Integrated reporting and analyst behaviour in diverse institutional settings

Purpose – Given the limited studies that have started to focus on contexts where integrated reporting (IR) is voluntarily adopted, this paper aims to explore the moderating role of institutional characteristics on the associationbetween voluntaryreport releaseandanalyst forecast accuracy. Design/methodology/approach – This study uses a quantitative empirical research method grounded on voluntary disclosure theory to provide empirical evidence on an international sample of companies choosing to release integrated reports. Preliminarily, a cluster analysis is used to group countries according to institutional patterns. Multivariate analyses detect the associations between report release choice and analysts ’ forecast accuracyacrossclusters.Multipleeconometricapproachesareusedtoaddresstheendogeneityconcerns. Findings – IR release is not informative for the market unless considering systematic variations across different institutional settings. Analysts ’ forecast is more accurate for IR adopters located in strong institutional enforcement settings than for all the other companies. In the strong institutional setting that is also characterized by a pluralistic society, IR release bene ﬁ ts for the market are conditioned by the fact that the choice to release IR depends on environmental, governance and social disclosure-based managers remuneration and disclosure requirements.Inweakinstitutionalsettings,IRreleaseisnotbene ﬁ cialfortheforecastaccuracy. Research limitations/implications – Academics and practitioners can gain understanding of the usefulnessof voluntary IR across different institutional settings. Originality/value – The study advances the understanding of the IR ’ s informativeness, overcoming the common dichotomous distinctions between strongandweak institutional settings.


Introduction
Integrated reporting (IR) represents a prominent initiative in the context of corporate reporting, receiving increasing attention by accounting scholars and practitioners (de Villiers et al., 2014Rinaldi et al., 2018). IR goes beyond traditional financial reporting in its explicit attention to non-traditional capitals (e.g. human capital, and social and relationship capital), strategy and business models and its emphasis on forward-looking information (Eccles and Krzus, 2014;de Villiers and Sharma, 2020). By combining financial and nonfinancial information in a single document, IR is intended to benefit a wide range of stakeholders (de Villiers and Maroun, 2017). However, financial stakeholders (e.g. investors and analysts) are expected to be the primary recipients of IR disclosure, as explicitly stated by the International Integrated Reporting Council (IIRC) in its framework (International Integrated Reporting Council, 2013;Flower, 2015;Rowbottom and Locke, 2016), and confirmed by report preparers, who underline the report's usefulness for equity and debt investors, analysts and rating agencies (Higgins et al., 2014;Lai et al., 2017Lai et al., , 2018. Some scholars are more sceptical and question the report's usefulness, even for financial stakeholders (Humphrey et al., 2017;Hsiao and Kelly, 2018;Slack and Tsalavoutas, 2018).
In line with the need for further research on IR "impacts" (Rinaldi et al., 2018), a number of studies have started exploring whether and how IR influences investors and other financial stakeholders, with a predominance of studies focused on the context where IR is mandatorily adopted (i.e. South Africa). These studies unambiguously demonstrate a positive impact of mandatory IR on financial stakeholders in terms of analysts' behaviour, and value relevance of IR disclosure (Barth, 2017;Zhou et al., 2017;Bernardi and Stark, 2018;Wang et al., 2020;Caglio et al., 2020). In contrast, the less numerous studies that have focused on contexts where IR is voluntarily released show sparse results. While Kim et al. (2017) find that IR adoption decreases the post-level of forecast dispersion compared with the pre-level of forecast dispersion, and Flores et al. (2019) show that IR adoption improves analysts' ability to make accurate earnings forecasts, Wahl et al. (2020) find no significant effect of IR publication on analyst earnings forecast accuracy. These results suggest further exploration within voluntary contexts, also considering that such contexts are not necessarily homogeneous in terms of country-level institutional characteristics, which are expected to affect determinants and consequences of nonfinancial disclosures (de Villiers and Marques, 2016).
To shed some light on the impact of voluntary release of integrated reports on financial stakeholders in diverse institutional settings, we explore the association between report release and analyst forecasts by considering the moderating role of country-level institutional characteristics, in line with the general argument that the disclosure's usefulness for the capital market is mediated by cross-country differences (Hope, 2003;Vanstraelen et al., 2003;Dhaliwal et al., 2012;Lee, 2017). While some studies have shown the influence of country-level institutional factors on the decision to voluntarily release an integrated report (Frías-Aceituno et al., 2014;Lai et al., 2016;Vaz et al., 2016;García-S anchez et al., 2019;Girella et al., 2019), very little attention has been paid to how these factors moderate the report's impact on financial stakeholders (Flores et al., 2019).
The impact of integrated report release on analyst behaviour is explored with reference to the forecast accuracy, which captures the preciseness of the analysts' predictions by comparing the companies that release reports with those that do not. While the present paper does not estimate the causal effect of integrated report release over the financial analysts' forecast, it aims to assess if accuracy of predictions systematically varies with report release in diverse institutional settings. To inspect the relation between integrated report release and analysts' behaviour, bringing out the moderating role of the institutional setting, we conduct an empirical quantitative analysis over three years. We identify companies releasing an integrated report by relying on the IR Examples Database provided by the IIRC, and then match a control sample of companies that do not release reports. We recognize that examining IR in a voluntary setting presents challenges in terms of selfselection as firms that expect to gain the most benefits from disclosing IR are more likely to MEDAR adopt it. Therefore, we use multiple econometric approachesincluding propensity score matching (PSM) and the Heckman two-stage model. Furthermore, we acknowledge that using an international sample might give rise to endogeneity bias by omitted correlated variables. Therefore, we use the two-stage least squares (2SLS) instrumental variable to address omitted variable concerns (Cui et al., 2018).
Among the different perspectives available for studying corporate discretionary disclosure strategies (Merkl-Davies and Brennan, 2007), we draw on voluntary disclosure theory (Healy and Palepu, 2001) and related literature on analysts' behaviour in different country-specific institutional settings (Lang and Lundholm, 1996;Dhaliwal et al., 2012) to hypothesize a negative association between integrated report release and analysts' forecast error (used as an inverse measure of forecast accuracy) [1]. We also hypothesize that the country-level institutional setting moderates the association between integrated report release and analysts' forecast error. We find evidence supporting both hypotheses.
This paper offers unique contributions by exploring the moderating role of the country-level institutional setting on the association between integrated report release and analysts' behaviour. Indeed, by pooling companies according to clusters that group countries with similar institutional characteristics, the paper disentangles the magnitude of the informativeness of report release across countries, thus offering a more nuanced partition than the distinction between shareholder and stockholder orientations considered in prior studies (Flores et al., 2019). We also develop an international sample including countries worldwide, instead of limiting the comparison to specific geographical areas. We provide straightforward evidence of IR impacts on analysts' behaviour, contrary to prior studies that did not find significant effects of voluntary release of integrated reports on analyst earnings forecast accuracy (Wahl et al., 2020).
The remainder of this paper is organized as follows. Section 2 theoretically frames our study in the light of voluntary disclosure theory, and then presents a literature review of pertinent IR studies and develops the research hypotheses. Section 3 describes the research strategy by presenting the sample and the research methods. Section 4 shows the main findings and the results of additional analyses, including tests addressing endogeneity concerns, while Section 5 discusses the findings in the light of the theoretical framework and prior studies. Section 6 concludes by highlighting the contributions, implications and limitations of the study.
2. Theoretical framework, prior research and hypothesis 2.1 Voluntary disclosure, integrated report release and analysts' forecasts Voluntary disclosure theory suggests that the disclosure of private information to investors enhances efficiency in resource allocation (Healy and Palepu, 2001) because it reduces information asymmetries claimed by agency theory (Verrecchia, 1983). According to the argument that the aim of a firm's disclosure policy is to increase the understanding of the company's performance and future perspectives (Holland, 1998), disclosure exceeding that mandatorily required by the regulation (Meek et al., 1995) is expected to add information that facilitates financial analysts' understanding of the company's performance and prospects (Lang and Lundholm, 1996;Hope, 2003). Acting "as intermediaries who receive and process [. . .] information for investors" (Schipper, 1991, p. 105), analysts help investors in allocating their capital (Nichols, 1989;Schipper, 1991).
Financial analysts' behaviour is observable through their forecasts, the accuracy of which is a proxy of the company's information availability (Lang et al., 2003). The more analysts' forecasts are precise, the more likely are investors able to allocate their capital efficiently, with positive effects on the capital market. The usefulness of voluntary disclosure to financial analysts can be detected by investigating the associations between voluntary disclosure and financial analysts' forecast accuracy. Lang and Lundholm (1996) Integrated reporting and analyst behaviour show that firms with more informative disclosure policies enhance analyst forecast accuracy, reducing the information asymmetry.
The decision to release the integrated report belongs to the overall company's disclosure policy, which is expected to be aimed at increasing the investors' set of information suitable to orient their decision-making processes, unless the costs of revealing information outweigh the benefit of the disclosure (Healy and Palepu, 2001). To the extent that IR is informative about the company's strategy, business model and prospects (de Villiers and Sharma, 2020), analysts' forecast error is likely to decrease when the company releases the report [2]. Indeed, investors and other financial investors (e.g. analysts and rating agencies) appear to be the main recipients not only within the IIRC's framework (International Integrated Reporting Council, 2013) but also in the preparers' view (Brown and Dillard, 2014;Flower, 2015;Thomson, 2015;International Integrated Reporting Council, 2017;Lai et al., 2017Lai et al., , 2018. The studies that have analytically explored the effect of IR on financial stakeholders' behaviour have been predominantly developed in the mandatory context of South Africa, where the South African Johannesburg Stock Exchange incorporated the move towards IR into its listing rules, under its King Code on Corporate Governance. These studies leave little doubt that, in this mandatory context, the integrated report is relevant for financial stakeholders (Barth, 2017;Zhou et al., 2017;Bernardi and Stark, 2018;Wang et al., 2020;Caglio et al., 2020). The less numerous studies that have considered contexts where IR is voluntarily adopted do not exhibit the same straightforward evidence (Kim et al., 2017;Flores et al., 2019;Wahl et al., 2020). However, following voluntary disclosure theory's line of reasoning, we expect that information contained in the integrated report is helpful for analysts in formulating accurate predictions for earnings. Consequently, we state the following directional hypothesis: H1. Voluntary release of the integrated report is negatively associated with analysts' forecast error.

Voluntary integrated report release and analysts' forecasts in diverse institutional settings
With the aim to gather nuanced evidence on the impacts of IR on the capital market, we explore whether the institutional characteristics of the countries in which reports are voluntarily released influence analysts' behaviour. Prior studies investigating voluntary disclosure in an international setting (Belkaoui and Karpik, 1989;Vanstraelen et al., 2003;Lee and Hutchison, 2005;Kothari et al., 2009;Dhaliwal et al., 2012;Lee, 2017) underline that the disclosure's usefulness for analysts is influenced by cross-country differences. Most studies capture such cross-country differences through the legal system (Frías-Aceituno et al., 2014;Simoni et al., 2020). These studies sustain that companies located in civil-law countries are oriented to satisfy stakeholders' information requirements because they are more stakeholder-oriented (Ball et al., 2000;Simnett et al., 2009). According to this approach, in civil-law countries, voluntary disclosure is more informative than the disclosure released by companies located in common-law countries (Dhaliwal et al., 2012). However, the legal system per se is not able to capture all the cross-country differences in voluntary disclosure's usefulness for the capital market, and the institutional factors are potentially more meaningful in explaining cross-sectional variation in the usefulness of corporate disclosure (DeFond and Hung, 2004). In the light of several institutional factors, Leuz et al. (2003) group countries into three different clusters that convey different levels of institutional enforcement. With reference to studies that focus on nonfinancial reporting (Lee and Hutchison, 2005;de Villiers and Marques, 2016), we rely on a broad range of World Bank Institute indicators capturing a variety of country-level institutional characteristics: voice and accountability; government effectiveness; regulatory quality; rule of law; and anti-MEDAR corruption. De Villiers and Marques (2016) demonstrate that these indicators, described in Section 3.2 of our research strategy, are able to influence corporate social responsibility (CSR) disclosure determinants and consequences for the financial stakeholders because they reflect how the institutional context favours companies' engagement with and investors' appreciation of nonfinancial issues. We use these same characteristics at the country level to capture the traits of institutional enforcement that could potentially ground different levels of IR informativeness because of some commonalities that IR shares with CSR reportinga voluntary, nonfinancial form of reporting [3].
Considering that IR is voluntarily adopted in several countries, we use voluntary report release as the empirical setting and inspect whether the different country-level institutional settings moderate the usefulness of the integrated report in analysts' behaviour. While a number of studies have investigated the role of institutional contexts in urging companies to release an integrated report (Frías-Aceituno et al., 2014;Lai et al., 2016;Vaz et al., 2016;García-S anchez et al., 2019;Girella et al., 2019), to the best of the authors' knowledge, no prior research has considered this moderating effect, with the only exception being Flores et al. (2019), who show that the governance regime moderates the impact of integrated report release on analysts' forecast errors, with reports prepared by firms belonging to countries with a shareholder-based governance regime (i.e. North America) having a greater impact on analysts' ability than those prepared within a stakeholder-based governance regime (i.e. Continental Europe). However, Wahl et al. (2020), using an international sample of listed companies, find no significant association between voluntary IR disclosure and analyst earnings forecast error. Therefore, in line with the need to disentangle the role of the institutional context characteristics in driving corporate reporting practices (Chen and Bouvain, 2009;Cordazzo et al., 2017), this study goes beyond Flores et al.'s (2019) dichotomy in two ways: first, it includes countries spread all over the world, and, second, clusters are grouped according to a wide range of country-level institutional factors, which are expected to jointly influence the pressure companies are subjected to when deciding whether to release integrated reports as well as the impact on analysts' behaviour (de Villiers and Marques, 2016).
Literature investigating the benefits of corporate disclosure for the capital market demonstrates that the more the institutional enforcement is strong, the more financial reporting disclosure eases analysts' forecast accuracy (Hope, 2003). Such a relation is controversial with reference to voluntary corporate disclosure. Dhaliwal et al. (2012) show that there is a positive association between standalone CSR reports and analyst forecast accuracy, and it is stronger in countries that are more stakeholder-oriented and in countries with more opaque financial disclosure. With reference to IR, Flores et al. (2019) show that the association between report release and analyst forecast error is stronger in countries that are more shareholder-oriented. While not focused on financial analysts' forecast accuracy, Obeng et al. (2020) have a different view, expecting that in stakeholder-oriented countries, report preparers may embrace the concept of IR more fully, with their more transparent IR leading to lower agency costs. Because our cluster analysis goes beyond the shareholder and stakeholder dichotomy in grasping the patterns connotating the institutional settings, we state the following nondirectional hypothesis: H2. Country institutional setting moderates the association between the integrated report's voluntary release and the analysts' forecast error.

Identification of integrated reporting adopters and sample selection
Similarly to other studies that include in their samples voluntarily released integrated reports (Kim et al., 2017;Melloni et al., 2017;Kiliç and Kuzey, 2018;Flores et al., 2019; Integrated reporting and analyst behaviour Girella et al., 2019), we select the companies whose report is included in the IIRC's IR Examples Database as at 1st January 2017. This database is publicly accessible from the IIRC official website, and we hand-collect from the corporate website of each of the IR adopters a proprietary data set in time-series over the years 2015-2017, including the first year of report release (even before 1 January 2017); the choice to release the report for each year; and the date of report release. Companies located in South Africa, where integrated reports are mandatorily required for listed companies, are excluded. The initial sample includes 181 international companies.
We restrict our analysis over the years 2015-2017 for two reasons. First, because most companies from the base sample started to publish an integrated report between 2013 and 2016, we can capture the usefulness of report release, referring to companies that had already issued at least one integrated report. Second, considering that the IIRC framework was issued in 2013, we aim at capturing the usefulness of the report released according to the framework, starting two years afterwards.
A sample of companies not releasing integrated reports (IR non-adopters) is selected according to the industry sector classification (based on the NAICS code); the geographical location (same country); and the size (based on the volume of assets). We find 226 companies meeting these requirements.
To address concerns that firms self-select into the IR adopters' group, we use PSM and nearest neighbour to form a matched control sample of IR non-adopter firms. For the first stage of PSM, we estimate the probability of being an IR adopter firm by using a probit model that includes firmand country-level variables. Specifically, we model the report release choice as a function of firm total assets, net income, leverage, industry and country's regulatory quality. Using the estimated coefficients from the probit model, we compute the expected probability of being an IR adopter firm for each firm-year observation and use these propensity scores to match the observations of the IR adopters to observations of the IR non-adopters.
For each company entered in the sample, we retrieve annual fundamental data and market data from Eikon. The scores depicting environmental, governance and social disclosure (ESG), board of directors' composition and ownership are retrieved from Bloomberg. The firm-specific data are collected from the Orbis database. The scores about the country institutional characteristics are retrieved from the World Bank database.
After filtering for analyst data and required control variables, and keeping the observations matched by the PSM, the final sample consists of 218 companies (139 IR adopters and 79 IR non-adopters), corresponding to 1,209 company-year observations over three years. Table 1 presents the sample selection and distribution along time. Panel A shows the sample selection; Panel B shows the companies' distribution along time and Panel C the observations' distribution along time.

Cluster analysis variables
We perform a descriptive country cluster analysis, grouping countries showing similar patterns of institutional characteristics. This analysis allows us to capture interactions among different factors and verify systematic patterns in the integrated report's impact on the capital market without relying on directional hypotheses. Drawing on de Villiers and Marques (2016), the cluster analysis is based on the World Bank Institute (2008) indicators at the country level, which capture the following institutional characteristics: voice and accountability; government effectiveness; regulatory quality; rule of law; and anticorruption.

MEDAR
The voice and accountability index (vae) and its standard deviation (vas) indicate the freedom of expression and association, and the related individual's possibility to publicly express concerns on various matters. This possibility also relates to nonfinancial matters that an integrated report is expected to deal with.
The government effectiveness index and its standard deviation capture the quality and the variability of the public services, the civil service, the policy formulation and implementation, and the credibility of the government's commitment to such policies. If governmental services are efficient and independent from political pressures, firms' engagement with nonfinancial issues disclosed by an integrated report is more likely to be genuine.
The regulatory quality index and its standard deviation indicate the ability and its variability of the government to formulate and implement sound policies and regulations that permit and promote private-sector development. The level of regulatory quality indicates the likelihood companies are in a position to deploy fair practices, which are expected to be disclosed by means of an integrated report.
The rule of law measures the extent and its variability to which agents have confidence in and abide by the rules of society, in particular, the quality of contract enforcement, the police and the courts. The likelihood of crime and violence enforcement measure developed by Kaufmann et al. (2010) is used as a proxy for institutional quality as it is meant to capture a country's ability to implement regulation and government policies. In contexts with high rule of law levels, companies are expected to be more expert in dealing with nonfinancial matters that an integrated report is expected to disclose.

Integrated reporting and analyst behaviour
The control of corruption index and its standard deviation indicate the extent and the variability to which public power is able to avoid conflicts of interest and opportunistic behaviours grounded on corruption. In contexts with high control of corruption, companies are expected to adopt transparent behaviour, also in their reporting practices.
The data retrieved from the World Bank database are based on surveys of firms and individuals, as well as the assessments of commercial risk rating agencies, nongovernmental organizations and a number of multilateral aid agencies. We use estimates for indicators that are the result of a maximum likelihood function and therefore are normally distributed. To capture the differences in the patterns across the cluster, we consider the standard deviations of most of the variables. The standard deviations indicate that the clusters do not show different variabilities in the index levels.

Regression models
We estimate ordinary least squares (OLS) regressions to test our hypotheses. The model has been widely applied in prior studies investigating the usefulness of voluntary disclosure for investors (Dhaliwal et al., 2012): Verifying H1, we expect b 1 in equation (1) to be negative and significantly associated with the dependent variable, thus indicating that integrated report release is likely to provide information useful to analysts to predict accurately the company's performance. We adopt hierarchical linear models to examine the moderator effects of the differences across the institutional enforcements. We augment the model with the interactions between the companies belonging to one of the three clusters and their choice to release the report. Specifically, the variable of interest (IR_release) is interacted with each of the three dummy variables (Cluster1, Cluster2 and Cluster3), indicating the belonging of each company to one of the three clusters (Cluster 1, Cluster 2 and Cluster 3). Both the interacted variables are dummies, and the base effects are calculated when both the interacted variables (IR_release and, respectively, Cluster1, Cluster2 and Cluster3) are 0. The interaction effect of interest is calculated when both the interacted variables are equal to 1. The three interacted variables are named, respectively, IR_release_Cluster1, IR_release_Cluster2 and IR_release_Cluster3. Coefficients are computed for all the baseline effects: Equation (2) is estimated to verify H2; we expect b z to be different from 0 and significantly associated with the forecast error, indicating that the report release is differently associated with the forecast error in relation to the cluster belonging. We estimate three b z values, one for each interacted variable (IR_release_Cluster1, IR_release_Cluster2 and IR_release_Cluster3).
We use robust standard errors to account for possible heteroscedasticity of the underlying data set. The standard errors are clustered at the firm level to allow errors of the same firm not to be correlated over time.

Variable definitions
We capture the integrated report's usefulness for the capital market by using financial analysts' forecast accuracy (Dhaliwal et al., 2012;Zhou et al., 2017). The analysts' forecast error (f_error) is used as an inverse measure of forecast accuracy. Forecast error is defined as the average of the absolute errors of all forecasts for target earnings, scaled by the share price at the fiscal year end (Behn et al., 2008): Subscripts i, t and j denote company i, year t and analyst j. The target earnings and forecasts are for the current year. FC is the analysts' earnings forecast for time t, and EPS is the actual earnings per share for time t. P is the price per share at the end of period t. The forecast horizon is limited to one year because of a lack of forecasts available for a longer time horizon. Only forecasts made after the report issue date are used to allow analysts to incorporate the information contained in the integrated report into their forecasts. The analyst forecast error (f_error) is winsorized at 95% to remove the skewness in the data.
The main dependent variable is IR_release, defined as a dummy variable equal to 1 if the company released a report, and 0 otherwise. The variable of interest (IR_release) is interacted with each of the three dummy variables indicating the belonging of the companies to the clusters, generating the variables IR_release_Cluster1, IR_release_Cluster2 and IR_release_Cluster3.
The model includes several control variables. Following Dhaliwal et al. (2011Dhaliwal et al. ( , 2012 and Zhou et al. (2017), the first set of control variables includes firm-level controls capturing specific firm financial features. Prior literature suggests that analysts' forecast accuracy is associated with the number of analysts following the company (Jiao et al., 2012); thus, we control for this factor by including the variable ln_nana, which is the natural logarithm of the number of analysts following each company each year. Jiao et al. (2012) suggest that a firm's performance volatility decreases the informativeness of firm reports, thus decreasing analysts' forecast accuracy. We include the variable l_eps_st, which is computed as the natural logarithm of the earnings per share's standard deviation. Resource constraint is likely to determine the informativeness of nonfinancial disclosure (Dhaliwal et al., 2011); therefore, we control for profitability, which is captured by return on assets (ROA), calculated as the ratio of earnings before interest and taxes (EBIT) to total assets. To control for financial risk that may injure corporate performance and, thus, performance predictability, we include the variable lev, which is measured by debt to asset ratio Caglio et al., 2020). Following Flores et al. (2019), analysts might react differently to companies that issue an IR depending on whether or not they operate in sensitive industries. We control for sensitive industries with the dummy Naics that is equal to 1 when companies are operating in industries where IR release might be driven by the incentive to polish the corporate image for legitimacy purposes, and 0 otherwise.
The second set of control variables refers to firm-level disclosure. As recent evidence (Bernardi and Stark, 2018) documents an association between sustainability disclosures and analysts' forecast accuracy after the introduction of an IR regime, we control for Bloomberg's ESG scores (ESG_env, ESG_gov and ESG_soc) indicating, respectively, the proxies for environment, governance and social performance disclosure. Following Zhou et al. (2017), we also control for the issuance of standalone CSR (or sustainability) reports, with the aim of verifying that information contained in the integrated report is incrementally Integrated reporting and analyst behaviour useful to analysts in addition to other nonfinancial reporting practices. The control variable CSR_report is equal to 1 if the company issues a standalone CSR/sustainability report, and 0 otherwise. Prior evidence shows that analysts' forecast errors decreased after firms mandatorily adopted International Financial Reporting Standards (IFRS) but only in countries with strong enforcement regimes and where domestic accounting standards differed significantly from IFRS (Byard et al., 2011). Therefore, we control for the financial accounting disclosure regime with the dummy variable IFRS, which is equal to 1 if the company prepares the annual report according to the IFRS, and 0 otherwise. Last, analysts might not respond to first-time report issuance because of a firm's lack of experience; thus, we control for the first release of the report by the companies in the sample (IR_first).
The third set of control variables captures corporate governance. Evidence shows that corporate governance is associated with greater voluntary disclosure (Eng and Mak, 2003) and is related to IR (Wang et al., 2020). Following Barth et al. (2017), we control for the board composition with three variables indicating the board size (BoardSize), the percentage of women sitting on the board of directors (W) and the percentage of non-executive board members (Nonex). Further, we control the ownership concentration because prior evidence shows that ownership structure is associated with nonfinancial disclosure: Höllerer (2013) finds a positive association between dispersed ownership and the decision to release standalone non-financial reports. Consistently, Cormier and Magnan (1999) and Cormier et al. (2005) find that concentrated ownership is associated with less environmental disclosures. The variable Controlled is a dummy variable equal to 1 if the company is controlled by concentrated ownership, and 0 otherwise.
The natural logarithm of continuous variables is used to remove the skewness in the data as shown by the skewness/kurtosis tests. Table 2 details the variables definition.

Cluster analysis results
We preliminarily perform a descriptive country cluster analysis, which groups countries with similar institutional characteristics. A k-means cluster analysis with three distinct country clusters is conducted. Three distinct country clusters are identified as a result of the non-hierarchical algorithm, which relies on the level and the variability of the institutional characteristics. The reliability of the clusters has been established performing the analysis multiple times (Ketchen and Shook, 1996), providing consistent outcomes. Table 3 shows the description of the three clusters obtained, detailing the features connotating each of the three clusters. Panel A shows the mean values of each of the institutional characteristics by cluster; Panel B shows the cluster membership of countries and Panel C shows the pervasiveness of integrated report release by cluster. The grouping shown in Panel B is consistent with the regional distinctions used in prior research to classify countries that overcome the simple distinction between common-and code-law traditions (Leuz et al., 2003;de Villiers and Marques, 2016). Cluster 1 includes the Anglo-Saxon countries (UK and USA), along with Hong Kong, which was formerly under British rule and has inherited parts of the institutional framework. Cluster 2 includes Northern Europe countries that are civil-law countries (Germany, Canada, Denmark, Luxembourg, Netherlands and Switzerland), countries under British rule that have common law (Australia and New Zealand) and Canada. Cluster 3 includes countries that have civil law: Brazil, China, Greece, Italy, Republic of South Korea, Namibia and Sri Lanka. Therefore, the legal system is not disentangling countries belonging to Cluster 1 and Cluster 2.
With reference to the institutional setting of the countries included in the three clusters, Cluster 1 is characterized by mid-range score levels of the indicators used to define the MEDAR country-specific institutional characteristics (i.e. voice and accountability, government effectiveness, regulatory quality, rule of law and anti-corruption). In particular, Cluster 1 presents average scores higher than those of the countries included in the cluster with the lowest scores (i.e. Cluster 3), while its main difference from Cluster 2 relates to the score of Dummy variable equal to 1 if the company is located in one of the countries belonging to Cluster 1 (namely, France, Hong Kong, Japan, the United Kingdom and the USA); and 0 otherwise Cluster2 Dummy variable equal to 1 if the company is located in one of the countries belonging to Cluster 2 (namely, Australia, Canada, Denmark, Germany, Luxembourg, The Netherlands, New Zealand and Switzerland); and 0 otherwise Cluster3 Dummy variable equal to 1 if the company is located in one of the countries belonging to Cluster 3 (namely, Brazil, China, Greece, Italy, Namibia, Republic of South Korea and Sri Lanka); and 0 otherwise IR_release_Cluster1 Interaction between IR_release and Cluster1 IR_release_Cluster2 Interaction between IR_release and Cluster2 IR_release_Cluster3 Interaction between IR_release and Cluster3 Firm-level financial controls ln_nana Natural logarithm of the number of analysts following each company each year l_eps_st Natural logarithm of the earnings per share standard deviation Lev Leverage is the ratio of total debt to the sum of total debt and the book value of common shareholders' equity ROA EBIT/total assets ratio Integrated reporting and analyst behaviour the "voice and accountability" indicator, which in Cluster 2 is considerably higher than in Cluster 1. In brief, Cluster 3 includes the countries that show the lowest levels of institutional enforcement, whereas Cluster 1 and Cluster 2 include the countries that show high levels of institutional enforcement. Further, in contrast to the countries in Cluster 1, the countries in Cluster 2 are characterized by high freedom of expression and related possibilities to publicly express concerns on various matters. In sum, Cluster 1 is characterized by strong institutional enforcement, Cluster 2 by strong institutional enforcement and a pluralistic society and Cluster 3 by weak institutional enforcement.

Descriptive analysis
We limit the analysis to the observations matched with the PSM. After matching, we obtain 1,209 firm-year observations related to 139 IR adopters and 79 IR non-adopters (as not all firms have complete data for every year, our panel is unbalanced). Table 4 shows the descriptive statistics. Panel A shows the statistics for the pooled sample; Panel B shows the statistics for the IR adopters and Panel C shows the statistics for the IR non-adopters. Panel D presents the difference in the means of the variables between the two sub-samples (IR adopters and IR non-adopters).  Integrated reporting and analyst behaviour of the companies are located in countries belonging to Cluster 1 (56%). Because the selection of the control sample is originally driven by the geographical location (other than industry and size), also the matched IR non-adopters are mostly located in Cluster 1 (73%).
We examine the difference in means between the matched firms for the variables included in our first-stage model. The difference is not significant for most of the variables.
Forecast error (f_error) is significantly lower in the IR adopters than IR non-adopters. This suggests that financial analysts following companies releasing reports predict the earnings per share with more accuracy compared with companies that do not release reports. However, the difference in the means of f_error between IR adopters and IR nonadopters is not significantly consistent across the clusters. Table 5 presents the simple correlations among the variables. The accuracy of the analysts' forecast is significantly and negatively related to IR_release and Cluster1, while it is positively and significantly related to Cluster2 and Cluster3.
These results provide preliminary evidence supporting our two hypotheses. However, given that the above comparison is univariate, we will turn to the multivariate regressions. Table 6 shows the results from the estimation for the model specifications measuring the analysts' forecast error. The model specifications report the association between the accuracy of the earnings forecasts and the integrated report release.

Multivariate results
Model 1 shows the specifications for equation (1). The coefficient of IR_release is negative and statistically significant (IR_release: À0.008, p-value: 0.00), suggesting that firms that release reports are associated with more forecast accuracy than companies that do not issue reports, providing preliminary indicators to accept H1.     Table 5. Notes: *p < 0.10, **p < 0.05, ***p < 0.01 Integrated reporting and analyst behaviour Models 2, 3 and 4 show the specifications for equation (2), where each interaction term is added individually. The baseline effect indicating the choice to release an integrated report (IR_release) is differently associated with the forecast error across the models. The interactions (IR_release_Cluster1, IR_release_Cluster2 and IR_release_Cluster3) are all significantly associated with the dependent variable, indicating that the cluster belonging plays a moderating role in the relationship between the choice to release a report and forecast accuracy, providing preliminary indicators to accept H2.

MEDAR
Model 2 shows that IR adopters located in Cluster 1 are negatively associated with the forecast error (IR_release_Cluster1: -0.023, p-value: 0.000). The baseline effect of Cluster1 indicates that firms located in countries characterized by strong institutional enforcement (Cluster 1) are negatively and significantly associated with the forecast error (Cluster1: À0.017, p-value: 0.00) regardless they are IR adopters or not. The baseline effect of IR_release is positive and significant (IR_release: 0.007, p-value: 0.004), indicating that the choice to release the IR, per se, is not beneficial for the analysts' forecast accuracy regardless the clusters' belonging.
Model 3 shows that IR adopters located in Cluster 2 are positively and significantly associated with the forecast error (IR_release_Cluster2: 0.014, p-value: 0.000). The baseline effect of Cluster2 indicates that firms located in countries belonging to Cluster 2 are positively and significantly associated with the forecast error (Cluster2: 0.023, p-value: 0.000), while the baseline effect capturing the choice to release IR is not significantly associated with the financial analysts' accuracy (IR_release: À0.003; p-value 0.25).
Model 4 shows that IR adopters located in Cluster 3 are positively and significantly associated with the forecast error (IR_release_Cluster3: À0.019, p-value: 0.000). The baseline effect shows that the choice to issue the IR is, per se, negatively and significantly associated with the forecast error (IR_release: À0.008; p-value: 0.00). The baseline effect of Cluster 3 indicates that firms located in countries belonging to Cluster 3 are negatively and significantly associated with the forecast error (Cluster3: À0.015, p-value: 0.000).
These results suggest that capital market benefits from the integrated report informativeness vary systematically across different institutional enforcement.

Additional analysis
We conduct a series of additional tests to address endogeneity concerns. Indeed, it is likely that the decision to release an integrated report is not random as this decision could be based on expectations of impact on future performance or even other firms' specific unobservable variables.
To address selection bias because of sample selection, we use a Heckman two-stage model (Heckman, 1979). We first develop a probit model predicting the integrated report release choice. The first-stage model includes a variable that satisfies the exclusion restriction. In our case, such a variable is related to the choice of integrated report release, but unrelated to forecast error. We argue that companies are more likely to release integrated reports depending on the investor protection at country level, while it is unlikely to affect the firm-level forecast error. Our first-stage probit model is specified below: where antiself is the measure of the investor protection for firm i in year t (Djankov et al., 2003); antiself is an index that measures the level of legal protection of minority shareholders MEDAR against expropriation by corporate insiders, taking into account vote by mail; shares not blocked or deposited; cumulative voting; oppressed minority; pre-emptive rights; and capital to call meeting. We use antiself as a variable satisfying the exclusion restriction, arguing that the level of investor protection is likely to drive companies' disclosure choice in two alternative directions. On the one hand, the higher the investor protection is, the more the companies are pressured to reduce agency conflict and information asymmetry, choosing to voluntarily release integrated reports. Conversely, high investor protection might indicate that there is no need for further measures to reduce information asymmetries because investors are already assured by extant practices. Either way, investor protection dictates a firm's disclosure practice, and thus is expected to be correlated with integrated report release choice. The other variables included in the first-stage probit are likely determinants of report release choice selected from prior literature (Obeng et al., 2020) and defined in Table 2. The variable lev indicates the leverage ratio and captures the information demand about firms' downside risks (Goss and Roberts, 2011). ROA and Size indicate, respectively, the company's profitability and total assets, capturing the resources available for IR disclosure (Zhou et al., 2017). We include the board size (BoardSize) and the percentage of non-executive directors (Nonex), following the literature (Frías-Aceituno et al., 2014). IFRS captures the adoption of accounting standards in preparing mandatory disclosure that several studies have found to be beneficial for forecast accuracy (Byard et al., 2011). Then, we use the inverse Mills ratio (IMR) estimated from the first-stage model as an additional regressor in equation (2). Table 7 shows the results of the Heckman two-stage model. Panel A presents the model in the first step, where the exclusion restriction variable antiself is negatively and significantly associated with the choice to release an integrated report, consistent with the motivation that the choice to release a report is inversely associated with the investor protection level (antiself: À0.851, p-value: 0.00). This result indicates that the pressure to reduce information asymmetry is significant in explaining the choice to release a report.
Panel B presents the model in the second step, where the coefficients of IR_release_Cluster1, IR_release_Cluster2 and IR_release_Cluster3 remain significantly associated with the dependent variable, with the same sign detected in the main analysis, confirming that the informativeness of IR for the market is systematically different across different institutional settings. The IMR is statistically significant, indicating that there is selection bias. Therefore, the two-step estimation is necessary to deal with the selection bias (Certo et al., 2016). In the second step, the true effect is recovered, and the results confirm our previous findings.
Heckman model performs poorly when other sources of endogeneity exist (Certo et al., 2016). To address other sources of endogeneity, such as omitted variables and simultaneity, we perform a 2SLS instrumental variables approach.
Obtaining appropriate instruments challenges the implementation of statistical approaches for dealing with omitted variables and simultaneity (Lang et al., 2012). We identify an instrumental variable related to the choice to release an integrated report that, at the same time, has not been identified as a determinant of the financial analysts' forecast accuracy. Specifically, we estimate the IR adopter/non-adopter choice as a function of the choice to grant top managers a variable remuneration based on the ESG performance. Companies granting the top managers a bonus linked to ESG performance are likely to be more facilitated in preparing integrated reports, which increases their incentives to release the report. ESG_LINKED_BONUS is a dummy variable equal to 1 if company i in year t grants the top managers a variable bonus linked to ESG performance, and 0 otherwise. Our first-stage model is specified below: Integrated reporting and analyst behaviour Notes: *p < 0.10, **p < 0.05, ***p < 0.01 Table 8 presents the results of the four models. Column 1 shows the first stage where the instrumental variable (ESG_LINKED_BONUS: 0.080; p-value: 0.000) is positively and significantly related to IR_release, consistently with our expectations. This result indicates that the ESG-based remuneration incentives granted to managers is significant in explaining the company's choice to release the IR. Notes: *p < 0.10, **p < 0.05, ***p < 0.01

Integrated reporting and analyst behaviour
Column 2 shows the second stage, which uses the predicted value of IR_release estimated in the first stage. The coefficient of the variables of interest (IR_release: 0.070; p-value: 0.021) in the second stage is not consistent with the main result, indicating that the choice to release the integrated report is not, per se, beneficial for the capital market. Thus, we cannot accept H1. This result supports the idea that the benefit of the IR is not observable unless the different institutional settings are taken into account.
The Cragg Donald Wald F statistic reveals that the first stage estimation is not weak according to the rule of thumb of Staiger and Stock (1997). Because the first stage model includes a single endogenous variable, we perform the refinement to the rule of thumb according to the Stock and Yogo (2005), confirming that the first stage is not weak. The pvalue of the under-identification test is less than significance level; therefore, we reject the null hypothesis of under-identification and we confirm that the model is exactly identified. The overidentification hypothesis is rejected because only one instrument is included.
We believe that the omitted variables and simultaneity problems might affect the crosssectional moderator role of the institutional setting captured in the main analysis by the interaction terms. Therefore, we identify two instrumental variables related to the choice to release an integrated report across different institutional setting that, at the same time, has not been identified as a determinant of the financial analysts' forecast accuracy.
Specifically, we estimate the endogenous interacted variables, respectively; IR_release_Cluster1, IR_release_Cluster2 and IR_release_Cluster3 as a function of the choice to grant top managers a variable remuneration based on the ESG performance (ESG_LINKED_BONUS) and the measure of the country level of disclosure requirements in security offerings (Disc) retrieved from the World Bank database. The variable Disc summarizes the disclosure requirements faced by firms referring to security offerings in a given country. The disclosure requirements in security offerings capture the ability to make it cheaper for investors to recover damages when information about the promoter of a security is wrong or omitted (La Porta et al., 2006). The variable Disc is an index calculated as the average of the following proxies: whether promoters can issue securities without delivering a prospectus describing the securities to potential investors in advance; insiders' compensation; ownership by large shareholders; inside ownership; contracts outside the normal course of business; and transactions with related parties (La Porta et al., 2006). The instrumental variable Disc is expected to capture cross-sectional variation in the disclosure requirements across different institutional settings [4]. We argue that the strength of the disclosure requirement is likely to drive companies' disclosure choice in two alternative directions. On the one hand, a high level of disclosure requirements urges companies to be equipped with corporate systems and data management suitable to supply information to prepare corporate disclosure meeting the requirements. Therefore, the more the disclosure requirements are strict, the more companies are likely to release the IR. Conversely, a demanding disclosure requirement might indicate that companies are already preparing a high standing corporate disclosure to meet the requirements. Therefore, the more the disclosure requirements are strict, the less companies are willing to invest additional effort to voluntary issue the IR. We expect that disclosure requirement drives firm's disclosure choice differently in diverse institutional setting, because the patterns of institutional characteristics are likely to urge companies to be or not willing to invest the extra effort to issue IR. We do not have expectation about the direction where diverse institutional settings patterns are likely to drive the companies' choices. However, the level of disclosure requirement is expected to be correlated with different integrated report release choice in diverse clusters.

MEDAR
As the endogenous variable to be estimated is an interacted term, following Wooldridge (2010), we model the first stage interacting each instrumental variable with, respectively, the dummies: Cluster1, Cluster2 and Cluster3. Our first-stage model is specified below: Table 9 presents the results of the models.
Columns 1, 3 and 5 show the first stage of the model specifications estimating the endogenous interaction, respectively, when IR adopter belongs to each of the three clusters. The first stage of each model's specifications shows that the instrumental variables are significantly related to the endogenous interacted variables.
In the first stage, the instrumental variable ESG_LINKED_BONUS interacted with each dummy variable, indicating the clusters' belonging (ESG_LINKED_BONUS_Cluster1, ESG_LINKED_BONUS_Cluster2, ESG_LINKED_BONUS_Cluster3) is positively and significantly associated with the endogenous variable across the clusters, indicating that the choice to grant manager bonus linked to ESG performance is significantly likely to predict the choice to release IR. This result indicates that the ESG-based managers' incentives are significant in explaining the companies' choice to release a report regardless the cluster belonging.
The instrumental variable Disc interacted with each dummy variable, indicating the clusters' belonging (Disc_Cluster1, Disc_Cluster2, Disc_Cluster3) is positively and significantly associated with the choice to release the IR in Cluster 1 (Disc_Cluster1: 0.649; p-value: 0.000) and in Cluster 3 (Disc_Cluster3: 0.151; p-value: 0.000), indicating that the more tighten the disclosure requirements, the more likely companies are choosing to release the integrated report. Interestingly the opposite phenomenon is observable in Cluster 2 (Disc_Cluster2: À1.995; p-value: 0.000), where the disclosure requirement at a country level is negatively and significantly associated with the likelihood the companies choose to release the integrated report. We interpret this phenomenon in light of the high "regulatory quality" characterizing Cluster 2. The high "regulatory quality" urges companies to prepare mandatory corporate reporting according to high standard of disclosure requirements and this is likely to discourage companies to embrace an additional voluntary reporting. This result indicates that the disclosure requirement is significant in explaining the companies' choice to release a report across different clusters.
The Cragg Donald Wald F statistic reveals that the first stage estimation is not weak according to the rule of thumb of Staiger and Stock (1997). The p-value of the underidentification test is less than significance level; therefore, we reject the null hypothesis of under-identification and we confirm that the model is exactly identified. The Sargan test indicates that overidentification hypothesis of all instruments is rejected.
Columns 3, 6 and 9 show the second stage of each model's specifications, which use the predicted value of IR_release estimated in the first stage.
In the second stage, the signs and the significance of the coefficients are different across the three clusters. Thus, we can assert that in contexts where IR is voluntary, the usefulness of IR in providing financial analysts with information that allows them to forecast the market performance is systematically different across countries with different institutional characteristics, confirming H2.
However, the coefficients of the interactions are not all consistent with prior findings. Specifically, the coefficients of the interactions are consistent with prior findings for Cluster  Notes: *p < 0.10, **p < 0.05, ***p < 0.01 Table 9.
In Cluster 2, the benefits of IR for the forecast accuracy emerge, indicating its informativeness. The latter result (in Column 4 IR_release_Cluster2: À0.019; p-value: 0.009) shows that the IR released is informative for the financial analysts also in strong institutional settings with pluralistic society. However, the IR informativeness depends on the fact that the choice to release IR is conditioned by the ESG-based managers' incentives and the disclosure requirements.
According to the literature (Certo et al., 2016), when both sample-induced and alternately sourced endogeneity exist, the 2SLS models provide less biased estimates than Heckman models. For this reason, the results obtained with the 2SLS models offer the best estimation of the coefficients of our interests. Relying on the results obtained with the 2SLS models, we reject H1 and we accept H2.

Discussion
Voluntary disclosure theory indicates that nonfinancial disclosure complements mandatory disclosure, providing additional information to stakeholders, including financial analysts. Our evidence indicates that IR's positive impact on the financial market is not confirmed unless the different institutional settings are taken into account.
Results support that the institutional settings play a moderation role in explaining the IR's informativeness for the market.
Analysts' forecast is more accurate for IR adopters located in strong institutional enforcement (Cluster 1) than for all the other companies.
In institutional settings characterized by a strong institutional enforcement and a pluralistic society (Cluster 2), the integrated report appears to be not informative for financial analysts providing no benefits to the forecast accuracy. In such a context characterized by the highest institutional enforcement, the high "regulatory quality" urges companies to prepare transparent mandatory corporate disclosure, thus disclosing all the information that financial analysts need to predict earnings. This inference is supported by the additional analysis showing that disclosure requirements discourage companies to release the IR. However, additional analysis shows that for those companies that are likely to choose to release IR, given the ESG-linked managers' incentives and the level of disclosure requirements, the forecasts are more accurate. Therefore, the additional analysis allows us to tease out the strength of the disclosure requirements at the country level and explains the different informativeness observable across strong institutional settings (Cluster 1 and Cluster 2).
Integrated report release is associated with less forecast accuracy in weak institutional enforcement, where companies are faintly urged to adhere to accounting and reporting rules, thus making corporate reporting less reliable. In such a context, analysts' forecasts are more complex and, consequently, more inaccurate (Hope, 2003). The integrated report turns out to be not informative for financial analysts to predict the performance of companies located in weak institutional enforcement.
These results are in line with Flores et al. (2019), whose study shows that the association between integrated report release and analyst forecast error is stronger in countries that are more shareholder-oriented than stakeholder-oriented. However, our results are able to show that institutional settings are essential to grasp IR informativeness for the market.
Our evidence reveals that the detected relations are affected by two endogeneity sources that we dealt with in the light of the literature.
First, a sample-bias endogeneity is found in estimating the probability of choosing to release an integrated report as a function of investor protection, consistent with de Villiers MEDAR and Marques (2016). This indicates that the market endowments to reduce information asymmetries contribute to drive the results obtained. The Heckman two-stage model restores the true effects confirming that cross-sectional variations in the institutional settings are essential to grasp IR informativeness for the market.
Second, endogeneity because of an omitted variable capturing the top managers' remuneration based on ESG performance and the disclosure requirements at the country level is found to be significant. This result indicates that IR potential for the capital market is affected by the fact that managers' incentives are linked to ESG performance, which is consistent with Bernardi and Stark (2018) and the institutional disclosure requirement, which is consistent with Hope (2003).
The explanatory power of the instrumental variables included in our study can be seen as an additional result that allows to infer the reasons clarifying what are the institutional traits that are likely to explain the IR's informativeness. Furthermore, the instrumental variables have not been used in previous evidence, in spite of they capture prior literature insights consistent with the theoretical framework. The inclusion of these terms in our study can be seen as a potential for the use of such variables in future empirical evidence.

Conclusion
Although IR has gained momentum in the tumultuous arena of nonfinancial reporting, empirical research on IR "impacts" (Rinaldi et al., 2018) is still limited and dominated by analyses focused on the context where IR is mandatorily adopted (i.e. South Africa). The less numerous studies that have focused on contexts where IR is voluntarily adopted show sparse results about the impact of IR on financial analyst behaviour. The variety of these results seems to be caused by the specific (voluntary) contexts explored. However, although the institutional setting is shown to influence the voluntary decision to release the integrated report (Frías-Aceituno et al., 2014;Lai et al., 2016;Vaz et al., 2016;García-S anchez et al., 2019;Girella et al., 2019), few studies consider the institutional setting as capable of influencing financial analyst behaviour. Obeng et al. (2020) capture the IR benefit for the capital market, disentangling shareholder from stockholder orientations, while Flores et al. (2019) provide unique research that considers the role of these country-level orientations in moderating the relationship between (voluntary) integrated report release and financial analysts' forecasts.
Starting from this premise, the present paper contributes to IR research on the voluntary adoption of this innovative form of reporting in an international setting (Flores et al., 2019;Obeng et al., 2020) by showing that voluntarily released integrated report informativeness is not uniform across countries.
In settings with a strong institutional enforcement (Cluster 1), IR adopters' performance are predicted by the financial analysts more accurately than all the other companies (IR nonadopters or IR adopters located elsewhere). In institutional settings characterized by a strong institutional enforcement and a pluralistic society (Cluster 2), the IR is not informative for the market. However, this finding is biased by endogeneity concerns of omitted variables. Once dealing with this bias, for those companies that are likely to choose to release IR given the ESG-linked managers' incentives and the level of disclosure requirements, the forecasts are more accurate. Therefore, the additional analysis allows us to tease out the strength of the disclosure requirements at the country level and explains the different informativeness observable across strong institutional settings. In countries with weak institutional enforcement (Cluster 3), IR is not beneficial for analysts' accuracy.
Our evidence calls for future research investigating the integrated report's usefulness for the capital market to take country-level institutional characteristics into systematic consideration. In particular, it highlights the importance of relying on specific institutional patterns, rather Integrated reporting and analyst behaviour than single country-level institutional characteristics, such as civil versus common law and the related stakeholder versus shareholder approach (Flores et al., 2019;Obeng et al., 2020).
More generally, the paper contributes to the research on IR "impacts" (Rinaldi et al., 2018) by unveiling the potential of IR for financial stakeholders. While some scholars have questioned the role of IR for stakeholders, including financial effects, the research shows that financial analysts rely on IR in predicting earnings per share. The research findings offer a comforting answer to the question of IR's usefulness for financial stakeholders (International Integrated Reporting Council, 2013;Higgins et al., 2014;Flower, 2015;Rowbottom and Locke, 2016;Lai et al., 2017Lai et al., , 2018, on the condition that the institutional setting urges report preparers to address investors' information needs and these needs are not satisfied by different sources of information.
Further, the paper contributes to the literature on voluntary disclosure and analyst forecasts by showing that the usefulness of voluntary nonfinancial disclosure claimed by voluntary disclosure theory varies according to the institutional setting. Our results show that IR's complementary role relative to mandatory corporate disclosure benefits forecast analysts' accuracy only where there is a strong institutional enforcement and once the ESGlinked managers' incentives and the strength of the disclosure requirements are considered.
The paper also has implications for practitioners in their understanding of the role played by IR for the capital market in diverse institutional settings. In particular, our findings offer insights for policymakers, standard setters and regulators involved in enhancing the quality and enforcement of nonfinancial disclosures to stimulate organizations to behave responsibly (Lai and Stacchezzini, 2021). While we are not in a position to comment on the impact of mandatory integrated report release, with some scholars recommending a rapid transition to mandatory nonfinancial reporting (Adams, 2020), our evidence should discourage the promotion of IR as a voluntary disclosure in contexts of weak institutional enforcement. Managers of companies operating in countries where IR is voluntarily prepared may find our findings interesting if they wish to initiate the voluntary release of reports, reasoning on the extent to which their IR may facilitate analysts' forecasts in relation to the institutional context in which they operate. This paper is not exempt from limitations. First, we built our base sample by considering companies that have published their reports in the IR Examples Database, thus not taking into consideration companies that have realized an integrated report not included in this database. We are aware of possible self-selection bias because of the absence of other searchable databases. However, this sample selection approach is similar to other studies that included in their samples voluntarily released integrated reports (Kim et al., 2017;Melloni et al., 2017;Kiliç and Kuzey, 2018;Flores et al., 2019;Girella et al., 2019). Second, we did not control our findings for the quality of the report. However, other studies have searched for impacts of both report release and report quality on analyst forecast accuracy, finding convergent results (Kim et al., 2017;Zhou et al., 2017). Future studies could mitigate the limitations of this study by expanding the base sample to integrated reports not published in the IR Examples Database and controlling for the quality of the report. Future research may also expand the timeframe of the analysis to inspect whether report release is increasingly capturing analysts' attention.
Notes 1. The detection of a negative association would support the voluntary disclosure theory perspective but also lead to reject the competing perspective of impression management (Merkl-Davies and Brennan, 2007), which suggests that IR discretionary disclosure manages users'