Textual Analysis of Sustainability Reports: Topics, Firm Value, and the Moderating Role of Assurance

Abstract

This study investigated how specific sustainability topics disclosed in standalone sustainability reports influence firm value and whether third-party assurance moderates this relationship. Drawing on signaling, agency, stakeholder, and legitimacy theories, we applied latent Dirichlet allocation (LDA) to extract latent topics from U.S. corporate sustainability reports. We analyzed their impact on Tobin’s Q using panel regressions and supplement our findings with discrete Bayesian networks (DBNs) and Shapley additive explanations (SHAP) to capture non-linear patterns. We identified six core topics: environmental impact, sustainable consumption, daily necessities, socio-economic impact, healthcare, and operations. The results revealed that topics of healthcare and daily necessities have immediate and sustained positive effects on firm value, while environmental and socio-economic impact topics demonstrate lagged effects, primarily two years after disclosure. The presence of assurance, however, produces mixed outcomes: it enhances credibility in some cases, but reduces firm value in others, especially when applied to environmental and socio-economic disclosures. This suggests a dual signaling effect of assurance, potentially increasing investor scrutiny when gaps in performance are highlighted. Our findings underscore the importance of topic selection, consistency in reporting, and strategic application of assurance in ESG communications to maintain stakeholder trust and market value.

1. Introduction

In recent years, sustainability disclosure has emerged as a central component of corporate communication, reflecting heightened investor interest in environmental, social, and governance (ESG) performance. A substantial body of literature links such disclosures to enhanced firm performance and value, with transparency, accountability, and stakeholder trust frequently identified as key drivers (Clarkson et al., 2020; Braam & Peeters, 2018). Despite the growing prevalence of sustainability reporting, relatively little attention has been paid to the specific content or thematic focus of these disclosures and how such content relates to firm value.

This study addresses that gap by examining the influence of distinct sustainability topics disclosed in standalone sustainability reports on firm value. We pursued three research questions: (1) What are the key topics disclosed in standalone sustainability reports? (2) How do these topics affect firm value? (3) Does third-party assurance moderate the relationship between these topics and firm value? These questions are especially relevant as firms face mounting pressure to provide disclosures that are both comprehensive and credible.

Alongside signaling, agency, and stakeholder perspectives, we incorporated organizational legitimacy theory to enrich our analysis of how sustainability topics shape firm value. Organizational legitimacy reflects the perception that a firm’s actions conform to socially accepted norms and values (Thomas & Lamm, 2012). Firms pursue legitimacy not only to retain stakeholder trust but also to secure long-term survival and competitiveness. Addressing material ESG topics—such as environmental impact, socio-economic engagement, and healthcare commitments—can help reinforce normative expectations while supporting strategic positioning.

Legitimacy theory, in both its strategic and institutional forms (Suchman, 1995), framed this inquiry. Strategically, sustainability disclosures and their assurance can act as deliberate signals of alignment with societal norms, thereby strengthening legitimacy among stakeholders. Institutionally, these same mechanisms may expose firms to regulatory, cultural, and normative expectations they cannot fully meet, risking legitimacy erosion. This dual-pathway perspective provides a theoretical rationale for why the moderating role of assurance is likely contingent on disclosure quality, stakeholder expectations, and contextual conditions.

We also drew on signaling theory, agency theory, and stakeholder theory to frame our hypotheses. From a signaling perspective, sustainability disclosures communicate a firm’s commitment to responsible conduct, reducing information asymmetry and potentially enhancing valuation (Kim & Park, 2023). Agency theory suggests that disclosures—and their assurance—can align managerial behavior with stakeholder expectations, mitigating agency problems (Al-Shaer & Zaman, 2019). Stakeholder theory supports the view that transparent, externally assured disclosures enhance legitimacy and long-term value (Peters & Romi, 2015).

The moderating role of assurance was central to our study. Prior research shows that assured reports tend to be more credible and useful to investors (Simnett et al., 2009; Moroney et al., 2012), and are associated with higher reporting quality (Clarkson et al., 2020), reduced capital constraints (Carey et al., 2021), and stronger reputations (Hazaea et al., 2022). Kim and Park (2023) further show that assurance can moderate the negative relationship between ESG performance and information asymmetry. We extended this line of work by investigating how assurance moderates the sustainability topic–firm value relationship.

Our study makes two primary contributions. First, methodologically, we employed latent Dirichlet allocation (LDA) to identify latent sustainability topics within textual disclosures, testing our hypotheses through regression analysis and validating results with discrete Bayesian networks (DBNs) and SHAP (Shapley additive explanations) analysis. Second, we examined the moderating effect of assurance on the link between sustainability topics and firm value—an area that remains underexplored. By doing so, we contribute to the literature on how both the content and credibility of sustainability disclosures influence firm outcomes, particularly within increasingly stakeholder-sensitive capital markets.

The remainder of this paper is structured as follows. Section 2 provides a comprehensive review of the extant literature and formulates our hypotheses. Section 3 covers methodology. Section 4 presents empirical results from the LDA, regression, DBN, and SHAP analyses. Section 5 offers a comprehensive discussion of our findings and their implications. Section 6 acknowledges the limitations of our study and suggests a path for future studies. Finally, Section 7 presents our concluding remarks and recommendations.

2. Background and Hypothesis Development

The relationship between environmental, social, and governance (ESG) disclosure and firm value is complex, multifaceted, and evolving. Foundational theories offer complementary insights into this relationship. From a signaling theory perspective, ESG disclosures act as credible signals to reduce information asymmetry between firms and investors (Spence, 1973; Cho et al., 2013). Firms that voluntarily disclose ESG topics signal superior management quality, long-term risk awareness, and ethical commitment.

Agency theory further posits that ESG disclosures—particularly when assured—reduce agency costs by providing transparent metrics to monitor managerial behavior (Jensen & Meckling, 1976; Al-Shaer & Zaman, 2019). Stakeholder theory complements this logic by framing ESG disclosures as responses to the expectations of diverse stakeholder groups whose support is essential to firm survival and legitimacy (R. E. Freeman, 2010; Donaldson & Preston, 1995).

Organizational legitimacy theory complements the above perspectives by highlighting how firms manage perceptions of appropriateness within socially constructed norms. From an instrumental perspective, legitimacy operates as a resource—enabling firms to attract capital, reduce regulatory scrutiny, and maintain stakeholder support (Jeong & Kim, 2019; Chiu & Sharfman, 2011). Simultaneously, from a normative standpoint, firms are expected to act in accordance with societal values and ethical obligations (Jackson, 2018). ESG disclosures serve both functions. For instance, when companies emphasize environmental stewardship or address socio-economic inequalities in their sustainability reports, they signal their alignment with public interest and values. These legitimacy signals can help firms maintain or recover reputational standing and trust. Moreover, third-party assurance may strengthen these perceptions, particularly when disclosure credibility is uncertain. Therefore, we propose that legitimacy-seeking behavior is a latent mechanism through which ESG topics and their assurance influence firm value, operating alongside signaling, agency, and stakeholder effects.

Despite growing consensus on the benefits of ESG disclosures, there is ongoing debate about whether and how the different ESG categories—environmental, social, and governance—contribute to firm value. Prior literature suggests that environmental disclosures may reduce regulatory risk and improve investor confidence (Clarkson et al., 2020); social disclosures foster reputation and employee engagement (Fatemi et al., 2018); while governance disclosures enhance board accountability and capital market discipline (Li et al., 2023). These topic-specific dynamics warrant disaggregated examination rather than assuming ESG as a monolithic construct. Accordingly, we hypothesized the following direct relationships:

H1a. 

Environmental topics disclosed in ESG reports will positively influence firm value.

H1b. 

Social topics disclosed in ESG reports will positively influence firm value.

H1c. 

Governance topics disclosed in ESG reports will positively influence firm value.

To address the temporal nature of market responses to ESG initiatives, we drew on J-curve logic. The ESG J-curve suggests that firms may experience an initial period of stagnant or negative financial response following ESG disclosures, as implementation costs rise or market actors adjust expectations. Over time, however, these efforts yield reputational gains, stakeholder loyalty, and operational efficiencies, resulting in upward valuation trends (Carnini Pulino et al., 2022). Incorporating a temporal lag in hypothesis testing accounts for the delayed realization of ESG benefits and aligns with empirical patterns in ESG valuation research.

Legitimacy theory provides further insight into how assurance may influence market perceptions of sustainability disclosures. From the strategic tradition, assurance can operate as a proactive tool for managing stakeholder impressions, enhancing the perceived credibility of disclosures, and reinforcing a firm’s alignment with prevailing social and environmental values. From the institutional tradition, assurance can subject firms to heightened scrutiny from regulators, rating agencies, and civil society actors, thereby reinforcing normative pressures but also increasing the risk of legitimacy loss if actual performance fails to meet these amplified expectations. This interplay between symbolic reinforcement and institutional constraint supports our expectation that assurance’s moderating effect is conditional on both the quality of disclosures and the broader stakeholder context.

While ESG disclosures may influence firm value, their impact is contingent on credibility. Assurance on ESG disclosures serves as a trust-enhancing mechanism that reinforces the perceived reliability of disclosed data. Signaling theory suggests that assurance functions as a high-cost signal of quality, particularly valuable under information asymmetry (Simnett et al., 2009). Empirical studies confirm that assurance enhances the value relevance of ESG disclosures (Carey et al., 2021; Vander Bauwhede & Van Cauwenberge, 2022). Therefore, we expected assurance to moderate the relationship between ESG topic categories and firm value, as follows:

H2a. 

Assurance positively moderates the relationship between environmental topics and firm value.

H2b. 

Assurance positively moderates the relationship between social topics and firm value.

H2c. 

Assurance positively moderates the relationship between governance topics and firm value.

3. Methodology

3.1. Study Sample

Our study focused on U.S. companies with a market capitalization of at least USD 20 billion. The U.S. is one of the world’s largest economies, and research conducted in the country has the potential to have a significant global impact on sustainability efforts. This focus is especially salient in light of the U.S. Securities and Exchange Commission’s recent adoption of rules aimed at enhancing and standardizing climate-related disclosures for publicly listed companies and public offerings (US Securities and Exchange Commission (SEC), 2024). While environmental, social, and governance (ESG) reporting in the United States remains largely voluntary and regulatory frameworks are still in their formative stages, the landscape is evolving rapidly. Consequently, companies must proactively acquaint themselves with forthcoming disclosure requirements and emerging sustainability priorities. These impending changes make our target sample of great interest to both scholars and practitioners. Additionally, we were motivated to test the boundaries of the influence of sustainability topics in a market with relatively minimal regulations. Researchers have the opportunity to replicate the study in markets with stringent regulations, such as European and Asian countries, for the purpose of conducting a comparative analysis.

Our initial sample pool comprised 297 companies, yielding 1454 firm-years for the fiscal years 2005 to 2017. After eliminating redundant data, non-English reports, obscured text, and inaccessible variables, we were left with 1208 firm-years (191 companies). Of these, 237 firm-years (58 companies) had assured reports and 971 firm-years (133 companies) had not. We sourced all financial variables from Standard and Poor’s Capital IQ and Compustat databases. Our control variables include total inventories (INV), long-term and current period debt (Leverage), the natural logarithm of total assets (LnTotal Assets), the natural logarithm of the ratio of current year revenue to prior year revenue (LnSalesGrowth), property plant and equipment (PPE), net income divided by total assets (Return on Assets), a variable that increments by 1 each year starting at 0 in the year 2006 (Trend), and environmental, social and governance scores. We obtained our dependent variable, Tobin’s Q, from the Compustat North America database.

Table 1. Sample selection of sustainability reports from CorporateRegister.com.

Number of companies in the USD 20 billion or above market capitalization (as of 31 December 2017)	297
Number of companies with at least 1 sustainability report	1454 firm-years (206 companies)
Number of firm-years dropped, due to unavailability of variables or due to MTB value being zero	246
Remaining firm years that we can work with	1208 firm-years (191 companies)
No. of firm-years that have assurance	37 firm-years (58 companies)
No. of firm-years that have NO assurance	971 firm years (133 companies)

below shows our sample selection from CorporateRegister.com

3.2. Topic Modeling Using LDA

Latent Dirichlet allocation (LDA) is a popular unsupervised machine learning method used for topic modeling in text analytics to model latent topics discussed in documents (Blei et al., 2003). LDA employs a three-level hierarchical Bayesian model to model documents as a finite mixture over an underlying set of topics. Each topic is in turn modeled as a mixture of topic probability distributions. This allows us to represent each document as a linear function of topic probabilities, which in turn are functions of the word frequencies.

In the LDA model, the parameters can be mathematically represented as follows:

$$p(\theta ,z,w|\alpha ,\beta )=p(\theta |\alpha )\prod _{n=1}^{N}p(z_n|\theta )p(w_n|z_n,\beta ),$$

In this equation, $p\left(\alpha \right)={\displaystyle \frac{\Gamma \left({\sum }_{i=1}^k{\alpha }_i\right)}{{\prod }_{i=1}^k\Gamma ({\alpha }_i)}}{\theta }_1^{{\alpha }_1-1}\dots {\theta }_k^{{\alpha }_k-1}$. This is known as the Dirichlet distribution, where $\theta$ is the topic mixture, $z$ is the set of $N$ topics, $w$ is the set of $N$ words, and {$\alpha ,\beta$} are hyperparameters. The model parameters and latent topics were estimated using a variational expectation-maximization algorithm (Blei et al., 2003). The quality of the discrimination between topics using quantitative metrics such as perplexity and coherence scores were inspected to decide the optimal number of topics in the document corpus, followed by assigning a label to each topic based on frequent words occurring in particular topics. Words such as articles and pronouns were ignored by the algorithm before the process of grouping words into topics. Further, while labeling the topics, commonly occurring nouns such as company, document, benefit, and people were given less weight compared to specific terms occurring at a higher frequency within a specific topic.

The LDA method is particularly relevant, as we wished to explore the nature of sustainability reporting of companies across industry sectors. LDA not only allows us to examine the distinct topic patterns in sustainability documents but also makes it possible to quantify the relative influence of topics in the discussion of each sustainability report. We used the genism library in Python software (version 3.12.6) to train the LDA algorithm over the sustainability text documents. LDA is an unsupervised method for grouping words across documents into coherent topics, and hence it does not require the authors to set any parameters for the model training. However, once the topics are determined, the words within each topic group are closely examined to label each topic. The labeling of topics involves manual inputs based on interpretation of the bag of words within each topic.

3.3. Regression Modeling

To test our two hypotheses, we fitted panel regression models using clustered robust errors. Tobin’s Q and its future values (Lead-1, Lead-2, and Lead-3 Tobin’s Q) served as our dependent variables. These proxies represented the current firm value and the anticipated firm values one, two, and three years ahead, respectively. For clarification, Lead-1 Tobin’s Q refers to future values of Tobin’s Q moved forward by one fiscal year for each company. In a similar vein, Lead-2 and Lead-3 denote Tobin’s Q advanced by two and three fiscal years, respectively1.

Our analysis recognized six latent topics (refer to Section 4). However, we excluded one topic (operations and logistics) from the model inputs to circumvent linear dependency errors. To illustrate, linear dependency arises because the sum of the LDA weights across all topics equals one. Therefore, given the other five estimated coefficients, the coefficient for the sixth topic became redundant.

Beyond the identified topics, we incorporated assurance indicators—specifically whether a firm sought assurance in a given year—into the model. Time trends and company-related covariates were also included following the established literature on firm value estimation (Albuquerque et al., 2019; Luo et al., 2013; Servaes & Tamayo, 2013; Simeth & Cincera, 2016). A comprehensive list and definitions of all variables utilized in our study can be found in Appendix A.

To cross-examine and validate the findings from the regression models, we conducted supplementary analyses (Section 4.2.1) using Bayesian networks, a directed acyclic graph representing condition probabilities among variables, and Shapley additive explanations, a game-theory feature-importance method that ranks important textual disclosure determinants of firm value. The supplementary analyses were conducted to validate the findings from the main analysis (i.e., regression modeling), while augmenting knowledge discovery using machine learning approaches that can capture non-linear relationships and multivariate interactions that may not be directly discernible using statistical linear models.

4. Analysis

Table 2. Descriptive statistics.

	Not Assured						Assured						Total
	Mean	SD	25th per	Median	75th per	N	Mean	SD	25th per	Median	75th per	N	Mean	SD	25th per	Median	75th per	N
Topic: Environmental	0.2811	0.2215	0.0967	0.2531	0.4205	971	0.4682	0.1956	0.3599	0.4609	0.6079	237	0.3178	0.229	0.1246	0.2967	0.4713	1208
Topic: Sustainable Consumption	0.1225	0.1589	0.0092	0.0447	0.1779	971	0.0876	0.1501	0.001	0.0091	0.0803	237	0.1156	0.1578	0.0059	0.0342	0.1723	1208
Topic: Daily Necessities	0.077	0.1596	0.0018	0.0083	0.0406	971	0.063	0.1391	0.0019	0.0075	0.0319	237	0.0743	0.1559	0.0018	0.0081	0.0388	1208
Topic: Socio-Economic Impact	0.291	0.2605	0.087	0.1921	0.4412	971	0.1722	0.2165	0.0303	0.0884	0.1923	237	0.2677	0.2567	0.069	0.1669	0.3899	1208
Topic: Healthcare	0.1318	0.3385	0	0	0	971	0.097	0.2966	0	0	0	237	0.125	0.3309	0	0	0	1208
LnTotalAssets	10.7148	1.3692	9.7898	10.5667	11.5678	971	10.9069	1.2749	10.0151	10.7489	11.7716	237	10.7525	1.3528	9.8484	10.5975	11.5967	1208
Leverage	0.9347	1.8901	0.523	0.6602	0.8342	971	0.6343	0.2422	0.4737	0.6108	0.7934	237	0.8758	1.702	0.5057	0.6467	0.8244	1208
PropPlant and Equipment	0.3181	0.4297	0.0702	0.1811	0.5052	971	0.3378	0.2453	0.1199	0.2934	0.5321	237	0.322	0.4003	0.0793	0.198	0.5194	1208
Inventory	0.0809	0.1269	0.0125	0.0415	0.1010	971	0.0582	0.0506	0.0190	0.0465	0.0862	237	0.0765	0.1163	0.0131	0.0445	0.0984	1208
Return on Assets	0.0652	0.0643	0.0242	0.0563	0.0963	971	0.0701	0.0673	0.0248	0.0619	0.1116	237	0.0662	0.0649	0.0242	0.0586	0.0997	1208
LnSalesGrowth	0.0462	0.1442	−0.0037	0.0428	0.0901	971	0.0173	0.1464	−0.0386	0.0325	0.0784	237	0.0405	0.145	−0.0091	0.0428	0.0881	1208
Trend	7.2225	3.5505	5	8	10	971	7.70042	3.43938	5	8	11	237	7.3162	3.5327	5	8	10	1208
ESG: ENV_Score	57.3522	20.763	42	57	74	971	62.37975	20.40769	44	65	82	237	58.3386	20.7816	43	59	75	1208
ESG: SOC_Score	42.4954	19.4415	25	40	61	971	51.03797	19.81339	33	46	71	237	44.1714	19.7998	27	43	62	1208
ESG: GOV_Score	51.4408	17.4645	36	50	69	971	55.38819	18.36290	37	57	70	237	52.2152	17.7062	37	50	69	1208

shows descriptive statistics by companies that were not assured, assured, and the total sample. We compared the 10 most frequent words used in sustainability documents in healthcare and IT companies (Figure 1). For this process, we removed proper nouns and conjunctions, articles, and other common stop words to retain terms that may have been related to discussion topics. As shown in Figure 1, terms such as employees, global, corporate, and business are common, while there are some distinct words in each industry sector. For example, the term health is most frequently used in sustainability reports of healthcare companies, as expected, while IT companies use terms such as data, environmental, energy, and emissions more often. We repeated this exploration for other industry sectors, which emphasized the need for modeling the topics in sustainability documents across industry sectors to see the distribution of topic themes. The appendix presents the correlation matrix. The correlations between sustainability-related topics and ESG scores range from 0.10 to 0.20, indicating a weak association between these variables. Notably, sustainability-related topics exhibit particularly low correlations with ESG scores. This finding suggests that analyzing latent sustainability-related topics offers additional insights and contributes to the existing literature on the relationship between ESG and firm outcomes.

The LDA topics were identified and labeled implementing the procedure adapted by Dyer et al. (2017). We first identified the number of topics using an elbow plot of perplexity score plotted over topic count. Then, the topic set identified with the saddle point in the plot was chosen, followed by manual inspection of words in each topic. Thereafter, these topics were grouped based on similar word membership and context. One of the authors and his student independently labeled each of the five resulting topics using a fixed set of labels and then cross-verified their assignments. Following labeling, the topic set and its word membership were verified by the other authors for contextual relevance and appropriateness.

Finally, six topics were identified using the LDA models. They can be represented as a function of words with the highest frequencies, as follows:

$$\begin{array}{l}Topi{c}_{ENVIRONMENTGLOBALIMPACT}\\ =0.016\ast report+0.011\ast employee+0.011\ast product+0.010\\ \ast use+0.009\ast business+0.009\ast global+0.009\ast include\\ +0.008\ast emission+0.008\ast environmental+0.008\\ \ast management\end{array}$$

$$\begin{array}{l}Topi{c}_{SUSTAINABLECONSUMPTION}\\ =0.012\ast employee+0.011\ast up+0.009\ast fuel+0.008\ast year\\ +0.007\ast company+0.007\ast vehicle+0.007\ast com+0.007\\ \ast safety+0.006\ast customer+0.006\ast use\end{array}$$

$$\begin{array}{l}Topi{c}_{SOCIOECONOMICIMPACT}\\ =0.012\ast employee+0.012\ast business+0.011\ast program\\ +0.010\ast community+0.009\ast help+0.008\ast corporate\\ +0.008\ast service+0.008\ast customer+0.008\ast report+0.008\\ \ast provide\end{array}$$

$$\begin{array}{l}Topi{c}_{HEALTHCARE}\\ =0.021\ast health+0.011\ast care+0.011\ast program+0.011\\ \ast patient+0.009\ast product+0.008\ast company+0.007\\ \ast employee+0.006\ast support+0.006\ast access+0.006\ast work\end{array}$$

$$\begin{array}{l}Topi{c}_{DAILYNECESSITIES}\\ =0.012\ast food+0.010\ast product+0.009\ast program+0.008\\ \ast work+0.008\ast water+0.007\ast company+0.007\ast store\\ +0.007\ast help+0.007\ast make+0.006\ast consumer\end{array}$$

$$\begin{array}{l}Topi{c}_{OPERATIONSANDLOGISTICS}\\ =0.020\ast energy+0.012\ast customer+0.009\ast employee\\ +0.009\ast program+0.009\ast company+0.007\ast include\\ +0.006\ast safety+0.006\ast use+0.006\ast emission+0.006\\ \ast project\end{array}$$

Across the topics, we observed common terms such as employee, corporate, business, management, etc., as expected in a typical company sustainability disclosure. However, there were distinct terms in each topic that were interrelated and shared common contexts. We labeled the topics based on the high-frequency words that were distinct to each topic and that shared a common context.

The first topic (environmental global impact) had a wide range of terms, such as sustainability, environmental, material, waste, global, emission, and water. This topic is related to the physical environment, sustainability, impacts of a corporation on global issues such as pollutant emission, energy crisis, etc. Almost all companies dedicate a section discussing the environmental and global impact of their business entities. The second topic (socio-economic impact) had some overlapping terms with the first topic, such as environmental and energy, but the other terms appeared to be more people-centric. That is, terms such as social, community, team, customer, diversity, support, and financial indicators are more of a social contribution by the corporation. The social contribution may be internal—helping employees, team building, etc.—extrinsic, i.e., helping the immediate community, customers, etc. The distinct terms in the third topic category (sustainable consumption) were use, reduce, fuel, vehicle, help, facility, provide, time, and support. Most of these words indicate a theme of giving back to society through responsible and reduced consumption of resources such as fuel, facilities, and other resources. The distinct terms in the fourth topic category (healthcare) included need, medical, medicine, help, health, care, patient, and access, which point to health-related needs of employees and community. The fifth topic (daily necessities) had frequent words that were similar to the socio-economic impact-related topic, but no terms related to finance or economics. Rather, the frequencies of terms such as food, product, program, work, water, and store were higher, indicating that this topic group pertains to the daily necessities of its employees, consumers, community, and other stakeholders of the corporation. While a few terms in the last topic (operation and logistics) were common to topic 1 and topic 2, such as energy, customer, work, and environmental, there were other terms, such as cost, project, safety, plan, plant, operation, service, use, etc., that indicated that this topic alludes to the logistics and operational functionality of any corporation. The six latent topics identified in ESG disclosures mapped to environmental (topic 1), social (topic 3), and governance (topics 2, 4, 5, and 6), thus enabling us to test our study hypotheses.

4.1. Main Findings

Our findings from the linear models are presented in

Table 3. No assurance: effect of sustainability topics on current and future Tobin’s Q.

Dependent Variables	TobinsQ	LeadTobinsQ	Lead2TobinsQ	Lead3TobinsQ
Topic: Environmental	0.6874	1.1636	2.2502 *	−0.1722
	(0.7432)	(0.8857)	(1.3322)	(0.7037)
Topic: Sustainable Consumption	−0.4974	0.0343	0.4493	−0.7538
	(0.8100)	(0.9180)	(1.3480)	(0.7322)
Topic: Daily Necessities	1.6644 **	2.3784 **	2.0504	0.0919
	(0.8379)	(0.9218)	(1.3870)	(0.7368)
Topic: Socio-Economic Impact	1.2003	1.3550	2.0117 *	0.0501
	(0.7704)	(0.8649)	(1.2135)	(0.6399)
Topic: Healthcare	2.5281 **	3.3210 ***	3.1931 **	−0.0227
	(0.9837)	(1.0847)	(1.5560)	(0.9031)
LnTotal Assets	−0.4421 ***	−0.5294 ***	−0.5298 ***	−0.3557 ***
	(0.0562)	(0.0604)	(0.0878)	(0.0613)
Leverage	−0.0384	0.0059	−0.1148	0.0648**
	(0.0653)	(0.0347)	(0.1240)	(0.0308)
PPE	0.0353	0.0359	0.4485	−0.3930*
	(0.3078)	(0.2456)	(0.5718)	(0.2143)
Inventory	−0.7194	−1.2029 **	−0.5411	−0.3588
	(0.6109	(0.5811)	(0.9607)	(0.4540)
Return on Assets	8.8346 ***	5.1387 ***	4.3817 **	13.3874 ***
	(1.1300)	(1.1564)	(1.8630)	(1.5203)
LnSalesGrowth	3.1253 ***	1.9986 ***	2.6683 ***	0.4208
	(0.7167)	(0.6262)	(1.0003)	(0.6855)
Trend	0.0899 ***	0.0576 **	0.1106 **	0.0879 ***
	(0.0276)	(0.0278)	(0.0479)	(0.0315)
ESG: Environmental Score	0.0126 ***	0.0126 ***	0.0127 *	0.0093 **
	(0.0041)	(0.0043)	(0.0066)	(0.0041)
ESG: Social Score	0.0109 *	0.0116 *	0.0025	0.0002
	(0.0061)	(0.0067)	(0.0099)	(0.0057)
ESG: Governance Score	−0.0233 ***	−0.0264 ***	−0.0133	−0.0120 *
	(0.0068)	(0.0073)	(0.0111)	(0.0066)
Constant	4.4232 ***	5.7734 ***	4.7776 ***	4.0206 ***
	(0.7617)	(0.7861)	(1.2620)	(0.7631)
Control for Industry Sector	Yes	Yes	Yes	Yes
Observations	1208	831	735	611
Number of Companies	191	159	157	137
R-sq Within	0.114	0.0332	0.0174	0.0195
R-sq Between	0.736	0.738	0.574	0.785
R-sq Overall	0.568	0.579	0.513	0.559

Standard errors are adjusted for heteroscedasticity and clustered at the firm level. *** p < 0.01, ** p < 0.05, * p < 0.1. Variables defined in Appendix A.

and

Table 4. Assurance and interactions—effect of sustainability topics on current and future Tobin’s Q.

Dependent Variables	TobinsQ	LeadTobinsQ	Lead2TobinsQ	Lead3TobinsQ
Topic: Environmental	0.8239	1.9162 *	3.0816 **	−0.1425
	(0.7798)	(0.9829)	(1.4328)	(0.7866)
Topic: Sustainable Consumption	−0.2692	0.9549	1.1858	−0.4869
	(0.8575)	(1.0263)	(1.4712)	(0.8179)
Topic: Daily Necessities	1.6896 **	2.9082 ***	2.1144	−0.0302
	(0.8548)	(0.9738)	(1.4545)	(0.7864)
Topic: Socio Economic Impact	1.0997	1.8865 **	2.2662 *	0.1086
	(0.7895)	(0.9263)	(1.2670)	(0.6810)
Topic: Healthcare	2.4497 **	3.8294 ***	3.2051 *	−0.0619
	(0.9956)	(1.1463)	(1.6219)	(0.9554)
Assurance	0.8126	2.1546*	1.6604	0.1523
	(1.0963)	(1.1467)	(1.7524)	(0.9605)
Assurance × Environmental topic	−1.6592	−2.8416 *	−3.4635	−0.3840
	(1.3809)	(1.4549)	(2.2585)	(1.4488)
Assurance × SustConsumption topic	−0.4255	−2.8195	−2.3380	−0.8200
	(1.7609)	(1.7725)	(2.6643)	(1.4651)
Assurance × DailyNecessities topic	1.4524	−0.8282	1.3195	1.5903
	(1.7341)	(1.6395)	(2.6841)	(1.3955)
Assurance × SocioEconImpact topic	0.2488	−2.3232 *	−1.5545	0.2321
	(1.2049)	(1.2512)	(2.0089)	(1.2987)
Assurance × Healthcare topic	1.3951	0.1518	3.0819	1.5266
	(1.8401)	(1.7750)	(2.8939)	(1.6840)
LnTotal Assets	−0.4356 ***	−0.5389 ***	−0.5199 ***	−0.3441 ***
	(0.0570)	(0.0628)	(0.0908)	(0.0645)
Leverage	−0.0328	0.0057	−0.0975	0.0601 *
	(0.0659)	(0.0347)	(0.1272)	(0.0313)
PPE	0.0168	0.0446	0.4209	−0.3500
	(0.3112)	(0.2469)	(0.5853)	(0.2180)
Inventory	−0.6416	−1.3163 **	−0.8699	−0.3906
	(0.6292)	(0.6020)	(0.9880)	(0.4745)
Return on Assets	8.1988 ***	4.6819 ***	3.3661 *	12.9267 ***
	(1.1620)	(1.1903)	(1.9084)	(1.5809)
LnSalesGrowth	2.7934 ***	2.1474 ***	2.7278 ***	0.3473
	(0.7410)	(0.6288)	(0.9945)	(0.6961)
Trend	0.0954 ***	0.0719 **	0.1344 ***	0.1050 ***
	(0.0278)	(0.0290)	(0.0490)	(0.0329)
ESG: Environmental Score	0.0122 ***	0.0136 ***	0.0125 *	0.0092 **
	(0.0041)	(0.0043)	(0.0066)	(0.0042)
ESG: Social Score	0.0110 *	0.0121 *	0.0058	0.0023
	(0.0062)	(0.0069)	(0.0101)	(0.0062)
ESG: Governance Score	−0.0226 ***	−0.0272 ***	−0.0158	−0.0137 **
	(0.0069)	(0.0075)	(0.0112)	(0.0069)
Constant	4.3120 ***	5.5305 ***	4.4040 ***	3.7789 ***
	(0.7631)	(0.8046)	(1.2809)	(0.7921)
Controls for Industry Sector	Yes	Yes	Yes	Yes
Observations	1208	831	735	611
Number of Companies	191	159	157	137
R-sq Within	0.101	0.0227	0.00771	0.0251
R-sq Between	0.750	0.754	0.601	0.796
R-sq Overall	0.586	0.578	0.502	0.586

Standard errors are adjusted for heteroscedasticity and clustered at the firm level. *** p < 0.01, ** p < 0.05, * p < 0.1. Variables defined in Appendix A.

. Table 3 reveals the main effects of sustainability topics on current, Lead-1, Lead-2, and Lead-3 Tobin’s Q. Table 4 explains the moderating role of assurance on the relationships between sustainability topics and firm value.

For both Table 3 and Table 4, we emphasize that the results presented should not be interpreted as evidence of causation. As noted earlier, our approach relied on textual analysis to identify key topics within standalone sustainability reports and to examine their association with firm value—both contemporaneously and in the future. While the results suggest a relationship between sustainability topics, assurance, and firm value (measured by Tobin’s Q), we do not claim a causal link.

We also highlight the role of control variables in our analysis. These variables help account for other factors that could influence both the independent and dependent variables, thereby reducing potential bias in our panel regression models. By including controls identified in prior literature, we aimed to isolate the specific associations of interest: namely the links between sustainability topics, assurance, and firm value.

Our panel regression models were designed to capture correlations within a structured framework that reflected the nature of panel data (i.e., repeated observations for the same firms over time). This approach allowed us to account for unobserved heterogeneity, endogeneity, and other factors that may have affected the relationships under investigation.

Table 3 indicates that including topics on environment and socio-economic impact in a company’s current period sustainability report significantly influences Lead-2 (p-value < 0.1), suggesting a value increase for the company two years later, given that there is no assurance sought. This lag effect was in line with our expectations. However, it is noteworthy that the impact of environment and socio-economic topics did not sustain beyond the second year.

Reporting on daily necessities elevates a company’s current and subsequent year’s value (p-value < 0.05), indicating immediate and sustained effects for at least two years. Furthermore, reporting on healthcare increases not only the current and next year’s firm value but also Lead-2 (respective p-values: model 1 < 0.05, model 2 < 0.01, model 3 < 0.05). This suggests that healthcare reporting has both an immediate and sustained effect, although the impact diminishes after the second year.

Interestingly, none of the sustainability topics showed significance for Lead-3 for companies not seeking assurance. This implies that while the lag effect persists for at least two years post-reporting, its influence dwindles beyond the third year. Therefore, firms not seeking assurance might benefit from consistent healthcare reporting due to its robust impact on current, Lead-1, and Lead-2 Tobin’s Q, thereby potentially enhancing their firm value. These results partially support Hypothesis 1 (H1).

However, when we introduced assurance as a moderating factor (Table 4), all topics except sustainable consumption showed significance. Reporting on environmental topics in the current sustainability report increases firm value in the next year and the second-year post-inclusion (p-values: model 1 < 0.1; model 2 < 0.05). This impact is more pronounced when compared to scenarios of no assurance or assurance alone. Similarly, including the socio-economic impact topic also enhances firm value in the subsequent year and the second-year post-inclusion (respective p-values: model 1 < 0.05; model 2 < 0.1).

The impact of daily necessities and healthcare reporting remained consistent when compared to no assurance or assurance alone, though the significance of healthcare reporting in the second-year post-inclusion slightly dropped (p-value < 0.05 to p-value < 0.1). Despite this minor change, healthcare reporting remains more influential than other topics.

Considering the absence of any sustainability topics showing significance for Lead-3, it is crucial to maintain consistent reporting on environmental, socio-economic impact, and daily necessities topics, given their impact on firm value. The diminished significance over time indicates the effect is not self-sustaining. Therefore, due to the strong impact on current, Lead-1, and Lead-2 Tobin’s Q, companies should consider annual reporting on healthcare topics to bolster their firm value. These findings partially support Hypothesis 2 (H2).

Surprisingly, the interaction of assurance with environmental topics and socio-economic impact has a negative influence on firm value in the next year and does not show significance in the following years (p-value in both cases < 0.1, Table 4). This might suggest that verified information on these topics could include unpredictable or unfavorable events, negatively impacting subsequent years’ firm value. To interpret these findings, we drew on signaling theory (Spence, 1973) and recent studies on assurance in sustainability reporting (Simnett et al., 2009; Peters & Romi, 2015). According to signaling theory, assurance can act as a mechanism to reduce information asymmetry between firms and investors by enhancing the credibility of ESG disclosures. However, this signal is not always straightforward. In some cases, assurance may draw attention to gaps or inconsistencies in disclosures that would otherwise go unnoticed, potentially raising investor skepticism rather than confidence (Christensen et al., 2021).

We propose the following dual-pathway explanation to better contextualize these findings: we argue that while third-party assurance may enhance the perceived credibility of ESG disclosures, it may also draw attention to inconsistencies or gaps, potentially increasing investor skepticism. This dual-pathway effect helps explain the nonsignificant interaction terms and underscores the complex and sometimes contradictory signaling role that assurance can play in shaping investor perceptions.

This interpretation reflects the growing complexity of assurance as a signal. Rather than uniformly enhancing firm value, assurance may send mixed signals, especially when the underlying disclosures lack substance or appear inconsistent. These findings align with prior work suggesting that the effectiveness of assurance depends on contextual factors such as disclosure quality, stakeholder expectations, and the perceived independence of the assurance provider (Moroney et al., 2012; Ballou et al., 2006).

In sum, our results highlight the nuanced role of assurance in the ESG disclosure landscape and suggest that its impact on firm value is conditional rather than universal.

4.2. Supplementary Analyses

4.2.1. Exploring Multivariate Associations

We trained a discrete Bayesian network (DBN) to characterize the conditional dependencies between industry sectors, assurance, ESG scores, sustainability topics in sustainability reports, and (present and future) firm values. A Bayesian network is a probabilistic graphical model that represents a set of variables and their conditional dependencies via a directed acyclic graph (DAG) (Koller & Friedman, 2009), and can be represented as follows:

$$P\left(X_1,X_2,\dots ,X_n\right)={\prod }_{i=1}^{n}P(X_i|{\pi }_i)$$

In this equation, ${\pi }_i$ are the parents of variable $X_i$. The two components of a Bayesian network are the graph structure and the numerical (conditional) probabilities assigned for each variable given its parents. Structural learning of the Bayesian network can either be achieved through expert inputs or using automated methods such as grow–shrink, greedy search, or genetic algorithms (Nagarajan et al., 2013). Our model had four tiers of variables: (i) industry sectors, (ii) ESG scores and assurance, (iii) sustainability topics, and (iv) present and future firm values. If the topic score for a sustainability document were greater than 0.15, we assumed the topic was present. Otherwise, we assumed it was absent. We initialized the automated learning by blacklisting directed edges from topics to industry sectors, assurance, ESG scores, and from firm values to topics. That is, we enforced a rule that the Bayesian network could not have a direct link from lower-tier to higher-tier variables from the outcomes of the inputs. Next, we enforced a rule that variables within the same tier were not linked to each other, as between-level effects were not our focus. After comparing the classification error for nodes using different structural learning methods, we considered the hill-climbing greedy search method (Tsamardinos et al., 2006) for generating the structure of the Bayesian network, shown in Figure 2. The pairwise conditional probabilities of each topic variable given its parents are shown as corresponding edge labels.

The DBN shows that topics included by firms in their sustainability reports are differently related to their industry sector membership. For example, sustainability reports of IT companies have 0.61 and 0.95 probabilities of including text related to socio-economic impact and environmental topics, respectively. Firms with assurance tend to include more text related to environmental topics than non-assured firms and less on sustainable consumption, daily necessities, and socio-economic impact topics. The model also shows that healthcare companies predominantly focus on healthcare topics in their sustainability reports, while firms in the energy sector emphasize health topics with a low probability of 15% in their sustainability reports. The results of the Bayesian network not only reaffirm certain findings based on the regression analysis but also uncover new multivariate associations.

4.2.2. Ranking the Predictive Contributions of Variables

We examined the contribution of each feature in predicting present and future values of Tobin’s Q using the Shapley additive explanations (SHAP) method (Lundberg & Lee, 2017). The SHAP method uses a coalitional game-theory approach where a prediction is explained by assuming that each feature is a player in a game where the prediction is the payout. We trained a state-of-the-art machine learning model called XGBoost (T. Chen & Guestrin, 2016) to predict the four outcomes independently using the input features, and then used SHAP to identify the contributions of the input features to the model’s prediction.

We used fivefold cross-validation to compare the XGBoost model’s predictive performance with other data-mining models, including random forest, gradient boosting machine (GBM), regression tree, and multivariate adaptive regression splines (MARS), and found XGBoost to outperform other predictive modeling methods for our dataset. We used fivefold cross-validation for grid-search-based parameter tuning of the XGBoost model for TobinsQ, LeadTobinsQ, Lead2TobinsQ and Lead3TobinsQ dependent variables. We have not shown the performance comparison tables for sake of brevity. The final hyperparameter values post-tuning for the XGBoost model were eta = 0.02, max_depth = 10, gamma = 0.01, subsample = 0.98, colsample_bytree = 0.86, nrounds = 1000, and early_stopping_rounds = 8. All models were trained using R software.

The feature contributions in

Table 5. Feature importance ranking for machine learning models predicting present and future firm values.

Rank	Feature Importance
	Present Firm Value as Outcome	Firm Value 1-Year Lead as Outcome	Firm Value 2-Year Lead as Outcome	Firm Value 3-Year Lead as Outcome
1	Company-related—ROA (48.2%)	Company-related—ROA (44.5%)	Company-related—ROA (40.7%)	Company-related—ROA (40%)
2	Company-related—LnTotalAssets (26.3%)	Company-related—LnTotalAssets (27.7%)	Company-related—LnTotalAssets (29.3%)	Company-related—LnTotalAssets (27.4%)
3	Industry—Financials (14.5%)	Industry—Financials (12.4%)	Industry—Financials (9%)	Industry—Financials (15%)
4	Time trend (9.9%)	Topic—Daily Necessities (6.7%)	Company-related—PPE (7.8%)	Time trend (7.6%)
5	Topic—Sustainable Consumption (6.2%)	Company-related—PPE (5.7%)	Time trend (7.1%)	Topic—Daily Necessities (7.4%)
6	Company-related—PPE (5.7%)	Time trend (5.7%)	Topic—Environmental (5.6%)	ESG GOV score (5.5%)
7	Company-related—logeSalesGrowth (5.2%)	Company-related—logeSalesGrowth (4.9%)	Topic—DailyNecessities (4.1%)	Topic—Environmental (4.9%)
8	Topic—Daily Necessities (4.4%)	Topic—Environmental (4.1%)	Topic—Socio-Economic Impact (3.9%)	Company-related—Leverage (4.4%)
9	ESG ENV score (3.2%)	ESG ENV score (3.7%)	ESG GOV score (3.8%)	Company-related—PPE (4.3%)
10	Topic—Environmental (3.2%)	ESG GOV score (3.4%)	Company-related—Leverage (3.1%)	ESG ENV score (3.5%)
11	Topic—Healthcare (3%)	Topic—Sustainable Consumption (3.3%)	Company-related—INV (2.8%)	Company-related—INV (3.3%)
12	Company-related—INV (2.7%)	Company-related—Leverage (2.8%)	ESG ENV score (2.8%)	Topic—Sustainable Consumption (3.2%)
13	Company-related—LEV (2.6%)	Company-related—INV (2.5%)	Topic—Sustainable Consumption (2.6%)	Industry—Energy (2.7%)
14	ESG GOV score (2.4%)	Topic—Healthcare (2.4%)	Company-related—logeSalesGrowth (2.6%)	Topic—Healthcare (2.2%)
15	Industry—Energy (1.8%)	Industry—Energy (2.2%)	Industry—Energy (2.3%)	Company-related—logeSalesGrowth (1.9%)
16	ESG SOC score (1.3%)	Topic—SocioEconomic Impact (1.7%)	Topic—Healthcare (2%)	ESG SOC score (1.6%)
17	Topic—Socio-Economic Impact (1.3%)	ESG SOC score (1.6%)	ESG SOC score (1.8%)	Topic—Socio-Economic Impact (1.3%)
18	Industry—IT (1%)	Industry—IT (1.1%)	Industry—IT (0.7%)	Industry—Consumer Discretionary (0.7%)
19	Industry—Consumer Discretionary (0.6%)	Industry—Consumer Discretionary (0.7%)	Industry—Materials (0.6%)	Industry—Healthcare (0.6%)
20	Industry—Industrials (0.4%)	Industry—Materials (0.4%)	Industry—Consumer Discretionary (0.5%)	Industry—Materials (0.5%)
21	Industry—Consumer Staples (0.3%)	Industry—Industrials (0.4%)	Industry—Industrials (0.5%)	Industry—Industrials (0.4%)
22	Industry—Materials (0.3%)	Industry—Healthcare (0.3%)	Industry—Healthcare (0.4%)	Assured (0.4%)
23	Assured (0.3%)	Assured (0.2%)	Assured (0.3%)	Industry—IT (0.2%)
24	Industry—Healthcare (0.2%)	Industry—Consumer Staples (0.1%)	Industry—Consumer Staples (0.1%)	Industry—Consumer Staples (0.1%)
25	Industry—Communication Services (0.1%)	Industry—Communication Services (0%)	Industry—Communication Services (0.1%)	Industry—Communication Services (0.1%)

provide insight into the relative importance of various predictors for firm value (Tobin’s Q) across different time horizons. Company-related factors—specifically ROA and total assets—emerged as the two most influential predictors across all four lagged firm value models. These were followed by the industry sector indicator for financials, which showed consistent predictive power, whereas most other industry sectors contributed minimally to the model’s performance. The time trend also appeared as a significant feature, particularly for current and short-term (one- and two-year) firm value predictions, underscoring the role of temporal or market-wide effects.

ESG disclosure topics accounted for between 1.3% and 4.9% of the models’ predictive performance, with the environmental topic standing out as one of the stronger contributors, consistent with findings from the dynamic Bayesian network (DBN) model. ESG (environmental, social, and governance) scores became more influential over longer time horizons (two- and three-year outcomes), suggesting that sustainability and governance factors exert a delayed, but meaningful impact on firm value.

While the findings from the SHAP and DBN analyses do not perfectly mirror those of the regression model—likely due to their ability to capture non-linear relationships not accounted for by linear models with first-order effects—they broadly support our main conclusions: ESG disclosure topics are associated with both immediate and lagged firm value outcomes.

5. Discussion

Our primary analysis employed linear regression models to test hypothesized relationships between specific sustainability topics, report assurance, and both current and lagged measures of firm value. This approach aligns with explanatory modeling, which seeks to identify patterns and gain insight into existing phenomena. To complement and validate our regression results, we also applied two independent machine learning techniques—Bayesian networks and SHAP—for descriptive and predictive modeling. These methods offer alternative perspectives on the link between sustainability reporting and firm value while also serving as robustness checks, mitigating the risk that our findings were driven by statistical artifacts (Venkatesh et al., 2013). Together, our use of explanatory and predictive modeling reflects a dual analytical lens: one that seeks to understand “what is” and another that forecasts “what could be.”

Our study’s underlying assumption was that companies engage in ESG activities with the aim of benefiting society. We argue that private businesses have a vested interest in pursuing societal good due to institutional pressures and isomorphic change (Slack & Hinings, 1994), in addition to the instrumental perspectives (I. Freeman & Hasnaoui, 2011; Cornell & Shapiro, 1987; Donaldson & Preston, 1995; Platonova et al., 2018), and strategic CSR (Baron, 2001; Bondy et al., 2012). These ontological assumptions are partially supported when we examine the association of sustainability topics and the firm’s current value. However, in some instances, we found stronger results when considering the time-lagged effect. These findings are consistent with our hypotheses, even though we observed some deviations.

We observed a positive association between environment and socio-economic topics and Lead-2 Tobin’s Q when no assurance was sought. While the lag effect was in line with our expectations, it is noteworthy that the influence of environment and socio-economic topics did not sustain beyond the second year. Evidently, the association of environment and socio-economic topics and firm value as measured by Tobin’s Q is not self-sustaining. This suggests that while legitimacy may be acquired over time, organizations need to engage actively in maintaining it. This observation aligns with Suchman’s (1995) conceptualization of legitimacy management, involving gaining, maintaining, and repairing legitimacy. We also observed that daily necessities and healthcare had an immediate and sustained effect on firm value, although the impact tapered after the second year. These differences likely reflect varying levels of stakeholder salience and topic visibility; for instance, healthcare disclosures may generate faster reputational benefits, whereas environmental initiatives typically entail longer-term strategic investment.

We also hypothesize that third-party assurance enhances the credibility and trustworthiness of sustainability disclosures. However, our findings present a more complex picture. While assurance acted as a moderating factor for most topics (with the exception of sustainable consumption), its impact on firm value was not uniformly positive. In fact, we observed a negative moderating effect in the year following disclosure for both environmental and socioeconomic topics, with no significant effect in subsequent years. This suggests a dual-pathway effect: on one hand, assurance may enhance perceived credibility; on the other, it may increase investor scrutiny and highlight inconsistencies, leading to skepticism.

These counterintuitive findings warrant further exploration within the framework of legitimacy theory. We find it crucial to consider two central traditions within this theory: the strategic tradition, focusing on how organizations manipulate symbols for societal support, and the institutional tradition, which underscores cultural pressures beyond organizations’ control (Suchman, 1995). Both perspectives offer plausible interpretations. From an institutionalist stance, the absence of mandatory sustainability reporting standards in the U.S. may reduce external pressure, causing voluntary disclosures to be met with investor skepticism. Conversely, the strategic perspective raises questions about the materiality and communicative effectiveness of the disclosures themselves, particularly if assurance is perceived as a superficial signal or even a form of greenwashing. The potential for third-party verification to backfire in such contexts underscores the contingent and sometimes paradoxical nature of assurance in sustainability reporting.

6. Limitations and Future Research Directions

Legitimacy theory is a complex and multidimensional phenomenon that can help us understand the real value of sustainability reporting. Applying the full spectrum of legitimacy theory, from the institutionalist perspective (external) to a strategic theorist perspective (internal), may allow researchers to explore the various dimensions of sustainability topics and gain insights into why and how firms engage in and disclose their sustainability practices or why they may assure or not assure their sustainability reports.

Makadok et al. (2018) emphasize the significance of boundary conditions within theories, which define the scenarios where a theory holds its effectiveness and those where its relevance weakens or disappears. The unexpected outcomes of our study could suggest limitations based on context. Could it be plausible that, without institutional pressure, the adoption of sustainability themes and the strategic decision to assure a report offer no discernible advantages or even have adverse effects? Replicating this study in a heavily regulated environment might provide valuable insights.

Future studies should explore functional fixation theory as an alternative conceptual framework to illuminate some of these findings further. Functional fixation refers to a conditioning effect where individuals become fixated on using certain data for specific functions, which may hinder their ability to use the same data for different purposes or their inability to incorporate additional data beyond traditional axioms (Haka et al., 1986). We opine that functional fixation theory may explicate some of the counterintuitive findings. For instance, in the context of assurance, we found a negative moderating effect on firm value a year after reporting on environmental and socio-economic topics. While we expect a lagged effect, the directionality is surprising. Possible explanations include impending regulatory changes or users’ naivety or insufficient understanding of ESG data. While some scholars attribute fixation to a paucity of experience or pertinent data (K. C. Chen & Schoderbek, 2000; Gupta & King, 1997; Waller et al., 1999), Luft and Shields (2001) discovered that accounting knowledge does not necessarily mitigate fixation on accounting—accounting itself influences the learning process. Their perspective suggests that stakeholder sophistication and comprehension play a pivotal role in data fixation.

The intricate dynamics of decision-making and data fixation in ESG engagement necessitate an intricate understanding through multiple theoretical lenses. The acknowledgment of the delayed benefits inherent in sustainability investments highlights the relevance of theories like functional fixation and diffusion of innovations. In-depth exploration of these theoretical frameworks in future studies promises to enhance the comprehensive understanding of sustainability topics.

From a methodological viewpoint, we used the latent Dirichlet allocation (LDA) method to identify sustainability topics within sustainability reports. LDA is a statistical model that identifies topics based on recurring words throughout all documents and then characterizes each document using topic probabilities (Blei et al., 2003). From our analysis, we recognized six topics in the corpus of sustainability reports, which we categorized as environmental, sustainable consumption, daily necessities, operations and logistics, socio-economic impact, and healthcare. We omitted the operations and logistics topic from model inputs to avoid linear dependency errors. However, as with other studies employing unsupervised topic modeling methods, there are certain limitations. First, LDA topics require more interpretation by the researcher, as sustainability reports encompass both qualitative and quantitative elements. Second, changes in topics within sustainability reports in this study are due to reasons other than regulation, as this area is not regulated (Muslu et al., 2019). Third, we acknowledge that some sectors may emphasize certain topics in their sustainability reports more than others: for instance, the healthcare sector may focus more on healthcare topics. Even though industry classification is mandatory, topic prioritization in sustainability reports is voluntary. We accounted for these potential variations by controlling for industry sectors in our model, thereby highlighting the effect of topic distributions in sustainability reporting on firm value. Fourth, our current research focused on market response to the broad discussion themes of ESG disclosures, rather than verified ESG performance outcomes. Future studies could incorporate linguistic deception cues and disclosure consistency metrics to better differentiate symbolic ESG behavior (e.g., sustainability reporting of ESG topics) versus substantive ESG behavior (i.e., third-party ESG ratings), which is beyond the scope of our current study. Lastly, Tobin’s Q is a useful measure of a firm’s valuation; however, it is not devoid of criticisms, similar to other popular quantitative measures. While Tobin’s Q formula is intuitive and proves to be a reliable indicator of effective management and operational practices of firms, it makes assumptions including perfect competition, frictionless markets, industry bias, and equal access to information.

7. Recommendations and Conclusions

This study advances theoretical, practical, and policy-oriented understanding of sustainability reporting by applying a multi-theoretical framework that included signaling theory, agency theory, stakeholder theory, and organizational legitimacy theory, alongside functional fixation theory to interpret time-lagged effects (Corley & Gioia, 2011). Our findings underscore the complex and intersectional nature of sustainability disclosures and affirm the value of analyzing not only whether firms report ESG information, but what they report and how those disclosures are received by the market.

Empirically, we show that the impact of sustainability topics on firm value is neither uniform nor enduring. Topics such as daily necessities and healthcare exhibited immediate and sustained positive effects on Tobin’s Q for up to two years, while environmental and socio-economic impact topics influenced firm value only after a two-year lag. However, none of the topics showed significant influence beyond the third year, suggesting that sustainability disclosures require consistent renewal to maintain investor relevance and legitimacy (Suchman, 1995). Assurance, although traditionally viewed as a credibility-enhancing mechanism, demonstrates mixed effects: when applied to environmental and socio-economic topics, it may inadvertently highlight disclosure gaps or performance shortfalls, thus triggering investor skepticism (Simnett et al., 2009; Christensen et al., 2021). As depicted in

Table 6. Summary of recommendations by stakeholder groups.

Stakeholder Group	Recommendations
Corporate Managers and Sustainability Officers	Prioritize *healthcare* and *daily necessities* in sustainability reports to enhance short-term firm value. Maintain consistent reporting on key topics to preserve investor attention and legitimacy. Apply third-party assurance judiciously, particularly for sensitive topics like environment and socio-economic issues.
Investors and Analysts	Incorporate a lag-based view (1–2 years) when assessing the impact of ESG disclosures on firm value. Evaluate the quality and substance of assured reports: assurance may signal risks if disclosures are weak or symbolic.
Policymakers and Regulators	Promote topic-specific ESG disclosure standards to improve comparability and relevance. Recommend annual or biennial reporting intervals to align with observed value timelines. Standardize assurance practices to ensure consistency and mitigate perceived greenwashing.
Researchers and Standard Setters	Explore theoretical extensions using functional fixation and legitimacy theory to explain counterintuitive effects. Replicate the study across regulatory contexts (e.g., EU, Asia) to test generalizability.

, these findings carry important implications for multiple stakeholders.

Corporate managers and sustainability officers should prioritize topics with demonstrated market impact—particularly healthcare and daily necessities. Consistent and credible reporting on these topics can enhance transparency, reputation, and stakeholder trust. Assurance should be applied strategically: when disconnected from substantive performance, it risks sending adverse signals that undermine firm value.

Investors and analysts are encouraged to adopt a time-lagged valuation perspective, recognizing that certain sustainability topics (e.g., environmental, socio-economic impact) deliver delayed financial benefits. Caution is warranted when interpreting assurance: it should not be assumed as a universal signal of quality, but rather contextualized within the depth and materiality of the disclosures it verifies.

Policymakers and regulators should consider moving beyond monolithic ESG mandates and instead promote topic-specific and industry-sensitive reporting standards. Given that the observed valuation effects wane after two to three years, setting annual or biennial reporting intervals may enhance disclosure effectiveness. Additionally, standardizing assurance practices could improve stakeholder confidence and reduce the risk of symbolic compliance.

Researchers and standard setters are urged to further investigate the boundary conditions under which assurance enhances or detracts from firm value (Makadok et al., 2018). Integrating functional fixation theory (Haka et al., 1986; Luft & Shields, 2001) could explain why investors may undervalue or misinterpret certain sustainability disclosures. Moreover, replicating this study in contexts with more stringent ESG regulations—such as the EU or Asia—could help assess the generalizability of our findings beyond the relatively light-touch regulatory environment of the U.S.

Although our study focused on U.S. firms, we controlled for industry-level variation and excluded country-level effects to mitigate contextual bias. Nonetheless, the inherent complexity of sustainability research, including the challenges of variable isolation and causal inference, should warrant caution against overgeneralization. Our results emphasize the need for dynamic, data-informed strategies in ESG communication and assurance, and suggest that sustainability’s contribution to firm value is best understood as conditional, topic-specific, and temporally dependent. Table 6 below provides a summary of recommendations by stakeholder groups.