The use of target prices to justify sell-side analysts' stock recommendations.

SYNOPSIS: This study examines a sample of 103 sell-side analysts' reports to document the frequency with which analysts disclose target prices as justifications for their stock recommendations. In addition, I investigate whether the degree of assessed overpricing or underpricing implied by target

prices is related to the favorableness of stock recommendations. I find that analysts use target price justifications in over two-thirds of the sample reports, and higher target prices are associated with more favorable stock recommendations. The most favorable recommendations (and target prices) are more likely to be justified by price-earnings ratios and expected growth while the least favorable recommendations are more likely to be justified with other qualitative statements. Further evidence suggests that analysts actually compute target prices using price-multiple heuristics such as "PEG." However, in reports that do not disclose target prices, estimates of target prices based on these heuristics are unable to j ustify the stock recommendations. Several explanations are proposed, including self-selection biases implying analysts do not disclose target prices when the disclosure would not support the recommendation or when analysts are less certain about underlying earnings forecasts.

Keywords: target prices; valuation; stock recommendations; earnings forecasts; sellside financial analysts; long-term growth; price-to-earnings ratios; PEG.

INTRODUCTION

Sell-side analysts presumably derive stock recommendations based on their own valuations of stocks. This study investigates the frequency with which analysts disclose these valuations in support of their stock recommendations and examines the properties of those valuations. A casual reading of analysts' reports reveals that they often mention valuations as "target" or "objective" prices, and the associated recommendations are supported by the difference between the target prices and the current trading prices. For example, CIBC Oppenheimer issued a report on Harcourt General when the price was $49.75 and concluded, "We continue to recommend Harcourt General with a price target of $64 per share. We believe that the stock is significantly undervalued at current levels." (1) The target price reflects the analyst's valuation of the stock, and it justifies the Buy recommendation in the report as one might expect.

Accepting the standard wording of recommendations, investors would expect that a Buy or Strong Buy recommendation indicates a stock that the analyst believes is currently underpriced, a Hold recommendation indicating a fairly priced stock, and a Sell recommendation indicating an overpriced stock. However, Sell recommendations are virtually nonexistent (Stickel 1998). Consequently, savvy investors might interpret Holds as essentially Sells, and Buys as Holds. However, would an analyst disclose a target price when he or she believes a company is overvalued? This may not be likely if previously documented optimistic bias in forecasts and recommendations also describes target prices (e.g., McNichols and O'Brien 1997). Explanations for this optimistic bias include a desire by analysts to "curry favor with management" (Francis and Philbrick 1993) and enhance investment banking relationships (Lin and McNichols 1998). To the extent that these incentives affect how analysts justify their recommendations, it is possibl e that there is an asymmetric use of target prices in support of stock recommendations.

Most prior research related to analysts' ability to identify mispriced stocks focuses on the stock recommendation, rather the analysts' valuations per se. For example, Womack (1996) documents significant market reactions to the release of analysts' recommendations, suggesting investors believe recommendations are valuation relevant. However, Bradshaw (2001) uses analysts' earnings forecasts as inputs into a residual income model, and finds that resulting valuations are unable to explain the associated stock recommendations, despite evidence that these valuations identify mispriced stocks (Frankel and Lee 1998). These findings suggest that analysts' private valuations differ from those of a residual income model, but there is little empirical work examining the valuations that analysts do provide. This paper investigates a random sample of analysts' reports to determine (1) how frequently analysts justify their recommendations with target prices, and (2) when they do not use target prices, what they use instea d.

Stock valuations are understood to reflect underlying financial information. Evidence in Bandyopadhyay et al. (1995) demonstrates a relation between earnings expectations and analysts' valuations for 128 Canadian companies. They find that near-term forecasted earnings revisions explain approximately 30 percent of the variation in target price revisions, while long-term earnings forecast revisions explain approximately 60 percent of the variation in target price revisions. Although valuations of forecasted financial information should theoretically be the basis for recommendations to buy or sell stocks, recent studies suggest that analysts might supplement (or possibly even replace) these valuations with other nonfinancial factors. For example, Amir and Lev (1996) document value-relevance of nonfinancial information for valuations in the telecommunications industry. Similarly, through questionnaires Barker (1999) finds that intangibles such as "quality of management" are often cited as a determinant of analyst s' assessments for a sample of U.K companies. In addition to the standard financial information, a skillful financial analyst incorporates such nonfinancial information into estimates of value. Thus, a secondary purpose of this paper is to examine the frequency with which analysts supplement their recommendations or target prices with other information.

I randomly select 103 companies across all industries and acquire the most recent analyst report that includes a recommendation. (2) Each report is read, and reasons for the recommendation are identified. Analysts disclose target prices in roughly two-thirds of the reports, and the tendency to disclose a target price is greater for more favorable recommendations. Moreover, the distribution of the ratio of the target price to actual trading price at the date of the report is positively related to the favorableness of the recommendation, consistent with analysts' recommendations reflecting the disclosed valuations. In addition to target prices, I find that analysts also justify recommendations using a number of other factors, with the average report containing two such justifications. Analysts appear to invoke nonfinancial factors more frequently when the underlying stock recommendation is less favorable (i.e., Hold).

The most prevalent bases for recommendations other than target prices are price-to-earnings (PIE) ratios and forecasted long-term earnings growth rates. Moreover, analysts frequently base their target prices on a combination of these two constructs. Together the P/E ratio and expected growth form a ratio frequently cited in the investment community and referenced in many of the surveyed reports, the PEG ratio, which is equal to the P/E ratio divided by the growth rate. Advocates of this ratio claim that a fairly valued stock should have a PEG ratio of 1. Implicitly, PEG advocates are using a valuation heuristic that calculates a target price as the earnings per share forecast times the growth rate. (3) For the sample, I find that PEG valuations are significantly related to analysts' disclosed target prices and stock recommendations. However, using PEG-based valuations as a proxy for analysts' undisclosed target prices, I find no analogous support for the recommendations in these reports. Supplemental tests s uggest that analysts' are less likely to disclose target prices when they are more uncertain about the forecasted earnings for the company.

The evidence in this paper is consistent with that in several prior studies. For example, Previts et al. (1994) analyze the content of 479 analysts' reports using the "Word Cruncher" computer program to document information needs of analysts. They find extensive use of various accounting data in their reports, particularly historical and forecasted earnings. Similarly, Govindarajan (1980) performs a content analysis to document whether analysts focus more on earnings or cash flow, and finds an overwhelming focus on earnings. Another related study is Block (1999), who surveys financial analysts and asks which analytical techniques they use. Analysts responded that they rarely use present value techniques in equity valuation. (Almost half of the respondents stated that they "never" used present value techniques.) The evidence presented here corroborates that finding with the low number of references to "formal" valuations, but frequent uses of PIE-based and PEG-based valuations.

An understanding of how analysts value stocks is of interest to a broad audience, particularly given the recent increase in the availability of analyst data to individual investors. For example, some of the most popular web sites are personal investment ones, such as TheStreet.com, Yahoo!Finance, and The Motley Fool, all of which include analysts' recommendations, target prices, and forecasts. An understanding of how analysts value and recommend stocks is of interest to the large number of current investors in the stock market. Moreover, accountants in particular will benefit from knowing how a group of sophisticated financial intermediaries utilizes accounting information.

The remainder of this paper is structured as follows. The next section discusses the sample selection criteria and coding procedures. Following that, I document the frequency of target price disclosure in the sample reports and the levels of those target prices. The next section examines differences between analysts' justifications for reports that disclose target prices vs. those that do not. Finally, I use the findings for the target price disclosure subsample to estimate pseudo-target prices for the target price nondisclosure subsample to shed light on why analysts omit target prices from these reports. The paper closes with a summary and suggestions for future research.

SAMPLE

The sample population includes all companies that meet the following criteria: (1) a recommendation and annual forecasts for one- and two-year ahead earnings in the First Call New Real-Time Earnings Estimates History Database, (2) accounting information available in the Compustat research files, and (3) share and price data available. in the CRSP daily returns file. These criteria create a diverse population from which I draw companies and obtain analyst reports. Based on the distribution of that population across two-digit SIC codes, I randomly select companies (without replacement) such that a sample of 100 companies is obtained. The random selection is executed within two-digit SIC codes such that the distribution of the random sample across two-digit SIC codes mirrors that of the merged First Call/Compustat/CRSP population (e.g., if 6 percent of the larger population firms are in SIC code 2600, then the random sample includes six firms from SIC code 2600). The final sample consists of 103 analysts' repor ts due to rounding of the percentages across SIC codes.

The source of the analysts' reports is the Investext database. When searching for an analyst's report on a particular company, I obtain the most recent report available for that sample company. (4) Most reports (67) are from the first calendar quarter of 1999. With two exceptions, the remainder of the reports (34) was released during 1998. The earliest report obtained was released in the fourth quarter of 1996. Thus, reports used in this study are characteristic of recent years only, and results should be interpreted accordingly.

If no reports are available for a company, I randomly select another company within the same two-digit SIC code. (5) In certain cases, an analyst report is obtained but excluded from the sample because it includes no discussion of the recommendation. There are 19 reports read but omitted from the sample because there is no discussion of a stock recommendation. In most cases, these reports are earnings forecast revisions only or discussions of recent earnings results and do not discuss a stock recommendation. This obviously introduces some bias into the final sample, in that the final sample includes only reports for which the analyst states a rationale for a stock recommendation.

The final sample covers 45 different two-digit SIC codes, with only six industries individually representing more than 5 percent of the sample. The sample also covers a large cross-section of brokerages. Forty-three different brokerages are represented, with five each accounting for more than 5 percent of the sample. (6)

TARGET PRICE DISCLOSURES

The distribution of the recommendations for the 103 reports is similar to that in prior research, which documents optimistic bias (e.g., Stickel 1998). There are no Sell recommendations in the sample. In contrast, the majority of recommendations are Buys (n = 42) or Strong Buys (n = 31), and the remainder are Hold recommendations (n = 30). Table 1 documents the frequency with which analysts provide target prices in support of these recommendations. Target prices are disclosed in a majority of reports (n = 67). However, the tendency to disclose a target price appears to be related to the favorableness of the recommendation, such that more favorable recommendations are more likely to be supported by a target price. For example, of the 30 Hold recommendations, only 8(27 percent of Holds) are accompanied by a target price. In contrast, 84 percent of the Strong Buys and 79 percent of Buys are supported by a target price. A [chi square] test confirms that the frequency of target price disclosures is not independen t of the stock recommendation (0.001 significance level).

The different disclosure levels of target prices across stock recommendations suggest that analysts are more inclined to provide them when their recommendations are more favorable (i.e., Buy or Strong Buy) than they are when their recommendations are less favorable (i.e., Hold). In Table 2, I provide an analysis of the ratio of the target price to actual trading price (TP/P) for all reports with disclosed target prices. This ratio should be a measure of perceived overvaluation or undervaluation by the analyst, and it is expected that more favorable recommendations are supported by higher values for TP/P.

Like the underlying stock recommendations, the target prices are quite optimistic with an overall mean (median) TP/P ratio of 1.36(1.29). Thus, analysts' mean (median) target prices for these companies are 36 percent (29 percent) above current trading prices. The minimum ratio of TP/P is 0.89 (a Hold), and the maximum is 2.46 (a Strong Buy). (7) More interesting, there is a monotonic increase in TP/P across the recommendation categories. TP/P is larger for Strong Buy than for Buy recommendations, and TP/P is larger for Buy than for Hold recommendations. Differences in TP/P between Hold and Buy recommendations are significant for both the mean and median. However, only the difference in the median TP/P between Buy and Strong Buy recommendations is statistically significant at conventional levels; the means are insignificantly different (p-value 0.1598). Overall, the results in Table 2 are consistent with target prices, when disclosed, being a primary justification for stock recommendations.

OTHER JUSTIFICATIONS FOR RECOMMENDATIONS

The evidence in Table 1 for target price disclosure is consistent with McNichols and O'Brien (1997), who hypothesize that analysts report selectively, leading to a truncation of "bad news" earnings forecasts and stock recommendations. The fact that target prices are less often disclosed for the least favorable (Hold) recommendations extends this notion to target prices as well. But how do analysts justify recommendations in the reports that do not disclose target prices? (8)

Panel A of Table 3 presents the distribution of the number of different (non-target price) justifications cited by analysts in support of their stock recommendations. On average, analysts provide approximately two explicit reasons for a recommendation. The minimum number of justifications is one, while the maximum number is six.

Panel B of Table 3 displays the frequency of particular justifications given in the research reports. For the entire sample, the most frequently cited justification for a recommendation is the price-to-earnings (PIE) ratio (over three-quarters of the reports). For example, an analyst often calculates a forward P/E ratio, and argues that the forward P/E ratio is sufficiently low to justify current purchase of the stock. (9) Analysts' projected three- to five-year long-term growth (LTG) rate in annual earnings per share is the second-most frequently cited justification (37 percent). Analysts tend to refer to a high (low) growth rate as support for an optimistic (less optimistic) recommendation. Following growth, the next most frequently cited justification is an "anticipated" earnings surprise relative to the consensus forecast (17 percent). (10)

The remaining explanations include various other price-multiples (i.e., price-to-earnings before income taxes, depreciation, and amortization--10 percent, price-to-cash flow--5 percent, price-to-book--4 percent, and price-to-sales--1 percent). Also, and not surprisingly, analysts cite miscellaneous industry-specific operating statistics such as subscriber growth or churn, dealer base, or number of exploratory wells (8 percent). Similarly, there were numerous unclassifiable reasons given such as recent accounting irregularities, court decisions, new contracts, or general macroeconomic conditions (18 percent). Nonetheless, the overriding message from Table 3 is that, across companies, P/E ratios and growth are systematically invoked by analysts when summarizing the investment potential of stocks.

In Panel B of Table 3, the second and third columns partition the sample based on whether a target price is disclosed, and the final two columns present a [chi square] statistic and p-value for whether the distribution of each justification differs between the two subsamples. Generally, there are two significant differences between the reports that contain a target price and those that do not. First, for the target price disclosure subsample, the P/E ratio is mentioned in 87 percent of the reports vs. only 56 percent of the target price nondisclosure subsample. Second, reports that do not contain a target price more frequently invoke miscellaneous industry-specific operating statistics as justification for the recommendations (17 percent for the target price nondisclosure subsample vs. only 3 percent for the target price disclosure subsample).

Panel C shows the frequency of various combinations of justifications that appear at least twice. As in Panel B, the most frequent justification for recommendations is the PIE ratio only (26 percent). The next most frequently cited combinations are (1) the forward PIE ratio coupled with the growth rate (21 percent) and (2) the forward P/E ratio, growth rate, and other information (8 percent). Overall, a majority of the combinations include both PIE ratios and the growth rate. In total, 36 of the 103 reports (35 percent) include both the PIE ratio and the growth rate as justification for the stock recommendations.

PEG

The combination of the P/E ratio and the LTG projection form a ratio referred to as the "PIE-to-Growth" or "PEG" ratio. Numerous financial sites exist that use PEG as a stock screen (e.g., http://www.stockselector.com, http://www.fool.com/pegulator/, etc.). Additionally, readings in the financial press indicate disparate attitudes toward PEG. Numerous articles support the use of PEG in stock selection; see, for example, Dorfman (1991) and Slater (1993). Even Lynch (1989, 198) advocates the use of PEG: "The P/E ratio of any company that's fairly priced will equal its growth rate.... We use this measure all the time in analyzing stocks for the mutual funds." However, as with any simple tool, PEG has its critics. For example, Serwer (1998) describes PEG as "that ingenious justification for ludicrous valuations." Formally, PEG is defined as:

PEG = P/E/LTG*100

where P/E is the forward P/E ratio and LTG is the projected three-to five-year annual growth rate in earnings per share. (11) The PEG ratio is purportedly used to assess whether a stock's forward PIE ratio is "out of line." (12) The rule-of-thumb is that PEG ratios around 1 would translate into Holds, below 0.5--Strong Buys, and above 1.5--Strong Sells. This sentiment was reflected in a recent article, which states, "Many Wall Street observers contend that a stock is potentially cheap if its earnings-growth rate exceeds its price/earnings ratio" (Jones 1999). The following is an example from the sample of an analyst discussing a target price based on P/E relative to growth:

In due course, we think AGT stock will reattain at least a 17-20x multiple on forward EPS, i.e., a P/E-to-EPS growth rate ratio of 1.0x-1.1x. Applying this postulated valuation to our 1999-2000 EPS estimates implies near-term upside potential to $20-23 and a 12-18-month target of $27-32. (13)

Another sample report states:

We believe the company deserves a PEG ratio near the top of the industry given its market dominance, broad product line and customer base, growth initiatives, and solid financial position. We value Dialogic's shares based on one times its long-term growth rate of 15%. We are initiating coverage of DLGC with a BUY rating for aggressive growth investors and a $32 price target based on 15 times our 2000 BPS estimate of $2.14. (14)

Assuming a "normal" PEG of 1, Equation (1) can be rearranged to solve for price, i.e., P = E*LTG*100, which is essentially how analysts derive target prices in the above examples.

PSEUDO-TARGET PRICES

The results presented so far suggest that recommendations are frequently determined by target prices relative to current prices, and that P/E ratios and growth projections are often the basis of target prices. In this section, I adopt a two-part analysis to investigate (1) whether target prices are systematically related to forward P/E ratios and growth projections for the target price disclosure subsample, and (2) whether it is possible to construct pseudo-target prices for the target price nondisclosure subsample that explain analysts' stock recommendations.

"Pseudo-target prices" are calculated using forward P/E ratios and growth projections as inputs. Based on the results from the review of analysts' reports, I consider four pseudo-target prices, based on (1) industry P/E ratios and (2) the PEG heuristic, combined with either the one-year or two-year ahead earnings forecasts. These pseudo-target prices are labeled [TP.sub.PE1], [TP.sub.PE2], [TP.sub.PEG1], and [TP.sub.PEG2], respectively. For example, [TP.sub.PE1] is computed as the median industry forward P/E ratio (from I/B/E/S) times the one-year ahead earnings forecast, and [TP.sub.PEG1] is computed as the one-year ahead earnings forecast times the growth projection. (15) [TP.sub.PE2] and [TP.sub.PEG2] are calculated analogously, but use the two-year ahead earnings forecast rather than the one-year ahead forecast. Five sample reports have forecasted losses at both forecast horizons and pseudo-target prices cannot be calculated for these reports, leaving a sample size of 98 for the remainder of the paper. (1 6) For cross-sectional comparison, all pseudo-target prices are scaled by actual price, forming a ratio comparable to TP/P discussed above.

For the target price disclosure subsample, Table 4 provides a Pearson correlation matrix for the stock recommendation, actual target price, and the four pseudo-target prices. (17) Confirming the results in Table 2, the correlation between stock recommendations (coded on a 1 to 5 scale, from "Strong Sell" to "Strong Buy," with higher values corresponding to more favorable recommendations) and TP/P is positive and significant (0.33). Three additional observations are noteworthy. First, the industry P/E-based target prices are uncorrelated with stock recommendations and only moderately correlated with the disclosed target prices. This is likely due to the simplifying assumption of using median forward P/E ratios as the earnings multiplier, which is a crude attempt to capture company-specific valuations. Second, and in contrast the P/E-based target prices, those based on the PEG heuristic are correlated with stock recommendations, with correlations of 0.39 ([TP.sub.PEG1]) and 0.38 ([TP.sub.PEG2]). Finally, PEG-ba sed target prices are also highly correlated with actual target prices, with [TP.sub.PEG2] showing a slightly higher correlation (0.56 for [TP.sub.PEG2] vs. 0.50 for [TP.sub.PEG1]).

Panel A of Table 5 benchmarks the pseudo-target prices against the actual disclosed target prices for the reports that included target prices. (18) Consistent with the correlations in Table 4, the industry-based P/E calculations do not appear to do a good job of mimicking the monotonic increase in TP/P across the stock recommendations. In contrast, both [TP.sub.PEG1]/P and [TP.sub.PEG2]/P track the increase in TP/P, with [TP.sub.PEG2]/P appearing to perform better. However, the PEG-based target prices appear to slightly underestimate the disclosed target prices for Hold and Buy recommendations and slightly overestimate the disclosed target price for Strong Buy recommendations, e.g., for Hold (Buy) [Strong Buy] recommendations [TP.sub.PEG2]/P is 0.88 (1.16) [1.57] relative to TP/P of 1.10 (1.36) [1.44]. For Hold and Strong Buy recommendations, the difference between TP and [TP.sub.PEG2] is not statistically significant (p-values = 0.2806 and 0.3254, respectively, not tabulated), although it is for Buy recommen dations (p-value = 0.0083, not tabulated). Overall, the results suggest that PEG-based target prices are a reasonable proxy for the target prices that are not disclosed by the analyst.

Despite the conclusion from Panel A that industry median P/E-based target prices do not appear to reflect the favorableness of the recommendation, they are included in Panel B of Table 5 for completeness. Focusing on the PEG-based pseudo-target prices in the last two columns, the evidence stands in contrast to that in Panel A for the target price disclosure subsample. Rather than reflecting a monotonic increase with the favorableness of the stock recommendation, the relations between PEG-based target prices and recommendations appear to be inverted U-shaped. The highest value for both ratios occurs for Buy recommendations, and the lowest for Strong Buy recommendations. Unlike the target price disclosure sample results, pseudo-target prices are unable to explain the stock recommendations of the target price nondisclosure sample (e.g., there are no significant t-statistics for tests of differences in pseudo-target prices between adjacent recommendations).

I offer three nonmutually exclusive explanations for why analysts might not disclose target prices for this subsample. For the Hold recommendations, it is likely that analysts do not disclose target prices because those target prices would essentially reflect "bad news," as evidenced by the low pseudo-target prices in Table 5. The predominance of Hold recommendations without target price disclosures is consistent with analysts behaving differently when their views are less favorable. Second, for the Strong Buy recommendations, the nondisclosure of target prices may reflect the fact that analysts' target prices (reflected by the low PEG-based pseudo-target prices in Table 5) would not have justified a Strong Buy. However, note that both of these explanations depend on the reliability of the pseudo-target price estimates as a proxy for the analysts' actual but unobserved valuations.

A third explanation for the nondisclosure of target prices is that the analysts may choose not to disclose target prices when they are less certain of the inputs into their valuations.(19) This reasoning is related to Hayes (1998), who models how analysts' incentives affect their production of information. When companies are expected to perform better, analysts are prompted to expend more effort to gather information, which may increase the precision of their forecasts.

For earnings forecasts ([FEPS.sub.1] and [FEPS.sub.2]) and long-term growth projections, Table 6 tabulates the mean standard deviation for all forecasts outstanding in the month prior to the date of the sample analyst reports.(20) In all cases, the consensus of the analysts is lower for the subsample of companies without disclosed target prices. However, only three of the differences in mean consensus measures are statistically different between the two subsamples (standard deviation of [FEPS.sub.1] and [FEPS.sub.2], and coefficient of variation for [FEPS.sub.2]).(21) Overall, the results in Table 6, combined with those previously discussed, are consistent with the nondisclosure of target prices being explained by a greater level of uncertainty about underlying earnings expectations. The uncertainty may be due to either a deliberate lack of effort by analysts (Hayes 1998) or inherently uncertain earnings expectations.

SUMMARY

This paper provides evidence on the rationale offered by analysts when releasing stock recommendations. Based on a sample of 103 analysts' reports across a large number of industries, I provide evidence that analysts frequently justify recommendations with target prices. Analysts issue more favorable recommendations for stocks with higher target prices relative to current prices. Further evidence suggests that these target prices are a function of earnings forecasts and projected long-term earnings growth rates. I also calculate pseudo-target prices for the target price nondisclosure subsample, but find that these estimates of analysts' undisclosed target prices either would have been "bad news" or would not have justified the recommendations.

The results imply that analysts rely heavily on their forecasts of accounting earnings in establishing target prices that support stock recommendations. Accountants should find comfort in the fact that the primary output from the accounting process seems to be very important in valuations of corporate stocks. However, it appears that analysts use their earnings forecasts in a relatively unsophisticated manner, relying on simple heuristics to derive valuations. The evidence presented here is primarily descriptive, thus further work examining the nature of analysts' valuations is warranted.

Several caveats apply to the interpretation of the results. First, the sample consists of a few companies from each of 45 different two-digit SIC codes. Other interesting results may be obtained by a larger sample study that can compare analysts' reports both across and within industries. Second, the sample includes a large number of brokerage houses. It would be interesting to examine the extent to which analysts' reports systematically differ across brokerage houses (e.g., underwriting relationships, regional vs. national). Third, the sample examined here covers a period during which the overall market has risen to unprecedented levels. Fourth, the inferences regarding why analysts do not disclose target prices are subject to a selection bias for the disclosed target prices, upon which the calculation of pseudo-target prices is based. Finally, it is worthwhile to note that use of price multiples is not as straightforward as it may appear. Analysts use substantial judgment to determine multiples that accoun t for diverse factors such as expected macroenvironmental trends, industry outlooks, competition, accounting differences, and other similar items not examined here. Accordingly, inferences based on small samples may be subject to a larger degree of measurement error.

TABLE 1

Disclosure of Target Prices Conditional on Stock Recommendation

                         Target Price Disclosed?
Stock Recommendation        No           Yes       Total

Hold                        22             8        30
Buy                          9            33        42
Strong Buy                   5            26        31
Total                       36            67       103

[chi square] = 27.65, p-value = 0.001

This table is a contingency table for whether an analyst report contains
a target price, partitioned by the stock recommendation. The [chi
square] statistic is a test of whether the disclosure of a target price
in a report is independent of the stock recommendation.
TABLE 2

Relative Optimism in Target Prices across Stock Recommendations

Stock                                        TP/P
Recommendation              n           Mean    [sigma]   Minimum

Hold                         8          1.10     0.15      0.89
Buy                         33          1.36     0.26      1.07
Strong Buy                  26          1.44     0.32      1.10
All                         67          1.36     0.29      0.89

[H.sub.A]: Buy>Hold?
 test statistic                         2.8
 (p-value)                            (0.0040)

[H.sub.A]: Strong Buy>Buy?
 test statistic                         1.0
 (p-value)                            (0.1598)

Stock                               TP/P
Recommendation               Median    Maximum

Hold                          1.08      1.43
Buy                           1.25      2.07
Strong Buy                    1.34      2.46
All                           1.29      2.46

[H.sub.A]: Buy>Hold?
 test statistic                2.3
 (p-value)                  (0.0119)

[H.sub.A]: Strong Buy>Buy?
 test statistic                1.7
 (p-value)                  (0.0470)

This table presents the distribution of target prices (TP) scaled by
actual trading price (P) for the subsample of reports that disclose a
target price (n=67). Share trading price is obtained from CRSP, and
reflects the closing share price on the trading day just preceding the
date of the stock recommendation. Tests of differences are shown in the
bottom section of the table and are one-sided tests for the alternative
hypothesis ([H.sub.A]) predicting that more favorable recommendations
are justified by a higher ratio, TP/P. Test statistics and p-values
under the "Mean" column reflect t-tests for differences in mean TP/P
between adjacent recommendation categories; test statistics and p-values
under the "Median" column reflect Z-tests for differences in median TP/P
between adjacent recommendation categories.
TABLE 3

Nontarget Price Justifications for Recommendations in Reports with and
without Target Price Disclosure

Panel A: Number of Justifications per Report

Mean  Minimum  10%  Median  90%  Maximum

1.95     1      1     2      3      6
Panel B

Percent of Reports Citing Various Recommendation Justifications

                                          Target Price Dislcosed?
                                          All Reports     No

Justification                             (n = 103)    (n = 36)

Price/Earnings                               76%          56%
Growth                                       37           31
Anticipated earnings surprise                17           22
Price/EBITDA                                 10           11
Miscellaneous industry-specific
  operating statistics                        8           17
Price/Cash flow                               5            3
Price/Book                                    4            3
Debt levels                                   2            3
Price/Sales                                   1            3
Other                                        18           17

                                          Target Price Dislcosed?
                                            Yes
                                                     statistic
Justification                             (n = 67)  [chi square]

Price/Earnings                               87%       12.3
Growth                                       40         1.0
Anticipated earnings surprise                15         0.9
Price/EBITDA                                  9         0.1
Miscellaneous industry-specific
  operating statistics                        3         6.1
Price/Cash flow                               6         0.5
Price/Book                                    4         0.2
Debt levels                                   1         0.2
Price/Sales                                   0         1.9
Other                                        19         0.1




Justification                             p-value

Price/Earnings                            0.0001
Growth                                    0.3290
Anticipated earnings surprise             0.3520
Price/EBITDA                              0.7250
Miscellaneous industry-specific
  operating statistics                    0.0130
Price/Cash flow                           0.4720
Price/Book                                0.6700
Debt levels                               0.6520
Price/Sales                               0.1700
Other                                     0.7330

Panel C: Combinations of Specific Justifications

                                          Target Price Dislcosed?
                                          All Reports     No

Justification                             (n = 103)    (n = 36)

Price/Earnings only                          26%          14%
Anticipated earnings surprise only            2            6
Miscellaneous industry-specific
  operating statistics only                   6           14
Price/Earnings and growth                    21           14
Price/Earnings, growth, and
  anticipated earnings surprise               2            0
Price/Earnings, growth, and other             8            6
Price/Earnings and anticipated
  earnings surprise                           6            8
Price/Earnings and price/EBITDA               2            6
Price/Earnings and other                      5            3
Growth and anticipated earnings surprise      2            6
Anticipated earnings surprise and other       3            0
Price/EBITDA and miscellaneous
  industry-specific operating statistics      4            3
Other combinations                           13           20

                                            100%         100%

                                          Target Price Dislcosed?
                                            Yes
                                                     statistic
Justification                             (n = 67)  [chi square]

Price/Earnings only                          33%        4.3
Anticipated earnings surprise only            0         3.8
Miscellaneous industry-specific
  operating statistics only                   2         6.6
Price/Earnings and growth                    25         1.8
Price/Earnings, growth, and
  anticipated earnings surprise               3         1.1
Price/Earnings, growth, and other             9         0.4
Price/Earnings and anticipated
  earnings surprise                           5         0.6
Price/Earnings and price/EBITDA               0         3.8
Price/Earnings and other                      6         0.5
Growth and anticipated earnings surprise      0         3.8
Anticipated earnings surprise and other       5         1.7
Price/EBITDA and miscellaneous
  industry-specific operating statistics      5         0.2
Other combinations                            7         3.5

                                            100%




Justification                             p-value

Price/Earnings only                       0.0371
Anticipated earnings surprise only        0.0514
Miscellaneous industry-specific
  operating statistics only               0.0104
Price/Earnings and growth                 0.1751
Price/Earnings, growth, and
  anticipated earnings surprise           0.2952
Price/Earnings, growth, and other         0.5388
Price/Earnings and anticipated
  earnings surprise                       0.4257
Price/Earnings and price/EBITDA           0.0514
Price/Earnings and other                  0.4722
Growth and anticipated earnings surprise  0.0514
Anticipated earnings surprise and other   0.1976
Price/EBITDA and miscellaneous
  industry-specific operating statistics  0.6703
Other combinations                        0.0610



This table summarizes the justifications cited by analysts in support of
103 stock recommendations. Panel A presents the distribution of the
number of justifications contained in each report. Panel B presents the
distribution of specific justifications. Panel C presents a tabulation
of various combinations of justifications. [chi square] statistics and
p-values in Panels B and C reflect a test for differences in the
distribution of justifications across the target price
disclosure/nondisclosure groups. Two categories, "Miscellaneous
Industry-Specific Operating Statistics" and "Other," are aggregates of
numerous justifications that are found in only one or two reports.
EBITDA is earnings before interests, taxes, depreciation, and
amortization.
TABLE 4

Pearson Correlation Matrix for Recommendations, Target Prices, and
Pseudo-Target Prices (n = 66)

                 REC  TP/P  [TP.sub.PE1]/P  [TP.sub.PEZ]/P

REC              --   0.33      -0.07^          -0.04^
TP/P                   --        0.24^           0.33
[TP.sub.PE1]/P                    --             0.82
[TP.sub.PE2]/P                                    --
[TP.sub.PEG1]/P
[TP.sub.PEG2]/P

                 [TP.sub.SEG1]/P  [TP.sub.PEG2]/P

REC                   0.39             0.38
TP/P                  0.50             0.56
[TP.sub.PE1]/P        0.42             0.23
[TP.sub.PE2]/P        0.41             0.45
[TP.sub.PEG1]/P        --              0.86
[TP.sub.PEG2]/P                         --

This table presents Pearson correlations for stock recommendations
(REC), target prices (TP), and pseudo-target prices ([TP.sub.i], where
i represents one of four different pseudo-target prices) for the
subsample with target price disclosures. [TP.sub.PE1] ([TP.sub.PE2])
is a pseudo-target price computed by multiplying the median forward
price-to-earnings ratio for the company's industry by the one-year
(two-year) ahead annual analyst earnings forecast. The industry median
forward price-to-earnings ratios are obtained from I/B/E/S and computed
in the month preceding the date of the applicable analyst report.
[TP.sub.PEG1] ([TP.sub.PEG2]) is a pseudo-target price computed based
on the PEG heuristic, obtained by multiplying the analyst's projected
long-term earnings growth (stated in percent) times the one-year
(two-year) ahead annual analyst earnings forecast. When the analyst
report is missing a long-term growth projection, I substitute the most
recent growth projection available for the company by that analyst or
the median growth projection from I/B/E/S for that company in the month
prior to the date of the analyst report. All target price variables are
scaled by stock price per share (P) as of the day prior to the day of
the analyst report. Share trading prices are obtained from CRSP. REC is
coded on a scale from 1 (Strong Sell) to 5
(Strong Buy). The sample size for all correlations is 66, reflecting the
initial 67 target price disclosure reports minus one with forecasted
losses at both horizons.

All correlations are significant at the 0.01 level or better except
those denoted by^, which are not significant.
TABLE 5

Mean Target Prices and Pseudo-Target Prices Partitioned by Stock
Recommendation

Panel A: Stock Recommendations Accompanied by a Target Price Disclosure

Stock
Recommendation                 n    TP/P    [TP.sub.PE1]/P

Hold                           8    1.10         0.95
Buy                           32    1.36         1.13
Strong Buy                    26    1.44         0.95
All                           66    1.36         1.03

[H.sub.0]: Hold = Buy?
 test statistic                     2.8          0.7
 (p-value)                        (0.0040)     (0.2356)

[H.sub.0]: Buy = Strong Buy?
 test statistic                     1.0          -1.4
 (p-value)                        (0.1598)     (0.9129)

Stock
Recommendation                [TP.sub.PE2]/P  [TP.sub.PEG1]/P

Hold                               0.92            0.73
Buy                                1.10            0.94
Strong Buy                         0.96            1.23
All                                1.02            1.03

[H.sub.0]: Hold = Buy?
 test statistic                    0.9              1.4
 (p-value)                       (0.1868)        (0.1009)

[H.sub.0]: Buy = Strong Buy?
 test statistic                    -1.2             2.6
 (p-value)                       (0.8850)        (0.0057)

Stock
Recommendation                [TP.sub.PEG2]/P

Hold                               0.88
Buy                                1.16
Strong Buy                         1.57
All                                1.29

[H.sub.0]: Hold = Buy?
 test statistic                     1.5
 (p-value)                       (0.0665)

[H.sub.0]: Buy = Strong Buy?
 test statistic                     2.4
 (p-value)                       (0.0115)
Panel B: Stock Recommendations Not Accompanied by a Target Price
Disclosure

Stock
Recommendation                 n  [TP.sub.PE1]/P  [TP.sub.PE2]/P

Hold                          20       1.02            0.96
Buy                            9       1.03            1.01
Strong Buy                     3       0.86            0.90
All                           32       1.01            0.97

[H.sub.A]: Buy > Hold?
 test statistic                        0.1             0.3
 (p-value)                           (0.4770)        (0.3883)

[H.sub.A]: Strong Buy > Buy?
 test statistic                        -0.8            -0.6
 (p-value)                           (0.7753)        (0.7229)

Stock
Recommendation                [TP.sub.PEG1]/P  [TP.sub.PEG2]/P

Hold                               0.96             1.20
Buy                                1.27             1.63
Strong Buy                         0.86             1.10
All                                1.04             1.32

[H.sub.A]: Buy > Hold?
 test statistic                     1.0              1.1
 (p-value)                       (0.1535)         (0.1408)

[H.sub.A]: Strong Buy > Buy?
 test statistic                    -0.9             -0.9
 (p-value)                       (0.8011)         (0.7953)

This table presents the distribution of target prices (TP) scaled by
actual trading price (P) for the subsample of reports, and the
distribution of pseudo-target prices (TPi, where i represents one of
four different pseudo-target prices) scaled by P. [TP.sub.PE1]
([TP.sub.PE2]) is a pseudo-target price computed by multiplying the
median forward price-to-earnings ratio for the company's industry by the
one-year (two-year) ahead annual analyst earnings forecast. The industry
median forward price-to-earnings ratios are obtained from I/B/E/S and
computed in the month preceding the date of the applicable analyst
report. [TP.sub.PEG1] ([TP.sub.PEG2]) is a pseudo-target price computed
based on the PEG heuristic, obtained by multiplying the analyst's
projected long-term earnings growth (stated in percent) times the
one-year (two-year) ahead annual analyst earnings forecast. When the
analyst report is missing a long-term growth projection, I substitute
the most recent growth projection available for the company by that
analyst or the median consensus growth projection from I/B/E/S for that
company in the month prior to the date of the analyst report. All target
price variables are scaled by stock price per share (P) as of the day
prior to the day of the analyst report. Share trading prices are
obtained from CRSP.

The sample is partitioned into subsamples according to whether a target
price was disclosed (Panel A) or was not disclosed (Panel B). Tests of
differences are shown in the bottom sections of each panel and are
one-sided t-tests for the alternative hypothesis ([H.sub.A]) predicting
that more favorable recommendations are justified by a higher ratio of
TP (or pseudo-target price) to P.
TABLE 6

Mean Levels of Analysts' Consensus for Earnings and Earnings Growth
Forecasts, for Recommendations with and without Target Price Disclosure

                        Target Price Disclosed?
Variable                   No           Yes       Difference  p-value

n                          32            66           --        --
Standard Deviations
  FEPS (1)               0.004         0.002        0.002     0.0405
  FEPS (2)               0.014         0.003        0.011     0.0189
  LTG                    0.045         0.041        0.004     0.3273

This table presents means for standard deviations of analysts' consensus
for earnings forecasts and longterm growth projections. Standard
deviations are obtained from I/B/E/S as of the month prior to the date
of the analyst' report for each sample company. [FEPS.sub.1]
([FEPS.sub.2]) refers to the analysts' one-year (two-year) ahead
earnings per share forecast. LTG is the analysts' projection of
long-term growth. The standard deviations of [FEPS.sub.1] and
[FEPS.sub.2] are scaled by the closing share price on the trading day
just preceding the date of the stock recommendation. Share trading
prices are obtained from CRSP. p-values are from one-sided t-tests for
the alternative hopothesis that the level of consensus is lower for
nontarget price disclosure reports.

Submitted: April 1999

Accepted: October 2001

(1.) Report dated February 8, 1999.

(2.) An alternative methodology is to conduct protocol analyses, in which actual analysts are asked to talk aloud as they prepare a company analysis. This approach is adopted by Biggs (1984) and Bouwman et al. (1987). However, in those studies, the primary sample variation is that of the individual analysts rather than the analyzed companies, and the researcher can never be certain that behavior observed in the laboratory corresponds to that outside the laboratory.

(3.) PEG is defined as the (PIE) + LTG, where LTG is the analysts' projection of the three- to five-year annual growth rate in earnings per share stated in percent (i.e., times 100). Setting PEG equal to 1 and rearranging terms to solve for "P," the PEG ratio (coupled with the assumption that a value of 1 is "normal") implies that price should equal E*LTG.

(4.) Actual data collection took place during the last week of April 1999. Thus, "most recent" is relative to this date. A list of the sample analyst reports, brokerage affiliation, and companies is available from the author.

(5.) In addition to brokerage reports, Investext also includes other investor reports that are not applicable to this study, and were ignored. Most relate to debt ratings, bank ratings, or corporate summary information. These included reports by Moody's Investor Services, Securities Data Company, Institutional Share. holder Services, Corporate Technology Information Services, Duff & Phelps Credit Rating Company, and Fitch Investors Service.

(6.) These are Merrill Lynch Capital Markets (16 percent), Painewebber (8 percent), Bancboston Robertson Stephens (6 percent), Morgan Stanley Dean Witter (6 percent), and Raymond James & Associates (6 percent).

(7.) Only this report has a target price that is less than the trading price. The next lowest observed TP/P is 1.06, also for a Hold recommendation.

(8.) I considered several company-specific determinants of target price disclosure such as company size, industry membership, and the existence of forecasted losses. Target price disclosure is unrelated to size and industry membership. There are five sample companies with forecasted losses and only one of these companies has a target price disclosure. The analyst for this company has a Buy recommendation; the recommendations for the other four companies include two Holds and two Strong Buys. There are no obvious similarities among the reports with forecasted losses.

(9.) A forward P/E ratio is the current price per share divided by the forecasted earnings per share.

(10.) For example, analysts occasionally state that their earnings estimates are "conservative" or "on the low side." The sample reports are selected primarily from recent years (i.e., 1998-1999), a period during which academic research and the popular press argue that analysts have indeed become less optimistic in their earnings forecasts than in previous years (Schonfeld 1998; Matsumoto 1999).

(11.) The formulation used here is the forward PIE ratio divided by the projected three-to five-year long-term earnings growth rate, although other ways of calculating PEG are possible (e.g., trailing PIE ratio, average historical growth, etc.).

(12.) In my reading of the sample reports, numerous analysts appear to use this heuristic as a basis for disclosed target prices either implicitly or explicitly.

(13.) Applied Graphics Technologies report by SG Cowen Securities Corporation, dated November 27, 1998.

(14.) Dialogic Corp. report by John G. Kinnard & Co., dated January 15, 1999.

(15.) In 12 reports, there is no long-term growth projection, and I obtain a growth projection from (1) an earlier report by the same analyst-company or (2) the median analyst projection in the prior month from I/B/E/S. The results are insensitive to deleting these observations from the tests.

(16.) Of the five reports with losses, only one has a disclosed target price. Thus, the remaining 98 reports are split into two subsamples of 66 (target price disclosure subsample) and 32 (target price nondisclosure sample).

(17.) Spearman correlations are qualitatively similar and are omitted for simplicity.

(18.) In unreported results, I also estimate a regression of TP on forecasted earnings, projected earnings growth rates, and interaction terms, where the interaction terms are the PEG-based valuations (i.e., FEPS*LTG). Those results confirm the inferences based on correlations and means. When the interaction terms (e.g., PEG valuations) are added to the regression, the coefficients on the earnings forecasts and growth projections are no longer significant but the coefficient on the interaction term is significant.

(19.) I thank an anonymous referee for this suggestion.

(20.) It is possible that all other analysts following the nontarget price sample companies did provide target prices in their reports. However, this possibility seems sufficiently remote that it may be dismissed.

(21.) I also investigated the coefficients of variation of the earnings and growth forecasts, and results are qualitatively similar to those for standard deviations.

REFERENCES

Amir, E., and B. Lev. 1996. Value-relevance of non-financial information: The wireless communications industry. Journal of Accounting and Economics 22 (1-3): 3-30.

Bandyopadhyay, S. P., L. D. Brown, and G. D. Richardson. 1995. Analysts' use of earnings forecasts in predicting stock returns: Forecast horizon effects. International Journal of Forecasting 11: 429-445.

Barker, R. G. 1999. The role of dividends in valuation models used by analysts and fund managers. The European Accounting Review 8: 195-218.

Biggs, S. F. 1984. Financial analysts' information search in the assessment of corporate earnings power. Accounting, Organizations and Society 9 (3/4): 313-323.

Block, S. B. 1999. A study of financial analysts: Practice and theory. Financial Analysts Journal (July/August): 86-95.

Bouwman, M. J., P. A. Frishkoff, and P. Frishkoff. 1987. How do financial analysts make decisions? A process model of the investment screening decision. Accounting, Organizations and Society 12 (1): 1-29.

Bradshaw, M. T. 2001. How do analysts use their earnings forecasts in generating stock recommendations? Working paper, Harvard Business School, Boston, MA.

Dorfman, J. R. 1991. Nail your stock picks down with PEGs. Wall Street Journal (October 11): CI.

Francis, J., and D. Philbrick. 1993. Analysts' decisions as products of a multi-task environment. Journal of Accounting Research 31: 216-230.

Frankel, R., and O. M. C. Lee. 1998. Accounting valuation, market expectation, and cross-sectional stock returns. Journal of Accounting and Economics 25: 283-320.

Govindarajan, V. 1980. The objectives of financial statements: An empirical study of the use of cash flow and earnings by security analysts. Accounting, Organizations and Society 5 (4): 383-392.

Hayes, R. M. 1998. The impact of trading commission incentives on analysts' stock coverage decisions and earnings forecasts. Journal of Accounting Research 36: 299-320.

Jones, S. D. 1999. Heard in the Northwest: Visio may have promise, despite slowdown as it builds sales force. Wall Street Journal (March 10): 2.

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

Lynch, P. 1989. One Up on Wall Street. New York, NY: Simon and Schuster. Matsumoto, D. A. 2000. Management's incentives to influence analysts' forecasts. Working paper, University of Washington, Seattle, WA.

McNichols, M., and P. C. O'Brien. 1997. Self-selection and analyst coverage. Journal of Accounting Research 35 (Supplement): 167-208.

Previts, G. J., R. J. Bricker, T. R. Robinson, and S. J. Young. 1994. A content analysis of sell-side financial analyst company reports. Accounting Horizons (June): 55-70.

Schonfeld, E. 1998. The guidance game. Fortune (December 21): 256-257.

Serwer, A. 1998. "It's the end of the world as we know it, and I feel fine." Fortune September 28: 317.

Slater, J. 1993. Making the most of PEG as an investment measure. The Independent--London (December 23): 3.

Stickel, S. E. 1998. Analyst incentives and the financial characteristics of Wall Street darlings and dogs. Working paper, La Salle University, Philadelphia, PA.

Womack, K. L. 1996. Do brokerage analysts' recommendations have investment value? The Journal of Finance (March): 137-167.

Mark T. Bradshaw is an Assistant Professor at Harvard Business School.

This paper is based on a supporting chapter of my dissertation at the University of Michigan. I thank Gene Imhoff (the editor), Bjorn Jorgensen, Tom Linsmeier (referee), Doug Skinner, Richard Sloan, Stephen Taylor, and an anonymous referee for helpful comments. Funding from the American Institute of Certified Public Accountants and the William A. Paton Scholarship Fund is greatly appreciated.

Corresponding author: Mark T. Bradshaw

Email: mbradshaw@hbs.edu