- Email: [email protected]
exit as an alternative explanation for the disciplining role of independent analysts
ARTICLE IN PRESS
Journal of Accounting and Economics 45 (2008) 317–323 www.elsevier.com/locate/jae
Endogenous entry/exit as an alternative explanation for the disciplining role of independent analysts Thomas Z. Lys, Jayanthi Sunder Kellogg Graduate School of Management, Northwestern University, 2001 Sheridan Road, Andersen Hall 517, Evanston, IL 60208-2002, USA Received 7 November 2007; received in revised form 7 February 2008; accepted 11 February 2008 Available online 17 February 2008
Abstract Gu and Xue [2008. The superiority and disciplining role of independent analysts. Journal of Accounting and Economics, this issue, doi:10.1016/j.jacceco.2008.02.002] study the disciplining effect of independent analysts on the accuracy and forecast relevance of the forecasts of non-independent analysts. One of the intriguing results is that while independent analysts issue inferior forecasts, their presence appears to reduce the forecast bias, improve the forecast accuracy and increase the forecast relevance of forecasts issued by non-independent analysts. We explore alternative explanations for the Gu–Xue results. Our evidence of endogenous entry and exit of independent analysts provides a more compelling explanation for the reported results. r 2008 Elsevier B.V. All rights reserved. JEL classification: G28; G29; M41; M43 Keywords: Endogenous entry; Disciplining; Independent
1. A straight-forward hypothesis with intriguing results Gu and Xue (2008, hereafter GX) study the role of independent analysts (i.e., analysts with no ties to companies whose earnings are being forecasted) in improving the quality of forecasts of non-independent analyst (i.e., analyst with potential conflicts of interest, for example investment banking relations). The basic premise of the paper is that independent analysts should, all else equal, issue less-biased and more precise forecasts. More importantly, the presence of independent analysts should have a disciplining effect on nonindependent analysts, resulting in less-biased and more accurate forecasts by non-independent analysts as well. The importance of GX’s research is highlighted by recent regulatory initiatives, culminating in the Global Research Settlement, which aim to address issues relating to the lack of independence of equity research analysts. While these regulatory initiatives by and large focus on stock recommendations, they are based on the assumption that lack of analyst independence impedes the quality of equity research. Therefore, GX set out to examine whether independent analysts are superior forecasters of earnings and whether their presence disciplines non-independent analysts following the same company. Corresponding author.
E-mail address: [email protected] (T.Z. Lys). 0165-4101/$ - see front matter r 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.jacceco.2008.02.001
ARTICLE IN PRESS 318
T.Z. Lys, J. Sunder / Journal of Accounting and Economics 45 (2008) 317–323
Several papers have compared the quality of research of independent analysts with non-independent analysts. GX adds to this stream of literature in a novel manner by studying the interaction between these two groups of analysts. This basic thesis is both simple and intuitive. Further, despite the richness of the equity research industry for addressing fundamental questions relating to the dynamics of competition, there is little prior research that examines the role of competition and GX take a first step in that direction. GX ask two questions. First, are independent analysts’ forecasts less-biased and more precise than those of non-independent analysts? Second, does the presence of independent analysts discipline non-independent analysts following the same firm by reducing bias, increasing precision, and increasing the credibility of forecasts of the latter group? GX address these issues by comparing the accuracy, bias, and earnings response coefficient (ERC) of earnings forecasts for firms that were followed by only non-independent analysts in some quarters to forecasts for the same firms when they were followed by both independent and non-independent analyst in other quarters. At first, the hypothesis tested by GX seems straight-forward: introducing independent analysts who compete with incumbent non-independent analysts increases competition and ought to improve the quality of the forecasts of the non-independent analysts. And indeed, GX report that, in general, the presence of independent analysts reduces the bias, increases the accuracy, and increases the ERC of forecasts issued by non-independent analysts. What makes their research intriguing is the result that, forecasts issued by independent analysts while less-biased, is significantly less precise than forecasts issued by non-independent analysts. Thus, GX document a puzzling phenomenon, namely that introducing an otherwise inferior competitor induces superior competitors to increase the quality of their output and become more credible. 2. Evidence and interpretation of results 2.1. Findings Using absolute forecast errors as a measure of forecast accuracy, GX find that independent analysts are less accurate than non-independent analysts.1 Moreover, using average forecast error as a measure of forecast bias, while both groups of analysts are on average pessimistically biased, independent analysts are less pessimistic than non-independent analysts. The accuracy results suggest that forecasts issued by independent analysts are in fact inferior to forecasts issued by non-independent analysts. As indicated in the introduction, GX main result, however, is that forecast accuracy of non-independent analysts is higher and their forecasts are less pessimistically biased in periods when independent analysts are also issuing forecasts compared to periods when independent analysts are not issuing forecasts. GX interpret this as the disciplining effect of independent analysts. With respect to credibility, GX find that the ERC for independent analysts is higher, particularly when the earnings news is bad, than the ERC of non-independent analysts. Moreover, GX find that the ERCs of nonindependent analysts are higher in the presence of an independent analyst following the same firm compared to the ERC of non-independent analysts when independent analysts are absent. This result is further evidence of the disciplining role of independent analysts. Based on these results, GX conclude that the presence of independent analysts has a disciplining effect on the non-independent analysts and therefore suggest that improving analyst independence enhances the quality of analyst research. With respect to the superiority of independent analysts, GX find mixed results with forecasts of independent analysts being less accurate but being more credible (having higher ERCs). However, GX argue that higher ERCs imply better quality forecasts since it suggests that the forecasts are better aligned with market expectations. The mixed evidence with respect to the superiority of independent analysts raises some interesting puzzles. How is it that the ERC of independent analysts is higher even though their forecast accuracy is lower than that of non-independent analysts? Further, how can the independent analysts discipline the non-independent analysts while their forecasts have lower forecast accuracy? We address the puzzling results using two 1
This is consistent with evidence on forecast accuracy in other recent papers such as Jacob, Rock and Weber (2005) and Ertimur, Sunder and Sunder (2007).
ARTICLE IN PRESS T.Z. Lys, J. Sunder / Journal of Accounting and Economics 45 (2008) 317–323
319
approaches. First, we examine the variables and tests used to establish the superiority and disciplining effect and second, we explore potential mechanisms through which independent analysts discipline the nonindependents and generate alternative explanations for the reported results. 2.2. Measures of forecast quality 2.2.1. Measuring accuracy independent of bias GX measure the accuracy of forecasts of a group of analysts using two metrics that both employ the absolute error. The first (ACCY1) is based on the absolute error in the mean forecast and the second, (ACCY2) uses the mean of individual absolute forecast errors. While these measures capture the distance between forecasted earnings and actual earnings, they are not independent of their measure of forecast bias (FE). Bias is measured as the signed forecast error and this creates a problem because a systematic bias mechanically results in a larger mean absolute error. As an alternative, accuracy can be measured using the mean-squared error, which adjusts for the mean forecast error and—therefore—is not affected by any systematic bias component. However, using the mean-squared error, we continue to find that non-independent analysts are better than independent analysts. 2.2.2. ERC as a measure of forecast quality The ERC measure captures the sensitivity of the market reaction to the earnings news and—therefore—as the measured earnings surprise gets closer to the true earnings surprise of the market, we would expect a higher ERC. GX find that the ERC associated with independent forecasts is higher than that for nonindependent forecasts suggesting that, investors pay greater attention to the independent analysts while arriving at the market expectation. In that case, the results in GX should be interpreted as suggesting that the credibility of independent analysts is higher with investors and the presence of independent analysts improves the credibility of non-independent analysts. This does not, however, necessarily imply that the credibility improves as a consequence of a change in forecasting behavior of the non-independent analysts, because investors may (possibly irrationally) believe that these forecasts are more credible. If the credibility of the forecasts issued by independent analysts is higher, one would expect a greater market reaction to forecast revisions issued by independent analysts. Contrary to this expectation, in unreported tests, we find that the reaction is greater for non-independent analysts. However, the magnitude of the reaction is greater in quarters when independent analysts are present. Because of these seemingly inconsistent results, we next discuss the actual implementation of the ERC tests and the implications for the reported results. 3. ERC tests The basic ERC test involves a regression of abnormal returns to the earnings announcement of a given firm on the consensus forecast error, control variables and the interactions of the forecast error with the control variables. The ERC is then the coefficient on the forecast error. GX use the size adjusted buy and hold abnormal return over a 90-day window (BHAR90) and a 3-day window (BHAR3) ending 1 day after the earnings announcement date (EAD). The most recent forecast at the time of the earnings announcement of each analyst is used in computing the consensus forecast error. The BHAR90 returns, therefore, straddle the forecasts (F), which in turn represents the time when market expectations are formed. Therefore, the abnormal returns do not merely reflect the response to the earnings surprise but also the abnormal returns prior to the formation of earnings expectations. Consequently, the coefficient on the earnings surprise can no longer be interpreted as the ERC. Below is a time line that represents the returns aggregation period and the timing of forecasts in a quarter. While GX do find significant results using BHAR90, their results using BHAR3 are not significant. This inconsistency raises questions because BHAR90 does not accurately align investors’ expectations (based on consensus forecasts) with the returns cumulation window. This problem of misaligned market returns and market expectations is—however—not present for the short horizon tests (BHAR3), which should result in lower measurement error. And yet in these tests that have lower measurement error, GX do not find any evidence of disciplining by or superiority of independent analysts. GX justify reliance on BHAR90 by citing
ARTICLE IN PRESS 320
T.Z. Lys, J. Sunder / Journal of Accounting and Economics 45 (2008) 317–323
BHAR90
F0
BHAR3
F
EAD EAD+1
FE Fig. 1. Time line of ERC tests.
Collins and Kothari (1989). However, Collins and Kothari compute the earnings surprise using a randomwalk model of earnings expectations (and not analyst consensus forecast at the EAD) and therefore the expectations are formed at the start of the returns cumulation window, and not at the forecast issuance date as is the case in the present paper. The inconsistency of the evidence across the two returns windows is puzzling. If independent analysts do not adequately update their forecasts during the quarter, then we might observe that the 90-day ERC alone is higher relative to non-independent analysts despite the lower forecast accuracy. In other words, in Fig. 1, F is not very different from F0, which is the forecast at the start of the 90-day window. This implies that the two groups respond to information during the quarter (e.g., as in stock prices) very differently and that in fact independent analysts under-react to news in prices. Ideally, one would want to use the buy-and-hold return from the issuance date of the last forecast (F) through EAD+1 (see Lys and Sohn (1990) for such an approach). Alternatively, one could solve the misalignment by using the expectations prior to the beginning of the return accumulation period. We implement this approach and compute the earnings expectation based on forecasts made at the start of the quarter (F0). While this aligns the returns window with the expectation formation date, it throws out all the information in analyst forecasts made during the quarter and leading up to the earnings announcement. The main ERC results that GX document using BHAR90 still hold. What remains unresolved in this story is the lack of results for BHAR3. Overall, the lack of consistent findings between BHAR90 and BHAR3 suggests that the forecasts of independent analysts are different and it may reflect poorer quality of forecast revisions of the independent analysts, since their latest forecasts seem to be reflecting market information at the start of the quarter. One has to question whether these mixed results allow the conclusion that independent analysts are superior. However, the disciplining results using BHAR90 are still present even when we use forecasts made at the start of the quarter (F0). This leads us to explore alternative explanations that may be causing the observed results. 4. Alternative explanations An underlying premise of the disciplining hypothesis in GX is that independent analysts do not suffer from conflicts of interest that non-independent analysts face. Therefore, the presence of non-conflicted forecasts by independent analysts makes it harder for non-independent analysts to continue to provide biased forecasts. This disciplining effect typically requires that non-independent analysts perceive that they will be compared unfavorably to the independent analysts if they continue to issue biased and inaccurate forecasts. This in turn assumes that the accuracy in forecasts of independent analysts is higher than that of non-independent analysts, which is contrary to what GX find. It is—therefore—puzzling that the forecasting behavior of nonindependent analysts would vary with the presence of independent analysts who are inferior forecasters. This raises the question of whether competition from inferior players affects the behavior of the other incumbent superior players and why such a reaction is rational. However, there are several possible alternative explanations for why GX find the disciplining effect even with inferior independent analysts. First, the competitive dynamics of the equity research industry forms the backdrop of their study, and this brings with it concerns regarding the endogeneity of entry and exit decisions of analysts following a firm. This is an issue that GX recognize but do not address in their paper, and we discuss this in greater detail in the next section.
ARTICLE IN PRESS T.Z. Lys, J. Sunder / Journal of Accounting and Economics 45 (2008) 317–323
321
Second, independent analysts may choose not to issue forecasts in quarters when the uncertainty regarding earnings is very high, while it may be harder for a big brokerage house to remain silent. This would then result in better quality forecasts in quarters when independent analysts choose to forecast relative to when they are silent, which would seem to support the disciplining hypothesis even though it is driven by the predictability of earnings. Third, while the independent analysts are less accurate, they are also systematically biased. Bias in and of itself contributes to inaccuracy but if it is anticipated, it can be easily undone by investors. Therefore, the measure of accuracy used by GX is not an independent statistic from the bias. If bias is an important contributor to the inaccuracy of independent analysts and if these independent analysts are otherwise better forecasters, then the market can rationally undo this systematic bias to arrive at a superior forecast. If— indeed—the accuracy adjusted for bias is higher for independent analysts, then the higher ERCs and disciplining role of independent analysts documented in the paper is not puzzling. We—therefore—examine the mean-squared error of the forecasts of the analysts to exclude the effect of any systematic bias and find that the forecast mean-squared error of independent analysts continues to be higher than that of nonindependent analysts, which suggests that non-independents are more accurate even after we control for bias. Therefore the disciplining role of independents is hard to justify. Finally, while the paper implies that the presence of independent analysts alters, on average, the behavior of all the non-independent analysts following the firm, it may be the case that the entry of independent analysts drives out the worst of the non-independent analysts. To address this concern, GX use the same group of nonindependent analysts with and without the presence of independent analysts and get similar results in Table 6 of their paper. 5. Endogenous entry and exit by independent analysts Since independent analysts are not randomly assigned to firm quarters, it is important to understand their decision to commence and/or abandon following a given firm. In defense of GX, there is very little guidance from prior literature that provides some structure for modeling the endogeneity of the entry–exit decision of analysts (McNichols and O’Brien (1997) being the lone exception). However, this raises an interesting issue that is worthy of future research, i.e. we need to separate out the conditions that lead to entry or exit of analysts from the competitive effects of their presence or absence. GX is unable to disentangle the two effects and therefore the results have to be viewed with caution. To get a first look at the issue, we reexamine the GX sample, which they generously shared with us (see GX for details on the data and variable measurement). We examine two alternative cases:(i) where IND ¼ 0 quarter precedes the IND ¼ 1 quarter, i.e. when the independent analyst enters (IND_ENTRY) and (ii) when IND ¼ 0 quarter follows the IND ¼ 1 quarter, i.e. when the independent analyst exits (IND_EXIT). We recognize that this is only an approximate approach for separating out entry versus exit cases, because it is possible that some of the independent analysts forecast selectively in certain quarters. For instance, if an independent analyst issues a forecast in t ¼ 8 (IND ¼ 1), did not issue a forecast in t ¼ 4 (IND ¼ 0) and again issued a forecast in t ¼ 0 (IND ¼ 1), and the matched quarter for t ¼ 0 is t ¼ 4, then we would classify this pair of observations as an entry observation even though in t ¼ 4, the analyst exited. Further, since forecasts that are more than 90-days old are dropped from the sample, some IND ¼ 0 quarters could in fact have an independent analyst who is not very active. Another caveat that we would like to highlight is that the sample excludes some important independent analysts, such as ValueLine who cover more than half of the firms in the sample and therefore, IND ¼ 0 quarters are misclassified in those cases. Recognizing potential measurement problems with the data, we examine the disciplining effect separately for cases when the independent analyst enters versus exits using the following specification which is a modified version of Eq. (4) in GX: BHAR90it ¼ a0 þ a1 FE_NIit þ a2 FE_NIit IND_ENTRY þ a3 FE_NIit IND_EXIT þ a4 INDit X X þ bj1 FE_NIit Controls þ bj2 Controls j
j
þ Time and industry dummies þ it
(1)
ARTICLE IN PRESS 322
T.Z. Lys, J. Sunder / Journal of Accounting and Economics 45 (2008) 317–323
The results are reported in Table 1. We find that the disciplining effect of independent analysts is concentrated in the entry subset. Further, the entry of independent analysts that is accompanied by the entry of other non-independent analysts appears to drive the results. This suggests that entry and exit of analysts might per se describe very different settings for the competitive spillover effects beyond the identity of the entering and exiting analyst(s). Further, it is important to understand why the independent analyst chose to enter in a given quarter and whether these underlying factors changed the competitive environment and/or the information environment as well, leading to the observed disciplining effect. However, GX do not control for this effect. As a result, their main results are likely, in fact, to be driven by exogenous events that lead analysts to enter or exit (or at the very least not to issue forecasts). As such, the interpretation of the GX results is severely limited.
Table 1 Disciplining effect of independent analysts: effect of entry versus exit by independent analysts BHAR90 Intercept FE_NI FE_NI IND_ENTRY FE_NI IND_EXIT IND
0.207*** 21.471*** 3.069** 1.592 0.0002
FE_NI Controls FE_NI LGSIZE FE_NI LGFOLLOW FE_NI BETA FE_NI MB FE_NI GROWTH FE_NI STD_ROE FE_NI PSST FE_NI |FE_NI|
1.660*** 2.306** 1.641 0.029 4.114 0.221 1.474 586.4***
Controls LGSIZE LGFOLLOW BETA MB GROWTH STD_ROE PSST |FE_NI| FE_1
0.002 0.013*** 0.011** 0.0001 0.003 0.005 0.015** 0.253 0.795**
Time and industry dummies N Adjusted R2
Yes 8346 0.185
P Model: BHAR90it ¼ a0+a1 FE_NIit+a2 FE_NIit IND_ENTRY+a3 FE_NIit IND_EXIT+a4 INDit+ j bj1 FE_NIit Controls+ P j bj2 controls+Time and industry dummies+eit—(1). BHAR90 is the size-adjusted buy-and-hold abnormal returns over the 90-day window up to 1 day after the earnings announcement, FE_NI is the mean forecast error (actual earnings minus the mean forecast) for the group of non-independent analysts, IND ¼ 1 indicates firm quarters followed by both independent and non-independent analysts, IND_ENTRY denotes firm quarters where IND ¼ 1 and was preceded by IND ¼ 0, i.e. the independent analyst entered and IND_EXIT denotes firms quarters where IND ¼ 1 and are followed by IND ¼ 0 quarters. LGSIZE is the logarithm of market capitalization (in $millions) at the beginning of the quarter, LGFOLLOW is the logarithm of the total number of analysts following the firm in a given quarter. BETA is the Scholes–Williams equity beta for the year provided by CRSP. MB is the market-to-book ratio at the beginning of the quarter. GROWTH is the average growth in book value of equity over the previous four quarters. STD_ROE is the firm-specific standard deviation of return-on-equity (earnings divided by book value of equity) during the sample period. PSST is the autoregressive coefficient estimate from the Foster (1977) model. FE_1 is the consensus forecast error in the previous quarter.
ARTICLE IN PRESS T.Z. Lys, J. Sunder / Journal of Accounting and Economics 45 (2008) 317–323
323
6. Conclusion GX ask an interesting question regarding the role that analyst independence plays in shaping the quality of earnings forecasts and provide evidence on the interaction effects between independent and non-independent analysts. However, given the mixed evidence on the superiority of independent analysts, it is not clear what the mechanism for their disciplining effect is. The authors need to be commended for taking a first step towards understanding competitive effects in the equity research industry, particularly the role that independent analysts play in improving the quality of research. However, the analysis raises several interesting and unanswered questions that future research should consider regarding the endogenous entry and exit of analysts and the impact that it has on the observed quality of research output. Additionally, an important question that remains unanswered is the inconsistent results in the ERC tests, especially the lack of results for the short window tests that have lower measurement error. References Collins, D., Kothari, S.P., 1989. An analysis of intertemporal and cross-sectional determinants of earnings response coefficients. Journal of Accounting and Economics 11, 143–181. Ertimur, Y., Sunder, J., Sunder, S.V., 2007. Measure for measure: the relation between forecast accuracy and recommendation profitability of analysts. Journal of Accounting Research 45 (3). Gu, Z., Xue, J., 2008. The superiority and disciplining role of independent analysts. Journal of Accounting and Economics, this issue, doi:10.1016/j.jacceco.2008.02.002. Jacob, J., Rock, S. and Weber, D., 2005. Conflicts of interest and research quality: evidence from earnings forecasts? Working paper, University of Colorado at Boulder. Lys, T., Sohn, S., 1990. The association between revisions of financial analysts’ earnings forecasts and security price changes. Journal of Accounting and Economics 13, 341–363. McNichols, M., O’Brien, P., 1997. Self-selection and analyst coverage. Journal of Accounting Research 35, 167–199.
Journal of Accounting and Economics 45 (2008) 317–323 www.elsevier.com/locate/jae
Endogenous entry/exit as an alternative explanation for the disciplining role of independent analysts Thomas Z. Lys, Jayanthi Sunder Kellogg Graduate School of Management, Northwestern University, 2001 Sheridan Road, Andersen Hall 517, Evanston, IL 60208-2002, USA Received 7 November 2007; received in revised form 7 February 2008; accepted 11 February 2008 Available online 17 February 2008
Abstract Gu and Xue [2008. The superiority and disciplining role of independent analysts. Journal of Accounting and Economics, this issue, doi:10.1016/j.jacceco.2008.02.002] study the disciplining effect of independent analysts on the accuracy and forecast relevance of the forecasts of non-independent analysts. One of the intriguing results is that while independent analysts issue inferior forecasts, their presence appears to reduce the forecast bias, improve the forecast accuracy and increase the forecast relevance of forecasts issued by non-independent analysts. We explore alternative explanations for the Gu–Xue results. Our evidence of endogenous entry and exit of independent analysts provides a more compelling explanation for the reported results. r 2008 Elsevier B.V. All rights reserved. JEL classification: G28; G29; M41; M43 Keywords: Endogenous entry; Disciplining; Independent
1. A straight-forward hypothesis with intriguing results Gu and Xue (2008, hereafter GX) study the role of independent analysts (i.e., analysts with no ties to companies whose earnings are being forecasted) in improving the quality of forecasts of non-independent analyst (i.e., analyst with potential conflicts of interest, for example investment banking relations). The basic premise of the paper is that independent analysts should, all else equal, issue less-biased and more precise forecasts. More importantly, the presence of independent analysts should have a disciplining effect on nonindependent analysts, resulting in less-biased and more accurate forecasts by non-independent analysts as well. The importance of GX’s research is highlighted by recent regulatory initiatives, culminating in the Global Research Settlement, which aim to address issues relating to the lack of independence of equity research analysts. While these regulatory initiatives by and large focus on stock recommendations, they are based on the assumption that lack of analyst independence impedes the quality of equity research. Therefore, GX set out to examine whether independent analysts are superior forecasters of earnings and whether their presence disciplines non-independent analysts following the same company. Corresponding author.
E-mail address: [email protected] (T.Z. Lys). 0165-4101/$ - see front matter r 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.jacceco.2008.02.001
ARTICLE IN PRESS 318
T.Z. Lys, J. Sunder / Journal of Accounting and Economics 45 (2008) 317–323
Several papers have compared the quality of research of independent analysts with non-independent analysts. GX adds to this stream of literature in a novel manner by studying the interaction between these two groups of analysts. This basic thesis is both simple and intuitive. Further, despite the richness of the equity research industry for addressing fundamental questions relating to the dynamics of competition, there is little prior research that examines the role of competition and GX take a first step in that direction. GX ask two questions. First, are independent analysts’ forecasts less-biased and more precise than those of non-independent analysts? Second, does the presence of independent analysts discipline non-independent analysts following the same firm by reducing bias, increasing precision, and increasing the credibility of forecasts of the latter group? GX address these issues by comparing the accuracy, bias, and earnings response coefficient (ERC) of earnings forecasts for firms that were followed by only non-independent analysts in some quarters to forecasts for the same firms when they were followed by both independent and non-independent analyst in other quarters. At first, the hypothesis tested by GX seems straight-forward: introducing independent analysts who compete with incumbent non-independent analysts increases competition and ought to improve the quality of the forecasts of the non-independent analysts. And indeed, GX report that, in general, the presence of independent analysts reduces the bias, increases the accuracy, and increases the ERC of forecasts issued by non-independent analysts. What makes their research intriguing is the result that, forecasts issued by independent analysts while less-biased, is significantly less precise than forecasts issued by non-independent analysts. Thus, GX document a puzzling phenomenon, namely that introducing an otherwise inferior competitor induces superior competitors to increase the quality of their output and become more credible. 2. Evidence and interpretation of results 2.1. Findings Using absolute forecast errors as a measure of forecast accuracy, GX find that independent analysts are less accurate than non-independent analysts.1 Moreover, using average forecast error as a measure of forecast bias, while both groups of analysts are on average pessimistically biased, independent analysts are less pessimistic than non-independent analysts. The accuracy results suggest that forecasts issued by independent analysts are in fact inferior to forecasts issued by non-independent analysts. As indicated in the introduction, GX main result, however, is that forecast accuracy of non-independent analysts is higher and their forecasts are less pessimistically biased in periods when independent analysts are also issuing forecasts compared to periods when independent analysts are not issuing forecasts. GX interpret this as the disciplining effect of independent analysts. With respect to credibility, GX find that the ERC for independent analysts is higher, particularly when the earnings news is bad, than the ERC of non-independent analysts. Moreover, GX find that the ERCs of nonindependent analysts are higher in the presence of an independent analyst following the same firm compared to the ERC of non-independent analysts when independent analysts are absent. This result is further evidence of the disciplining role of independent analysts. Based on these results, GX conclude that the presence of independent analysts has a disciplining effect on the non-independent analysts and therefore suggest that improving analyst independence enhances the quality of analyst research. With respect to the superiority of independent analysts, GX find mixed results with forecasts of independent analysts being less accurate but being more credible (having higher ERCs). However, GX argue that higher ERCs imply better quality forecasts since it suggests that the forecasts are better aligned with market expectations. The mixed evidence with respect to the superiority of independent analysts raises some interesting puzzles. How is it that the ERC of independent analysts is higher even though their forecast accuracy is lower than that of non-independent analysts? Further, how can the independent analysts discipline the non-independent analysts while their forecasts have lower forecast accuracy? We address the puzzling results using two 1
This is consistent with evidence on forecast accuracy in other recent papers such as Jacob, Rock and Weber (2005) and Ertimur, Sunder and Sunder (2007).
ARTICLE IN PRESS T.Z. Lys, J. Sunder / Journal of Accounting and Economics 45 (2008) 317–323
319
approaches. First, we examine the variables and tests used to establish the superiority and disciplining effect and second, we explore potential mechanisms through which independent analysts discipline the nonindependents and generate alternative explanations for the reported results. 2.2. Measures of forecast quality 2.2.1. Measuring accuracy independent of bias GX measure the accuracy of forecasts of a group of analysts using two metrics that both employ the absolute error. The first (ACCY1) is based on the absolute error in the mean forecast and the second, (ACCY2) uses the mean of individual absolute forecast errors. While these measures capture the distance between forecasted earnings and actual earnings, they are not independent of their measure of forecast bias (FE). Bias is measured as the signed forecast error and this creates a problem because a systematic bias mechanically results in a larger mean absolute error. As an alternative, accuracy can be measured using the mean-squared error, which adjusts for the mean forecast error and—therefore—is not affected by any systematic bias component. However, using the mean-squared error, we continue to find that non-independent analysts are better than independent analysts. 2.2.2. ERC as a measure of forecast quality The ERC measure captures the sensitivity of the market reaction to the earnings news and—therefore—as the measured earnings surprise gets closer to the true earnings surprise of the market, we would expect a higher ERC. GX find that the ERC associated with independent forecasts is higher than that for nonindependent forecasts suggesting that, investors pay greater attention to the independent analysts while arriving at the market expectation. In that case, the results in GX should be interpreted as suggesting that the credibility of independent analysts is higher with investors and the presence of independent analysts improves the credibility of non-independent analysts. This does not, however, necessarily imply that the credibility improves as a consequence of a change in forecasting behavior of the non-independent analysts, because investors may (possibly irrationally) believe that these forecasts are more credible. If the credibility of the forecasts issued by independent analysts is higher, one would expect a greater market reaction to forecast revisions issued by independent analysts. Contrary to this expectation, in unreported tests, we find that the reaction is greater for non-independent analysts. However, the magnitude of the reaction is greater in quarters when independent analysts are present. Because of these seemingly inconsistent results, we next discuss the actual implementation of the ERC tests and the implications for the reported results. 3. ERC tests The basic ERC test involves a regression of abnormal returns to the earnings announcement of a given firm on the consensus forecast error, control variables and the interactions of the forecast error with the control variables. The ERC is then the coefficient on the forecast error. GX use the size adjusted buy and hold abnormal return over a 90-day window (BHAR90) and a 3-day window (BHAR3) ending 1 day after the earnings announcement date (EAD). The most recent forecast at the time of the earnings announcement of each analyst is used in computing the consensus forecast error. The BHAR90 returns, therefore, straddle the forecasts (F), which in turn represents the time when market expectations are formed. Therefore, the abnormal returns do not merely reflect the response to the earnings surprise but also the abnormal returns prior to the formation of earnings expectations. Consequently, the coefficient on the earnings surprise can no longer be interpreted as the ERC. Below is a time line that represents the returns aggregation period and the timing of forecasts in a quarter. While GX do find significant results using BHAR90, their results using BHAR3 are not significant. This inconsistency raises questions because BHAR90 does not accurately align investors’ expectations (based on consensus forecasts) with the returns cumulation window. This problem of misaligned market returns and market expectations is—however—not present for the short horizon tests (BHAR3), which should result in lower measurement error. And yet in these tests that have lower measurement error, GX do not find any evidence of disciplining by or superiority of independent analysts. GX justify reliance on BHAR90 by citing
ARTICLE IN PRESS 320
T.Z. Lys, J. Sunder / Journal of Accounting and Economics 45 (2008) 317–323
BHAR90
F0
BHAR3
F
EAD EAD+1
FE Fig. 1. Time line of ERC tests.
Collins and Kothari (1989). However, Collins and Kothari compute the earnings surprise using a randomwalk model of earnings expectations (and not analyst consensus forecast at the EAD) and therefore the expectations are formed at the start of the returns cumulation window, and not at the forecast issuance date as is the case in the present paper. The inconsistency of the evidence across the two returns windows is puzzling. If independent analysts do not adequately update their forecasts during the quarter, then we might observe that the 90-day ERC alone is higher relative to non-independent analysts despite the lower forecast accuracy. In other words, in Fig. 1, F is not very different from F0, which is the forecast at the start of the 90-day window. This implies that the two groups respond to information during the quarter (e.g., as in stock prices) very differently and that in fact independent analysts under-react to news in prices. Ideally, one would want to use the buy-and-hold return from the issuance date of the last forecast (F) through EAD+1 (see Lys and Sohn (1990) for such an approach). Alternatively, one could solve the misalignment by using the expectations prior to the beginning of the return accumulation period. We implement this approach and compute the earnings expectation based on forecasts made at the start of the quarter (F0). While this aligns the returns window with the expectation formation date, it throws out all the information in analyst forecasts made during the quarter and leading up to the earnings announcement. The main ERC results that GX document using BHAR90 still hold. What remains unresolved in this story is the lack of results for BHAR3. Overall, the lack of consistent findings between BHAR90 and BHAR3 suggests that the forecasts of independent analysts are different and it may reflect poorer quality of forecast revisions of the independent analysts, since their latest forecasts seem to be reflecting market information at the start of the quarter. One has to question whether these mixed results allow the conclusion that independent analysts are superior. However, the disciplining results using BHAR90 are still present even when we use forecasts made at the start of the quarter (F0). This leads us to explore alternative explanations that may be causing the observed results. 4. Alternative explanations An underlying premise of the disciplining hypothesis in GX is that independent analysts do not suffer from conflicts of interest that non-independent analysts face. Therefore, the presence of non-conflicted forecasts by independent analysts makes it harder for non-independent analysts to continue to provide biased forecasts. This disciplining effect typically requires that non-independent analysts perceive that they will be compared unfavorably to the independent analysts if they continue to issue biased and inaccurate forecasts. This in turn assumes that the accuracy in forecasts of independent analysts is higher than that of non-independent analysts, which is contrary to what GX find. It is—therefore—puzzling that the forecasting behavior of nonindependent analysts would vary with the presence of independent analysts who are inferior forecasters. This raises the question of whether competition from inferior players affects the behavior of the other incumbent superior players and why such a reaction is rational. However, there are several possible alternative explanations for why GX find the disciplining effect even with inferior independent analysts. First, the competitive dynamics of the equity research industry forms the backdrop of their study, and this brings with it concerns regarding the endogeneity of entry and exit decisions of analysts following a firm. This is an issue that GX recognize but do not address in their paper, and we discuss this in greater detail in the next section.
ARTICLE IN PRESS T.Z. Lys, J. Sunder / Journal of Accounting and Economics 45 (2008) 317–323
321
Second, independent analysts may choose not to issue forecasts in quarters when the uncertainty regarding earnings is very high, while it may be harder for a big brokerage house to remain silent. This would then result in better quality forecasts in quarters when independent analysts choose to forecast relative to when they are silent, which would seem to support the disciplining hypothesis even though it is driven by the predictability of earnings. Third, while the independent analysts are less accurate, they are also systematically biased. Bias in and of itself contributes to inaccuracy but if it is anticipated, it can be easily undone by investors. Therefore, the measure of accuracy used by GX is not an independent statistic from the bias. If bias is an important contributor to the inaccuracy of independent analysts and if these independent analysts are otherwise better forecasters, then the market can rationally undo this systematic bias to arrive at a superior forecast. If— indeed—the accuracy adjusted for bias is higher for independent analysts, then the higher ERCs and disciplining role of independent analysts documented in the paper is not puzzling. We—therefore—examine the mean-squared error of the forecasts of the analysts to exclude the effect of any systematic bias and find that the forecast mean-squared error of independent analysts continues to be higher than that of nonindependent analysts, which suggests that non-independents are more accurate even after we control for bias. Therefore the disciplining role of independents is hard to justify. Finally, while the paper implies that the presence of independent analysts alters, on average, the behavior of all the non-independent analysts following the firm, it may be the case that the entry of independent analysts drives out the worst of the non-independent analysts. To address this concern, GX use the same group of nonindependent analysts with and without the presence of independent analysts and get similar results in Table 6 of their paper. 5. Endogenous entry and exit by independent analysts Since independent analysts are not randomly assigned to firm quarters, it is important to understand their decision to commence and/or abandon following a given firm. In defense of GX, there is very little guidance from prior literature that provides some structure for modeling the endogeneity of the entry–exit decision of analysts (McNichols and O’Brien (1997) being the lone exception). However, this raises an interesting issue that is worthy of future research, i.e. we need to separate out the conditions that lead to entry or exit of analysts from the competitive effects of their presence or absence. GX is unable to disentangle the two effects and therefore the results have to be viewed with caution. To get a first look at the issue, we reexamine the GX sample, which they generously shared with us (see GX for details on the data and variable measurement). We examine two alternative cases:(i) where IND ¼ 0 quarter precedes the IND ¼ 1 quarter, i.e. when the independent analyst enters (IND_ENTRY) and (ii) when IND ¼ 0 quarter follows the IND ¼ 1 quarter, i.e. when the independent analyst exits (IND_EXIT). We recognize that this is only an approximate approach for separating out entry versus exit cases, because it is possible that some of the independent analysts forecast selectively in certain quarters. For instance, if an independent analyst issues a forecast in t ¼ 8 (IND ¼ 1), did not issue a forecast in t ¼ 4 (IND ¼ 0) and again issued a forecast in t ¼ 0 (IND ¼ 1), and the matched quarter for t ¼ 0 is t ¼ 4, then we would classify this pair of observations as an entry observation even though in t ¼ 4, the analyst exited. Further, since forecasts that are more than 90-days old are dropped from the sample, some IND ¼ 0 quarters could in fact have an independent analyst who is not very active. Another caveat that we would like to highlight is that the sample excludes some important independent analysts, such as ValueLine who cover more than half of the firms in the sample and therefore, IND ¼ 0 quarters are misclassified in those cases. Recognizing potential measurement problems with the data, we examine the disciplining effect separately for cases when the independent analyst enters versus exits using the following specification which is a modified version of Eq. (4) in GX: BHAR90it ¼ a0 þ a1 FE_NIit þ a2 FE_NIit IND_ENTRY þ a3 FE_NIit IND_EXIT þ a4 INDit X X þ bj1 FE_NIit Controls þ bj2 Controls j
j
þ Time and industry dummies þ it
(1)
ARTICLE IN PRESS 322
T.Z. Lys, J. Sunder / Journal of Accounting and Economics 45 (2008) 317–323
The results are reported in Table 1. We find that the disciplining effect of independent analysts is concentrated in the entry subset. Further, the entry of independent analysts that is accompanied by the entry of other non-independent analysts appears to drive the results. This suggests that entry and exit of analysts might per se describe very different settings for the competitive spillover effects beyond the identity of the entering and exiting analyst(s). Further, it is important to understand why the independent analyst chose to enter in a given quarter and whether these underlying factors changed the competitive environment and/or the information environment as well, leading to the observed disciplining effect. However, GX do not control for this effect. As a result, their main results are likely, in fact, to be driven by exogenous events that lead analysts to enter or exit (or at the very least not to issue forecasts). As such, the interpretation of the GX results is severely limited.
Table 1 Disciplining effect of independent analysts: effect of entry versus exit by independent analysts BHAR90 Intercept FE_NI FE_NI IND_ENTRY FE_NI IND_EXIT IND
0.207*** 21.471*** 3.069** 1.592 0.0002
FE_NI Controls FE_NI LGSIZE FE_NI LGFOLLOW FE_NI BETA FE_NI MB FE_NI GROWTH FE_NI STD_ROE FE_NI PSST FE_NI |FE_NI|
1.660*** 2.306** 1.641 0.029 4.114 0.221 1.474 586.4***
Controls LGSIZE LGFOLLOW BETA MB GROWTH STD_ROE PSST |FE_NI| FE_1
0.002 0.013*** 0.011** 0.0001 0.003 0.005 0.015** 0.253 0.795**
Time and industry dummies N Adjusted R2
Yes 8346 0.185
P Model: BHAR90it ¼ a0+a1 FE_NIit+a2 FE_NIit IND_ENTRY+a3 FE_NIit IND_EXIT+a4 INDit+ j bj1 FE_NIit Controls+ P j bj2 controls+Time and industry dummies+eit—(1). BHAR90 is the size-adjusted buy-and-hold abnormal returns over the 90-day window up to 1 day after the earnings announcement, FE_NI is the mean forecast error (actual earnings minus the mean forecast) for the group of non-independent analysts, IND ¼ 1 indicates firm quarters followed by both independent and non-independent analysts, IND_ENTRY denotes firm quarters where IND ¼ 1 and was preceded by IND ¼ 0, i.e. the independent analyst entered and IND_EXIT denotes firms quarters where IND ¼ 1 and are followed by IND ¼ 0 quarters. LGSIZE is the logarithm of market capitalization (in $millions) at the beginning of the quarter, LGFOLLOW is the logarithm of the total number of analysts following the firm in a given quarter. BETA is the Scholes–Williams equity beta for the year provided by CRSP. MB is the market-to-book ratio at the beginning of the quarter. GROWTH is the average growth in book value of equity over the previous four quarters. STD_ROE is the firm-specific standard deviation of return-on-equity (earnings divided by book value of equity) during the sample period. PSST is the autoregressive coefficient estimate from the Foster (1977) model. FE_1 is the consensus forecast error in the previous quarter.
ARTICLE IN PRESS T.Z. Lys, J. Sunder / Journal of Accounting and Economics 45 (2008) 317–323
323
6. Conclusion GX ask an interesting question regarding the role that analyst independence plays in shaping the quality of earnings forecasts and provide evidence on the interaction effects between independent and non-independent analysts. However, given the mixed evidence on the superiority of independent analysts, it is not clear what the mechanism for their disciplining effect is. The authors need to be commended for taking a first step towards understanding competitive effects in the equity research industry, particularly the role that independent analysts play in improving the quality of research. However, the analysis raises several interesting and unanswered questions that future research should consider regarding the endogenous entry and exit of analysts and the impact that it has on the observed quality of research output. Additionally, an important question that remains unanswered is the inconsistent results in the ERC tests, especially the lack of results for the short window tests that have lower measurement error. References Collins, D., Kothari, S.P., 1989. An analysis of intertemporal and cross-sectional determinants of earnings response coefficients. Journal of Accounting and Economics 11, 143–181. Ertimur, Y., Sunder, J., Sunder, S.V., 2007. Measure for measure: the relation between forecast accuracy and recommendation profitability of analysts. Journal of Accounting Research 45 (3). Gu, Z., Xue, J., 2008. The superiority and disciplining role of independent analysts. Journal of Accounting and Economics, this issue, doi:10.1016/j.jacceco.2008.02.002. Jacob, J., Rock, S. and Weber, D., 2005. Conflicts of interest and research quality: evidence from earnings forecasts? Working paper, University of Colorado at Boulder. Lys, T., Sohn, S., 1990. The association between revisions of financial analysts’ earnings forecasts and security price changes. Journal of Accounting and Economics 13, 341–363. McNichols, M., O’Brien, P., 1997. Self-selection and analyst coverage. Journal of Accounting Research 35, 167–199.