Corporate Disclosures in Pandemic Times. The Annual and Interim Reports Case

ABSTRACT

The paper examines if COVID-19 crisis has brought changes in companies’ approach on corporate reporting, with focus on annual reports. The research method is based on text mining techniques in order to build measures of readability and tone of uncertainty of annual reports, given the information published by the companies listed on four stock exchanges from Europe, namely the Bucharest Stock Exchange, ATHEX Stock Exchange, IBEX-35, and WIG-20 between 2017–2020. Findings emphasize, through text mining, multivariate analysis, and topic modeling, that the analyzed reports are less extensive in times of pandemic and tend to become more generic. Among firms’ financial performance metrics considered in our models, we found that there is a significant association only between annual reports textual characteristics and respectively, firm size, price earnings ratio and accruals reported. We prove as well significant stock exchange effects and industry effects. Our results show a slight decrease in annual reports readability, while the tone of uncertainty is more prominent within firms listed on less mature stock exchanges.

KEYWORDS:

• COVID-19
• financial reporting
• economic crisis
• disclosures
• textual analysis

JEL:

• G34
• H12
• M41
• M42

1. Introduction

The novel coronavirus caused a vast economic decline worldwide. This new context has seriously affected the daily businesses and operations of companies and highlighted the need for better communication and disclosure policies with shareholders and other stakeholders. In this sense, the annual report remains one of the main drivers of communication that companies have with their interested parties regarding the company’s performance and, most important, the company’s ability to continue as a going concern in an environment marked by uncertainty.

As noted by Leuz and Wysocki (2016), disclosures represent an ideal tool for reduction of information asymmetry between managers and firms’ shareholders and stakeholders. However, benefits of higher financial transparency is strongly conditioned by the nature of the disclosures and the content. Instead, under the premises of integrated or sustainability reporting frameworks, managers have started using annual report as powerful marketing tools for the firms they lead, with the scope of signaling positive effects on the market and having benefits, such as lower cost of capital (Christensen, Hail, and Leuz 2019).

In times of crisis, annual reports become even more important, as they represent the main tool for managers to communicate financial results to shareholders and stakeholders as well. However, the focus is oriented toward risk management disclosure and presentation of macroeconomic context that might affect forward-looking information. Those directions lead to a change in the tone of annual reports, which is expected to be more negative, showing that accounting estimated reported is highly affected by external factors, and less by managers’ ability (Oskouei and Sureshjani 2021).

Furthermore, managers could choose to prepare obfuscated annual reports, to hide as much as possible potential future evolutions of financial performance, to avoid transmitting any negative signal to the market, as during COVID-19 pandemic period shares liquidity has become significantly volatile (Hatmanu and Cautisanu 2021). Moreover, capital markets’ systemic impact of COVID-19 becomes globally contagious, especially under uncertain conditions, when news attention negative impact on financial markets is higher than investors’ rational expectation (Engelhardt et al. 2020).

Our aim in this paper is to bring some insights on the changes in the annual reports characteristics, determined by the current COVID-19 pandemic crisis. In this context, the focus of our paper is the analysis of firm-specific financial determinants on annual reports readability and tone of uncertainty of entities listed on four stock exchanges, including Bucharest Stock Exchange (BSE) and ATHEX, two stock exchanges that recorded the highest growth in the world, considering the evolution of stock markets in 87 countries with available data (Wall Street 2019). According to the ranking published by FTSE Russell in September 2020, Romania and Greece are included in the emerging markets category. In order to compare the data, we included in the study the stock exchanges from Spain and Poland, respectively the entities listed on IBEX-35 and WIG −20, classified by FTSE Russell among the developed stock markets (FTSE 2020).

Our article comes with additional insights within the literature. It adds empirical evidence on the area of analysis of readability of annual reports and sentiment analysis, for annual reports made public by firms listed on European capital markets. To our knowledge, only Lang and Stice-Lawrence (2015) analyzed another sample of European companies. It also fills in the gap on the literature by analysis annual reports, as most of the literature is limited to the 10-K disclosure analysis published by American companies. The article is the first to analyze the effect of COVID-19 on annual reports. Only Wang and Xing (2020) analyzed the effect of COVID-19 on corporate conference calls and management announcements. Nonetheless, as of our knowledge, it is the first study that addresses the problem of topic modeling on annual reports disclosed by firms listed in the four European capital markets considered in the study, to arise issues related to the risk of generic disclosures, in the context of actual COVID-19 pandemic.

This paper is structured as follows: Section 2 presents the literature review and the hypothesis development, testing the ways in which corporate disclosures are influenced by the novel coronavirus context; Section 3 presents the research methodology, Section 4 discusses the results obtained, and Section 5 summarizes the main findings, conclusions, and avenues for future research directions.

2. Literature Review

The pandemic and its effects represent a topic of great interest because many view the COVID-19 outbreak as an unprecedented event that the world has witnessed since the inception of the stock markets. COVID-19 affected the activity of companies from all business sectors, being considered an event similar to the financial crisis that began in 2008 (Wang and Xing 2020). Much has been written about the crisis, authors providing insights and perspectives, most agreeing that in uncertain times some accounting techniques used in financial reporting led to its deepening. Hence, in this novel coronavirus crisis it is interesting to study how firm’s financial reports are elaborated and released into the market, to mitigate the negative impact of a pandemic on firm’s performance before shareholders and stakeholders.

Overall, the literature shows that over time, the length of annual reports has increased (Dyer, Lang, and Stice-Laurence 2017; Lang and Stice-Lawrence 2015). Those premises lead to changes in the length of the annual reports, because of complexity of models used for accounting estimates (Lim, Chalmers, and Hanlon 2018).

In the current context, telling the story of COVID-19 through annual reports implies more professional judgments, that lead to lengthier annual reports, aimed to describe properly postulates, assumptions, and methods used in accounting estimation and financial reporting. Gould and Arnold (2020) stated that some companies tend to report financial effects of the pandemic through interim financial statements first, which involves a greater use of accounting estimates. Therefore, we believe that these new elements that need to be highlighted in the annual reports, given the current pandemic context, will require more robust and detailed corporate disclosures and the first hypothesis that will be tested will be H₁: in the context of COVID-19 pandemic restrictions, firms’ financial characteristics determine significant changes in the length of annual reports.

However, as noted by Dyer, Lang, and Stice-Laurence (2017), the literature underlines a trend of increasing length of annual reports, in parallel with deterioration of readability and increase of tone of ambiguity.

Researchers highlight that performing entities disclose readable reports in order to impress (Li 2008). Moreover, there are studies that show a positive correlation between managerial ability and annual reports readability, especially in case of firms reporting positive profitability (Hasan 2020). There are also studies that show readability and complexity of annual reports are significantly driven by the accounting standards (Dyer, Lang, and Stice-Laurence 2017). The increase in innate component (common complexity) of reporting complexity (readability) is mainly driven by accounting regulation (Dyer, Lang, and Stice-Laurence 2017; Efretuei 2020), while the increase of obfuscation (uncommon complexity) is more associated to management choice that depends mainly on capital markets incentives (Li 2008; Loughran and McDonald 2013; Miller 2010) and investors disclosure choices. Through more complex and less readable disclosures, managers tend to hide poor performance, by amplifying annual reports obfuscation (Li 2008; Lo, Ramos, and Rogo 2017), with impact on firms’ cost of capital (Ertugrul et al. 2017).

Therefore, we believe that the readability of the annual and interim reports is influenced by the COVID-19 pandemic and the second hypothesis that will be tested will be H₂: in the context of COVID-19 pandemic restrictions, firms’ financial characteristics determine significant changes in the readability of annual reports.

Given the importance of making accurate disclosures, reporting in times of uncertainty represents a major challenge. Considering that COVID-19 pandemic negatively impacted companies around the world, the approach by which the “bad news” are disclosed into the market through annual and interim reports represents a topic of great interest. The way the information is presented and structured is essential to ensure significant impact on investors’ decision, especially through the lens of impression managements techniques, as some strategies are aimed to obfuscate bad news, while most of them are designed to emphasize good news (Merkl-Davies and Brennan 2017).

The tone of the discourse used when preparing annual reports is also important, as behavioral theories show that negative words have a higher impact compared with the positive words (Tetlock, Saar-Tsechansky, and Macskassy 2008). Consequently, the positive bag-of-words are preferred by managers when preparing annual reports. However, there is little evidence on studying the tone of uncertainty of annual reports in the literature, as researchers focus more on either the use of dichotomous positive versus negative predefined lists of words (Loughran and McDonalld 2016), or the use of ambiguous words (Dyer, Lang, and Stice-Laurence 2017; Ertugrul et al. 2017).

Loughran and McDonald (2011) have shown that managers present information on a more positive tone in the annual reports, which is more likely associated with negative stock returns. Davis et al. (2015) noted as well that managers have incentives to set-up their optimistic or pessimistic tone on corporate communication of firms’ results, especially when these information are more sensitive to stock prices, reflecting that the tone of disclosures became negative only after the financial crisis from 2008. However, Wang and Xing (2020) have shown, referring to annual reports related to the COVID-19 pandemic period, that low level of uncertainty tone in the earnings announcements has been made by the management.

Therefore, we believe that now, more than ever, in this uncertain times generated by the COVID-19 pandemic, managers should avoid negative or ambiguous words when reporting the consequences of the global pandemic through annual statements and the third hypothesis that will be tested will be H₃: the novel coronavirus created a context marked by uncertainty that has a significant influence on disclosure tone of the annual reports.

Efretuei (2020) has shown that the tone of uncertainty of annual reports is negatively related to the corresponding readability, showing that managers prefer to amplify the obfuscation information, by using more words related to uncertainty and ambiguity, describing an increasing trend of this negative relation overtime. In the actual context of COVID-19 pandemic crisis, the issue of information asymmetry, through its components of moral hazard and adverse selection that arise, is highly condoned by industry-specific and domestic institutional framework as well.

Over time, economies have become increasingly interconnected and interdependent, which has exacerbated the crisis. In this context, it is vital that the information transmitted by companies into the market to be transparent and real so that the business environment could mitigate the negative impact of a pandemic on firm’s performance.

3. Methodology Research

3.1. Sample Data

Our research methodology is similar to Lang and Stice-Lawrence (2015). The focus of the paper is the analysis of firm-specific determinants in annual and interim reports readability and keywords frequency, considering the period between 2017 and the first quarter of 2020, as depicted in Figure 1.

Figure 1. Sample distribution per sector and per nature of report.

Source: own projection.

In order to emphasize local specifics of institutional and cultural framework, we used a sample that consists of firms listed on different European capital markets, respectively the Romanian capital market (BSE), Greek capital market (ATHEX), the Spanish capital market (IBEX-35), and Polish capital market (WIG-20), as depicted in Table 1.

Table 1. Sample distribution on capital markets and years

Capital market	ATHEX	BSE	IBEX	WIG	Cluster period
Firms included on sample
2017	18	78	30	18	Before COVID-19 pandemic
2018	19	77	29	18
2019	16	78	12	13	During COVID-19 pandemic
2020	16	77	12	13
Invalid format of reports	23	2	41	22
Total reports analyzed	69	310	83	62

Source: Own projection.

Annual reports and interim reports were downloaded manually from firm’s websites, or capital markets websites where they are listed. Overall, we included in our analysis 524 observations, controlling for industry effects and capital markets effects as well. The data are split into two main groups, namely data related to period before COVID-19 crisis, concerning period 2017 to 2018, and respectively data related to period during COVID-19 crisis, concerning period 2019–2020.

3.2. Variables Definition

The data used in econometric models estimated in our study consist of some key financial performance indicators and several characteristics of the annual reports, as described in Table 2, determined based on annual reports and output generated running text mining procedures.

Table 2. Variables definition

Category	Name	Description
Dependent variables	Length	Similar to Lang and Stice-Lawrence (2015) and Dyer, Lang, and Stice-Laurence (2017) papers, we consider the number of pages per each annual report included in our analysis as a measure of disclosure transparency; the measure reflects the logarithm of the number of pages of each annual report analyzed.
	Readability	As noted by Dyer, Lang, and Stice-Laurence (2017), the measure of readability, together with the measure of complexity of financial disclosures, represents a mean of analyzing annual reports understandability; the score is determined as the average number of words per sentence, per each annual report.
	Uncertainty	Similar to Wang and Xing (2020) paper, using the word list developed by Loughran and McDonald (2011), we determined weight of words related to uncertainty, on the number of total tokens (with number of characters between 4 and 25) extracted from analyzed annual reports and steamed; there are also alternative solutions to measure sentiment analysis on documents, as emerging computational text mining models and technologies give the researcher the possibility to get more detailed information, relevant for identifying the dimensions of each bag-of-words that offer currently a narrow perspective on the context the words are used in the text.
Explanatory variables	ROA	Similar to Lang and Stice-Lawrence (2015) and Wang and Xing (2020), we consider in our analysis this performance measure, determined by dividing net income to total assets, based on information disclosed by annual or interim reports.
	Leverage	In line with Ertugrul et al. (2017) we consider firms’ cost of capital is significantly influenced by the effect of readability of annual reports; this measure is defined as the ratio between debts and equity reported by firms on their annual reports;
	Accruals	Francis et al. (2005) have underlined the positive relation between firms’ earnings quality and disclosures quality, including indirect relation with disclosures readability; additionally, Jaafar and Hussainey (2012) underlined the substitute relation between those two elements, reason why we test if accruals amplitude can influence annual reports length and readability;
	PER	It expresses the price earnings ratio as of year-end reporting; as noted by Miller (2010), the level of financial reports readability is significantly associated with stocks return volatility and indirectly with firms’ market liquidity.
	Size	Reflects the size of the firms, using natural logarithm of the total assets as an indicator; the more complex business models are, including the business strategy, the higher is the complexity of financial disclosure and the lower financial disclosures readability (Lim, Chalmers, and Hanlon 2018);
	Growth	As noted by Loughran and McDonalld (2016), managers prefer to set-up their optimistic or pessimistic tone on the annual reports, depending on the nature of the results; this measure, defined by the rate of growth in total assets, shows the effect of restructuration in firms’ operations because of COVID-19 pandemic restrictions and the potential of future growth in revenue as well, seen as positive results.
	Corporate governance score	There is a unanimous position on the literature about the role of corporate governance on monitoring for the quality of financial reporting output. Through the composite corporate governance score calculated extracted from Refinitiv database, we check for implications of effective governance mechanisms implemented by firms, on corporate reporting, as a higher governance index is expected to lead to higher quality of annual reports (Leuz and Wysocki 2016). Board of directors with high experience and expertise, efficient function of internal audit, or effective internal controls are just few basic pillars for accurate, clear, and relevant corporate reports.
	Analysts coverage	The number of analysts following firms analyzed is extracted from Refinitiv database. If firms’ corporate reports are analyzed by a higher number of analysts, managers are limited on their opportunities to smooth financial performance, leading to higher quality of corporate reports (Leuz and Wysocki 2016). Additionally, to avoid any misunderstanding when analysts decide on their forecasts, managers become more careful on the understandability of annual reports. Otherwise, they are subject to potential negative signals transmitted on the market.
	Fixed effects: - industry	Variables used to reduce the shock of unobserved features of the industry and institutional framework of stock markets, as those dimensions are significant on designing and preparing financial disclosures (Leuz and Wysocki 2016).
	- stock market

Source: Own projection.

3.3. Construction of the Linear Regression Models

The regression model used to test the hypotheses related to annual reports characteristics has the following configuration:

(1) $Disclosur{e_{characteristics}}_{it}={\alpha }_0+{\sum }_{k=1}^6{\alpha }_k\cdot Financia{l_{indicator}}_{it}+{\epsilon }_{it}$(1)

where $Financia{l_{indicator}}_{it}$ represents the financial measures of firms’ characteristics, respectively ROA, Leverage, Accruals and Size, Growth and PER as defined in Table 2, while $Disclosur{e_{characteristics}}_{it}$ reflects each of the dependent variables defined in Table 2, respectively, Length, Readability and Uncertainty.

Estimation of the econometric models is made using the OLS method. However, in order to control for potential implications of probability distribution of independent variables, we estimate the econometric models looking for firm financial drivers using the quantile regression method as well, with focus on 1^st quantile estimates, 2^nd quantile estimates, and respectively 3^rd quantile estimates (Wooldridge 2010).

For reliability purpose, we proceed to results triangulation, by using an alternative measure of readability of annual reports, considering the Flesch reading ease score, which is determined based on formula $FRE=206.835-1.015\cdot \frac{totalwords}{totalsentences}-84.6\cdot \frac{totalsyllables}{totalwords}$. A higher $FRE$ score means better readability of annual reports. Therefore, this alternative measure of document readability is the opposite of the main readability measure of our study, which consists of number of words per sentence.

To illustrate specifics of each area of activity on financial disclosures characteristics, we control our model for fixed effects as well, as described in econometric model below:

$Disclosur{e_{characteristics}}_{it}={\beta }_0+{\sum }_{k=1}^6{\alpha }_k\cdot Financia{l_{indicator}}_{it}$

$+{\sum }_{l=1}^5{\alpha }_{l+6}\cdot Corporategovernanc{e_{indicator}}_{it}$

(2) $+{\sum }_{m=1}^8{\alpha }_{m+12}\cdot Industryeffect{s_l}_t$(2)

where $Industryeffect{s_l}_t$ reflects the fixed/random specific effects of each area of activity analyzed. Additionally, we look for impact of corporate governance metrics that describe better firms’ institutional particularities designed to look for the quality of financial reports.

3.4. Entropy Analysis of Annual Reports

In order to measure how uniform extracted topics are addressed in annual reports and if there is a pattern related to industry specific, or capital markets specific, we proceed to Latent Dirichlet Analysis (LDA), in order to extract distinct topics addressed in a text. After performing LDA for different number of topics extracted, we have decided on the extraction of 45 topics from the corpus analyzed, as the value of perplexity score tends to increase less (Hofmann and Klinkenberg 2014).

Each topic extracted from the corpus is characterized by a set of performance metrics. The essential metric considered in our study is the distribution of topics, based on topics coherence obtained for each annual report. The coherence of a topic is the measure whether the words in a topic tend to co-occur together. The larger the value topic coherence the lower the indication that words co-occur in different topics simultaneously. Once determined each topic coherence per each annual report, we calculate the Shannon entropy score, using the relation below:

(3) $Entrop{y_{annualreport}}_{it}=-{\sum }_{k=1}^n{p}_k\left(x\right)\cdot log{p}_k\left(x\right)$(3)

where $p_k\left(x\right)$ is the proportion (coherence metric) from the document that topic $k$ is addressed in annual report disclosed by firm $i$ for year $t$. The interpretation of this score is that the higher the entropy of an annual report, the more uniformly are addressed in the annual report all topics extracted from the corpus running the LDA. Instead, the lower the score of entropy is, the more attention is given in the annual report to only few topics extracted from the corpus analyzed.

Different extensions of the LDA procedure are used as well to directly link the stock prices with topics addressed in the corporate communications, so that stock returns or firm profitability can be explained easily by related topics extracted from the corpus analyzed (Li et al. 2020).

Our approach is similar with Dyer, Lang, and Stice-Laurence (2017) that have explained the length of the first three topics by firms’ financial characteristics and annual reports textual characteristic as well. However, in this study we do not focus on explaining length of topics extracted from annual reports, but rather on justifying annual reports readability by the diversity of topics addressed.

4. Results and Discussion

4.1. Descriptive Statistics

In Table 3 we summarize the descriptive statistics for the sample analyzed, considering two separate panels of data, respectively. Panel A that consists of data analyzed related to period 2017 to 2018 and Panel B including data analyzed related to period 2019 to 2020. We observe that there is spread in the practice of financial disclosures, as the interval of variation for either the number of pages (length), or the level of understanding of financial disclosures (readability) is high, especially in case of readability score.

Table 3. Descriptive statistics

	Panel A: before COVID-19 pandemic				Panel B: during COVID-19 pandemic
	Mean	Std. Dev.	VIF	Normality test		Mean	Std. Dev.	VIF	Normality test
				Stat.	Sig.				Stat.	Sig.
Disclosures characteristics
Length	5.131	0.746	-	0.081	.000^a	4.581	0.815	-	0.000	.200^a,b
Readability	13.31	5.195		0.137	.000^a	12.77	5.162		0.132	.000^a
Uncertainty	0.072	0.060		0.249	.000^a	0.084	0.058		0.227	.000^a
Firm characteristics
Size	19.96	3.090	1.102	0.130	.000^a	19.58	2.903	1.048	0.118	.000^a
Leverage	2.843	20.77	1.089	0.143	.000^a	1.859	7.288	1.112	0.199	.000^a
ROA	1.282	3.007	1.030	0.239	.000^a	0.020	0.082	2.116	0.145	.000^a
Growth	0.032	0.132	1.033	0.271	.000^a	0.057	0.214	1.106	0.189	.000^a
PER	1.929	1.396	1.116	0.190	.000^a	18.62	2.821	1.057	0.231	.000^a
Accruals	−0.313	4.921	1.016	0.485	.000^a	−0.006	0.098	1.938	0.173	.000^a

^aLilliefors Significance Correction.

^bThis is a lower bound of the true significance.

Source: Authors’ calculation.

Instead, the standard deviation is relatively small in case of the number of pages of financial disclosures, which means that there is only a slight spread on the practice of financial reporting along capital markets or different areas of activity. In average, the annual reports have about $e^{5.131}\approx 169$ pages, in case of annual reports related to Panel A. On the other side, annual reports related to Panel B record significant decrease to an average of 98 pages. Similar but slight changes are observed in case of reports readability, as the average number of 13.31 words per sentence corresponding to Panel A annual reports decrease to the number 12.77 words per sentence, in case of Panel B annual reports, but with higher deviation between reports analyzed.

The results in Table 3 show a small percentage of words reflecting the sentiment of uncertainty, as the average on Panel A related annual reports is about 7.2%, while for Panel B annual reports is slightly higher, respectively 8.4%, which show managers avoid, as much as possible, words that lead to unclear or uncertain messages transmitted to shareholders and stakeholders as well. However, the slight increase highlights the systemic effect of COVID-19 pandemic on business operations and financial disclosures as well.

Related to firms’ characteristics summarized in Table 3, notable results for our purpose concern the changes determined by COVID-19 pandemic. The statistics show that firms suffer from negative effects of COVID-19 pandemic, through a decrease of profitability, a decrease of financial leverage and a decrease of firms’ market value. On the other side, the high standard deviation of the growth variable shows that some firms are less affected than other, either because of their specific of activity that has benefit from COVID-19 pandemic, such as IT sector (Efretuei 2020), or because of business models short-term reorientation that implied also a slight increase on firms’ assets, leading to changes on production capacity. Nonetheless, we observe a significant decrease of accruals once the COVID-19 pandemic crisis has started, suggesting us less earnings management.

Our results are similar with Jiang, Pittman, and Saffar (2019) who show that in times of crisis, described by high uncertainty on policy making, firms’ disclosures become lengthier, less readable and with a more prominent uncertainty tone on discourse, in order to avoid compliance costs or litigation costs (Christensen, Hail, and Leuz 2019).

In Figure 2, we observe slight changes, especially related to the length of the annual reports that seem to be lower, or to the readability of the annual reports that seem to deteriorate in time, especially in case of IBEX-35 listed entities. It seems that the more mature the capital markets are, the higher the readability of annual reports is, in parallel with a lower uncertainty tone of discourse. The highest score of annual reports readability is found within entities listed on IBEX-35 capital market, considered as a developed capital market, with a high level of value and investor segmentation. The highest tone of uncertainty is observed within entities listed on BSE capital market, which is more related to cultural dimension of the corporate reporting, rather than capital markets incentives.

Figure 2. Marginal means conditioned by COVID-19 pandemic effects.

Source: authors’ projections.

Additionally, in Figure 2 we observe that higher coherence of topics seems to appear in case of annual reported after COVID-19 pandemic, which show topics addressed in annual reports are equally weighted, suggesting a more general approach considered, with less focus on specific topics, similar with the concern Dyer, Lang, and Stice-Laurence (2017) have highlighted. Overall, we observe higher entropy of firms’ annual reports operating in industry area, IT, commerce, and utilities.

4.2. Correlation Analysis

In Table 4 we provide the correlations between disclosure characteristics and firms’ financial performances. The results show unanimously that firms size represents essential factor for the length and the readability of annual reports, as business strategy related to more complex business models asks for more complex annual reports (Lim, Chalmers, and Hanlon 2018). Instead, the correlation between firm size and the annual reports tone of uncertainty is significantly negative, showing that annual reports disclosed by big firms show lower tone of uncertainty.

Table 4. Pearson correlation matrix

	Variable	Length	Readability	Uncertainty
Panel A: before COVID-19 pandemic	Size	0.607**	0.576**	−0.647**
	Lev.	−0.067	0.026	−0.106
	ROA	0.472**	0.323**	−0.411**
	Accruals	−0.086	0.065	−0.097
	Growth	−0.031	0.038	−0.030
	PER	0.097	0.181**	−0.270**
Panel B: during COVID-19 pandemic	Size	0.372**	0.523**	−0.552**
	Lev.	0.052	0.179**	−0.136*
	ROA	0.102	0.142*	0.018
	Accruals	−0.028	0.042	0.056
	Growth	0.081	0.120	−0.127
	PER	0.185**	0.086	0.029

Source: Authors’ calculation.

We observe a significant positive correlation as well between readability of annual reports and firms’ leverage, and a significant negative correlation between annual reports tone of uncertainty and firms’ leverage, related to period during COVID-19 pandemics. Those correlations suggest to us that, in times of crisis, annual reports disclosed by firms with higher debt focus more on less uncertain information disclosed and try to transmit a more comprehensive message to the creditors.

We highlight in Table 4 a slight positive correlation between firms’ profitability and the length of annual reports and their readability, for the period before COVID-19 pandemic. Firms’ profitability is negatively correlated with annual reports tone of uncertainty. Like Loughran and McDonald (2011), our results suggest that managers prefer to focus their annual reports on good news, rather than bad news. However, readability of annual reports published during the COVID-19 pandemic is less influenced by firms’ profitability.

Instead, we see a low statistically significant positive correlation between price earnings ratio and the length of the annual reports, only in case of period related to COVID-19 pandemic, which suggest that managers are slightly sensitive to capital markets in times of crisis.

4.3. Assessing Firm-level Determinants of Financial Disclosures

In Table 5 we provide evidence on the association between measures of annual reports and respectively firms’ financial performance indicators. We addressed the existence of heteroscedasticity in our models, by performing the Modified Breusch-Pagan test. Model 1 estimates are not heteroscedastic, as their $pvalue$ is less than the threshold of 1%, reason why we estimate the model using the White-Huber standard errors regression procedure, to ensure efficient regression coefficients estimates, concerning standard errors (Wooldridge 2010).

Table 5. Multivariate analysis on financial drivers of annual reports textual measures

		Panel A before COVID-19 pandemic			Panel B after COVID-19 pandemic
Dependent variable		Length	Readability	Uncertainty	Length	Readability	Uncertainty
Models		1	2	3	4	5	6
Constant		3.059*	−9.38*	0.361*	3.473*	−7.559*	0.343*
		(0.306)	(1.969)	(0.020)	(0.402)	(2.300)	(0.023)
ROA		0.028	1.570	0.001	1.160	13.12*	−0.032
		(0.117)	(1.658)	(0.017)	(0.858)	(4.914)	(0.050)
Leverage		0.039	0.050	−0.002	−0.078	0.693	−0.005
		(0.069)	(0.599)	(0.006)	(0.084)	(0.483)	(0.005)
Accruals		−0.006	0.121**	−0.002*	−0.838	−1.865	0.009
		(0.006)	(0.050)	(0.001)	(0.715)	(4.098)	(0.042)
Growth		−0.028	−0.898	0.009	0.049	−0.565	−0.025***
		(0.226)	(2.013)	(0.020)	(0.253)	(1.449)	(0.015)
PER		−0.014	0.175	−0.004***	0.042	−0.035	0.004**
		(0.033)	(0.208)	(0.002)	(0.035)	(0.198)	(0.002)
Size		0.101*	1.11*	−0.014*	0.052*	1.006*	−0.013*
		(0.015)	(0.103)	(0.001)	(0.020)	(0.116)	(0.001)
Fixed effects		Y	Y	Y	N	N	Y
Model validation
Sample size		282	282	282	234	227	227
R^2 adjusted		0.318	0.424	0.519	0.133	0.353	0.441
F stat		9.989	15.21	21.83	3.577	10.19	14.28
P		0.000	0.000	0.000	0.000	0.000	0.000
Modified BP Test - heteroscedasticity	Stat.	11.88	2.984	1.143	2.762	0.113	0.303
	p	0.001	0.084	0.285	0.097	0.736	0.582
Hausman test - fixed effects	Stat.	13.20	23.94	38.61	10.87	5.186	58.06
	p	0.040	0.001	0.000	0.092	0.520	0.000
Breusch-Pagan test - random effects	Stat.	3.596	1.724	19.06	1.313	0.110	16.154
	p	0.058	0.189	0.000	0.252	0.741	0.000

* Significant for 1% significance level;

** Significant for 5% significance level;

Source: Authors calculations

All models estimated in Table 5 are statistically significant, as their level of significance ($Ftestpvalue.$) is lower than the acceptable level of 1%. Instead, we observe a relatively low $R^2$ adjusted, varying between 13.3% in case of model 4 to 51.9% valid for model 3.

Based on Haussmann test results, summarized in Table 5, we consider the models with industry fixed effects more relevant for our analysis in case of our models, except for model 4 and model 5, as their $Sig.<0.05$ (Lee et al., Lee 2019). For model 4 and model 5 we have performed the Lagrange Multiplier (LM) test to check if random effects regression model is more relevant than the fixed effects or the pooled regression model. Based on statistics provided in Table 5, we observe that the models look for random effects ($Sig.>5\mathrm{\%}$), which shows that the specific of each firms’ business model and management related corporate reporting policy approach is more relevant than the industry-specific effect (Lee 2019).

Firm size effect is significant for the readability, the length, and the uncertainty tone of the annual reports, no matter the panel data analyzed. The results show that the more complex is the firm’s business model, the less readable the annual reports will be. Our results are similar with Dempsey et al. (2012) results, who have confirmed a positive association as well between firm size and readability of annual reports.

The results also highlight that the higher the firm size, the lower the uncertainty tone of the annual report will be, suggesting that issuers illustrate in the annual reports less information about the uncertain economic perspectives. The COVID-19 pandemic has affected annual reports, as our results show slight changes in the readability and tone of uncertainty of annual reports, reflecting an increase in readability and a less pronounced tone of uncertainty in case of larger firms.

Managers’ sensitivity to capital markets is visible according to our results only on the tone of uncertainty of annual reports, described by the PER (price earnings ratio) coefficients, which differs on the panels analyzed. While during the period of COVID-19 pandemic, higher stock price earnings ratio determines more pronounced tone of uncertainty of annual reports ($Coef=0.004,Sig.<1\mathrm{\%}$), during the period before the COVID-19 pandemic, higher PER determines more positive tone of annual reports ($Coef=-0.004,Sig.<1\mathrm{\%}$). The results show that managers would like to emphasize better the positive financial results in conditions of crisis, showing the results are obtained in terms of significant negative constraints, such as the negative economic effects of COVID-19 pandemic.

Our results show as well a negative association between the tone of uncertainty and the level of accruals reported ($Coef=-0.002,Sig.<1\mathrm{\%}$), for the period prior to COVID-19 pandemic. This result reveals that the lower the accruals are reported, the higher is the uncertainty tone of annual report, as there is less evidence of earnings management, because managers might explain poor financial performance referring to negative shocks on economic environment. However, the marginal effect on the tone of uncertainty on annual reports is relatively small, suggesting that managers prefer to refer less to economic environment uncertainty on annual reports, as negative information has a higher impact compared with the positive words (Tetlock, Saar-Tsechansky, and Macskassy 2008).

Our results show as well, a positive association between the level of accruals and the readability of annual reports ($Coef=0.121,Sig.<5\mathrm{\%}$), valid for period before COVID-19 pandemic. In our study, higher readability score indicates less readable reports according to our readability score definition. The results we obtained suggest that, managers reporting higher level of accruals present the information through less readable reports. Those results confirm again the relation of substitution between accruals quality and disclosures quality, as lower accruals suggest higher earnings quality, positively related with higher readability of annual reports (Jaafar and Hussainey 2012).

The results describe as well a positive association between firms’ profitability and the length of annual reports ($Coef=13.21,Sig.<1\mathrm{\%}$), but only in case of reports published during COVID-19 pandemic. Our results highlight that higher profitability means less readable annual reports. The results suggest that managers might use less readable discourse in preparing annual reports, to hide information about earnings management that has led to higher firm’s profitability, similar to Davis et al. (2015), Lang and Stice-Lawrence (2015).

Nonetheless, for the period during COVID-19 pandemic, we confirm a negative statistically significant association between the tone of annual reports and the rate of growth in sales ($Coef=-0.025,Sig.<10\mathrm{\%}$). This result provides us with evidence that the higher the level of growth in sales, the lower the tone of uncertainty in annual reports. As noted by Davis et al. (2015), managers have incentives to set-up their optimistic or pessimistic tone on corporate communication of firms’ results, especially when this information is more sensitive to stock prices, such as the increase in sales.

4.4. Robustness Analysis Controlling for Distribution Values Effect

In Table 6 we provide summary statistics of the quantile regression estimation models, for the period before the ongoing COVID-19 pandemic. The results confirm once again that there is a positive association between annual reports readability and firm size. The higher the firms size, the lower the annual reports readability. The result implies that firms of larger size involve more complex business models which are hardly to be reflected in sufficiently readable disclosures.

Table 6. Quantile analysis on financial drivers for period before COVID-19 pandemic

	1^st	2^nd	3rd	1st	2nd	3rd
Dependent variable	Readability			Uncertainty
Models	7	8	9	10	11	12
Constant	−10.51*	−9.177*	−3.520*	0.309*	0.347*	0.329*
	(1.440)	(1.347)	(1.850)	(0.027)	(0.016)	(0.016)
ROA	−0.320	0.836	1.975	0.020**	0.014	0.006
	(0.578)	(6.597)	(3.569)	(0.008)	(0.009)	(0.007)
Leverage	−0.155	−0.133	−0.761	0.008	0.003	0.000
	(0.285)	(0.474)	(0.584)	(0.005)	(0.005)	(0.004)
Accruals	0.057*	0.088*	0.127*	−0.002*	−0.002*	−0.001*
	(0.018)	(0.022)	(0.019)	(0.000)	(0.000)	(0.000)
Growth	−0.014	−1.282	2.080	−0.007	0.025	0.009
	(1.048)	(1.719)	(2.858)	(0.043)	(0.016)	(0.013)
PER	−0.116	0.012	0.138	−0.008*	−0.001	0.000
	(0.142)	(0.175)	(0.294)	(0.003)	(0.002)	(0.002)
Size	1.091*	1.092*	0.923*	−0.012*	−0.014*	−0.012*
	(0.074)	(0.069)	(0.090)	(0.001)	(0.001)	(0.001)
Model validation
R^2 adjusted	0.248	0.359	0.284	0.242	0.355	0.119
Sparsity	7.468	7.350	10.45	0.118	0.090	0.106
Quasi-LR statistic	121.1	208.5	127.9	101.1	211.4	54.91
p	0.000	0.000	0.000	0.000	0.000	0.000

Source: Authors’ calculations.

We obtain robust results as well for the association between the tone of uncertainty of annual reports and firm size. These results show that firms of larger size do not focus on emphasizing the uncertainty of economic environment when presenting annual results, probably because of their higher power of negotiation with the creditors and the other stakeholders.

We get robust results as well related to the association between readability of annual reports and the level of firms’ reported accruals, which show that the higher the accruals, the less readable annual reports are. We observe that estimated regression coefficients increase from one percentile to another ($0.057<0.088<0.127$). The results show that firms of smaller size tend to be more precise on reporting, in order to explain better accounting estimates reported and avoid any future litigation cost, or capital market penalization.

Robust results are obtained also concerning the association between the uncertainty tone of annual reports and the level of accruals, showing higher level of accruals determines lower tone of uncertainty of annual reports, with the aim to transmit positive signal to investors, that should not warn investors of inaccurate earnings quality. In association with the results related to readability of annual reports, we observe that higher level of accruals determine managers to prepare obfuscated annual reports, but in a more positive tone that would likely determine less concerns for investors. Otherwise, higher investors’ concerns would determine them to do further investigation and put additional pressure on firms’ management team.

In Table 7 we provide summary statistics of the quantile regression estimation models, for the period before current COVID-19 pandemic started. The results look robust, compared with the estimates obtained using OLS method, but only related to firm size positive marginal effect on the readability of annual reports and firm size negative effect on the tone of uncertainty of annual reports.

Table 7. Quantile analysis on financial drivers for period after COVID-19 pandemic

	1st	2^nd	3rd	1st	2nd	3^rd
Dependent variable	Readability			Uncertainty
Models	13	14	15	16	17	18
Constant	−11.64*	−8.697*	−3.411	0.314*	0.322*	0.179*
	(2.354)	(1.867)	(2.671)	(0.017)	(0.017)	(0.063)
ROA	−3.266	−0.778	1.740	0.068	0.028	0.003
	(4.162)	(4.799)	(9.419)	(0.084)	(0.051)	(0.044)
Leverage	0.489**	0.123	−0.512	0.000	0.004	−0.001
	(0.274)	(0.331)	(0.360)	(0.006)	(0.004)	(0.004)
Accruals	6.764**	2.431	3.979	−0.046	0.025	0.010
	(3.507)	(3.695)	(6.631)	(0.065)	(0.043)	(0.039)
Growth	0.768	1.009	−0.744	−0.046*	−0.001	−0.017
	(1.879)	(1.562)	(2.014)	(0.015)	(0.018)	(0.019)
PER	0.034	−0.018	0.354	0.001	0.005**	0.006***
	(0.196)	(0.227)	(0.409)	(0.003)	(0.002)	(0.003)
Size	1.091*	1.055*	0.932*	−0.013*	−0.013*	−0.003
	(0.118)	(0.094)	(0.124)	(0.001)	(0.001)	(0.004)
Model validation
R^2 adjusted	0.165	0.236	0.180	0.233	0.222	0.011
Sparsity	9.373	9.072	16.476	0.119	0.086	0.106
Quasi-LR statistic	62.61	101.0	51.01	95.14	111.6	12.25
P	0.000	0.000	0.000	0.000	0.000	0.057

Source: Authors’ calculations.

The results from quantile regression show robustness for the marginal effect of firm size on readability of annual reports, no matter the percentile estimate. Instead, the negative impact of firm size on the tone of uncertainty of annual reports is statistically significant only for 1^st and 2^nd percentile, suggesting that only in case of annual reports with less pessimistic tone the firm size matters.

4.5. Robustness Analysis Controlling for Corporate Governance Implications

The main purpose of this analysis is to check if the entropy score of annual reports improves our regression models $R^{2}adjusted$, as annual reports readability is not influenced only by managers’ preferences reflected in firms’ financials, but also depends on the managers ability to present a clear, concise and coherent message with focus only on the key aspects of the business model (Hasan 2020).

In Table 8 we provide the statistics for models estimation, triangulating our results by using an alternative for readability score, respectively the Flesch readability ease score, presenting in parallel results for both measures of readability of annual reports. Additionally, we control for annual reports entropy and firms’ corporate governance framework as well.

Table 8. Control variables effect on annual reports readability score

		Panel A:		Panel B:
		Before COVID-19 pandemic		During COVID-19 pandemic
Dependent variable		Flesch score	Readability	Flesch score	Readability
Models		(19)	(20)	(21)	(22)
Constant		40.59*	5.431	33.07*	−3.64
		(6.773)	(3.436)	(8.93)	(4.018)
Entropy		−0.921	−2.807*	1.224	0.125
		(0.986)	(0.500)	(1.346)	(0.606)
Leverage		0.474	0.242	0.169	0.278
		(1.121)	(0.569)	(1.043)	(0.469)
Accruals		0.111	0.128*	7.559	4.838
		(0.094)	(0.048)	(6.814)	(3.066)
PER		0.357	0.140	0.955**	−0.026
		(0.396)	(0.201)	(0.441)	(0.199)
Size		0.071	0.812*	−0.064	0.777*
		(0.268)	(0.136)	(0.371)	(0.167)
Corporate governance score		−0.085*	0.022	−0.164*	0.044*
		(0.030)	(0.015)	(0.045)	(0.020)
Analysts coverage		0.075	−0.042	0.446**	−0.014
		(0.119)	(0.060)	(0.191)	(0.086)
Fixed effects		Y	Y	Y	Y
Model validation
Sample size		220	220	252	252
R^2 adjusted		0.053	0.492	0.106	0.356
F stat/Quasi-LR stat		1.980	18.08	2.823	9.486
P		0.020	0.000	0.001	0.000
Modified BP Test- heteroskedasticity	Stat.	0.44	13.127	3.158	0.316
	p	0.507	0.000	0.076	0.574
Hausman test - fixed effects	Stat.	13.13	25.99	19.01	13.19
	p	0.069	0.001	0.008	0.068
Breusch-Pagan test - random effects	Stat.	1.452	2.882	3.434	0.028
	p	0.228	0.090	0.064	0.868

Source: Authors’ calculations.

We observe the negative association between annual reports readability and their entropy score is statistically significant only for models estimated for period before COVID-19 pandemic ($Coef=-2.807,Sig.<1\mathrm{\%}$). The results show that the higher the entropy score of annual reports, the lower is their readability, meaning that if annual reports address multiple topics, the readability is significantly affected as there is no clear orientation of the financial reporting discourse. It seems that managers prefer to follow a standard approach on corporate reporting, looking especially for compliance with generally accepted reporting standards, such the GRI standards, rather than providing relevant information which could reflect better the impact of particularities of firms’ business model on financial performance.

Robust results are again confirmed related to the positive association between readability of annual reports and firm size as well, no matter the period analyzed and if corporate governance control variables are included in the estimated models.

Instead, the Flesch reading ease score is statistically affected only by firms’ corporate governance score. This measure of readability is more sensitive to the complexity of the words used, rather than the number of words used per sentence. Our results could suggest that there isn’t a pattern on how managers try to prepare obfuscated annual reports using fog words, meaning there isn’t a systematic preoccupation of managers to select fog words just to reduce the readability of annual reports.

However, it seems more effective mechanisms of corporate governance lead to less readable annual reports, with a higher marginal effect on the period of COVID-19 pandemic. Higher corporate governance score shows more mature and effective governance mechanisms. The results show that annual reports are less readable because of longer sentences, but not because of the use of complex words. Corroborated, the results suggest us that management is highly accountable for financial performance, which makes them to obfuscate annual reports. As TOP management incentive plans often consist of equity compensation, they are interested to stimulate attractiveness of the stocks on the capital markets. A root-cause could be that corporate governance mechanisms are effective more in the area of management accountability, but less effective in the area of shareholders or board of director’s oversight.

Instead, same results can suggest as well that strategic shareholders as well can put pressure on preparers inside the firm on how to prepare financial statements, as the use of fewer complex words is conditioned by higher stock market returns.

Industry effects are confirmed along all estimated models that control for effect of financial performance on annual reports readability score, based on Hausman test, for which the p value is less than 5%. Our results are similar with Efretuei (2020) who have underlined the effects of the specific of several areas of activity, such as financial services, technology, tourism, or utilities. Our results show in Figure 3 significant negative fixed effects in case of firms operating in commerce, or logistics. Instead, we observe positive fixed effects in case of firms operating in financial services, logistics or tourism area. These results show that annual reports are less readable in case of firms operating in commerce or logistics area, while firms operating in financial services area seem to be more concise and clearer.

Figure 3. Fixed effects of industry on estimated models.

Source: authors’ calculations.

Additionally, we observe from a temporal perspective that the COVID-19 pandemic has changed drastically firms’ approach on corporate reporting for firms operating in tourism, or financial services, leading to significantly less readable annual reports. Instead, in case of firms operating in IT and communications, the change in the readability of annual reports is even higher, as it goes from a positive value to a negative value, suggesting higher level of readability of annual reports. Therefore, those firms can be used for benchmark for the other firms, with the limitation of some basic constraints, such as the business model complexity, and corporate governance mechanisms.

5. Conclusions

This paper discussed the impact of firms’ financial drivers on annual reports characteristics, respectively the length of the annual report, the level of readability, and the degree of negative tone reflected on analyzed annual reports. The analysis is designed to highlight how the current COVID-19 pandemic affects annual reports and how managers use them to emphasize or to hide firms’ financial results.

Through multivariate analysis and text mining techniques, our paper reveals that the analyzed reports are less extensive in content and less readable. The results show increase in obfuscate information, especially in times of pandemic. However, the results are affected as well by industry fixed effects, as COVID-19 pandemic has affected distinctly the areas of economic activity. Firm size and accruals reported determine a significant positive marginal effect on annual reports readability, while firm size, accruals, and price earnings ratio determine a negative association with the tone of uncertainty of annual reports, especially during the period prior to COVID-19 pandemics. Nonetheless, we underline that users of financial reports should be more careful on annual reports discourse, including the tone preparers use, as this could be already a sign of potential obfuscation on financial reporting.

Finally, we conclude that the lack of a common reporting framework in pandemic conditions to build a disclosure index strictly related to the size of COVID-19 presentations in the report represents one of the main limitations of our research. Future research can empirically investigate the breakdown of a composite ESG score index to understand the extent to which annual reports readability is influenced by financial information vs. non-financial information. Future research can also empirically investigate the correlation analysis of financial and non-financial indicators with the characteristics of the annual reports, for the pre and post COVID-19 period.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This research was supported by the West University of Timisoara through FDI 0414 funds.