Has the introduction of IFRS improved accounting quality? A

Transcription

1 Has the introduction of IFRS improved accounting quality? A comparative study of five countries Corresponding author: Andreas Jansson, Assistant Professor, PhD, School of Business and Economics, Linnaeus University, Växjö, Sweden andreas.jansson@lnu.se Phone: Fax: Micael Jönsson, Research Assistant, School of Business and Economics, Linnaeus University, Växjö, Sweden Christopher von Koch, Assistant Professor, PhD, School of Business and Economics, Linnaeus University, Växjö, Sweden 1

2 Has the introduction of IFRS improved accounting quality? A comparative study of five countries Abstract This paper investigates whether the implementation of International Financial Reporting Standards (IFRS) has increased accounting quality. Previous research has primarily explored the effects of IFRS on accounting quality as measured through the use of value relevance, timely loss recognition and earnings management. In contrast, this paper employs a measure of accounting quality that is based on the use of accounting information, namely, the performance of financial analysts. The sample encompasses nearly 2,500 publicly traded firms, all followed by analysts, from The sample covers five European countries (Sweden, Netherlands, France, Germany and the United Kingdom (the UK)), each with different legal and accounting traditions. We use quantile regressions to estimate the impact of IFRS while simultaneously considering that most prediction errors are small and are most likely random and unaffected by the accounting standard being followed. Our results suggest that IFRS have had no effect on analysts average ability to accurately forecast firms earnings per share. In all countries except the UK, IFRS have led to higher consistency in analyst forecasts. The impact of IFRS is not more pronounced in firms that are more affected by their asset measurement methods. The results suggest that in countries where prior GAAP differ from IFRS, IFRS may have the effect of presenting more consistent but not more accurate pictures of firms for analysts. Keywords: IFRS, accounting quality, analyst forecasts, comparative study 2

3 1. Introduction With regulation EC No 1606/2002, the European Union (EU) decided that all publicly traded companies shall prepare their consolidated accounts in conformity with the international accounting standards [IAS] (EU, 2002: Article 4) for each financial year beginning on or after 1 January More specifically, this requirement means that these companies must apply IAS, International Financial Reporting Standards (IFRS) and Standing Interpretations Committee/International Financial Reporting Standards Interpretation Committee (SIC/IFRSIC) interpretations issued by the International Accounting Standards Board (IASB). This requirement most likely constitutes the single most significant change in accounting standards to have ever occurred in Europe and is popularly referred to as the introduction of IFRS (for a comprehensive description of the implementation of IFRS, see, for example, Armstrong et al. 2010). In this paper, we empirically examine the impact of the introduction of IFRS on the accuracy and dispersion of financial analysts forecasts in five EU countries. The past decade has seen a large amount of empirical research regarding what constitutes high-quality accounting (see Soderstrom and Sun, 2007 for a review). For many European countries, the introduction of IFRS has entailed substantial changes in accounting methods, and this change has prompted a major natural opportunity to examine factors thought to affect accounting quality. Consistently, academics around the world are now extensively studying the effects of IFRS on accounting quality (see, for example, Armstrong et al., 2010; Ball, 2006; Barth et al., 2008; Bartov et al., 2005; Byard et al., 2011; Daske and Gebhardt, 2006; Daske et al., 2008; Ding et al., 2007; Hung and Subramanyam, 2007; Jeanjean and Stolowy, 2008; Jiao et al., 2012). Results from these studies are mixed. On the one hand, IFRS appear to have a positive effect on accounting quality, but the results are contingent on country- or firm-specific characteristics. In general, IFRS require more extensive and 3

4 sophisticated disclosures than were afforded by prior local standards, and this requirement may have a positive influence on the quality of financial reports. On the other hand, certain aspects of IFRS, such as the greater flexibility in choice of accounting methods that it offers in comparison with some EU countries previous local standards, may be negatively affecting accounting quality (e.g., Ormrod and Taylor, 2004). Previous research has commonly used earnings management, timely loss recognition and value relevance as indicators of accounting quality (cf. Barth et al., 2008), although metrics such as quality indices and appropriateness also appear. All of these metrics fail to directly capture the usefulness of the information to accounting users. In this paper, we approach the effect of IFRS on accounting quality from a different angle by determining whether the introduction of IFRS has allowed users of accounting information to make better predictions regarding firm performance. More specifically, we examine whether the introduction of IFRS has allowed financial analysts to formulate better forecasts of firm performance. This decision usefulness dimension of accounting quality is in line with a purpose of accounting that is stressed in the conceptual framework of the IASB, which states that The objective of general purpose financial reporting is to provide financial information about the reporting entity that is useful to existing and potential investors, lenders and other creditors in making decisions about providing resources to the entity. Those decisions involve buying, selling or holding equity and debt instruments, and providing or settling loans and other forms of credit (IASB, 2010: OB2). Financial analysts are often argued to fulfill an important function in financial markets by reducing the information asymmetry between firms and investors through their intermediary 4

5 role (Lang and Lundholm, 1996). However, financial analysts ability to perform their task is contingent on the information shared between themselves and the firm not being too asymmetrical (cf. Krishnaswami and Subramaniam, 1999). Available evidence suggests that financial analysts rely extensively on accounting information to make forecasts (Block, 1999; Roger and Grant, 1997); analysts are also considered to be sophisticated users of accounting information (e.g., Schipper, 1991). Prior research has focused primarily on the role of voluntary disclosure in reducing this information asymmetry (Barry and Brown, 1985; Glosten and Milgrom, 1985; Lang and Lundholm, 1996; Merton, 1987), but the introduction of IFRS also allows for the exploration of impacts that mandatory financial accounting may have on analysts ability to reduce information asymmetry. Higher quality financial accounting could be expected to allow extensive users of accounting data to formulate higher quality assessments of firms, leading to less overall information asymmetry. Because the impact of IFRS on accounting quality appears to vary among countries, it is unlikely to have the same effects on analysts performance in all countries where it is introduced. Barth et al. (2008), Byard et al. (2011), Daske et al. (2008) and Preiato et al. (2010), for example, suggest that the enforcement of accounting standards, which can vary among countries (La Porta et al., 1998), is pivotal for the realization of quality increases through the introduction of IFRS. Therefore, in this study, we examine the impact of IFRS on accounting quality in five EU countries: Sweden, the UK, Germany, France and the Netherlands. These countries have regulatory systems with different origins and with varying enforcement strength (La Porta et al., 1998; Preiato et al., 2010), which is reflected in their varying financial accounting traditions (Nobes, 1983). Our paper is related to a number of previous studies. Ashbaugh and Pincus (2001), who examined 80 non-us firms that voluntarily adopted IAS during the period, found 5

6 that analysts forecast accuracy increases after firms adopt IAS and that the convergence in firms accounting policies that is achieved by adopting IAS is positively associated with a reduction in analysts forecast errors. Some early studies have examined the effects of mandatory IFRS adoption on analysts performance. For example, Byard et al. (2011) found that forecast errors and forecast dispersion decrease, but only in countries with both strong enforcement regimes and domestic accounting standards that differ significantly from IFRS. Similar conclusions are suggested in Horton et al. (2012) and Preiato et al. (2010) who demonstrate that IFRS can have a positive impact on forecast accuracy if enforcement is strong. Consistent with this notion, Tan et al. (2011) demonstrate that the accuracy of foreign financial analysts forecasts increases when IFRS are implemented, particularly if the difference between previous local Generally Accepted Accounting Principles (GAAP) and IFRS is significant; however, domestic analysts accuracy is not affected by IFRS. Beuselinck et al. (2010) and Yang (2010) examine the impact of IFRS on private and public information precision. They found a positive association between IFRS and public information precision, which is consistent with an increase in accounting quality. However, Beuselinck et al. (2010) suggest that the effect is more significant in countries in which IFRS entail a significant change, whereas Yang (2010) suggests that the improvement is more significant in countries that already had high-level disclosure standards. Glaum et al. (forthcoming), who examined the impact of IFRS on analysts following Germans firms, separated the effect between improved disclosure and other changes in the firms information environments. They found that improved disclosure had a modest and positive effect on analysts accuracy but suggest that improvements in the quality of earnings, improvements in firms investor relations and changes in analyst behavior have contributed more to the improvement of analyst accuracy. Overall, the literature suggests that IFRS have a positive but not uniform impact. 6

7 Our results, which are based on a sample comprising 2,447 public companies from 5 EU countries during the period of , suggest that IFRS have no impact on forecast accuracy but appear to reduce forecast dispersion. However, the effect of this dispersion is small. In fact, the effect on dispersion is nonexistent for the UK, the country that exhibits the smallest difference between previous GAAP and IFRS. Furthermore, the effect is only visible for the median part of the distribution of forecast dispersions; for large as well as small dispersions, there is no significant effect. Overall, our results suggest that IFRS appear to provide a more consistent but not more accurate picture of firms, a conclusion that is also strengthened by the finding that these effects appear to not be driven by IFRS ability to better represent the underlying economic value of a firm. Four aspects of our approach distinguish our study from prior empirical studies on the effects of IFRS. First, we contribute to the literature using a sample of forecast accuracy and forecast dispersion that encompasses an extensive time period. As Preiato et al. (2010) suggest, analyst performance differs significantly over time for reasons that are unrelated to accounting regulation. It is therefore important to include both periods of generally strong and periods of generally poor analyst performance from both the pre and post IFRS adoption period to better isolate the effects of IFRS. Second, to analyze our dataset, we use an estimation technique a quantile regression model that is robust to the problem of skewness. As Yang (2010) argues, skewness, which can bias estimates, is a serious problem with this type of data. Yang (2010) suggests that a median regression model could remedy this problem. The use of a median regression model allows us to estimate coefficients without manipulating the dataset, which would not be possible with an OLS regression model of estimation. Third, our estimation technique also allows us to predict the effect of independent variables on different magnitudes of forecast errors and forecast dispersions. This approach allows us to detect the effect of 7

8 IFRS on large and small errors without assuming the same impact across the entire distribution. It is reasonable to assume that the impact of IFRS is not uniform in this respect because most small errors or dispersions are likely to be random and therefore impossible to eliminate through the use of improved accounting. Fourth, we use a different proxy for the firm-level impact of IFRS. Although prior studies have used industry (e.g., Beuselinck et al., 2010) and differences in reported earnings in reconciliation accounts (e.g., Horton et al., 2012) as proxies for the firm-level impact of IFRS, we use an accounting figure that is highly affected by IFRS as a proxy: intangible assets. Intangible assets, which are difficult for an analyst to valuate because they are typically unique, are regularly valued at fair value (i.e., market value or a proxy thereof) in IFRS; traditionally, these assets have been valued at historical cost. This method allows us to generate new evidence on what aspect of IFRS might affect analyst performance. Our results have a number of theoretical and practical implications. The results indicate the need to distinguish between two aspects of accounting quality: how it affects users accuracy and how it affects users consistency. Accounting quality measures should be developed to accommodate this distinction. The fact that IFRS primarily affect consistency implies that fair-value appraisals of asset values are used by analysts when predicting firm performance and that IFRS have therefore created a more level playing field despite their lack of a significant effect on predictive value. 2. Hypotheses development 2.1 IFRS and forecast accuracy Our approach relies on the assumption that higher quality accounting will be reflected in higher quality forecasts by financial analysts. Revsine et al. (2004) and Schipper (1991) argue 8

9 that analysts are considered to be among the most important and influential users of financial reports and among the most important information intermediaries between firms and investors. Considering their information-processing ability and access to resources, analysts are typically viewed as sophisticated users of accounting information and as being less likely (than naïve investors) to misunderstand the implications of such information (e.g., Schipper, 1991). Therefore, if financial analysts receive access to higher quality financial reports, they should be able to make better predictions, measured as higher forecast accuracy. Lang and Lundholm (1996) provide evidence of the relationship between disclosure and decreased information asymmetry. Firms with more informative disclosure policies enjoy a larger analyst following, more accurate earnings forecasts, less dispersion among forecasts, and less volatility in forecast revisions. Firms that provide firm-specific information, in particular, are associated with more accurate earnings forecasts and less forecast dispersion. Soderstrom and Sun (2007) argue that the accounting standard being followed affects accounting quality. This relationship implies that the introduction of a new accounting standard should affect the accounting quality of a firm on the margin. The introduction of IFRS has necessitated an accounting standard change for most EU-member countries and, in turn, this change should be reflected in accounting quality. In general, the change to IFRS has entailed a shift toward more valuation of assets according to fair value instead of historical cost. This shift should mean that IFRS provide a better picture of the underlying economic value for firms in the EU because changes in the value of assets generally will be accounted for on a regular basis. However, at the same time, fair value accounting is likely to provide managers with more discretion in accounting, which might diminish the quality of accounting because of increased earnings management (Ormrod and Taylor, 2004). 9

10 Empirical research (e.g., Armstrong et al., 2010; Ball, 2006; Barth et al., 2008; Bartov et al., 2005; Byard et al., 2011; Daske and Gebhardt, 2006; Daske et al., 2008; Ding et al., 2007; Hung and Subramanyam, 2007; Jeanjean and Stolowy, 2008) has attempted to compare the quality measured primarily in terms of value relevance, earnings management and timely loss recognition of IFRS and that of previous standards. The overall quality assessment is not conclusive (cf. Soderstrom and Sun, 2007), perhaps because of the two opposing effects of IFRS (Ormrod and Taylor, 2004). Nevertheless, in terms of supporting high-quality decision making by financial analysts, it is reasonable to assume that a better representation of underlying economic value will outweigh the negative effects of an increase in management discretion. Thus, our first hypothesis is as follows: H1: The introduction of IFRS increases the accuracy of analysts earnings forecasts 2.2 IFRS and forecast dispersion Forecast dispersion (measured as the standard deviation in analysts forecasts), which is an indication of the extent of analysts disagreement regarding a firm s upcoming earnings, can be used as a proxy for investor uncertainty prior to the release of key information (Ramnath et al., 2008).According to Krishnaswami and Subramaniam (1999), this dispersion is a measure of information asymmetry. They claim that when information asymmetry between a firm and its market is high, it is difficult for the market to evaluate or predict the firm s performance. This difficulty increases firm uncertainty. We assume that higher quality accounting will decrease information asymmetry and therefore decrease firm uncertainty, leading to less forecast dispersion. With their stronger orientation toward fair value accounting, IFRS are likely to provide a better representation of a firm s underlying economic value and should therefore decrease information asymmetry. However, 10

11 increased opportunities for managerial discretion could create a degree of uncertainty. The effect of IFRS on forecast dispersion may also depend on whether analysts use more public or private information (Heflin et al., 2003; Irani and Karamanou, 2003). If public information is the primary source used, there should be less dispersion because public information is available to all. However, if analysts seek to gain advantage by gathering private information in response to an increase in public information, policies for the improvement of accounting information may increase dispersion. It could also be the case that analysts choose to herd when earnings are more uncertain, leading to less forecast dispersion for firms with less predictable earnings (Ramnath et al., 2008). Using the BKLS model, Yang (2010) empirically tests whether analysts use more public or private information after IFRS adoption in 18 countries. Yang concludes that both public and private information increase after mandatory IFRS adoption and that the overall effect is a decrease in forecast dispersion among analysts. Interestingly, public and private information increase more in common law countries than in civil law countries, indicating that dispersion can increase in some countries while decreasing in others. Because available evidence suggests that financial analysts use financial reports as a primary source (Block, 1999; Roger and Grant, 1997) of firm information and because it is likely that a better representation of the underlying economic value will outweigh the negative effects of an increase in management discretion, our second hypothesis is as follows: H2: The introduction of IFRS reduces the dispersion of analysts earnings forecasts 2.3 The relative impact of IFRS The last decade has seen the emergence of considerable research discussing the influence of legal and institutional settings on accounting quality (e.g., Byard et al., 2011; Soderstrom and 11

12 Sun, 2007). In several cases, the research is based on the assumption that these settings have a significant impact on accounting quality, which, in turn, affects analysts ability to make accurate forecasts. Other studies have also emphasized firm-specific characteristics, as well as the importance of reporting firms incentives as salient factors in the success of IFRS. Byard et al. (2011) and Jeanjean and Stolowy (2008), for example, stress the importance of firms reporting incentives, which are influenced by legal institutions, various market forces, firms operating characteristics and the like. In line with Zeff (2007), it is therefore reasonable to assume that country-specific differences such as business and financial culture, accounting culture, auditing culture and regulatory culture are likely to affect the success of IFRS implementation. We should therefore expect the introduction of IFRS to have different effects on accounting quality in different legal and institutional settings. For instance, one variable aspect is the strength of accounting standard enforcement (Preiato et al., 2010), which influences managerial discretion. Studies by Francis et al. (2003) and Hope (2003), among others, found that common law countries (i.e., the UK and Ireland) have stronger enforcement mechanisms than code law countries (i.e., the rest of the EU). The international accounting literature also indicates that accounting quality is higher in countries with a common law origin (Ali and Hwang, 2000; Ball et al., 2000; Leuz et al., 2003), and enforcement mechanisms appear to influence the expected quality of financial reporting under IFRS (Ball et al., 2000; Ball, 2006; Barth et al., 2008; Byard et al., 2011; Dao, 2005; Daske et al., 2008). Barth et al. (2008) conclude that the potential benefits of the introduction of IFRS are difficult to attain without the existence of effective enforcement mechanisms (cf. Byard et al., 2011; Preiato et al., 2010). 12

13 Another important factor is the quality of the accounting standard previously used in a given country. If the standard was of low quality, the positive effects of changing to a standard that better reflects a firm s underlying economic value would be expected to be more significant (Byard et al., 2011). The magnitude of differences between IFRS and the local GAAP that they have replaced varies considerably (Bae et al., 2008; Nobes, 1983), and it is therefore reasonable to expect IFRS effect on analyst performance to vary among countries. Thus, our third set of hypotheses is as follows: H3a: The larger the difference between IFRS and the previous GAAP, the more the introduction of IFRS increases forecast accuracy H3b: The stronger the enforcement of accounting standards, the more the introduction of IFRS increases forecast accuracy H4a: The larger the difference between IFRS and the previous GAAP, the more the introduction of IFRS decreases forecast dispersion H4b: The stronger the enforcement of accounting standards, the more the introduction of IFRS decreases forecast dispersion 2.4 The moderating effect of intangible assets IFRS imply a higher degree of fair value accounting. In particular, IFRS systematically employ fair value accounting for classes of assets such as intangible assets, financial instruments, investment property and biological assets. Therefore, the potential improvement triggered by IFRS in terms of the correspondence between firms accounting and underlying economic value is likely to be more sizeable for firms with a higher proportion of such assets. In the absence of fair value accounting data, intangible assets are particularly difficult for 13

14 analysts to value because they are often unique and, as such, lack official market values. Therefore, it is reasonable to assume that the higher the proportion of a firm s intangible assets, the more analysts forecast accuracy will improve and forecast dispersion diminish following the introduction of IFRS. Our fifth and sixth hypotheses are therefore as follows: H5: The higher the proportion of intangible assets, the more the introduction of IFRS increases the accuracy of analysts earnings forecasts H6: The higher the proportion of intangible assets, the more the introduction of IFRS reduces the dispersion of analysts earnings forecasts 3. Method and sample 3.1 Sample Our sample consists of 2,447 publicly traded companies from five European countries. The countries were chosen primarily because of differences in the countries accounting histories (Nobes, 1983). Our third and fourth hypotheses identify the two relevant dimensions as being the difference between previous GAAP and IFRS and the enforcement of accounting standards. Table 1 reports measures of these two dimensions pertaining to the five countries in the sample Insert Table 1 about here

15 The measure of difference between previous GAAP and IFRS (henceforth GAAP-difference) is provided by Bae et al. (2008), who report an index measuring conformity with IAS consisting of 21 specific items in The higher the number, the larger the difference is, the maximum value being 21. The measure is likely to overstate the change resulting from mandatory IFRS adoption because many local standard setters began to harmonize local GAAP with IAS before the mandatory adoption period. However, the measure is likely to capture the difference between IFRS and the more long-term, prior accounting tradition in each country, which, according to Kvaal and Nobes (2012), have a persistent influence on accounting despite IFRS adoption. We also report a measure of accounting standard enforcement developed by Preiato et al. (2010), which is based on 19 items and a scale ranging from 0 to 27, capturing effects of the strength of the audit function and the accounting enforcement body, respectively. The table reports the measure for the year 2005, when IFRS were made mandatory in our sample countries. Our sample countries display variation on the dimensions of GAAP-difference and enforcement. The UK and the Netherlands exhibit the smallest GAAP-difference, whereas France, Germany and Sweden display larger differences. The UK is regarded as having strong enforcement. According to Preiato et al. (2010), Sweden and the Netherlands have relatively weak enforcement, whereas France and Germany have stronger enforcement. Together, these results suggest four clusters: (i) France and Germany, which display both a relatively significant GAAP-difference and relatively strong enforcement; (ii) the UK, which displays a negligible GAAP-difference but strong enforcement; (iii) Sweden, which exhibits a sizeable GAAP-difference and weak enforcement; and (iv) the Netherlands, which displays a small GAAP-difference and weak enforcement. This clustering would suggest that based on hypotheses three and four, we could expect the strongest impact on analyst performance to 15

16 occur in France and Germany. Conversely, the effect could be expected to be negligible in the Netherlands. Sweden and the UK appear to be opposites. Therefore, if GAAP-difference is more important to IFRS impact than enforcement, we would expect the effect on analyst performance to be more pronounced in Sweden than in the UK and vice versa. 3.2 Variables and descriptive statistics Analysts performance is usually measured in the financial literature in terms of forecast accuracy and/or the correctness of stock recommendation or price target, depending on how the final output is viewed. Schipper (1991) discusses the reasonable belief that analysts earnings forecasts should relate to their stock recommendations, which suggests that forecasts and valuation estimates (relative to current price) should be positively related to stock recommendations. However, research into this area does not support these conjectures. Bradshaw (2004) demonstrates that recommendations and valuation estimates in the US are either insignificantly or negatively related, depending on the specification. EPS predictions are likely to suffer from less bias and better reflect analysts use of accounting information and, therefore, accounting quality. To make a forecast, an analyst processes much information from a firm; the degree of accuracy is therefore highly related to the level of informational asymmetry between the analyst and the firm. Accordingly, forecast accuracy is selected as our first performance variable. Because the literature also suggests that forecast dispersion could be viewed as a measure of information asymmetry (Krshnaswami and Subramaniam 1999), forecast dispersion is our second performance variable. In accordance with Lang and Lundholm (1996), the first dependent variable, forecast accuracy, is calculated as the negative of the absolute value of the actual earnings minus the 16

17 analyst s earnings forecast, scaled by the stock price at the beginning of the year, and forecasted EPS t is the mean analyst forecast of the earning per share during period t. ActualEPSt ForecastedEPS Forecast Accuracy = t Stock price at the beginning of the fiscal year Forecast accuracy is defined as the negative of the scaled absolute forecast error. In other words, more accurate forecasts are represented by higher (less negative) values, i.e., lower forecast error, with zero representing a perfect forecast. Analysts usually forecast the earnings per share (EPS) of a particular fiscal year several times before the actual figures are released. The frequency of the forecasts differs in accordance with the analyst. The Institutional Brokers Estimate System (I/B/E/S) collects forecast data from individual analysts around the world once a month and uses those data to calculate statistics such as the mean, median, and standard deviation. Only the final estimates of the analysts are included in the monthly calculation. Thus, the I/B/E/S database provides calculated statistics of analysts EPS forecasts once a month. In this study, we utilize the general methodology for collecting forecast data (see, for example, Lang and Lundholm, 1996) using the final calculated mean of an analyst s EPS forecasts before the first quarterly EPS report is released. For example, for a firm with a fiscal yearend of December 31, 2009, we use the mean forecast calculated in March 2009 as the forecast data for the actual EPS on December 31, The second dependent variable, forecast dispersion, is the inter-analyst standard deviation of forecasts, scaled by the stock price at the beginning of the year, also in line with Lang and Lundholm (1996). Standard deviations are always positive numbers, but we have changed the sign so that the logic in use for the scale of forecast dispersion is the same as that in use for forecast accuracy, i.e. so that a lower (more negative) forecast dispersion represents a higher standard deviation. 17

18 Because the forecast measures are scaled with stock price, cross-company comparisons are possible. To test our hypotheses, we use two test variables, accounting standard followed and interaction term, which is the interaction between the standard followed and the proportion of intangible assets, and six control variables. Accounting standard followed is a dummy variable for which 1 is used for IFRS and 0 is used for every other accounting standard. This variable is a firm-level variable, which means that early IFRS adopters are identified as IFRS users even though this usage is not mandatory. The variable is set to 1 one year after a firm s implementation of IFRS because analysts could not have based their adoption-year predictions on IFRS accounting, as the EPS predictions we use were formulated before the first IFRS-based quarterly report. We obtain the interaction term by multiplying the dummy variable by the proportion of intangible assets to total assets. The six control variables (see Table 2), selected on the basis of prior research into factors that normally affect analysts performance (Lang and Lundholm 1996), are as follows: number of analysts, market value, trading volume, earnings surprise, profit/loss, and standard deviation of return on equity (std ROE) Insert Table 2 about here The number of analysts is determined by simply a count of those following the company and providing earnings forecast, again in line with Lang and Lundholm (1996). We control for firm size using market value and trading volume. Firm size is used in the literature as a proxy for several factors. Size should reflect information availability and therefore be positively related to forecast accuracy. Brennan and Hughes (1991) also found empirical evidence 18

19 between firm size and analysts following a firm, and Lang and Lundholm (1993) found that firm size and performance variability likely correlate with disclosure policy. Market value is measured as the company s market value at the beginning of the fiscal year and is commonly used to control for size. However, we also utilize trading volume as a control for size because it may be more indicative of the number of analysts following a firm, as analysts are often paid indirectly through trading activity. Trading volume refers to the company s absolute daily trading volume during the first month of the fiscal year. Earnings surprise, which is the variation in a firm s results from one year to another, is calculated as the absolute value of the year s earnings per share minus the previous year s earnings per share, scaled by the share price at the beginning of the fiscal year. EPS t is the earnings per share during period t (of a given year), and EPS t-1 is the earnings per share during period t-1 (the previous year). EPS 1 EarningsSurprise = t EPS t Stock price at the beginning of the fiscal year According to Lang and Lundholm (1996), earnings surprise controls for the likely effect that major events, such as a firm's introduction of a new product, have on forecasts. In these circumstances, realized earnings are most apt to deviate from expected earnings, and it is likely that analysts will not be able to make accurate forecasts. Hope (2003) suggests that it is much more difficult to predict future earnings for firms with negative earnings. We therefore use a control variable, loss, a dummy variable that has a value of 1 if the company reported a loss and 0 otherwise. King et al. (1990) found that the number of analysts following firms is likely to be related to variations in return. Fewer analysts follow firms that experience significant fluctuations in profitability. In other words, a negative relationship exists between the number of analysts and variations in profitability. 19

20 Thus, standard deviation of return on equity is the final control variable in our regressions, and it is measured as the company s return on equity over the previous three years. Table 3 contains descriptive statistics for the sample. Within each country, we select all publicly traded companies that were followed by analysts. We do not attempt to follow individual analysts because we assume that the ability to predict values for a specific firm improves over time and therefore might bias the results. From this population, we then extracted those companies with at least one year of both IFRS reporting and non-ifrs reporting. Persuaded by Byard et al. s (2011) argument regarding the necessity of examining the impact of IFRS on analysts forecasts over a longer time period, we chose the period of However, it should be noted that because some companies existed during this entire time period and others for only a couple of years, our sample is an unbalanced panel data set. From our full sample, we obtained 17,449 valid observations of forecast accuracy and 15,003 valid observations of forecast dispersion. We are missing observations of forecast dispersion because those firms followed by only one analyst cannot display forecast dispersion. Forecast accuracy is calculated as a mean of all analysts predictions for a specific firm (the number of analysts ranges from 1 to 49). As Table 3 shows, the mean forecast error for the full sample is approximately 6.3 percent, with the worst analyst performance occurring in Germany and the best in the UK. The mean forecast dispersion is highest in Germany and lowest in the UK Insert Table 3 about here

21 An inspection of the data on forecast accuracy and forecast dispersion reveals that these two variables are highly skewed and therefore not normally distributed. An inspection of kurtosis values suggests the same conclusion. For the full sample, the skewness is for forecast accuracy and for forecast dispersion. These values are obtained because most of the observed forecast errors and dispersions are small. The picture is similar for individual countries. It is therefore more rational to focus on median values, which are provided in Table 4, which also contains significance tests to determine whether pre and post IFRS medians differ from each other. We observe a significant decrease in forecast accuracy post IFRS in the full sample and in Germany and a significant increase in Sweden and the UK. Forecast dispersion exhibits a more consistent pattern, with significant decreases in Sweden and France Insert Table 4 about here Models and estimation techniques We estimate the following two equations: Q (Forecastaccuracy ) = α + β Number of analysts + β Market value + β Trading volume θ it 1 it 2 it 3 it + β Profit/Loss + β Earnings surprise + β Std dev ROE + β Accounting standard followed 4 it 5 it 6 it 7 it + β Intangible/Total Assets + β Interaction term + ε 8 it 9 it it (1) 21

22 Q (Forecast dispersion ) = α + β Number of analysts + β Market value + β Trading volume θ it 1 it 2 it 3 it + β Profit/Los s + β Earnings surprise + β Std dev ROE + β Accounting standard followed 4 it 5 it 6 it 7 it + β Intangible/Total assets + β Interaction term + ε 8 it 9 it it (2) Equation 1 estimates the effect of our two test variables (accounting standard followed and interaction term) and the six control variables for forecast accuracy. Equation 2 estimates the effect of these same variables on forecast dispersion. We estimate both models for the entire sample, as well as for individual countries. Because of the skewness in our dependent variables, ordinary least squares (OLS) regression models, fixed and random effects models, and other regression models yielding estimates that predict the conditional mean of the dependent variable risk the obtainment of biased estimation results. We therefore use a median regression model, namely, the quantile regression estimation technique (Koenker and Baset, 1978), which, because it is a regression model that does not produce estimates that predict the mean, is robust against this type of problem. Another benefit of median regression models is their lack of sensitivity to outliers, which allows for the use of all observations in estimations. The quantile regression model is a technique for estimating the θth quantile (i.e., percentile) of a variable (in this case, the dependent variables forecast accuracy and forecast dispersion), conditional on the values of the predictor variables. This method allows us to estimate the effect of the variable accounting standard followed in various percentiles of the distribution of the variables forecast accuracy and forecast dispersion, not only at their respective means. In other words, quantile regression enables us to estimate the effect of a changing accounting standard on the size of θth quantile forecast errors and dispersions while mitigating the effects of skewness produced by the many smaller forecast errors and dispersions. In this study, we estimate quantile regression models 22

23 predicting the 10th, 50th (i.e., the median) and 90th percentile for the full sample, as well as for individual countries. Thus, when predicting the 10th percentile, we study the effect of IFRS on only the largest forecast errors or highest dispersions. In addition to statistical arguments, theoretical arguments justify the use of this procedure. Small forecast errors and small forecast dispersions are likely to be random and independent of poor-quality financial reporting. Hence, one could argue that whatever steps are taken to introduce improvements, it is most likely not possible to eliminate small errors. 4. Results Table 5 provides the correlations of all variables. The two dependent variables, forecast accuracy and forecast dispersion, are significantly correlated, indicating a relationship whereby an increase in forecast accuracy correlates with a decrease in forecast dispersion. Therefore, when analysts reach greater consensus, their EPS forecasts become more accurate. None of these variables is significantly correlated with accounting standard followed or the interaction term. Accounting standard followed and the interaction term correlate negatively with forecast accuracy, which is counter to the hypothesized relationship, in which the introduction of IFRS would increase forecast accuracy, particularly in firms with a higher degree of intangible assets. The table also shows that all six control variables significantly correlate with forecast accuracy. The signs indicate that the number of analysts, market value, and trading volume are associated with increased forecast accuracy, whereas unprofitable firms, earnings surprise and variation in profitability appear to worsen forecast accuracy. These correlations all occur in the expected direction, as observed earlier in Table 1. The table provides similar results for forecast dispersion, with the exception of trading volume, which is not significant. There appear to be no problems with multicollinearity; the highest correlation among the independent variables is 0.35 (for market value and number of analysts). However, 23

24 as expected, a high positive correlation exists between accounting standard followed and the interaction term. Nonetheless, all variance inflation factor (VIF) values are lower than 2; thus, there is no reason to believe that this correlation is affecting the accuracy of the estimations Insert Table 5 about here The complete results of the regressions are presented in appendices 1-6. The strongest predictors of forecast accuracy and forecast dispersion are the control variables profit/loss and earnings surprise, which have the expected signs in all cases except two (earnings surprise for Sweden and the Netherlands, in appendix 6). Overall, the models have much higher explanatory power when predicting the 10 th percentile (Pseudo R 2 from for estimations of equation 1 and for estimations of equation 2) than when predicting the 50 th percentile (Pseudo R 2 from for equation 1 and for equation 2) or the 90 th percentile (Pseudo R 2 from for equation 1 and for equation 2). This result indicates that smaller errors and dispersions are more random than systematic and, therefore, should be little affected by a change in accounting standard. Table 6, which summarizes the results for the test variables from appendices 1-3, indicates that IFRS have no measurable impact on forecast accuracy, either overall or for individual countries, and that there are no measurable differences between firms with varying degrees of intangible assets. None of the coefficients is significant and there is no consistency among the signs of the coefficients. Therefore, in general, we fail to find evidence in support of H1 or H5. Because we find no systematic difference between countries, the results support neither 24

25 H3a nor H3b. IFRS appear to have no impact on forecast accuracy, regardless of prior GAAPdifference or legal enforcement Insert Table 6 about here Table 7, which summarizes the results for the test variables from appendices 4-6, shows a slightly more consistent pattern regarding forecast dispersion. All coefficients of the variable accounting standard followed are positive and many are significant, indicating that IFRS appears to result in diminished forecast dispersion. In particular, IFRS appears to have affected forecast dispersion in the 50 th percentile; the only exception is the UK, for which there is a positive but insignificant coefficient. In the 10 th percentile and the 90 th percentile, the effect is not equally consistent, but there is a significant positive effect in France and for the full sample at the 10 th percentile level and a significant positive effect in the Netherlands at the 90 th percentile level. However, the interaction term exhibits no consistent pattern. The signs are mixed, and only one coefficient is significant, although it has a sign that is the opposite of that expected Insert Table 7 about here These results provide some support for H2 but no support for H6. Although IFRS appear to have a measurable impact on forecast dispersion, the degree of intangible asset appears not to 25

26 affect this relationship. When examining cross-country differences, we note that in the UK, forecast dispersion has not been affected by IFRS. The UK exhibits the smallest GAAPdifference of all of the countries in the sample, which might explain why there is no effect on forecast dispersion. At the same time, the UK has the strongest enforcement, which in theory, should indicate a stronger effect. Another country with a low GAAP-difference is the Netherlands, in which we detect an effect both in the 10 th percentile and the 50 th percentile, which suggests that the impact of GAAP-difference is not so straightforward. The Netherlands has the weakest enforcement in the sample but still displays positive significant coefficients. The only other country that has two significant coefficients, in the 50 th percentile and in the 90 th percentile, is France, which exhibits both a high GAAP-difference and strong enforcement. That there is an effect in Sweden, a country with a large GAAP-difference but weak enforcement, but not the UK, suggests that GAAP-difference is more important than enforcement. Although the results for the Netherlands may appear to discredit this interpretation, it is possible that GAAP-difference has a non-linear positive effect, suggesting support for H4a. It is difficult to argue that enforcement has an effect based on our results, indicating that H4b is likely false. Broadly speaking, our results indicate that analysts have become more uniform in their forecasts since the introduction of IFRS, suggesting that uncertainty among these professionals has decreased. This decreased uncertainty appears not to be driven by IFRS asset valuation methods better representation of firms underlying economic value because the effect is not more pronounced in firms with higher degrees of intangible assets. 26

27 5. Robustness analysis As an alternative procedure, we have also estimated our equations using OLS regressions. Because contrary to quantile regression, OLS is sensitive to skewness and outliers, the sample was winsorized. A total of 2.5 percent of each tail was altered during this process, which should resolve the outlier problem. The sample is still highly skewed, so the risk of biased estimators remains and the result should be interpreted with caution. The estimated models are provided in appendices 7-8. The result for forecast accuracy differs slightly from that obtained using quantile regression. There is a small but significant positive effect in Sweden and France, whereas there is a significant and negative effect in the UK. We find no significant effect for the sample as a whole. Obviously, these effects are not consistent and, because of the remaining skewness of the sample, this result might be viewed as being uncertain. However, it is possible to argue that this result suggests that IFRS have increased forecast accuracy in Sweden and France while decreasing it in the UK. The interaction term is significant and positive for Germany but not for any other country or overall. The result for forecast dispersion is practically the same as that obtained using quantile regression. We record a positive effect of IFRS in all countries except the UK. Three countries (Sweden, France and Germany) exhibit significant interaction terms, but these point in different directions. The OLS estimations thus do little to challenge the overall pattern identified in the data. Although a few more coefficients become significant, there is no consistent pattern, which causes us to suspect that they are spurious. 27