Financial analyst journal: - Rutgers University



The Impact of Stock Recommendation-Earnings Forecast Consistency on Forecast Accuracy and Recommendation Profitability

Lawrence D. Brown

Georgia State University

and

Kelly Huang

Georgia State University

October 2009

We gratefully acknowledge the helpful suggestions of Jeffrey Callen, Xia Chen, Qiang Cheng, Ole-Kristian Hope, Hai Lu, Gordon Richardson, Kent Womack, Franco Wong and the participants of the University of Toronto and University of Wisconsin workshops.

The Impact of Stock Recommendation-Earnings Forecast Consistency on Forecast Accuracy and Recommendation Profitability

Abstract

We examine how the consistency between an analyst’s stock recommendation and her one-year-ahead earnings forecast impacts forecast accuracy and stock recommendation profitability. Defining consistency as the analyst’s stock recommendation and earnings forecast are both above (below) the prevailing consensus, we show analysts are consistent less than 60% of the time. We find consistent analysts to make relatively more accurate forecasts and more profitable recommendations. We benchmark the importance of consistency on recommendation profitability in two ways. First, we compare ex ante consistency with ex post earnings forecast accuracy. We show that consistency is more valuable for recommendation profitability than is forecast accuracy, and that perfect foreknowledge of earnings forecast accuracy is more valuable when it is combined with consistency. Second, we compare consistency with ex ante boldness. We demonstrate that consistency is more valuable for recommendation profitability than is boldness and that boldness is more valuable when it is combined with consistency.

The Impact of Stock Recommendation-Earnings Forecast Consistency on Forecast Accuracy and Recommendation Profitability

1. Introduction

We investigate how stock recommendation-earnings forecast consistency impacts earnings forecast accuracy and recommendation profitability. We define consistency as the same analyst’s stock recommendation and earnings forecast are above (below) the respective consensus. Otherwise, the analyst is said to be inconsistent regarding the stock recommendation-earnings forecast in question.[1]

We show that consistent analysts, who constitute nearly 60% of our sample, make more accurate earnings forecasts and more profitable recommendations. Our evidence is consistent with the notion that consistent analysts are more likely than inconsistent analysts to focus on firm fundamentals and/or possess better information regarding firms’ future prospects, making their earnings forecasts and recommendations more precise.[2]

The literature examining the relation between stock recommendation profitability and earnings forecast accuracy (Loh and Mian 2006; Ertimur, Sunder and Sunder 2007) shows analysts who make more accurate earnings forecasts ex post make more profitable stock recommendations ex ante. We benchmark the impact of ex ante consistency on stock recommendation profitability by comparing it with ex post accuracy’s impact on stock recommendation profitability.[3] We find consistency is more valuable than accuracy for recommendation profitability, and accuracy is much more useful when it is employed in conjunction with consistency.

The literature examining the relation between stock recommendation profitability and recommendation boldness (Jagadeesh and Kim 2009, Loh and Stulz 2009) has shown that bold recommendations are more profitable than non-bold recommendations. As a second way to benchmark the impact of consistency on recommendation profitability, we examine boldness, another ex ante measure for improving recommendation profitability. We show consistency is more valuable than boldness for recommendation profitability, and that boldness is far more useful when it is used in conjunction with consistency.

In sum, stock recommendation-earnings forecast consistency is important for both accuracy of analysts’ earnings forecasts and profitability of their stock recommendations. Indeed, it is more important for accuracy than most factors heretofore shown to explain accuracy, and it is a more important determinant of recommendation profitability than either forecast accuracy or recommendation boldness. Moreover, forecast accuracy and boldness are far more valuable when they are used in conjunction with consistency.

Our study has several important implications. First, we add to the forecasting literature by showing that consistency is an important determinant of analyst forecast accuracy. Second, we add to the stock recommendation literature by providing a simple rule investors can use to increase their expected returns from stock recommendations, and a benchmark academics can use to compare the profitability of other trading rules. Third, we add to earnings forecast accuracy-recommendation profitability literature by showing that investors should heed the advice of accurate analysts who are consistent, but ignore advice of accurate analysts who are inconsistent. Fourth, we supplement the literature on profitability of bold recommendations by showing that investors are far better off heeding the advice of bold, consistent analysts than bold, inconsistent analysts.

We proceed as follows. Section 2 discusses the related literature and puts forth our hypotheses. Section 3 describes our data selection and research design. Section 4 presents our results. Section 5 reports our additional analyses and section 6 concludes.

2. Related Studies and Hypotheses

Early studies showed that stock returns are associated with both analysts’ stock recommendations (e.g., Womack 1996) and their earnings forecasts (e.g., Lys and Sohn 1990). Later studies investigated the informativeness of analyst reports (e.g., Francis and Soffer 1997, Brav and Lehavy 2003, Asquith, Mikhail, and Au 2005). These studies generally concluded that each output of an analyst’s report contains distinct information (e.g. recommendations, earnings forecasts, and target prices) and the market’s reaction to each output depends on the nature of the other outputs in the report.

More recent studies consider the role of earnings forecasts as inputs in generating recommendations and assess if analysts translate their earnings forecasts into profitable recommendations. Bradshaw (2004) shows analysts rely on simple heuristics, e.g., price-earnings-growth ratios, for their recommendations, and investors can earn higher returns by using present value models incorporating analysts’ short-term and long-term earnings forecasts. Barniv, Hope, Myring, and Thomas (2009) and Chen and Chen (2009) examine how recent regulations have impacted analysts’ use of earnings forecasts, and conclude they have mitigated the influence of investment banking relationships and strengthened analysts’ translational effectiveness.

Prior research suggests that analysts are more biased in their stock recommendations than their earnings forecasts (e.g. Dugar and Nathan 1995, Lin and McNichols 1998). More recently, Malmendier and Shanthikumar (2007) find that analysts with conflicting interests may distort recommendations upwards to trigger small-investor purchases and to please management, but may distort forecasts downwards before earnings announcements to allow managers to report earnings that meet or beat the earnings forecast. Ke and Yu (2009) confirm that conflicts of interest, such as investment banking and intuitional ownership pressure, lead analysts to not use their forecasts in their recommendations, and behavioral biases, such as analysts’ use of investor sentiment, lead analysts to rely too little on their earnings forecasts for their recommendations. Bagnoli, Clement, Crawley, and Watts (2009) show analysts do not effectively utilize investor sentiment, and issue less profitable recommendations when they use sentiment, ignoring firm fundamentals. In sum, both analysts’ economic incentives and cognitive biases lead them to sub-optimally use value relevant information in making their stock recommendations. We suggest that consistency between two information signals, earnings forecasts and recommendations, indicates that analysts make more efficient use of value-relevant information in both their earnings forecasts and their stock recommendations.

Apart from analyst biases, the precision of the information analysts receive affects their use of the information. All signals contain a mixture of information and noise. The lower the amount of noise relative to the amount of information, the greater is the signal’s precision and thus its potential usefulness. Consistency between an analyst’s earnings forecast and her recommendation reflects the strength and precision of her information. Information signals that are stronger and/or more precise are more likely to be translated into consistent recommendations and earnings forecasts. If an analyst’s earnings forecast is more precise, it should be more accurate. If an analyst’s stock recommendation is more precise, it should be more profitable. More formally, our first two hypotheses are:

H1:  Consistent analysts’ earnings forecasts are more accurate.

H2:  Consistent analysts’ stock recommendations are more profitable.

Loh and Mian (2006) demonstrate that analysts who make more accurate earnings forecasts ex post make more profitable stock recommendations ex ante. Ertimur, Sunder, and Sunder (2007) extend this literature by considering analysts’ conflicts of interest, value relevance of earnings, and regulatory regime. They show the relation between earnings forecast accuracy and recommendation profitability is related to these three factors, but they do not examine whether recommendation-earnings forecast consistency impacts the relation between forecast accuracy and recommendation profitability.

Ball and Brown (1968) and scores of subsequent studies show that pre-knowledge of one-year-ahead earnings is value relevant. The “Ball and Brown effect” suggests that investors with advance knowledge of ex post forecast accuracy should take long (short) positions in stocks that accurate analysts contend will have earnings increases (decreases). Thus, investors should heed stock recommendations of accurate analysts if their stock recommendations are consistent with their earnings forecasts. On the other hand, when accurate earnings forecasters make recommendations that are inconsistent with their earnings forecasts, stock prices will move in the direction of their earnings forecasts so investors should pay less heed to their stock recommendations. More formally, our third hypothesis, which pertains to accurate earnings forecasters, is:

H3: Stock recommendations of accurate earnings forecasters are more relatively more profitable when they are consistent with their earnings forecasts.

Analyst herding has received considerable attention in the forecasting literature (Hong, Kubik, and Solomon 2000, Clement and Tse 2005). Some studies conclude that analysts herd towards consensus based on little information (e.g. Welch 2000). Jagadeesh and Kim (2009) and Loh and Stulz (2009) find recommendation changes that are further away from the consensus recommendation generate higher returns. Other studies (e.g., Chen and Jiang 2006) argue that herding is attributable to analysts’ responses to common information and, after controlling for common information, analysts move away from the consensus (i.e., they “anti-herd”). We expect that when analysts are consistent, their bold recommendations are more likely to be based on more precise information regarding firm fundamentals. Our fourth hypothesis, which pertains to bold analysts, is:

H4: Stock recommendations of bold analysts are relatively more profitable when they are consistent with their earnings forecasts.

3. Data Description and Research Design

We obtain stock recommendations and earnings forecasts from I/B/E/S and stock returns from CRSP. Because IBES initiated its recommendations in 1993, our sample is for the 15-year period, 1993-2007. IBES codes recommendations using a five-point scale, ranging from 1 (strong buy) to 5 (strong sell). We require our sample to meet several requirements. First, analysts must make their recommendations and earnings forecasts on the same day so we can define recommendation-forecast consistency unambiguously. Second, firms must be followed by at least three analysts during a calendar year to allow reliable estimates of both the recommendation consensus and the earnings forecast consensus. Third, we only include analysts who issue at least three forecasts for the firm during a particular calendar year to ensure that the analyst follows the firm actively. Fourth, we eliminate cases when the earnings forecast equals the forecast consensus or the recommendation equals the recommendation consensus.[4] Fifth, we require appropriate stock return data for examining the capital market’s reaction to recommendations. These requirements result in 67,447 analyst recommendation-forecast observations for our primary univariate tests. Sample sizes differ for different tests due to various data requirements and research design choices so we report sample sizes in all tables for purposes of clarity.

We classify recommendation-forecast consistency based on the information in the forecast (recommendation) relative to the respective consensus. This measure is based on the assumption that analysts know the consensus earnings forecast and the consensus recommendation, and convey their incremental information through their own earnings forecasts and recommendations. Consistency (hereafter CON) exists if both the stock recommendation is above (below) its consensus and the earnings forecast is above (below) its consensus. To calculate consensus, we use the mean of forecasts (recommendations) issued during the 90-calendar days prior to the joint recommendation forecast issue date.

Consistent with Loh and Mian (2006) and Ertimur et al. (2007), we classify a forecast as accurate (hereafter ACC) if it is ex post more accurate than the consensus. We investigate our first hypothesis by regressing ACC on CON and other determinants of analyst forecast accuracy (Mikhail, Walther and Willis 1997; Clement 1999; Jacob, Lys and Neale 1999; Brown 2001), i.e., firm experience (FEXP), number of firms (NFIRMS) followed, number of forecasts the analyst makes for the firm (FREQ), size of brokerage house the analyst works for (BSIZE), forecast horizon (HORIZON), and past accuracy (PACC).[5] To allow for comparisons of regression coefficients and to control for firm and year effects, we scale all variables (except ACC and HORIZON) to range from 0 to 1 using the method in Clement and Tse (2005). The transformed variables for analyst i take the form:

Characteristics i,j,t = Raw-Characteristic i,j,t - Min-Characteristic j,t

Range of Characteristic j,t

where Raw-characteristic is the value for analyst i and Min-characteristic (Range of characteristic) is the minimum value (range) of all analysts following firm j in year t. For forecast horizon, we rank the raw values of horizon into deciles and scale them into a range of (0, 1).[6] We model accuracy (ACC = 1 if accurate and 0 if INACC) with the following logistic regression with the scaled independent variables defined above.

Prob (ACC=1) = α0 + α1CON + α2 FEXP + α3NFIRMS + α4 FREQ+ α5BSIZE + α6 HORIZON + α7PACC (1)

If consistent analysts are more likely to be accurate (our first hypothesis), the coefficient on CON should be positive and significant.

We perform our return analyses over the 32-trading day period, the day before the recommendation-earnings forecast day through the 30th day after it. We start with the day before the recommendation-earnings forecast as information in analyst research often leaks to the capital markets before its official announcement (Irvine, Lipson, and Puckett 2007).[7] Recommendation profitability (PFT) is based on the cumulative raw returns minus cumulative value-weighted market returns from day (-1, 30), where day 0 is the joint recommendation-earnings forecast day. We measure recommendation profitability based on two alternative trading strategies. The first strategy takes long positions in buy recommendations, and short positions in hold and sell recommendations. The second strategy takes long positions in recommendations that are above the recommendation consensus and short positions in those that are below the recommendation consensus.

We assess the incremental effect of consistency on recommendation profitability (H2) by running the following multiple regression:

PFT = α0 + α1CON + α2 FEXP + α3NFIRMS + α4FREQ + α5BSIZE + α6LFR (2)

All the independent variables with the exception of LFR, the leader-follower-ratio, have been defined above. LFR represents the timeliness of analyst forecasts (Cooper, Day, and Lewis 2001) and is defined as the cumulative number of days between the two immediately preceding forecasts and the particular analyst’s forecast divided by the cumulative number of days between the two immediately succeeding forecasts and the particular analyst’s forecast. We include it for comparability with Ertimur et al. (2007). In model (2), α0 is the average return to inconsistent recommendations and α1 is the incremental return to consistent recommendations. If consistent analysts make more profitable recommendations (our second hypothesis) the coefficient on CON should be positive and significant.

We assess the impact of CON on recommendation profitability by benchmarking it against the impact of ACC on recommendation profitability. We run the following regression, which is identical to equation (2) except that it substitutes ACC for CON.

PFT = α0 + α1ACC + α2 FEXP + α3NFIRMS + α4FREQ + α5BSIZE + α6LFR (3)

The evidence in Loh and Mian (2006) and Ertimur et al. (2007) suggests that ACC should be positive and significant. We expect this result with our sample. In order to test our third hypothesis, we run the following regression, which contains CON, ACC and CON_ACC and the other control variables in equations (2) and (3). [8]

PFT = α0 + α1CON + α2 ACC + α3 CON_ACC + α4FEXP + α5NFIRMS + α6FREQ + α7 BSIZE + α8LFR (4)

α0 represents profitability to recommendations issued by inconsistent, inaccurate analysts; α1 represents profitability to recommendations issued by consistent, inaccurate analysts; α2 represents profitability to recommendations issued by accurate, inconsistent analysts; and α3 represents profitability to recommendations issued by consistent, accurate analysts. If accuracy is more useful for recommendation profitability when analysts are consistent (our third hypothesis), α3 should exceed α2.

We also assess the impact of consistency for recommendation profitability by benchmarking it against boldness’s impact on recommendation profitability. We classify recommendations as bold if the distance between the individual recommendation and the prevailing consensus recommendation is greater than one level (e.g., strong buy relative to a consensus of hold). If so, BOLD is a dummy variable equal to one 1 (otherwise 0). We run the following regression, which is identical to equation (3) except it uses BOLD in lieu of ACC.

PFT = α0 + α1BOLD + α2 FEXP + α3NFIRMS + α4 FREQ + α5 BSIZE + α6LFR (5)

The findings of Jegadeesh and Kim (2009) and Loh and Stulz (2009) suggest that BOLD should be positive and significant. We expect to find this result in our sample. In order to test our third hypothesis, we run the following regression, which is similar to equation (4) except it contains BOLD and CON_BOLD in lieu of ACC and CON_ACC. [9]

PFT = α0 + α1CON + α2 BOLD + α3 CON_BOLD + α4FEXP + α5NFIRMS + α6FREQ + α7 BSIZE + α8LFR (6)

α1 is profitability to inconsistent, non-bold recommendations; α1 is profitability to consistent, non-bold recommendations; α2 is profitability to bold, inconsistent ones; and α3 is profitability to consistent, bold recommendations. If boldness is more useful for recommendation profitability when analysts are consistent (our fourth hypothesis), α3 should exceed α2.

4. Results

4.1 Consistency Frequency between Recommendations and Earnings Forecasts

Table 1 reports the consistency frequency (CON%) between recommendations and earnings forecasts. Panel A shows that analyst consistency exists 58.9% of the time. Panel B shows consistency for each of the five recommendation levels: strong buy, buy, hold, sell and strong sell. Similar to Barber, Lehavy, McNichols, and Trueman (2006), most recommendations (35,847) are buys and strong buys; few (5,056) are sells and strong sells. Analysts are more likely to be consistent when recommending investors take short positions (hold, sell or strong sell) rather than long positions (buy or strong buy). We henceforth refer to hold, sell and strong sell recommendations as “all sells” and buy or strong buy recommendations as “all buys.” Panel C shows consistency frequency for those recommendations above versus below the recommendation consensus. Consistency is more evident for recommendations below the consensus (CON% = 64.4%) than above it (52.8%). Panel D reports consistency frequency by year. There is little variation during our 15-year sample period, 1993-2007, ranging from a low of 55.0% in 1993 to a high of 61.2% in 2002.

-------------------------

Insert Table 1 here

-------------------------

4.2 Consistency and Earnings Forecast Accuracy

Table 2 contains information pertinent to testing our first hypothesis. Panel A provides the mean and standard deviation of CON and six well-known determinants of forecast accuracy: firm experience (Mikhail, Walther, and Willis 1997), number of firms followed (Clement 1999), forecast frequency (Jacob, Lys, and Neale 1999), brokerage house size (Clement 1999), forecast horizon (Mikhail et al. 1997), and past accuracy (Brown 2001), for two accuracy-groups: analysts who are more accurate than the forecast consensus (ACC) and analysts who not more accurate than the consensus (INACC). All of the well-known determinants of accuracy have their expected signs and five are significant at the 10% level (one-tailed test) or better. Accurate analysts are more experienced; follow fewer firms; work for larger brokerage houses; have shorter forecast horizons; and are more likely to have been accurate earnings forecasters the prior year.[10] Moreover, in line with our first hypothesis, analyst consistency is significantly, positively related to accuracy.

Panel B is a correlation matrix of ACC (equal to 1 if the analyst earnings forecast is more accurate than the consensus; 0 otherwise) and the seven variables in panel A. The correlations reveal that ACC has its expected relation with all variables, and five of the seven are significant (NFIRMS and FREQ are the exceptions). Several of the accuracy determinants are highly correlated so we include them all in the multivariate model that tests our first hypothesis.

Panel C reports results of a logistic regression of ACC on the seven variables in panel A. Except for FREQ, the other six variables are significant (10% level or better) with their expected signs. More importantly, in conformance with our first hypothesis, CON is positive and significant, revealing that consistency is an incremental determinant of accuracy in the presence of other determinants shown by prior literature. The last column of Panel C shows marginal effects of the seven independent variables.[11] The marginal effects range from a low of 0 for forecast frequency to the absolute value of 0.04 for consistency and forecast horizon. In sum, CON is a significant determinant of accuracy, and it is at least as important as any others we examine.

------------------------

Insert Table 2 here

------------------------

4.3 Returns to Consistent and Inconsistent Signals: Univariate Results

Table 3 shows cumulative abnormal returns (CAR) to stock recommendations for the 32-day holding period, (-1, 30). Panel A provides results for all recommendation types, both unconditional and conditional on signal consistency. Consistent with prior research (e.g. Womack 1996), “all buy” recommendations yield positive returns and “all sell” recommendations yield negative returns. Ignoring signal consistency, returns to a strategy of long positions in “all buy” and short positions in “all sell” recommendations yields a return of 5.5%. In conformity with our second hypothesis, recommendations by consistent analysts are more profitable than those by inconsistent analysts. The former generates an 8.0% return while the latter provides nearly an 80% lower return (i.e., 1.7%). The 6.3% difference for 32 trading days is economically (and statistically) significant.

Panel B shows returns to a strategy of taking long positions in recommendations above the consensus and short positions in recommendations below the consensus.[12] The results are even more dramatic than those in Panel A. Taking a long position in those recommendations above the consensus and a short position in those below the consensus provides 9.7% and -0.4% returns for consistent and inconsistent analysts, respectively.

------------------------

Insert Table 3 here

------------------------

4.4 Returns to Accurate and Inaccurate Forecasts: Univariate Results

Loh and Mian (2006) and Ertimur et al. (2007) have shown that if one had ex ante knowledge of ex post analyst earnings forecast accuracy, one is better off heeding the recommendations of those analysts whose earnings forecasts will be more accurate. To benchmark the impact of consistency on recommendation profitability, we compare it to accuracy’s impact on recommendation profitability. Table 4 presents the results.

Panel A provides evidence consistent with Loh and Mian (2006) and Ertimur et al. (2007) that accuracy is related to recommendation profitability. More particularly, taking a long (short) position in “all buy” (“all sell”) recommendations yields a 7.2% return for accurate earnings forecasters, which is nearly three times that for inaccurate forecasters (i.e., 2.6% return). The difference in recommendation profitability based on accuracy of 4.6% is about three-quarters of the difference based on consistency (i.e., 6.3%; see Panel A of Table 3). Panel B reveals similar evidence to Panel A. The difference in recommendation profitability partitioned on accuracy taking a long (short) position in recommendations above (below) the consensus is 4.2%, only about two-fifths of the difference for consistency (i.e., 10.1%; see Panel B of Table 3).

-------------------------

Insert Table 4 here

-------------------------

4.5 Returns to Stock Recommendation Conditional on Accuracy and Consistency

Table 5 presents some univariate results pertinent to our third hypothesis. Panel A shows results based on taking long positions for “all buy” recommendations and short positions for “all sell” recommendations; Panel B shows results based on taking long (short) positions for all recommendations above (below) the consensus. Both panels show results conforming to our third hypothesis. Conditional on an analyst being an accurate earnings forecaster, investors are far better off by heeding her advice if she is consistent (12.0% in Panel A; 14.3% in Panel B) than if she is inconsistent (-0.09% in Panel A; -5.2% in Panel B).

It is evident that the relation between recommendation profitability and earnings forecast accuracy is more nuanced than heretofore recognized. Conditional on analysts being accurate earnings forecasters, investors should do what they recommend if they are consistent. But if accurate earnings forecasters are inconsistent, investors may lose money, on average because returns after the joint recommendation-earnings forecast date conform to the “Ball and Brown effect,” i.e., in general stock prices move in the direction of future earnings surprises.

------------------------

Insert Table 5 here

------------------------

4.6 Multivariate Returns to Consistency and Accuracy

Table 6 presents multiple regression results for equations (2), (3) and (4). Panel A shows results based on taking long positions for “all buy” recommendations and short positions for “all sell” recommendations; Panel B shows results based on taking long (short) positions for all recommendations above (below) the consensus. All three models include factors that are similar to those used by Ertimur et al. (2007).[13] Because results are qualitatively similar in Panels A and B, we only discuss Panel A results. Model 1 indicates that consistent stock recommendations outperform inconsistent ones by 2.3% (t = 16.3) in the 32-day window. Consistent with Ertimur et al. (2007), the coefficient on ACC in model 2 is positive and significant (1.2%; t = 7.85), suggesting that, on average, recommendations by accurate earnings forecasters are more profitable. Model 3 shows that inconsistent, inaccurate analysts generate an average return of 1.1%; consistent, inaccurate analysts generate an average return of -2.0%; accurate, inconsistent analysts yield an average return of -2.3%; and consistent, accurate analysts generate an average return of 3.4%.[14] Consistent with our third hypothesis, recommendations of accurate forecasters are more profitable if they are consistent than if they are inconsistent (difference = 5.70%, F-statistic = 756.66).

-----------------------

Insert Table 6 here

-----------------------

4.7 Returns to Bold and Nonbold Forecasts: Univariate Results

Jagadeesh and Kim (2009) and Loh and Stulz (2009) show recommendations by bold analysts are more profitable than those by nonbold analysts. To obtain insight into the impact of consistency on recommendation profitability, we examine if the consistency effect is incremental to the boldness effect. Since consistency and boldness are ex ante measures, they can be used together to create an implementable trading strategy. Table 7 presents results. Panel A shows results based on taking long positions for “all buy” and short positions for “all sell” recommendations; Panel B shows results based on taking long (short) positions for all recommendations above (below) the consensus.

Both panels provide evidence consistent with Jagadeesh and Kim (2009) and Loh and Stulz (2009) that bold analysts make more profitable recommendations. More particularly, Panel A shows that taking a long position in “all buy” recommendations and a short position in “all sell” recommendations yields a 7.9% return for bold analysts, nearly two-thirds greater than for nonbold analysts (i.e., a 4.8% return). This difference of 3.1% is significant at the 1% level. Panel B provides similar evidence to that in Panel A. The difference in recommendation profitability partitioned on boldness, which takes long (short) positions in recommendations above (below) the consensus is 3.4%.[15] The differences in both panels based on analyst boldness (BOLD versus NONBOLD) are less than those based on analyst earnings forecast accuracy (see ACC versus INACC in Table 4), which in turn are less than those based on analyst consistency (see CON versus INCON in Table 3). Thus, CON is more important for recommendation profitability than ACC, and ACC is more important for recommendation profitability than BOLD.[16]

-----------------------

Insert Table 7 here

-----------------------

4.8 Returns to Stock Recommendations Conditional on Boldness and Consistency

Table 8 presents univariate results pertinent to our fourth hypothesis. Panel A shows results based on taking long positions for “all buy” recommendations and short positions for “all sell” recommendations; Panel B shows results based on taking long (short) positions for all recommendations above (below) the consensus. Both panels show evidence consistent with our fourth hypothesis. Conditional on the analyst being bold, investors make over four times as much by heeding her advice if she is consistent (10% and 10.7% in Panels A and B) than inconsistent (2.4% and 2% in Panels A and B).

------------------------

Insert Table 8 here

------------------------

4.9 Multivariate Returns to Consistency and Boldness

Table 9 presents multiple regression results for equations (2), (5), and (6) which underlie Models 1, 2, and 3. Panel A shows results based on taking long positions for “all buy” recommendations and short positions for “all sell” recommendations; Panel B shows results based on taking long (short) positions for all recommendations above (below) the consensus. The Panel A and B results are qualitatively similar so we only report the Panel A results. Model 1 shows that consistent recommendations outperform inconsistent ones 2.3% (t = 16.3) in a 32-day window. In model 2 the coefficient on BOLD is 1.2% (t =7.85), suggesting that bold recommendations are more profitable than nonbold ones. Model 3 shows recommendations by inconsistent, nonbold analysts yield average returns of -1.1%; consistent, nonbold analysts generate average returns of 0.9%; bold, inconsistent analysts provide average returns of -0.7%; and consistent, bold analysts provide average returns of 2.4%. In conformance with our fourth hypothesis, bold recommendations are more profitable when made by consistent rather than inconsistent analysts (difference = 3.10%, F-statistic = 129.19).

-----------------------

Insert Table 9 here

-----------------------

5. Additional Analyses

5.1 Conflicts of Interest, Value Relevance of Earnings, and Regulatory Reforms

Ertimur et al. (2007) show the relation between earnings forecast accuracy and recommendation profitability depends on conflicts of interest, value relevance of earnings, and regulatory reforms. More specifically, they find that: (1) forecast accuracy impacts recommendation profitability only if earnings are value relevant; (2) conflicts of interest differentially affect buy, hold and sell recommendations; and (3) regulatory reforms have mitigated conflict of interest problems. Barniv et al. (2009) and Chen and Chen (2009) show analysts are more likely to use their earnings forecasts in their recommendations after implementation of Regulation FD and the Global Settlement Act. To examine if our results are robust to conflicts of interest, value relevance of earnings, and regulatory reforms, we retest our hypotheses for three partitions of our data: (1) conflicts of interest, (2) value relevance of earnings, and (3) regulatory regime.

Similar to Ertimur et al. (2007), we measure conflict of interest using the Carter and Manaster (1990) brokerage house reputation ranking modified by Loughran and Ritter (2004).[17] Analysts are classified as TOPTIER if the maximum ranking of their investment banks in our sample period is 9.1. Analysts with lower ranks are classified as investment bank, IBANK, and analysts at brokerage firms without CM reputation rankings are classified as NONIBANK. Following Ertimur et al., we measure value relevance of earnings in three ways. Earnings are considered to be less value relevant if (1) the firm reports a loss (LOSS), (2) the firm’s book-to-market ratio is in the quintile of our sample (LOWBTM), or (3) R-square from a rolling regression of 15 month returns ending three months after the fiscal year end on the current year earnings level and earnings change falls in the first quintile of our sample (LOWRSQ).[18] We also divide our sample period into three time periods: Pre-FD, Post-FD, and Post-GS. Recommendations issued before October 23, 2000 are classified as PRE_FD,; those after May 10, 2002 are classified as POST_GS; and the rest as POST_FD. Panels A through D of Table 10 respectively provide results for hypotheses one through four. The first row of Table 10 shows the key number from the earlier relevant tables that tested the hypothesis based on our full sample. Rows two to four show similar results based on the investment banking types: TOPTIER, IBANK, and NONIBANK. Rows five and seven show similar results based on the value relevance groups: LOSS, LOWBTM, and LOWRSQ. Rows eight to ten provide similar results for the regulatory regimes: PRE-FD, POST-FD, and POST-GS.

Panel A contains findings regarding our first hypothesis. It shows the coefficient of CON is equal to 0.173 based on the full sample from the Table 2 Panel B logistic regression of ACC on CON and the six well-known determinants of ACC. The coefficients of CON based on the nine subsamples are listed below the main results. The CON coefficients are significantly positive across all nine sub-samples, indicating our findings regarding our first hypothesis are robust to conflicts of interest, value relevance of earnings, and regulatory regime.

For simplicity, in order to test our recommendation profitability hypotheses (H2, 3, 4), we report results for recommendations profitability from taking long (short) positions on above (below) the recommendation consensus (i.e., we confine our analysis to Panel B of Tables 3, 6 and 9). The first row of Panel B provides evidence concerning our second hypothesis. It shows the differential profitability of consistent and inconsistent recommendations is 3.5% for our full sample. The differential profitability is significant across all our nine sub-samples, ranging from 2.3% to 6.5%, revealing that our findings for our second hypothesis are robust to conflicts of interest, value relevance of earnings, and regulatory regime.

Panel C provides evidence concerning our third hypothesis. The first row shows the difference of 8.9% between recommendations issued by accurate but inconsistent (ACC) and those issued by accurate and consistent (CON_ACC) in Panel B of Table 6. The differential profitability between these two groups is significant across all sub-samples ranging from 7.6% to 12.7%. Thus, our findings regarding our third hypothesis are robust to conflicts of interest, value relevance of earnings, and regulatory regime.

Panel D provides evidence concerning our fourth hypothesis. The first row shows the difference of 3.3% between bold and inconsistent (BOLD) and bold and consistent recommendations (CON_BOLD) in Panel B of Table 9. The differential profitability between these two groups is significant across all but POST_FD subsamples and the differences range from 0.5% to 5.1%. Thus, our findings regarding our fourth hypothesis are robust to conflicts of interest, value relevance of earnings, and regulatory regime.

-------------------------

Insert Table 10 here

-------------------------

5. 2 Robustness Checks

We perform a number of robustness checks of our results. First, we investigate whether forecasts issued by consistent analysts are timelier. We argued above that consistent analysts possess a greater amount of and/or more precise private information. Thus, we hypothesized that consistent analysts are more accurate, and we obtained such evidence. Another proxy for information quality of a forecast is timeliness. Cooper et al. (2001) shows that timely analysts elicit greater market responses than accurate forecasts. Brown and Hugon (2009) find team analysts trade off forecast accuracy for timeliness, consistent with the notion that timeliness is an important dimension of earnings forecast quality. To test whether consistent forecasts are timelier, we run the following regression, which is similar to our first regression except that our dependent variable is LFR, not ACC.

LFR = α0 + α1 CON + α2 FEXP + α3NFIRMS + α4 FREQ+ α5BSIZE + α6HORIZON + α7PACC (7)

The coefficient on CON is 0.398 (t =3.98) suggesting that consistent analysts are more timely. This result confirms our conjecture that consistent analysts possess a relatively greater amount of and/or more precise private information.

Second, we control for the information content of earnings forecasts. The greater recommendation profitability associated with consistent recommendations may simply reflect the additive effect of the two information signals: forecasts and recommendations. To examine this, we rerun models (2) to (6) by adding REV_EARN, which is defined as the difference between the forecast and consensus scaled by the stock price two days before the forecast issuance date. Our results are consistent with our second to fourth hypotheses.

Third, we replace FEXP, FREQ, NFIRMS, and NIND in equations (1) – (6) with the prior year t-1 data so that all our variables except ACC are known ex ante. Our results are consistent with all four hypotheses. Finally, we rerun our analyses presented in Panel A table 6 and Panel A table 9 by omitting hold recommendation because it is unclear whether hold recommendation should be treated as sells. Our results remain consistent with our second to fourth hypotheses.

Fourth, we examined CON in conjunction with ACC and CON along with BOLD, but we did not examine all three together. In order to see whether CON (BOLD) has the largest (smallest) impact on recommendation profitability when all three are in the same equation, we run the following equation:

PFT = α0 + α1CON + α2 ACC + α3BOLD + α4FEXP + α5NFIRMS + α6FREQ + α7 BSIZE + α8LFR (8)

When PFT is calculated as the returns to long “all buy” and short “all other”, we find that the respective coefficients on CON, ACC, and BOLD are 0.022 (t-value = 15.82), 0.017 (t-value = 12.51), and 0.011 (t-value = 7.41), the same ordering we obtained with our pair-wise comparisons of CON with ACC in Table 6 Panel A and CON with BOLD in Table 9 Panel A. When PFT is calculated as returns to longing recommendations above the consensus and shorting recommendations below the consensus, we find that the respective coefficients on CON, ACC, and BOLD are 0.034 (t-value = 23..60), 0.017 (t-value = 12.78), and 0.013 (t-value = 8.28), the same ordering as seen with our pair-wise comparisons of CON with ACC in Table 6 Panel B and CON with BOLD in Table 9 Panel B.

Finally, we examined the impact of ACC on recommendation profitability, recognizing that it was an ex post measure that cannot be used as a trading rule. We now examine the impact of past accuracy (PACC) which can be used as a trading rule. When we rerun the Model 2 regression in Table 6 substituting PACC for ACC, we find PACC is insignificant (coefficient = 0.001; t-value = -0.74 for Panel A regression; coefficient = 0.002; t-value = -1.26 for Panel B regression).[19] We also rerun equation (8) substituting PACC for ACC. For the Panel A regression we obtain the following coefficient ordering: CON (coefficient = 0.028; t-value = 15.82), BOLD (coefficient = 0.012; t-value = 6.47), and PACC (coefficient = 0.002; t-value = 1.28). For the Panel B regression we get the following coefficient ordering: CON (coefficient = 0.043; t-value = 23.70), BOLD (coefficient = 0.013; t-value = 6.82), and PACC (coefficient = 0.004; t-value = 1.83). Thus, unlike private knowledge of future accuracy, public knowledge of past accuracy is not valuable for the purposes of increasing recommendation profitability.

6. Conclusions

Stock recommendations are generally issued along with earnings forecasts. We construct a simple measure of consistency of these two outputs of analysts’ reports. We define consistency as both an analyst’s recommendation and earnings forecast are above (below) the prevailing consensus. We show consistency impacts both earnings forecast accuracy and stock recommendation profitability. Consistent analysts issue more accurate earnings forecasts and their recommendations are more profitable conditional upon either earnings forecast accuracy or recommendation boldness.

We have shown that consistency, unconditional on either forecast accuracy or boldness, provides investors with an average abnormal 32-trading day return of 9.7% when taking long (short) positions for recommendations that are above (below) the consensus. These returns can serve as a benchmark for evaluating other trading strategies, such as those based on intrinsic value models. Consistency is much simpler to use than intrinsic value models. Consistency merely requires an analyst’s stock recommendation and earnings forecast along with the recommendation consensus and the earnings forecast consensus. Intrinsic value models require the analyst’s one- and two-year ahead earnings forecasts, long-term growth forecast, dividend payout and cost of capital forecasts, book value, stock price, and additional assumptions (e.g., clean surplus).

We have confined our analysis to one particular type of consistency, namely that between an analyst’s stock recommendation and her one-year-ahead earnings forecast. Future research should consider other types of consistency, such as between earnings and cash flow forecasts, and recommendations and a vector of earnings forecasts (e.g., one-year ahead; two-year ahead; long-term growth).

References

Ashbaugh-Skaife, H., D. W. Collins, and R. LaFond. 2006. The effects of corporate governance on firms’ credit ratings. Journal of Accounting Economics 42: 203–43.

Asquith, P., M. Mikhail, and A. Au. 2005. Information content of equity analyst reports. Journal of Financial Economics 75: 245-282.

Bagnoli, M., M. Clement, M. Crawley, and S. Watts. 2009. The profitability of analysts’ stock recommendations: What role does investor sentiment play? Working paper, Purdue University.

Ball, R. and P. Brown. 1968. An empirical evaluation of accounting income numbers. Journal of Accounting Research, 6: 159-178.

Barber, B., R. Lehavy, M. McNichols, and B. Trueman, 2006. Buys, holds, and sells: The distribution of investment banks’ stock ratings and the implications for the profitability of analysts’ recommendations. Journal of Accounting and Economics 41: 87-117.

Barniv, R., O. Hope, M. Myring, and W. Thomas. 2009. Do analysts practice what they preach and should investors listen? Effects of recent regulations. Accounting Review 84: 1015-1039.

Bradshaw, M. 2004. How do analysts use their earnings forecasts in generating stock recommendations? Accounting Review 79: 25-50.

Brav, A and R. Lehavy. 2003. An empirical analysis of analysts’ target prices: Short-term informativeness and long-term dynamics. Journal of Finance 58: 1933-1967.

Brown, L. 2001. How important is past analyst forecast accuracy? Financial Analysts Journal 57: 44-49.

Brown, L. and A. Hugon. 2009. Team earnings forecasting. Review of Accounting Studies 14: 587-607.

Carter, R., and S. Manaster. 1990. Initial public offerings and underwriter reputation. Journal of Finance 45: 1045-1067.

Chen, C., and P. Chen. 2009. NASD rule 2711 and changes in analysts’ independence in making stock recommendations. Accounting Review 84: 1041-1071.

Chen, Q., and W. Jiang. 2006. Analysts' weighting of private and public information. Review of Financial Studies 19: 319-355.

Clement, M. 1999. Analyst forecast accuracy: do ability, resources, and portfolio complexity matter? Journal of Accounting and Economics 27: 285−303.

Clement, M., and Tse, S. 2005. Financial analyst characteristics and herding behavior in forecasting. Journal of Finance 60: 307-341.

Cooper, R.A., T.E. Day and C.M. Lewis. 2001. Following the leader: A study of individual analysts’ earnings forecasts. Journal of Financial Economics 61: 383-416.

Dugar, A., and S. Nathan. 1995. The effect of investment banking relationships on financial analysts’ earnings forecasts and investment recommendations. Contemporary Accounting Research 12: 131-160.

Ertimur, Y., J. Sunder, and S. Sunder. 2007. Measure for measure: The relation between forecast accuracy and recommendation profitability of analysts. Journal of Accounting Research 45: 567-606.

Francis, J., and L., Soffer. 1997. The relative informativeness of analysts’ stock recommendations and earnings forecast revisions. Journal of Accounting Research 35: 193-211.

Frankel, R. and C.M.C. Lee. 1998. Accounting valuation, market expectation, and cross-sectional stock returns. Journal of Accounting and Economics 25: 283-319.

Hong, H., J. Kubik, and A., Solomon. 2000. Security analysts’ career concerns and herding of earnings forecasts. Rand Journal of Economics 31: 121–144.

Irvine, P., M. Lipson, and A. Puckett. 2007. Tipping. Review of Financial Studies 20: 741-768.

Jacob, J., T. Lys, and M. Neale. 1999. Expertise in forecasting performance of security analysts. Journal of Accounting and Economics 28: 51-82.

Jagadeesh, N and W. Kim. 2009. Analysts Herd? An analysis of recommendations and market reactions. Review of Financial Studies, Forthcoming.

Ke, B and Y.Yu. 2009. Why don’t analysts use their earnings forecasts in generating stock recommendations? Working paper, Pennsylvania State University.

Lin, H., and M. McNichols. 1998. Underwriting relationships, analysts’ earnings forecasts and investment recommendations. Journal of Accounting and Economics 25: 101-127.

Loh, R., and G. Mian. 2006. Do accurate earnings forecasts facilitate superior investment recommendations? Journal of Financial Economics 80: 455-483.

Loh, R.L. and R.M. Stulz. 2009. When are analyst recommendation changes influential? Working paper, Ohio State University.

Loughran, T., and J. Ritter. 2004. Why has IPO underpricing changed over time? Financial Management Autumn: 5-37.

Lys, T., and S. Sohn. 1990. The association between revisions of financial analysts' earnings forecasts and security-price changes. Journal of Accounting and Economics 13: 341-363.

Malmendier, U., and D. Shanthikumar. 2007. Do security analysts speak in two tongues? Working paper, University of California, Berkeley and Harvard University.

Mikhail, M., B., Walther, and R., Willis. 1997. Do security analysts improve their performance with experience? Journal of Accounting Research 35: 131-157.

Ohlson, J. 1995. Earnings, book values, and dividends in security valuation. Contemporary Accounting Research Spring: 661-687.

Womack, K. 1996. Do brokerage analysts’ recommendations have investment value? Journal of Finance 51: 137-167.

|Table 1 |

|Consistency between Analysts' Stock Recommendations and Their Earnings Forecasts |

| | | | | | | |

|Panel A: Consistency for full sample |

| | |Total No. of Observations | |No. of Consistent Observations| |CON % |

|All Recommendations |  |67,447 |  |39,744 |  |58.90% |

| | | | | | | |

|Panel B: Consistency by each of the five recommendation levels |

|Recommendations | |Total No. of Observations | |No. of Consistent Observations| |CON % |

|Strong buy | |17,516 | |10,013 | |57.2% |

|Buy | |18,331 | |9,311 | |50.8% |

|Hold | |26,544 | |16,719 | |63.0% |

|Sell | |3,006 | |2,175 | |72.4% |

|Strong sell | |2,050 | |1,523 | |74.3% |

| | | | | | | |

|Panel C: Consistency by recommendations relative to recommendation consensus |

|Recommendations | |Total No. of Observations | |No. of Consistent Observations| |CON% |

|Above Consensus | |31,888 | |16,845 | |52.8% |

|Below Consensus |  |35,559 |  |22,896 |  |64.4% |

|Panel D: Consistency by year |

|Year | |Total No. of Observations | |No. of Consistent Observations | |CON% |

|1993 | |1,014 | |558 | |55.0% |

|1994 | |3,000 | |1,795 | |59.8% |

|1995 | |3,418 | |1,979 | |57.9% |

|1996 | |3,495 | |2,060 | |58.9% |

|1997 | |3,867 | |2,257 | |58.4% |

|1998 | |4,278 | |2,502 | |58.5% |

|1999 | |4,358 | |2,568 | |58.9% |

|2000 | |3,839 | |2,269 | |59.1% |

|2001 | |3,778 | |2,273 | |60.2% |

|2002 | |5,025 | |3,077 | |61.2% |

|2003 | |6,484 | |3,850 | |59.4% |

|2004 | |6,688 | |3,930 | |58.8% |

|2005 | |6,117 | |3,638 | |59.5% |

|2006 | |6,208 | |3,603 | |58.0% |

|2007 |  |5,878 |  |3,383 |  |57.6% |

Notes: This table shows the number of recommendation-forecast pairs in our sample, partitioned by recommendation level, above versus below the consensus, and years. Data are from I/B/E/S for the 15 years 1993-2007. CON% indicates the percent of recommendations that are consistent with the same analyst’s earnings forecast defined by CON, a binary variable, which defines an analyst as consistent if both her recommendation and her earnings forecast are above or below their respective prevailing consensus. The forecast (recommendation) consensus is the mean of all forecasts (recommendations) issued for the firm during the 90-day period prior to but excluding the joint recommendation-forecast issuance date.

|Table 2 |

|Analyst Determinants of Earnings Forecast Accuracy |

| |

| |ACC | |INACC | |DIFF |

| |n = 26,423 | |n = 14,345 | | | |

|  |

| |

|Panel C: Logistic Regression of Forecast Accuracy on Seven Determinants of Accuracy |

|Prob (ACC=1) = α0 + α1CON + α2FEXP + α3NFIRMS + α4FREQ + α5BSIZE + α6HORIZON + α7PACC (1) |

| |Expected Sign |Parameter |Chi-Square |Significance |Marginal Effect |

|Intercept |? |0.496 |97.64 | ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download