Download Target price forecasts: Fundamentals and behavioural factor

Target price forecasts: Fundamentals and
behavioral factors
Peter Clarkson
UQ Business School, The University of Queensland and
Beedie School of Business, Simon Fraser University
[email protected]
Alexander Nekrasov
The Paul Merage School of Business, University of California, Irvine
[email protected]
Andreas Simon
Graziadio School of Business and Management, Pepperdine University
[email protected]
Irene Tutticci
UQ Business School, The University of Queensland
[email protected]
Keywords: target price forecasts, 52-week high, investor sentiment, rounding
JEL Classification: D81, D82, G12, G14, M41.
Current Version: 8 December 2013
*Acknowledgements: We appreciate helpful comments from Mark Arnold, Philip Brown, David Hirshleifer,
Russell Lundholm, John Nowland, Gordon Richardson, Siew Hong Teoh, and seminar participants at The
Australian National University, The University of British Columbia, the University of California Irvine, The
Chinese University of Hong Kong, the Hong Kong Polytechnic University, The University of Melbourne, The
University of Queensland, Simon Fraser University, the University of Western Australia, and the 2009
AFAANZ Conference.
1
Target price forecasts: Fundamentals and behavioral factors
ABSTRACT
This paper reveals both fundamental and behavioral factors as playing an important role in analysts‟
target price formation. Analysts‟ forecasts of short-term earnings and long-term growth are shown to
be important explanatory variables for target prices; equally, the following behavioral factors that are
salient but irrelevant to fundamental value are also shown to explain target price levels and
especially target price biases: the 52-week high price, recent market sentiment, and rounding. Here,
increases in the 52-week high and market sentiment measures of one standard deviation correspond
to increases in the positive target price bias of 4.8% and 14.7%, respectively, while rounding of the
target price corresponds to an increase of 13.7%. We also find that analysts place greater weight on
these behavioral factors in settings with greater task complexity and/or resource constraints, and
when they rely on valuation heuristics as opposed to more rigorous valuation methodology, and that
this greater weight is associated with increased optimistic bias. Finally, our results show that
analysts‟ target prices are useful in predicting future stock returns beyond earnings forecasts and
commonly used risk proxies. However, in an internally consistent fashion, the informativeness of
target prices for future returns is significantly reduced when greater weight is placed on either the 52week high or recent market sentiment in the target price formation process.
Keywords: target price forecasts, 52-week high, investor sentiment, rounding
JEL Classification: D81, D82, G12, G14, M41.
2
1. INTRODUCTION
As financial market information intermediaries, equity analysts provide forecasts of a number of
key summary measures relevant to firm value.1 These summary measures are forecast on the basis of
various observable and unobservable inputs. Our study focuses on one of the key summary measures
disseminated by analysts - target price forecasts. To shed light on factors associated with analysts‟
target price forecasts, we consider the roles that both fundamental inputs and behavioral factors play in
their derivation, their accuracy and their association with future returns. Our research is motivated by
Gleason et al. (2013) who direct future researchers to consider analysts‟ use of valuation heuristics
using other judgment and decision-making settings, as their results question the extent to which
analysts rigorously adhere to methods of fundamental valuation.
Analysts‟ target price forecasts are a prediction of a stock‟s future price, generally over the 12
months following the release date (Asquith et al., 2005). This forecast is a point estimate that provides
investors with a benchmark against which to directly compare stock price in the short run.2 As Gleason
et al. (2013, pg. 86) state, target prices are more “granular, verifiable, and comparable across analysts
than buy-sell recommendations.” Until recently, research on analysts‟ forecasts has tended to overlook
analysts‟ target prices.3 One reason for this may be the expectation of a natural overlap between stock
recommendations, analysts‟ forecasts of fundamentals (earnings forecasts), and target prices. However,
prior research indicates an element of „mismatch‟ between target prices and stock recommendations,
which leads to questions about the extent to which the two measures substitute for each other (Asquith
et al., 2005; Bradshaw, 2002; Gleason et al., 2013). Further, since many analysts invest additional time
and effort in generating a target price and then publicly announce this forecast, it is likely they perceive
a net benefit in doing so.
1
Brown (1993) and Bradshaw (2011) provide broad reviews of the literature.
Target prices have become increasingly available to both researchers and individual investors. For example, finance.yahoo.com displays
consensus target price forecasts on the summary page for most companies that are followed by analysts.
3
Research on analysts‟ forecasts of target prices includes studies by Bradshaw (2002), Brav and Lehavy (2003), Bradshaw et al. (2012),
Bonini et al. (2010), and Gleason et al. (2013).
2
1
In a survey of the content of analysts‟ reports, Asquith et al. (2005) find that the three most
common methods used by analysts to determine target prices are earnings or cash flow multiples,
discounted cash flow models, and asset multiples. Their investigation reveals that price earnings
multiples are used in 99% of the analysts‟ reports they examine. Prior research also provides some
evidence of analysts‟ forecasts of short-term earnings forecasts and long-term growth being associated
with target price (Bandyopadhyay et al., 1995; Gleason et al., 2013). Directly, Gleason et al. (2013)
show that incorporating analysts‟ forecasts of earnings and long-term growth in the residual income
model or PEG ratio heuristic results in pseudo target prices that have substantial explanatory power for
actual analysts‟ target prices (with adjusted R2s in excess of 50% in most years from 1997 to 2003).
Prior research has additionally established that target price announcements have information
content for investors (Brav and Lehavy, 2003; Asquith et al., 2005). Notwithstanding, two recent
empirical studies by Bonini et al. (2010) and Bradshaw et al. (2012) find the accuracy of target price
forecasts to be relatively limited.4 We might expect that if an analyst‟s estimates of future earnings and
long-term growth are more accurate, they would translate to better target price forecasts. In fact,
research demonstrates that forecasts of earnings and stock recommendations have some level of
accuracy and that investors who follow the investment advice of the most accurate earnings forecasters
can earn abnormal returns (Ertimur et al., 2007; Loh and Mian, 2006; Simon and Curtis, 2011).
However, Bradshaw et al. (2012) find that although analysts do exhibit persistent, differential abilities
in forecasting stock prices, these differences are economically small. Gleason et al. (2013) find
superior target-price investment performance when analysts are most likely to be using rigorous
valuation techniques (the residual income model) rather than simple heuristics (the PEG ratio
heuristic). Consistent with Asquith et al. (2005), their findings also indicate that the majority of
analysts use simple heuristics and do not exhibit significant ability in forecasting future stock prices.
4
Bradshaw et al. (2012) show that returns based on target prices on average exceed actual returns by 15%, and absolute target price
forecast errors average 45%. Asquith et al. (2005) find that only 54% of target price forecasts from the Institutional Investor’s “All
American” team analysts are achieved at any time during the one-year forecast horizon.
2
There is growing evidence in the finance and accounting literature that stock prices are affected by
behavioral factors such as the 52-week high price and market sentiment. Thus, it is quite possible that
prices forecasted by analysts are also affected by behavioral factors. Here, the effect could either be due
to analysts‟ intentional attempt to improve forecast accuracy, or due to analysts not being immune to
behavioral biases affecting average investors. Equally, given the difficulty of forecasting future stock
prices and the possibility that analysts do not have sufficiently strong incentives to spend the additional
time and effort required to rigorously forecast target prices (Bradshaw et al., 2012; Gleason et al.,
2013), it is possible that an analyst may reduce effort by making a number of simplifying assumptions
about the fundamental value of the firm and how the stock is likely to be priced by the market over the
next 12 months. On finding that some analysts appear to rely on many valuation heuristics, Gleason et
al. (2013) suggest that analysts may reduce a more complex task to a simpler judgment in order to save
time and effort. In this scenario, we would then expect a target price to be a noisy measure of
fundamental value, reflecting aspects of the analyst‟s estimate of the fundamental value of the stock
based on some valuation process as well as the analyst‟s expectation of what the market price will be in
the short run. This interpretation is consistent with theories in the behavioral literature that suggest
individuals are more likely to rely on anchors and reference points when a decision-making task is
complex (Tversky and Kahneman, 1974). Specifically, this literature indicates that estimates often start
from an initial value (an anchor) and are then adjusted to obtain a final value (Tversky and Kahneman,
1974). Anchors are generally arbitrary, uninformative values that are readily available to the individual.
In sum, compared to earnings forecasts, target price forecasts provide a direct and powerful setting to
study the effect of behavioral factors on important market intermediaries such as financial analysts. 5
The behavioral factors that we consider in this study are those commonly cited as indicators in
financial markets: (1) the 52-week high price, (2) recent market sentiment, and (3) rounded numbers.
5
Since reported earnings are determined by accounting which does not reflect the behavioral factors that affect stock price, there is less
reason to expect that analysts‟ earnings forecasts will reflect behavioral factors since earnings forecast accuracy is rewarded.
3
These choices are supported by Shiller (2005) who suggests that possible judgment anchors for stock
price levels include past prices, particularly where the price has had a suitable amount of attention; the
nearest milestone of an index such as the Dow Jones; and the nearest round number. A number of
empirical studies have linked investors‟ judgment regarding investment decisions to psychological
anchors such as the 52-week high (Baker et al., 2009; Li and Yu, 2012; Zuckerman, 2009). Empirical
studies have also shown that market sentiment affects stock returns at the firm and market levels (Baker
and Wurgler, 2006, 2007) and helps to explain bias in analyst forecasts (Bagnoli et al., 2009; Ke and
Yu, 2009; Hribar and McInnes, 2012). Finally, a body of research provides evidence of clustering in
stock prices and analyst forecasts. Of particular relevance is the inclination of individuals in Western
markets to round values, in the presence of uncertainty, to end in 0 or 5 (Ikenberry and Weston, 2007;
Dechow and You, 2012).
In conjunction with our analysis of the level of the target price forecast, we also examine whether
these three ex ante behavioral factors help explain ex post target price forecast errors (the difference
between the target price and the 12-month-ahead stock price). From an ex post perspective, target
prices may turn out to be incorrect due to many factors including changes in expectations of future
earnings. In this study, we are interested in how the initial inputs (both fundamental and behavioral)
relate to the forecast error. We argue that if we can identify ex ante factors that systematically bias
target price forecasts, then analysts may be able to improve their target price estimates. Finally, we
examine the usefulness of target price forecasts to investors by investigating the relation between target
price forecasts and future stock returns conditional on our ex ante behavioral variables.6
Our sample consists of 26,746 target price forecasts for U.S. firms over the period 1999 to 2007
representing data from 4,148 firms and 3,518 analysts. Analyst data are sourced from Thomson Reuters
I/B/E/S. Our results indicate that, as expected, analysts‟ forecasts of short-term earnings and long-term
6
Since target price forecasts are equivalent to forecasts of (ex-dividend) returns, a “good” target price forecast should predict future (exdividend) return.
4
growth help explain target price forecasts. Importantly, we also find that the behavioral factors are
important determinants, with a higher 52-week high price, higher recent market sentiment, and
rounding being associated with higher target price forecasts. 7 Further, each set of factors has
incremental explanatory power relative to the other. In this sense, the results reveal a direct role for the
behavioral factors incremental to any effect that they may indirectly have on target price through their
influence on forecasts of short-term earnings or long-term earnings growth. Further, we find that the
roles played by the behavioral factors are enhanced when stocks are more difficult to value (measured
by firm size and earnings volatility), when analysts are less well positioned to make a forecast
(measured by analyst experience and brokerage size), or when analysts rely on less rigorous valuation
techniques (as revealed by the valuation model ratio (VRM) (Gleason et al, 2013).
The results from our model of ex post target price errors indicate that there is a larger positive bias
when forecasts of short-term and long-term earnings growth are high. In conjunction, we also find, in a
manner consistent with their enhanced role in the target price formation process, that higher values of
the 52-week high and recent market sentiment, and the use of rounding, are each associated with a
larger positive bias in target price. An increase in the 52-week high and recent market sentiment of one
standard deviation corresponds to an increase in target price bias of 4.8% and 14.7%, respectively,
while rounding of the target price corresponds to an increase of 13.7%, effects that are economically
meaningful when compared with the mean target price bias of 15.6%. Further, the model incorporating
just the behavioral factors explains more of the bias in target price than a model containing just the
fundamentals, consistent with reliance on behavioral factors producing less reliable forecasts than
reliance on the fundamentals. Lastly, in a parallel fashion to our findings for the target price forecast,
the bias introduced by the behavioral factors increases with greater task complexity, greater resources
constraints, and when analysts rely on less rigorous valuation techniques (as revealed by the VRM).
7
Although rounded target prices could be a result of unbiased and rational rounding to reflect relatively large standard errors of target
price estimates, we hypothesize and find that analysts‟ rounding exhibits a significant asymmetry with the tendency to round target price
forecasts upward.
5
Finally, we find a positive association between target prices and future stock returns. However, this
association is significantly weaker when the 52-week high is high relative to the current price and when
recent market sentiment is relatively more positive. Thus, while target price forecasts appear
informative for future returns, the degree to which they are, is significantly improved when the
behavioral factors play a relatively reduced role in the target price derivation process and thus when
they make a reduced contribution to the optimistic bias in target price. Arguably, this is as expected
given that there is little theoretical basis for a relation between the behavioral factors and future returns.
Overall, we view our results as consistent with the nature of a target price, which is not designed to
be an accurate estimate of fundamental value but rather is intended to indicate where the analyst
believes price is likely to go over the next year. From this perspective, roles for both fundamental and
behavioral factors emerge. For example, if analysts are not convinced that the stock price will reflect
the fundamental value in the short term, they may adjust the target price forecast to fit recent market
expectations. Alternatively, the results may reflect analysts‟ reliance on short cuts to generate a forecast
for which there is possibly limited direct benefit of estimating accurately (Bradshaw et al., 2012), or in
circumstances where they are less well positioned to make the forecast. Finally, they also fit with the
general finding of prior research that analysts‟ forecasts are optimistically biased. Our results suggest
that the optimistic bias may, at least to a degree, result from analysts‟ in part relying on past stock and
market highs. Thus, the results consistently support the inference that although analysts use
fundamental inputs in deriving target price forecasts, they are also influenced by highly visible
reference points such as the 52-week stock price high, recent investor sentiment, and round numbers,
and further that reliance on these factors explains some of the observed bias in target prices and
ultimately their usefulness as predictors of future returns.
Our study contributes to the literature in several ways. First, it documents the importance of
behavioral factors in the target price formation process. Much of the prior literature has focused on the
6
analysts‟ use of fundamentals. This paper shows that the 52-week high price, recent market sentiment,
and rounding play economically significant roles beyond the fundamental factors.
Second, it sheds light on the source of target price errors. Our findings indicate that the 52-week
high, market sentiment, and target price rounding are more important in explaining target price biases
than are analysts‟ forecasts of earnings and long-term growth. Third, we contribute to the literature on
the usefulness of analysts‟ target prices in predicting future stock returns. Prior literature finds that
target prices can be used to predict future stock returns when analysts use more rigorous valuation
techniques or more accurate earnings forecasts (Gleason et al., 2013). We find that target prices have
significant explanatory power with respect to future stock returns beyond earnings forecasts and
commonly used risk proxies, but more so when the influence of the behavioral factors is relatively low.
Finally, our paper contributes to the growing literature on the economic effects of behavioral factors.
Extant research finds that behavioral factors affect stock prices, the exercise of options, and mergers
and acquisitions, to name a few. Here, we document the effect of these factors on target prices that are
issued by important market participants, financial analysts, and viewed as informative by investors.
The remainder of this document is organized as follows. In the next section, we review the
literature that informs our expectations regarding the target price derivation process. Section 3
describes our sample data, and Section 4 describes the methodology. The empirical results are
presented in Section 5 and Section 6 concludes.
2. RELATED RESEARCH
A number of studies relating to investment decisions have relied on theories from psychology to
explain why prices might reflect information other than fundamental value (e.g., Baker et al., 2009;
Heath et al., 1999). Tversky and Kahneman (1974) describe three heuristics that are used by
individuals to predict value: representativeness, availability, and anchoring. Representativeness refers
7
to the tendency of individuals to rely on one item because it is highly representative of (resembles)
another item. Within our context, for example, past stock price highs and market sentiment could be
considered highly representative of future prices. Tversky and Kahneman refer to the “illusion of
validity,” in which confidence in a decision is based on the fit between a predicted outcome and the
input information. Availability refers to the tendency of individuals to rely on events that are easily
brought to mind. Characteristics that enhance availability are the familiarity of an occurrence and how
recently the event took place. Finally, anchoring refers to the tendency of individuals to make an
estimate by starting from an initial value, which is then adjusted to determine the final result.
Each of these heuristics could be used to describe the use of short-term indicators by analysts in
their determination of a target share price forecast. Forecasting a target price involves a large number
of uncertainties with different probabilities of eventuating. Applications of financial theory often
employ historic values to infer expectations of future values because of the difficulty of generating
values of future events. In this sense, past prices and investor sentiment could be seen as representative
of future prices. Further, because recent prices and investor sentiment are easily observable by analysts,
they satisfy the description of availability.
These heuristics suggest that the 52-week high and recent market sentiment may serve as reference
points or judgment anchors for estimating target price.8 Baker et al. (2009) utilize this theory in
explaining merger and acquisition activity. In particular, they identify the 52-week high as a judgment
anchor. A number of the justifications they provide for this metric as an anchor are also salient to issues
faced by analysts in estimating target prices.
Specifically, the 52-week high is a value that is widely published and is the focus of analysts,
investors and corporate executives. In fact, the 52-week high price appears prominently in the summary
8
The use of these behavioural anchors may be done consciously or unconsciously by analysts in forming target price. This study does not
make this distinction. It is unlikely that the analyst would reveal that their valuation was in some way linked to the 52-week high price
unless they believed that there was some rationality to this. In fact, use of the 52-week high price as a value indicator is evident in trading
strategies of technical analysts and is known as the „breakout strategy‟.
8
information on the front page of many analysts‟ reports. It is often viewed as a high price that the
company has achieved and may recover through good management. We therefore expect that the 52week high could well provide an anchor point for analysts‟ target price forecasts, a point that is
adjusted for other considerations such as their forecast of fundamental value and recent market
sentiment. A number of studies have found the prior year‟s high price to be salient in investment
decisions. For example, Baker et al. find it to be influential in the valuation of mergers and acquisitions
while Heath et al. find that the exercise of options by employees doubles when the share price exceeds
the prior year high. In explaining returns to momentum investing, George and Hwang (2004) suggest
that investors may anchor on the 52-week high as a reference point from which to assess the
incremental value of new information, since traders are slow to react, or overreact, to good news. Using
nearness to the 52-week high as a proxy for recent good news, they find that this is a better predictor of
future returns than are past returns, suggesting that investors focus on price levels rather than price
changes. When price is close to its 52-week high, there is a reluctance to bid it higher even where a
higher price would be supported by information. Eventually the price does increase, revealing the
information. A similar behavior occurs in the event of bad news. Thus, price is sticky at values closest
to, and farthest away from, the 52-week high. Li and Yu (2012) attempt to predict aggregate excess
market returns and find that both nearness to the 52-week high and nearness to the historical high
dominate the ability of past market returns to predict future aggregate market returns, with the strongest
results for horizons of less than one year.
A burgeoning literature also provides links between analysts‟ forecasts and market sentiment.
Using Baker and Wurgler‟s (2007) market sentiment index, a number of studies show that sentiment
explains bias in analysts‟ earnings forecasts, long-term growth forecasts, and recommendations (e.g.,
Bagnoli et al., 2009; Ke and Yu, 2009; Hribar and McInnes, 2012). Bagnoli et al. show that analysts‟
stock recommendations are correlated with investor sentiment when analysts follow a greater number
9
of industries and companies, and when they issue a greater number of earnings forecasts. One possible
interpretation is that analysts take short cuts in order to produce a certain quantity of research in a
limited time. However, the results also appear to suggest an underlying belief on the part of analysts
that sentiment affects asset prices or that reliance on sentiment can substitute for a more labor intensive
analysis of fundamental value. Nevertheless, while higher market sentiment might provide a rationale
for higher stock recommendations, Bagnoli et al. find that reliance on it is associated with less
profitable stock recommendations.
A few recent studies have provided evidence that investor sentiment influences short-term earnings
forecasts. For example, Hribar and McInnes (2012) find that investor sentiment affects analyst
expectations across all stocks, with earnings forecasts becoming more optimistic when sentiment is
high and less optimistic when sentiment is low. Ke and Yu (2009) show that the translation of analysts‟
forecasts into profitable stock recommendations is adversely affected by periods of extreme investor
sentiment and a high reliance on trading commissions. They also find that institutional ownership and
insider trading have a negative effect.
In a study of stock returns, Baker and Wurgler (2007) show that certain types of firms are more
likely to be sensitive to market-wide sentiment: small firms, young firms, high volatility firms,
unprofitable firms, non-dividend-paying firms, extreme growth firms, and distressed firms. They argue
that such firms are more difficult to value, which means that valuation mistakes are more likely. They
also point out that stocks of such firms tend to be the riskiest and costliest to arbitrage. This means that
prices will not always reflect fundamental value.
Financial markets research also demonstrates the impact of anchoring on round numbers.
Following Shiller‟s (2005) comment that market participants will rely on the nearest round number
when making judgments about value, Westerhoff (2003) applies this anchor to model traders‟ behavior
in foreign exchange rate markets and finds that anchoring leads to periods of excessive volatility and
10
persistent exchange rate misalignment. In the analyst literature, Dechow and You (2012) show that
46.3% of EPS forecasts in a sample of 809,129 forecasts over the period 1984 to 2009 are rounded to
the nearest five cents. Their results suggest that analysts are more likely to round earnings per share
(EPS) forecasts when the economic significance of rounding is smaller, when the followed firm is
unlikely to generate large brokerage fees, and when the number of analysts following the firm is
greater. They argue that analysts are motivated to use round numbers where the costs of generating a
more accurate forecast of EPS outweigh the benefits, and suggest that rounding reflects analysts‟ views
of the lack of precision in forecasting to the nearest penny.
Finally, evidence of price clustering in equity and other asset markets is also well-established, with
a number of rational and behavioral explanations being advanced to explain the phenomenon.9 Rational
explanations argue that market participants round numbers in circumstances of uncertainty to reduce
various costs (e.g., the resolution hypothesis by Ball et al., 1985 and the negotiation hypothesis by
Harris, 1991). Ikenberry and Weston (2007) find that the extent of clustering on the NYSE and Nasdaq
around increments of 5 and 10 cents cannot be completely explained by rational economic theories and
suggest that the phenomenon is also a function of a fundamental psychological bias for prominent
numbers. They refer to the psychology literature that finds rounding bias in the context of various
numeric-based tasks. This literature suggests that some numbers are easier to process than others and
that this is reflected by the rounding of numbers during tasks that are time-based, involve large
numbers, and have high levels of difficulty (Shepard et al., 1975; Hornick and Zakay, 1994; Loomes,
1988). On this basis, we predict that price clustering will also be observed in analysts‟ target price
forecasts, since the task of forecasting future stock prices involves a high level of uncertainty and high
information seeking costs.
In sum, we expect that analysts‟ forecasts of fundamentals, the 52-week high price, and recent
9
Evidence of price clustering has been found for the US equity markets (Harris 1991; Christie and Schulz, 1994; Grossman et al., 1996),
gold markets (Ball et al., 1985; Grossman et al., 1996), foreign exchange markets (Grossman et al., 1996), and the London equity index
futures and options markets (Gwilym et al., 1998).
11
market sentiment will each be positively related to their forecasts of target price. We also expect that
the price clustering and inherent optimism observed in analysts‟ forecasting activities will result in
analysts rounding their target prices upward more often than downward. Further, the use of behavioral
factors is likely to induce bias in target price; thus, we expect that a greater reliance on these factors
will lead to larger target price errors and thereby likely introduce noise into the association between the
target price and future stock returns.
3. METHOD
3.1 Target Price Forecast
If, in addition to fundamentals, analysts utilize the identified behavioral factors (52-week high
stock price, recent investor sentiment, and rounded numbers) as reference points or judgment anchors
when setting their target price forecasts, we would expect their usage to bias target prices away from
fundamental value. To test this conjecture, we estimate the following regression model:
TPijt = 0 + 1 STE1ijt + 2 DIFFSTEijt + 3 LTGijt + 4 52WHit + + 5 SENTt
+ 6 ROUNDijt + γ CONTROLit + εijt
(1)
where
TPijt
=
analyst j‟s target price forecast for firm i at time t deflated by the closing price
on the trading day before the target price announcement date;
STE1ijt
=
analyst j‟s forecast at time t of one-year-ahead earnings for firm i scaled by the
closing price on the trading day before the target price announcement date;
DIFFSTEijt
=
the difference between analyst j‟s forecast at time t of two-year-ahead and oneyear-ahead earnings for firm i scaled by the closing price on the trading day
before the target price announcement date;
LTGijt
=
analyst j‟s growth rate forecast at time t for firm i‟s earnings over the next three
to five years;
52WHit
=
the highest stock price for firm i over the 52-week period preceding the target
price announcement date t scaled by the closing price on the trading day before
the target price announcement date;
12
SENTt
=
the monthly Baker and Wurgler (2007) investor sentiment index for the month
prior to the target price announcement month;10
ROUNDijt
=
a categorical variable set equal to 1 if analyst j‟s target price forecast for firm i
is rounded to the nearest dollar at time t, and 0 otherwise;
and CONTROL is a vector of nine control variables considered relevant to target price (discussed
below). All variables are measured at the time the target price forecast is released. The models are run
as pooled time-series, cross-sectional regressions with corrections for clustering of standard errors by
firm and month (Petersen, 2009). For ease of reference, Table 1 summarizes the independent variables
of primary interest along with their predicted signs for this and subsequent regression models.
Within this model, analysts‟ forecasts of fundamentals are captured by STE1, DIFFSTE, and
LTG.11 Based on the conceptual arguments underlying a valuation model and the results of prior
research (e.g., Bandyopadhyay et al., 1995), we expect a positive association between each of these
variables and the target price forecast measure, TP (i.e., 1, 2, 3 > 0). Alternatively, the variables
designed to capture the selected behavioral factors are 52WH, SENT, and ROUND. Here, if analysts use
the 52-week high as a reference point in forming their target price forecasts, we would expect the
coefficient on 52WH to be positive (i.e., 4 > 0). Similarly, if investor sentiment has an influence on the
setting of the target price, we expect the coefficient on SENT to be positive (i.e., 5 > 0). Finally, if as
documented in prior literature, analysts are optimistic in their forecasting behavior, we would expect
them to disproportionately round upward to the next higher dollar amount. Thus, we expect the
coefficient on ROUND to be positive (i.e., 6 > 0). Note, given model specification, any role played by
the behavioral factors in the setting of the target price is incremental to their influence on either shortterm earnings or long-term earnings growth.
10
Monthly investor sentiment is accessed from www.stern.nyu.edu/~jwurgler. The investment sentiment variable is originally defined
using annual data in Baker and Wurgler (2006) in equation (3) and denoted as SENTIMENT┴.
11
We use DIFFSTE instead of two-year-ahead earnings forecast, STE2, because STE1 and STE2 are highly correlated at around 87%.
DIFFSTE captures the incremental contribution of STE2 over STE1 and can be interpreted as a measure of short-term earnings growth.
13
In addition to the fundamental and behavioral factors, our econometric model also includes nine
control variables considered relevant to target price derivation. These control variables relate to both
analyst/brokerage characteristics and firm-specific characteristics. The analyst/brokerage characteristics
we include in the model are:
REP
=
analyst reputation, measured as a categorical variable set equal to 1 if the analyst
is named as an “All American” team analyst by the Institutional Investor, and 0
otherwise;
EXP
=
analyst experience in forecasting firm i, measured as the number of years in
which the analyst has issued target price forecasts for the firm;
B_SIZE
=
brokerage size, measured as the number of analysts associated with a particular
broker in a given year; and
CONFLICT
=
a measure of possible investment banking-related conflicts of interest that may
influence analysts‟ opinions which, following Ertimur et al. (2007), is based on
the Carter and Manaster (1990) rankings as updated by Loughran and Ritter
(2004); following Gleason et al. (2013), we assign a value of 1 if the brokerage
firm has the top investment banking reputational rank, 0.5 if it has a lower
reputational rank, and 0 if it does not have a Carter-Manaster reputational rank.
Prior research indicates that analysts‟ estimates of summary measures – forecasts of earnings and
stock recommendations – vary in accuracy and often display optimistic bias (Loh and Mian, 2006;
Ertimur et al., 2007; Bradshaw, 2002; Bradshaw et al., 2012), with various factors at both the analyst
and brokerage levels identified as contributing to this variation. We thus include the three common
analyst- and broker-level variables identified above (REP, EXP, and B_SIZE) to control for the
influence of these factors on analysts‟ target price forecasts. Each has generally been positively
associated with the accuracy of forecasting and recommendations (Stickel, 1990; Park and Stice, 2000;
Gleason et al., 2013). Since the optimistic bias in analysts‟ forecasts and stock recommendations has
also been associated with investment banking relations, we additionally include CONFLICT to capture
any inflation in the target price induced by such bias (Lin and McNichols, 1998; Dechow et al., 2000;
O‟Brien et al., 2005; Gleason et al., 2013). Note, since each of these characteristics is associated with
bias in analysts‟ forecasts and stock recommendations, we expect them to also be related to both the
bias in, and the level of, TP since characteristics associated with more upward bias should result in a
14
higher target price relative to the current market price. As such, we include the same analyst and broker
controls in both the target price model and the target price error model (discussed in the next section).
The control variables included in the model to capture firm-specific characteristics are:
β
=
the firm‟s CAPM beta, estimated from a regression of firm returns minus the riskfree (one-month T-bill) rate on the value-weighted market index minus the risk-free
rate over a period of 60 months preceding the target price month;
SIZE
=
firm size, measured as the log of market capitalization on the trading day before the
target price announcement date t;
BM
=
the book-to-market ratio, calculated as the book value of equity divided by the
stock price at the end of the previous fiscal year;
RETt-1
=
the past one-year return calculated as the 250-trading-day buy-and-hold return
ending one day before the target price announcement date; and
RETstd
=
return volatility, measured as the daily return volatility over the 250-trading-day
period ending one day before the target price announcement date.
These firm-specific characteristics are included as they have been shown to be associated with
analyst following and their ability to forecast accurately. Target price, as a forecast of value, will be a
function not only of the forecast of earnings and long-term growth but also of expectations of risk, both
systematic and unsystematic (Kerl, 2011). We include β, BM, SIZE and RETstd to proxy for a firm‟s risk
(Lui et al., 2007). Firms with higher betas, higher book-to-market ratios, and smaller market
capitalization are viewed as riskier and therefore more difficult to value accurately. Also, higher risk
can result in a higher discount factor being used to discount future cash flows and hence a lower
fundamental value and target price. We include past returns, RETt-1, to control for analyst bias in target
price arising from selecting past winners to follow. Hayes (1998) argues that analysts‟ incentives to
gather and provide information are strongest for stocks that are expected to perform well. Ertimur et al.
(2007) provide empirical evidence that analysts initiate recommendations following high returns and
prior growth. Again, the inclusion of these firm-specific characteristics in both the target price and
target price error models is based on the argument that characteristics associated with more upward
bias should result in a higher target price forecast and hence greater target price forecast error.
15
Finally, we conduct our analysis in three stages. First, we consider four variants of our primary
model (equation 1) in which we include the sets of fundamental and behavioral factors, and the set of
control variables, in various combinations. Second, for robustness purposes, we extend the model to
include a lagged target price forecast and then alternatively to include the two pseudo-target prices
developed by Bradshaw (2004) and Gleason et al. (2013), one based on the residual income model
(RIM) (VRIP) and one based on PEG ratio heuristic (VPEGP). Here, we follow the procedures
described in Gleason et al. to determine each of these pseudo target prices. Third, since our arguments
envisage an enhanced role for the behavioral anchors in situations where stocks are more difficult to
value and/or when analysts are less well positioned to make a forecast, we re-run the model with a
series of interaction effects. Specifically, we interact both the fundamental and behavioral factors in
turn with each of firm size (SIZE), earnings volatility (EV), analyst experience (EXP), and brokerage
size (B_SIZE) under the argument that smaller firms and firms with greater earnings volatility will be
more difficult to value, while analysts with less experience and from smaller brokerages will be in a
weaker position to make a forecast. In conjunction, we also run a model in which we interact these
factors with the valuation model ratio (VMR) developed by Gleason et al. As described by Gleason et
al., the VMR facilitates inferences about valuation model usage, with values of VMR exceeding one
indicative of the analysts favoring the PEG valuation heuristic and alternatively, for values less than
one, favoring RIM. Based on the findings in Gleason et al. of an improved target price process when
the forecasts are derived using a more rigorous valuation technique rather than an heuristic, we expect
analyst reliance on the behavioral factors to be positively related to the VMR. For this final analysis,
we set the operational notion of VMR equal to 1 if the firm‟s VMR exceeds the median VMR and 0
otherwise.
16
3.2 Target Price Error
The fact that analysts‟ target price forecasts are influenced by the behavioral factors does not
necessarily imply that this influence is detrimental to the target price forecast accuracy. To test whether
the influence of the behavioral factors is indeed detrimental, we estimate the following target price
error model:
TPerrorijt = δ0 + δ1 STE1ijt + δ2 DIFFSTEijt + δ3 LTGijt + δ4 52WHit + δ5 SENTt
+ δ6 ROUNDijt + γ CONTROLit + υijt
(2)
where all variables with the exception of TPerror are as previously defined, and the vector of control
measures includes the same analyst/brokerage and firm-specific characteristics employed in the target
price model above (equation (1)).12 As noted, Table 1 provides a summary of the primary independent
variables along with their descriptions and predicted signs.
Following Bradshaw et al. (2012), we define target price error, TPerrorijt as the difference between
the one-year-ahead stock price (Pt+1) and the target price forecast, each scaled by the closing price on
the trading day before the target price announcement date (Pt) (i.e., TPerrort = (Pt+1/ Pt) – TPt). A
negative value of TPerror therefore implies an optimistic bias in the target price forecast. Alternatively,
if a firm is delisted, Pt+1 is not available. In these instances, to avoid look-ahead bias, we use the return
to the firm‟s shares over the period prior to delisting (Rt+1), determining the target price forecast error
as (Rt+1 + 1) – TPt.
As argued within the context of the target price forecast model above, we expect that a higher 52week high share price and more positive recent investor sentiment will each map into a higher target
price forecast. Further, we argue that rounding to the nearest dollar will more often involve rounding
up, since forecasts tend to exhibit an optimistic bias. To the extent that these forces lead analysts to a
more optimistic target price forecast, we expect a reliance on them to result in a larger (more negative)
12
As explained in the previous section, both sets of characteristics have been shown to be associated with biases in analysts‟ forecasts and
recommendations, so they are also likely to be related to both the level of and errors in analysts‟ target prices. As we discussed in Section
4 and revealed in Table 3, our sample target price forecasts do indeed exhibit a positive (optimistic) bias, on average.
17
target price forecast error. Thus, we expect the coefficients on each of 52WH, SENT, and ROUND to be
negative (i.e., δ4 < 0, δ5 < 0, and δ6 < 0). Again, the role played by the behavioral factors is incremental
to any error they introduce into forecasts of short-term earnings and long-term earnings growth.
Finally, we adopt the same three stage strategy underlying the analysis of the target price forecast
as described in the previous section. Again, the models are run as pooled time-series cross-sectional
regressions with standard errors clustered by firm and month (Petersen, 2009).
3.3 Predictability of Future Stock Returns
If our findings reveal that reliance on salient but economically irrelevant behavioral measures
induces biased target prices, we might expect the association between target prices and future stock
returns to decline as the influence of these measures increases. If, for example, when the 52-week high
price is significantly above the current stock price, it influences analysts pulling their target price
forecasts away from the fundamental value implied by the forecasts of fundamentals, we might expect
to find a weak, or perhaps even no, association between target prices and future returns. In contrast,
when the 52-week high is close to the current price, the effect of the 52-week high on the target price
forecast should be relatively small. We might then expect a relatively stronger positive association
between target prices and future returns for such stocks. In a similar fashion, we would expect that
when recent market sentiment is relatively high, there will be a larger „sentiment induced‟ bias in target
prices and therefore a weaker relation between the target price forecast and future returns.
To test this conjecture, we estimate the following regression model of future stock returns on target
prices and control variables:
RETt+1 = λ0 + λ1 TPijt + λ2 TPijt*52WHit + λ3 TPijt*SENTt + λ4 TPijt*ROUNDijt
+ λ5 STE1ijt + λ6 DIFFSTEijt + λ7 52WHit + λ8 SENTt + λ9 TPijt*ROUNDijt
+ φ CONTROL*it + πijt
(3)
where RETt+1 is the raw ex-dividend stock return measured over the 250-trading-day period following
the target price announcement day (t) using daily ex-dividend returns from CRSP, CONTROL* is a
18
vector of control variables limited to firm-specific variables (, lnSIZE, BM, and RETt-1),13 and all
variables are as previously defined. Within the context of this model, if the target price forecast has
incremental predictive power with respect to future stock price, we would expect a positive association
between TP and future returns (i.e., λ1 >0). Further, if higher values of 52WH and SENT have an
increased influence on analysts thereby pulling their target price forecasts further away from the value
implied by the forecasts of fundamentals, we would expect the coefficients on the interaction terms,
TP*52WH and TP*SENT, to be negative (i.e., λ2 >0 and λ3 >0). Alternatively, since the rounding of a
target price is unlikely to predict future stock return, we make no prediction for the coefficient on
TP*ROUND. Note, we exclude LTG because, as Table 2 reveals, the requirement for long-term growth
forecasts significantly reduces sample size and thus the power of the returns tests.14 We estimate this
model as a pooled time-series cross-sectional regression and correct t-statistics for clustering of
standard errors by firm and month (Petersen, 2009). Again, Table 1 provides a summary of the primary
independent variables, along with their predicted signs.
4. SAMPLE DATA
To construct our sample, analyst forecasts of target prices, earnings, and long-term growth were
collected from the Thomson Reuter I/B/E/S Detail History file over the period January 1999 to
December 2007. Financial data were then collected from CRSP and COMPUSTAT, monthly returns,
daily returns, the risk-free rate and the market value of equity from CRSP and the book value of equity
from COMPUSTAT. The final sample consists of 26,746 target price forecasts for 4,148 U.S. firms
issued by 3,518 analysts. Table 2 presents the sample selection criteria. As revealed, the sample size is
13
Since we do not expect that individual analyst and broker specific characteristics will help to explain firm-level future returns, we
include only firm specific controls that have been found to be associated with future stock returns.
14
Since returns tests usually have relatively low explanatory power (adjusted R2s), sample size is especially important for such tests. As
discussed in Section 5.5, our results for the returns model are robust to the inclusion or exclusion of these variables.
19
greatly reduced by the requirement that an analyst must also provide short-term earnings and long-term
growth forecasts on the date of their target price forecast.15
Table 3 presents descriptive statistics. All variables except future ex-dividend stock returns,
RETt+1, are winsorized at the top and bottom 1% level. As revealed, the mean (median) value of the
target price forecast scaled by the closing price on the previous day (TP) is 1.228 (1.185), with a
standard deviation of 0.276. Thus, on average, a target price forecast is a prediction that the stock price
will exceed its current level by 22.8% in the next 12 months. The mean (median) target price forecast
error, also scaled by the previous day closing price (TPerror), is -0.156 (-0.139), with a standard
deviation of 0.566. These statistics reveal a positive bias, with the scaled target price forecast exceeding
the scaled realized price by 15.6% on average.16 Finally, the mean (median) one-year-ahead exdividend stock return over the sample period is 7.6% (3.5%).
Turning to the behavioral factors, we find that the mean 52-week high price (52WH) is 51.7%
higher than the stock price on the day prior to the target price announcement, while the median is
18.5% higher. Perhaps coincidentally, the median value of 52WH (1.185) is the same as the median
value for the target price forecast measure, TP. However, the range for TP is narrower than the range of
52WH. These summary statistics appear consistent with our expectation that analysts rely on past price
data. The proximity of the target price forecast to the 52-week high (PROX_52WH) is, on average, 151
days (0.414 times 365 days).
To further illustrate the relation between TP and 52WH, we plot the density distribution of the
target price relative to the 52-week high (Figure 1, Panel A). The plot reveals a sharp spike and the
absolute maximum of the density at the 52-week high. A binomial test rejects the null hypothesis that
the number of observations in the interval that includes 52-week high equals the average of the two
15
Even though the sample is greatly reduced by the requirement for LTG forecasts, Jung et al. (2012) find that analysts that publicise
LTG forecasts are signalling effort in their recommendations. Since it seems less likely that such analysts would consciously rely on
behavioral factors, this sampling constraint is likely to skew our analyses against finding a role for the identified behavioural factors.
16
The direction and magnitude of the bias are consistent with Bradshaw et al., 2012, who find that the average target price error in their
sample is -15%.
20
immediately adjacent intervals at less than the 1% level.17 To check the robustness of this result, we
also change the scale variable from the stock price to the 52-week high price (Panel B) and the size of
each interval from 0.01 to 0.001 (Panel C). The plots in both Panel B and Panel C exhibit similar
spikes, and absolute maximums, in the target price density at the 52-week high.
Returning to Table 3, we find that the sample mean (median) for investor sentiment, SENT, is
0.086 (-0.060). Thus, while the mean is suggestive of positive investor sentiment, on average, the
negative median value indicates that observations for more than one-half of the sample are derived
from periods when investor sentiment was low. Additionally, the mean value of the rounding measure
(ROUND) is 0.874. Thus, for our sample, 87.4% of the target price forecasts are presented as an even
dollar amount (i.e., rounded to the nearest dollar figure).
Finally, Table 3 reveals that less than 10% of our analysts are classified as “All American” team
analysts (the mean value of REP is 8.8%) and that analysts have on average approximately two years of
experience forecasting target prices for a given firm in our sample (based on their target price forecasts
collected by I/B/E/S, the mean value of EXP is 2.097). Investment banking relations for our sample
brokers is also toward the lower end, with a median value for CONFLICT is 0.500.
Lastly, Table 2 indicates we lose 169,642 observations due to the lack of a long-term growth
forecast. To assess whether there are any substantial differences between our sample and the sample of
firms without LTG forecasts, we compare summary statistics between the two samples. The
untabulated results indicate that the greatest differences relate to analyst and broker characteristics.
Analysts who provide LTG forecasts tend to be less experienced (the mean value of EXP is 2.097 years
for the LTG sample and 2.600 years for the non-LTG sample). Further, the CONFLICT measure has a
mean of 0.352 for the LTG sample and a mean of only 0.277 for the non-LTG sample, suggestive of
greater investment banking activity between the firm and the broker for firms with LTG forecasts.
17
The binomial test is similar to the test Burgstahler and Dichev (1997) use to examine the discontinuity of earnings distribution around
zero.
21
5. RESULTS
5.1 Univariate Correlations
Pearson pair-wise correlations for the variables used in the models are presented in Table 4.
Consistent with our expectation that analysts appeal the 52-week high price when forecasting target
price, the table shows an economically significant correlation of 29.2% between the TP and 52WH. The
investor sentiment measure is also positively associated with analysts‟ forecasts of target price and
long-term growth, and the 52-week high price. Turning to fundamentals, the two-year-ahead growth,
DIFFSTE, and long-term growth, LTG, are positively correlated with TP, while higher forecasts of
STE1 are associated with lower TP. The latter result is likely to be driven by a trade-off between the
next-year and longer-term forecasts, with higher one-year-ahead earnings forecasts being associated
with lower two-year-ahead and long-term growth forecasts. This is consistent with the expectation that
extreme earnings revert to an average level over time, and it highlights the importance of the
multivariate analysis presented in the next section.
Correlations for the control variables (not tabulated) indicate the following relations. Analyst
reputation (REP) is positively associated with experience (EXP) and broker size (B_SIZE). That is, “All
American” team analysts are more experienced and more likely to work for larger brokers. Further,
analysts are more likely to provide lower target price, DIFFSTE, and LTG forecasts when they are “All
American” team analysts, have more experience, or are from larger brokerages. Investment banking
relations are associated with optimistic bias in target price forecasts as indicated by a negative
correlation between the CONFLICT measure and the target price error.
5.2 Target Price Forecast Model Results
Table 5 presents results based on the target price forecast model (equation (1)). As described in
Section 3.1, we conduct our analysis in three stages. First, we consider four variants of the model in
22
which we include fundamental and behavioral factors in various combinations. Following, for
robustness purposes, we extend the model to include a lagged target price forecast and then
alternatively to include the two pseudo-target price forecasts developed by Gleason et al. (2013).
Finally, we run a series of models with various interaction effects to explore whether the roles played
by the behavioral factors are enhanced when stocks are more difficult to value, when analysts are less
well positioned to make a forecast, or when analysts rely on less rigorous valuation techniques.
5.2.1 Primary Model Results
Results for the following four variants of the primary model are presented in Panel A of Table 5:
Model 1 includes only the fundamental factors, Model 2 includes only the behavioral factors, Model 3
includes both sets of factors, and finally, Model 4 includes both sets of factors along with the control
variables. Here, to begin, the adjusted R2s are 16.5% for Model 1 with only the fundamental factors,
9.8% for Model 2 with only the behavioral factors, and 22.7% for the composite model (Model 3).
Thus, while it appears that the fundamental factors play a stronger role in explaining target price
forecasts, the behavioral factors also play an (incrementally) important role. This conclusion is
supported by the Vuong test statistics (not reported) comparing the adjusted R2s among the three
models, with the null hypothesis of no difference being rejected at the 1% level for all comparisons.18
Turning to the individual factors, given the similarity in results across models, for parsimony, we
restrict our discussion to the results for Model 4, the complete model inclusive of the control variables.
For the fundamental factors, the coefficients on STE1, DIFFSTE, and LTG are 0.986, 4.060, and 0.411,
each significant at better than the 0.001 level. Thus, as expected within a valuation framework, the
target price forecast is increasing in forecasts of both short-terms earnings and long-term earnings
growth. Equally, the coefficients on the behavioral factors 52WH, SENT, and ROUND of 0.035, 0.038,
and 0.029, respectively, are also each significant at better than the 0.001 level. Thus, incremental to any
18
Similar inference obtain when the three models are run with the control variables. Here, the adjusted R2‟s are 22.8% and 16.9% for the
analogues of Models 1 and 2, and the Vuong tests now comparing Models 1, 2, and 4 are again all significant at the 1% level.
23
indirect effect they might have through the fundamentals, each of the identified behavioral factors also
appears to play a direct role in the target price formation process, with the forecast increasing in the 52week high (52WH) and in the level of recent investor sentiment (SENT). Additionally, the target price
forecast is higher when it is presented as a round number. The magnitude of regression coefficients
indicates that an increase in 52WH and SENT of one standard deviation corresponds to an increase in
the target price (scaled by price) of 4.0% (0.035*1.151) and 2.3% (0.038*0.618), respectively, while an
increase of ROUND from zero to one corresponds to an increase in the target price of 2.9% (0.029*1).
Thus, the identified behavioral factors explain an economically meaningful proportion of the overall
target price variation (from Table 3, the standard deviation of TP is 27.6%).
5.2.2 Additional Considerations
First, to control for unobserved analysts incentives and characteristics, we add the one-year lagged
value of TP (LagTP) for the same analyst and firm to the model. The results are presented as Model 5
in Panel B of Table 5. As revealed, the lagged measure (LagTP) is highly significant (0.406, p < 0.001)
and its inclusion results in a significant increase in model adjusted R2 (0.370 relative to 0.251 Model 4
in Panel A). Importantly, the results for both the fundamental and behavioral factors are consistent with
those for the primary analysis reported in Panel A, with all coefficients remaining significant at the
0.001 level. Thus, results and conclusion are robust to the inclusion of the lagged measure of the
dependent variable as a control within our econometric model.
Second, as an alternative approach to address issues surrounding model specification, we add the
two pseudo target prices constructed by Gleason et al. (2013) to the model, one based on the residual
income model (RIM) (VRIP) and one based on PEG ratio heuristic (VPEGP). Here, we find that the
inclusion of these two measures does not alter our base conclusions and adds relatively little to the
model‟s explanatory power. Specifically, the model adjusted R2 increases only marginally with the
inclusion of the pseudo target price forecasts (to 0.256 from 0.251 in the comparable Model 4 of Panel
24
A). Further, with the exception of the coefficient on ROUND (which now has a p-value of 0.002), each
of the fundamental and behavioral factors remains significant in the predicted direction at better than
the 0.001 level. Thus, results and conclusions are also robust to the inclusion the two pseudo target
prices constructed by Gleason et al. within our econometric model.19
5.2.3 Interactive Model Results
Panel C of Table 5 presents results for a series of five interactive models designed to provide
insights into the settings where the behavioral factors are most important in the development of the
target price forecast. Here, the first two sets of columns present results for the models where the
fundamental and behavioral factors are interacted with firm-related characteristics (size and earnings
volatility), the third and fourth sets of columns present results where the factors are interacted with
analyst and brokerage characteristics (analyst experience and brokerage size), and the fifth set of
columns where they are interacted with the VRM ratio designed to capture the relative use by analysts
of the RIM valuation model versus the less rigorous PEG valuation heuristic.
The results are largely (although not universally) consistent with expectations. To illustrate,
consider first the results for the model interacted with firm size presented in the first set of columns.
Here, the coefficients on the STE1 and LTG interaction terms are 0.770 and 0.091, respectively, while
the coefficients on the 52WH and SENT interaction terms are -0.032 and -0.017. In each instance, the
coefficient is statistically significant at better than the 0.01 level. Thus, on balance, for larger firms
which are arguably easier for analysts to value given their better information environments, analysts
place greater weight on the fundamental factors and reduced weight on the behavioral factors.
A similar conclusion follows from the results for the model interacted with analyst experience
presented in the third set of columns. Here, the coefficients on STE1 and LTG are positive and
19
When we alternatively include the two pseudo target prices in the model in lieu of the fundamental factors, our conclusions regarding a
role for the behavioral factors is unaltered. The coefficients on each of these factors remains significant at better than the 0.001 level.
Additionally, the coefficients on each of the two pseudo target prices is also significant at better than the 0.001 level.
25
significant while the coefficients on 52WH and ROUND are negative and significant. Thus, again
consistent with expectations, on balance analysts with greater experience and thereby who are better
positioned to make a target price forecast, place greater weight on the fundamental factors and reduced
weight on the behavioral factors. As a final illustration, for the model interacted with VMR presented
in the fifth column, the coefficient on LTG is negative and significant (-0.355, p = 0.005) while the
coefficients on 52WH (0.020, p = 0.021) and ROUND (0.034, p = 0.001) are positive and significant.
Thus, when analysts are more likely to use valuation heuristics rather than a rigorous valuation model,
they rely more on the behavioral factors 52WH and ROUND.
In sum and to re-iterate, the results from these analyses are largely supportive of the view that
analysts place greater weight on the behavioral factors (and reduced weight on the fundamental factors)
in settings with greater task complexity and/or resource constraints, and when they use less rigorous
valuation methodology to develop their target price forecasts.
5.3 Target Price Forecast Error Model Results
Table 6 presents results for the target price error model (equation (2)). Here, we examine whether
analysts‟ ex ante forecasts of fundamentals and reliance on behavioral anchors help to explain the
optimistic bias in target price forecasts documented in Table 3. The format of the presentation parallels
that used for the target price forecast results discussed in the previous section. Specifically, Panel A
presents results for the same four variants of the base model, Panel B extends the model to include
alternatively one-year lagged target price error or the two pseudo target prices developed by Gleason et
al. (2013) (VRIP and VPEG), and Panel C presents results for the same five interactive models.
5.3.1 Primary Model Results
To begin, the results for the primary models presented in Panel A reveal the adjusted R2s to be
5.4% for the model with only the fundamental factors (Model 1), 11.0% for the model with only the
26
behavioral factors (Model 2), and 14.4% for the composite model (Model 3). Here again, the test
statistics for the Vuong tests (not reported) indicate that the null hypothesis of no difference in adjusted
R2s is rejected at the 1% level for all pairwise comparisons.20 These results reveal two points of note.
First, the behavioral factors play a stronger role in explaining target price forecast error than do the
fundamental factors. Second, there is a role for both sets of factors, given that the composite model
exhibits a statistically significant increase in explanatory power over either separate factor model.
Turning to the individual factors, again given the similarity of results across models, we focus on
the complete model, Model 4. For the fundamental factors, the coefficient on STE is insignificant
(0.246, p = 0.353) while in contrast, the coefficients on DIFFSTE (-2.274, p < 0.001) and LTG (-0.461,
p < 0.001) are both negative and significant. These findings are consistent with analysts overweighting
longer term forecast measures factors when forecasting target prices and/or with the documented
analyst optimism in longer-term forecasts (Frankel and Lee, 1998; Hughes et al., 2008; Gode and
Mohanram, 2012), each of which would result in a more optimistic target price forecast (recall, more
negative values of TPerror indicate greater optimism). For the behavioral factors, as expected all
coefficients are negative and statistically significant. The coefficients on 52WH, SENT and ROUND are
-0.042, -0.238, and -0.137, respectively, each significant at better than the 0.001 level. Thus, the results
indicate that there is indeed a larger optimistic bias in the target price forecast when the 52-week high
price is higher, when recent market sentiment is more positive, and when the target price forecast is
presented as a round number. Again, this effect is incremental to any effect that the behavioral factors
might have the forecasts of fundamentals. The magnitude of regression coefficients indicates that an
increase in 52WH and SENT of one standard deviation corresponds to an increase in the target price
optimistic bias (scaled by price) of 4.8% (0.042*1.151) and 14.7% (0.238*0.618), respectively, while
an increase of ROUND from zero to one corresponds to an increase in the target price optimistic bias of
20
The same inferences pertain when the models are run the control variables. The adjusted R2‟s are 8.2% and 13.6% for the analogues of
Models 1 and 2, respectively, with the Vuong tests among Models 1, 2, and 4 again significant at the 1% level for all comparisons.
27
13.7% (0.137*1). Hence, the identified behavioral factors explain an economically significant
proportion of the overall variation in target price errors and the average optimistic bias (as Table 3
reveals, the standard deviation of TPerror is 56.6% and the mean value of TPerror is -15.6%).
5.3.2 Additional Considerations
First, as revealed in the first column of Panel B in Table 6, when the target price error model is
extended to include the one-year lagged measure of TPerror, results for both the fundamental and
behavioral factors are entirely consistent with those reported in Panel A for the primary analysis. For
the fundamental factors, the coefficient on STE1 remains insignificant while the coefficients on
DIFFSTE and LTG remain negative and highly significant. Equally, for the behavioral factors, the
coefficient on each remains negative and highly significant. Additionally, the coefficient on the lagged
TPerror measure is negative and highly significant, and the model adjusted R2 increases modestly to
0.181 from 0.151 reported for Model 4 in Panel A. Thus again, results and conclusions are robust to the
inclusion of the lagged measure of the dependent variable.
Second, the second column of Panel B reveals that results and conclusions are also robust to the
addition of the two pseudo target prices to the model. With the addition of VPEGP and VRIP, the
model adjusted R2 increases only slightly to 0.152 (from 0.151). As with the primary analysis reported
in Model 4 of Panel A, the coefficients on each of DIFFSTE and LTG remain negative and highly
significant, as do the coefficients on each of the three behavioral factors. %).
5.3.3 Interactive Model Results
Panel C of Table 6 presents results for the series of the five interactive target price error models.
Here, the results essentially parallel those for the interactive target price models (presented in Panel C
of Table 5) and as such, are largely (but again not universally) consistent with expectations. For
example, the coefficients on 52WH and ROUND are positive and significant within each of the models
28
interacted with firm size, analyst experience, and brokerage size. Thus, for firms which are larger and
hence arguably easier to value, and for analysts arguably better positioned to make a target price
forecast (those with more experience analysts and those from larger brokerage firms), these factors
contribute less to the optimistic bias in target price forecasts (consistent with analysts having placed
reduced weight on these two behavioral factors in making their forecasts as documented in Table 5).
Continuing, for the model interacted with earnings volatility, the coefficient on SENT is negative
and highly significant, consistent with analysts placing greater weight on recent market sentiment for
firms with greater earnings volatility and thereby arguably harder to value. Within this model, the
coefficients on 52WH and ROUND are also negative but insignificant. Finally, for the model interacted
with the valuation model ratio (VMR) presented in the fifth column, the coefficients on 52WH and
ROUND are negative and highly significant, indicative that with the use of valuation heuristics as
opposed to rigorous valuation methodology, the increased emphasis on the 52-week high price and the
greater tendency to round upward leads to greater optimism in the target price forecast.
In sum, following on from the results presented in Panel C of Table 5 which reveal that analysts
place greater weight on the behavioral factors (and reduced weight on the fundamental factors) in
settings with greater task complexity and/or resource constraints, and when they use less rigorous
valuation methodology to develop their target price forecasts, the results for the interactive target price
error models indicate that this increased emphasis on the behavioral factors maps into increased
optimism in the forecasts and hence greater error.
5.4 Returns Analysis Results
As the final step, we consider the usefulness of target price forecasts by examining their
association with subsequent stock returns, measured as the ex-dividend stock return over the 250trading-day period following the target price announcement day. Of perhaps equal or greater interest, in
conjunction we also examine whether the influence of two of our anchors, the 52-week high and
29
investor sentiment, reduces their usefulness. The intuition underlying this latter aspect can be illustrated
using the 52-week high: when the 52-week high price is high relative to the current stock price, we
expect analysts‟ target prices to be more heavily influenced by the 52-week high and hence, to be a
poorer predictor of future stock return; in contrast, when the 52-week high price is close to the current
price, we expect a stronger association between target prices and future returns. A similar story pertains
relating to the level of recent market sentiment.
The results for this final step in our investigation into target price forecasts are presented in Table
7. In terms of our primary focus, Model 1 includes only TP, Models 2, 3, and 4 add the interaction
terms between TP and each of the three behavioral factors (52WH, SENT, and ROUND) individually,
and finally Model 5 includes TP and all three interaction terms. All models include the sets of
fundamental and behavioral factors, as well as the set of control variables. Here, across all models, as
might be expected within a valuation framework, the coefficients on the fundamental factors, STE1 and
DIFFSTE, are uniformly positive and significant at better than the 0.001 level, a finding consistent with
the notion that short-term earnings forecasts provide explanatory power for future returns. For the
behavioral factors, the coefficient on 52WH is insignificant across all models, the coefficient on SENT
is consistently negative and significant at better than the 0.001 level, and the coefficient on ROUND is
uniformly negative and typically significant. The finding for SENT is consistent with recent market
sentiment being negatively associated with future stock returns (Baker and Wurgler, 2007). Finally,
model adjusted R2s range from 7.0% to 7.7%.
Turning directly to evidence regarding the usefulness of target price forecasts for predicting future
returns, across all models the coefficient on TP is positive and significant at better than the 0.001 level.
Thus, in a general sense, it appears that target price forecasts can be used as a reference point for
developing profitable trading strategies. For example, in Model 5 (the complete model), its coefficient
30
is 0.194 (p < 0.001). This coefficient indicates that an increase in TP of one standard deviation
corresponds to an increase in the future one-year ex-dividend return of 5.4% (0.194*0.276).
Importantly, however, the results for the interaction terms also indicate that, as conjectured, the
strength of the association is conditional on the levels of the 52-week high and recent market sentiment,
with the coefficients on the interaction terms with 52WH and SENT each negative and significant. For
example, in Model 5, the coefficient on TP*52WH is -0.041 (p = 0.002) and the coefficient on
TP*SENT is -0.086 (p = 0.038). Thus, the target price forecast, TP, appears to be most informative in
terms of future returns when 52WH and SENT are relatively low, and its informativeness decreases as
these measures increase. Here, intuitively, these findings fit well with the findings reported in Tables 5
and 6 that higher values of both 52WH and SENT map into higher target price forecasts and ultimately,
into larger optimistic bias in the forecast.
Thus, overall, the results for this part of our investigation suggest that the target price forecast is
incrementally informative for future stock returns over short-term earnings forecasts, but more so when
the influence of the behavioral factors, the 52-week high and recent market sentiment, on the target
price formation process is relatively more muted. When the 52-week high stock price is high relative to
the current price and/or when recent market sentiment is relatively more positive, both instances where
early results presented above suggest that these behavioral factors play a greater role in the
development of the target price forecast and ultimately manifest in greater optimistic forecast bias, we
find the informativeness of the target price forecast for future returns to be significantly diminished.
5.5 Additional Considerations
This section reports results from a series of tests designed to reinforce arguments and conclusions
based on our primary analyses reported in the previous sections. First, arguments supportive of a role
for the 52-high in the target price formation process imply that its influence should increase with
proximity to the target price forecast date since the more recent it is, the more likely the analyst will be
31
able to bring it to mind in determining the target price. To consider the effect of proximity, we interact
all measures within both the target price and target price errors models with proximity (PROX_52WH),
measured as minus one times the number of days between the occurrence of the 52-week high price
and the target price forecast date, scaled by 365. The results, reported in Panel A of Table 8, indicate
that the weight placed on the 52WH variable increases with proximity in both models. From the first set
of columns for the target price model, the coefficient on 52WH*PROX_52WH is 0.047 (p = 0.014)
while from the second set of columns for the target price error model, the coefficients on
52WH*PROX_52WH is -0.062 (p = 0.019). Of note, in conjunction with the relatively greater weight
on the 52WH measure, the weights placed on the other two behavioral measures, SENT and ROUND,
are significantly reduced with increased proximity, as is the weight placed on the long-term growth
measure (LTG). Thus, as suggested, with increased proximity, analysts appear to rely more heavily on
the 52-week high price in developing their target price forecast but in an apparent tradeoff, rely less on
recent market sentiment and are less inclined towards rounding.
Second, we investigate whether the association between target price and the fundamental factors is
affected by the level of either 52WH or SENT. For example, larger values of either could lead analysts
to reduce the weight they place on the fundamentals. The results are reported in Panel B of Table 8. For
the target price model, the weight placed on STE1 is reduced when either 52WH or SENT is high while
the weight placed on longer-term forecasts is increased. One possible interpretation is that analysts
increase the weight on longer-term forecasts in order to “support” the higher target price forecast that
derives with higher values of 52WH or SENT. Interestingly, for the 52WH interaction model, the
coefficient on SENT*52WH is insignificant whereas for the SENT interaction model, the coefficient on
52WH*SENT is negative and significant, suggestive that 52WH becomes less important when SENT is
high but alternatively, the role of SENT is unaffected in the face of a high 52WH. The interaction
coefficient on ROUND is positive and significant, consistent with an increased tendency to round target
32
price estimates upward when 52WH and SENT are high. Finally, for the target price error model, on
balance the results follow logically from those for the target price model with the effect of STE1 on
target price optimism reduced and the effect of LTG increased when 52WH and SENT are high.
Third, to reinforce arguments and conclusions that the behavioral factors incrementally and
directly affect target price forecasts rather than just through the valuation inputs, we ran regressions
with each of the valuation inputs as the dependent variable (STE1, DIFFSTE, and LTG) and then each
of the pseudo prices deflated by the closing price on the trading day before the target price
announcement date as the dependent variable (VRIP/P and VPEGP/P). Each model included the three
behavioral factors and the set of nine control variables. The results (not tabulated) reveal that none of
the behavioral factors has a consistently positive effect on either the fundamental forecasts, with the
effects often negative or insignificant. For the pseudo price models, only SENT has a positive
association with VRIP/P or VPEGP/P. Opposite expectations, both 52WH and ROUND are negatively
associated with VPEGP/P which, based on the prior research, is the most likely valuation model used
by analysts. Thus, overall, these results reaffirm our conclusion that the behavioral factors directly
affect target price forecasts rather than acting only indirectly through the fundamentals.
Finally, to examine whether our results and conclusions are robust to several of our design
choices, we undertake the following additional sensitivity analysis. First, to mitigate the potential
influence of outliers, we estimate both the target price and target price error models using quintile ranks
of all continuous independent variables. Our results (untabulated) remain qualitatively and statistically
the same. Second, the returns regressions have been run with LTG excluded from the model. This
allows a substantial increase in the sample size and the power of our statistical analysis. When we
alternatively include LTG in the returns model, we find its coefficient to be statistically insignificant.
More importantly, its inclusion does not alter the central results and conclusions drawn from the returns
analyses. Third, we consider an alternative measure of experience (EXP) based on the number of years
33
for which the analyst‟s forecasts for any firm have been posted on I/B/E/S. Here also the results are
qualitatively unaffected.
Next, in addition to measuring target price error as the difference between the one-year-ahead price
and the target price scaled by the current stock price, we examine the robustness of the results by
replacing the one-year-ahead price with (i) the maximum stock price achieved over the next year and
(ii) the average stock price over the next year. The results based on these alternative measures of target
price errors are qualitatively very similar. Prior literature also uses {0,1} indicator variables to measure
target price errors (Bradshaw et al., 2012). These indicator measures have two disadvantages. First,
they reward conservative estimates rather than accurate estimates. Very low target prices are likely to
have zero errors while target prices that are close to but slightly exceed the future stock price have an
error of one. Second, they are crude measures since they do not take into account the magnitude of
error, which is central to our research question. Consistent with these disadvantages, not all of the
results are significant when we use these measures: the results for the investor sentiment and target
price rounding remain significant, while the results for the 52-week high becomes insignificant.
Lastly, we run the analysis using the 52-week low as a reference point. While we do not expect
analysts to use it as a reference point since prior research indicates that analyst forecasts tend to be
positively biased, we consider it because it is commonly mentioned in the media. The untabulated
results show that this price point is not associated with target price.
6. CONCLUSION
In this study, we seek insights into the target price forecast derivation process, focusing
specifically on level, error, and return predictability. Using a valuation model framework, we expect
analysts‟ target price forecasts to impound their forecasts of fundamentals, specifically short-term
earnings forecasts and long-term growth forecasts. However, we argue that target price forecasts are
34
equally affected by information based on behavioral factors. The specific behavioral factors we
consider, drawn from the prior literature, are the level of the 52-week high stock price, recent market
sentiment, and rounding.
In support of a role for behavioral anchors, we argue that by its nature, a target price forecast is not
intended to be an accurate estimate of the fundamental value. Rather, it is intended to provide a sense
of where analysts believe that the price is likely to go over the next year. Thus, if analysts are not
convinced that the stock price will reflect the fundamental value over the short term because of
exogenous factors, they are likely to adjust the forecast appropriately in light of the identified
behavioral factors. In conjunction, to the extent that analysts can benefit from reducing additional time
and effort associated with rigorous target price forecasting, they may also defer to the use of the readily
available signals contained within these factors.
Our empirical results indicate that analysts‟ forecasts of short-term earnings and long-term growth
explain target price forecasts as expected. We also find that our behavioral factors are important
determinants of target price forecasts, with higher 52-week high prices, more positive recent market
sentiment, and rounding each being associated with higher target price forecasts. In terms of relative
contribution, perhaps not unexpectedly we find that the explanatory power of a model incorporating
just fundamentals is greater than that of a model incorporating just the behavioral factors, but
importantly that each set of factors plays an incremental role. Additionally we find, as expected, that
the behavioral factors play an enhanced role when the underlying firm is more difficult to value, when
the analyst is less well positioned to make the forecast, and when they are more reliant on valuation
heuristics rather than rigorous valuation methodology.
Turning to target price errors, we find in a parallel fashion that there is a larger positive bias when
forecasts of short-term and long-term growth are high. Importantly, we also find that high values of the
52-week high (relative to the current stock price), more positive recent market sentiment, and the use of
35
rounding are each associated with a larger optimistic bias in target prices. Of interest and in contrast
with the findings for the target price level, the model incorporating only the behavioral factors explains
more of the target price bias than a model containing only the fundamentals. In conjunction, paralleling
the target price forecast results, we find that the roles of the behavioral factors in target price forecast
error is greater when they place a greater role in the target price formation process (i.e., when the
underlying firm is more difficult to value, when the analyst is less well positioned to make the forecast,
Finally, we find an association between target prices and future stock returns, but that the strength
of this association is significantly reduced when the role of the behavioral factors in the target price
formation process is enhanced.
In sum, taken together, our findings suggest that analysts not only use forecasts of firm
fundamentals but appeal to the recent high share price, recent market sentiment, and round numbers
when determining their target price forecasts. They also indicate that reliance on these factors explains
some of the bias observed in target prices and that when the influence of these factors is reduced, target
prices are more useful predictors of future returns. Overall, we find evidence that both fundamental and
behavioral factors play an important role in the target price formation process.
36
REFERENCES
Asquith, P., M. Mikhail, and A. Au. 2005. Information content of equity analyst reports. Journal of
Financial Economics 75 (2): 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.
Available at SSRN: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1430617.
Baker, M., and J. Wurgler. 2006. Investor sentiment and the cross-section of stock returns. Journal of
Finance LXI (4): 1645-1680.
Baker, M., and J. Wurgler. 2007. Investor sentiment and the stock market. Journal of Economic
Perspectives 21 (2): 129-151.
Baker, M., X. Pan, and J. Wurgler. 2009. A reference point theory of mergers and acquisitions.
Working paper, Harvard University. Available at SSRN: http://ssrn.com/abstract=1364152.
Ball, C., W. Torous, and A. Tschoegl. 1985. The degree of price resolution: The case of the gold
market. Journal of Futures Markets 5: 29-43.
Bandyopadhyay, S., L. Brown, and G. Richardson. 1995. Analysts' use of earnings forecasts in
predicting stock returns: Forecast horizon effects. International Journal of Forecasting 11 (3):
429-445.
Bonini, S., L. Zanetti, R. Bianchini, and A. Salvi. 2010. Target price accuracy in equity research,
Journal of Business Finance and Accounting 37 (9-10): 1177-1217.
Bradshaw, M. 2002. The use of target prices to justify sell-side analysts' stock recommendations.
Accounting Horizons 16 (1): 27-41.
Bradshaw, M. 2010. Analysts‟ forecasts: What do we know after decades of work? Working paper,
Boston
College.
Available
at
SSRN:
http://papers.ssrn.com/sol3/papers.cfm?
abstract_id=1880339.
Bradshaw, M., L. Brown, and K. Huang. 2012. Do sell-side analysts exhibit differential target price
forecasting ability? forthcoming, Review of Accounting Studies.
Brav, A., and R. Lehavy. 2003. An empirical analysis of analysts' target prices: Short-term
informativeness and long-term dynamics. Journal of Finance 58 (5): 1933-1968.
Burgstahler, D., and I. Dichev. 1997. Earnings management to avoid earnings decreases and losses.
Journal of Accounting and Economics 24: 99-126.
Brown, L. 1993. Earnings forecasting research: Its implications for capital markets research.
International Journal of Forecasting 9(3): 295–320.
Carter, R. and S. Manaster. 1990. Initial public offerings and underwriter reputation. Journal of
Finance XLV(4): 1045-1067.
Christie, W., and P. Schultz. 1994. Why do NASDAQ market makers avoid odd-eighth quotes?
Journal of Finance, Vol. 49: 1813–1840.
Dechow, P., and H. You. 2012. Analysts' motives for rounding EPS forecasts. The Accounting Review
in-Press: http://dx.doi.org/10.2308/accr-50226.
37
Dechow, P., A. Hutton, and R. Sloan. 2000. The relation between analysts' long-term earnings forecasts
and stock price performance following equity offerings. Contemporary Accounting Research 17
(1): 1-32.
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(3):
567-606.
Frankel, R., and C. Lee. 1998. Accounting valuation, market expectation, and cross-sectional stock
returns. Journal of Accounting and Economics 25: 283–319.
George, T., and C. Hwang. 2004. The 52-week high and momentum investing. Journal of Finance 59:
2145-2176.
Gleason, C., W., Johnson, and H. Li, 2013. Valuation model use and the price target performance of
sell-side equity analysts. Contemporary Accounting Research, 30(1): 80-115.
Gode, D., and P. Mohanram. 2012. Improving the relationship between implied cost of capital and
realized returns by removing predictable analyst forecast errors. Forthcoming, Review of
Accounting Studies.
Grossman, S., M. Miller, K. Cone, D. Fischel, D. Ross. 1997. Clustering and competition in asset
markets. Journal of Law and Economics, Vol. 40, 1997, pp. 23–60.
Gwilym, O., A. Clare and S. Thomas. 1998. Extreme price clustering in the London equity index
futures and options markets. Journal of Banking and Finance 22: 1193-1206.
Harris, L. 1991. Stock price clustering and discreteness. Review of Financial Studies 4 (3): 389-415.
Hayes, R. 1998. The impact of trading commission incentives on analysts' stock coverage decisions and
earnings forecasts. Journal of Accounting Research 36: 299-320.
Heath, C., S. Huddart, and M. Lang. 1999. Psychological factors and stock option exercise. The
Quarterly Journal of Economics 114 (2): 601-627.
Hornick, C., and D. Zakay. 1994. The influence of prototypic values on the validity of studies using
time estimates. Journal of Market Research Society 36: 145-147.
Hribar, P., and J. McInnis. 2012. Investor sentiment and analysts' earnings forecast errors, Management
Science (Special Issue on Behavioral Economics and Finance) 58 (2): 293-307.
Hughes, J., J. Liu, and W. Su. 2008. On the relation between predictable market returns and predictable
analyst forecast errors. Review of Accounting Studies, 13, 266–291.
Ikenberry, D., and J. Weston. 2007. Clustering in US stock prices after decimalization. European
Financial Management 14 (1): 30-54.
Jung, B, P. Shane, and Y. Yang. 2012. Do financial analysts' long-term growth forecasts matter?
Evidence from stock recommendations and career outcomes. Journal of Accounting and
Economics 53(1-2): 55-76.
Ke, B., and Y. Yu. 2009. Why don't analysts use their earnings forecasts in generating stock
recommendations? Working paper, Nanyang Technological University. Available at SSRN:
http://ssrn.com/abstracts=10111449.
Kerl, A., 2011. Target price accuracy. BuR - Business Research 1(1): 74-96.
38
Li J., and J. Yu. 2012. Investor attention, psychological anchors, and stock return predictability.
Journal of Financial Economics 104: 401-419.
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 (2): 455-483.
Loomes, G. 1998. Different experimental procedures for obtaining valuations of risky actions:
Implications for utility theory. Theory and Decisions 25: 1-23.
Loughran, T., and J. Ritter. 2004. Why has IPO underpricing changed over time? Financial
Management 33 (2): 5-37.
Lui, D., S. Markov, and A. Tamayo. 2007. What makes a stock risky? Evidence from sell-side analysts'
risk ratings. Journal of Accounting Research, 45 (3): 629-665.
O'Brien, P., M. McNichols, and H. Lin. 2005. Analyst impartiality and investment banking
relationships. Journal of Accounting Research 43 (4): 623-650.
Park, C., and E. Stice. 2000. Analyst forecasting ability and the stock price reaction to forecast
revisions. Review of Accounting Studies 5: 259-272.
Petersen, M. 2009. Estimating standard errors in finance panel data sets: Comparing approaches.
Review of Financial Studies 22 (1): 435-480.
Shepard, R., D. Kilpatric, and J. Cunningham. 1975. Internal representation of numbers. Cognitive
Psychology 7: 82-138.
Shiller, R. 2005. Irrational Exuberance. Second Edition (First Edition, 2000), Broadway Books, United
States.
Simon, A., and Curtis, A. 2011. The use of earnings forecasts in stock recommendations: Are accurate
analysts more consistent? Journal of Business Finance and Accounting 38 (1-2): 119-144.
Stickel, S. 1990. Predicting individual analyst earnings forecasts. Journal of Accounting Research 28:
409-417.
Tversky, A., and D. Kahneman. 1974. Judgment under uncertainty: Heuristics and biases. Science, 185
(4157): 1124-1131.
Westerhoff, F. 2003. Anchoring and psychological barriers in foreign exchange markets. Journal of
Behavioral Finance 4: 65-70.
Zuckerman, R. 2009. Investor overconfidence, turnover, volatility and the disposition effect: A study
based on price target updates. Working paper, Rutgers University. Available at SSRN:
http://ssrn.com/abstract=1514964.
39
Figure 1
Histogram of the Difference between Target Price and 52-Week High, TP-52WH
Panel A: Histogram of TP-52WH scaled by the stock price, 0.01 intervals
Frequency
8000
H0: p<0.001
a
6000
4000
2000
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
-0.05
-0.1
-0.15
-0.2
-0.25
-0.3
-0.35
-0.4
-0.45
-0.5
0
Difference between TP and 52WH
Panel B: Histogram of TP-52WH scaled by the 52-week high price, 0.01 intervals
Frequency
a
H0: p<0.001
8000
6000
4000
2000
0.35
0.4
0.45
0.5
0.35
0.4
0.45
0.5
0.3
0.25
0.2
0.15
0.1
0.05
0
-0.05
-0.1
-0.15
-0.2
-0.25
-0.3
-0.35
-0.4
-0.45
-0.5
0
Difference between TP and 52WH
0.3
0.25
a
0.15
0.1
0.05
0
-0.05
-0.1
-0.15
-0.2
-0.25
-0.3
-0.35
-0.4
-0.45
H0: p<0.001
0.2
1200
1000
800
600
400
200
0
-0.5
Frequency
Panel C: Histogram of TP-52WH scaled by the stock price, 0.001 intervals
Difference between TP and 52WH
a
p-value relates to the two-tail binomial test of the null hypothesis that the number of observations in the interval that
includes zero is the average of the number of the observations in the two immediately adjacent intervals.
The figure plots the frequency of the difference between the target price and the 52-week high stock price. The difference is
scaled by the closing stock price on the trading day before issuance of the target price (Panels A and C) and by the 52-week
high stock price (Panel B). The size of each interval is 0.01 (Panels A and B) and 0.001 (Panel C). The vertical axis
corresponds to the number of observations in each interval.
40
Table 1
Signed Predictions for Variables of Interest in the Three Econometric Models
=======================================================================
Variable
Variable description
Equation (1)
Equation (2)
Equation (3)
TP
TPerror
RETt+1
N/A
N/A
+
TP
Target price scaled by current price
STE1
Analyst forecast of one-year-ahead
earnings
+
?
+
DIFFSTE
Difference between two-year-ahead
and one-year-ahead earnings forecast
+
?
+
LTG
Long-term growth forecast
+
?
N/A
52WH
Past 52-week high price
+
–
?
SENT
Monthly investor sentiment
+
–
?
ROUND
Indicator of rounded target price
+
–
?
This table provides a summary of the independent variables of interest for our regression models along with their description and
predicted signs.
Table 2
Sample Selection Criteria
Starting sample of target price forecasts:
Twelve-month forecast horizon, US firm, announcement date is known,
target price is not null, issued between 1999 and 2007
525,617
Less: Observations with insufficient data on CRSP or COMPUSTAT or stock
prices less than one dollar
(40,653)
Less: Observations without one-year-ahead forecasts on the target price date
(232,763)
Less: Observations without two-year-ahead forecasts on the target price date
(55,813)
Less: Observations without long-term growth forecasts on the target price date
Final sample (includes 3,518 analysts and 4,148 firms)
(169,642)
26,746
This table reports the reduction in the sample size due to our data requirements. The target prices, one-year-ahead, two-year-ahead, and
long-term earnings growth forecasts are from I/B/E/S, as of June, 1 2009. The sample period is January 1997 – December 2007.
41
Table 3
Descriptive Statistics
Variable
Mean
P10
P25
Median
p75
p90
StdDev
TP
1.228
0.987
1.086
1.185
1.304
1.498
0.276
TPerror
-0.156
-0.810
-0.451
-0.139
0.130
0.442
0.566
RETt+1
0.076
-0.485
-0.215
0.035
0.287
0.617
0.516
STE1
0.046
0.012
0.030
0.048
0.066
0.086
0.046
DIFFSTE
0.013
0.003
0.006
0.009
0.014
0.026
0.020
LTG
0.192
0.080
0.118
0.150
0.240
0.350
0.124
Dependent Variables
Main Variables
52WH
1.517
1.008
1.052
1.185
1.543
2.212
1.151
PROX_52WH
-0.414
-0. 942
-0. 748
-0.348
-0. 068
-0. 011
0.348
SENT
0.086
-0.502
-0.292
-0.060
0.306
1.028
0.618
ROUND
0.874
0
1
1
1
1
0.332
REP
0.088
0
0
0
0
0
0.284
EXP
2.097
1
1
1
3
4
1.554
B_SIZE
62.30
10
20
48
100
126
50.42
CONFLICT
0.414
1.232
0.000
0.269
0.000
0.609
0.500
1
0.500
1.712
1.000
2.644
0.337
0.832
10,637.6
224.1
571.1
1,805.5
7,121.9
22,847.5
29,592.7
Control Variables
β
SIZE ($m)
BM
0.379
0.091
0.186
0.327
0.512
0.734
0.286
RETt-1
RETstd
0.287
0.028
-0.346
0.013
-0.103
0.018
0.144
0.025
0.456
0.034
0.973
0.048
0.766
0.015
This table reports the mean, median, standard deviation (StdDev), 10th percentile (P10), 25th percentile (P25), 75th percentile (P75), and
90th percentile (P90) for all variables based on a sample of 26,746 target price forecasts for U.S. firms over the period 1999 – 2007.
Variable definitions: TP is the analyst target price forecast scaled by the closing price on the trading day before the target price
announcement date; TPerror is the target price error measured as the difference between the one-year-ahead stock price and the target price
forecast, each scaled by the closing price on the trading day before the target price announcement date; RETt+1 is the ex-dividend stock return
measured over the 250-trading-day period following the target price announcement day; STE1 is the analyst forecast of one-year-ahead
earnings scaled by the closing price on the trading day before the target price announcement date; DIFFSTE is the difference between twoyear-ahead and one-year-ahead earnings forecasts scaled by the closing price on the trading day before the target price announcement date.
LTG is the analyst forecast of earnings growth rate over the next three to five years; 52WH is the highest stock price over the 52-week period
preceding the target price announcement date scaled by the closing price on the trading day before the target price announcement date;
PROX_52WH is the proximity of the date on which 52-week high price is experienced to the target price forecast date, measured as minus
one times the number of days between the occurrence of the 52-week high price and the target price forecast date, scaled by 365; SENT is the
monthly Baker and Wurgler (2007) investor sentiment index for the month prior to the target price announcement month; ROUND is a
categorical variable set equal to 1 if the target price forecast is rounded to the nearest dollar, and 0 otherwise; REP is analyst reputation,
measured as a categorical variable set equal to 1 if the analyst is named as an “All American” team analyst by the Institutional Investor, and
0 otherwise; EXP is analyst experience in forecasting firm i, measured as the number of years in which the analyst has issued target price
forecasts for the firm; B_SIZE is brokerage size, measured as the number of analysts associated with a particular broker in a given year;
CONFLICT is a measure of possible investment banking-related conflicts of interest that may influence analysts‟ opinions which following
Gleason et al. (2013) based on the Carter and Manaster (1990) rankings as updated by Loughran and Ritter (2004), we assign a value of 1 if
the brokerage firm has the top investment banking reputational rank, 0.5 if it has a lower reputational rank, and 0 if it does not have a CarterManaster reputational rank β is the firm‟s CAPM beta, estimated from a regression of firm returns minus the risk-free (one-month T-bill) rate
on the value-weighted market index minus the risk-free rate over a period of 60 months preceding the target price month; SIZE is firm
market capitalization (in millions of dollars) on the trading day before the target price announcement date; BM is the book-to-market ratio,
calculated as the book value of equity divided by the stock price at the end of the previous fiscal year; RETt-1 is the past one-year return
calculated as the 250-trading-day buy-and-hold return ending one day before the target price announcement date; and RETstd is the return
volatility, measured as the daily return volatility over the 250-trading-day period ending one day before the target price announcement date.
42
Table 4
Correlations among Main Variables of Interest
TP
-
TP
TPerror
-0.518
-
TPerror
STE1
DIFFSTE
***
STE1
-0.158
***
DIFFSTE
0.309
***
LTG
0.290
***
52WH
0.292
***
SENT
***
0.194
ROUND
RET_250
-0.001
-0.001
0.155***
-0.130***
-0.200***
-0.206***
-0.301***
-0.070***
-0.034***
-
-0.509***
-0.310***
-0.252***
-0.046***
0.024***
-0.073***
-
0.129***
0.186***
0.006
-0.056***
-0.082***
-
0.139***
0.114***
-0.029***
0.256***
-
-0.298***
0.016**
-0.248***
-
0.021***
-0.055***
-
-0.033***
LTG
52WH
SENT
ROUND
-
RETt+1
This table reports the correlation among the main variables of interest based on a sample of 26,746 target price forecasts for U.S. firms over the period 1999 – 2007.
Variable definitions: TP is the analyst target price forecast scaled by the closing price on the trading day before the target price announcement date; TPerror is the target
price error measured as the difference between the one-year-ahead stock price and the target price forecast, each scaled by the closing price on the trading day before the target
price announcement date; STE1 is the analyst forecast of one-year-ahead earnings scaled by the closing price on the trading day before the target price announcement date;
DIFFSTE is the difference between two-year-ahead and one-year-ahead earnings forecasts scaled by the closing price on the trading day before the target price announcement date;
LTG is the analyst forecast of earnings growth rate over the next three to five years; 52WH is the highest stock price over the 52-week period preceding the target price
announcement date scaled by the closing price on the trading day before the target price announcement date; SENT is the monthly Baker and Wurgler (2007) investor sentiment
index for the month prior to the target price announcement month; ROUND is a categorical variable set equal to 1 if the target price forecast is rounded to the nearest dollar, and 0
otherwise; and RETt+1 is the ex-dividend stock return measured over the 250-trading-day period following the target price announcement day.
*** **
, , and * denote significance at the 1%, 5%, and 10% levels, respectively.
43
Table 5
Target Price Model Results
Panel B: Additional Considerations
Panel A: Primary Model Results
Fundamentals
Behavioral
Control Variables
Adjusted R2
Model 1
Model 2
Model 3
Model 4
Model 5
Model 6
***
***
***
***
***
INTERCEPT
0.501
(< 0.001)
0.999***
(< 0.001)
STE1
1.222***
(< 0.001)
0.104
(0.652)
4.060***
(< 0.001)
DIFFSTE
3.367***
(< 0.001)
2.943***
(< 0.001)
0.561***
(< 0.001)
0.411***
(< 0.001)
LTG
0.301***
(< 0.001)
0.256***
(< 0.001)
0.062***
(< 0.001)
0.048***
(< 0.001)
0.035***
(< 0.001)
52 WH
0.026***
(< 0.001)
0.032***
(< 0.001)
0.053**
(< 0.001)
0.001
(0.905)
0.049***
(< 0.001)
0.018***
(< 0.001)
0.038***
(< 0.001)
0.029***
(< 0.001)
SENT
0.038***
(< 0.001)
0.031***
(< 0.001)
---
N
N
N
Y
0.022***
(< 0.001)
0.017***
(0.002)
0.406***
(< 0.001)
0.165
0.098
0.227
0.251
INTERCEPT
1.026
(< 0.001)
1.129
(< 0.001)
0.936
(< 0.001)
0.985
(< 0.001)
STE1
0.536***
(0.015)
---
0.778***
(< 0.001)
0.986***
(< 0.001)
DIFFSTE
4.423***
(< 0.001)
---
4.245***
(< 0.001)
LTG
0.618***
(< 0.001)
---
52 WH
---
SENT
---
ROUND
---
Fundamentals
Behavioral
ROUND
Lagged TP
LagTP
Pseudo Prices
VRIP
---
0.065***
(0.001)
VPEGP
---
0.036***
(< 0.001)
Y
Y
0.370
0.256
Control
Variables
Adjusted R2
44
Panel C: Interaction Model Results
Interacted with firm size
Variable
Primary
Coefficients
Interaction
Coefficients
Interacted with
earnings volatility
Primary
Coefficients
Interaction
Coefficients
Primary
Coefficients
Interaction
Coefficients
Primary
Coefficients
Interaction
Coefficients
Primary
Coefficients
Interaction
Coefficients
STE1
0.769***
[< 0.001]
0.770***
[< 0.001]
1.412***
[< 0.001]
-0.570**
[0.011]
0.665***
[0.001]
0.694***
(< 0.001)
0.732***
[0.002]
0.601***
[0.006]
1.017***
[< 0.001]
-0.051
[0.734]
DIFFSTE
3.728***
[< 0.001]
0.389
[0.257]
3.592***
[< 0.001]
0.599*
[0.057]
3.852***
(< 0.001)
0.480
[0.317]
4.581***
[< 0.001]
-1.389***
[< 0.001]
4.599***
[< 0.001]
0.999***
[0.005]
LTG
0.383***
[< 0.001]
0.091***
[0.009]
0.373***
[< 0.001]
0.058
[0.103]
0.334***
(< 0.001)
0.216***
(< 0.001)
0.427***
[< 0.001]
-0.049
[0.207]
0.278***
[< 0.001]
-0.355***
[< 0.001]
52WH
0.053***
[< 0.001]
-0.032***
[0.004]
0.038***
[< 0.001]
-0.006
[0.479]
0.041***
(< 0.001)
-0.014*
[0.082]
0.039***
[< 0.001]
-0.011
[0.207]
0.044***
[< 0.001]
0.020**
[0.021]
SENT
0.049***
[< 0.001]
-0.017***
[0.006]
0.032***
[< 0.001]
0.015**
[0.026]
0.029***
(< 0.001)
0.019***
[0.001]
0.031***
[< 0.001]
0.011*
[0.058]
0.037***
[< 0.001]
-0.003
[0.549]
ROUND
0.022***
[0.001]
0.017
[0.167]
0.015***
[0.009]
0.028***
[0.006]
0.040***
(< 0.001)
-0.025**
[0.047]
0.037***
[< 0.001]
-0.018*
[0.094]
0.043***
[< 0.001]
0.034***
[0.001]
Control
Variables
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
0.254
0.258
0.974***
[< 0.001]
Interacted with VMR
1.019***
[< 0.001]
0.260
1.012***
(< 0.001)
Interacted with
brokerage size
INTERCEPT
Adjusted R2
0.983***
[< 0.001]
Interacted with
analyst experience
0.261
0.989***
[< 0.001]
0.259
This table presents results based on the target price forecast model described by equation (1) based on a sample of 26,746 target price forecasts for U.S. firms over the period 1999 – 2007.
Variable definitions: TP is the analyst target price forecast scaled by the closing price on the trading day before the target price announcement date; STE1 is the analyst forecast of one-yearahead earnings scaled by the closing price on the trading day before the target price announcement date; DIFFSTE is the difference between two-year-ahead and one-year-ahead earnings forecasts
scaled by the closing price on the trading day before the target price announcement date; LTG is the analyst forecast of earnings growth rate over the next three to five years; 52WH is the highest
stock price over the 52-week period preceding the target price announcement date scaled by the closing price on the trading day before the target price announcement date; SENT is the monthly
Baker and Wurgler (2007) investor sentiment index for the month prior to the target price announcement month; and ROUND is a categorical variable set equal to 1 if the target price forecast is
rounded to the nearest dollar, and 0 otherwise.
*** **
, , and * denote significance at the 1%, 5%, and 10% levels, respectively. p-values (in parentheses) are based on a two-tailed t-test corrected for clustering of standard errors by firm
and month (Petersen, 2009).
45
Table 6
Target Price Error Model Results
Panel B: Additional Considerations
Panel A: Primary Model Results
Model 1
Fundamentals
Behavioral
Control Variables
Adjusted R2
Model 2
**
Model 3
***
Model 4
Model 5
**
Model 6
***
0.602
(< 0.001)
0.153**
(0.021)
0.117
(0.632)
-0.213
(0.589)
INTERCEPT
-0.013
(0.666)
0.059
(0.031)
0.169
(< 0.001)
0.141
(0.033)
STE1
0.773**
(0.038)
---
0.493*
(0.082)
0.246
(0.353)
DIFFSTE
-2.141***
(< 0.001)
---
-2.210***
(< 0.001)
-2.274***
(< 0.001)
DIFFSTE
-1.342***
(0.008)
-2.890***
(< 0.001)
LTG
-0.781***
(< 0.001)
---
-0.634***
(< 0.001)
-0.461***
(< 0.001)
LTG
-0.220***
(0.001)
-0.623***
(< 0.001)
52 WH
---
-0.063***
(< 0.001)
-0.043***
(0.001)
-0.042***
(< 0.001)
52WH
-0.037***
(0.001)
-0.039***
(< 0.001)
SENT
-----
-0.234***
(< 0.001)
-0.128***
(< 0.001)
-0.238***
(< 0.001)
-0.137***
(< 0.001)
SENT
ROUND
-0.239***
(< 0.001)
-0.113***
(< 0.001)
-0.238***
(< 0.001)
-0.136***
(< 0.001)
---
N
N
N
Y
-0.242***
(< 0.001)
-0.120***
(< 0.001)
-0.475***
(< 0.001)
0.054
0.110
0.144
0.151
INTERCEPT
Fundamentals
STE1
ROUND
Lagged TPE
LagTPE
Pseudo Prices
VRIP
---
-0.058
(0.193)
VPEGP
---
0.048***
(0.003)
Y
0.181
Y
0.152
Control Variables
Adjusted R
2
46
Panel C: Interaction Model Results
Interacted with firm size
Variable
Primary
Coefficients
Interaction
Coefficients
Interacted with
earnings volatility
Primary
Coefficients
Interaction
Coefficients
Interacted with
analyst experience
Primary
Coefficients
Interaction
Coefficients
Primary
Coefficients
Interaction
Coefficients
Primary
Coefficients
Interaction
Coefficients
0.191***
[0.008]
STE1
0.567**
[0.040]
-1.137***
[< 0.001]
-0.143
[0.547]
0.404
[0.164]
0.648**
[0.018]
-1.033***
[< 0.001]
0.653**
[0.036]
-0.990***
[0.001]
0.229
[0.416]
0.401*
[0.064]
DIFFSTE
-2.098***
[< 0.001]
0.255
[0.684]
-2.209***
[< 0.001]
-0.259
[0.702]
-1.985***
[< 0.001]
-0.855
[0.282]
-2.469***
[< 0.001]
0.596
[0.331]
-3.277***
[< 0.001]
-1.449**
[0.013]
LTG
-0.460***
[< 0.001]
-0.027
[0.742]
-0.304***
[< 0.001]
-0.156
[0.111]
-0.488***
[< 0.001]
0.136
[0.282]
-0.531***
[< 0.001]
0.188**
[0.014]
-0.146**
[0.026]
0.955***
[< 0.001]
52WH
-0.056***
[< 0.001]
0.026**
[0.039]
-0.015
[0.331]
-0.018
[0.210]
-0.057***
[< 0.001]
0.034**
[0.012]
-0.056***
[< 0.001]
0.038***
[0.003]
-0.055***
[< 0.001]
-0.029***
[0.009]
SENT
-0.238***
[< 0.001]
-0.006
[0.718]
-0.173***
[< 0.001]
-0.166***
[< 0.001]
-0.221***
[< 0.001]
-0.047***
[0.006]
-0.234***
[< 0.001]
0.000
[0.980]
-0.234***
[< 0.001]
0.010
[0.360]
ROUND
-0.153***
[< 0.001]
0.048**
[0.024]
-0.133***
[< 0.001]
-0.006
[0.754]
-0.156***
[< 0.001]
0.059***
[0.010]
-0.156***
[< 0.001]
0.035*
[0.081]
-0.174***
[< 0.001]
-0.096***
[< 0.001]
Control
Variables
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
0.154
0.100
[0.124]
0.163
0.156
0.170***
[0.009]
Interacted with VMR
INTERCEPT
Adjusted R2
0.194***
[0.002]
Interacted with
brokerage size
0.159
0.126**
[0.045]
0.164
This table presents results based on the target price error model described by equation (2) based on a sample of 26,746 target price forecasts for U.S. firms over the
period 1999 – 2007.
Variable definitions: TPerror is the target price error measured as the difference between the one-year-ahead stock price and the target price forecast, each scaled by the closing price
on the trading day before the target price announcement date; STE1 is the analyst forecast of one-year-ahead earnings scaled by the closing price on the trading day before the target price
announcement date; DIFFSTE is the difference between two-year-ahead and one-year-ahead earnings forecasts scaled by the closing price on the trading day before the target price announcement
date; LTG is the analyst forecast of earnings growth rate over the next three to five years; 52WH is the highest stock price over the 52-week period preceding the target price announcement date
scaled by the closing price on the trading day before the target price announcement date; SENT is the monthly Baker and Wurgler (2007) investor sentiment index for the month prior to the target
price announcement month; and ROUND is a categorical variable set equal to 1 if the target price forecast is rounded to the nearest dollar, and 0 otherwise.
*** **
, , and * denote significance at the 1%, 5%, and 10% levels, respectively. p-values (in parentheses) are based on a two-tailed t-test corrected for clustering of
standard errors by firm and month (Petersen, 2009).
47
Table 7
Relation between Target Prices and Future Returns
TP*52WH
0.051
(0.359)
0.079***
(< 0.001)
---
TP*SENT
---
0.031
(0.596)
0.120***
(< 0.001)
-0.043***
(0.002)
---
TP*ROUND
---
---
-0.088**
(0.035)
---
0.786***
(< 0.001)
1.424***
(< 0.001)
0.011
(0.432)
-0.210***
(< 0.001)
-0.079***
(< 0.001)
-0.021**
(0.018)
-0.006*
(0.055)
0.027
(0.106)
-0.022
(0.277)
0.803***
(< 0.001)
1.444***
(< 0.001)
0.018
(0.249)
-0.215***
(< 0.001)
-0.079***
(< 0.001)
-0.017**
(0.044)
-0.007**
(0.015)
0.019
(0.234)
-0.031
(0.121)
0.070
0.071
Intercept
TP
STE1
DIFFSTE
52WH
SENT
ROUND
Beta
LnSize
BM
RETt-1
Adjusted R2
0.034
(0.526)
0.136***
(< 0.001)
---
0.018
(0.757)
0.107***
(< 0.001)
---
0.767***
(< 0.001)
1.413***
(< 0.001)
0.008
(0.562)
-0.148***
(< 0.001)
-0.077***
(< 0.001)
-0.019**
(0.023)
-0.006*
(0.064)
0.024
(0.142)
-0.023
(0.251)
-0.032
(0.139)
0.785***
(< 0.001)
1.421***
(< 0.001)
0.011
(0.430)
-0.210***
(< 0.001)
-0.040
(0.139)
-0.021**
(0.018)
-0.006*
(0.056)
0.027
(0.103)
-0.022
(0.276)
-0.010
(0.868)
0.194***
(< 0.001)
-0.041***
(0.002)
-0.086**
(0.038)
-0.023
(0.289)
0.782***
(< 0.001)
1.430***
(< 0.001)
0.014
(0.329)
-0.154***
(< 0.001)
-0.049*
(0.074)
-0.015*
(0.059)
-0.007**
(0.019)
0.017
(0.287)
-0.032
(0.116)
---
0.075
0.070
0.077
This table reports results for the returns model, equation (3).
Variable definitions: TP is the analyst target price forecast scaled by the closing price on the trading day before the target price
announcement date; STE1 is the analyst forecast of one-year-ahead earnings scaled by the closing price on the trading day before the
target price announcement date; DIFFSTE is the difference between two-year-ahead and one-year-ahead earnings forecasts scaled by the
closing price on the trading day before the target price announcement date; 52WH is the highest stock price over the 52-week period
preceding the target price announcement date scaled by the closing price on the trading day before the target price announcement date;
SENT is the monthly Baker and Wurgler (2007) investor sentiment index for the month prior to the target price announcement month;
and ROUND is a categorical variable set equal to 1 if the target price forecast is rounded to the nearest dollar, and 0 otherwise.
The dependent variable in all regressions is the ex-dividend stock return over the target price forecast horizon
(RETt+1). Control variables included in this model are β, SIZE, BM, and RETt-1. These control variables are defined in
Table 3.
*** **
, , and * denote significance at the 1%, 5%, and 10% levels, respectively. p-values in parentheses are based on a
two-tailed t-test corrected for clustering of standard errors by firm and year (Petersen, 2009).
48
Table 8
Additional Considerations
Panel A: Model Conditioned on Proximity of 52-week high (PROX_52WH)
Target Price Model (TP)
Variable
Primary
Coefficients
Target Price Error Model (TPerror)
Interaction
Coefficients
Primary
Coefficients
Interaction
Coefficients
INTERCEPT
0.974***
(< 0.001)
STE1
0.864***
(< 0.001)
0.304
(0.233)
0.091
(0.786)
0.683*
(0.091)
DIFFSTE
3.956***
(< 0.001)
0.250
(0.637)
-2.182***
(< 0.001)
-0.507
(0.497)
LTG
0.490***
(< 0.001)
-0.149***
(0.004)
-0.578***
(< 0.001)
0.245**
(0.012)
52WH
0.027***
(0.004)
0.047**
(0.014)
-0.029**
(0.018)
-0.062**
(0.019)
SENT
0.045***
(< 0.001)
-0.019**
(0.011)
-0.318***
(< 0.001)
0.144***
(< 0.001)
ROUND
0.052***
(< 0.001)
-0.047***
(< 0.001)
-0.160***
(< 0.001)
0.051**
(0.039)
0.017
(0.110)
---
-0.076**
(0.042)
---
Y
Y
Y
Y
PROX_52WH
Control Variables
Adjusted R2
0.109
(0.113)
0.258
0.159
Panel B: Models Interacted with 52-week High (52WH) and Market Sentiment (SENT)
Target Price Model (TP)
Target Price Error Model (TPerror)
Interacted with 52WH
Interacted with SENT
Interacted with 52WH
Interacted with SENT
Variable
Primary
Coef
Interact
Coef
Primary
Coef
Interact
Coef
Primary
Coef
Interact
Coef
Primary
Coef
Interact
Coef
INTERCEPT
0.754***
(< 0.001)
---
0.957***
(< 0.001)
---
0.370*
(0.053)
---
0.054
(0.698)
---
STE1
1.567***
(< 0.001)
-0.222***
(0.001)
1.182***
(< 0.001)
-0.842***
(< 0.001)
-0.218
(0.635)
0.176**
(0.038)
-0.163
(0.668)
1.222***
(< 0.001)
DIFFSTE
3.963***
(< 0.001)
-0.084
(0.257)
3.951***
(< 0.001)
0.421**
(0.033)
-2.848***
(0.002)
0.129
(0.421)
-2.855***
(< 0.001)
-0.763
(0.148)
LTG
0.279***
(0.001)
0.095**
(0.019)
0.423***
(< 0.001)
-0.048
(0.380)
-0.128
(0.495)
-0.185**
(0.050)
-0.371***
(< 0.001)
-0.220***
(0.001)
52WH
0.202***
(< 0.001)
---
0.057***
(< 0.001)
-0.020***
(0.006)
-0.192***
(< 0.001)
---
-0.012
(0.793)
0.002
(0.933)
SENT
0.032***
(< 0.001)
0.004
(0.242)
0.201***
(< 0.001)
---
-0.144***
(0.003)
-0.057**
(0.019)
-0.042
(0.836)
---
ROUND
-0.009
(0.392)
0.023***
(< 0.001)
0.026***
(0.001)
0.025**
(0.023)
-0.117***
(< 0.001)
-0.009
(0.403)
-0.129***
(< 0.001)
-0.053***
(0.004)
Controls
Y
Y
Y
Y
Y
Y
Y
Y
Adjusted R2
0.280
0.266
0.171
0.172
PROX_52WH is the proximity of the date on which 52-week high price is experienced to the target price forecast date, measured
as minus one times the number of days between the occurrence of the 52-week high price and the target price forecast date, scaled by
365. All remaining variables are as defined in Tables 5 and 6.
*** **
, , and * denote significance at the 1%, 5%, and 10% levels, respectively. p-values in parentheses are based on a
two-tailed t-test corrected for clustering of standard errors by firm and year (Petersen, 2009).
49