Climate Sentiment Analysis on the Disclosures of the Corporations Listed on the Johannesburg Stock Exchange

Abstract

International organizations have highlighted the importance of consistent and reliable environment, social and governance (ESG) disclosure and metrics to inform business strategy and investment decisions. Greater corporate disclosure is a positive signal to investors who prioritize sustainable investment. In this study, economic and climate sentiment are extracted from the integrated and sustainability reports of the top 40 corporates listed on the Johannesburg Stock Exchange, employing domain-specific natural language processing. The intention is to clarify the complex interactions between climate risk, corporate disclosures, financial performance and investor sentiment. The study provides valuable insights to regulators, accounting professionals and investors on the current state of disclosures and future actions required in South Africa. A time series analysis of the sentiment scores indicates a noticeable change in the corporates’ disclosures from climate-related risks in the earlier years to climate-related opportunities in recent years, specifically in the banking and mining sectors. The trends are less pronounced in sectors with good ESG ratings. An exploratory regression study reveals that climate and economic sentiments contain information that explain stock price movements over the longer term. The results have important implications for asset allocation and offer an interesting direction for future research. Monitoring the sentiment may provide early-warning signals of systemic risk, which is important to regulators given the impact on financial stability.

1. Introduction

In 2015, the 193 member states of the United Nations adopted the 2030 Agenda for Sustainable Development, which consists of 17 sustainable development goals (SDGs). The SDGs target essential human needs such as health, poverty, education, gender equality, climate change, sustainable industrialization and consumption, conservation of life on land and below water—reducing the inequality between countries—and forming partnerships to promote the implementation of the goals. Progress to achieve the goals has been notably slow (United Nations, 2024), which is concerning, given the negative consequences of depleting natural resources, the impact on livelihoods, the exacerbation of poverty and the spread of diseases due to a failure to achieve these goals (Filho et al., 2020). In 2024, South Africa ranked 115 with an SDG index score of 63.4, indicating the percentage of SDGs that have been achieved to date (Sachs et al., 2024).

Achieving the SDGs depends on sustainable business and management practices (Mahajan et al., 2024). Corporations play a critical role in contributing to the successful adoption of the SDGs (JSE, 2022). Increased disclosure by companies regarding their sustainability practices is a positive indicator for investors whose strategies prioritize the SDGs (Y. Li et al., 2023; Megeid, 2024; Pandey et al., 2025). However, a study conducted in the US cautions that investors are not significantly influenced by climate-related risks unless policymakers intervene (Faccini et al., 2023).

P. Chen and Dagestani (2023) note that the increasing pressure on companies to disclose sustainability-related issues may lead to a greater risk of greenwashing. Greenwashing refers to misleading sustainability disclosures that exaggerate the positive aspects of a company’s practices while obscuring the more negative aspects. Greenwashing improves the perceived quality of the corporate disclosure, leading to a positive impact on firm value.

To address the risks around sustainability reporting, the International Financial Reporting Standards (IFRS) Foundation launched the International Sustainability Standards Board (ISSB). The objective was to enhance the quality of sustainability reporting by aligning it to existing regulations pertaining to financial disclosures. On 26 June 2023, the ISSB released IFRS S1 and S2 that became effective from 1 January 2024. Companies are mandated to disclose decision-useful information regarding sustainability-related (IFRS S1) and climate-related (IFRS S2) risks and opportunities that may affect their risk profiles (International Sustainability Standards Board, 2023a, 2023b).

Sentiment extracted from climate-related publications produce interesting insights. Sources of climate risk sentiment include social media (Rosenberg et al., 2023; Upadhyaya et al., 2023; Uthirapathy & Sandanam, 2023); news sources (Faccini et al., 2023); or company financial reports (Bingler et al., 2022; Van der Lugt et al., 2025).

In this study, climate-related sentiment data are extracted from the Financial Times to gauge sentiment based on recent global events (2024 to 2025). Next, the integrated and sustainability reports of the companies that make up the top 40 all-share index on the Johannesburg Stock Exchange (JSE) were sourced. The JSE top 40 index consists of the largest 40 companies ranked by market value (JSE, 2020). The study considers how the climate-related sentiment has changed between 2010 and 2024 and the main themes that drive the change. The study demonstrates that the evolution of reporting standards has positively influenced climate-related disclosures, indicating that businesses are recognizing their climate-related risks and translating them into strategic actions. This study provides valuable insights for regulators, accounting professionals and investors regarding the current state of corporate disclosure and future actions needed in South Africa.

Previous South African studies have established the impact of investor sentiment on the stock market, where investor sentiment is measured by market variables such as share turnover, equity issue ratio, rand/dollar bid-ask spread and interest rates (Moodley et al., 2025; Naidoo et al., 2025). Studies on corporate disclosures typically employed textual analysis, focusing on the readability of the texts by examining factors such as sentence length and syllables per word (Enslin et al., 2025) or an analysis of the number of occurrences of keywords (Van der Lugt et al., 2025). This study extends the existing research by incorporating domain-specific natural language processing (NLP) techniques. Climate sentiment is extracted from the corporate disclosures with ClimateBERT, a large language model developed by Webersinke et al. (2022); FinBERT is used to extract economic sentiment (Araci, 2019). BERT stands for bidirectional encoder representations from transformers where surrounding text is used to establish context (Zhou et al., 2024).

The research contributes to clarifying the complex interactions between climate risk, corporate disclosures, financial performance and investor sentiment. The economic and climate sentiment extracted from the corporate disclosures are shown to contain information that may inform asset allocation strategies or serve as early-warning signals of potential systemic risk for regulators.

The outline of this study is as follows. Section 2 presents a literature review. Section 3 outlines the data sources, and Section 4 summarizes the methodology. The results are explored in Section 5, and conclusions are drawn in Section 6.

2. Literature Review

Climate-related risks and the transition to a low-carbon economy have significant impacts on all sectors of the economy, prompting corporations to adjust their strategies towards more sustainable investments and projects. The JSE (2022) notes the fundamental role of the corporations in the country’s sustainability goals. The sustainability disclosures are very important to enable an accurate assessment and pricing of risks by investors.

The adoption of sustainability reporting is often studied based on institutional and stakeholder theory. Institutional theory suggests that corporations adopt disclosure practices based on coercive (regulatory demands), normative (industry norms) and mimetic (imitation of industry leaders) isomorphism. Stakeholder theory postulates that corporate disclosure is tailored for the needs of different stakeholder groups such as investors, employees and the community (Herold, 2018; Wukich et al., 2024; Galleli & Amaral, 2025).

The importance of consistent and reliable ESG metrics to inform business strategies and investment decisions is accentuated by organizations such as the Organization for Economic Cooperation and Development (OECD) (OECD, 2025). ESG ratings were developed as a measure of company performance against the sustainability indicators. Companies with good ESG ratings may be perceived as less risky and more sustainable and thus attract more capital. Morningstar Sustainalytics for instance derives ESG ratings by considering the magnitude of a company’s unmanaged ESG risks by analyzing (1) material ESG issues such as human capital and occupational health and safety; (2) corporate and stakeholder governance; (3) systemic ESG issues and (4) idiosyncratic risks (Pop & Vosburg, 2024). Research indicates significant discrepancies in the outcomes produced by different ESG rating agencies, suggesting the ESG performance is measured largely by business effort rather than actual effect (Eren et al., 2022; OECD, 2025). Huang and Lin (2022) found that ESG scores are higher for companies located in counties where more people believe in global climate change.

Climate change and global warming pose both physical risk (catastrophic weather events and chronic risk such as the ocean level rising) and transition risk (moving away from fossil fuels to achieve net zero by 2050) and are considered a systemic risk that affects financial stability. It is a major concern for central banks and financial regulators (H. Li et al., 2021; Jourde & Moreau, 2023; Monnin et al., 2024). Supervisory authorities have started to explore the potential of capital buffers to address climate-related systemic risks (Ikeda & Monnin, 2024).

The systemic risk impacts on asset pricing are explored in Campiglio et al. (2023); they argue that both physical and transition risks can trigger a revaluation of financial assets. This aligns with the numerous research studies that examine the relationship between ESG factors and corporate financial performance (Kale & Akkaya, 2016; Antoniuk & Leirvik, 2021; Ai & Gao, 2023; S. Chen et al., 2023; Bagh et al., 2024; Macdonald & van Vuuren, 2024). South African studies show mixed results. Masongweni and Simo-Kengne (2024) show that composite ESG scores do not significantly affect the financial performance of South African companies. Social and governance aspects have a positive association with company performance, but the impacts of environmental factors are limited. Studies such as those by Chininga et al. (2024) and Short and Ndlovu (2025) identified a positive relationship between ESG activities and corporate performance, suggesting that incorporating ESG ratings in asset allocation may enhance fund performance.

Sentiment analysis with NLP on company disclosures provides new insights on industry trends. Existing research underscores the importance of using domain-specific NLP to ensure the accuracy of the outcomes (Sham & Mohamed, 2022; Rosenberg et al., 2023). In this study, ClimateBERT is used to extract climate-related sentiment. The ClimateBERT language model has been pretrained on over 2 million paragraphs of climate-related texts from various sources such as common news, research articles and corporate disclosures (Webersinke et al., 2022). Economic sentiment analysis is performed with FinBERT that was trained on a large financial corpus (Araci, 2019).

This study is a sophisticated extension of existing research on corporate disclosures, which typically employ textual analyses that focus on assessing the readability of the texts (Enslin et al., 2025). The study offers innovative insights into the relationships between ESG ratings, economic and climate sentiment, corporate disclosure, financial performance and systemic risk, providing valuable information for regulators, investors and other stakeholders.

3. Data

This section offers a summary of the data utilized in this study. Data sources include news, company integrated and sustainability reports, sustainability ratings, market data and the natural disasters database.

3.1. EBRD Ratings

The European Bank for Reconstruction and Development (EBRD) provides guidance on the level of environmental and social risk in specific industries (EBRD, 2014). Each industry is assigned an environmental, social and overall risk indicator of low, medium and high. The EBRD ratings were assigned to each corporation in the study based on their industry classification and is summarized in Table 1.

The environmental rating captures the impact of human activity on the environment such as atmosphere, water, plants and animals. The social rating considers aspects such as public health, safety, security and gender equality. The overall EBRD rating reflects a balanced assessment of both the environmental and social impacts.

3.2. ESG Ratings

The ESG ratings for the JSE top 40 corporations were sourced from Morningstar Sustainalytics (https://www.sustainalytics.com/) (accessed on 1 May 2025). They provide an online platform to extract the most recent ESG ratings. The data were sourced in May 2025. The ESG ratings are summarized in Table 1.

3.3. Financial Integrated and Sustainability Reports

The disclosures provide decision-useful information on the corporations’ operating environment, emerging risks and strategies to address it.

Table 1 provides a summary of the data available for each of the corporations in the JSE top 40 index in May 2025. The annual integrated and sustainability reports were sourced from each company’s website. The annual reports were extracted from 2010 to 2024 (where available) and converted to text using the PyMuPDF Python (version 1.26.3) library for data conversion of PDF documents. Annual reports prior to 2010 were found to lack the sustainability focus.

Each corporation was assigned an alias based on a very high-level industry classification. Companies with multiple listings increased the number of corporations included in the study to 43.

Table 1. Summary of the largest corporations that form part of the JSE top 40 index.

Alias	Sector	EBRD— Environment	EBRD— Social	EBRD— Overall	ESG Rating Category	Financial Year End	Financial Report Data Available
Bank_1	Banks	Low	Low	Low	Medium	31-December	2010 to 2024
Bank_2	Banks	Low	Low	Low	Medium	28-February	2010 to 2024
Bank_3	Banks	Low	Low	Low	Low	30-June	2010 to 2024
Bank_4	Banks	Low	Low	Low	Low	30-June	2016 to 2024
Bank_5	Banks	Low	Low	Low	Low	31-March	2011 to 2024
Bank_6	Banks	Low	Low	Low	Low	31-March	2011 to 2024
Bank_7	Banks	Low	Low	Low	Low	31-December	2010 to 2024
Bank_8	Banks	Low	Low	Low	Low	31-December	2013 to 2024
Bank_9	Banks	Low	Low	Low	Low	30-June	2010 to 2024
Bank_10	Banks	Low	Low	Low	Medium	31-March	2010 to 2024
Insurance_1	Life Insurance	Low	Low	Low	Low	30-June	2010 to 2024
Insurance_2	Life Insurance	Low	Low	Low	Medium	31-December	2010 to 2024
Insurance_3	Life Insurance	Low	Low	Low	Medium	31-December	2012 to 2024
RealEstate_1	Real Estate Investment	Medium	Medium	Medium	Negligible	31-December	2017 to 2024
RealEstate_2	Real Estate Investment Trusts	Medium	Medium	Medium	Negligible	30-June	2010 to 2024
Consumer_1	Beverages	Low	Low	Low	Medium	31-December	2010 to 2024
Consumer_2	Tobacco	Low	Low	Low	High	31-December	2010 to 2024
Consumer_3	Retailers	Low	Low	Low	Low	31-March	2011 to 2024
Consumer_4	Retailers	Low	Low	Low	Low	30-June	2016 to 2024
Consumer_5	Personal Care, Drug and Grocery Stores	Low	Low	Low	Medium	31-August	2010 to 2024
Consumer_6	Personal Care, Drug and Grocery Stores	Low	Low	Low	Medium	30-June	2010 to 2024
Consumer_7	Personal Care, Drug and Grocery Stores	Low	Low	Low	Low	30-June	2016 to 2024
Consumer_8	Personal Goods	Low	Low	Low	Low	31-March	2010 to 2024
Industrial_1	General Industrials	Medium	Medium	Medium	Low	31-December	2010 to 2024
Industrial_2	General Industrials	High	High	High	High	30-June	2013 to 2024
Industrial_3	Pharmaceuticals and Biotechnology	High	High	High	Medium	30-June	2020 to 2024
OilGas_1	Chemicals	High	High	High	High	30-June	2010 to 2024
OilGas_2	Oil, Gas and Coal	High	High	High	High	31-December	2010 to 2024
Mining_1	Industrial Metals and Mining	High	High	High	Medium	31-December	2010 to 2024
Mining_2	Industrial Metals and Mining	High	High	High	Medium	30-June	2010 to 2024
Mining_3	Industrial Metals and Mining	High	High	High	High	31-December	2011 to 2024
Mining_4	Precious Metals and Mining	High	High	High	Low	31-December	2010 to 2024
Mining_5	Precious Metals and Mining	High	High	High	High	31-December	2016 to 2024
Mining_6	Precious Metals and Mining	High	High	High	Medium	31-December	2010 to 2024
Mining_7	Precious Metals and Mining	High	High	High	Medium	30-June	2010 to 2024
Mining_8	Precious Metals and Mining	High	High	High	Medium	30-June	2010 to 2024
Mining_9	Precious Metals and Mining	High	High	High	Medium	31-December	2013 to 2024
Tech_1	Software and Computer Services	Low	Low	Low	Low	31-March	2019 to 2024
Tech_2	Software and Computer Services	Low	Low	Low	Low	31-March	2020 to 2024
Tech_3	Telecommunications Service Providers	Low	Medium	Medium	Low	31-March	2019 to 2024
Tech_4	Telecommunications Service Providers	Low	Medium	Medium	Low	31-December	2010 to 2024
Tech_5	Telecommunications Service Providers	Low	Medium	Medium	Low	31-March	2010 to 2024

Table 1. Summary of the largest corporations that form part of the JSE top 40 index.

Alias	Sector	EBRD— Environment	EBRD— Social	EBRD— Overall	ESG Rating Category	Financial Year End	Financial Report Data Available
Bank_1	Banks	Low	Low	Low	Medium	31-December	2010 to 2024
Bank_2	Banks	Low	Low	Low	Medium	28-February	2010 to 2024
Bank_3	Banks	Low	Low	Low	Low	30-June	2010 to 2024
Bank_4	Banks	Low	Low	Low	Low	30-June	2016 to 2024
Bank_5	Banks	Low	Low	Low	Low	31-March	2011 to 2024
Bank_6	Banks	Low	Low	Low	Low	31-March	2011 to 2024
Bank_7	Banks	Low	Low	Low	Low	31-December	2010 to 2024
Bank_8	Banks	Low	Low	Low	Low	31-December	2013 to 2024
Bank_9	Banks	Low	Low	Low	Low	30-June	2010 to 2024
Bank_10	Banks	Low	Low	Low	Medium	31-March	2010 to 2024
Insurance_1	Life Insurance	Low	Low	Low	Low	30-June	2010 to 2024
Insurance_2	Life Insurance	Low	Low	Low	Medium	31-December	2010 to 2024
Insurance_3	Life Insurance	Low	Low	Low	Medium	31-December	2012 to 2024
RealEstate_1	Real Estate Investment	Medium	Medium	Medium	Negligible	31-December	2017 to 2024
RealEstate_2	Real Estate Investment Trusts	Medium	Medium	Medium	Negligible	30-June	2010 to 2024
Consumer_1	Beverages	Low	Low	Low	Medium	31-December	2010 to 2024
Consumer_2	Tobacco	Low	Low	Low	High	31-December	2010 to 2024
Consumer_3	Retailers	Low	Low	Low	Low	31-March	2011 to 2024
Consumer_4	Retailers	Low	Low	Low	Low	30-June	2016 to 2024
Consumer_5	Personal Care, Drug and Grocery Stores	Low	Low	Low	Medium	31-August	2010 to 2024
Consumer_6	Personal Care, Drug and Grocery Stores	Low	Low	Low	Medium	30-June	2010 to 2024
Consumer_7	Personal Care, Drug and Grocery Stores	Low	Low	Low	Low	30-June	2016 to 2024
Consumer_8	Personal Goods	Low	Low	Low	Low	31-March	2010 to 2024
Industrial_1	General Industrials	Medium	Medium	Medium	Low	31-December	2010 to 2024
Industrial_2	General Industrials	High	High	High	High	30-June	2013 to 2024
Industrial_3	Pharmaceuticals and Biotechnology	High	High	High	Medium	30-June	2020 to 2024
OilGas_1	Chemicals	High	High	High	High	30-June	2010 to 2024
OilGas_2	Oil, Gas and Coal	High	High	High	High	31-December	2010 to 2024
Mining_1	Industrial Metals and Mining	High	High	High	Medium	31-December	2010 to 2024
Mining_2	Industrial Metals and Mining	High	High	High	Medium	30-June	2010 to 2024
Mining_3	Industrial Metals and Mining	High	High	High	High	31-December	2011 to 2024
Mining_4	Precious Metals and Mining	High	High	High	Low	31-December	2010 to 2024
Mining_5	Precious Metals and Mining	High	High	High	High	31-December	2016 to 2024
Mining_6	Precious Metals and Mining	High	High	High	Medium	31-December	2010 to 2024
Mining_7	Precious Metals and Mining	High	High	High	Medium	30-June	2010 to 2024
Mining_8	Precious Metals and Mining	High	High	High	Medium	30-June	2010 to 2024
Mining_9	Precious Metals and Mining	High	High	High	Medium	31-December	2013 to 2024
Tech_1	Software and Computer Services	Low	Low	Low	Low	31-March	2019 to 2024
Tech_2	Software and Computer Services	Low	Low	Low	Low	31-March	2020 to 2024
Tech_3	Telecommunications Service Providers	Low	Medium	Medium	Low	31-March	2019 to 2024
Tech_4	Telecommunications Service Providers	Low	Medium	Medium	Low	31-December	2010 to 2024
Tech_5	Telecommunications Service Providers	Low	Medium	Medium	Low	31-March	2010 to 2024

3.4. Market Data

Daily market data were extracted from https://za.investing.com/ for the period between January 2010 and December 2024, accessed on 1 May 2025. The market dataset includes corporate share prices, the JSE top 40 index, the rand/dollar exchange rate and the 5-year SA bond yield.

3.5. SARB Composite Business Cycle Indicator

The SARB composite business cycle indicator provides early warnings of turning points in the business cycle and is used as a measure of systemic risk. It is available online at https://www.resbank.co.za/, accessed on 1 May 2025.

3.6. Financial Times News Articles

Climate-related news articles were sourced from the Financial Times (https://www.ft.com/climate-capital, accessed on 1 May 2025) for the period between July 2024 and July 2025. There were 700 articles in total during this period.

3.7. Natural Disasters

The EM-DAT natural disasters database contains all the global disasters. It was initially a joined initiative between the Centre for Research on the Epidemiology of Disasters (CRED) and the World Health Organization (WHO) but is now linked to the University of Louvain in Belgium. The EM-DAT natural disasters database is available online at https://www.emdat.be/, accessed on 1 May 2025.

4. Methodology

This section summarizes the methodologies employed to derive a systemic index from corporate stock price returns, and the sentiment scores from climate-related texts.

4.1. Systemic Index

A systemic index was constructed from the share price data of the JSE top 40 constituents for the period between January 2010 and December 2024. To reduce volatility, the data were summarized into monthly average prices $\overline{P_t}$. The monthly stock returns $R_t$ were derived using $R_t={\displaystyle \frac{\overline{P_t}}{\overline{P_{t-1}}}}-1$ and then converted to Gaussian to ensure all the data are on the same scale. The stock return data were cleaned by smoothing out stock splits and outliers. The missing data were forward-filled.

A principal component analysis (PCA) was performed to reduce the dimensionality. The eigenvalues and eigenvectors were derived from the Gaussian stock returns. The systemic index was then derived using

$${Systemic}_t={\sum }_{j=1}^m{\displaystyle \frac{{\lambda }_j}{{\lambda }_{Tot}}}{\sum }_{k=1}^n{\beta }_{j,k}^2{Z}_k,$$

(1)

where ${Systemic}_t$ denotes the systemic index, ${\lambda }_j$ the eigenvalues, ${\lambda }_{Tot}$ the total variance explained by the $m$ principal components (PCs) selected to derive the systemic index, $n$ denotes the number of stocks included in the analysis, $Z_k$ the Gaussian stock return and ${\beta }_k^2$ denotes the weight contribution of each stock return such that ${\sum }_{k=1}^n{\beta }_{j,k}^2=1$. In this application, the first 7 PCs were selected, capturing between 60 and 70% of the total variance.

The systemic index is considered a more accurate proxy for the stock market systemic risk compared to the JSE top 40 index, as the constituents of the JSE index change over time and the historical data may not adequately represent the considered in this study. Figure 1 confirms a strong relationship between the systemic index ${Systemic}_t$ and the JSE top 40 index returns (also converted to Gaussian).

4.2. Corporate Disclosures: Economic Sentiment

The integrated and sustainability reports published by the JSE top 40 corporations were converted into text. Each sentence was tokenized, and stop words were removed. The words were then lemmatized. Text processing enabled the NLP to produce more accurate results. Each sentence was then assigned a negative, neutral or positive sentiment. The economic sentiment was extracted using FinBERT, which assigned a score between −1 and 1, denoting the level of negative or positive sentiment. In the analyses, all the ‘neutral’ comments were excluded.

Two sentiment variables were derived for each corporation and year, for which financial reports were available. The first variable, proportion positive, is derived as

$$ProportionPositive={\displaystyle \frac{T_P{-T}_N}{T_P{+T}_N}},$$

(2)

where $T_P$ denotes the total number of positive sentiment sentences and $T_N$ denotes the total number of negative sentiment sentences. The outcomes are between 1 and −1, where 1 is observed if all sentences were positive, and −1 if all sentences were negative.

The second variable, average economic sentiment, looks at the average economic sentiment score assigned to each sentence:

$$AverageEconomicSentiment={\displaystyle \frac{1}{n}}{\sum }_j{E}_j,$$

(3)

where $E_j$ denotes the economic sentiment assigned to sentence $j$ and $n$ denotes the total number of sentences considered.

The sentiment variables were derived for each corporation listed in Table 1 and referred to as the idiosyncratic sentiment scores. Sector-level sentiment scores were also derived as the average sentiment score across all corporations in the specific sector.

4.3. Corporate Disclosures: Climate Sentiment

ClimateBERT was used to determine whether the corporations climate-related disclosures were mainly a negative risk, a positive opportunity, or neutral. The sentiment was assigned to each sentence. The proportion-positive, climate-related sentiment was derived using Equation (2); the average economic sentiment was derived using Equation (3) based on the economic sentiment score extracted with FinBERT for the specific sentence. All neutral sentences were excluded from the analysis.

Two filters were considered to identify texts related to climate risk: the first filter applied a ClimateBERT filter on all the texts that considered climate-related disclosure; the second filter was a manual classification where climate-related sentences were identified based on the keywords summarized in Figure 2.

The final list of sentiment variables is summarized in

Table 2. Definition of the sentiment variables derived from the corporate disclosures.

Variable Code	Algorithm	Sentiment Filer	Filter Applied to Each Sentence	Sentiment Score
$X_1$	FinBERT	No neutral economic	No Filter	Average Economic Sentiment
$X_2$	FinBERT	No neutral economic	No Filter	Proportion Positive
$X_3$	FinBERT	No neutral economic	Manual Climate Comments	Average Economic Sentiment
$X_4$	FinBERT	No neutral economic	Manual Climate Comments	Proportion Positive
$X_5$	FinBERT	No neutral economic	ClimateBERT identified	Average Economic Sentiment
$X_6$	FinBERT	No neutral economic	ClimateBERT identified	Proportion Positive
$X_7$	ClimateBERT and FinBERT	Only risk or opportunity	No Filter	Average Economic Sentiment
$X_8$	ClimateBERT	Only risk or opportunity	No Filter	Proportion Positive
$X_9$	ClimateBERT and FinBERT	Only risk or opportunity	Manual Climate Comments	Average Economic Sentiment
$X_{10}$	ClimateBERT	Only risk or opportunity	Manual Climate Comments	Proportion Positive
$X_{11}$	ClimateBERT and FinBERT	Only risk or opportunity	ClimateBERT identified	Average Economic Sentiment
$X_{12}$	ClimateBERT	Only risk or opportunity	ClimateBERT identified	Proportion Positive

. The “Algorithm” column shows whether FinBERT or ClimateBERT (or both) were used to calculate the sentiment score. For instance, to derive $X_7$, the sentences were filtered based on the sentiment assigned by ClimateBERT (risk or opportunity), and then Equation (3) was used to derive the sentiment score based on the corresponding FinBERT score, which ranged between −1 and 1 for each sentence. The sentiment score for $X_8$ was obtained by applying the sentiment assigned by ClimateBERT in Equation (2).

The variables $X_1$ and $X_2$ are economic sentiment variables; all the other variables are considered climate sentiment variables because either a climate-related filter was applied, or ClimateBERT was used to extract the sentiment.

The sentiment scores were derived for each corporation as well as the average at a sector level.

4.4. News: Climate Sentiment

The climate-related news data, extracted from the Financial Times, were analyzed with ClimateBERT which detects climate-related risks (negative), opportunities (positive) or neutral sentiment. Each article was assigned a climate sentiment score by calculating the proportion positive climate-related sentences using Equation (2). The news sentiment provides valuable information around recent trends in climate-related news.

5. Results

5.1. Recent Climate-Related News Trends

Global climate-related news data were sourced from the Financial Times for the period July 2024 to July 2025. Figure 3 shows the distribution of the climate sentiment score across the news articles. There is significantly greater emphasis towards negative sentiment, indicating that the articles focus more on climate-related risks than on opportunities.

Figure 4 shows the time series behavior of the climate sentiment. The red line shows the 10-day moving average to achieve a better assessment of the trend over time. The climate sentiment fluctuates around a mean level of −40%, and there has been no change in the trend over the last year. This is in line with the Network for Greening the Financial System (NGFS)’s (2024) warnings of increasingly adverse effects of climate change and the negative consequences observed in many countries.

5.2. Economic vs. Climate Sentiment

In this section, the relationship between the climate and economic sentiment extracted from the corporate disclosures is analyzed. The economic sentiment variables are $X_1$ and $X_2$, as defined in Table 2. All the remaining variables are either derived from the ClimateBERT sentiment analysis or have a climate-related filter applied, which makes them climate sentiment variables.

The correlation analysis in

Table 3. Pearson correlation between the economic and climate sentiment variables.

	${\mathit{X}}_1$	${\mathit{X}}_2$	${\mathit{X}}_3$	${\mathit{X}}_4$	${\mathit{X}}_5$	${\mathit{X}}_6$	${\mathit{X}}_7$	${\mathit{X}}_8$	${\mathit{X}}_9$	${\mathit{X}}_{10}$	${\mathit{X}}_{11}$	${\mathit{X}}_{12}$
$X_1$	100%	100%	40%	40%	46%	47%	93%	83%	38%	41%	38%	39%
$X_2$		100%	40%	41%	45%	47%	92%	83%	38%	42%	37%	38%
$X_3$			100%	98%	58%	59%	44%	44%	86%	69%	57%	55%
$X_4$				100%	58%	59%	45%	45%	85%	71%	57%	57%
$X_5$					100%	98%	48%	42%	56%	52%	88%	71%
$X_6$						100%	50%	44%	58%	53%	88%	73%
$X_7$							100%	92%	51%	53%	49%	52%
$X_8$								100%	49%	62%	44%	57%
$X_9$									100%	78%	67%	65%
$X_{10}$										100%	60%	74%
$X_{11}$											100%	84%
$X_{12}$												100%

establishes a strong relationship between the economic sentiment variables and the climate sentiment variables where no filters are applied.

Figure 5 shows the relationship between economic sentiment and climate sentiment with contour plots. The darker areas on the plots denote more observations. There is a clear positive relationship between economic and climate sentiment. The relationship is stronger when using no filters on both economic sentiment ($X_1$) and climate sentiment ($X_8$). Applying the ClimateBERT filters on the climate sentiment ($X_{12}$) lead to increased volatility of the sentiment scores.

5.3. Corporate Disclosure over Time

In this section, the time series behavior of the sentiment extracted from the corporate disclosures is analyzed. This analysis reveals whether there has been a change in sentiment over time.

Table 4. Descriptive statistics for the sentiment variables extracted from the corporate disclosures.

Statistics: 2010 to 2024	${\mathit{X}}_1$	${\mathit{X}}_2$	${\mathit{X}}_3$	${\mathit{X}}_4$	${\mathit{X}}_5$	${\mathit{X}}_6$	${\mathit{X}}_7$	${\mathit{X}}_8$	${\mathit{X}}_9$	${\mathit{X}}_{10}$	${\mathit{X}}_{11}$	${\mathit{X}}_{12}$
Minimum	−0.63	−0.91	−0.90	−1.00	−0.63	−1.00	−0.70	−1.00	−0.90	−1.00	−0.84	−1.00
Mean	0.25	0.32	0.26	0.37	0.28	0.40	0.05	−0.26	0.05	−0.29	0.12	−0.15
Median	0.26	0.34	0.26	0.38	0.30	0.41	0.05	−0.32	0.03	−0.33	0.13	−0.20
Maximum	0.64	0.83	0.88	1.00	0.85	1.00	0.67	0.80	0.88	1.00	0.81	1.00
Standard Deviation	0.17	0.22	0.20	0.26	0.20	0.26	0.22	0.33	0.27	0.40	0.28	0.42
Statistics: 2020 to 2024	${\mathit{X}}_{\mathbf{1}}$	${\mathit{X}}_{\mathbf{2}}$	${\mathit{X}}_{\mathbf{3}}$	${\mathit{X}}_{\mathbf{4}}$	${\mathit{X}}_{\mathbf{5}}$	${\mathit{X}}_{\mathbf{6}}$	${\mathit{X}}_{\mathbf{7}}$	${\mathit{X}}_{\mathbf{8}}$	${\mathit{X}}_{\mathbf{9}}$	${\mathit{X}}_{\mathbf{10}}$	${\mathit{X}}_{\mathbf{11}}$	${\mathit{X}}_{\mathbf{12}}$
Minimum	−0.63	−0.91	−0.90	−1.00	−0.63	−1.00	−0.70	−1.00	−0.90	−1.00	−0.84	−1.00
Mean	0.25	0.33	0.27	0.38	0.29	0.41	0.07	−0.24	0.07	−0.26	0.12	−0.13
Median	0.27	0.35	0.27	0.38	0.31	0.43	0.07	−0.26	0.06	−0.30	0.14	−0.15
Maximum	0.58	0.79	0.88	1.00	0.85	1.00	0.57	0.72	0.88	1.00	0.81	1.00
Standard Deviation	0.17	0.23	0.20	0.27	0.20	0.27	0.23	0.34	0.27	0.40	0.28	0.43

summarizes the descriptive statistics of the sentiment variables. The statistics were derived across all corporations for the full period between 2010 and 2024 and for a more recent period between 2020 and 2024.

The full-period statistics do not differ significantly from the more recent period. The sentiment variables to which additional filters were applied exhibit more volatile outcomes, as evidenced by the higher standard deviation.

Figure 6 illustrates the trend of selected corporations for the $X_4$ and $X_8$ sentiment variables. The positive trend observed for some of the corporations denotes a shift in the disclosures from climate risk in earlier years, to climate opportunities in recent years. The trends are more easily observed for $X_8$, where no climate-related filters were applied.

A simplistic regression analysis is used to determine the number of corporations that demonstrate an improvement in their climate sentiment over the 10-year period between 2014 and 2024.

Table 5. The slope indicates whether the climate sentiment from the disclosures has improved over time.

Alias	Slope		Alias	Slope		Alias	Slope
Bank_1	0.05	*	RealEstate_2	−0.02	**	Mining_1	0.06	*
Bank_2	0.01	**	Consumer_1	0.01		Mining_2	−0.02	*
Bank_3	0.04	*	Consumer_2	−0.01	*	Mining_3	−0.01
Bank_4	−0.04	**	Consumer_3	−0.05	*	Mining_4	0.03	*
Bank_5	0.03	**	Consumer_4	−0.05	*	Mining_5	−0.01
Bank_6	0.03	**	Consumer_5	−0.04	*	Mining_6	0.00
Bank_7	0.05	*	Consumer_6	0.04	*	Mining_7	0.03	*
Bank_8	0.03	*	Consumer_7	0.00		Mining_8	0.04	*
Bank_9	0.01		Consumer_8	−0.02	**	Mining_9	0.01	*
Bank_10	−0.02		Industrial_1	−0.02	*	Tech_1	−0.11	*
Insurance_1	−0.01	*	Industrial_2	−0.01		Tech_2	−0.07	*
Insurance_2	0.10	*	Industrial_3	0.09	*	Tech_3	−0.04	***
Insurance_3	0.07	*	OilGas_1	0.01		Tech_4	−0.02
RealEstate_1	0.04	*	OilGas_2	0.03	*	Tech_5	−0.01	***

* Denotes significance at a 1% level; ** 5%; *** 10%.

summarizes the slope as estimated from the regression equation $X_8=slope\times year$, where $X_8$ denotes the climate sentiment variable defined in Table 2 and $year=\left\{2014,2015,\dots ,2024\right\}$ denotes the financial reporting year. The results indicate significant improvement in the sentiment from the banking, mining and the oil and gas sectors. Sectors such as consumer and technology do not exhibit any clear trends because the estimated slopes are either negative or not statistically significant.

Word clouds were used to identify the key topics that explain the changes in disclosure narrative over time. Three periods were considered: the climate risks in 2014–2016 and 2020–2021, and the climate opportunities in the 2023–2024 disclosures. The word clouds in Figure 7 show clear differences in the narrative between the different industries.

A major focus of the banking industry during the 2014–2016 period was the impact of lower demand for oil and commodities due to a slowdown in the Chinese economy, which adversely affected the economic growth of many African countries. Droughts in countries such as South Africa led to higher food prices and general difficult economic conditions. There were further cost pressures due to rising electricity prices, which reduced consumer spending. The 2020–2021 narratives focused on the severe impact of the COVID-19 pandemic. There was also an increased focus on the impact of transitioning to a low-carbon economy, particularly given the dependence on coal electricity generation. With the coal industry being a significant employer, disclosures highlighted the risk of increased unemployment. The climate opportunities explored in the 2023–2024 narrative focused on financing renewable energy and sustainable clients who support decarbonization innovations.

The mining disclosures initially focused on the increased production costs resulting from higher electricity costs and the energy supply shortages during the 2014–2016 period. Poor water infrastructure led to water quality degradation, while reduced rainfall and water restrictions had a negative impact on mining operations. There were increased environmental costs from disposing of waste in landfill sites and from managing the impacts of acidic mine water contaminating drinking water. During the 2020–2021 period, there was increasing water scarcity due to severe droughts and reservoirs running dry in some of the countries that the mines were operating in. There was an increased demand for greener and cleaner energy and transport. The companies also considered the changes in demand for metals and minerals due to the societal demands related to climate change. The climate opportunities explored during 2023–2024 considered ways to reduce the carbon footprint and new projects such as reforestation, research to improve soil health and renewable electricity initiatives. The oil and gas sector had similar focal points with climate opportunities, including the integration of renewable electricity into operations and investment into minerals that are crucial in new low-carbon technologies such as green hydrogen production.

The climate disclosures of the consumer industry exhibit no clear trends over time. During the 2014–2016 period, the weakness of the USD/ZAR exchange rate put pressure on imports and increased the costs of goods and services. The industry was negatively impacted by constrained consumer spending due to the rising costs of fuel and electricity. During 2020–2021, carbon pricing started to affect the cost of operations. Trading was negatively affected by increased power disruptions and the new regulatory directives on packaging led to increased operational costs. Water scarcity and severe droughts impacted on the agricultural supply chain. During 2023–2024, the focus shifted to the increased use of solar renewable energy, investment in solar battery capacity and continued investment into sustainable packaging and recycling.

5.4. Disclosure Impact on Share Price

Many studies have found a relationship between ESG sentiment and stock market returns (Erhemjamts et al., 2024; Ugurlu-Yildirim & Dinc-Cavlak, 2024; Gaies et al., 2025). In this section, the impact of the disclosures on investor sentiment and the subsequent impact on the stock price is analyzed for the South African market.

Table 6. Summary of the Spearman rank correlation between the share price returns with the economic and climate sentiment variables.

	Industry	${\mathit{X}}_1$	${\mathit{X}}_8$	${\mathit{X}}_{12}$	Number of Observations
Share Return at FYE	Banking	11%	10%	0%	140
	Consumer	12%	3%	−1%	91
	Insurance	7%	−6%	−28%	41
	Mining	12%	−2%	−5%	125
	Oil and gas	−2%	1%	−8%	30
	Pharma	−13%	3%	−2%	31
	Real estate	−18%	−25%	−6%	23
	Technology	17%	6%	−5%	47
Share Return 5d after FYE	Banking	−8%	−3%	9%	140
	Consumer	−6%	0%	13%	91
	Insurance	0%	1%	14%	41
	Mining	7%	−4%	−6%	125
	Oil and gas	−30%	−25%	−18%	30
	Pharma	−31%	−36%	−5%	31
	Real estate	5%	−2%	−26%	23
	Technology	20%	26%	3%	47
Share Return Month after FYE	Banking	−8%	−18%	−8%	140
	Consumer	−1%	−1%	12%	91
	Insurance	−36%	−36%	−27%	41
	Mining	−4%	−7%	−7%	125
	Oil and gas	−35%	−31%	−30%	30
	Pharma	0%	−7%	−21%	31
	Real estate	14%	−1%	−42%	23
	Technology	1%	5%	−12%	47
Share Return 3 Months after FYE	Banking	−5%	−13%	5%	140
	Consumer	6%	6%	−5%	91
	Insurance	−58%	−60%	−19%	41
	Mining	−9%	−18%	−15%	125
	Oil and gas	−28%	−28%	−25%	30
	Pharma	−13%	−28%	−19%	31
	Real estate	14%	3%	−16%	23
	Technology	−7%	−2%	14%	47
Share Return at FYE	All industries	8%	6%	3%	528
Share Return 5d after FYE	All industries	−7%	−11%	−3%	528
Share Return Month after FYE	All industries	−10%	−15%	−10%	528
Share Return 3 Months after FYE	All industries	−9%	−13%	−5%	528

summarizes the Spearman rank correlation between the share price returns at various periods around the FYE and for the different sectors, with the economic sentiment variable $X_1$ and the climate sentiment variables $X_8$ and $X_{12}$. The correlations are generally low when analyzing the trend over all the sectors. Where higher correlations are detected for certain sectors, they are considered artifacts of the relatively low sample size. The industries with more observations confirm no clear trend between the share price and the sentiment variables.

Figure 8 illustrates the relationship between the share price and the climate sentiment variable $X_{12}$. The different colors on the plot denote the three EBRD environmental ratings (low, medium and high). The cluster analysis confirms no clear relationship between climate sentiment and the share price; however, there is some trend when comparing the behavior based on the EBRD ratings. During the 2013–2018 period, companies with high EBRD ratings generally published negative sentiment associated with climate-related risks, whereas the companies with medium EBRD ratings had positive climate sentiment (focused on climate-related opportunities). Companies with low EBRD ratings were spread across positive and negative climate sentiment. During 2022–2024, the trends for the companies with low- and medium ratings still exhibited similar behavior, but the high rating companies had shifted towards more positive climate sentiment.

The cluster analysis based on the ESG ratings is summarized in Figure 9. It was only performed for the 2022–2024 period because only the recent (2024) ESG ratings were available. In contrast to the EBRD ratings, Figure 9 shows no clear trend between companies with different ESG ratings. The EBRD ratings are assigned based on industry and only cover the environmental factors, whereas the ESG ratings incorporate idiosyncratic risks and consider additional ESG issues such as governance. Generally, companies with low ESG ratings also have low EBRD ratings. However, sectors with high EBRD ratings may have improved ESG ratings due to idiosyncratic factors.

5.5. Predictive Power of Economic and Climate Sentiment

Corporate disclosures provide information on the risks, opportunities and strategies for future business growth. The disclosures empower investors to make informed decisions on business potential.

In this section, an exploratory regression analysis is conducted to assess whether economic and climate sentiment can effectively explain movements in stock prices. The regression is performed on the average monthly share price returns:

$$Z_t=\alpha +{\beta }_1{Z}_{t-1}+{\beta }_2{USDZAR}_t+{\beta }_3{bond}_t+{\beta }_4{Systemic}_t+{\beta }_5{SARB}_t+{{\beta }_{6}Sentiment}_t,$$

where $Z_t$ denotes the monthly share return, $\alpha$ denotes the regression intercept, ${\beta }_i$ denotes the regression coefficients, ${USDZAR}_t$ denotes the South Africa rand per dollar exchange rate returns, ${bond}_t$ denotes the 5-year South African bond yield, ${Systemic}_t$ denotes the monthly systemic index derived as per Equation (1), ${SARB}_t$ denotes the SARB leading composite indicator of the business cycle and ${Sentiment}_t$ denotes the economic or climate sentiment variable. The specific sentiment variable used in each regression is recorded in the last column of

Table 7. Exploratory regression results explaining stock price moves.

Corporate	$\mathit{\alpha }$	${\mathit{\beta }}_1$	${\mathit{\beta }}_2$	${\mathit{\beta }}_3$	${\mathit{\beta }}_4$	${\mathit{\beta }}_5$	${\mathit{\beta }}_6$	R2 Value	Sample Size	Sentiment Variable Included
Tech_3	0.110 ***	0.761 *	0.113	0.096	0.588 **	0.074	0.106 **	72%	180	X12 Sector
Bank_7	−1.275 *	0.047	−0.376 *	−0.014	1.095 *	0.037	3.254 *	58%	180	X1 Idiosyncratic
Bank_7	−0.135 **	0.095 ***	−0.344 *	−0.008	1.192 *	0.036	0.194 *	56%	180	X1 Sector
Bank_7	0.198 **	0.089 ***	−0.328 *	0.002	1.216 *	0.015	0.974 *	55%	180	X8 Idiosyncratic
Bank_8	−1.014 **	0.047	−0.297 *	−0.089	1.209 *	0.043	2.539 ***	55%	144	X1 Idiosyncratic
Bank_8	−0.033	0.057	−0.307 *	−0.075	1.226 *	0.037	0.984 ***	55%	144	X8 Idiosyncratic
Bank_7	−0.129 **	0.104 **	−0.342 *	0.003	1.198 *	0.021	0.159 *	55%	180	X8 Sector
Bank_7	−0.109 ***	0.101 ***	−0.361 *	0.008	1.111 *	0.007	0.095 ***	53%	180	X12 Sector
Bank_8	−0.157 *	0.072	−0.263 *	−0.077	1.223 *	0.067	0.187 *	53%	180	X1 Sector
Bank_8	−0.154 *	0.075	−0.260 *	−0.067	1.234 *	0.054	0.163 *	52%	180	X8 Sector
Bank_3	−0.136 **	0.068	−0.204 *	−0.133 **	1.238 *	0.075	0.099 ***	52%	180	X1 Sector
Bank_3	−0.135 **	0.068	−0.202 *	−0.128 ***	1.247 *	0.068	0.090 ***	52%	180	X8 Sector
Insurance_3	−0.745 *	−0.066	−0.121	−0.135 ***	1.365 *	−0.024	1.851 *	49%	156	X1 Idiosyncratic
RealEstate_1	0.111	0.492 *	0.037	0.050	0.948 *	−0.005	0.162 **	49%	180	X8 Sector
Insurance_3	0.018	−0.055	−0.120	−0.129 ***	1.405 *	−0.037	0.802 *	48%	156	X8 Idiosyncratic
RealEstate_1	0.110	0.525 *	0.047	0.049	0.936 *	−0.002	0.089 ***	47%	180	X12 Sector
Insurance_3	0.031	−0.040	−0.130	−0.123	1.396 *	−0.053	0.514 **	47%	156	X12 Idiosyncratic
Insurance_3	−0.155 *	−0.029	−0.105	−0.105	1.388 *	−0.030	0.126 **	46%	180	X1 Sector
Insurance_3	−0.154 *	−0.023	−0.111	−0.106	1.386 *	−0.028	0.118 **	46%	180	X8 Sector
Bank_9	0.182	0.042	−0.131 ***	−0.143 ***	1.239 *	−0.021	0.653 *	45%	180	X12 Idiosyncratic
Bank_9	−0.321 *	0.050	−0.111	−0.157 ***	1.253 *	0.002	1.404 **	45%	180	X1 Idiosyncratic
Bank_9	0.278	0.047	−0.120	−0.155 ***	1.227 *	−0.022	1.004 **	45%	180	X8 Idiosyncratic
Bank_9	−0.152 *	0.068	−0.108	−0.153 ***	1.292 *	−0.011	0.134 **	44%	180	X1 Sector
Bank_9	−0.148 *	0.070	−0.107	−0.145 ***	1.295 *	−0.021	0.109 ***	44%	180	X8 Sector
Consumer_3	−0.366 **	0.113 ***	−0.303 *	−0.194 *	0.662 *	0.012	0.998 ***	43%	156	X1 Idiosyncratic
Bank_1	−0.111 ***	0.115 **	−0.279 *	−0.104	0.942 *	0.076	0.133 **	43%	180	X1 Sector
Bank_1	−0.538 **	0.125 **	−0.285 *	−0.103	0.946 *	0.077	1.733 **	43%	180	X1 Idiosyncratic
Bank_1	0.168	0.117 **	−0.273 *	−0.101	0.958 *	0.067	0.654 **	43%	180	X8 Idiosyncratic
Mining_1	0.178	0.142 **	−0.074	0.148 ***	1.288 *	0.191 *	0.981 *	42%	180	X8 Idiosyncratic
Mining_1	−0.624 *	0.135 ***	−0.076	0.148 ***	1.256 *	0.188 *	2.104 *	42%	180	X1 Idiosyncratic
Mining_1	−0.021	0.151 **	−0.092	0.152 **	1.242 *	0.181 *	0.526 **	41%	180	X12 Idiosyncratic
Mining_1	−0.151 **	0.146 **	−0.089	0.140 ***	1.208 *	0.190 *	0.148 *	41%	180	X1 Sector
Mining_1	−0.149 **	0.156 **	−0.093	0.145 ***	1.213 *	0.189 *	0.130 **	40%	180	X8 Sector

* a 1% significant level; ** 5%; *** 10%.

. The strong relationship between the economic and climate sentiment variables established in Section 5.2 meant that only one sentiment variable can be considered at a time to ensure no multicollinearity issues in the regression. Only regressions where the variance inflation factors (VIFs) of all the variables were below 5 were considered.

The sentiment variables were considered both at a company level, denoted as ‘idiosyncratic’, and at the sector level, where the average sentiment across all companies in the sector was used and is denoted as ‘sector’ in Table 7.

The regressions were performed for each company listed in Table 1, and each one of the three sentiment variables, $X_1$, $X_8$ and $X_{12}$, as defined in Table 2. These three variables were included because $X_1$ is considered an economic sentiment variable; $X_8$ is a climate sentiment variable; and $X_{12}$ is a climate sentiment variable with more strict filters applied.

Only regressions in which the sentiment variable had a positive sign (${\beta }_4>0)$ and a statistically significant impact in explaining the share price return were selected. Positive sentiment is expected to have a positive impact on the share price. Table 7 lists all the outcomes where $R^2\ge 40\%$.

${USDZAR}_t$ and ${bond}_t$ were included based on the results from previous studies, such as those by Bossman et al. (2022), Javangwe et al. (2022) and Bonga-Bonga and Mpoha (2025); these studies showed useful market variables able to explain share price returns. ${USDZAR}_t$ is expected to exhibit a negative relationship with the stock price, indicating a currency appreciation is correlated with improved investor sentiment and thus higher stock returns. ${bond}_t$ is expected to have a negative relationship with the stock market, where a drop in investor sentiment leads to less investment in stocks and more investment in bonds.

The systemic index and SARB business cycle were included to capture systemic risks.

The regression analyses demonstrate that the sentiment variables produce statistically significant results for approximately half the corporations, mostly from the banking and mining sectors. The idiosyncratic- and sector-level sentiment variables are equally important; with the economic sentiment variables significant in 60% of the regressions compared to the 40% for the climate sentiment variables. The significance of the sentiment variables suggests that the disclosures provide a clear representation of the climate-related risks, thereby reducing the likelihood of greenwashing.

The analyses were performed over a 15-year period, which generally meant around 180 monthly observations. The sentiment variables are only available annually because they were extracted from the annual disclosures. The generally low $R^2$ values indicate that additional variables are necessary to accurately predict share price returns. It may also be an artifact of the extended time period over which the regressions are conducted. Using subsets for the different parts of the economic cycle may lead to improved outcomes. This is confirmed when considering the correlation between the actual and fitted stock price returns over time.

Figure 10 shows how the correlation varies among selected corporations over different periods. It shows that Bank_7 has correlations that fluctuate between 60% and 90%; that the correlation of Consumer_3 has been steadily increasing; and that OilandGas_2 shows low correlation in the most recent period. There are very different trends for the different corporations. Generally, the $R^2$ values reported in Table 7 provide an indication of the average correlation over the full period.

5.6. Discussion

The global climate news data have a strong bias towards negative sentiment, which indicates a focus on climate-related risks compared to opportunities over the past year (from July 2024 to July 2025). This may be due to the increased pressure by organizations such as the United Nations Framework Convention on Climate Change (UNFCCC) and the annual Conference of the Parties (COP) meetings, where global milestones are agreed by all member countries.

Interestingly, the EM-DAT natural disasters database does not demonstrate a significant increase in natural disasters over the period between 2000 and 2025. Figure 11 shows that the overall numbers of global disasters are at similar levels in 2024 to that of the early 2000s. Figure 12 shows that the total number of people affected has also not increased substantially. When only considering the most recent data from 2018 onwards, there has been an uptick in natural disasters.

The impact of new regulations from organizations such as the ISSB around the disclosure of decision-useful sustainability and climate risk information is clear from the increased volume of information available on the climate-related risks and opportunities that inform corporate strategies.

An analysis of climate sentiment extracted from JSE-listed corporations in the banking, mining and oil and gas sectors reveals a significant improvement in sentiment over time, transitioning from an initial focus on risks in 2010 to an exploration of opportunities in 2024.

Sectors such as consumer and technology do not exhibit any clear trends in the disclosures over time, but they also generally have good ESG and EBRD ratings. Companies with good ESG ratings are perceived as less risky and more sustainable.

The enhancements of the banking sector disclosures align with the SARB (2024) guidance and the strategic move towards more sustainable green lending (Gambacorta et al., 2024). Systemic risk and indirect exposure to climate risks are managed closely by these institutions.

The improved disclosure of the mining sector is in line with international studies that show that corporations in carbon-intensive industries, such as energy, materials and utilities, tend to disclose more climate-related information (Ding et al., 2023). The higher-risk nature of these corporations, as indicated by the ESG ratings, require increased disclosure to show how strategies are aligned to the global SDGs.

The word clouds emphasize the key topics that explain the changes in the disclosure narratives across the different industries. In recent years, the banking sector has focused on opportunities to finance renewable energy and sustainable clients whose efforts support achieving the SDGs. The mining and oil and gas sectors focused on new projects to reduce the carbon footprint, research to improve soil health, renewable electricity initiatives and investment into minerals that are crucial in new low-carbon technologies such as green hydrogen production. Many studies argue that low-carbon sustainable technology will open new opportunities in mining (Hodgkinson & Smith, 2021; Dou et al., 2023).

Corporate disclosures do not elicit an immediate reaction from investors based on the correlation study between the sentiment variables and the stock price at various periods following the FYE. The correlations were generally low, which contrasts with many other international studies where such a relationship was observed (Erhemjamts et al., 2024; Ugurlu-Yildirim & Dinc-Cavlak, 2024; Gaies et al., 2025). It is envisaged that this will change over time given new investment trends and increased reputational risks associated with climate.

The relationship between climate sentiment and the share price was explored further with a cluster analysis based on the EBRD environment ratings. The analysis indicated that companies with high EBRD ratings (mining and oil and gas sectors) generally published negative climate sentiment, whereas companies with medium EBRD ratings (real estate and general industrials) had positive climate sentiment. Companies with low EBRD ratings (banks, consumer, technology) are spread across positive and negative climate sentiments. Interestingly, the same cluster analysis based on the ESG ratings exhibited no trends at all. This may be explained given that the EBRD ratings are assigned based on industry and only cover environmental factors, whereas the ESG ratings incorporate idiosyncratic risks and other ESG-related issues such as governance. Companies with high EBRD ratings can have improved ESG ratings due to idiosyncratic factors. The results are in line with studies such as that by Masongweni and Simo-Kengne (2024), which established that ESG scores do not significantly affect the financial performance of South African companies.

An exploratory regression study indicated that climate and economic sentiment may contain information that can explain stock price moves over the longer term. The sentiment variables produced statistically significant results for around half the corporations, mostly from banking and mining. The significance of the sentiment variables indicates the corporate disclosures provide a clear picture of the climate-related risks and thus less risk of greenwashing. The results have important implications in asset allocation and provide an interesting direction for future research. Monitoring the sentiment may provide early-warning signals of systemic risk in certain industries, which is of interest to regulators given its impact on financial stability (H. Li et al., 2021; Jourde & Moreau, 2023; Monnin et al., 2024).

The study has some limitations that must be considered. Firstly, the sample of corporations is relatively small with only the JSE top 40 companies included, leading to a strong bias to the banking and mining sectors. The sample thus represents the behavior of the largest corporations in South Africa, but it is not necessary representative of all the sectors. Secondly, the sentiment variables are only available annually. Extracting sentiment at a higher frequency could improve the outcomes of the regression analyses, but then alternative data sources will have to be utilized. Finally, even though domain-specific NLP modules were used, there are still limitations in the accuracy of these algorithms.

6. Concluding Remarks

Corporate disclosures were analyzed using a domain-specific NLP, which assigned a positive, negative or neutral sentiment to each sentence. Both economic and climate sentiment were extracted using the pre-trained Python libraries FinBERT and ClimateBERT, respectively. The study extends existing research studies that primarily utilize textual analysis.

The study highlights the current state of climate-related disclosure and highlight areas to focus on in the future for regulators, policymakers and other stakeholders. The sentiment analysis of corporate disclosures highlighted a shift from climate-related risks in the earlier years to climate-related opportunities in recent years, specifically for the banking and mining sectors. The study revealed the corporations’ ability to adapt their strategies to climate-related risks. The trends were less pronounced in sectors with good ESG ratings.

The research contributes to clarifying the complex interactions between climate risk, disclosures, financial performance and investor sentiment. We found no evidence that climate disclosures caused a significant investor reaction immediately following the financial year-end; however, an exploratory regression study revealed that climate and economic sentiments contain information that explains stock price movements over the longer term. This has important implications in asset allocation and quantifying systemic risk. Systemic risk may affect the stability of the financial system and is closely monitored by the regulators. The results also indicate a promising direction for future research.

For corporations where the sentiment variables were shown to explain share price returns, this suggests that the disclosures provide a clear understanding of climate-related risks, thereby reducing the likelihood of greenwashing. However, this was only observed for around half the corporations in the study, which indicates that further work is needed on climate-related disclosures. Even low-risk corporations are exposed to these risks, albeit indirectly through supply chains or policy changes.