À 0.31 3.22 À 0.76

15.54 4.42

Recommendation reiterations

0.221 À 0.110

All target price revisions 0.094 1.129 0.686 98.7%

5.34 À 3.30

0.70 38.62 18.86

0.780 À 0.152 À 0.009

Most favorable target price revisions 0.478 1.161 99.0%

17.33 À 3.08 À 0.28

3.07 33.07

Least favorable target price revisions À 0.080 0.216 À 0.311

1.232 0.690 95.4%

À 0.31 2.74 À 4.87

22.26 9.48

Recommendation downgrades

À 0.325 0.275 À 0.132

All target price revisions 1.179 0.471 93.7%

À 1.32 3.69 À 2.35

21.44 6.95

Most favorable target price revisions À 0.297 1.141 0.442 0.244 0.103 91.1%

À 0.99 17.16 5.36 2.71 1.48

Least favorable target price revisions À 0.354 0.219 À 0.538

1.332 0.614 90.7%

À 0.94 1.93 À 6.29

15.86 5.94

results reported in Table IV, we ¢nd no evidence of abnormal postevent return for

all three recommendation revision categories.

Next, within each recommendation category, we present regression results for

portfolios in which we condition on the magnitude of the target price revision at

the time of the portfolio formation. Consider ¢rst recommendation upgrades. It

can be seen that ¢rms with the highest target price revision tend to comove with

1954 The Journal of Finance

the returns of small growth ¢rms, and that the portfolio intercept, 0.799 percent,

is large both economically and statistically.When we consider ¢rms in the lowest

tercile of target price revisions, we ¢nd that the portfolio covaries with the re-

turns of small, value ¢rms but ¢nd no evidence of abnormal performance. This

evidence is consistent with the view that over our sample period small growth

¢rms exhibited strong price appreciation whereas value stocks in general per-

formed poorly.

In the case of recommendation reiterations, we still ¢nd strong evidence of ab-

normal performance for the high target price revision.The other portfolios in this

case yield insigni¢cant estimates of abnormal returns. Much like with recom-

mendation upgrades, it can be seen that the high target price revision portfolio

tends to comove with the returns of small, low book-to-market ¢rms, while the

lowest target price revision portfolio returns covary with the returns of small

value ¢rms. Finally, the evidence within recommendation downgrades is consis-

tent with the event-time analysis, with insigni¢cant intercepts for both highest

and lowest target price revisions. Interestingly, the estimated factor loadings for

¢rms with a recommendation downgrade indicate that the key di¡erence

between ¢rms with the lowest and highest target price is their exposure to the

momentum factor. While both sets of ¢rms comove with the returns of small

growth ¢rms, those that receive an upward (downwards) revision to their target

price behave like other ˜˜winner™™ (˜˜loser™™) ¢rms.19

The abnormal return evidence presented in this section is consistent both with

an irrational ˜˜underreaction™™ interpretation (e.g., Jegadeesh and Titman (1993))

and with rational learning (Brav and Heaton (2002)). We caution, however, that

our three-year sample period coincides with the highs of the bull market in the

United States, which might be viewed historically as an unusual period. There-

fore, an alternative viable view is that the evidence of postevent returns is unique

to this period and unlikely to persist as more data becomes available.We leave for

future research a detailed study of these explanations.

III. Modeling the Long-term Relation between Market and Target Prices

We extend the analysis in Section II with an investigation of the long-term dy-

namics of the stock and target prices. Our objectives are to further investigate

the extent to which analyst price targets are systematically related to ¢rms™ fun-

damental values and to quantify the low-frequency dynamics of these price series.

Because both sets of prices are nonstationary, we employ a cointegration

framework and estimate the linear combination of the price series that is station-

ary. This linear combination is termed as the price system™s ˜˜long-term relation™™

19

In additional unreported tests, we repeat the regression analysis but with value-weighted

portfolio returns. Speci¢cally, we construct our portfolios as explained above but weigh the

component monthly returns with the lagged market capitalization of the constituent ¢rms. We

¢nd that across the three recommendation classi¢cations, the pattern of positive (negative)

performance subsequent to a high (low) revision in the target price is qualitatively the same

as in Table V.

An Empirical Analysis of Analysts™ Target Prices 1955

and is parameterized as the ratio of target and market prices.20 Since target

prices are predominantly one-year-ahead prices, this ratio can be interpreted as

the analysts™estimate of the ¢rm™s ex ante return.

Cointegration also implies that any price deviations from the long-term ratio

are stationary as well.That is, if on a given date the ratio of the two prices is equal

to the long-term relation, then a shock to any of the variables will lead to a price

path that will settle back to the long-term relation.We analyze the price system™s

reaction to deviations from this long-term ratio by examining which price series

corrects to the long-term relation, once the system has been perturbed away.That

is, are analysts the ones reacting to a deviation from the long-term relation by

adjusting their target prices, or, do stock prices contribute towards most of the

adjustment?

Our empirical analysis is conducted on a time series of individual ¢rms™ weekly

stock prices and consensus target prices. Consensus target price is calculated as

the average target price across all brokerage houses. Target prices outstanding

for more than 90 days are excluded from the consensus.We choose a weekly inter-

val to avoid microstructure problems associated with daily data. Given the long-

term nature of the cointegration analysis, we require each ¢rm to have a mini-

mum of 500 trading days (approximately 104 consecutive weeks) with continuous

stock and target price data.21 The ¢nal sample consists of 900 ¢rms.

To set the stage for our methodological approach and to build intuition for the

full-sample analysis, we begin by examining the joint price behavior of a single

¢rm, IBM.

A. Basic Setup and Application to IBM™s Stock and Target Price Behavior

Figure 2 depicts the time series of weekly market prices for IBM over our sam-

ple period, January 1997 through December 1999.We also plot the weekly consen-

sus target price and the ratio of target to market price, denoted TP/P. Inspection

of this ¢gure provides a key insight. While both price series are nonstationary

(we are unable to reject the null of nonstationarity with an augmented Dickey^

Fuller test), it is evident that market and target prices share a long-run common

20

Our use of the term ˜˜long-term relation™™ di¡ers from that used in the traditional cointe-

gration literature (Engle and Granger (1987)) in which the term ˜˜long-term equilibrium™™ has

been employed. The motivation for the use of the latter term is the idea that when two coin-

tegrated economic variables drift apart from this equilibrium, economic forces eventually

drive them back to it (e.g., Lee, Myers, and Swaminathan (1999), Hasbrouck (2002)). Finally,

as Campbell and Shiller (1988) argue, such an ˜˜equilibrium™™ can occur simply because one

variable is a rational expectation of the future value of another variable that follows an inte-

grated process. Since target prices are, by de¢nition, analysts™ forecasts of future prices, we

have therefore chosen to avoid using the term ˜˜long-term equilibrium™™ in our analysis in favor

of ˜˜long-term relation.™™

21

While the choice of a two-year minimum period reduces the number of ¢rms that we can

study, it is a necessary requirement, as we are interested in studying the long-term dynamics

of target and market prices. We have replicated the cointegration analysis using a minimum of

250 trading days and obtained similar results. In addition, the analysis was conducted on

size-sorted portfolios containing all ¢rms in our database. Our conclusions regarding the dy-

namics of market and target prices remain unchanged.

1956 The Journal of Finance

160 2

1.8

140

1.6

Stock Price and Target Price ($)

IBM's Target Price/ Price Ratio

120

1.4

Target Price/Price

100

1.2

IBM's Consensus Target Price

80 1

0.8

60

0.6

40

0.4

IBM's Stock Price

20

0.2

0 0

970106

970218

970331

970512

970623

970804

970915

971027

971208

980120

980302

980413

980526

980706

980817

980928

981109

981221

990201

990315

990426

990607

990719

990830

991011

991122

Date

Figure 2. IBM™s weekly stock price, target price, and target price-to-price ratio.

This ¢gure depicts weekly market price and consensus target price for IBM. The market

prices are from CRSP and the weekly consensus target price is the average of all outstand-

ing target prices issued over the preceding 90 days. All target prices are from the First Call

database. We also plot the ratio of the target price to market price at each point in time,

denoted by TP/P. The vertical axis on the left-hand side corresponds to the price series

while the axis on the right-hand side corresponds to the TP/P ratio.

relation. This association is captured by our third variable, TP/P. It can be seen

that target prices, which are forward looking, are consistently higher than cur-

rent market prices, but that the ratio of the two price series £uctuates about a

common value at approximately 1.2 (see the right-hand-side vertical axis). When

either price series deviates from this ratio, the system tends to revert back to this

value over time.

Using a cointegration approach (Engle and Granger (1987)), we provide direct

evidence regarding both the long-term relation and the manner with which these

prices correct toward this long-term relation.To do that, we assume that analysts™

ex ante expected return, and therefore the mean of the TP/P ratio, is constant

over our sample period. We denote the time t target price by tpt and the market

price by pt.The (2 ‚ 1) vector xt has elements tpt and pt. Both variables are assumed

integrated of order one, that is, stationary after ¢rst di¡erencing. Cointegration

of the two price series means that there is a (2 ‚ 1) vector b such that the following

linear relationship holds: b 0 xt ¼ 0.22

22

We have conducted Johansen™s cointegration rank test for each of the 900 price series that

we later study in Section III.C and ¢nd that we are able to reject the null of no cointegration

essentially for all ¢rms.

An Empirical Analysis of Analysts™ Target Prices 1957

We begin by writing the price system in an error correction form:

X

p

Dxt ¼ Pi DxtÀi þPxtÀ1 þ et ; 8t ¼ 1; . . . ; T; °3Þ

i¼1

where Dxt À i is a (2 ‚ 1) vector of ¢rst di¡erences lagged i periods and etBN(0,O),

assumed independent over time. The matrices Pi and O are both (2 ‚ 2), and the

latter is assumed positive de¢nite. P is the (2 ‚ 2) long-run impact matrix that con-

tains the cointegrating vector. Inclusion of this term in this vector autoregressive

regression (VAR) follows from the Granger Representation Theorem. P rank is™s

equal to the number of cointegrating relations which in our simple case is just one.

Following earlier literature, we parameterize P as follows:

P ¼ ab0 ; °4Þ

where both a and b are (2 ‚ 1) vectors. The vector b contains the cointegrating

coe„cients, and for identi¢cation we normalize its ¢rst element to À 1, that is,

b( À 1,b). For example, if, on average, analysts forecast that target prices are 20

percent higher than current market prices, then b ¼ [À 1 1.2]. The vector a is the

vector of weights in theVAR regression.This vector can be interpreted as a vector

of ˜˜adjustment coe„cients™™; that is, the elements in a allow us to quantify how

target prices and market prices react to past deviations from the long-term rela-

tion. By obtaining estimates of the elements in a, we can quantify the way in

which the two time series contribute to the correction of the system back to the

long-term relation.

B. Cointegration Results for IBM

We present the cointegration results in Panel A of TableVI.We focus our atten-

tion on b, the parameter capturing the long-run ratio of target prices relative to

market prices, as well as on the (2 ‚ 1) vector a, capturing the response coe„-

cients of each price variable to deviations from the long-run relation. The long-

term relation for IBM is 1.23, indicating that, ex ante, analysts expected that

IBM™s annual return would be 23 percent.

The estimates of the vector of response coe„cients, a, provide interesting evi-

dence on the manner with which target and market prices adjust to the long-term

relation. The target price response, atp, is large and positive (0.17, with a t-

statistic ¼ 8.15), while the market price response, amarket, is statistically indistin-

guishable from zero. This ¢nding can be interpreted as follows. Suppose market

and target prices are currently 100 and 123 dollars, respectively, and that a 4 -dol-