A Note on the Use of Analysts Forecasts to

A Note on the Use of Analysts Forecasts to Measure Firm Information Environments John Nowland School of Economics & Finance Queensland University of Technology GPO Box 2434, Brisbane 4001, Australia. Ph: +61-7-3138-4241. Fax: +61-7-3138-1500. Email: j.nowland@qut.edu.au Andreas Simon Orfalea College of Business California Polytechnic State University San Luis Obispo California, USA Ph: +1-858-531-8482 Email: a.simon@business.uq.edu.au A Note on the Use of Analysts Forecasts to Measure Firm Information Environments ABSTRACT Since Lang and Lundholm (1996) a number of studies have used analyst variables to measure firm information environments. However, a critical assumption in these studies is that the composition and abilities of analysts providing forecasts does not change over time. We use the illustrative case of foreign firms cross-listing on US stock exchanges to show the confounding effect of a change in analyst composition on the accuracy of analysts forecasts. Our analysis indicates that on average only 30% of analysts continue following the company from the year before to 2-3 years after cross-listing. Furthermore, the act of cross-listing is associated with a shift in analyst coverage away from a group of analysts who are less accurate forecasters and toward a group of analysts who are more accurate forecasters. This shift in analyst coverage by itself results in an improvement of 9.66% in analyst forecast accuracy. The effect is greater (18.99%) in firms from emerging markets where there is a greater change in analyst composition. To remove this bias, we suggest using the forecasts of a fixed group of analysts, i.e. only those that follow companies over the entire period of interest. JEL codes: G14, G15, G34. Keywords: Analyst following, analyst forecast accuracy, cross-listing, information environment, transparency. 2 Introduction Since Lang and Lundholm (1996) established a link between firm disclosure practices and analyst behaviour, a number of studies have used analyst following and the properties of analysts forecasts (e.g. accuracy and dispersion) to measure firm information environments. For example, Lang et al. (2003) examine the change in the information environments of foreign firms around US cross-listings. Bailey et al. (2003), Helfin et al. (2003) and Irani and Karamanou (2003) explore the effect of Regulation FD on firm information environments. International studies have also used analyst variables to examine the effect of new regulations on corporate transparency (Brown et al., 1999; Chang et al., 2007; Nowland, 2008). In general, higher analyst following and higher forecast accuracy represent an improvement in firm information environments. However, a critical assumption in these studies is that the composition and abilities of analysts providing forecasts during the period of interest are consistent. If there is a change in the composition or quality of analysts during the period, this could introduce a bias to the properties of analysts forecasts. For example, if there is an event that causes an improvement in the quality of analysts following the company, this will result in an improvement in analyst forecast accuracy, which will confound the effect of the event on the accuracy of analysts forecasts. To eliminate this bias we propose that only the forecasts of continuing analysts (those who follow the company over the entire period of interest) be used in the analysis. To illustrate the potential effect of this bias, we re-examine the effect of US cross-listings on the information environments of foreign firms. Previous research has documented an increase in analyst following and analyst forecast accuracy around cross-listing (Baker et al., 2002; Lang et al., 2003). This has been attributed to the 3 commitment by firms to the improved disclosure practices required by US markets. However, unlike previous studies we split analysts into old, continuing and new analysts in the seven-year period around cross-listing. Old analysts are those who only follow the company prior to listing. New analysts only follow the company after it cross-lists. Continuing analysts follow the company throughout the period. Our analysis indicates that on average only 30% of analysts continue following the company from the year before to 2-3 years after cross-listing. This represents a substantial shift in the composition of analysts around cross-listing. To explore the forecasting ability of the old, new and continuing analysts we examine their forecast accuracy and dispersion in the years around cross-listing. Any differences in the relative ability of the groups will influence conclusions made about the change in firm information environments around cross-listing. We find that the forecast accuracy of old analysts is significantly lower in the year before and the year of cross-listing, while the forecast accuracy of new analysts is significantly higher in the year of and the year after cross-listing. This result illustrates that the act of crosslisting is associated with a shift in analyst coverage away from a group of analysts who are less accurate forecasters and toward a group of analysts who are more accurate forecasters. This shift by itself could explain at least part of the documented improvement in the information environments of cross-listed firms. To further understand the shift in analyst coverage around cross-listing we investigate the characteristics of the old, new and continuing analysts. Before crosslisting, we find that old analysts have less experience, follow fewer US companies and forecast in fewer markets than continuing analysts. In addition to their poor forecasting performance, their lack of experience and international expertise, may contribute to their decision to stop following the company after cross-listing. After 4 cross-listing we find that new analysts have less experience, follow more US companies and forecast in fewer markets than continuing analysts. This suggests that it is not experience and international expertise that is helping the new analysts accurately forecast the earnings of the newly cross-listed companies. Rather, it seems that the new analysts are selectively choosing to follow companies that they are able to forecast accurately. Furthermore, a number of studies show that the effect of cross-listing can differ for companies from developed versus emerging markets (Doidge et al., 2004; Fernandes and Ferreira, 2008). In particular, the effect of cross-listing is expected to be greater for companies from emerging markets as they are expected to improve their transparency the most. We therefore examine the shift in analyst following separately for companies from developed and emerging markets. We find that the shift in analyst composition is greater for companies from emerging markets. Therefore it is possible that the relatively larger shift in analyst composition for firms from emerging markets may partially explain the greater effect cross-listing has on the information environments of firms from emerging markets Finally, we conduct analysis similar to Lang et al. (2003) by examining the effect of cross-listing on analyst forecast accuracy. However, we use both the forecasts of all analysts and just the forecasts of continuing analysts. The forecasts of continuing analysts are expected to be a cleaner test of the effect of cross-listing on company information environments as they are not influenced by the shift in analyst coverage. The forecasts of all analysts include both the effect of the shift in analyst composition and the effect of cross-listing on company transparency. Our interest is in the difference between these two results, which represents the bias due to the change in analyst composition around cross-listing. 5 We find that using the forecasts of all analysts rather than just continuing analysts biases the results towards finding an improvement in forecast accuracy around cross-listing. The size of this bias is 9.66% of average forecast accuracy. In the case of emerging market firms, where the shift in analyst composition is greater, the bias represents 18.99% of average forecast accuracy. To remove this bias, we suggest using the forecasts of a fixed group of analysts, i.e. only those that follow companies over the entire period of interest. Finally, while it is not the primary focus of this paper and we use different modelling techniques to Lang et al. (2003), we do not find evidence of an improvement in analyst forecast accuracy around cross-listing. One possible explanation for this is that the cross-listed companies already started improving their disclosure and transparency in preparation for cross-listing. Data & Variables As an illustrative case, this study examines the analyst following and analyst earnings forecasts of foreign firms from 3 years before to 3 years after cross-listing on US exchanges. Firms are included in the sample if they were cross-listed on the New York Stock Exchange (NYSE) or Nasdaq at the end of 2006 and if analyst earnings forecasts are available from I/B/E/S for at least one year before and after cross-listing. Cross-listed companies are sourced as American Depositary Receipts (ADRs) from the Bank of New York website.1 As I/B/E/S coverage is from 1988 to 2006, crosslisting dates range from 1987 to 2005. The main sample comprises 1,189 firm-year observations of 191 firms from 31 countries. At the analyst level this includes 81,924 annual earnings forecasts from 7,790 analysts. The regression analysis is conducted 1 We do not include non-exchange listings as they are not required to meet the same new disclosure requirements as exchange listings. We do not include direct listings as the sample of ADRs provides us with sufficient data across an adequate number of countries to illustrate our case. 6 on a reduced sample due to the need to match current year financial variables and three years of prior stock price data from Compustat with each observation from the main sample. This reduced sample comprises 715 firm-year observations of 144 firms from 28 countries. Table 1 provides the number of firms, number of firm-year observations and descriptive statistics of the analyst variables for the main sample and total assets (in billions of US dollars) for the reduced sample by country of origin. The emerging market column identifies countries as emerging markets (yes/no) as per Fernandes and Ferreira (2008). Analyst following is defined as the number of analysts providing annual earnings forecasts of the company during the fiscal year. Analyst forecast accuracy is the negative of the absolute value of average forecast earnings per share less actual earnings per share all divided by actual earnings per share. If analysts have multiple forecasts in an estimation period, we use only the most recent forecast. Analyst forecast dispersion is the standard deviation of earnings forecasts divided by actual earnings per share. Both forecast accuracy and dispersion are windsorized at the 1st and 99th percentiles to mitigate the effect of small actual earnings figures. All results are consistent if year-end share price is used as the denominator instead of actual earnings, however the sample size is significantly reduced. The greatest number of sample firms and firm-year observations come from the United Kingdom (31, 194), Brazil (17, 124) and France (17, 100). The average sample firm has an analyst following of 21.27 analysts, analyst forecast accuracy of 0.3934 and forecast dispersion of 0.4774. The average firm size in the reduced sample is US$59.45 billion. There is considerable variation in the sample with average analyst following ranging from 5.42 in Russia to 41.55 in Germany. Average firm size 7 also ranges from US$0.01 billion in Korea and Venezuela to US$163.27 billion in Switzerland. Methodology and Results Analyst Coverage around Cross-listing To examine analyst composition around cross-listing, analysts are split into old, continuing and new analysts for each firm-year observation. Old analysts are those who forecast the earnings of the company at least once before the date of cross-listing and then do not forecast again after cross-listing. New analysts provide earnings forecasts of the company at least once after cross-listing but not before cross-listing. Continuing analysts forecast company earnings at least once before and after crosslisting. Table 2 Panel A shows analyst following across the seven-year period around cross-listing. The result for all analysts shows that analyst following steadily increases from before to after cross-listing. This is consistent with the findings of previous research that cross-listing is associated with increased analyst coverage. However, the breakdown between old, continuing and new analysts (also displayed in Figure 1) shows that there is a significant change in analyst composition around cross-listing. A large number of analysts following the companies before cross-listing drop out in the year of cross-listing. In the years after cross-listing more and more of the analyst following is comprised of new analysts. On average, of the 19.93 analysts that follow companies in the year before cross-listing, only 7.01 (35%) and 4.93 (25%) remain in 8 the two and three years after cross-listing respectively.2 This represents a substantial change in analyst composition from old to new analysts around cross-listing. However, a change in analyst composition will only have implications for measures of firm information environments (e.g. analyst forecast accuracy and dispersion) if the relative forecasting ability of the old, new and continuing analysts are different. We therefore examine the forecast accuracy and dispersion of the old, new and continuing analysts in the years around cross-listing. Panel B presents the results for analyst forecast accuracy. The analysis indicates that old analysts have lower forecast accuracy than continuing analysts in the year before and the year of cross-listing. New analysts have significantly higher forecast accuracy than continuing analysts in the year of and the year after crosslisting. This result illustrates that the act of cross-listing is associated with a shift in analyst coverage away from a group of analysts (old analysts) who are less accurate forecasters and toward a group of analysts (new analysts) who are more accurate forecasters. This shift by itself could explain at least part of the documented improvement in the information environments of cross-listed firms. Three years after cross-listing there is also a significant difference between new and continuing analysts. In this case, new analysts have lower forecast accuracy than continuing analysts. This is most likely due to survivorship bias i.e. the superior forecasting ability of a small number of continuing analysts who have been following the company for a long period of time (Mikhail et al., 1997). Panel C examines forecast dispersion around cross-listing. While forecast dispersion is not a direct measure of analyst ability it does provide an indication of the consistency of analysts forecasts. For example, if analysts have a common forecasting 2 As a comparison, the authors measured the average proportion of analysts that continue following a company over a 3-4 year period. Over a number of control groups the average was 40-50%. 9 approach or are herding (staying close to the average forecast) then dispersion will be low. If there is variation in forecasting approaches then dispersion will be high. The only significant results are that new analysts have significantly higher forecast dispersion than continuing analysts in the first and second years after cross-listing. In summary, this section has documented a substantial shift in analyst composition around cross-listing away from a group of analysts who are less accurate forecasters and toward a group of analysts who are more accurate forecasters. This shift by itself could explain at least part of the documented improvement in the information environments of cross-listed firms. Our initial explanation for this is shift is that old analysts must foresee that their relatively poor forecasting performance is unlikely to improve and that their best option is to stop following the company at cross-listing. After cross-listing, new analysts are found to have better forecasting accuracy than continuing analysts even though they have no experience forecasting the company. It is unlikely that the new analysts are herding as the forecast dispersion of new analysts forecasts is not lower than continuing analysts after cross-listing (Graham, 1999). It could be that the new analysts are selectively choosing to follow companies that they are able to forecast accurately (McNichols and O’Brien, 1997). Another possibility is that the new analysts have extensive experience following other international companies, which gives them an edge over continuing analysts. These explanations are further explored in the next section on analyst characteristics. Analyst Characteristics The results of the previous section indicate that cross-listing has two effects on analyst coverage: (1) old analysts who have lower forecast accuracy stop following the company at cross-listing, and (2) new analysts who have higher forecast accuracy 10 start following the company after cross-listing. This section explores potential explanations for these effects by analysing the characteristics of old, new and continuing analysts at the analyst level in the three-year period around cross-listing. We examine five analyst characteristics related to experience, busyness and international expertise. As Mikhail et al. (1997) document a relationship between analyst experience and forecast accuracy, we examine two measures of analyst experience. A broad measure of experience (analyst experience), which is the number of years since the analyst had their first forecast recorded on I/B/E/S, and a company-specific measure of experience (company experience), which is the number of years the analyst has been forecasting the company. If analysts are busy, i.e. they follow more companies, they may produce less accurate forecasts as they have less time to devote to each forecasted company. We measure busyness as the number of companies the analyst forecasts during the same year (companies followed). The remaining two characteristics measure the international expertise of analysts. As the cross-listed companies are now part of both their home market and US markets, we measure the expertise of analysts in following both US companies and companies from other foreign markets. Percentage of US companies followed measures the number of US companies forecasted divided by all companies forecasted. Number of markets measures the number of different markets (represented by different forecasting currencies) in which the analyst forecasts companies. Table 3 presents the differences between old and continuing analysts in Panel A, and new and continuing analysts in Panel B. Only analyst-firm-year observations in the three year period around cross-listing are examined as this is where significant differences between the groups were found in the previous analysis. Before cross- 11 listing, we find that old analysts have less experience (broad and company specific), follow fewer US companies and forecast in fewer markets than continuing analysts. There is no significant difference in the busyness of old and continuing analysts. Combined with their poor forecasting performance, these results suggest that the old analysts have less experience and less expertise in following US and foreign companies. If the old analysts foresee that the company will be followed by a new group of analysts with greater experience or ability than themselves after cross-listing then it is logical that they would decide to stop following the company. After cross-listing we find that new analysts have less experience, follow more US companies and forecast in fewer markets than continuing analysts. There is no significant difference in the busyness of new and continuing analysts. This suggests that it is not experience and international expertise that is helping the new analysts accurately forecast the earnings of the newly cross-listed companies. Rather, it is more likely that the new analysts are selectively choosing to follow companies that they are able to forecast accurately. Developed versus Emerging Market Companies A number of recent studies show that the effect of cross-listing is different for firms from different home markets. Doidge et al. (2004) show that the effects of crosslisting are greater for firms from countries with weaker investor protection. This is because they are expected to improve their investor protection and transparency the most. Fernandes and Ferreira (2008) also show that the effects of cross-listing are different for firms from developed and emerging markets. In particular they show that firms from emerging markets have a greater increase in analyst coverage. 12 We therefore examine the change in analyst composition separately for companies from developed and emerging markets. The main reason for this analysis is to determine if the change in analyst composition is consistent between firms from developed and emerging markets. If the change is greater for emerging market firms, then this may partially explain the greater effect cross-listing has on the information environments of firms from emerging markets. Table 4 reports analyst following around cross-listing for firms from developed and emerging countries. There are a total of 746 firm-year observations of 118 firms from 16 developed markets and 443 firm-year observations of 73 firms from 15 emerging markets. Panel A presents the analysis for firms from developed countries and shows that of the 23.34 analysts that follow companies in the year before cross-listing, only 9.07 (39%) and 6.24 (27%) remain in the two and three years after cross-listing respectively. Panel B presents the analysis for firms from emerging markets and shows that of the 14.21 analysts that follow companies in the year before cross-listing, only 3.78 (27%) and 2.76 (19%) remain in the two and three years after cross-listing respectively. This indicates that the shift in analyst composition is greater for firms from emerging markets. The effect of this greater shift in analyst composition on the forecast accuracy of firms from emerging markets is further explored in the regression analysis in the next section. Implications for Company Information Environments As the change in analyst composition around cross-listing has implications for conclusions drawn about the effect of cross-listing on the information environments of foreign companies, we conduct analysis similar to Lang et al. (2003) by examining the effect of cross-listing on analyst forecast accuracy. However, we use both the 13 forecasts of all analysts and just the forecasts of continuing analysts. The forecasts of continuing analysts are expected to be a cleaner test of the effect of cross-listing on company information environments as they are not influenced by the shift in analyst composition. The forecasts of all analysts include both the effect of the shift in analyst composition and the effect of cross-listing on company transparency. Our interest is in the difference between these two results, which represents the bias due to the change in analyst composition around cross-listing. The regression analysis is conducted on a reduced sample due to the need to match current year financial variables and three years of prior stock price data from Compustat with each observation from the main sample. Our model includes the same control variables as Lang et al. (2003), with an additional variable to control for the difficulty analysts have in forecasting company losses. Our analysis also includes fixed firm effects to control for any omitted time-invariant firm-level variables. Therefore, our analysis compares the time series effect of cross-listing on firms, whereas the main focus of the analysis in Lang et al. (2003) is on the cross-sectional differences between cross-listed and non-cross-listed firms. The model is: (1) where Accuracy is analyst forecast accuracy of firm i in year t, PLIST is a dummy variable equal to one in the years after cross-listing, SIZE is the natural logarithm of total assets in millions of US dollars, RISK is the standard deviation of monthly returns over the past three years, RE is the return-earnings correlation over the past three years, ES is earnings surprise measured as actual earnings this period minus actual earnings last period standardized by end of year share price and LOSS is a dummy variable equal to one if the company made a loss. 14 Table 5 presents the results for all firms and firms from developed and emerging markets separately. Our primary interest is the difference in the coefficients on PLIST between the regressions using the forecasts of all analysts and continuing analysts only. This represents the bias due to the change in analyst composition around cross-listing. The difference for all companies (between the first and fourth regressions) is 0.0402. This equals 9.66% of the average analyst forecast accuracy of all firms. The difference for developed market firms (between the second and fifth regressions) is 0.0406, which equals 9.55% of the average forecast accuracy of developed market firms. The difference for emerging market firms (between the third and sixth regressions) is 0.0973, which equals 18.99% of the average forecast accuracy of emerging market firms. Therefore, we find that using the forecasts of all analysts rather than just continuing analysts biases the results towards finding an improvement in forecast accuracy around cross-listing. The size of this bias is not immaterial and is greater for emerging market firms, where the shift in analyst composition is greater. To remove this bias, we suggest using the forecasts of continuing analysts only, i.e. only those that follow companies over the entire period of interest. While it is not the primary focus of this paper, we do not find any evidence of an improvement in analyst forecast accuracy of firms around cross-listing. Our only marginally significant result is that analyst forecast accuracy is worse after crosslisting for firms from developed markets when using the forecasts of continuing analysts only. Possible explanations for the difference in our results compared to Lang et al. (2003) are that they focused on cross-sectional analysis, whereas our analysis focuses on the time series impact of cross-listing. Lang et al. (2003) explain that firms 15 may improve their disclosure and transparency in preparation for cross-listing, so there may be minimal differences around the cross-listing event. Finally, the following robustness checks were completed with results consistent to those presented. Year-end share price was used as the denominator of forecast accuracy instead of actual earnings. An additional variable controlling for the timeliness of the forecasts was included but found to be insignificant. The models were run using different econometric specifications e.g. pooled OLS with industry, country and year dummies, and random firm effects. Conclusion A number of recent studies have used analyst following and the properties of analysts forecasts (e.g. accuracy and dispersion) to measure firm information environments. However, in doing so they have implicitly assumed that the composition and quality of analysts does not change over the period of interest. We use the example of foreign firms cross-listing on US exchanges to illustrate the potential effect of a shift in analyst composition on analyst forecast accuracy. We document a substantial change in analyst composition around cross-listing, which includes a shift away from a group of analysts who are less accurate forecasters and toward a group of analysts who are more accurate forecasters. We show that this shift in analyst composition biases the results towards finding an improvement in forecast accuracy around cross-listing. The size of this bias is 9.66% of average forecast accuracy. In the case of emerging market firms, where the shift in analyst composition is greater, the bias represents 18.99% of average forecast accuracy. To remove this bias, we suggest using the forecasts of a fixed 16 group of analysts, i.e. only those that follow companies over the entire period of interest. These results indicate that researchers need to be careful when using analysts forecasts to measure company transparency. While using the forecasts of a fixed group of analysts will remove the bias due to a change in analyst composition, there may be other biases that need to be considered, such as a learning effect (an improvement in analyst forecast accuracy the longer an analyst follows a company) documented by Mikhail et al. (1997). To minimize the potential effect of these biases on reported results, researchers investigating company transparency should use a number of different measures, such as abnormal return volatility or the stock price informativeness measure of Fernandes and Ferreira (2008). 17 References Bailey, W., Li, H., Mao, C.X. and Zhong, R. (2003) Regulation fair disclosure and earnings information: market, analyst and corporate responses, Journal of Finance, 58, 2487 - 2514. Baker, H.K., J. Nofsinger and D. Weaver, 2002, International cross-listing and visibility, Journal of Financial and Quantitative Analysis 37, 495-521. Brown, P., Taylor, S.L. and Walter, T.S. (1999) The impact of statutory sanctions on the level and information content of voluntary corporate disclosure, Abacus, 35, 138 – 162. Chang, J., Cho, Y.J. and Shin, H.H. (2007) The change in corporate transparency of Korean firms after the Asian crisis: an analysis using analysts’ forecast data, Corporate Governance: An International Review, 15-6, 1144-1167. Doidge, C., Karolyi, G. A. and Stulz, R. (2004) Why are foreign firms listed in the U.S. worth more?, Journal of Financial Economics, 71, 205 - 238. Fernandes, N. and M. Ferreira, 2008, Does international cross-listing improve the information environment, Journal of Financial Economics 88, 216-244. Graham, J. R., 1999, Herding among investment newsletters: Theory and evidence, Journal of Finance 54 (1), 237-268. Helfin, F., Subramanyam, K.R. and Zhang, Y. (2003) Regulation FD and the financial information environment: early evidence, Accounting Review, 78, 1 – 37. Irani, A.J. and I. Karamanou: 2003. “Regulation fair disclosure, analyst following and analyst forecast dispersion”, Accounting Horizons, 17, 15 – 29. Lang, M., K. Lins and D. Miller, 2003, ADRs, analysts and accuracy: Does crosslisting in the US improve a firm’s information environment and increase market value?, Journal of Accounting Research 41, 317-346. 18 Markow, C. and A. Tamayo, 2006, Predictability in financial analyst forecast errors: Learning or irrationality?, Journal of Accounting Research 44, 725-761. McNichols, M. and P. O’Brien, 1997, Self-selection and analyst coverage, Journal of Accounting Research 35, 167-199. Mikhail, M. B., B. R. Walther and R. H. Willis, 1997, Do security analysts improve their performance with experience?, Journal of Accounting Research 35 (3), 131-157. Nowland, J., 2008, The effect of national governance codes on firm disclosure practices: Evidence from analysts earnings forecasts, Corporate Governance: An International Review, 16, 475-491. 19 20 Table 1 – Descriptive Statistics Emerging market indicates if the country is classified as an emerging market (yes/no). No. firms is the number of firms in the main sample. No. obs is the number of firm-year observations in the main sample. Analyst following is defined as the number of analysts providing earnings forecasts of the company during the fiscal year. Forecast accuracy is the negative of the absolute value of average forecast earnings per share less actual earnings per share all divided by actual earnings per share. Forecast dispersion is the standard deviation of earnings forecasts divided by actual earnings per share. Total assets is in US$ billions for companies in the reduced sample. Analyst forecast accuracy and dispersion are windsorized at the 1st and 99th percentiles. Data is from I/B/E/S and Compustat. Country Argentina Australia Belgium Brazil Chile China Colombia Denmark Finland France Germany Hong Kong Hungary India Ireland Italy Japan Korea México Netherlands New Zealand Norway Philippines Russia South Africa Spain Sweden Switzerland Taiwan United Kingdom Venezuela Total Emerging Market Y N N Y Y Y Y N N N N Y Y Y N N N Y Y N N N Y Y Y N N N Y N Y No. firms 4 6 1 17 6 4 1 1 3 17 15 3 1 10 4 1 14 2 10 11 1 3 1 2 7 2 2 6 4 31 1 191 No. Analyst obs Following 24 6.04 31 9.32 7 31.71 124 16.44 34 16.24 23 24.30 5 6.00 7 16.71 20 25.75 110 29.29 96 41.55 17 15.29 5 14.00 55 15.91 26 7.62 6 12.50 88 12.85 9 21.11 61 19.75 70 27.27 7 22.29 17 11.24 6 21.83 12 5.42 38 7.13 14 38.21 12 26.67 41 34.34 23 15.04 194 22.14 7 14.14 1189 21.37 Forecast Accuracy -0.7416 -0.3301 -0.5259 -0.4735 -0.3705 -0.3197 -0.5753 -0.0533 -0.5393 -0.5049 -0.3308 -0.5186 -0.1977 -0.2410 -0.4537 -0.0537 -0.3432 -0.1144 -0.5124 -0.3605 -0.3410 -0.5375 -0.7961 -0.7430 -0.4460 -0.3389 -0.2887 -0.3171 -0.3717 -0.2937 -0.6035 -0.3934 Forecast Dispersion 0.8697 0.2739 0.3652 0.6333 0.9403 0.7128 1.7137 0.3699 0.5200 0.4386 0.2587 0.3418 0.1220 0.2014 0.6674 0.4057 0.2240 0.2381 0.5527 0.3709 0.6760 0.6527 0.7743 0.6374 0.4640 0.6612 0.5077 0.3579 0.4539 0.5086 1.6643 0.4774 Total Assets 0.60 0.99 7.59 3.61 3.41 5.99 3.15 11.97 30.45 147.62 0.46 1.79 17.40 8.19 0.01 2.88 122.93 4.20 1.27 5.94 2.26 3.03 10.26 29.57 163.27 2.69 107.89 0.01 59.45 21 Table 2 – Analyst Coverage around Cross-listing Analyst following is defined as the number of analysts providing earnings forecasts of the company during the fiscal year. Analyst forecast accuracy is the negative of the absolute value of average forecast earnings per share less actual earnings per share all divided by actual earnings per share. Analyst forecast dispersion is the standard deviation of earnings forecasts divided by actual earnings per share. Analyst forecast accuracy and dispersion are windsorized at the 1st and 99th percentiles. Old analysts are those who forecast the earnings of the company only before cross-listing. New analysts provide earnings forecasts of the company only after cross-listing. Continuing analysts forecast company earnings before and after cross-listing. Old-Cont represents the difference between old and continuing analysts. New-Cont represents the difference between new and continuing analysts. Analyst forecast data is from I/B/E/S. Asterisks represent significance of t-tests for differences in means at the 1%***, 5%** and 10%* levels. Panel A – Analyst Following Years Around Cross-listing Analysts Old Continuing New All analysts 19.33 19.85 19.93 -3 14.92 4.41 -2 13.23 6.62 -1 10.05 9.88 0 4.03 12.64 4.16 20.83 10.23 12.16 22.39 7.01 16.70 23.71 4.93 18.78 23.72 1 2 3 Panel B – Analyst Forecast Accuracy Years Around Cross-listing Analysts Old Continuing New Old - Cont New - Cont 0.0102 -0.0003 -0.0237** -3 -0.3570 -0.3672 -2 -0.4101 -0.4098 -1 -0.3697 -0.3460 0 -0.4447 -0.4130 -0.3979 -0.0317*** 0.0151* 0.0354** 0.0118 -0.0303* -0.4583 -0.4229 -0.3660 -0.3542 -0.2710 -0.3013 1 2 3 Panel C – Analyst Forecast Dispersion Years Around Cross-listing Analysts Old Continuing New Old - Cont New - Cont -0.0056 0.0151 -0.0162 -3 0.4311 0.4367 -2 0.4370 0.4219 -1 0.3505 0.3667 0 0.2766 0.3347 0.3772 -0.0581 0.0425 0.0451* 0.0872*** 0.0418 0.4528 0.4979 0.4134 0.5006 0.4166 0.4584 1 2 3 22 Table 3 – Analysts Characteristics Characteristics of analysts in the three years around cross-listing. Analyst experience is the number of years since the analyst had their first forecast recorded on I/B/E/S. Company experience is the number of years the analyst has been forecasting the company. Companies followed is the number of companies the analyst forecasts during the same year. Percentage of US companies followed measures the number of US companies forecasted divided by all companies forecasted. Number of markets measures the number of different markets (represented by different forecasting currencies) in which the analyst forecasts companies. Old analysts are those who forecast the earnings of the company only before cross-listing. New analysts provide earnings forecasts of the company only after cross-listing. Continuing analysts forecast company earnings before and after crosslisting. Analyst forecast data from I/B/E/S. Asterisks represent significance of t-tests for differences in means at the 1%***, 5%** and 10%* levels. Panel A – Old versus Continuing Analysts Old Mean Std Continuing Mean Std Means Tests Difference t-stat Analyst experience Company experience Companies followed % of US companies Number of markets No. of analysts 3.24 1.60 14.98 9.59 2.12 2657 2.99 1.98 46.05 26.15 2.60 3.67 1.88 14.85 12.96 2.32 4378 3.19 2.15 19.94 28.22 1.98 -0.43 -0.28 0.13 -3.37 -0.20 -5.70*** -5.52*** 0.14 -5.10*** -3.38*** Panel B – New versus Continuing Analysts New Mean Std Continuing Mean Std Means Tests Difference t-stat Analyst experience Company experience Companies followed % of US companies Number of markets No. of analysts 2.15 0.34 13.64 14.15 2.18 3125 2.76 0.87 26.46 31.32 1.94 4.37 2.49 14.22 12.75 2.31 4512 3.26 2.26 18.80 27.73 1.94 -2.22 -2.15 -0.58 1.40 -0.13 -32.14*** -58.09*** -1.06 2.01** -2.91*** 23 Table 4 – Developed versus Emerging Market Companies Analyst following is defined as the number of analysts providing earnings forecasts of the company during the fiscal year. Old analysts are those who forecast the earnings of the company only before cross-listing. New analysts provide earnings forecasts of the company only after cross-listing. Continuing analysts forecast company earnings before and after cross-listing. Analyst forecast data is from I/B/E/S. Panel A – Analyst Following of Developed Market Companies Years Around Cross-listing Analysts Old Continuing New All 21.99 23.11 23.34 -3 16.58 5.41 -2 14.75 8.36 -1 10.83 12.51 0 4.17 15.51 4.65 24.33 12.67 13.07 25.74 9.07 18.70 27.77 6.24 22.18 28.42 1 2 3 Panel B – Analyst Following of Emerging Market Companies Years Around Cross-listing Analysts Old Continuing New All 14.49 14.13 14.21 -3 11.89 2.60 -2 10.56 3.57 -1 8.74 5.47 0 3.80 7.93 3.36 15.09 6.10 10.61 16.71 3.78 13.58 17.36 2.76 13.12 15.88 1 2 3 24 Table 5 – Regression Analysis of Analyst Forecast Accuracy Fixed firm effect regressions of analyst forecast accuracy for all analysts and continuing analysts only on the following variables - PLIST is a dummy variable equal to one in the years after cross-listing, SIZE is the natural logarithm of total assets in millions of US dollars, RISK is the standard deviation of monthly returns over the past three years, RE is the return-earnings correlation over the past three years, ES is earnings surprise measured as actual earnings this period minus actual earnings last period standardized by end of year share price and LOSS is a dummy variable equal to one if the company made a loss. Data from I/B/E/S and Compustat. P-values are in parentheses. All analysts All PLIST SIZE RISK RE ES LOSS Intercept Adj-R2 Continuing analysts only Emerging 0.0982 (0.28) 0.0543 (0.09) -0.4773 (0.10) 0.0383 (0.37) 0.0029 (0.99) -0.9533 (0.00) -0.3768 (0.25) 0.5156 221 All -0.0451 (0.32) -0.0028 (0.89) -0.1858 (0.40) -0.0316 (0.17) 0.2075 (0.21) -0.6764 (0.00) 0.0243 (0.92) 0.3813 670 Developed -0.0904 (0.09) 0.0189 (0.44) 0.6120 (0.09) -0.0599 (0.03) 0.1320 (0.49) -0.4992 (0.00) -0.2102 (0.46) 0.3227 474 Emerging 0.0009 (0.99) -0.0019 (0.95) -0.5327 (0.05) 0.0426 (0.32) -0.2040 (0.59) -0.9899 (0.00) 0.0416 (0.91) 0.5481 196 Developed -0.0498 (0.34) 0.0140 (0.57) 0.5700 (0.12) -0.0590 (0.03) 0.1603 (0.40) -0.5123 (0.00) -0.1915 (0.50) 0.3209 494 -0.0049 (0.91) 0.0191 (0.33) -0.1592 (0.47) -0.0311 (0.18) 0.2504 (0.14) -0.6716 (0.00) -0.1956 (0.45) 0.3814 715 n 25