Complex information and accounting standards: Evidence from UK narrative reporting

The application of International Financial Reporting Standards (IFRS) has been introduced in many countries to enhance efficiencies in financial markets and improve communication in financial reporting. However, extant studies have suggested that the introduction of IFRS has increased narrative complexity, owing to the demand for more reporting. Considering that accounting complexity can be either informative (enhancing understanding) or non-informative, thereby causing obfuscation, this study performs an empirical analysis to highlight which of the two types of complexities may be affected by IFRS application. Using the setting of IFRS adoption in the UK and a word list-adjusted component of the fog index, this study decomposes complexity into two components: information (common complexity) and obfuscation (uncommon complexity). The results reveal that IFRS adoption has increased the common complexity of accounting narratives (information) but does not necessarily increase obfuscation. The study’s contribution is twofold: methodological through the decomposition of complexity using the term weighting concept and policy-related by identifying areas of increased narrative comparability in IFRS reports. Moreover, the study’s application of complexity decomposition to IFRS is novel. Future studies may apply this by using the identified information and obfuscation components to investigate the economic consequences of IFRS-associated complexity.


Introduction
This study investigates changes in the narrative complexity of corporate annual reports associated with the mandatory adoption of accounting standards. There are two motivations for this study. The first is the increase in complex financial narrative disclosures 1 (Lo et al., 2017), and the associated economic consequences (Asay et al., 2017;Callen et al., 2013;De Franco et al., 2015;Lee, 2012). The second is the role of accounting standards in increasing the complexity of narrative disclosures (Dyer et al., 2017). Here, the complexity of narrative disclosures refers to report readability, specifically the use of complex words, while accounting standards refer to the International Financial Reporting Standards (IFRS). Extant research on narrative complexity in IFRS compliant annual reports shows that IFRS increase complexity (Lang & Stice-Lawrence, 2015) and that this increase can be driven by increased disclosures commingled with complex narratives. In this study, we address the gap in the literature in relation to whether this increased complexity is informative or obfuscating. This study adopts the view that reporting incentives define reporting choices, implying that IFRS, despite being aimed at report comparability, allow discretion in reporting (Leuz & Wysocki, 2016). Firms exercise reporting discretion according to their expectation of positive outcomes. For example, Li (2008) shows that obfuscation of narratives increases with negative changes in earnings.
Within this incentives view, firms have the choice to provide either informative or obfuscated disclosures (Guay et al., 2016), moderated by the application of IFRS (Neel, 2017). This study is based on the premise that narrative complexity commingles the two latent components of information and obfuscation (Bushee et al., 2018). The information component of narrative complexity represents the technical nature of accounting narratives − for example, the use of complex words such as "financial" or "incorporated". The obfuscation component represents obscurity or perplexity and is intended to reduce the informativeness of narrative disclosure − for example, the excessive use of non-technical complex words. This study decomposes narrative complexity into i) an informative component measured as "common complexity", that is, complex words that are commonly used in annual reports, and ii) an obfuscation component measured as "uncommon complexity", or complex words that are not regularly used in annual reports. This study uses both components to examine whether the adoption of IFRS increases the narrative complexity of annual reports.
The findings show that narrative complexity associated with the adoption of IFRS appears to increase in terms of informative complexity. Further analysis shows that this increase in narrative complexity is significantly associated with IFRS adoption after controlling for the expected increase over the years, that is, the changes in complexity with IFRS adoption and the changes in reported economic effects with IFRS adoption. However, there is an increase in narrative obfuscation 2 that tends to disappear after controlling for time effects. The observed increase in obfuscation appears to be consistent with a general increase in obfuscation over time.
This study contributes to the existing literature in two respects. First, in terms of methodology, this study applies a measure of complexity based on the reporting incentives view to analyse the two components of narrative complexity: incentive to obfuscate and incentive to inform. It does this using the term weighting concept (Jegadeesh & Wu, 2013). Second, with regard to policy, this study contributes to the International Accounting Standard Board's (IASB, 2017) Disclosure Initiative project and supports the European Financial Reporting Advisory Group's (EFRAG) recommendation to address complexity in the conceptual framework (EFRAG, 2014) by indicating which components of complexity are influenced by IFRS adoption. The results show that the application of IFRS increases the informative component of disclosure. The IASB's Disclosure Initiative project is focused on addressing three aspects of the disclosure problem: (i) not enough relevant information, (ii) irrelevant information, and (iii) ineffective communication of information. 3 This study finds that the application of IFRS does not complement the disclosure problem. In fact, the observed increase in "common complexity" suggests an increase in the informativeness and, to an extent, comparability of narratives post-IFRS adoption.
The remainder of the paper is structured into six sections. Section 2 reviews the relevant literature. Section 3 presents the narrative complexity classification. Section 4 develops the hypothesis. Section 5 discusses the research design. Section 6 presents the empirical results, and Section 7 concludes the paper.

Literature review
Studies show that the reading difficulty of annual reports has increased over the years (Dyer et al., 2017). Li (2008) finds that firms whose annual reports are less readable or difficult to read have lower earnings persistence, while firms whose annual reports are easier to read have more persistent positive earnings. Even if companies apply the same reporting standards, differences arise in reporting outcomes owing to internal and external factors that shape incentives (Leuz & Wysocki, 2016). Supporting the view that reporting incentives influence readability, Lo et al. (2017) find that increased reading difficulty is associated with firms that have managed to surpass the prior year's earnings, concluding that obfuscation incentives increase the reading difficulty of narratives. Similarly, Dempsey et al. (2012) conclude that narrative complexity is higher in the reports of poorly performing firms.
Both internal and external factors shape reporting incentives. Internal factors include the company's operating characteristics, while external factors include legal institutions and enforcement regimes or capital market forces (Leuz & Wysocki, 2016). In applying IFRS, firms with incentive to provide informative disclosures tend to increase the information components of their narrative disclosures, while obfuscation incentives are more likely to increase the non-informative components of disclosures. A number of studies find that bigger, more complex, and more volatile firms tend to have more complex disclosures (Li, 2008). However, application of IFRS can moderate this complexity through disclosure requirements. Dyer et al. (2017) analyse the narrative trends in annual reports over the years and report poorer readability. They find that regulatory requirements explain most of the changes and topics; for example, fair value and risk factor disclosures.
Studying IFRS and readability, Cheung and Lau (2016) examine the notes appended to financial statements and report that they have become lengthier following the adoption of IFRS. Interestingly, they observe that narrative disclosures have increased in length, while readability scores have decreased. A lower readability score implies improved readability of the text. Thus, in addition to improving disclosure quantity, IFRS may increase the informative components of disclosure. Accounting regulations such as IFRS are viewed as governing extraction cost (Bloomfield, 2002). The incomplete revelation hypothesis (IRH) is based on a "noisy rational expectations" model, which assumes that the cost of extracting information from publicly available data increases with information complexity (Bloomfield, 2002). However, by introducing IFRS, information becomes standardised, comparable, and accessible to users, thereby fulfilling the standards' role of increasing the informative components of disclosure.
Even if IFRS were to be introduced with the intention of reducing extraction cost, 4 a firm's reporting incentives would affect the application of the standards; moreover, firms' discretion in application provides obfuscation opportunities. Firms increase the complexity of their narrative disclosures when there are benefits to be gained by doing so. Numerous studies show that narrative complexity is associated with market incentives (Lawrence, 2013;Miller, 2010;Rennekamp, 2012;Tan et al., 2015;You & Zhang, 2009) and investor disclosure choices (Asay et al., 2017;Filzen & Peterson, 2015). According to You and Zhang (2009), investors' under-reaction to 10-K disclosures appears sluggish and tends to be more prominent for firms with complex disclosures. Filzen and Peterson (2015) reveal that analysts' reliance on narratives increases with the complexity of financial disclosures. Lawrence (2013) and Miller (2010) show that the complexity of disclosures affects retail and small investors' reactions and disclosure choices.
IFRS also increase the narrative complexity of disclosures. Richards and Staden (2015) hypothesise that the complexity of accounting standards tends to increase the associated narrative complexity in annual reports, contrary to Cheung and Lau (2016), who find that IFRS adoption improves the readability of narrative disclosures. Most of these studies use the "fog index" readability measure, which is widely used in the accounting literature to measure the reading difficulty of accounting narratives (e.g., Li, 2008;Lo et al., 2017). The fog index uses a combination of complex words and complex sentences based on the assumption that, on average, more words per sentence and more syllables per word make a document more difficult to read.
The impact of IFRS on narrative complexity appears to be mixed, but there is consensus that the accounting standard increases disclosure length. Lang and Stice-Lawrence (2015) find that the "fog index" of annual reports has increased over the years and attribute this trend to the disclosure of complex topics, such as financial instruments, currencies, and taxes. The authors also study the benefits of IFRS and show that firms with a higher fog index experience fewer benefits from IFRS. 5 If IFRS are associated with increased complexities characterised by the fog index, and a higher fog index is associated with poor economic outcomes − such as a reduction in IFRS benefits, earnings management, and lower earnings − then IFRS application would likely provide discretionary opportunities for increasing reporting outcomes and variation in narrative complexity. Christensen et al. (2013) find increased liquidity benefits with enforcement during the IFRS application period. In the United Kingdom (UK), enforcement was characterised by the change from reactive to proactive reviews. A reactive approach is a limited review of companies brought to the regulator's notice, while a proactive approach involves identification of a broad sample of companies for review (Christensen et al., 2013). 6 Proactive reviews evaluate 15% of interim and annual financial reports, which creates new incentives for narrative obfuscation. This is because in a proactive regime, consistent with the expectation of the management obfuscation hypothesis (Li, 2008), firms that have incentives to manage earnings through poor application of IFRS tend to obfuscate reported disclosures to circumvent selection. Meanwhile, firms that are more likely to enhance reporting outcomes with IFRS adoption are also more likely to report clear and informative disclosures.
Isolated benefits of IFRS are observed in countries with smaller differences between local accounting standards used pre-IFRS adoption and IFRS. For instance, using a sample of UK firms, Brochet et al. (2013) observe that IFRS increase overall comparability. Lang and Stice-Lawrence (2015) observe an increase in the narrative comparability of reports following IFRS adoption and attribute this to an increase in relative word similarities across reports. While the UK has had limited differences in standards pre-and post-IFRS adoption (Bae et al., 2008), the exogenous effect of increased comparability influences disclosure behaviour (Brochet et al., 2013). Bae et al. (2008) show that the pre-IFRS UK accounting standards are similar to IFRS, indicating limited changes in disclosure. The motivation for a change in disclosure post-IFRS adoption implies increased uniformity in standards across multiple countries (Brochet et al., 2013). Firms tend to signal private information using narratives because standards have become more accessible and comparable. The annual reports of UK firms have both regulated and unregulated narrative components, observed as a potential mechanism for discretionary reporting (private information).
IFRS have focused less on wider narrative reporting, with no separate standard for it (only a practice statement, "The IFRS Management Commentary Practice Statement"). This creates the expectation that IFRS regulation will increase the complexity of narrative reporting because it is not standardised and, thus, allow obfuscation. Conversely, the application of standards can increase narrative complexities through improved discussions on complex financial disclosures, such as fair value disclosures (Dyer et al., 2017), and an increase in narrative comparability (Lang & Stice-Lawrence, 2015). In the next section, we discuss the decomposition of complexity in narrative reporting and how it is applied in this study.

Narrative complexity classification
In this study, we utilise the concept of "word complexity" (derived from the fog index) 7 for classifying narrative complexity. In particular, this study is based on the premise that the theoretical construct of narrative complexity is composed of two latent components, information and obfuscation (Bushee et al., 2018), and that the target audience tends to understand only one component of complex words in annual reports (Loughran & McDonald, 2014). This complex component is labelled the informative component. Thus, this study decomposes the observed complex words in narratives into two components: i) information labelled as common complexity and ii) obfuscation labelled as uncommon complexity.
To measure these common and uncommon complexity components, the theoretical concept of "term weighting" is applied. The core idea underlying the term weighting technique is that the effectiveness or impact of a word in any text is inversely related to its occurrence; that is, if a word occurs with higher frequency, it should be weighted with less impact than a word that occurs less frequently in the text (Manning & Schutze, 2003). For example, the word "and" occurs frequently in most texts, but its high frequency does not equate with its impact on those who read the text because the more readers see the word "and", the more likely they are comfortable with the meaning of the word.
The process of term weighting involves the calculation of the TF.IDF weight of each word, which considers the occurrence frequency of words both in the defined text and in the defined corpus. In this context, the term "defined text" is a company's annual report, whereas "defined corpus" is the universe of all company annual reports in the sample. By calculating the frequency of a word in the corpus, we can consider whether that word is concentrated in the defined text or in the defined corpus. The purpose is to account for the likelihood that the readers of similar texts are "comfortable" with the word. The occurrence of words with higher TF.IDF weights is less frequent in the defined corpus than words with lower TF.IDF weights.
Based on the concept of term weighting, more "familiar" words in the annual report corpus are expected to have lower weights, and less "familiar" words are expected to have higher weights. The term TF.IDF has been applied in accounting narrative studies to moderate narrative measures. Specifically, Loughran and McDonald (2011) apply it to moderate the impact of high-frequency words in the repetition of tonal words in 10-K disclosures. TF.IDF originates from the concept of "term weighting" in the literature on information retrieval. 8 It is composed of two terms: TF and IDF. The first term, TF, attenuates the impact of high-frequency complex words in a document. It is a normalised frequency of the word in the document applied. The second term, IDF, modifies the impact of a complex word in a document based on its commonality within a defined corpus (Loughran & McDonald, 2011). It indicates whether the word is common or rare in the corpus of documents. The assumption is that the impact of a complex word reduces as its frequency increases within the corpus. Thus, a high-IDF term decreases the impact of the first term, TF. Manning and Schutze (2003) denote TF.IDF as (i) highest when a term (complex word) occurs repeatedly in a few documents in the defined corpus, 9 (ii) lower when the occurrence is less frequent in a document or occurs in multiple documents, and (iii) lowest when the term occurs in virtually all documents.
This study applies the theoretical assumptions of "term weighting" to identify the two categories of complexity. It measures uncommon complexity based on the assumption that complex words with high TF.IDF weights tend to increase obfuscation. This is because high TF.IDF words are less frequent in the defined corpus and, thus, are less familiar to the users of the corpus (Manning & Schutze, 2003). Following this assumption, we expect that words with high frequency are more likely to be familiar to readers, consistent with prior literature assumptions (e.g., Loughran & McDonald, 2014). If a complex word occurs in a firm's annual report but is not common in the identified corpus, it is assumed that its persistent use is associated with obfuscation. Kim et al. (2019) find that an adjusted word complexity component of the fog index, which is adjusted based on the concept of "familiar words", is significantly related to the post-filing date return volatility as opposed to the traditional fog index.
Conversely, words with low TF.IDF scores have greater narrative complexity, 10 although they are more informative. Complex words in this category occur more frequently in the annual report corpus and are considered familiar to investors. Based on Manning and Schutze's (2003) definition, TF.IDF terms are lowest when they occur in virtually all documents. In this study, the implication is that words identified as low TF.IDF complex words tend to occur in all documents in the defined corpus. Thus, we identify words with low TF.IDF scores to test the association between the application of IFRS and common or uncommon complexity. 11 The concept of TF.IDF is an appropriate measure to decompose words theoretically into informative and obfuscating components of complexity, given its extensive application in the information retrieval literature to moderate the impact of words in a corpus. 12 Based on the theoretical observation that narrative complexity measures two latent components (information and obfuscation), this study deems that words with low TF.IDF weights indicate information. These words are either informative (necessary technical details) or common (persistent across reports). 13 To identify low TF.IDF words, this study draws from the identification of 52 words that account for a quarter of the total count of complex words in annual report narratives 14 (Loughran & McDonald, 2014). Loughran and McDonald (2014) identify them as "familiar words" that users of annual reports are accustomed to, and their complexity is informative rather than obfuscating.

Hypothesis development
The study aims to answer the following research question: what is the role of accounting standards in mitigating obfuscation and enhancing the informative component of complex annual report narratives? The hypothesis is developed based on the theoretical views of information and obfuscation (Asay et al., 2018;Bushee et al., 2018;Lo et al., 2017;Stone & Parker, 2016). Information is interpreted as driven by the increased complexity of technical disclosures, while obfuscation is interpreted as driven by increased complexity not attributable to informative disclosures (Bloomfield, 2008;Bushee et al., 2018;Merkl-Davies & Brennan, 2007). The analysis of complexity is at the word level; this implies that increased disclosure of uninformative complex words indicates increased complexity (obfuscation), while disclosure of informative technical words represents information (Courtis, 2004).
Several studies report that IFRS adoption increases the fog index (Lang & Stice-Lawrence, 2015;Richards & Staden, 2015). However, it is unclear whether this increase is driven by the increased disclosure of informative complex words or the increased disclosure of obfuscating complex words. With IFRS application, discretionary opportunities in disclosure can lead to either informative reporting outcomes or uninformative outcomes depending on the reporting incentive (Christensen et al., 2015;Gao & Sidhu, 2017;Leuz & Wysocki, 2016). If IFRS application increases the complexity of reports, and firms have lower IFRS benefits such as reduced liquidity and analyst following , it is more likely that IFRS application increases the obfuscation component of complexity. Moreover, lower liquidity and analyst following indicate that a firm's information environment is obscure. However, if IFRS foster improved communication (Tarca, 2004), clear disclosure relative to local requirements, and increased technical disclosure (Lang & Stice-Lawrence, 2015), then IFRS adoption would be more likely to increase the disclosure of complex informative words.
The complexity index that is used to measure information and obfuscation is a function of weighted word occurrence and syllabification. The first input is a component of the fog index studied by Lo et al. (2017) in the context of financial performance. Li (2008) finds a negative relationship between firm profitability and the fog index. Bloomfield (2008) provides several alternative explanations for Li's (2008) observed results, including obfuscation and information. Li's (2008) findings are consistent with the observation that management uses linguistic obfuscation to make textual disclosures more complex when performance is poor. Bushee et al. (2018) provide a theoretical view by decomposing the two components using the fog index, where information is aimed at explaining disclosure complexity as a necessary detail. This study uses a component measure of the fog index to estimate the complexity index.
Specifically, this study uses the complex-words component of the fog index to calculate "common" and "uncommon" complexity. Loughran and McDonald (2014) analyse the readability of accounting narratives and find that the component of complex words in the fog index does not explain the market's reaction to disclosures. The authors explain that investors easily understand words such as "management" and "corporation" despite that they are termed "complex". While these words have two or more syllables, they are familiar to users of such reports. The authors show that these complex words account for the first quartile of the most frequently occurring complex words in annual reports. Bonsall et al. (2017) identify the components of plain English attributes that capture narrative complexity, namely, the bog index. The word "bog" is designed to capture "word difficulty", which "bogs" readers down. It is calculated using a general vocabulary, identifying words that are consistent with the notion of "familiar" words in annual reports (Bonsall et al., 2017). Both studies find that certain words that are often identified as complex are familiar to investors, and the use of such words in annual reports is informative rather than obfuscating.
Following this notion, if IFRS adoption increases the disclosure of "familiar" words in annual reports, it would increase the information component of narrative complexity, and this is identified as common complexity (information). Alternatively, if the adoption of IFRS increases the disclosure of complex words that are not identified as "familiar" in annual reports, it would increase obfuscation, and this is identified as uncommon complexity. Given that IFRS application is expected to improve informativeness through better quality technical disclosures, it is likely to persist after first-time adoption. Therefore, we propose the following testable hypothesis: 15 Hypothesis 1 (H 1 ): Informative disclosures in annual reports (common complexity) are positively correlated with IFRS application for 2 years after first-time IFRS adoption.

Research design
The empirical analysis tests whether increased narrative complexity following IFRS application is associated with the disclosure of complex words that are more informative in annual reports. The general equation used to test H 1 is: The sampling process and variables are discussed in the below subsections.

Sample selection and data preparation
To test H 1 , this study uses the annual reports of a sample of companies listed on the FTSE All-Share (FTALSH) index. This study assesses the changes in disclosure complexity between the pre-and post-IFRS adoption periods. It begins with firms listed on the The adoption year is excluded to eliminate the effects of narrative reporting from the results for first-time IFRS adoption. The narrative data for each firm must include 4 years of data to enable a matched comparison of the pre-and post-adoption years.
To retrieve the annual report narratives for each firm in the FTALSH index, the reports were downloaded from the Thomson ONE database and the respective company websites of firms included in the sample. In line with the methods used in the readability literature, the text extraction process excludes sections with more than 50% numerical content as opposed to textual content (Li, 2008;Miller, 2010). The data collection process excludes interim reports presented as annual reports. For firmyear observations, the study eliminates annual reports with fewer than 2000 words from the sample and firm-year observations with missing or extreme values. Furthermore, the study eliminates firm years with no matching financial statement data (see Table 1 for details of the data collection and sampling process). The final study sample comprises 939 firm years following incorporation of financial statement data results. The sample size varies across the control variables owing to data availability. The sample sizes are reported for the regression analysis. Table 2 presents the definitions of the key variables and control variables applied in the regression. The identified control variables are as applied in similar studies on the readability of annual reports (e.g., Li, 2008). The data collection steps indicating the estimation of both common and uncommon complexity are discussed in Appendix I. The list of words identified as low TF.IDF is provided in Appendix II.

Variable definitions
H 1 involves the influence of mandatory IFRS adoption regulation regarding textual word complexity (common and uncommon complexity). Therefore, it requires measures of word complexity decomposed into two components. This study does not measure readability (fog index) or length but, rather, focuses on word complexity in evaluating the informative and obfuscating components driving complexity following IFRS application. To provide a benchmark for the increase in information during the sample years, the logarithm of the number of words is captured as a control variable. Table 3 provides summary statistics of the narrative and control variables. Panel A provides descriptive statistics for the complete sample. Panel B shows descriptive statistics for the pre-and post-IFRS group. Reports, on average, contain 36,000 words. From panel B, it can be observed that the length of reports is smaller in the pre-IFRS period, compared to the post-IFRS period. The numcompwords variable shows an increase in IFRS years, which may indicate the disclosure of complex topics. Deconstructing this further, the variables hightfidf comp words and lowtfidf comp words both increase in the post-IFRS years. This indicates that both the informative and obfuscating components of complexity, as measured in this study, increase during the IFRS application years. The variable comp words is a component of the fog index and captures average word complexity. As seen in Table 3, the average fog index of the sample of reports is 22, with a minimum of 16 and maximum of 31. This is consistent with prior literature, which observes similar levels of fog index scores. 16 Profit_loss is an indicator variable where 1 = profit firms and 0 = loss firms. The mean of 0.92 for profit_loss indicates that the sample is skewed toward profit-making firms, comprising 92% of the sample. Table 4 presents correlation results of the variables included in the testing of H 1 in the regressions. The variables hightfidf and lowtfidf are positively correlated, although the correlation is below 10%, indicating that they are not interchangeable. The correlations between the narrative variables and determinant variables are consistent with the expectations of this study. Consistent with prior studies, earnings are negatively associated with the word complexity measures (Li, 2008). However, the correlation is not significant for the hightfidf measure, indicating the broad nature of the complexity measure. The primary variable of interest year_ifrs is positively correlated with the narrative variable comp words (21%), which is consistent with the view that accounting standards increase narrative disclosures in annual reports (Dyer et al., 2017). hightfidf and lowtfidf are positively correlated with IFRS, consistent with the proposed hypothesis, H 1 , that narrative   Table 5 presents the results of the determinants of the dependent variables included as controls in the hypothesis testing in the regressions. Consistent with prior studies (Hasan, 2018;Li, 2008), earnings and earnings volatility are significantly associated with complexity variables, including compwords and lowtfidf. However, they are not significantly  associated with the occurrence of hightfidf words. This is expected because the lowtfidf word complexity variable is more likely to capture financial words related to earnings reporting. Prior literature reports mixed results on the relationship between earnings and narrative length. Li (2008) finds a negative relationship between earnings and report length, while Lang and Stice-Lawrence (2015) show that profitable companies have lengthy reports. The results in Table 5 are consistent with a positive association between earnings and length, indicating that firms with positive earnings tend to have more informative (longer) annual reports than firms reporting lower earnings. This finding is consistent with the results of Lang and Stice-Lawrence (2015), 17 who studied an international sample. In addition, this difference is expected because the content of annual reports differs significantly between corporates in the United States (US) and the UK, as reports in the UK are unstructured in comparison to the US 10-K filings (El-Haj et al., 2019). Geographical complexity tends to increase the use of complex words. However, lowtfidf words show a negative association with geographical complexity, indicating that firms with more geographic segments have less "common" complex words. The significant association between geographical complexity and hightfidf shows that firms with more geographic segments are likely to have higher variation in narrative reporting. They are more likely to report words that are less common within the corpus of all reports and more likely unique to the firm. This is consistent with the expectations of TF.IDF assumptions. Based on the TF.IDF literature, words with higher TF.IDF occur less frequently in the universe of the corpus considered but could have a higher occurrence in one or a small number of documents in the sample. We also note the positive association between bus_comp and the length of t statistics in parentheses. * p < 0.05, ** p < 0.01, *** p < 0.001. Table 5 presents results from estimating the determinants of the key dependent variables. See Table 2 for variable definitions. The regression analysis is performed with robust standard errors to obtain unbiased estimates under heteroskedasticity. We first present results of a model without firm fixed year effects in Table 5. Doing so permits us to investigate the impact of variables that exhibit little within-firm variation over our sample period. However, given that our controls variables have limited variables that exhibit these characteristics, we perform the remaining analysis with firm fixed effects.

Determinants
reports, which supports the view that firms with more complex operations have longer reports. This supports the ontological explanation that 'some aspect of reality … that requires more dictionary entries' in its description tends to be more complex (Bloomfield, 2002). This finding may be caused by complex firms needing more text to explain their context. Most of the determinant variables appear to be consistent with the predicted signs, which are based on prior literature (e.g., Li, 2008). Most of the variables in the regression are significantly associated at the 5% level with the length or complexity variables, making them suitable as controls for testing H 1 in the regression. 18 The results show that textual disclosures in annual reports increase in length following IFRS adoption, which is consistent with other studies (e.g., Lang & Stice-Lawrence, 2015). The dependent variables' word complexity throughout the entire report (comp words) increases with IFRS adoption. Specifically, the decomposed variables of interest are significantly and positively associated with the IFRS adoption years. This indicates that IFRS adoption increases both common and uncommon complexity, as measured using t statistics in parentheses. * p < 0.05, ** p < 0.01, *** p < 0.001. Table 6 presents the results from estimating the regression of complexity variables and IFRS adoption. year_ifrs is an indicator variable equal to 1 for the post-IFRS adoption years and 0 for pre-adoption years. All control variables are included in the regression model. See Table 2 for variable definitions. The regression analysis is performed with robust standard errors to obtain unbiased estimates under heteroskedasticity. the word list in Appendix II. The coefficients of both variables, lowtfidf and hightfidf, are positive and significant. The results are deemed significant at the 1% level. The significant coefficient of lowtfidf is consistent with H 1 , namely, that informative disclosures in annual reports (common complexity) are positively correlated with IFRS application in the 2 years following first-time IFRS adoption. While we do not make predictions regarding hightfidf, the results indicate a significant positive association between IFRS and obfuscating disclosures in annual reports (uncommon complexity).

Narrative complexity and IFRS adoption with economic and time effects
From the findings, a concern arises as to whether the results capture an increase in informative and obfuscating components of annual reports, given the economy-wide effects that are external to the firm and occur over time. Thus, the regression analysis is estimated on the assumption that complexity increases over time, as recorded in previous studies (Lang & Stice-Lawrence, 2015;Li, 2008). To exclude these external effects, we introduce two  Table 7 shows that the coefficients of year_ifrs remain significant for all of the dependent variables except for the hightfidf complex measure. The increase in hightfidf word complexity appears to be driven primarily by time1, indicating that after IFRS adoption, there is no significant increase in uncommon complexity. The variable hightfidf indicates uncommon complexity, while lowtfidf indicates common complexity. The variable lowtfidf remains positive and significant, supporting the hypothesis that IFRS application increases the disclosure of informative complex words (common complexity) in the 2 years post-IFRS adoption. There is a 0.6% increase in the disclosure of informative complex words in the 2 years post-IFRS adoption. The results of the variable hightfidf add to those of other studies; for example, Kim et al. (2019) show that complex words that investors are not "familiar" with add noise to investors' information processing and are not associated with financial information. The significant association between hightfidf and time1 indicates that this is also driven by wider time trends that have allowed an increase in the disclosure of these words. It can also be associated with new words that have been incorporated into the financial corpus not captured in the hightfidf word list.

Further analysis -The impact of firm performance and firm complexity
To extend the analysis of obfuscation and information, the study reports additional regression results in Table 8; interaction terms are included in the regression model in Equation 1a. This aids in better understanding the relationship between the variables driving increased word complexity and the impact following IFRS adoption. The regression model used to test H 1 models the level of word complexity based on the identified firm characteristics and whether the reported narratives are in the pre-or post-IFRS adoption period. The regression equation is estimated as follows: firmcharacteristics are labelled Controls in Equation 1a. Li (2008) finds primary results consistent with the IRH that management makes bad news more costly to extract by writing excessively long and complex reports. This is in line with the management obfuscation concept that where earnings are poor, reports are likely to be more complex. Above, we find that IFRS reports are more complex; however, we do not know whether this effect is increased by firms reporting poorer earnings upon application of IFRS. If firms with poor performance use more complex words in their disclosures post-IFRS and these increase with IFRS adoption, then it supports the view that IFRS adoption itself does not change observed reporting obfuscation. Bloomfield (2002) offers an ontological explanation that firms use excessively complex language because the context to be described requires more dictionary categories, for example, good news being inherently easier to communicate than bad news (Lo et al., 2017). Likewise, in the post-IFRS period, it is possible that firms with complex operations * p < 0.05, ** p < 0.01, *** p < 0.001. Table 7 presents results from estimating the regression of complexity variables, including time1 and time2 variables as independent variables. time1 indicates the increasing complexity of annual reports over time, and time2 indicates increasing complexity in the post-IFRS adoption years. All control variables are included in the regression model. See Table 2 for variable definitions. The regression analysis is performed with robust standard errors to obtain unbiased estimates under heteroskedasticity.
tend to use a higher proportion of complex words owing to increased disclosure classes, whereas firms with less complex operations tend to use fewer complex words. Another likelihood is that operational complexity affects the level of word complexity during both the pre-and post-IFRS adoption periods, but it may increase with the adoption of IFRS given, for example, the accounting standards for specific business segment reporting. This supports the perspective that the level of word complexity in annual reports post-IFRS adoption is associated with increased informative disclosures. We test this by adding an interaction term to Equation 1b in which two firm characteristics, business complexity (bus_comp) and earnings (earns) variables, are multiplied by the variable year_ifrs: A significant coefficient for the interaction term indicates that the effect of bus_comp/ earns on the level of word complexity varies with the values of the year_ifrs variable. To test the business complexity and earnings effect, both the interaction term and the individual variables bus_comp and earns, respectively, without interaction (in this case, b 2 ), are t statistics in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 Table 8 presents results from estimating the regression of the complexity variables and the independent variables bus_obf and earns_obf, which are the interaction of year_ifrs with business complexity and earnings, respectively. All control variables are included in the regression model. See Table 2 for variable definitions. The regression analysis is performed with robust standard errors to obtain unbiased estimates under heteroskedasticity.
relevant. Adding an interaction term to the model changes the interpretation of the coefficient of b 2 . In Equation 1b, that is, without an interaction term, and consistent with prior literature that tests obfuscation using performance, the interpretation of this variable shows the unique impact of earnings on the level of complexity. Including an interaction term in the model shows that the effect of earnings on word complexity at different levels of year_ifrs is not limited to b 2 but also depends on the values of b 3 . This implies that b 2 excludes the effect of b 3 , as it is held constant. b 2 is interpreted as the unique effect of earnings on complexity only when year_ifrs = 0. The significance of b 2 indicates that earnings change complexity during the pre-IFRS period, while the coefficient of b 3 indicates that earnings change complexity during the post-IFRS period. The interpretation of b 1 remains the main variable of interest; that is, year_ifrs increases word complexity after controlling for the interaction of year_ifrs and firm characteristics. Table 8 reports the results for the regression analysis of equation 2. To explain specific IFRS associations, Table 8's results include firm fixed effects, time fixed effects, and time measured variables used to test H 1 to control for firm and time factors affecting the complexity of annual report disclosures during the IFRS adoption period. Table 8 presents the regression tests for H 1 after controlling for the interaction of firm variables and IFRS adoption. This is to ensure that the observed results are not driven by increased word complexity attributable to reporting disclosures on firm complexity and firm performance post-IFRS application.
Row 3 in Table 8 indicates that the interaction of earnings and the IFRS variable is not associated with increased complexity of lowtfidf. When we control for firm-specific effects and include the time variable, it appears from the results that firm performance does not drive the disclosure reporting with more complex and informative words. This supports the IFRS objective of transparency, showing that despite the strong association between earnings and complexity, as reported in Table 4, firm performance and IFRS adoption do not drive common complexity. This provides evidence of achieving the IFRS objectives of improved report transparency and comparability, with respect to the disclosure of "common" complex words in annual report narrative disclosures notwithstanding firm earnings.
The results support H 1 , that post-IFRS application, informative complexity increases after controlling for earnings-driven complexity, as shown by the significant positive coefficient of year_ifrs and lowtfidf. However, there is an association between the earnings IFRS interaction term and hightfidf words, indicating that the interaction of earnings and IFRS adoption is positively associated with disclosures of uncommon complexity. This denotes that uncommon complexity is largely driven by earnings reporting, supporting the concept of uncommon complexity as an extraction cost. Accordingly, complex words that are not "familiar" to investors are more likely to be positively associated with earnings narrative management post-IFRS adoption. The results suggest that with IFRS adoption, obfuscation is driven by earnings; however, the findings do not make specific predictions regarding the obfuscation component of disclosure, as the research design extracts informative words and categorises words that are not considered informative as obfuscation.
Row 2 shows that the interaction of business complexity and IFRS adoption (bus_obf) is not associated with word complexity in annual reports. In this study, business complexity is measured as the number of business segments, implying that complexity associated with IFRS adoption is not driven by business complexity. This further strengthens the results of IFRS-specific effects on complex words. This result also includes the time variable, showing that with the increase in complexity over time − that is, increased disclosures for larger firms and changes in disclosure needs − IFRS increase word complexity.

Conclusions
This study assesses the impact of IFRS adoption on narrative complexity by decomposing complexity into two components: information and obfuscation. The study's findings provide evidence that IFRS adoption increases the use of complex words in disclosures, and this increased word complexity grows with the informativeness of reports, identified as common complexity. In other words, narrative complexity, when modelled as common complexity, increases with IFRS adoption. The findings also show that increasing narrative obfuscation is identified as uncommon complexity; however, this effect reduces when the assumption that time increases word complexity is introduced into the model. The study shows that the effect of IFRS on disclosure complexity is not driven by obfuscation with respect to earnings and business complexity. While we observe an association between earnings and uncommon complexity, this association is not related to IFRS adoption.
The increase in word complexity is persistent with IFRS adoption, indicating increased complexity of disclosures for users of annual reports. However, this increased word complexity is driven by an increase in the common complexity component of narrative complexity. This contribution is novel and complements prior studies by identifying a component of complexity using an adjusted word complexity component of the widely studied fog index. The study further shows that this component is associated with the IFRS disclosure complexity observed in prior studies. This study does not establish direct causality owing to the existence of mandatory IFRS requirements, the introduction of the new standards, or firms' reporting incentives around this period. However, it reports a significant increase in the complexity and informativeness of reports based on the observed significant increase in common complexity. The study also attempts to control for these effects in the proposed research design.
Future studies can extend this research by enhancing the lowtfidf words (common complexity words). This can be done by incorporating words that are not present in Loughran and McDonald's (2014) list of informative words. For example, studies can apply the list adjustment in Kim et al. (2019). In addition, IFRS was adopted in the UK in 2005 and has evolved over the years with amendments to the standards, such as IFRS 9 and IAS 39 Financial Instruments, 20 which are perceived as inherently complex (Lang & Stice-Lawrence, 2015). Future studies could complement this work by analysing the changes in the informative and obfuscating components of complexity, focusing on the change in IFRS over time.
This study is relevant for the "Better Communication" component of the IASB's Disclosure Initiative project, which recognises that valuable information is lost in voluminous disclosures. While there is agreement that narrative complexity is a significant concern that reduces accessibility to accounting narratives, it is unclear how accounting standards drive complexity. Further research into the determinants of this increase in common complexity with the adoption of IFRS can guide standard setters on the specific standards driving narrative complexity. The results indicate that IFRS annual reports tend to become more comparable with the increase in common complexity in the post-IFRS adoption period. The IASB recommendation is to actively supply information to reduce complexity, for example, by providing references and, in some instances, further explanation. However, the use of words is a factor worth considering in the IASB's recommendations. Specifically, this study shows that IFRS themselves drive increased use of complex words, although these words are informative. Further debate and discussion are required within the IASB to determine whether the findings of an increase of complex words with IFRS complexity persisting for years post-IFRS, the predictions are likely to persist for more than 2 years after the adoption year. 16. Studies have reported average fog index scores from approximately 20 (Lang & Stice-Lawrence, 2015;Li, 2008) up to 22 (Dyer et al., 2017). The type of corpus (management report or annual report), the sample period of the study (later years have higher fog index scores), and the institutional context (US or UK) tend to affect the average fog index reported. 17. Lang and Stice-Lawrence (2015) use an indicator variable coded 1 if the firm reports losses. 18. Variables that are not significantly associated are included to maintain consistency with prior research on readability and accounting variables. 19. For example, in the UK, there was a change in the enforcement regime aligned to the regulation of IFRS adoption (Christensen, 2013). This may impact the form and content of narrative disclosures not due to the application of IFRS but, rather, to align to new enforcement regime requirements. 20. See section 20 [4] of Gore-Browne on EU Company Law, which details the changes in standards since the adoption of IFRS in 2005 (Efretuei, 2020b). 21. This is a Linux-based platform that converts a pdf file into a text file. It has an additional check using OCR whereby the pdf files cannot be converted due to inclusion of embedded text. 22. This module, developed by James (2018), is used to split text into words, paragraphs, segments, and tiles. The Perl module is part of a Perl programme and is used to analyse the annual reports and produce a file of split words for each annual report. 23. Lingua:: EN::Syllable::syllable estimates the number of syllables in the word passed to it.
Further information on the code can be found in Bowers and Fast (1999).
outputs a text file with words split into new lines. The criterion for identifying words is that a word must have letters and at least one vowel sound. The next step is to input the converted text files into a computer programme that separates the words in the file into two groups: group A is a text file with common complexity words, identified using the high-frequency financial words (Loughran & McDonald 2014) listed in Appendix II; group B is a text file excluding the common complexity words − uncommon complexity words defined as complex words with three or more syllables and not present in group A. Following this, the text files of groups A and B are input into a Perl programme, "Lingua::EN::Syllable", which estimates the number of complex words for each group. 23 The data steps are as follows: .
Step 1: . Download PDF annual reports . PDF-to-text conversion using OCR for text preservation .
Step 3: . Extract common complexity words for group A text files and exclude common complexity words from group B text files using author-written Perl code. . Step 4: . Compute word complexity weighting for lowtfidf complex words and hightfidf complex words using Perl programme https://metacpan.org/pod/Lingua::EN::Syllable . This produces two measures of word complexity: . lowtfidfcommon complexity (information) . hightfidfuncommon complexity (identified as obfuscation) .
Step 5: . Run regression analysis using word complexity weighting as the dependent variable and an IFRS adoption dummy as the independent variable, including relevant controls.