finance [14], is used as a proxy for financing constraints.

3.2.4. Mediating Variable: DT

DT is constructed from annual report text as:

$$D{T}_{i,t}=\ln \left(1+{\sum }_{k=1}^{5}Fre{q}_{k,i,t}\right),$$

where the five dimensions are artificial intelligence, blockchain, cloud computing, big data, and digital platforms. This text-based approach follows established practices for measuring firm digitalization from corporate disclosures [12]. Conceptually, we interpret DT as an operational capability that can strengthen measurement, coordination, and process control, thereby supporting energy-related operational productivity, that is, improving the economic value generated per unit of energy input, rather than implying a purely engineering-based notion of physical energy use efficiency [11]. The digital transformation dictionary and the detailed construction of the text-based digital transformation index are described in Appendix C.

3.2.5. Control Variables

We include a standard set of control variables commonly used in corporate finance and governance studies: board size (Boardsize), independent director ratio (Indep, measured on a 0–100 scale), revenue growth (Growth, computed from operating revenue), return on assets (ROA), and leverage (Lev). These control variables capture governance structure, growth opportunities, profitability, and capital structure, which are plausibly related to both managers’ climate-related attention (as reflected in the MD&A language) and firms’ EE [26,27].

3.3. Model Specification

3.3.1. Baseline Specification

We estimate the association between climate risk perception and energy productivity using a two-way fixed-effects framework:

$$E{E}_{i,t}=\mathrm{\beta }\hspace{0.17em}CR{P}_{i,t}+{\mathrm{\theta }}^{\mathrm{\prime }}{X}_{i,t}+{\mathrm{\mu }}_i+{\mathrm{\lambda }}_t+{\mathrm{\epsilon }}_{i,t}$$

(1)

where ${\mathrm{\mu }}_i$ are firm fixed effects and ${\mathrm{\lambda }}_t$ are year fixed effects. Standard errors are clustered at the firm level. This design absorbs time-invariant firm heterogeneity and common macro-shocks and is standard in panel corporate finance applications [28]. As a robustness check, we also consider two-way clustering by firm and year, following the finance econometrics literature [29].

3.3.2. Moderation Tests

To test boundary conditions, we estimate interaction models as follows:

$$E{E}_{i,t}={\mathrm{\beta }}_{1}CR{P}_{i,t}+{\mathrm{\beta }}_2{M}_{i,t}+{\mathrm{\beta }}_3\left(CR{P}_{i,t}\times M_{i,t}\right)+{\mathrm{\theta }}^{\mathrm{\prime }}{X}_{i,t}+{\mathrm{\mu }}_i+{\mathrm{\lambda }}_t+{\mathrm{\epsilon }}_{i,t}$$

(2)

where $M_{i,t}$ is either analyst coverage (AnaRep) or financing constraints (WW). The coefficient ${\mathrm{\beta }}_3$ captures whether the CRP–EE association varies with the information environment or financial frictions.

3.3.3. Mediation Test

We examine the mediating role of DT using a three-model regression approach with the same fixed-effects structure as that in the baseline specification. First, we estimate whether climate risk perception predicts DT:

$$D{T}_{i,t}=a\hspace{0.17em}CR{P}_{i,t}+{\mathrm{\varphi }}^{\mathrm{\prime }}{X}_{i,t}+{\mathrm{\mu }}_i+{\mathrm{\lambda }}_t+u_{i,t}$$

(3)

Second, we estimate whether DT is associated with energy productivity:

$$E{E}_{i,t}=b\hspace{0.17em}D{T}_{i,t}+{\mathrm{\theta }}^{\mathrm{\prime }}{X}_{i,t}+{\mathrm{\mu }}_i+{\mathrm{\lambda }}_t+v_{i,t}$$

(4)

Third, we include both CRP and DT in the EE equation:

$$E{E}_{i,t}=c^{\mathrm{\prime }}\hspace{0.17em}CR{P}_{i,t}+b^{\mathrm{\prime }}\hspace{0.17em}D{T}_{i,t}+{\mathrm{\theta }}^{\mathrm{\prime }}{X}_{i,t}+{\mathrm{\mu }}_i+{\mathrm{\lambda }}_t+{\mathrm{\epsilon }}_{i,t}$$

(5)

Evidence of mediation requires that CRP significantly predicts DT in Equation (3) and that DT is significantly associated with EE in Equation (4). We interpret mediation as partial if, in Equation (5), DT remains statistically significant while CRP remains statistically different from zero, indicating that DT partially explains the association between CRP and EE.

3.3.4. Dynamic Specifications and Identification

We implement dynamic specifications that relate the current CRP to future energy productivity and use lagged CRP as an alternative timing structure to mitigate concerns regarding timing and reverse causality. Furthermore, we adopt an instrumental-variable strategy, in which CRP is instrumented by province-by-year extreme weather exposure constructed from meteorological thresholds, to strengthen causal interpretations. Weather-based measures are commonly treated as plausibly exogenous shocks in climate-economy research [30], and climate-policy real effects are known to propagate through financial friction [17]. In the first stage, we include the same fixed-effects structure as in the baseline model. In the second stage, we estimate the fitted CRP effect on EE.

Table 1. Variable names and definitions.

Variable Type	Variable	Symbol	Definition/Measurement
Dependent variable	Energy productivity	EE	Natural logarithm of the ratio of operating revenue to total energy consumption in tons of standard coal equivalent (tce):$E{E}_{i,t}=\ln \left({\displaystyle \frac{Operating Revenu{e}_{i,t}}{Energy Consumptio{n}_{i,t}^{tce}}}\right)$$\mathrm{Operating} \mathrm{revenue} \mathrm{is} \mathrm{obtained} \mathrm{from} \mathrm{CSMAR}. Energy Consumptio{n}_{i,t}^{tce}$ is constructed by converting firm-disclosed physical consumption of coal, oil products, natural gas, and electricity into tce using China’s standard coal equivalent conversion coefficients and then summing across energy types.
Independent variable	Climate risk perception	CRP	Text-based climate risk perception index constructed from MD&A disclosures using a Word2Vec-expanded climate-risk lexicon and TF–IDF weighting; the resulting index is standardized by year (yearly z-score).
Moderating variable	Analyst report attention	AnaRep	$\mathrm{Analyst} \mathrm{report} \mathrm{count} \mathrm{for} \mathrm{a} \mathrm{firm}-\mathrm{year}, \mathrm{measured} \mathrm{as} \ln \left(1+\mathrm{Analyst} {\mathrm{Reports}}_{it}\right)$, where reports are issued by analysts covering the focal firm; data from CSMAR.
	Financing constraints	WW	Whited–Wu (WW) financial constraints index.
Mediating variable	Digital transformation	DT	$\mathrm{Digital} \mathrm{transformation} \mathrm{index} \mathrm{based} \mathrm{on} \mathrm{annual} \mathrm{report} \mathrm{text} \mathrm{mining} \mathrm{from} \mathrm{CSMAR}: D{T}_{it}=\ln \left(1+{\sum }_{k=1}^{5}Fre{q}_{k,it}\right),$ where the five dimensions cover artificial intelligence, blockchain, cloud computing, big data, and digital-technology applications (keyword frequencies summed across dimensions).
Control variable	Revenue growth	Growth	$\mathrm{Operating} \mathrm{revenue} \mathrm{growth} \mathrm{rate} \mathrm{based} \mathrm{on} \mathrm{CSMAR} \mathrm{operating} \mathrm{revenue} \left(\mathrm{winsorized}\right): \left(O{R}_{it}-O{R}_{i,t-1}\right)/O{R}_{i,t-1}$.
	Board size	Boardsize	Number of directors on the board (CSMAR).
	Independent director ratio	Indep	Percentage of independent directors on the board (0–100), from CSMAR.
	Return on assets	ROA	Current-period return on assets: net income divided by total assets.
	Leverage	Lev	Total liabilities divided by total assets.

is the variable names and definitions.

4. Results

4.1. Descriptive Statistics

Table 2. Summary statistics.

	N	Mean	SD	Min	P50	Max
EE	10,226	0.568	0.609	0.014	0.347	3.017
CRP	11,796	0.008	1.07	−0.536	−0.42	5.278
AnaRep	11,796	2.34	1.53	0	2.565	5.677
WW	11,796	−0.986	0.287	−1.259	−1.059	0
DT	11,796	1.579	1.407	0	1.386	6.306
Growth	11,796	0.148	0.347	−0.492	0.093	2.054
Board size	11,796	8.967	1.857	3	9	18
Ind. director ratio	11,796	37.642	5.849	16.67	36.36	80
ROA	11,796	0.048	0.072	−0.645	0.04	0.969
Leverage	11,796	0.474	0.2	0.008	0.486	0.997

summarizes the sample characteristics of the key variables. The full panel comprises 11,796 firm-year observations. However, EE can be computed for 10,226 observations because some firms do not report energy consumption information. All continuous variables are trimmed by winsorizing at the 1st and 99th percentiles to limit the impact of extreme values. EE varies markedly across firms and over time, consistent with substantial heterogeneity in firms’ operating environments and energy-related operational management. By construction, CRP is standardized within each year and thus centers around zero, whereas the other variables display distributions commonly observed in Chinese A-share firm-level panel data.

4.2. Correlation Analysis

Table 3. Correlation matrix.

	EE	CRP	AnaRep	WW	DT	Growth	Board Size	Ind. Director Ratio	ROA	Leverage
EE	1
CRP	0.112 ***	1
AnaRep	0.326 ***	−0.056 ***	1
WW	−0.140 ***	−0.039 ***	−0.039 ***	1
DT	0.050 ***	−0.120 ***	0.134 ***	0.009	1
Growth	0.044 ***	0.025 **	0.156 ***	−0.103 ***	0.044 ***	1
Board size	0.147 ***	0.101 ***	0.066 ***	−0.061 ***	−0.092 ***	−0.029 ***	1
Ind. director ratio	0.107 ***	−0.060 ***	0.034 ***	−0.005	0.054 ***	−0.006	−0.399 ***	1
ROA	0.007	−0.059 ***	0.410 ***	0.014	0.031 ***	0.267 ***	−0.009	0.002	1
Leverage	0.408 ***	0.170 ***	−0.088 ***	−0.126 ***	−0.066 ***	0.009	0.099 ***	0.029 ***	−0.399 ***	1

**, *** indicate significance at the 5%, and 1% levels.

presents the Pearson correlation coefficients computed using the regression sample with listwise deletions. Several patterns are noteworthy. First, EE is positively correlated with CRP (0.112) and analyst report attention (AnaRep; 0.326) and negatively correlated with financing constraints (WW; −0.140). EE is positively correlated with DT (0.050). Second, CRP is weakly negatively correlated with AnaRep (−0.056) and WW (−0.039), but positively correlated with leverage (0.170) and board size (0.101). Third, AnaRep is positively correlated with ROA (0.410), indicating a weak-to-moderate association, consistent with analysts concentrating on more profitable firms. Overall, the correlation magnitudes are generally weak to moderate, suggesting that multicollinearity is unlikely to be a first-order concern. Nevertheless, the main inferences rely on the multivariate fixed-effects specifications reported below.

4.3. Baseline Regressions

Table 4. Baseline results.

	(1)	(2)	(3)
Variables	EE	EE	EE
CRP	0.025 ***	0.023 ***	0.031 ***
	(0.008)	(0.008)	(0.011)
CRP × AnaRep		0.005 **
		(0.002)
AnaRep		0.044 ***
		(0.004)
CRP×WW			−0.129 ***
			(0.037)
WW			−0.093 ***
			(0.015)
Board size	0.020 ***	0.019 ***	0.020 ***
	(0.005)	(0.005)	(0.005)
Growth	0.064 ***	0.057 ***	0.059 ***
	(0.007)	(0.007)	(0.007)
Ind. director ratio	0.002 *	0.002 *	0.002 *
	(0.001)	(0.001)	(0.001)
ROA	0.663 ***	0.466 ***	0.637 ***
	(0.062)	(0.055)	(0.062)
Leverage	0.398 ***	0.385 ***	0.379 ***
	(0.048)	(0.044)	(0.047)
Firm FE	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
Observations	10,226	10,226	10,226
Adjusted R2	0.929	0.922	0.930
Robust standard errors are in parentheses Standard errors clustered at the firm level are in parentheses.

*** p < 0.01, ** p < 0.05, * p < 0.1.

summarizes the two-way fixed-effects results linking climate risk perception (CRP) to EE. Firm- and year-fixed effects are included in every specification, and inferences are based on firm-clustered standard errors. In column (1), CRP has a positive and statistically significant coefficient (0.025, p < 0.01), indicating that firms expressing stronger climate risk perception in their MD&A narratives tend to exhibit higher energy productivity. This positive CRP–EE relationship remains evident after adding the interaction terms in columns (2) and (3).

Column (2) evaluates whether the information environment conditions this relationship by interacting CRP with analyst report attention (AnaRep). The interaction coefficient, CRP × AnaRep, is positive and significant (0.005, p < 0.05), indicating that the CRP–EE association is stronger when analyst scrutiny is more intensive. AnaRep also loads positively on EE (0.044, p < 0.01), which is consistent with greater analyst attention being associated with higher energy-related operational productivity, measured as the operating revenue generated per unit of energy input.

Column (3) examines whether financing constraints attenuate the association between CRP and EE. The interaction term CRP × WW is negative and highly significant (−0.129, p < 0.01), indicating that the positive relationship between climate risk perception and energy productivity is weaker among financially constrained firms. The main effect of WW is negative (−0.093, p < 0.01), which is consistent with financing frictions being associated with lower energy productivity.

Across all columns, the control variables display economically plausible signs: board size, growth, profitability (ROA), and leverage are positively associated with EE, whereas the independent director ratio is weakly positive. The adjusted R-squared values are high, reflecting the strong explanatory power of firm- and year-fixed effects for within-firm variation in energy productivity. Overall, the baseline results provide consistent evidence of a positive relationship between climate risk perception and operational energy productivity, amplified by analyst scrutiny but dampened by financing constraints.

4.4. Robustness Checks

4.4.1. Alternative Indicators

Table 5. Robustness check results: Alternative indicators.

	Alternative CRP			Alternative Dependent Variable
	(1)	(2)	(3)	(4)
Variables	EE	EE	EE	EI
CRPO	3.637 **	3.052 **	2.41
	(1.558)	(1.531)	(1.523)
CRP				−0.099 **
				(0.036)
CRPO × AnaRep		0.968 **
		(0.424)
AnaRep		0.044 ***
		(0.004)
CRPO × WW			−17.261 ***
			(5.135)
WW			−0.088 ***
			(0.014)
Board size	0.020 ***	0.019 ***	0.020 ***	−0.051 ***
	(0.005)	(0.005)	(0.005)	(0.011)
Growth	0.064 ***	0.057 ***	0.020 ***	−0.226 ***
	(0.007)	(0.007)	(0.005)	(0.026)
Ind. director ratio	0.002 *	0.002 *	0.002 *	−0.005 *
	(0.001)	(0.001)	(0.001)	(0.002)
ROA	0.667 ***	0.469 ***	0.647 ***	−2.032 ***
	(0.062)	(0.054)	(0.062)	(0.224)
Leverage	0.399 ***	0.386 ***	0.379 ***	−1.524 ***
	(0.048)	(0.044)	(0.047)	(0.160)
Firm FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
Observations	10,226	10,226	10,226	10,226
Adjusted R2	0.928	0.932	0.929	0.920
Robust standard errors are in parentheses Standard errors clustered at the firm level are in parentheses. EI denotes energy intensity, defined as energy consumption per unit of operating revenue, which is the inverse measure of energy productivity.

*** p < 0.01, ** p < 0.05, * p < 0.1.

re-examines the main findings using a different construction of climate risk perception. We substitute the TF–IDF-weighted index with CRPO, which is based on the unweighted frequency of climate-risk terms in MD&A texts. This exercise checks whether the baseline results are not an artifact of the TF–IDF weighting procedure but instead reflect the underlying climate-risk-related attention. CRPO is measured using a different scale from that used for the standardized baseline CRP; therefore, the coefficient magnitudes are not directly comparable. Thus, the interpretation focuses on the sign and statistical significance.

In column (1), CRPO is positively related to EE (3.637, p < 0.05), mirroring the baseline direction. Column (2) shows that the interaction CRPO × AnaRep is positive and significant (0.968, p < 0.05), indicating that the CRPO–EE relationship is more pronounced under stronger analyst scrutiny. Column (3) reports a negative interaction between CRPO and financing constraints (CRPO × WW = −17.261, p < 0.01). Although the average coefficient of CRPO is not statistically different from zero in column (3), the significant interaction implies that the positive association is concentrated among firms with looser financial constraints and diminishes as the constraints tighten.

Column (4) further replaces the dependent variable with energy intensity (EI), defined as energy consumption per unit of operating revenue, which is the inverse measure of energy productivity. The coefficient on CRP remains statistically significant and with the expected opposite sign (−0.099, p < 0.05), indicating that higher climate risk perception is associated with lower energy intensity. This result is consistent with the baseline findings and supports the robustness of the main conclusions to an alternative dependent-variable specification.

Overall, the sign patterns for the interaction terms and the direction of the main association are aligned in Table 5, supporting the robustness of our baseline conclusions to alternative constructions of climate risk perception and energy-related performance measures.

4.4.2. Endogeneity Tests

Table 6. Endogeneity test results.

	Reverse Causality		2SLS
Variables	EE	$\mathit{E}{\mathit{E}}_{\mathit{i},\mathit{t}+1}$	First Stage: CRP	Second Stage: EE
$CR{P}_{i,t-1}$	0.032 ***
	(0.011)
CRP		0.029 ***
		(0.010)
EWnat			0.180 ***
			(0.039)
$\mathrm{C}\mathrm{R}\mathrm{P}\mathrm{I}\mathrm{V}$				0.089 *
				(0.048)
Boardsize	0.017 ***	0.013 ***	0.025 ***	0.018 ***
	(0.004)	(0.004)	(0.004)	(0.005)
Growth	0.072 ***	0.068 ***	0.074 ***	0.065 ***
	(0.009)	(0.008)	(0.022)	(0.007)
IndDirectorRatio	0.002 *	0.002	−0.006 ***	0.002 *
	(0.001)	(0.001)	(0.001)	(0.001)
ROA	0.654 ***	0.600 ***	0.353 ***	0.680 ***
	(0.064)	(0.057)	(0.107)	(0.062)
Lev	0.359 ***	0.367 ***	0.740 ***	0.382 ***
	(0.052)	(0.047)	(0.040)	(0.048)
Firm FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
Observations	8615	8615	10,226	10,226
Adjusted R2	0.943	0.942	0.075	0.074
Kleibergen–Paap rk Wald F			18.06
Robust standard errors are in parentheses Standard errors clustered at the firm level are in parentheses.

* p < 0.10, *** p < 0.01.

reports the results of the endogeneity tests addressing concerns regarding reverse causality and omitted variables. Columns (1) and (2) implement dynamic specifications. In column (1), lagged climate risk perception is positively related to current energy productivity ($CR{P}_{i,t-1}$ = 0.032, p < 0.01). In column (2), current climate risk perception predicts next-period energy productivity ($CR{P}_{i,t}$ = 0.029, p < 0.01). These timing patterns are consistent with climate risk perception preceding subsequent energy productivity improvements, as reflected in the higher operating revenue generated per unit of energy input.

Columns (3) and (4) report two-stage least squares estimates of instrument firm-level CRP with leave-one-out extreme-weather exposure (EWnat), defined as the leave-one-out average province-year extreme-weather exposure computed from other provinces (excluding the firm’s own province). This construction is intended to reduce the mechanical links between local contemporaneous shocks and the instrument. In the first stage, EWnat significantly predicts CRP (0.180, p < 0.01), and the Kleibergen–Paap rk Wald F statistic is 18.06, suggesting that the instrument is not weak. In the second stage, the fitted component of CRP remains positively associated with energy productivity (0.089, p < 0.10). Overall, the dynamic specifications and IV results are directionally consistent with the baseline evidence. However, the exclusion restriction cannot be directly verified; therefore, we interpret them as supportive associations rather than definitive causal estimates. The detailed construction of the extreme-weather-based instrumental variable is described in Appendix D.

Firm- and year-fixed effects are included throughout, and standard errors are clustered at the firm level.

5. Heterogeneity Analysis

Table 7. Heterogeneity tests.

	(1)	(2)	(3)	(4)	(5)	(6)
	SOE	NSOE	HP	NHP	High-Tech	Non-High-Tech
CRP	0.044 **	0.039 ***	0.063 ***	0.012	0.051 ***	0.019
	(0.017)	(0.015)	(0.017)	(0.014)	(0.014)	(0.017)
Boardsize	0.016 ***	0.021 ***	0.011	0.025 ***	0.014 ***	0.023 ***
	(0.006)	(0.006)	(0.007)	(0.006)	(0.005)	(0.007)
Growth	0.088 ***	0.039 ***	0.093 ***	0.057 ***	0.064 ***	0.069 ***
	(0.010)	(0.008)	(0.019)	(0.007)	(0.009)	(0.011)
IndDirectorRatio	0.001	0.004 ***	0.003	0.002 *	0.001	0.002 *
	(0.001)	(0.001)	(0.002)	(0.001)	(0.001)	(0.001)
ROA	0.900 ***	0.513 ***	0.593 ***	0.600 ***	0.615 ***	0.637 ***
	(0.107)	(0.064)	(0.104)	(0.065)	(0.064)	(0.096)
Lev	0.401 ***	0.348 ***	0.381 ***	0.410 ***	0.389 ***	0.303 ***
	(0.072)	(0.048)	(0.115)	(0.052)	(0.052)	(0.067)
Firm FE	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Observations	5448	4636	2773	7435	5205	5000
Adjusted R2	0.942	0.908	0.939	0.93	0.929	0.937
Standard errors are in parentheses
Standard errors clustered at the firm level are in parentheses.

* p < 0.10, ** p < 0.05, *** p < 0.01.

examines whether the baseline association between climate risk perception and energy productivity varies across firm ownership, pollution intensity, and technological orientation. We re-estimate the baseline two-way fixed-effects specification within each subsample and cluster standard errors at the firm level.

Columns (1) and (2) show that CRP is positively associated with energy productivity in state-owned enterprises (SOEs) and non-state-owned enterprises (NSOEs). The coefficients are 0.044 (p < 0.05) and 0.039 (p < 0.01) for SOEs and NSOEs, respectively. This pattern suggests that climate-risk-related managerial attention is linked to energy-related operational productivity improvements across ownership types, and this association appears to be more precisely estimated among NSOEs.

Columns (3) and (4) report the subsample results for high-polluting (HP) and non-high-polluting (NHP) industries, respectively, based on the industry pollution classification. The results show pronounced heterogeneity by pollution intensity. In high-polluting industries (HP), CRP exhibits a larger and highly significant coefficient (0.063, p < 0.01). By contrast, the CRP coefficient in non-high-polluting industries (NHP) is small and not statistically significant (0.012). This result is consistent with the notion that when firms face stronger regulatory and transition pressure exposure, which is more typical in high-polluting sectors, climate risk perception is more likely to translate into measurable operational upgrading.

Columns (5) and (6) show that the CRP–EE association is concentrated in high-tech firms. The CRP coefficient is positive and significant in the high-tech subsample (0.051, p < 0.01); however, it becomes statistically non-significant among non-high-tech firms (0.019). This heterogeneity aligns with the concept that firms with stronger technological capability and absorptive capacity are better positioned to convert climate-risk attention into energy-productivity-enhancing operational adjustments.

Overall, the heterogeneity results reinforce the baseline findings while highlighting meaningful cross-sectional variations. The CRP–EE association is systematically stronger in settings where transition pressure and operational stakes are higher (high-polluting sectors) and implementation capacity is likely to be greater (high-tech firms).

6. Mediation Analysis

Table 8. Mediation analysis results.

Dependent Variable	DT	EE	EE
	Mediating Effect
Variables	(1)	(2)	(3)
CRP	0.109 ***		0.038 ***
	(0.031)		(0.010)
DT		0.030 ***	0.029 ***
		(0.005)	(0.005)
Board size	0.030 **	0.019 ***	0.019 ***
	(0.011)	(0.005)	(0.005)
Growth	0.009	0.064 ***	0.063 ***
	(0.023)	(0.007)	(0.007)
Ind. director ratio	−0.003	0.002	0.002 *
	(0.002)	(0.001)	(0.001)
ROA	0.333	0.664 ***	0.648 ***
	(0.217)	(0.081)	(0.078)
Leverage	0.111	0.398 ***	0.396 ***
	(0.129)	(0.051)	(0.051)
Firm FE
Year FE
Observations	10,226	10,226	10,226
Adjusted R2	0.797	0.929	0.930
Standard errors are in parentheses
Standard errors clustered at the firm level are in parentheses.

* p < 0.10, ** p < 0.05, *** p < 0.01.

reports the results of the mediation analysis examining whether DT serves as a channel through which climate risk perception (CRP) is associated with EE. Following the standard sequential approach, we estimate (i) the effect of CRP on DT, (ii) the effect of DT on EE, and (iii) the joint regression including both CRP and DT to assess coefficient attenuation consistent with partial mediation. All specifications include firm- and year-fixed effects, with standard errors clustered at the firm level.

Column (1) shows that CRP is positively associated with DT. The coefficient of CRP is 0.109 (p < 0.01), indicating that firms that express higher climate risk perceptions in MD&A reports tend to exhibit stronger digital transformation intensity. This finding supports the “capability-building” segment of the mediation pathway.

Column (2) shows that DT is positively associated with EE. The coefficient of DT is 0.030 (p < 0.01), suggesting that firms with stronger digital transformation tend to achieve higher energy productivity, which is consistent with digital capability facilitating operational measurement, coordination, and process optimization.

Column (3) includes DT and CRP, both of which remain positive and significant (DT: 0.029, p < 0.01; CRP: 0.038, p < 0.01). Notably, the CRP coefficient in the joint model is smaller than the standalone CRP effect reported in the baseline regressions, indicating that part of the CRP–EE association operates through DT, while a substantial direct association remains. Collectively, these three-step results are consistent with partial mediation. Climate risk perception is linked to stronger digital transformation, which, in turn, is associated with improved energy productivity; however, DT does not fully account for the overall CRP–EE relationship.

7. Conclusions and Implications

7.1. Discussion

This study examines whether managers’ climate risk perceptions, as extracted from MD&A narratives, are associated with firms’ energy-related operational productivity, reflected in EE, a revenue-based measure defined as operating revenue generated per unit of energy input. Consistent with an attention-based view indicating that the focus of decision-makers shapes resource allocation and organizational priorities [16], our baseline estimates show a positive and statistically significant association between climate risk perception and energy productivity. This finding is economically meaningful because energy productivity is an operational endpoint that typically depends on internal routines, monitoring, and process control rather than purely symbolic responses. Prior evidence shows an “energy-management gap” in which economically meaningful improvements in energy-related performance remain unrealized because of managerial and organizational frictions [10]. Our results suggest that heightened climate-related attention is associated with tighter energy management and improved operational outcomes, extending the climate-finance literature demonstrating the economic relevance of firm-level climate exposure derived from corporate texts [4,6] from “exposure measurement” toward “operational adaptation.” Our dependent variable is revenue-based; therefore, the estimated relationship should be interpreted as reflecting the higher economic value generated per unit of energy input, which may incorporate pricing and product-mix variations, in addition to changes in physical energy use.

We further demonstrate that the information environment conditions the attention–outcome link. The interaction results indicate that stronger analyst scrutiny amplifies the positive association between climate risk perception and energy productivity. This pattern is consistent with analyst coverage functioning as an external monitoring and information-production mechanism in capital markets [19] and with evidence that analysts can acquire incremental information through deeper engagement, such as site visits [13]. Although analyst coverage can also induce managerial myopia and reduce long-horizon innovation investment [20], our estimates indicate that, in the context of energy-related operational outcomes, the monitoring channel is empirically dominant in our setting. This pattern is consistent with the idea that improvements in energy productivity can generate relatively near-term cost and risk benefits, and may therefore be less exposed to an “innovation crowd-out” mechanism.

Financial frictions reflect the other key boundary condition. We find that financing constraints, proxied by the Whited–Wu index [14], significantly weaken the association between climate risk perception and energy productivity. This aligns with the broader evidence that financial constraints and spillovers propagate the real effects of climate-related policy shocks [9,17]. Moreover, policy-based evidence from China shows that green credit guidelines improve total factor energy performance, and financing conditions are integral to firms’ ability to undertake upgrading-oriented investments [21]. Collectively, our results support an “implementation feasibility” interpretation, in which climate-related attention may be necessary but not sufficient for operational upgrading when external finance is costly and internal funds are scarce.

We also provide evidence of a capability channel. The mediation results indicate that climate risk perception is positively associated with digital transformation, which, in turn, is positively associated with energy productivity. When climate risk perception and digital transformation are included, the climate risk perception coefficient remains significant, but the mechanism operates through a partial pathway. This interpretation is consistent with the view that digital technologies improve productivity and operational performance at the firm level [8,11]. This is consistent with the domain-specific evidence that corporate digital technology adoption can reduce energy intensity and improve energy-related operational performance in Chinese listed firms [22], which can plausibly translate into higher revenue generated per unit of energy input in our setting. Conceptually, digital transformation can be viewed as a set of capabilities, including data, connectivity, analytics, and digital tools that strengthen measurement, coordination, and process control. These capabilities can support operational upgrading under climate transition pressure by enabling more precise energy use monitoring, identifying energy waste, and optimizing processes, thereby improving the economic value generated per unit of energy input.

Our findings have three key implications. First, MD&A-based climate risk perception measures can provide predictive content for energy-related operational outcomes, consistent with the broader evidence that firm-level climate exposure extracted from disclosure language is economically meaningful [4]. Second, external information intermediaries matter. Stronger analyst coverage appears to discipline the translation from climate attention to operational upgrading, suggesting that information transparency and monitoring can complement climate transition objectives [13,19]. Third, green transition is partly a financing problem. Policies that relieve financing frictions and expand access to green credit may materially improve firms’ capacity to implement upgrades that enhance energy productivity [17,21].

7.2. Conclusions

Using a large panel of Chinese A-share non-financial firms, we find robust evidence that higher climate risk perception is associated with higher energy productivity. This association strengthens with greater analyst scrutiny and weakens with tighter financing constraints and is partially mediated by digital transformation. The results are robust to an alternative climate risk perception measure, dynamic specifications that mitigate reverse-causality concerns, and an instrumental variable strategy that exploits plausibly exogenous weather variations commonly used in climate–economy research designs [30]. Overall, this study contributes to the climate-finance and sustainability accounting literature by linking climate-related managerial attention, measured based on corporate narratives, to a concrete energy-related operational productivity outcome and identifying information and financing frictions as key boundary conditions.

7.3. Limitations and Future Research

This study has several measurement and identification limitations. First, the energy productivity measure is based on revenue per unit of energy, which may be influenced by changes in prices, product mix, or sales strategies, rather than reflecting purely technical energy efficiency. Although this measure is widely used in firm-level studies, it may not fully capture physical energy efficiency. Second, climate risk perception is constructed using MD&A disclosures and may reflect both managerial beliefs and strategic disclosure incentives. More productive firms may have stronger incentives to communicate climate-related information for reputational reasons, which raises potential endogeneity concerns. Third, although our endogeneity testing design includes firm- and year-fixed effects, dynamic specifications, and an instrumental variable approach grounded in weather-based variation [30], the exclusion restriction cannot be proven and may be challenged if extreme weather affects energy productivity through channels unrelated to climate narratives. Future studies could combine disclosure-based measures with quasi-experimental shocks to disclosure mandates, climate regulations, or energy pricing to strengthen identification.