Skip to content

Advertisement

  • Research
  • Open Access

Sustainable strategy for corporate governance based on the sentiment analysis of financial reports with CSR

Financial Innovation20184:2

https://doi.org/10.1186/s40854-018-0086-0

  • Received: 3 July 2016
  • Accepted: 8 February 2018
  • Published:

Abstract

Focusing only on shareholders’ financial return is not consistent with the concept of sustainable corporate governance. In contrast to financial performance, corporate social responsibility (CSR) is a non-financial performance index. Financial reports consist of both financial and non-financial disclosures. These disclosures help investors make decisions. This paper characterizes the interaction between the sentiment analysis of financial reports and CSR scores. The classification accuracy through SVM exceeds 86%. The empirical study shows that the financial report sentiment based on the PESTEL model, Porter’s Five Forces model, and Value Chain (Primary and Support Activities) significantly correlates to the CSR score.

Keywords

  • Financial report
  • CSR score
  • Sentiment analysis
  • Object library

Background

Lawlor (2001) warned that even if an organization is “as pure as the driven snow in its own operations, it should expect public scrutiny of the practices of suppliers and associates”. Bad news about companies such as Nike, GAP, Reebok, and Hennes & Mauritz (H&M) attracted negative media attention in the late 1990s. In 2012, Apple became a high-profile target for non-government organizations concerned about working conditions in multinational supply chains. These concerns were related to the China-based Foxconn organization, where the vast majority of its 1.2 million employees were involved in assembling Apple products (according to Reuter’s reports). At the request of 250,000 petitioners, Apple was persuaded to ask the Fair Labor Association to investigate the working conditions at the Foxconn factory in China. Walmart was heavily criticized for its workplace practices in the US. In 2012, the National Employment Law Project (NELP) published the “Chain of Greed” report (Cho et al., 2012) about Walmart’s worker exploitation in the US. Some responsibility for the conditions of the poorly paid garment makers in Bangladesh has also been placed on the western retailers who sold the garments made by these oppressed workers after the collapse of the Rana Plaza building caused international outrage.

Since the Global Financial Crisis (GFC), and amid a lingering recession that has intensified pressure from shareholders, companies are devising new corporate social responsibility (CSR) models that are more aligned with their core business goals and services. For example, blue-chip companies, such as Visa and Unilever, are creating new markets in the developing world by closely aligning social causes with their overarching corporate strategies. CSR has a strong role to play in the provision of information for risk management purposes. The Carbon Disclosure Project (CDP) was formed in 2000. Based in New York and London, the CDP focuses on the implications of climate change for shareholder value and commercial operations. CDP believes that carbon emissions and climate change represent significant business risks and, therefore, an organization’s policies and performance in relation to climate change should be factored into investment decisions. The decision to report information (whether it be CSR or financial information) is the concept of “accountability”. It is defined as the duty to provide a report, or an account, of the actions and decisions made about those areas of activity for which an organization is deemed to be responsible. The Commission of European Communities (CEC) states that CSR is “a concept whereby companies integrate social and environmental concerns in their business operations and in their interaction with their stakeholders on a voluntary basis. Being socially responsible means not only fulfilling legal expectations, but also going beyond compliance and investing more into human capital, the environment, and the relations with stakeholders” (CEC 2001) , p. 6. The above definition is also consistent with the definition of CSR provided by the World Business Council on Sustainable Development: “The commitment of business to contribute to sustainable economic development, working with employees, their families, the local community, and society at large to improve their quality of life” (Holme & Watts, 2000), p. 10.

The 3 main pillars of sustainability include environmental, social, and economic sustainability. Social and environmental performance can affect an organization’s future reputation, brands, and its ability to attract talented staff and maintain consumer and public support (CPA Australia, 2015a). Based on the Stakeholder Theory, corporate governance includes both financial performance and CSR. CSR refers to a wide range of activities that an organization undertakes, from charity donations to the management of carbon emissions. Based on the Signaling Theory, a financial report (FR) is the fundamental tool investors use to make decisions. It contains financial ratios as well as textual notes. Through this textual information, investors can learn about the firm’s social responsibility. Sentiment analysis is a relatively mature technique. However, China’s capital market is growing and Chinese is a complex language. We developed a sustainable strategy for corporate governance based on the sentiment analysis of Chinese financial reports with CSR. We solved 3 main problems: (1) Identifying the subjective object description in financial reports, (2) building an object library based on strategic models, and (3) characterizing the interaction between the CSR score and financial report sentiment categories.

Literature review

Sustainability is widely recognized as one of the most important challenges facing the world today (Wan, Arief, & Rajagopalan, 2014). For most of the world’s largest companies, reporting on non-financial information appears to be a continuing trend. The communication of the social and environmental dimensions of a company plays a key role in the sustainable development of an organization, and therefore should be investigated in depth (Enrique & Michaela, 2015).

Multinational corporations play a prominent role in shaping the environmental trajectory of the planet. The integration of environmental costs and benefits into corporate decision-making has an enormous, but as yet unfulfilled, potential to promote sustainable development. Now is the time to take advantage of an explosion of sustainability commitments from business leaders and expanding pressure for sustainable practices from shareholders, financial institutions, and consumers (Kareiva, McNally, & McCormick Steve et al., 2015). CSR initiatives are crucial for achieving part of an inclusive growth vision. Organizations that are proactive rather than reactive may help achieve a sustainable inclusive growth via various CSR initiatives (Radhakrishnan, Chitrao & Nagendra, 2014). Research on CSR disclosure points to an increasing lack of completeness and decreasing amount of credibility in the information reported, as well as concerns about overall reporting practices. The evidence supports increasing skepticism about the use of CSR reporting practices as tools used to enhance perceived accountability (Giovanna, Silvia, & Federica, 2015). Standalone CSR reporting by retail companies appears to positively influence perceptions of a company’s reputation and may lead to increased appeal for socially responsible investors (Dennis & Na, 2014). Disclose CSR could increase transparency and non-financial accountability in capital markets. The empirical study reveals that family ownership reduces the level of CSR disclosure (Safaee M & M S Gerayli, 2017). Combining stakeholder theory with CSR research is a historic accomplishment (Zhang, Liang, & Yin, 2012). Chinese empirical data showed that CSR behavior is one of the key factors influencing consumers’ purchasing decisions (Ma, 2011). Data from listed Chinese companies in Shanghai demonstrated that the previous years’ CSR positively affects the current year’s corporate financial performance (CFP), and the current year’s CFP positively affects the CSR (Zhang, Jin, & Li, 2013). An empirical analysis showed that, assuming CSR is exogenous, the current CSR plays a significant positive role in promoting the current CFP, and vice versa (Yin, Liu, & Chen, 2014).

There is a vast amount of financial information on companies’ financial performance available to investors in electronic form (Salahuddin & Gow, 2016; Loughran & McDonald, 2011; Robert, Schumaker, & Hsinchun, 2009; Kloptchenko et al., 2004). While the automatic analysis of financial figures is common, it has been difficult to automatically extract information from the textual part of financial reports. The textual part contains more information than numerical part in an annual report (Chen et al., 2012; Hobson, Mayew, & Venkatachalam, 2012; Chan & Franklin, 2011; Feng, 2010; Feng, 2006; Kloptchenko et al., 2004). In recent decades, and with the advent of the eXtensible Business Reporting Language (XBRL), financial reports have experienced a great change in terms of the unified reporting process. Nevertheless, the unstructured part of financial reports, the footnotes, remains a barrier to accurate automatic and real-time financial analysis (Heidari & Felden, 2015). Currently, the application of data mining to auditing is at an early stage of development and researchers typically take a scatter-shot approach to it, investigating patterns in financial statement disclosures, text in annual reports and MD&As, and journal entries without the appropriate guidance, such as the lessons drawn from investigations on known fraud patterns (Gray & Debreceny, 2014). The sentiment (tone, opinion) has been assessed using several categorization schemes in order to explore the various aspects of language used in the annual reports of US companies. The results indicated that the sentiment information is an important forecasting determinant of financial performance and, thus, can be used to support the decision-making process of corporate stakeholders (Hajek, Olej, & Myskova, 2014). (Wang et al. 2016) investigated the correlation between mutual funds’ scale and return in China by text mining a large volume of online financial reports. They further employed K-means clustering for fund categorization, which enables the reliable examination of correlations between the fund’s scale and return. The findings highlighted the uniqueness of emerging markets while providing interesting guidelines for exploiting big data analytics for financial studies.

Corporate social responsibility is crucial for sustainable corporate governance. Researchers have realized the potential value of the textual analysis of financial reports for detecting fraud, managing risk, and forecasting future performance. However, few researchers have performed empirical studies on the sentiment analysis of financial reports based on strategic factors and CSR. This paper will illustrate how to perform an effective sentiment analysis of financial reports.

Methods

Research framework

The research framework is shown in Fig. 1 and includes 4 steps: (1) data collection and preprocessing, (2) subjective sentence identification, emotional trend classification, and object description extraction, (3) the object library of the financial reports, and (4) empirical analysis.
Fig. 1
Fig. 1

Research Framework

Data collection and preprocessing

To develop the sample for the transferrable object library, we used firm-year observations from different industries, including real estate, automobile, medicine, electric, communications, and energy. The final sample contained 50 financial reports from 2008 to 2013. The data came from www.cninfo.com.cn, which is the designated disclosure website of the China Securities Regulatory Commission. To distinguish our work from existing quantitative (financial ratios) research and mine the potential information contained in the textual content of the financial reports, we did not extract common words like “we” or “company” or industry-specific words and pure financial terminology such as “automobile”, “real estate”, or “asset-liability ratio”. The classification precision based on the sentence corpus was higher than that of the paragraph or document (Zheng, 2014); we extracted 32,767 sentences from the textual part of the financial reports based on the full-stop punctuation.

Identification of the object description

Using the approach found in (Bao and Datta’s 2014) work, we recruited 10 graduate students to label the sample sentences. Each student labels around 4000 sentences, among which 800 sentences are repetitive. Before labeling, they are briefed on the definition and trained on a number of real labeled examples. We keep communication and discussion during the whole process. The consistency of repetitive sentences reaches 80%. We only retain the subjective sentences whose labels are agreed upon by other annotators. The final sample set contained 2001 subjective sentences with a general emotional tendency and 4181 groups of related object descriptions. The expression of the subjective emotion was very complicated, and each sentence may not only have described the situation and emotional expression of a single aspect of the company. Although the variety of sentences means that some sentences will tend to approve some aspects of the company while others tend to criticize aspects of the company, a sentence’s general emotional tendency can be identified. The emotion tendency of the internal parts of 1 sentence can be different from the whole emotional tendency of the sentence. Therefore, we labeled the whole emotional tendency, related object description of each sentence, and emotional tendency of each object description (shown in Table 1).
Table 1

Examples of object description identification

Sentence emotion

Sentence

Object

Description

Description emotion

+ 1

2013 年,面对复杂的外部环境和日益激烈的行业竞争,公司坚持“有质量的增长”,规模稳步增长,经营效率不断提升。(Faced with complicated external environment and increasingly fierce industry competition in 2013, the company adheres to the growth of quality, makes a steady rise in scale, and constantly improves operational efficiency.)

external environment

complicated

−1

industry competition

increasingly fierce

−1

scale

steady rise

+ 1

operational efficiency

constantly improve

+ 1

+ 1

公司稳健的经营风格、审慎的财务管理以及良好的信用积累获得国际投资者认可,为拓展海外融资渠道创造了良好条件。(The firm’s stable operation style, prudent financial management and good credit accumulation are approved by international investors, which help create favorable conditions to expand overseas financing channels.)

operation style

Stable

+ 1

financial management

Prudent

+ 1

credit accumulation

Good

+ 1

−1

受市场形势变化以及公司积极进行产品结构的调整,主动关停部分老车型的影响,报告期内,公司产销规模出现了下滑。(The company’s production and marketing scale have declined during the reporting period, due to the influence of changes in market conditions and active close of partly old models in order to adjust product structure.)

production and marketing scale

Decline

−1

−1

主要归因于化工产品价格同比下跌幅度较大,化工板块业绩同比下降。(The performance of chemical sector decreased due to price fall of chemical products.)

product price

Decrease

−1

−1

这些都可能会对本公司生产经营和效益带来较大的影响。(All of these could lead to a large influence on the company’s product operation and efficiency.)

product operation and efficiency

large influence

−1

Note: the original corpus is in Chinese, and the contents in brackets are corresponding translation. Similarly hereinafter

Let us use the first sentence to illustrate this process of identification. Even though there is only one sentence (divided by full stop punctuation), the subjective expression involves 4 groups of the object description, which should be listed separately. Among them, 2 groups of emotional expression are negative descriptions about the company’s external environment, but the whole emotional tendency of this sentence is positive. Therefore, the general emotion of this sentence should be labeled as positive, while its constituent parts are labeled as having different emotional tendencies. We obtained a set of 1724 positive sentences and 277 negative sentences. Among the object description groups, there were 3482 expressing positive emotions and 699 expressing negative emotions. For both the sentence level and object description level, around 85% of this sample expressed positive emotions and only 15% expressed negative emotions. After removing the repeated items, there were 2637 objects and 1728 descriptions. The difference in these results may be explained by the fact that the describing objects in financial reports can differ in thousands of ways, while the emotional tendency is mainly expressed as either “good” or “bad”.

SVM testing

Text consists of characters and punctuation, constituting words, phrases, sentences, paragraphs, sections, chapters, and documents. Before the computer could automatically deal with textual content, we needed to find an idealized method of formal representation that reflected the potential content of the text and then help it to identify different texts. The Support Vector Machine (SVM), a supervised learning method, is a prominent method for text classification (Pang et al., 2002; Mullen & Collier, 2004; Whitelaw et al., 2005; Ni et al., 2007). The basic idea of the SVM is to use a training set to find the hyperplane in vector space and then separate the data points into as many different categories as possible. Through a trained SVM model, the emotional tendency of the testing data could be automatically predicted.

We labeled the subjective sentences of financial reports from different industries. After removing the data with missing values, we obtained a matrix of 1966*1728 (sentence*description). Based on the proportion of 3:1, we divided the data into a training set and a testing set.

Table 2 shows the SVM classification result. The precision ratio was 97.36% and recall ratio was 88.04% when identifying a positive emotional tendency. The precision ratio was 35.29% and recall ratio was 73.17% when identifying a negative emotional tendency. The total precision ratio reached 86.83% and the F-measure value was 51.81%. The positive precision was much higher than that for the negative identification, which can be explained by the aforementioned statistical results. When preparing financial reports, managers usually have a positive attitude or mood (instead of a negative or adverse emotional tendency) towards the company’s financial position, operation results, and cash flow. Thus, most sentences in the sample set expressed positive emotion and few were negative. This tendency was also reflected in the calculation of “the precision ratio of identifying the negative emotional tendency”. The denominator contained a large number of misjudged samples that the machine recognized as − 1 while the human label was + 1, which led to the lower precision ratio for negative expressions.
Table 2

Testing result of the SVM

 

Human label as + 1

Human label as − 1

Machine label as + 1

405

11

Machine label as − 1

55

30

Object library of the financial reports

SWOT strategic analysis is a common method for evaluating corporate performance, which contains an analysis of the internal environment and external environment. In the above section, we divided the financial reports into sentences based on full stop punctuation, then identified the emotional tendency of the sentences, and finally extracted the related object description. These objects were words or phrases that we wanted to classify. Our process was as follows.

First, we labeled these objects as internal or external.

If the text described the company’s external information, it belonged to the external strategic analysis. The strategic analysis of the external environment is usually subdivided into the macro environment analysis and industry environment analysis. The former analyzes the whole environment, which can influence all kinds of industries, so we used the PESTEL model for these data. The latter aims at the strategic factors in a specific industry outside of a company that affect the company being analyzed as well as all the other companies engaged in similar economic and operational activities in the same industry; for these data, we adopted Porter’s Five Forces model.

If the sentence described the company’s internal situation, it belonged to the internal strategic analysis. Companies cannot control external environment factors. The strategic analysis of the internal environment relates to controllable factors, which can be improved by management and other control measures. The Value Chain includes the Primary and Support Activities and is used for corporate internal analysis.

There were 11 (6 + 5) external object categories, shown in Table 3. After removing the repeated items, the statistical results indicated that, among the factors in the corporate external environment, companies were concerned about the economic factor, industry rivalry, and power of the buyer the most, which are issues related to company earnings.
Table 3

External object library of financial reports

Object category

Examples

count

Political

政策措施(policy measure)、宏观调控(macro-control)、监管政策(supervision policy)、利好政策(beneficial policy)

37

Economic

资金状况(capital position)、供需关系(supply and demand)、市场走势(market trend)、融资环境(financing environment)

200

Social

人口老龄化(population aging)、居民收入(resident income)、消费方式(consumption pattern)、消费意愿(consumption intention)

31

Technological

产品研发(R&D)、技术水平(technological level)、创新能力(innovation ability)、研发耗资(R&D cost)

16

Environment

安全事件(security affair)、环境污染(environment pollution)、自然灾害(natural disaster)、能源生产(energy production)

18

Legal

行业标准(industry standard)、价格管制(price control)、质量标准(quality standard)、限购措施(restriction measure)

11

Threat of new entrant

新兴市场(emerging market)、新产能扩展(new productivity extension)、跨界竞争(cross-border competition)、垄断性(monopoly)

5

Threat of substitute

外部刺激(external stimulus)、业务竞争(business competition)、产品更新(product update)、替代品价格(price of substitute)

11

Industry rivalry

市场分化(market segmentation)、行业竞争(industry competition)、经营压力(operation pressure)、市场规模(market size)

115

Power of supplier

原料价格(material price)、生产成本(production cost)、物价上涨(price increase)、市场供应(market supply)

45

Power of buyer

市场需求(market demand)、消费水平(consumption level)、潜在需求(potential demand)、下游监控(downstream monitor)

69

Subtotal:

 

558

Table 4 shows that the internal objects are more related to a firm’s infrastructure, marketing and sales, and operations and technology development than human resource management and services. The factors of least importance are inbound logistics, outbound logistics, and procurement.
Table 4

Internal object library of financial reports

Object category

Examples

count

Inbound logistics

资金成本(capital cost)、资源投入(resource investment)、原料基地(material base)、能源损耗(energy consumption)

39

Operations

运作效率(operation efficiency)、产品结构(product structure)、经营效果(management achievements)、业务流程(business process)

372

Outbound logistics

库存结构(inventory structure)、配送能力(logistics ability)、订单(order)、物流结构(logistics structure)

73

Marketing & Sales

销售结构(sales structure)、营销能力(marketing ability)、品牌影响力(brand influence)、渠道管理(channel management)

438

Service

服务能力(service ability)、资源保障(resource guarantee)、增值服务(value-added service)、售后故障率(failure after sale)

128

Procurement

采购流程(purchasing process)、采购成本(purchase cost)、原料采购(material purchase)、原料成本(material cost)

73

Technology development

研制水平(technological level)、技术垄断(monopoly)、创新体系(innovation system)、新产品开发(R&D)

252

Human resource management

薪酬分配(salary distribution)、绩效考核(performance assessment)、招聘渠道(recruitment channel)、培训环境(training condition)

146

Firm infrastructure

经营策略(business strategy)、公司治理(corporate governance)、管理效率(management efficiency)、财务结构(financial structure)

558

Subtotal:

 

2079

There were 345 sentences describing the external environment and 859 related groups of the object descriptions. There were 1656 sentences describing the internal environment and 3322 related groups of object descriptions, which were 4.8 times and 3.9 times as great, respectively, as the results from the external analysis. Among the related descriptions of the external environment, 41.68% described positive emotions and 58.32% described negative emotions. In contrast, 94.04% described positive emotions regarding the internal environment and 5.96% described negative emotions regarding the internal environment. These results demonstrated that the main sentiment expressed in these financial reports was about internal information disclosure, mainly good news about corporate performance. Compared with the internal disclosure, the description of the external environment revealed much more emotion that is negative.

Results and discussion

We developed a regression model for the CSR score and financial report sentiment, shown below:
$$ Log\left({CSR}_{it}\right)=\alpha . Log\left({CSR}_{it-1}\right)+\beta . Control\_{variables}_{it}+\gamma . FR\_{categories}_{it}+\varepsilon $$

The CSR score data was taken from www.hexun.com. Control variables included the ROE, BM ratio, and company size, which were obtained from the China Stock Market & Accounting Research (CSMAR) database. The empirical data set consisted of 16 companies (000002 Vanke, 600050 China Unicom, 600028 SINOPEC, 601857 CNPC, 600104 SAIC Motor, 000800 FAW CAR, 000927 TJ FAW, 000572 HaiMa Motor, 601607 Shanghai Pharma, 000538 Yunnan Baiyao, 000423 DEEJ, 600085 TongRenTang, 000651 GREE, 600690 Haier, 600839 Sichuan Changhong, and 000016 KONKA A) in various industries over a period of 2 years, 2012 and 2013. Based on the object library we built and emotional tendencies classified in the above sections, we calculated the financial reports’ sentiment value of each category using Models 1–6.

Model 1:
$$ Log\left({CSR}_{it}\right)=\alpha . Log\left({CSR}_{it-1}\right)+\beta 11.{ROE}_{it}+\beta 12.{BM}_{it}+\beta 13. Log\left({Size}_{it}\right)+\varepsilon $$
Model 2:
$$ Log\left({CSR}_{it}\right)=\alpha . Log\left({CSR}_{it-1}\right)+\beta 11.{ROE}_{it}+\beta 12.{BM}_{it}+\beta 13. Log\left({Size}_{it}\right)+\gamma 11.{FR}_{it}+\varepsilon $$
Model 3:
$$ Log\left({CSR}_{it}\right)=\alpha . Log\left({CSR}_{it-1}\right)+\beta 11.{ROE}_{it}+\beta 12.{BM}_{it}+\beta 13. Log\left({Size}_{it}\right)+\gamma 21.{External}_{it}+\gamma 22.{Internal}_{it}+\varepsilon $$
Model 4:
$$ {\displaystyle \begin{array}{c} Log\left({CSR}_{it}\right)=\alpha . Log\left({CSR}_{it-1}\right)+\beta 11.{ROE}_{it}+\beta 12.{BM}_{it}+\beta 13. Log\left({Size}_{it}\right)\ \\ {}+\gamma 31.{PESTEL}_{it}+\gamma 32.{Porter}_{it}+\gamma 33.{ValueChain}_{it}+\varepsilon \end{array}} $$
Model 5:
$$ {\displaystyle \begin{array}{c} Log\left({CSR}_{it}\right)=\alpha . Log\left({CSR}_{it-1}\right)+\beta 11.{ROE}_{it}+\beta 12.{BM}_{it}+\beta 13. Log\left({Size}_{it}\right)\ \\ {}+\gamma 41.{PESTEL}_{it}+\gamma 42.{Porter}_{it}+\gamma 43.{Primary}_{it}+\gamma 44.{Support}_{it}+\varepsilon \end{array}} $$

Model 6:

Log(CSR it ) = α. Log(CSRit − 1) + β11. ROE it  + β12. BM it  + β13. Log(Size it ) + γ51. Political it  + γ52. Economic it  + γ53. Social it  + γ54. Technological it  + γ55. Environment it +γ56. Legal it  + γ57. Entrant it  + γ58. Substitute it  + γ59. Rivalry it  + γ60. Supplier it + γ61. Buyer it  + γ62. Inbound it  + γ63. Operations it  + γ64. Outbound it + γ65. Marketing it  + γ66. Service it  + γ67. Procurement it  + γ68. Technology it + γ69. Human it  + γ70. Infrastructure it  + ε

Table 5 presents the descriptive statistics for our sample. The units of the CSR and Size are obviously different from the other variables, so we used the log value of these 2 variables for the regression.
Table 5

Summary statistics of the sample data

Variable

Min

Max

Mean

Standard Deviation

CSR

0.61

86.66

59.00

23.27

Control Variables

    

 ROE

−0.10

0.31

0.11

0.11

 BM

0.17

1.46

0.83

0.35

 Size

14.74

21.10

17.51

1.60

FR Categories

    

 FR

−0.09

1.00

0.71

0.27

 External

−1.00

1.00

−0.10

0.57

 Internal

0.42

1.00

0.90

0.16

 PESTEL

−1.00

1.00

0.11

0.68

 Porter

−1.00

1.00

−0.40

0.60

 Primary

0.29

1.00

0.88

0.20

 Support

0.40

1.25

1.00

0.17

 Political

−1.00

1.00

0.00

0.76

 Economic

−1.00

1.00

0.09

0.54

 Social

0.00

1.00

0.26

0.44

 Technological

−1.00

1.00

0.13

0.42

 Environment

−1.00

1.00

0.03

0.47

 Legal

−1.00

1.00

−0.13

0.49

 Entrant

−1.00

1.00

0.00

0.25

 Substitute

−1.00

1.00

−0.06

0.44

 Rivalry

−1.00

1.00

−0.45

0.57

 Supplier

−1.00

1.00

−0.39

0.56

 Buyer

−1.00

1.00

0.03

0.64

 Inbound

−1.00

1.00

0.37

0.60

 Operation

0.00

1.00

0.88

0.27

 Outbound

0.00

1.00

0.64

0.48

 Marketing

0.00

1.00

0.85

0.27

 Service

−1.00

1.00

0.78

0.47

 Procurement

0.00

1.00

0.58

0.48

 Technology

0.00

1.00

0.83

0.35

 Human

0.00

1.00

0.80

0.39

 Infrastructure

−0.33

1.00

0.87

0.31

Table 6 shows the experiment results, among which Model 5 demonstrates a statistical significance between the CSR score and financial report sentiment based on the categories of the PESTEL, and Porter’s Five Forces, Primary Activities and Support Activities. It outperformed the other models. The signs of the 4 variables’ coefficients were consistent with the aforementioned statistical results. When preparing a financial report, managers express positive emotions in accordance with Porter’s Five Forces and Support Activities, which show the company’s opportunities and strengths. Managers may also complain about the threat from the macro environment and weakness of the Primary Activities.
Table 6

Regression results of different models

Variable category

Variable

Model 1

Model 2

Model 3

Model 4

Model 5

Model 6

Intercept

 

−1.4860

−1.4306

− 1.5804

−1.0108

− 0.9943

0.5900

Control variables

Log(CSRt-1)

0.3392·

0.3174

0.3250·

0.3378·

0.2761·

0.3483

ROE

− 0.2841

− 0.1069

0.0012

0.1019

−0.7234

− 0.2213

BM

−0.4080·

−0.3889

− 0.3698

−0.4782·

− 0.7948**

−0.7326

Log(Sizet-1)

2.3923

2.4235

2.4579

2.1105

2.1349

0.7702

FR categories

FR

 

−0.1311

    

External

  

−0.1072

   

Internal

  

−0.0406

− 0.0598

  

PESTEL

   

−0.1812

−0.2689*

 

Porter

   

0.0825

0.2839*

 

Primary

    

−1.1081*

 

Support

    

1.4174**

 

Political

     

−0.1021

Economic

     

0.0174

Social

     

−0.0595

Technological

     

−0.0220

Environment

     

0.0359

Legal

     

0.1794

Entrant

     

0.0997

Substitute

     

0.1795

Rivalry

     

0.2797

Supplier

     

0.0553

Buyer

     

−0.1574

Inbound

     

−0.1462

Operation

     

0.3617

Outbound

     

−0.1118

Marketing

     

0.1446

Service

     

−0.0820

Procurement

     

0.3196

Technology

     

0.4783

Human

     

−0.2218

Infrastructure

     

−0.3058

Adjusted R2

 

0.3660

0.3484

0.3386

0.3608

0.5203

0.0283

Variable “Internal” in Model 3 is actually “ValueChain” in model 4, their value are the same

·* and ** indicate p < 10%, p < 5% and 1% respectively

Based on Model 5, we calculated the predicted values for the CSR scores. The comparison between predicted values and true values of the CSR scores is plotted as Fig. 2, which shows a consistent trend.
Fig. 2
Fig. 2

The Comparison of CSR scores

Managerial implications

Milton Friedman, the noted economist, argued that the primary responsibility of managers is to maximize shareholder wealth. However, any director or senior manager who believes that acting for the exclusive benefit of its shareholders will lead to long-term success and satisfactory corporate governance is mistaken. It is often forgotten that Friedman also mentioned that this pursuit of shareholder wealth should be “within the rules”.

In the real world, ignoring stakeholders’ interests will result in poor performance and could even result in corporate bankruptcy. Based on the Stakeholder Theory, an organization’s success depends on the successful management of all the relationships that the organization has with its stakeholders. These stakeholders include not just shareholders (for corporate entities), but also other parties who could be affected by the operations of an entity, such as its employees, competitors, customers, suppliers, lenders, government, community, and environment. Paying attention to a broad range of stakeholders is important from a moral or ethical perspective and the survival of a firm depends on the management of a range of relationships.

This research contributes to stakeholders’ comprehensive decision-making ability in regards to corporate accountability. The attitude of a manager towards CSR interacts with the emotional tendency expressed in the financial report. Stakeholders can learn more about a company’s sustainable accountability through the sentiment analysis of its financial report. Companies can also do better by paying attention to CSR and by pursuing sustainable corporate governance.

Conclusion

We developed an object library for financial reports that consisted of 20 categories based on strategic analysis. Regression models of the CSR score and financial report sentiment of different categories demonstrate that when financial report sentiment is classified into 4 categories based on the PESTEL model, Porter’s Five Forces model, the Primary Activities and Support Activities, it adds statistically significant explanatory power to the CSR score. The result is transferrable to different industries. In the future, we can apply the method for creating an object library of financial reports to other financial disclosures.

Declarations

Funding

National Natural Science Foundation of China (No. 71371144, 71601119, 71771177).

Authors’ contributions

Yuan Song collected data and wrote the main paper. Hongwei Wang gave the main idea and improvement suggestions. Maoran Zhu contributed to the experiment analysis. All the authors read and approved the final manuscript.

Authors’ information

Yuan Song is a PhD candidate at School of Economics and Management, Tongji University, Shanghai 200,092, China. Her research interests include corporate governance, auditing and textual analysis.

Hongwei Wang is currently a professor of Information Systems in the School of Economics and Management at Tongji University, China. His research interests include sentiment analysis, social media, and business intelligence.

Maoran Zhu is an associate professor of Management Science in the School of Economics and Management at Tongji University, China. His research interests include E-commerce and business intelligence.

Competing interests

The authors declare that they have no competing interests.

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

Authors’ Affiliations

(1)
School of Economics and Management, Tongji University, Shanghai, 200092, China

References

  1. Bao Y, Datta A (2014) Simultaneously discovering and quantifying risk types from textual risk disclosures. Manag Sci 60(6):1371–1391View ArticleGoogle Scholar
  2. Bonson E, Bednarova M (2015) CSR reporting practices of Eurozone companies. Revista de Contabilidad-Spanish Accounting Review 18(2):182–193View ArticleGoogle Scholar
  3. CEC (Commission of European Communities) (2001), Promoting a European framework for corporate social responsibility, Green Paper, Commission of European Communities, Brussels, accessed June 2014, http://eur-lex.europa.eu/summary/chapter/employment_and_social_policy.html?root_default=SUM_1_CODED=17
  4. Chan SWK, Franklin J (2011) A text-based decision support system for financial sequence prediction. Decis Support Syst 52(1):189–198View ArticleGoogle Scholar
  5. Chen CL, Liu CL, Chang YC, Tsai HP (2012) Opinion Mining for Relating Subjective Expressions and Annual Earnings in us financial statements. J Inform Sci Engineering 29(4):743–764Google Scholar
  6. Cho E H, Christman A, Emsellem M, Ruckelshaus C K, & Smith R (2012), Chain of greed: How Walmart’s domestic outsourcing produces everyday low wages and poor working conditions for warehouse workers, accessed June 2014, http://www.nelp.org/page/-/Justice/2012/ChainOfGreed.pdf?nocdn=1
  7. CPA Australia 2015. CPA Program Global Strategy and Leadership [M]., Version 15aGoogle Scholar
  8. Gray Glen L, Debreceny Roger S (2014) A taxonomy to guide research on the application of data mining to fraud detection in financial statement audits. Int J Account Inf Syst 15(4):357–380View ArticleGoogle Scholar
  9. Heng W, Junjie W, Shi Y, Jian C (2016) On Characterizing Scale Effect of Chinese Mutual Funds via Text Mining. Signal Process 124:266–278View ArticleGoogle Scholar
  10. Hobson JL, Mayew WJ, Venkatachalam M (2012) Discussion of analyzing speech to detect financial misreporting. J Account Res 50(2):393–400View ArticleGoogle Scholar
  11. Holme R, Watts P (2000), Corporate Social Responsibility: Making good business sense, World Business Council on Sustainable Development. http://www.ceads.org.ar/downloads/Making%20good%20business%20sense.pdf. Accessed Feb 2018.
  12. Kai-guo Y, Xiao-qin L, Hua-dong C (2014) Study on the relationship between corporate social responsibility and financial performance from the endogenous perspective-evidence from Chinese listed companies. China Soft Science 06:98–108 (in Chinese)Google Scholar
  13. Kareiva Peter M, McNally Brynn W, Steve MC, Tom M, Mary R (2015) Improving global environmental management with standard corporate reporting. Proc Natl Acad Sci U S A 112(24):7375–7382View ArticleGoogle Scholar
  14. Kloptchenko A, Eklund T, Karlsson J, Back B, Vanharanta H, Visa A (2004) Combining Data and Text Mining Techniques for Analysing Financial Reports. International Journal of Intelligent Systems in Accounting, Finance and Management 12(1):29–41View ArticleGoogle Scholar
  15. Lawlor P (2001). Chasing the triple bottom line, IT Web News, 27 July, accessed June 2014, http:// allafrica.com/stories/200107270552.html
  16. Li F (2006). Do stock market investors understand the risk sentiment of corporate annual reports?. Ssrn Electronic Journal Google Scholar
  17. Li F (2010) Textual analysis of corporate disclosures: a survey of the literature. J Account Lit 29:143–165Google Scholar
  18. Loughran T, Mcdonald B (2011) When is a liability not a liability? Textual analysis, dictionaries, and 10-Ks. J Financ 66:35–65View ArticleGoogle Scholar
  19. Ma L-l (2011) A study on the mechanism of the impact of the Company' s social responsibility on Consumers' willingness of purchase. Management World 05:120–126 (in Chinese)Google Scholar
  20. Maryam H & Felden Carsten (2015). Impact of text mining application on financial footnotes analysis. 10th international conference on design science research in information systems and technology (DESRIST). Dublin, IRELANDGoogle Scholar
  21. Michelon G, Pilonato S, Ricceri F (2015) CSR reporting practices and the quality of disclosure: an empirical analysis. Crit Perspect Account 33:59–78View ArticleGoogle Scholar
  22. Mohammad S, Jeff G (2016) The effects of internet usage, financial development and trade openness on economic growth in South Africa: a time series analysis. Telematics Inform 33(4):1141–1154View ArticleGoogle Scholar
  23. Mullen T & Collier N (2004). Sentiment analysis using support vector machines with diverse information sources. Proceedings of EMNLP-2004, Barcelona, Spain, 412-418Google Scholar
  24. Ni X, Xue G, Ling X, Yu Y & Yang Q (2007). Exploring in the weblog space by detecting informative and affective articles. Proc. of the 16th Int. Conf. On world wide web, 281–290Google Scholar
  25. Pang B, Lee L & Vaithyanathan S (2002). Thumbs up? sentiment classification using machine learning techniques The Conference on Empirical Methods in Natural Language Processing Philadelphia, US, 79–86Google Scholar
  26. Patten DM, Na Z (2014) Standalone CSR reporting by U.S. retail companies. Account Forum 38:132–144View ArticleGoogle Scholar
  27. Petr H, Vladimir O, Renata M (2014) Forecasting corporate financial performance using sentiment in annual reports for stakeholders’ decision-making. Technol Econ Dev Econ 20(4):721–738View ArticleGoogle Scholar
  28. Robert P, Schumaker, Chen H (2009) Textual analysis of stock market prediction using breaking financial news: the Azfin text system. ACM Trans Inf Syst 27(2):1139–1141Google Scholar
  29. Safaee M, Gerayli MS (2017) Family control and corporate social responsibility: evidence from Iranian companies. Int J Fin Engineering 04(04)Google Scholar
  30. Srinivasan R, Pradnya C, Asha N (2014) Corporate social responsibility (CSR) in market driven environment. Procedia Economics and Finance 11:68–75View ArticleGoogle Scholar
  31. Te Liew W, Adhitya A, Srinivasan R (2014) Sustainability trends in the process industries: a text mining-based analysis. Comput Ind 65:393–400View ArticleGoogle Scholar
  32. Whitelaw C, Garg N & Argamon S (2005). Using appraisal groups for sentiment analysis. Proc. Of the 14th ACM Int. Conf. on Information and Knowledge Management, 625-631Google Scholar
  33. Zhao-guo Z, Liang Z-g, Kai-guo Y (2012) Research on corporate social responsibility from the stakeholder perspective. China Soft Science 02:139–146 (in Chinese)Google Scholar
  34. Zhang Zhao-guo, Jin Xiao-cui, Li Geng-qin (2013). An empirical study on the interactive and inter-temporal influence between corporate social responsibility and corporate financial performance. Accounting Research, 08, 32–39+96. (in Chinese)Google Scholar
  35. Zheng LJ (2014) Users’ sentiment analysis and application of Chinese online reviews. Tongji UniversityGoogle Scholar

Copyright

© The Author(s). 2018

Advertisement