Does Integrated Reporting Matter to the Capital Market?

Information Quality and Analysts' Earnings Forecasts

As outlined above, this study examines whether the different levels of alignment with the <IR> framework in companies' integrated reports arising from managements' disclosure discretions matters to the capital market. Owing to the discretionary nature of the disclosures at this early stage of the journey, we employ voluntary disclosure theory in developing our hypothesis.

Voluntary disclosure theory asserts that voluntary disclosures help to improve the information environment of companies by enhancing analysts' understanding of companies' prospects (e.g., Beyer et al., 2010). Following the theory, empirical studies generally document a positive relationship between disclosure quality and analysts' earnings forecasting properties, such as lower forecast error and dispersion (Barron et al., 1999; Barth et al., 2001; Bradshaw et al., 2008; Hope, 2003; Lang and Lundholm, 1993; Plumlee, 2003). The general argument from these studies is that better disclosure quality enhances analysts' understanding of the company's performance and future outlook and helps analysts interpret the disclosures in an informed and similar manner, which in turn results in an improved forecast accuracy and a lower forecast dispersion (Hope, 2003; Lang and Lundholm, 1996). Moreover, better disclosures lower the costs of processing and interpreting the disclosures, and thus enhance analysts' ability to correctly incorporate all pertinent information, which in turn results in improved earnings forecasts (Lehavy et al., 2011).

Evidence from prior literature also reveals that analysts use non-financial information in their earnings forecasting (Dhaliwal et al., 2012; Nichols and Wieland, 2009; Orens and Lybaert, 2007; Simpson, 2010). The supply of non-financial information is beneficial in reducing analyst earnings forecast error and dispersion (Dhaliwal et al., 2011, 2012; Nichols and Wieland, 2009) and analysts issue more optimistic recommendations for companies with higher CSR ratings (Ioannou and Serafeim, 2015). In fact, considerable evidence shows that analysts use financial information and non-financial information interactively in their earnings forecasting (Coram et al., 2011; García-Meca and Martinez, 2007; Ghosh and Wu, 2012; Maines et al., 2002; Orens and Lybaert, 2010; Pflugrath et al., 2011; Simpson, 2010). This means that both types of information are integral to analysts' earnings forecasting.

Theoretically, the ability of analysts to forecast earnings should improve with the amount of disclosure (regardless of whether disclosures are financial or non-financial) as long as they help analysts to assess companies' future performance. However, analysts are known to have cognitive limitations in information processing and the complexity of the task is found to adversely affect analyst earnings forecast error and dispersion. For example, analysts' forecasts have been found to be less accurate if they are associated with complex changes to tax laws (Plumlee, 2003), and more complex accounting choices can negatively affect forecast accuracy and increase dispersion (Bradshaw et al., 2008). Further, although analysts exert extra effort in generating reports when analyzing companies that produce less readable 10-K filings, these less readable 10-Ks are still associated with greater dispersion, lower accuracy, and greater overall uncertainty in analyst earnings forecasts (Lehavy et al., 2011).

Thus, if such problems exist for analysts relying on complex financial information, then adding non-financial information into their decision-making processes could exacerbate the adverse effects. This is especially the case when the correlation between non-financial information and financial information is not well articulated, which adds significantly to the total task complexity and thus the issue of ‘overload’ for analysts. Empirical evidence shows that analysts tend to underreact to information in non-financial measures (e.g., customer acquisition cost, average revenue per user) even though those measures can have significant predictive ability for future earnings (Rajgopal et al., 2003; Simpson, 2010).

Furthermore, the relevance of the information format has also been well documented in psychology and accounting studies, revealing that the way information is systematically presented affects the way people think about that information. Informationally equivalent disclosures that vary only in their ease of processing can have differential effects on market prices (Hopkins, 1996; Hodge, 2001; Hodge et al., 2006; Kelton et al., 2010; Koonce and Mercer, 2005). The insights from these studies suggest that even if <IR> is simply an improved re-formatting of information currently required, it could still beneficially impact upon users' information processing, including sophisticated users like financial analysts. By incorporating material non-financial information with financial information into one report, <IR> can increase the salience of those material non-financial information items, which users may ignore if they are reported separately (Agnew and Szykman, 2005; Hirshleifer and Teoh, 2003).

To summarize, disclosure theories and associated empirical evidence demonstrate that both financial and non-financial information have the potential to improve analysts' forecasting abilities if they are value-relevant and/or they help to reduce information acquisition and processing costs. Studies using psychology theories suggest that analysts' earnings forecasting abilities could be hindered by voluminous reporting and by their failure to fully incorporate non-financial information into their decision making.

Information Quality and the Cost of Equity Capital

Theoretical studies have established both direct and indirect links through which financial information can affect the cost of equity capital. The direct links include risk sharing (Merton, 1987) and the reduction of estimation/information risk (Barry and Brown, 1984, 1985; Brown, 1979; Coles et al., 1995). This line of research suggests that providing better information will reduce the estimation risk and therefore the cost of capital. The indirect links include the effect on market liquidity and information asymmetry (Baiman and Verrecchia, 1996; Diamond and Verrecchia, 1991; Easley and O'Hara, 2004; Verrecchia, 2001) and the effect on companies' real decisions (Lambert et al., 2007).

Despite theoretical conjecture on the negative link between discretionary disclosures and the cost of equity capital, the empirical evidence on the link is less consistent and robust for a definitive and unambiguous conclusion to be drawn (Beyer et al., 2010; Botosan, 2006; Core, 2001; Healy and Palepu, 2001; Kothari, 2001; Leuz and Wysocki, 2008). However, overall, the empirical studies generally provide support for the theoretical negative link between the level/quality of discretionary disclosures and the cost of equity capital, although the results appear to be sensitive to many factors such as the presence of market intermediaries (Botosan, 1997; Griffin and Sun, 2014), the types and frequency of disclosures (Botosan and Plumlee, 2002; Kothari et al., 2009), missing control variables such as earnings quality (Francis et al., 2008), and countries' institutional factors (Chen et al., 2009; Francis et al., 2005; Hail and Leuz, 2006, 2009).

With the increasing trend of supplementing financial information disclosure with non-financial information disclosure, research has extended voluntary disclosure theory to non-financial information, with the assumption that this non-financial information is value-relevant (Margolis et al., 2009; Margolis and Walsh, 2003; Orlitzky et al., 2003). Following this line of thought, a number of studies have examined the impact of CSR disclosures/performance on the cost of equity capital and have generally documented a negative relationship 6 (Dhaliwal et al., 2011, 2013; El Ghoul et al., 2011; Plumlee et al., 2015). While these recent studies on CSR disclosures/performance yield useful insights into how non-financial information affects investor behaviour, they suffer from some limitations. First, these studies generally examine the issuance status of the CSR report (issued or not issued) without further distinguishing between the difference in the quality of those reports. Second, while the issuance of a CSR report signals progress toward expanding current corporate reporting to include environmental, social, and governance issues, there are criticisms that these reports are not easily related to a company's strategy and business model, and are thereby less effective in communicating the company's performance to investors (Eccles and Krzus, 2010; Serafeim, 2014).

Despite ample evidence suggesting an interconnected relationship between non-financial information and financial information, most existing research examines them in isolation. Indeed, this propensity reflects the current bifurcated state of corporate reporting. The concept of <IR> aims to advance and enhance corporate reporting by placing emphasis on the interconnections between different types of information currently reported in separate strands. <IR> is still an emerging phenomenon so therefore empirical research about it is both recent and sparse. Among the limited empirical studies on <IR>, Serafeim (2014) provides evidence on the value of this form of reporting by examining the investor base of companies that practice <IR>. Using the data on Thompson Reuters Asset4 database to proxy for <IR>, Serafeim notes an association between <IR> and the investor clientele in terms of the investment horizon, but he also notes that it is not clear how investors change capital allocation decisions based on the information within integrated reports. Furthermore, as the data from the Asset4 database do not provide links to the content elements of integrated reports, Serafeim is not able to determine which elements of <IR> are most effective in attracting long-term investors.

Our study advances the literature by examining two direct outcomes of user decisions, that is, the impact on analysts' earnings forecasting tasks and the change in reporting companies' cost of equity capital. Further, examining the level of alignment of companies' integrated reports with the <IR> framework allows for a more in-depth analysis of the relationship between the disclosure quality of the integrated report and the associated economic outcomes.

<IR> and Analyst Earnings Forecast Accuracy and Dispersion

The preceding literature review suggests analysts' earnings forecasting ability could be improved in three ways: (i) by having new value-relevant information; (ii) by having better disclosed information, which reduces information acquisition costs, and/or a cognitive effect in processing and interpreting the information; and (iii) by having better presented information, which facilitates the incorporation of all relevant information into the user decision-making process.

<IR> places emphasis on narrating companies' value-creation stories during the short, medium, and long term by providing information relating to corporate strategy, business model, and future outlook (IIRC, 2013a). The long-term focus of <IR> corresponds well to the information demands of analysts with their long-term earnings forecasting horizon. In addition to current corporate reporting requirements, the integrated report can also provide new information, which is supported by the findings from an investor survey that was undertaken, and is described in the methodology section. Thus, the integrated report is expected to provide both new and useful information to analysts for their earnings forecasting tasks.

The principles of <IR> mean that it provides more than additional value-relevant information. In particular, the key underlying principles of <IR> are materiality, conciseness, and connectivity (IIRC, 2013a). The materiality principle helps to de-clutter the report by including only substantial matters affecting a company's value-creation ability. This in turn reduces the costs of information acquisition and processing, thereby relieving information overload faced by analysts. The conciseness principle stresses the need for cross-referencing between elements of the report and shifting detailed standard information to other platforms/documents. This helps to reduce the cognitive effort exerted by analysts in analyzing and interpreting the information. The connectivity principle means that the relationships among key elements included in the report are explicitly and clearly presented and articulated. This is not only useful in enhancing analysts' ability to incorporate all value-relevant information into their decision making, but also in easing analysts' information analyzing processes. In sum, the combined aim of these <IR> principles is to alleviate the information overload problem encountered by users by de-cluttering the report and highlighting the connections between different parts of the reports.

Accordingly, the integrated report not only potentially contains new value-relevant information which is helpful in assessing the long-term prospects of companies, but also information that is easier to process, thus enabling analysts to incorporate all pertinent information. In this way, analysts can have a better understanding of the company's performance and future outlook and thus make improved forecasts.

However, the effectiveness of <IR> is dependent on the quality of the integrated report, and more specifically on how well the prescribed <IR> principles are followed. In this sense, integrated reports that are more closely aligned with the <IR> principles are expected to be more helpful to analysts.

This discussion leads to the first set of hypotheses to be tested in this study:

Following on from the previous discussion, if the integrated report contains useful value-relevant information, which is easy to understand and interpret, the reduced level of ambiguity will enhance the consensus among analysts during their earnings forecasting tasks. This effect is hypothesized as follows:

<IR> and the Cost of Equity Capital

Collectively, theoretical arguments support the notion that <IR> could help reduce the cost of equity capital if the principles of <IR> are adequately followed, which is formally hypothesized as follows:

It is recognized that companies' annual reports are not the only avenue through which companies disseminate information to stakeholders. Companies use other channels such as websites, conference calls, management forecasts, and media to communicate with information users. Nonetheless, market intermediaries, such as analysts, play an important role in analyzing, processing, and disseminating information about companies. The theoretical model of risk sharing noted by Merton (1987) suggests that disclosures by lesser known companies can make investors aware of their existence and therefore enlarge their investor base, which in turn improves risk sharing and lowers the cost of capital. In this sense, such an effect is likely to be less relevant for large companies, which are likely to have advanced information-sharing mechanisms such as a larger analyst and investor following. Voluntary disclosure theory also posits that the benefit of additional disclosure depends on the company's information environment (Diamond and Verrecchia, 1991; Lambert et al., 2007; Verrecchia, 1983). This is particularly relevant to our study as the disclosures in annual reports may be more important for some companies than others.

This asymmetric effect has been identified empirically (e.g., Botosan, 1997; Griffin and Sun, 2014). Our study is similar to Botosan (1997) in that we use disclosure in annual reports as a proxy for companies' overall disclosure quality. Thus, the effect of <IR> on the cost of equity capital could differ between companies with different information environments. This is formally hypothesized as follows:

Sample and Data

The sample selection process started with all listed companies on the JSE with fiscal years ending in 2009 to 2012 that are also included on Global Compustat. This list of companies was then merged with corresponding analyst data from I/B/E/S. The annual fundamental data, market data, and exchange rate data were obtained from Global Compustat. All data were downloaded from Wharton Research Data Services (WRDS).

After filtering for analyst data and required control variables, the final sample consisted of 443 company-year observations over four years (132 unique companies) for analyst forecast error 7 and dispersion analysis and 430 company-year observations (130 unique companies) for cost of equity capital analysis.

Research Model

Hypotheses 1(a) and 1(b) are tested by estimating the following OLS regression:

(1)

(2)

The models follow those used in Dhaliwal et al. (2012) with the addition of the issuance status of a standalone CSR report as a control variable, which was examined in Dhaliwal et al. (2012) and found to help reduce analyst forecast error. All variables are analyzed in their changes form 8 rather than levels form to address the potential endogeneity concern (Dhaliwal et al., 2011), as this approach controls for unobserved company characteristics (whether constant or time-variant) which might be correlated with analyst forecast properties. Further, to ensure that the results are not driven by the potential endogenous relation between changes in <IR> and ICC, all independent variables are lagged by one year.

Hypothesis 2 is tested by estimating the following OLS regression:

(3)

Equation 3 is developed from Dhaliwal et al. (2011) and similarly, all variables are analyzed in their changes form and a lead–lag approach is used in the model to ameliorate endogeneity concerns. To test Hypothesis 2(a) on the role of the information environment in the relationship between <IR> and ICC, the sample companies are partitioned into high versus low analyst sub-samples based on the sample median number of analysts following the company. The regression in equation 3 is run separately for the two sub-samples.

Variable Definition–Dependent Variables

Analyst forecast error (FCERROR) Consistent with Dhaliwal et al. (2012), the analyst forecast 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 made in the 12 months after the fiscal year end of the integrated report, scaled by the share price 9 at the fiscal year end:

Subscripts i, t, and j denote company i, year t, and forecast j, respectively. Indicator Y takes three values—zero, one and two—to denote whether the target earnings and forecasts are for current year, one-year ahead, or two-years ahead. 10 FC is the analyst earnings forecast for time t, and EPS is the actual earnings per share for time t. For consistency, both FC and EPS are obtained from the I/B/E/S database. The forecast horizon is restricted to a maximum of two years for the same reason noted in Dhaliwal et al. (2012): analysts typically do not make forecasts for periods beyond the third fiscal year. Further, as the sample period of this study is from 2009 to 2012, the sample size significantly decreases for three-year-ahead forecasts. Only forecasts made after the fiscal year-end month (1 to 12) 11 are used to allow analysts to incorporate the information contained in the integrated reports into their forecasts.

The natural logarithm of analyst forecast error (FERROR) is used in the regression to remove the skewness in the data as shown by the Skewness/Kurtosis tests, and the histogram suggests that the analyst forecast error is significantly right skewed. This approach is similar to that adopted in Lehavy et al. (2011).

Independent Variable of Interest

Control Variables

The control variables follow Dhaliwal et al. (2011, 2012) and Francis et al. (2008). The definitions of these control variables can be found in Appendix 1. All continuous variables are winsorized at the 1 and 99 percentile to deal with potential outliers. 16

Sample for Cost of Equity Capital Analysis

Table 1, Panel C provides summary statistics for the cost of equity capital analysis. The average ICC for JSE listed companies during the period 2009 to 2012 is 0.137, which is slightly higher than that of US listed companies (which is usually around 0.11, see Easton and Monahan, 2005), but is closer to the results for South African companies in international studies (i.e., 0.16, see Hail and Leuz, 2006). Overall, the cost of capital shows a moderate downward trend across the sample years. Similar to the results in Panel A, the average total score for integrated reports is 6.283 and displays a slight increase from 2009 (3.479) to 2010 (4.733), but almost doubles in the year 2011 (7.493), reflecting the effect of mandatory JSE regulation on <IR> adoption by companies.

Hypothesis 2(a) hypothesizes that the negative relationship between the level of alignment of reports and the cost of equity capital will be more significant among companies with a low analyst following. Panel D of Table 1 displays a mean comparison between companies with a high (low) analyst following. 17 The results show that companies with a high analyst following have a lower ICC (t = 4.619) and higher <IR> disclosure scores (t = 11.196); are larger in size (t = 13.803); have a lower book-to-market ratio (t = 5.806); are more highly leveraged (t = 2.679); have a lower growth rate (t = 2.816); and are more likely to issue standalone CSR reports (t = 3.550).

Panel E of Table 1 presents the summary statistics for the yearly change in all variables used for the cost of equity capital analysis. Owing to the use of the changes form, the sample size reduces from 430 to 294. On average, there is a decrease in the ICC of 0.8% (∆ICC_PEG) and an increase in the integrated report disclosure score of nearly 2 (∆IR_TOTAL) across the sample years. When the difference between companies with differing levels of analyst following 18 is examined, there are no significant differences between the two groups except for the changes in the integrated report disclosure score (∆IR_TOTAL, t = 4.213, p < 0.000) and issue of a standalone CSR report (∆CSR, t = 1.982, p < 0.048). This shows that companies with a larger analyst following improved in their disclosure scores. This is consistent with the findings from Graham et al. (2005), that companies with a large analyst coverage view reputation for transparent reporting as an important motivation for voluntary disclosure. Meanwhile, companies with a larger analyst following issue fewer standalone CSR reports, which is likely due to their perception that the practice of <IR> is sufficient (thus rendering the standalone CSR report to be less relevant).

The Pearson correlations (untabulated) suggest a negative relationship between analyst forecast error (FCERROR) and disclosure scores (IR_TOTAL) (−0.196, p < 0.000). Similarly, a negative relationship is identified between analyst forecast dispersion (FDISP) and disclosure scores (−0.099, p < 0.038), although with a much smaller magnitude. A negative relationship between the disclosure score of integrated reports (IR_TOTAL) and the cost of equity capital measures is also identified (−0.161, p < 0.0008). Other variables all have expected signs and none of the unexpected correlations are high enough to raise potential multicollinearity concerns.

Empirical Results

Sensitivity Analysis

Alternative Measures of the Cost of Equity Capital

In this section, we report results using several alternative measures of the cost of equity capital: first, the GLS model developed from Gebhardt et al. (2001), second, the CT model developed from Claus and Thomas (2001), and third, the simultaneous estimation method developed from Easton et al. (2002). All three alternative measures produce results consistent with the main analysis, albeit not statistically significant. The insignificant results from using these three alternative measures could potentially be due to the different assumptions that go into these measures, and perhaps more importantly, because ICC_PEG is much more parsimonious in calculation compared to any of the other measures, suffers fewer measurement errors, and is recommended as one of the superior models in Botosan and Plumlee (2005).

As a second type of alternative measure, the realized (ex-post) return (calculated as the 12-month monthly market-adjusted cum-dividend returns calculated from daily security prices after fiscal year-end) is used as the dependent variable to replace the implied (ex-ante) cost of equity capital. One of the most commonly used measures to evaluate the validity of accounting-based estimates of the ICC is to test their association with realized returns (Easton and Monahan, 2005; Guay et al., 2011). Further, the ex-post realized return has the advantage of suffering fewer measurement errors than the implied (ex-ante) cost of equity capital. Thus, the realized returns are commonly used as an additional analysis to see if the results corroborate those from using the ex-ante implied return, that is, the implied cost of equity capital (Francis et al., 2008). The regression results (untabulated) reveal that companies with higher <IR> scores are associated with a lower return, consistent with the expectation that these companies are regarded as having lower risks for investors, and are therefore willing to invest with lower returns. This is supported by the findings in Jones et al. (2007), that higher levels of sustainability disclosure are associated with negative abnormal returns, and in Hong and Kacperczyk (2009), that sin stocks (alcohol, tobacco, and gaming) have higher expected returns because they are neglected by norm-constrained investors and face greater litigation risk. The sub-sample analysis also produces similar results to the main analysis in that the negative relationship between the disclosure score and the monthly return is only significant among those with a low analyst following. Therefore, the result on the realized return corroborates those found on the ICC in the main analysis, thus providing further support for Hypotheses 2 and 2a.

Alternative Measures of Analyst Earnings Forecast Error and Dispersion

The measure of forecast error and dispersion in the main analysis is consistent with previous studies such as Lang and Lundholm (1996), Hope (2003), Lehavy et al. (2011) and Dhaliwal et al. (2012). However, Cheong and Thomas (2011) challenged the routinely followed way of measuring analyst forecast error on scale (usually by share price or actual/forecast EPS). Their reasoning was that both analyst forecast error and dispersion vary little with scale for EPS forecasts in a number of large markets around the world. In view of the findings by Cheong and Thomas (2011), we used the unscaled/raw analyst forecast error as an alternative dependent variable in the sensitivity analysis.

The results for Hypotheses 1(a) and 1(b) hold, supporting the conclusion that improvement in the disclosure score of the integrated report helps improve analyst forecast accuracy but not forecast dispersion.

Alternative Measures of the Level of Alignment of Integrated Reports with the <IR> Framework

In the main analysis, the level of alignment of integrated reports with the <IR> framework is measured using the total disclosure score of those reports ranging from 0−31. Although the use of a continuous score is deemed appropriate and necessary given the significant variations in the level of alignment of integrated reports produced, a potential downside is that such a measure is noisy given the limitations of a user-defined coding framework and the inevitable subjectivity that goes into the coding process. To address the aforementioned limitation in the total disclosure score (IR_TOTAL), a dichotomous variable (HIGHIR) is created which equals one if the disclosure score on the integrated report produced by the company is greater than the sample median disclosure score and zero if it is not. Although the level of variation is restricted in the use of the dichotomous variable (HIGHIR), it is believed to have the benefit of being less noisy. The use of the dichotomous variable produces qualitatively similar results to those obtained from the main analysis, except when testing Hypothesis 1(b) on the relationship between the level of integrativeness and analyst earnings forecast dispersion. In the latter case a significant and negative relationship was identified (t = −2.084, p < 0.05) as opposed to insignificant results using the continuous disclosure score (IR_TOTAL), lending some support for Hypothesis 1(b).

Another limitation of the total disclosure score (IR_TOTAL) is the potential bias derived from the check-list approach used in the scoring process. In order to circumvent such a limitation in the total disclosure score (IR_TOTAL), an impression score (IR_IMPR) was also collected during the coding process. The coder was required to score the report from 0–7 based on his/her overall impression of the report against the principles of an integrated report set out by the IIRC. Thus, the impression score (IR_IMPR) is inherently more subjective than the total disclosure score (IR_TOTAL) but suffers less bias due to not employing the check-list approach of coding. The use of the impression score (IR_IMPR) as the independent variable of interest produces results qualitatively similar to those obtained from the main analysis.

Additional Analyses

Current Year and Two-year-ahead Analyst Earnings Forecast

In the main analysis, only the one-year-ahead analyst earnings forecast properties are used as the dependent variable. However, as an additional analysis, the current year and two-year-ahead forecast error and dispersion are also calculated and regressed on the disclosure score of sample companies. Results (untabulated) demonstrate that the disclosure scores of companies' integrated reports are not significantly related to either the current year or the two-year-ahead analyst forecast error and dispersion.