The use and trend of emotional language in the banks ’ annual reports : the state of the global financial crisis

This study is of an exploratory nature as it seeks to explore the extent to which the language of emotions in the banks’ annual reports is affected by the global financial crisis (GFC). The language of emotions was analyzed using eight categories (trust, anticipation, sadness, anger, fear, disgust, surprise and joy) in annual reports of 12 listed banks from six countries in the Middle East area (namely, Jordan, Kingdom of Bahrain, United Arab Emirates, Sultanate of Oman, Kuwait, Kingdom of Saudi Arabia) from 2002 to 2017. The final data set consists of 192 bank-year observations. The study time was divided into three periods (pre, during and post GFC). In addition, the study enriches accounting literature by being the first study to test Pollyanna hypothesis using emotion analysis. The results of the study show that the percentage of emotional words in banks’ annual reports (2002–2017) represents almost 22% on average. The trust, anticipation and fear categories were the most affected than other emotional categories during GFC. While the trust category decreased, both the fear and anticipation categories increased. Other findings of the study show that regardless of GFC, emotional words of trust and anticipation categories in banks’ annual reports have dominated the emotional words of the disgust and surprise categories. Therefore, Pollyanna hypothesis is supported. In contrast to the emotional words of the joy category in banks’ annual reports which has not dominated the sadness category. In this case, Pollyanna hypothesis is rejected.


INTRODUCTION
The global financial crisis (GFC) at the beginning of 2008 has had negative and serious impacts on many financial institutions in all countries of the world, especially the banks that were among the victims of this crisis. Many banks, especially in the United States (US), have been affected by GFC. Some banks have tried to survive, while others have collapsed (Alnajjar et al., 2010; Wall Street Journal, 2007; Olaniyi & Olabisi, 2011). The financial system in the banks faced severe financial difficulties during GFC. The crisis was preceded by a direct growth in the debt ratio with a sharp rise in the prices of assets and real estate and a severe shortage of liquidity. The top management in banks did not notice that risk until some banks collapsed (Norris, 2007;Wall Street Journal, 2007). Because of the great effects of GFC on various aspects of business and financial markets, many researchers and analysts have examined these effects in different ways that led to the richness of accounting literature. For example, Nazir et al. (2012) analyzed the impact of GFC on the banks' financial performance in Pakistan, Chan-Lau (2010) investigated the effect of GFC on the Chilean bank-ing system, while Olaniyi and Olabisi (2011) reported the negative effects of GFC on the financial performance of the banking sector in Nigeria, namely deposits, advances, and investments. The report of the United Nations (UN) in 2011 titled "World Economic Situation and Prospects 2011" pointed out that "the effect of GFC is expected to continue into 2011 and 2012 as weaknesses in major developed economies continue to provide a drag on the global recovery and pose risks for world economic stability in the coming years" (UN, 2011, p. iii). The importance of the topic of GFC gives the motivation for the current study that seeks to explore the impact of GFC on the use of the language of emotions in banks' annual reports from six countries in the Middle East area over 16 years (from 2002 to 2017). In addition, it tests the validity of Pollyanna hypothesis, which argues that regardless of the performance, or circumstances of the companies, managers will continue to use positive expressions rather than negative ones (Hildebrandt & Snyder, 1981).
The beginning of the 80's has witnessed a great deal of interest by many researchers to study corporate annual reports, the language, content, and style. Corporate annual reports occupy a special and unique position for a large segment of users such as shareholders, investors, analysts, lenders, regulators and others (Arfin, 1993; Thomas, 1997). Consequently, accounting literature (Loughran & McDonald, 2011;Chen, 2014;Huang et al., 2014; offers several studies on the analysis of language and the content of firms' annual reports. The results of these studies indicate that firms' annual reports contain a stock of information about corporate financial performance as well as other information indicating financial risks or fraud. For example, Jaeschke et al. (2018, p. 174) pointed out that "the language used in the Securities and Exchange Commission (SEC) filings has been shown to provide information about a firm's market and litigation risks". In this regard, Tama-Sweet (2014, p. 276) stated that "the tone of earnings announcements provides information to market participants and that this information is impounded into stock price". While, Smith and Taffler (2000) argued that the content of the chairperson's statement is associated with a firm's financial performance. Prior studies reported a direct link between managers' language or tone in financial disclosures and a firm's performance such as Li (2008) who found that firms with low profits and high risks used complex language in their disclosures to hide their problems. Similar results were reported by Subramanian et al. (1993) who provided evidence on companies' financial reports with good financial performance are more readable and clearer than financial reports of companies with poor performance. Hope et al. (2016) also reported these results. In this line, Kohut and Segars (1992) found that firms with good financial performance use different communication strategies in their annual reports with a different number of words and sentences, which may help to future predict their performance. However, the results of Hildebrandt and Snyder (1981) supported "Pollyanna Hypothesis" that suggests that the language of the Chairman of the Board to shareholders in the annual report will often be positive regardless of the firm's financial performance (Hildebrandt & Snyder, 1981, p. 5).
The importance of the current study steams from a number of aspects as follows. First, several studies have addressed the content and language of financial disclosures and documented a direct link between such disclosures and different critical matters such as risks, losses, stock prices, illegal acts, fraud, financial performance and financial crisis (Li, 2008 The current study provides a valuable contribution to corporate disclosure literature by using a unique analysis, an emotion analysis, which has not been used in previous studies in the Middle East area. The current study is different from other studies such as Lee et al. (2014) who relied on the sentiment analysis to investigate the association between the textual analyses of financial reports for a sample of US firms and the prediction of stock prices. Similar to Lee et al. (2018) who used the text and data mining to examine the association between sales performance and the text patterns of the US firms' annual reports. "Sentiment analysis is a technique for determining whether a sentence's tone is positive, negative or neutral" (Lee et al., 2018, p. 12). However, in the current study, the emotion analysis as a new and different technique is conducted because it is broader than the sentiment analysis. While emotional analysis includes several emotional words such as confidence, expectation, surprise, fear, and others, the

LITERATURE REVIEW
The idea of studying the language style or tone in financial reports dates to relatively old studies. For example, Hildebrandt and Snyder (1981) studied the frequency of use of positive and negative words in financial reports as an attempt to answer the question of whether the language used by companies in their financial reports varies according to their performance. At the same time, do companies use more positive words than negative ones in their reports even if their financial performance is bad? A similar study was conducted by Thomas (1997). The results of both studies showed that companies use positive words more than negative words regardless of their financial performance. Moreover, a number of prior studies indicated that using the language analysis of financial reports provides valuable information to various decision makers on several matters such as the future of the company, financial performance, risks, corporate corruption and others (Henry, 2008;Rogers et al., 2011;Jaeschke et al., 2018). Despite the association between management language in annual reports or disclosure tone and firm performance, which has started to receive major interest in accounting literature, the findings are mixed.
A stream of studies (Davis et al., 2012) documented a relationship between managers' use of optimistic language to express the firm's future performance, in their reports to stockholders and investors. For example, the results of Huang et al. (2014) indicated that managers use a biased language, which is the optimistic language when the company's financial performance is bad. Kothari et al. (2009) reported that the firm's stock return volatility is significantly associated with a negative tone in corporate disclosures. Tetlock et al. (2008) found a relationship between a firm's future earnings and the tone of corporate disclosure, which suggests that managers use the tone, which reflects future stock returns. Chen (2014) used the text analysis of firms' annual reports focusing on narrative sections in these reports to detect fraud. The author filtered fraudulent language in texts of these reports and differentiated between fraudulent and non-fraudulent annual reports. Chen (2014) concluded that the language in firm's annual reports could indicate the possibility of fraud. This is similar to Goel et al. (2010) who investigated fraud in corporate financial disclosure and found that the writing style of fraudulent firms is different from non-fraudulent firms because fraudulent firms use long paragraphs with long sentences so that their reports are less readable. In Egypt, Aly et al. (2018) found that financial performance of Egyptian firms is one of the main determinants of tone disclosure where Egyptian firms disclose more information when they have good news related to their financial performance. In US, the results of Jaeschke et al. On the other hand, Rogers et al. (2011) presented conflicting results. The authors pointed out that companies facing legal problems use a more optimistic language in their financial reports. In the same line, Nelson and Pritchard (2016) showed that firms with litigation risks provide more readable financial disclosures than other firms. Schleicher and Walker (2010) showed that firms with losses use more positive tone in their disclosures compared to firms with earnings which provide a more negative tone. Using sentiment analysis, Lee et al. (2018) showed that corporate sales performance is not associated with the style of language in the annual report, in other words, a positive or negative tone.

Theoretical framework of the study
Several theoretical perspectives provide different explanations for managers' motivations for corporate disclosure such as stakeholder theory, legitimacy theory, agency theory and signaling theory. However, agency theory is one of the most widely used theories to explain corporate disclosure and managers' behavior and their choice of a certain information to disclose or to avoid the disclosure of other information. Agency theory offers a number of advantages for corporate disclosure such as reducing information asymmetry (Watson et al., 2002), conflicts of interest between managers and owners (Jensen & Meckling, 1976), and the cost of capital (Easley & O'Hara, 2004;Rajgopal & Venkatachalam, 2011). In line with agency theory, it can be argued that corporate disclosure provides an opportunity for the company to create a certain image of the company in the minds of investors and convince them that the company is working well to maximize their benefits (Abraham & Cox, 2007). The current study tests the extent to which the emotional language in the annual reports of a sample of banks is affected by GFC. It is argued that managers use the emotional language in their annual reports to deliver a specific message to the company's stakeholders. This message may be related to financial performance and other matters. Accounting literature provides evidence through some studies on the use of language styles in financial reporting in a biased manner related to the financial performance of the company (

Hypotheses development
The general Pollyanna hypothesis in business suggests "Regardless of the financially good or bad years of a corporation, the communication in the annual letters to the stockholders will be predominantly positive" (Hildebrandt & Snyder, 1981, p. 5).
The current study tested the validity of Pollyanna hypothesis in terms of using the emotional language in banks' annual reports. It used eight basic emotional categories (joy against sadness, anger against fear, trust against disgust, and surprise against anticipation) that suggested by Plutchik (1980, p. 1) (more details on these emotional categories are explained in the research method section).
Based on the previous discussion, the following hypotheses are suggested: H1: The emotional words of the trust category dominate the emotional words of the disgust category in banks' annual reports regardless of GFC.
H2: The emotional words of the joy category dominate the emotional words of the sadness category in banks' annual reports regardless of GFC.
H3: The emotional words of the anticipation category dominate the emotional words of the surprise category in banks' annual reports regardless of GFC.
H4: The emotional words of the fear category dominate the emotional words of the anger category in banks' annual reports regardless of GFC.

Emotion analysis (the lexicon-based approach)
Since the main objective of the current study is to explore the effect of GFC on the use and trend of emotional words in 192 banks' annual reports from six countries over 16 years (2002-2017), emotion analysis was used. This analysis is seen as the science of studying and extracting information that expresses a person's feeling through speaking, text, film, photo and other aspects. Recognition of emotional language from text is a new field of study that is connected to sentiment analysis (Ortony et al., 1988 . In 1980, Plutchik introduced the concept of "wheel of emotion" that depends on how several emotions can be mixed into one another to make new emotion". Plutchik (1980) proposed eight basic emotions: "each composed of two basic ones: joy against sadness; anger against fear; trust against disgust; and surprise against anticipation" (Plutchik, 1980, p. 1). These eight basic emotions suggested by Plutchik (1980) were used in the current study.
In general, there are two approaches for the problem of extracting emotions automatically. First, the lexicon-based approach which 1 For more details, see Mohammad (2018) and Saif et al. (2012). depends on ready words or phrases bag from known lexicon and calculating emotional words or phrases in a document using a software (Aman & Szpakowicz, 2007). Second, the text classification approach which creates data from the classifying of texts or sentences (Pang et al., 2002). In this study, the lexicon-based approach is followed in which the dictionaries of words are annotated with the word's emotional orientation where emotions can be expressed in textual forms using words. It has a rich library with emotional words that are the most common in social sciences.
R-software was used that offers "NRC" lexicon via the "tidy" and "syuzhet" packages, in the current study, to identify the emotional words expressed in the text 1 . Such packages include several words for emotional expressions that were collected in eight categories (joy against sadness; anger against fear; trust against disgust; and surprise against anticipation) and two sentiments (positive against negative). The eight categories are the same as those suggested by Plutchik (1980).

The sample
The aim of this study is to explore the impact of GFC on the use of emotional language in the banks' annual reports through three periods Total number of selected banks 12

The descriptive analysis
The current study seeks to explore the extent to which the language of emotions in the banks' annual reports is affected by GFC. Consequently, the study period from 2002 to 2017 (16 years totally), was divided into three periods (pre, during and post) of GFC as mentioned early in the current study. Moreover, eight emotional categories were used (trust, anticipation, surprise, fear, anger, sadness, joy and disgust). To recognize the trend of all emotional words in banks' annual reports (from 2002 to 2017), Figure 1 was prepared.
In Figure 1, the percentage of emotional words in the annual reports of some banks (AUB, DIB, MASQ, HBMO, THBK, GBK and SAAB) in the pre-GFC was 20% to 22%, while it was 23% to 25% during GFC, and in the post-GFC it amounted 22% to 24%.   It should be noted that all the graphs in Figure 1 indicate that the percentage of emotional words in banks' annual reports (from 2002-2017) represents almost 25% on average. Such a percentage is relatively large, which needs further analysis to determine the percentage and the use of each emotional category (eight categories were used in the current study), in addition to knowing whether the use of each emotional category was affected by the three different periods of GFC.
Consequently, Table 3 shows the descriptive analysis of the eight categories of emotional words of sampled banks in the study. In Table 3 Table 3 related to the period from 2002 to 2017. Figure 2 shows the distribution of eight emotional categories in banks' annual reports (from 2002-2017). In Figure 2, Boxplot reflects the distributional characteristics of emotional words in terms of median, skewness and outliers as well as the level of the emotional words. It can be noted that the emotional words of the trust category are the highest and most frequently used in the annual reports of the sampled banks. In contrast, the emotional words of the disgust category are the least commonly used in these annual reports. For example, in AUB, the trust category is ranked first with median percentage of 42%. The anticipation category is the second with median percentage of 25% and the third category is fear with median percentage of 13%. In addition, the fourth category is joy with median percentage of 11%, the fifth category is sadness with median percentage of 7%, the sixth category is anger with median percentage of 5%, the seventh category is surprise with median percentage of 3% and the eighth category is disgust with median percentage of 1%.

The distribution and trend of emotional words in banks' annual reports
This result is logical and predictable as it is known that managers are trying to gain investor confidence by using the language that generates confidence and trust in their annual reports. Therefore, the emotional words of the trust category are used extensively and consistently throughout all banks (from 2002-2017). At the same time, managers try to avoid any language style that can alienate investors such as the emotional words of the disgust, anger, sadness, and surprise categories. Such categories are the least commonly used in banks' annual reports. The anticipation category is ranked second in terms of use by managers in banks' annual reports. Moreover, the emotional words of the fear and joy categories in all banks' annual reports overall 16 years appear very close to each other and have the same behavior. These results are consistent with a number of studies, which argue that managers use language or tone in annual reports to impact different users of these reports and to send a specific message to them (see, for     The period of the current study was divided into three periods, namely pre-GFC from 2002 to 2007; during GFC from 2008 to 2012; post-GFC from 2013 to 2017. To identify if there are differences in the use of emotional words in these three periods, the Wilcoxon test was used to test the four hypotheses that were developed earlier in the current study. The Wilcoxon rank-sum test is a nonparametric alternative to the two-sample t-test that is used when the sample size is small and does not require the normal assumption (Sprent & Smeeton, 2007). With respect to H1, Table 4 shows the p-values for the one tail Wilcoxon sum-rank test. The number of significant tests for the trust category has dominated the disgust category at 1% and 5% in period 1 (pre-GFC), it is 12 out of 12 (100%). Similar results are shown in period 2 (during GFC) and in period 3 (post-GFC) it is 12 out of 12 (100%). Based on these results, the Pollyanna hypothesis is confirmed. With respect to H2, Table 5 shows the p-values for the one tail Wilcoxon sum-rank test. The number of significant tests for the joy category has dominated the sadness category at 1% and 5% in period 1 (pre-GFC), it is 5 out of 12 (42%). Similar results are shown in period 2 (during GFC) and in period 3 (post-GFC) it is 5 out of 12 (42%). Consequently, the results do not support H2, therefore, the Pollyanna hypothesis is not confirmed.
Concerning H3, Table 6 reports the p-values for the one tail Wilcoxon sum-rank test. The number of significant tests for the anticipation category has dominated the surprise category at 1% and 5% in period 1 (pre-GFC), it is 12 out of 12 (100%). The number of significant tests in period 2 (during GFC) is 11 out of 12 (92%) and in period 3 (post-GFC), it is 12 out of 12 (100%). Based on these results, the H3 is supported, therefore, the Pollyanna hypothesis is confirmed.
For H4, Table 7 reports the p-values for the one tail Wilcoxon sum-rank test. The number of significant tests for the fear category has dominated the anger category at 1% and 5% in period 1 (pre-GFC), it is 2 out of 12 (17%). Therefore, the Pollyanna hypothesis for period 1 is not confirmed. On the other hand, in period 2 the number of significant tests at 1% and 5% is 10 out of 12 (83%) and in period 3 it is 9 out of 12 (75%). Thus Pollyanna hypothesis is supported for periods 2 and 3.
The overall Wilcoxon rank-sum test for the whole sample for each period is given in Table 8, which has shown that the Pollyanna hypothesis is strongly significant for hypotheses H1, H2, H3 and H4 (during period 2 and period 3), while for H4 (in period), it is very near 0.05. It can be noted that the results in Table 8 are supported with those shown in previous tables (4, 5, 6 and 7).

CONCLUSION AND FUTURE RESEARCH
There is no doubt that GFC is one of the important and serious matters that affected the various aspects of life in the world's economic, political and financial areas which was a strong incentive for a number of researchers in different countries to study the effects of GFC in many ways. The current study is one of the first studies in the Middle East area that addresses the impact of GFC on the use of emotional language in annual reports using a sample of banks from six countries. The current study used emotion analysis of banks' annual reports through three periods (pre-GFC, during GFC, and post-GFC) based on eight categories suggested by Plutchik (1980), namely joy, sadness, anger, fear, trust, disgust, surprise, and anticipation. The results of the study showed that the percentage of emotional words in banks' annual reports (from 2002-2017) represents almost 25% on average. Details for this percentage are as follows: • the trust category accounts for 36.5%-46%; • the anticipation category is scored for 21%-23.9%; • the fear category accounts for 8.31%-15.3%; • the joy category ranges from 8.53% to 11.7%; • the anger category accounts for 4.62%-7.75%; • the sadness category is 4.88%-8.59%; • the surprise category accounts for 1.64%-2.91%; and • the disgust category is 0.81-2.43%.
Moreover, the graphs showed that three categories, namely trust, anticipation and fear, were the most affected over other emotional categories during GFC where the trust category decreased, while both the fear and anticipation categories increased. To test the four hypotheses of the current study, the Wilcoxon test was used. It was found that the trust category has dominated the disgust category in three periods, thus H1 and Pollyanna hypothesis are accepted. Similarly, H3 is accepted where the anticipation category has dominated the surprise category in all periods of the study. Consequently, Pollyanna hypothesis is confirmed. In contrast, H2 and H4 were rejected.
The study has some limitations. For example, the study was based on a sample of 12 banks from six countries in the Middle East area. This sample is relatively small. Consequently, the generalization of the study's results may be difficult. This matter can be justified by the difficulty of obtaining annual reports for 16 years for many listed banks in the stock exchanges of the six countries under study. In addition, the study used graphics to show the use of emotional words in each bank through 16 years. Consequently, a limited number of banks is required. The current study opens new horizons to study the relationship between the use of emotional language in financial reports and many key issues such as financial performance, stock prices, corporate governance, risk disclosure and others.