Booking Sustainability: Publicly Traded Companies as Catalysts for Public Goods Provision in Brazil

Abstract

This study assesses the extent of public goods provision by Brazilian firms and how this behavior has changed over time. We use text data of publicly traded companies’ annual standardized financial declarations from 2010 and 2022 and apply natural language processing techniques to extract ESG (environmental, social, and governance) keywords related to the provision of public goods. Context and sentiment analyses were used to supplement the information extracted from the raw keyword counts; these analyses were conducted using diverse regression techniques. We found a pronounced increase in keyword mentions over time; in particular, “responsibility” and “sustainability” appeared more frequently. Virtually all firms became more dedicated to ESG practices, particularly those that had a low frequency of ESG mentions in a positive context. Overall, it seems that large Brazilian corporations have embedded comprehensive ESG policies into their business practices, thus aligning their strategies with those of pioneering multinationals.

1. Introduction

Clean air, security in public places, and the Amazon rainforest are examples of public goods, which is defined as their use being both non-excludable and non-rival (Samuelson 1954). A well-known problem surrounding public goods is their under-provision; everyone can exploit them, but individuals have little incentive to contribute to their existence. At present, the recognition of humanity’s detrimental impact on the environment—driven by business activities that deplete natural resources for private goods production—has reached such a level that the current geological epoch is often referred to as the Anthropocene. This term reflects the profound influence of human actions on the Earth’s systems, underscoring the urgent need to change our behavior and production practices (Keys et al. 2019).

Thus, the following questions arise: How can society increase the amount and quality of public goods such that overall welfare gains can be realized and the risks of the Anthropocene are mitigated? Can corporations and their private capital be instrumentalized to contribute to the supply and conservation of public goods? Liberal economists in the tradition of Milton Friedman argue that the only social responsibility of private firms is to increase their profits. Empirical studies seem to have found that a positive relation between investments in social projects and the profits of firms is valid only under certain restrictions (Flammer 2015; Servaes and Tamayo 2013). However, when a firm’s shareholders have a preference for socially responsible behavior, it may be rational for its managers to devote profits to the provision of public goods (Morgan and Tumlinson 2019). Another explanation may be that firms are competing for socially responsible consumers (Bagnoli and Watts 2003).

The present study provides novel evidence surrounding the provision of public goods1 by companies in Brazil. We applied natural language processing techniques to extract a unique dataset of keywords from firms’ standardized financial declarations. These keywords allowed us to determine the ESG policies that are related to the provision of public goods. We then used quantitative and qualitative methods to analyze the penetration of different ESG-related topics among Brazilian firms. We also documented how the pattern has changed over time using data from 2010 and 2022. To the best of our knowledge, there is no comparable study that summarizes ESG policies and the change in patterns among firms in the Brazilian market. Another contribution of the present study is the innovative methodological approach that is used; this approach shows how progress can be made, even in the absence of established metrics for ESG policies at the firm level, without the need to conduct interviews or other direct forms of primary data collection.

The following arguments highlight the contributions and relevance of the present study. First, the extensive natural resources and biodiversity of Brazil mean that we can position sustainability as a vital concern for its companies, presenting a distinctive opportunity for them to take a leadership role in this domain (Possebon et al. 2024). Second, Brazil presents a compelling case for studying public goods and sustainability, given its huge internal market and status as a developing country. Both in terms of population size and aggregate GDP, Brazil ranks among the top ten countries worldwide. Third, the nation has a significant population of consumers with relatively low purchasing power (de Almeida et al. 2021), who often prioritize price over other attributes of a product and its producer. These dynamics pose unique challenges for the promotion of sustainability practices among firms, as consumer demand for ESG (environmental, social, and governance) factors may be less pronounced than it is in wealthier economies. Additionally, the well-documented positive correlation between environmental protection and income extends to consumer behavior regarding ESG practices (Magnani 2000). As income levels rise, we expect that consumers will tend to place greater emphasis on environmental and social responsibilities, making Brazil a key context for examining how these factors interact in an emerging market setting. Fourth, there is increasing interest in ESG issues from the financial market, recognizing both the need for greater sustainability and the opportunity for innovative practices in companies.2 For example, there is a specific green bond market in Brazil which, despite some relevant growth in recent years, faces structural challenges (see Yamahaki et al. (2022)). Finally, the provision of public goods by private entities gains greater relevance in developing economies, due to institutional weaknesses and resource constraints (Bradly and Nathan 2019). The present descriptive study offers an orientation for investors, entrepreneurs, and policy makers interested in the recent development of ESG policies among major Brazilian firms.

To document the voluntarily provision of public goods by the market, we relied on information provided in publicly traded firms’ standardized financial declarations. For the purpose of conducting a rigorous empirical analysis, we needed to define a few informative keywords that captured the activities of firms with respect to public goods provision. Natural language processing techniques allowed us to extract keyword count variables, which were then analyzed with graphical tools, locally weighted regression, and quantile regressions. The proper selection of keywords is arguably a subjective step and involves a trade-off between ensuring concision and being too general, possibly going beyond the scope of public goods. We selected the following seven keywords, which we believe provide a fair approximation of the topic: “education”, “health”, “environment”, “sustainability”, “responsibility”, “transparency”, and “poverty”. The first four—“education”, “health”, “environment”, and “sustainability”—were deliberately chosen due to their recognizability as public goods, defined by their non-exclusivity and non-rival characteristics. These terms were crucial for examining how resources that collectively benefit society are distributed and maintained. Conversely, the remaining three keywords—“responsibility”, “transparency”, and “poverty”—were selected for their frequent association with the provision and governance of public goods. These terms were pivotal in exploring how societal efforts and policies contribute to the enhancement of public welfare. We complement the keyword count analyses with a contextual description using word clouds and sentiment analyses, following the approach taken in Mohammad and Turney (2010).

We found that the 363 publicly traded firms in our sample have increasingly cited public-goods-related keywords. In 2010, there was an average of 20 keyword mentions in the management report part of the financial declarations. In 2022, that number was equal to 32, corresponding to a 60% increase. Among the most-cited keywords that drove the observed expansion were “responsibility” and “sustainability”. Our results indicate that the increased reporting of public goods is supported by virtually all market participants, not just by a few key players. It is also interesting to note that those firms with the lowest mentions of keywords in a positive context in 2010 registered the highest growth rates for those keyword counts. Hence, there seems to be broad change in the market and a substantial recognition of the necessity to provide public goods.

Our context analysis revealed a transition from broad sustainability themes to more targeted and strategic areas, indicating a deeper integration of these concepts into companies’ overall planning and operations. This shift may reflect a more holistic approach, embedding sustainability into the core of business practices and long-term objectives. Despite this evolution, the consistent presence of terms related to social and environmental factors underscores the ongoing acknowledgment of their interconnectedness.

Our findings are well-aligned with individual perceptions about climate change, as the recent poll by Spektor et al. (2023) indicates. While most Brazilians accept that climate change is real and driven by human activity, opinions are split on the seriousness of its effects. Of the total population, 56% believe that climate change will have a negative impact on their daily lives. The perceived changes made by large Brazilian companies show that the country is increasingly aligning with the global trend of prioritizing sustainability as a key business strategy, reflecting a growing awareness of the importance of responsible and sustainable practices. This trend was substantially fueled by the introduction of a radically different corporate model in 2009, when Paul Polman—the CEO of Unilever—committed to reducing the company’s environmental impact, while enhancing its social contributions (Polman and Winston 2022). The readiness of society to embrace this innovative approach allowed Unilever to achieve returns that were significantly higher than those of its competitors and to establish itself as one of the most respected companies globally.

The present findings complement the literature on ESG practices by Brazilian companies, which has highlighted many aspects that are pertinent to this discussion; however, a comprehensive review of ESG policies is absent. A number of papers have related ESG practices to firms’ profitability, risk, or contagion (Fdez-Galiano and Feria-Dominguez 2024; Garcia et al. 2017). We are unaware of other research on ESG practices in Brazil using keyword, contextual, and comparative analyses, as we undertook in the present study.3 The studies by Possebon et al. (2024) and Carvalhal and Nakahodo (2023) examined the relation between firms’ ESG practices and returns on the stock market. Schleich (2022) focused on human resource policies and showed that, although many leading companies have implemented ESG elements, there remains a considerable disconnect between their designs and real-world applications. Measures to improve the health and safety of employees are among the most common measures, but many firms also involve their stakeholders and suppliers. In other areas, such as diversity and discrimination management, much more progress is needed. The study by Martins (2022) used firm-level data to compare the Brazilian market with other emerging economies, focusing on the role of competition. He found that a large external negative shock—one which disrupted international competition—forced Brazilian firms to reduce their investments in ESG. Other studies had previously found a positive role of competitive pressure in ESG policies; however, this pressure was imposed by public opinion, peers, or investors (Baldini et al. 2018; Matos 2020).

Most of these studies used numeric indicators such as investment values or ESG scores, whereas we used self-reported but audited information, contained in text form. Our objective text metrics permitted a stable comparison over time that was not subject to subjective classification of firms’ specific projects. See Kotsantonis and Serafeim (2019) for an insightful discussion on the difficulties and inconsistencies that are encountered when measuring ESG practices. The authors of Berg et al. (2022) presented evidence on the diverging results of ESG ratings by different agencies.

The following papers on ESG reporting in other countries are particularly similar to ours, as they also used NLP techniques. Research in this direction, however, is still in its infancy. The authors of Gupta et al. (2024) used NLP models with the aim of predicting the current ESG focus of Indian firms based on information disclosed in their sustainability reports. Maibaum et al. (2024) reflected on methodological issues and the performance of different techniques, including large language models, in extracting information on firms’ sustainability politics from unstructured data, such as annual reports. We avoided such methodological problems by relying on a simplified keyword extraction approach. Pikatza-Gorrotxategi et al. (2024) relied on information in news articles and linked the style of ESG-related reporting to changes in companies’ reputation over time. Finally, Schimanski et al. (2024) classified ESG-related texts and trained an NLP model, which was then shown to be capable of explaining the differences between existing ESG ratings.

The rest of this paper is organized as follows: Section 2 explains the methodology and data used in the present empirical research. Section 3 presents the results of our keyword count and context analyses. Section 4 concludes the paper and points to avenues for future research.

2. Materials and Methods

The research question of the present study was as follows: Are publicly traded companies increasingly dedicated to providing public goods? To this end, we used a context analysis approach based on textual data from standardized financial statements published in 2010 and 2022. The basic idea is that these statements reveal significant and reliable information about a firm’s realized or planned operations. We defined several keywords that are related to public goods, and then applied both qualitative and quantitative methods to analyze the changes in the frequency of and the context around these keywords.

2.1. Data

We focused on publicly traded companies for the following reasons: First, publicly traded firms are much larger and have a much lower exit rate compared to other types of companies (Ehrl 2021). Therefore, we argue that the sample selection was not related to the provision of public goods. Only 3 of the 366 firms identified in the year 2010 were not active anymore in 2022. Second, public companies are required to present standardized annual documents, such as the standardized financial declaration (SFD). The SFD is a mandatory document under art. 133 of Law 6404/76, and it has a standardized structure—despite some degree of discretion regarding the content—in which companies communicate their main results to their shareholders. A firm can thus choose whether or not to include descriptions referring to the provision of public goods by the company. The structure and purpose of the SFD has not changed over time, permitting a consistent analysis of the years 2010 and 2022. Third, the SFD is audited by external auditors and receives substantial attention from its stakeholders, which means that the reported information is highly reliable and representative of the firm’s actions.

The SFDs were downloaded manually from the website of the CVM. In Brazil, the CVM (Comissão de Valores Mobiliários) is the Securities and Exchange Commission that is responsible for regulating and overseeing the securities market, ensuring its transparency, integrity, and the protection of investors. The SFD documents bring together different information, such as registration data, financial statements, and management reports, which were exploited in our context analysis. The management report is one the main sources of annual information for the shareholders of companies and other strategic audiences. This report provides a comprehensive view of a company’s short-, medium-, and long-term strategies and directly impacts shareholder decisions surrounding investments in a company (Silva et al. 2007).

Storage, naming, and identification of the pages of the management reports were undertaken manually, which limited the scope of this research to two years; in particular, 2010 and 2022 are the first and last years available for which the CVM offers SFDs with complete information about the status of the companies. Note that not all publicly traded firms included a management report in their SFD in both years, limiting the number of companies included in this analysis to 363.

2.2. Keyword Extraction

Through a bibliographical review of texts that used the criteria of exclusivity and rivalry to define their types of goods, we had to make a choice regarding the terms that best represented public goods for the subsequent content analysis. The items that stood out were related to the following categories: (i) social actions seeking to deliver education to local communities and employees; (ii) health policies for employees and assistance to communities in need4; (iii) actions aimed at environmental education and/or application of sustainable practices; and (iv) investments through a companies’ own foundations or third parties, as a way of supporting the provision of public goods by these foundations.

These evaluations led to four obvious candidates for the keywords, namely “education”, “environment”, “health”, and “sustainability”. Another key term that seemed essential was (social) “responsibility”. The term “poverty” should account for the consideration of communities in need. Finally, “transparency” is a general aspect of public goods provision and it indicates a degree of dialogue with stakeholders external to the company, and this was included as the seventh keyword, although the frequency of occurrences show that the two latter terms were by far the least utilized. At the top of the list in

Table 1. Frequency of keywords.

	(1)	(2)	(3)	(4)	(5)
Keyword	2010	(in %)	2022	(in %)	Total
education	725	10.14	789	6.85	1514
environmental	2046	28.63	2276	19.77	4322
health	1002	14.02	1833	15.92	2835
poverty	22	0.31	23	0.20	45
responsibility	2546	35.62	4691	40.75	7237
sustainability	546	7.64	1466	12.73	2012
transparency	260	3.64	435	3.78	695
Total	7147	100	11,513	100	18,660

Notes: The table shows the aggregate absolute and relative frequencies of keyword occurrences for the years 2010 and 2022, extracted by NLP techniques from 363 firms’ standardized financial declarations.

are “responsibility” and “environmental”, with 7237 and 4322 occurrences, respectively, in all SFDs from 2010 and 2022.

The appearance of the terms can be exemplified by the following excerpts from three administration reports: “Our foundation is the belief that education is the path to promoting equal opportunities, personal and collective fulfillment, as well as the means to build a dignified, fair, and productive society”. (Banco Bradesco, 2022);

“As we are a company that advocates health and well-being, we invest heavily in healthy and educational actions”. (Pague Menos, 2010); “Guidelines (...) reduce the impact of effluents, the population and waste in the environment” (CAGECE, 2022).

The algorithm that we developed to perform the context analysis was based on natural language processing (NLP) methods in the programming language R, and proceeded as follows for each firm and year:

• Loading the text within the management report section of the SFD using the pages identified from the index.
• Removing the head and footer line.
• Cleaning the text involved removing numbers, special characters, common abbreviations, and single letters.
• Splitting the text into sentences using the remaining “.” characters.
• Counting all remaining words.
• Keeping only sentences that contained at least one of the seven keywords.
• Counting the frequency of each keyword.
• Performing a sentiment analysis (described in detail below) and classifying each sentence according to its context as positive, negative, or neutral.
• Counting the frequency of each keyword in each context.

The extracted keyword counts were used to derive descriptive statistics in the form of key statistical moments, such as the mean and the standard deviation. We also provide graphical analyses of the entire keyword distribution at the firm level.

2.3. Quantile Regressions

We completed the investigation into shifts in the distribution of public-goods-related keywords with the following quantile regressions, according to Koenker and Bassett (1978):

$$\widehat{\beta }\left(\tau \right)=\underset{\beta }{\mathrm{argmin}}\sum {\rho }_{\tau }(y_{it}-\mathbb{I}{(year=2022)}_t^{\top }\beta )$$

(1)

where $y_{it}$ is either the frequency of one specific keyword or all keywords relative to the total number of words in the management report of firm i in year t. Due to the large number of total words, we multiplied our dependent variable by 10,000. A coefficient of $\beta =1$ would thus indicate that, in 2022, for each 10,000 words in the report, the keyword appears one more time than it did in 2010. ${\rho }_{\tau }$ is the check function for the quantiles $\tau \in [15\%;30\%;50\%,75\%;85\%]$; this determines the chosen percentile at which the distribution is analyzed. The only explanatory variable is an indicator for the year, equal to 2022. Therefore, the coefficient estimate for $\beta \left(\tau \right)$ indicates whether the distribution at the percentile $\tau$ shifted upwards, downward, or remained stable over time.

Quantile regressions complemented our analysis, as they allowed us to explore how the shifts over time varied across different points in the distribution of the use of “sustainability”. Unlike ordinary least squares (OLS) regression, which focuses solely on the mean, quantile regressions provide insights into the entire distribution, revealing shifts that may be significant for firms that feature a low, medium, or high usage of “sustainability”. This approach was particularly relevant for our study, as it enabled us to identify whether changes in word usage over time were not only evident in an visual comparison across distribution, but whether the effects were statistically significant. For instance, a term may exhibit a notable increase in usage in the upper quantiles, while remaining stable in the lower ones, indicating diverging patterns that OLS could obscure. These quantile regressions thus provided a rigorous econometric tool to measure how much the distribution of keywords shifted and the increased the robustness of our findings.

2.4. Sentiment Analysis

Sentiment analysis is a computational NLP technique used to determine the emotional tone behind a body of text. There are numerous approaches and techniques that can be used to perform sentiment analysis (see Medhat et al. (2014) for a review). Existing algorithms are capable of identifying several sentiments like anger, joy, or sadness. While the most sophisticated results are achieved in the most-utilized language, English, software packages cope well with a wide range of other languages as well.

To analyze our research question, we aimed to classify the context of sentences from the management reports as either positive, negative, or neutral. This approach kept the margin for error low and permitted a relatively simple, transparent, and objective analysis. We followed the approach by Mohammad and Turney (2010) and applied the R package syuzhet: get_nrc_sentiment, which is also available in the Portuguese language. This package is lexicon-based and, thus, does not rely on classifications from machine learning.

In fact, the authors of Mohammad and Turney (2010) manually compiled a moderate-sized lexicon of words with their sentiment context. They asked participants online to describe a given word as positive or negative (among other sentiments) in exchange for a monetary reward. Their procedure thus used “human intelligence tasks”. They also carefully excluded potentially false responses and thus derived a relatively reliable tool for sentiment analysis that is frequently applied in different research fields. The original English lexicon is available in many other languages. The one utilized in the present research in Portuguese language features 13,901 classified words.

3. Results

3.1. Keyword Counts

Our first result was derived from the number of keyword occurrences and their change over time. Table 1 shows the aggregate absolute and relative number of keyword occurrences, while

Table 2. Descriptive statistics—keywords per firm.

	(1)	(2)	(3)	(4)
Variable	Mean	Std. Dev.	Min.	Max.
	Panel A: Year 2010
all words	25,059	16,874	2115	90,287
all keywords	19.69	24.47	2	219
education	2.00	7.30	0	125
environmental	5.64	10.72	0	112
health	2.76	6.13	0	56
poverty	0.06	0.34	0	4
responsibility	7.01	4.85	2	29
sustainability	1.50	3.41	0	26
transparency	0.72	1.33	0	9
	Panel B: Year 2022
all words	31,682	20,797	3110	156,798
all keywords	31.72	31.65	5	195
education	2.17	9.04	0	141
environmental	6.27	9.84	0	65
health	5.05	10.01	0	82
poverty	0.06	0.35	0	4
responsibility	12.92	7.42	4	58
sustainability	4.04	6.48	0	48
transparency	1.20	2.02	0	16

Notes: The table shows the absolute frequency of keyword occurrences and the total number of words in the financial statements of each firm for the years 2010 and 2022. The number of firms in each year is equal to 363.

displays the descriptive statistics that were calculated at the firm level. Both Tables clearly show that firms used more words related to public goods in their annual reports. The 363 firms under consideration mentioned the keywords 7147 times in 2010 and 11,513 times in 2022. These frequencies correspond to an average of 19.7 and 31.7 times per report, respectively. The variation among firms was quite high, as the standard deviation in both years was at least as large as the mean. Only one of the keywords—“responsibility”—was used by all firms in both 2010 and 2022.

“Sustainability” and “responsibility” are the keywords that showed the highest growth in usage over time. On the other hand, mentions of “education”, “environmental”, and “poverty” showed little or virtually no increase between 2010 and 2022. Note that the standard deviations of these keyword counts are relatively high compared to those of the other keywords, indicating great heterogeneity among the firms in this respect. These findings are remarkable, given that a lack of higher/technical education and poverty/inequality are key areas in which Brazil needs to catch up to developed countries (de Almeida et al. 2021). Moreover, environmental protection is a recurring issue of public debate in the European Union, and this did not seem to receive much attention in Brazil according to the management reports.

Table 2 also contains the total number of words in the management reports and indicates that these texts are becoming more extensive in general. The average number of words in 2010 was 25,059 and grew to 31,682 per firm. The variation in text length was substantial, as the minimum, maximum, and standard deviation indicate. Consequently, we see that the frequency of keywords related to public goods increased over time, but the management reports were also lengthier. It is thus debatable whether firms are simply feeling the need to publish more extensive reports for the public or whether, in fact, firms are becoming more dedicated to public goods provision. In any case, we devote attention to this issue by analyzing the frequency of keywords relative to the total number of words in the following steps.

3.2. Keyword Distribution Across Firms

The descriptive statistics presented above only gave us a few indications about the dispersion of keyword occurrences across firms. Figure 1 reveals more about the heterogeneity among the publicly traded companies. The left graph in Figure 1 plots the histogram for the variable ”all keyword occurrences” in each management report separately for the two years under consideration. Recall that, as observed in Table 2, all of the firms reported at least two keywords.

The X-axis in Figure 1 represents the total number of keyword occurrences, while the Y-axis indicates the number of firms that fall within each occurrence range. This visual representation allows for an assessment of how keyword usage varied among firms, highlighting patterns such as the concentration of firms with low, moderate, or high keyword occurrences. We see that the distribution was highly right-skewed and peaked at around 7 in 2010 and at around 12 in the year 2022. However, the number of firms that fell into these categories diminished over time. In 2010, there were 131 firms in the first bin, with between two and seven mentions of “sustainability” in their management reports. In 2022, this the number of firms in the same category diminished to 45. The median values are thus considerably smaller than the means, which also implies that the distributions have long, thin right-hand tails. In 2010, 33 companies used more than 50 keywords and 5 of them used more than 100; these numbers increased to 76 and 15, respectively, in 2022.

Overall, the previous considerations, as well as the data graphically represented in Figure 1, show that the distribution of keyword occurrences shifted to the right. We can thus report that the increases in the average and total occurrences of “sustainability” were not driven by just a few firms. The entire segment of publicly traded firms produced standardized financial statements with more extensive discussions of public goods.

This right-shift was particularly strong for the keyword “responsibility”. The right-hand graph in Figure 1 shows that fewer firms used the expression less than nine times (which is where the two distributions intersect). The upper tail of the 2022 distribution is far more outstretched and it unites more reports with high occurrences. For example, only nine companies used the word “responsibility” more than 20 times in 2010, while this number grew to 46 by 2022. The distributions of the remaining keywords are omitted for brevity, but they showed similar results, although there were less-pronounced shifts over time.

Table 3. Quantile regressions—keyword distributions.

Percentiles:	(1)	(2)	(3)	(4)	(5)
	15	30	50	70	85
	dep. var.: relative frequency of all keywords
Year = 2022	2.128 ***	2.260 ***	3.087 ***	2.712 ***	1.958 **
	(0.248)	(0.365)	(0.413)	(0.613)	(0.786)
Observations	726	726	726	726	726
Pseudo $R^2$	0.055	0.039	0.045	0.025	0.011
	dep. var.: relative frequency of “education”
Year = 2022	−0.087 **	−0.096	−0.100	−0.100	0.052
	(0.036)	(0.071)	(0.105)	(0.201)	(0.263)
Observations	300	300	300	300	300
Pseudo $R^2$	0.005	0.002	0.001	0.001	0.000
	dep. var.: relative frequency of “environmental”
Year = 2022	−0.038	−0.230	−0.311	−0.610	−1.413 **
	(0.119)	(0.150)	(0.197)	(0.483)	(0.576)
Observations	464	464	464	464	464
Pseudo $R^2$	0.000	0.004	0.004	0.007	0.020
	dep. var.: relative frequency of “health”
Year = 2022	0.092 **	0.056	0.224	0.224	0.680
	(0.044)	(0.086)	(0.144)	(0.289)	(0.485)
Observations	404	404	404	404	404
Pseudo $R^2$	0.004	0.000	0.004	0.001	0.010
	dep. var.: relative frequency of responsibility
Year = 2022	1.079 ***	1.125 ***	1.367 ***	1.696 ***	2.577 ***
	(0.092)	(0.145)	(0.182)	(0.267)	(0.558)
Observations	726	726	726	726	726
Pseudo $R^2$	0.090	0.062	0.057	0.052	0.066
	dep. var.: relative frequency of sustainability
Year = 2022	0.137 ***	0.277 ***	0.501 ***	0.737 ***	1.242 ***
	(0.047)	(0.082)	(0.128)	(0.269)	(0.319)
Observations	345	345	345	345	345
Pseudo $R^2$	0.011	0.017	0.031	0.037	0.061
	dep. var.: relative frequency of “transparency”
Tear = 2022	−0.015	0.019	0.009	0.072	0.014
	(0.027)	(0.031)	(0.050)	(0.074)	(0.139)
Observations	299	299	299	299	299
Pseudo $R^2$	0.001	0.001	0.000	0.001	0.000

Notes: Table 3 reports the estimated coefficients for the indicator variable for the year 2022 according to Equation (1). Each of the coefficients stem from a separate quantile regression, where the dependent variable is the firm level count of the keywords relative to the total number of words (in ten thousands) in the management reports, as indicated at the top of the different panels in the table. Coefficients were estimated using the maximum likelihood and the corresponding pseudo R² indicates the fit of the regression. Standard errors are shown in parentheses. ** p < 0.05, *** p < 0.01.

reports the estimated coefficients for the indicator variable for the year 2022, according to the quantile regression in Equation (1). Each of the coefficients stems from a separate quantile regression, where the dependent variable is the share of all keywords in the management reports of each firm, as indicated at the top of the different panels in the Table; that is, to complement the previous analysis, the quantile regressions used the count of keywords relative to the total number of words in the management report, multiplied by 10,000 to avoid tiny coefficients. According to Table 2, one out of every thousand words was a keyword related to public goods. The estimated coefficient in the first regression in Table 3 (2.1) can thus be interpreted as follows: In the 15th percentile of the distribution of the relative keywords count, we observe that, compared to the year 2010, in 2022, firms mentioned 2.1 more keywords per 10,000 words. Given that the average length of a report was about 30,000 words, the overall increase by more than six mentions in relation to public goods is quite substantial. Importantly, the estimated coefficients for the distribution of all keywords are highly significant at every percentile analyzed. We can thus confirm with statistical rigor that the entire distribution shifted to the right. Therefore, it is not only the average firm but, instead, the entire market in Brazil that is now more concerned about sustainability than it was a decade ago.

Once we analyze the occurrence of keywords separately, the changes over time are quite heterogeneous. The overall increase in public goods keywords was driven by the words “responsibility” and “sustainability”. The estimated coefficients for the distribution of “responsibility” are particularly large and close to the numbers observed for all keywords. No statistically significant increase was observed for mentions of “education”, “health”, “environmental”, and “transparency”. We conclude that it is not possible to identify a specific area where the firms’ actions and investments were directed. However, it is commendable that public enterprises see themselves as being in a position of greater responsibility and communicate their role to the stakeholders.

3.3. Keywords in Context

The following exercises are presented with the intention of providing more context to the simple keyword occurrence counts analyzed thus far. We begin by presenting word clouds using the 50 most frequent terms that appeared in sentences with at least one of the keywords. Words in larger font and in the center of the clouds had a higher importance because they appeared more frequently. These images were created using the R package wordcloud2.

The word cloud in Figure 2 from the 2010 management reports reveals a compelling snapshot of the priorities and actions of companies engaged in sustainability efforts. Key terms such as “social”, “environmental”, “sustainable”, and “socio-environmental” feature prominently, indicating a strong focus on integrating social responsibility with environmental stewardship. This emphasis suggests that companies were increasingly recognizing the importance of balancing their operational impacts with broader societal and ecological concerns.

Central to the image are words related to management and reporting, such as “management”, “report”, “indicators”, and “performance”. These terms highlight a structured approach to sustainability, where firms were implementing comprehensive management systems to track and enhance their environmental and social performance. The presence of terms like “program”, “project”, and “actions” further emphasizes the fact that companies were not just setting goals but actively working on initiatives and strategies to achieve them.

The word cloud also reflects a broad commitment to various aspects of sustainability. Terms like “energy”, “water”, “waste”, and “preservation” indicate a focus on resource management and environmental conservation. Additionally, words such as “employees”, “collaborators”, and “communities” point to an emphasis on engaging stakeholders and fostering positive relationships within the organizational ecosystem. This suggests that firms were aiming to create value not only through their products and services but also by enhancing their impact on society and the environment.

Overall, the word clouds illustrate a multifaceted approach to sustainability, where companies integrated social and environmental considerations into their core operations. Through focusing on rigorous management practices, stakeholder engagement, and resource conservation, firms demonstrated a commitment to sustainable development that aligns with both their operational goals and broader societal values.

Analyzing the word cloud from 2022 in the right-hand graph of Figure 2, we see a nuanced shift in the focus and context of sustainability-related discussions compared to the 2010 word cloud. Both sets of keywords emphasize the integration of sustainability into corporate practices, but there are notable differences in the emphasis and terminology used in each period.

For 2022, the word cloud highlights a continued focus on sustainability with prominent terms such as “ESG” (environmental, social, and governance), “innovation”, “commitment”, and “socio-environmental”. The inclusion of “ESG” signifies an advanced understanding and formalization of sustainability criteria, reflecting the increasing importance of standardized environmental and social governance metrics in corporate reporting and strategy. Terms like “innovation”, “strategy”, and “initiatives” suggest a progressive approach, where companies are not only adhering to the established practices, but are also exploring new ways to integrate sustainability into their business models. Additionally, the presence of “innovation” and “neoenergy” suggests that firms are not only maintaining traditional practices, but are also exploring new technologies and solutions to address environmental challenges.

Compared to the 2010 word cloud, which emphasizes terms like “social”, “environmental”, “management”, and “quality”, the 2022 cloud introduces a more sophisticated and forward-looking discourse. The 2010 keywords were centered on foundational aspects of sustainability management and quality indicators. In contrast, the 2022 keywords highlighted a deeper engagement with global trends and cutting-edge solutions. The emphasis on “global” and “neoenergy” reflects a heightened awareness of international sustainability challenges and advancements in energy technology.

The 2022 word cloud also shows a shift in focus from general sustainability concepts to more specific and strategic areas. Terms such as “strategy”, “initiatives”, and “commitment” suggest that companies were integrating sustainability into their broader strategic planning and operational frameworks. This evolution points to a more comprehensive approach, where sustainability is embedded into the core of business operations and long-term goals. However, the key expressions “social” and “socio-environmental” appear in both word clouds, indicating a consistent recognition of the intersection between social and environmental factors.

Overall, the transition from the 2010 to the 2022 word cloud illustrates a progression towards a more dynamic and integrated approach to sustainability. Firms are increasingly aligning their strategies with global standards, leveraging innovation, and adopting formal ESG frameworks. This shift reflects a growing recognition of sustainability as a critical component of modern business strategies, driven by both evolving practices and emerging global trends.

3.4. Keywords and Sentiments

The final investigation, presented here, aims to place the keyword counts in greater context and infer more about the strategic behavior of firms. Figure 3 and Figure 4 exploit the results of the sentiment analysis. That is, we classified each sentence with a keyword occurrence as having either a positive, negative, or neutral context, following the methodology of Mohammad and Turney (2010).

Figure 3 shows the distributions of the absolute and relative number of keywords with a positive context in both years. The two distributions with the absolute frequency of keywords in the left-hand-side graph are very similar to the distributions where the context is not distinguished. For both 2010 and 2022, the distribution is positively skewed. Once again, we observe a clearly visible right-shift, indicating that firms mentioned public-goods-related keywords more frequently in 2022. Obviously, not all of the keywords appear in a positive context, which is why the keyword frequency values are lower than those in the distribution of all keywords (see Figure 1).

The right-hand graph of the figure illustrates the relative frequency of keywords per firm with positive contexts. In this case, the number of positive contexts is presented in relation to the number of all keyword occurrences. The graph shows that, on average, only in slightly more than half of the cases was the context positive. These distributions, however, are quite symmetric and resemble a normal distribution. The dispersion diminished over time, and we observe that the density of the positive share is more concentrated around its mean of 55%. In other words, there are less firms that refer to public-goods-related keywords in an exclusively positive or negative context. The market instead seemed to become more balanced when it comes to describing the company’s activities and ambitions regarding sustainability.

Figure 4 focuses on the changes within firms, rather than on the changes in the entire market, completing our previous results. It shows the growth rates of the number of keyword mentions in the firms’ reports between 2010 and 2022. The keyword occurrences are again distinguished by being in either a positive or negative context. Firms were then ranked by the number of positive mentions in 2010, where lower ranks correspond to higher usage of “sustainability”. We smoothed the relation between the growth rate and the rank using a locally weighted regression line.

Note first that the growth rates are strictly positive for both contexts. This observation is in line with our previous interpretation of firms using all of the analyzed keywords more frequently over time. The pattern of the growth rate of keywords used in a positive context, however, is striking and reveals an additional insight into the behavior of firms. The curve shows a clear and steep increase from lower to higher ranks. Companies that were already concerned with sustainability issues in 2010 thus are the ones who only slightly adapted their reports. However, firms which showed little concern about the environment and social issues in 2010 presented the highest increases, with growth rates of up to 400%. These market participants apparently felt a strong need to adjust their strategy and include ESG issues on the agenda.

We did not observe a similar pattern regarding the growth rates of keywords with a negative context. Instead, the locally weighted regression line has a hump-shaped form. The growth rate of keywords being used in negative contexts was close to 50% for firms with the lowest rank. Firms that used fewer keywords in positive contexts in 2010 presented higher growth rates of almost 200%, while this rate decreased again among firms with the largest rank. We thus observed a more balanced and less heterogeneous adjustment of keywords used in negative contexts.

Examples for such negative contexts that may apply to the broad market include challenges in the implementation of their strategy. An actual execution may be hindered by factors like limited resources, inadequate infrastructure, or resistance within the organization. Another reason may be economic or regulatory pressures that force firms to comply with standards exogenously. Finally, keywords such as “socio-environmental”, “security”, and “impacts” could highlight potential risks and negative outcomes associated with sustainability initiatives (see the word clouds in Figure 2). For example, socio-environmental projects might encounter significant challenges, such as conflicts with local communities, unintended environmental degradation, or issues with social equity.

4. Discussion

The present study explored the extent of public goods provision by large private corporations in Brazil. We used natural language processing methods, distributional analyses, and diverse regression techniques to identify patters in the annual standardized financial statements of publicly traded firms. Our study revealed a notable increase in the frequency of sustainability-related keywords in these reports when comparing data between 2010 and 2022. This trend was not limited to a select group of companies; rather, it spanned the entire corporate landscape, indicating a broad-based shift in how large private entities addressed sustainability. Particularly striking was the catch-up effect observed among firms that were initially less engaged in sustainability policies in the year 2010. By 2022, these companies appeared to be making significant efforts to integrate sustainability into their operations and reporting. However, the context in which sustainability was mentioned was not uniformly positive. Firms seemed to increasingly acknowledge not only their contributions to sustainable development but also the risks and challenges associated with it, reflecting a more nuanced and realistic approach to sustainability in corporate governance.

The increase in the use of terms such as “sustainability” and “responsibility” in reports by publicly traded firms in Brazil, contrasted with the relatively stable usage of words such as “education”, “transparency”, “environmental”, and “health”, could indicate a strategic shift in how firms are framing their contributions to public goods. This pattern may reflect a growing emphasis on aligning with global sustainability trends and priorities, while specific areas such as “education” and “health", which are equally crucial to public welfare, receive less attention. This could suggest that, while firms are increasingly integrating sustainability into their reporting practices, there may be a need to broaden the focus to ensure that all aspects of public goods provision are adequately addressed. Our observations highlight an opportunity for firms to deepen their commitment across a wider range of public goods, ensuring a more balanced and comprehensive approach to aligning corporate necessities with investments in areas that create benefits for stakeholders.

The evidence presented in this research may be useful for policymakers concerned about fostering a more sustainable business environment in Brazil. From a welfare perspective, one may argue that the higher the number of ESG-related public goods, the better. However, the extent to which private companies contribute to this aim is their own decision and depends on the regulatory framework, the existing incentives provided by the government, and the objective functions of each firm. On the one hand, we saw that, under the current legislation, firms already seem to have a strong intrinsic motivation to become more sustainable and responsible. This outcome may be related to existing public policies, as well as to strategical management decisions, aiming to satisfy consumer, investor, and employee preferences. Among the several incentives for corporate investments in ESG issues that have been implemented over the last decade, we highlight the following: (1) The National Bank for Economic and Social Development (BNDES) offers subsidized loans for long-term investments in sustainable projects, including renewable energy and energy efficiency initiatives. (2) Programs promoting renewable energy projects such as the Prosol and Paiere were created after enacting the law 12.187/2009, which established national policy directives on climate change. On the other hand, more incentives may be necessary to mobilize private investments in education, health, and poverty reduction. Similarly to the initiatives cited above, policymakers could consider offering subsidized interest rates and tax reductions for corporate projects that promote significant contributions to public goods in the areas of education and health. Regarding subsidies, however, policymakers should bear in mind the fact that the type of subsidy determines the outcome—these types of interventions do not always lead to the desired positive effects (Lee and Do Chung 2022). In some cases, subsidies can even have negative impacts on the environment and regional inequality (Barde and Honkatukia 2005; Dupont and Martin 2006).

The findings of this research have significant implications for corporate accountability and investment strategies among Brazilian firms. The pronounced increase in ESG keyword mentions indicates a growing commitment to public responsibility, enhancing stakeholder trust and transparency. Companies should leverage this momentum by investing in sustainability and social responsibility initiatives, recognizing that robust ESG practices can provide a competitive advantage. Additionally, the alignment of local firms with the strategies of pioneering multinationals serves as a benchmark, encouraging others to evaluate and enhance their own ESG commitments. Furthermore, the positive sentiment surrounding these keywords highlights shifting consumer preferences, suggesting that businesses can benefit from emphasizing their sustainability efforts in marketing and branding.

To support and enhance these developments, several policy recommendations are proposed here. Policymakers should consider establishing standardized ESG reporting frameworks, to ensure consistency and comparability among firms, facilitating informed investment decisions. Implementing tax incentives or subsidies for companies actively engaging in public goods provision and ESG initiatives could encourage broader participation in sustainable practices. Public awareness campaigns could educate stakeholders on the importance of ESG practices, fostering a culture of responsibility. Additionally, promoting collaboration between the public and private sectors could leverage resources and expertise to effectively address social and environmental challenges. Finally, establishing mechanisms for monitoring and evaluating the effectiveness of ESG initiatives will ensure that firms not only increase their ESG mentions but also have meaningful impacts on society and the environment.

The present study focused exclusively on mentions of keywords related to public goods provision. Future research could address the investments and real impacts of these projects, both for the firm and for society as a whole. We prioritized a relatively ample time horizon, sample size, and the objectivity provided by standardized financial declarations. Our approach also cannot determine whether the investment in public goods occurs systematically in order to change the organizational structure of the business, or whether it is simply a way to avoid criticism from shareholders and the general public (Kondo 2006). In other words, it is not possible to analyze whether the measures are substantive or merely symbolic. Case studies may provide deeper and more up-to-date insights, but their downside is limited generalizability (see, for example, Mafra et al. (2024) and Pereira et al. (2024)). Another limitation is induced through the consideration of publicly traded companies. Although these firms are among the largest and most powerful in the country, we miss a wide range of ESG projects advanced by other types of organizations, the public domain, as well as small- and medium-sized firms.