International Review of Economics and Finance 35 (2015) 235248

Contents lists available at ScienceDirect

International Review of Economics and Finance

journal homepage: www.elsevier.com/locate/iref

Analyst valuation and corporate value discovery

Yih-Wenn Laih a, Hung-Neng Lai b,, Chun-An Li c
a
b
c

Department of Finance, Ling Tung University, No. 1, Ling Tung Rd., Taichung City 40852, Taiwan
Department of Finance, National Central University, Jhongli City, Taoyuan County 32001, Taiwan
Department of Finance, Nation Yunlin University of Science and Technology, 123 University Road, Section 3, Douliou, Yunlin 64002, Taiwan

a r t i c l e

i n f o

Article history:
Received 27 February 2013
Received in revised form 8 October 2014
Accepted 9 October 2014
Available online 18 October 2014
JEL classication:
G14
G32

a b s t r a c t
This paper examines rm-level valuations by nancial analysts and by the market, using a traditional vector error-correction model (VECM) or threshold vector error-correction model
(TVECM) to obtain the information shares of the two parties. While investors' valuations lead nancial analysts' valuations in most rms, the reverse is not uncommon. A cross-sectional analysis
reveals that analyst forecasts are more valuable for rms with less trading, less uncertainty, and
weaker association between prices and earnings.
2014 Elsevier Inc. All rights reserved.

Keywords:
Analyst forecast
Valuation
Information shares
Residual income model

1. Introduction
This paper examines rm-level value discovery between analyst-forecast-based and market valuations. Similar to price discovery
in nancial markets for gathering and interpreting news (Baillie, Booth, Tse, & Zabotina, 2002), value discovery incorporates the news
to determine the value of the rm.
A rm's stock price theoretically reects both supply and demand sides in the market and is usually regarded as investors' viewpoints of corporate valuation. If the capital market is efcient in reecting all available information, then nobody can outperform the
market in assessing a rm's value. However, given the fact that information collection is costly, it is possible that a certain group of
people may value the rm better than the market (Grossman & Stiglitz, 1980).
This paper focuses on the interplay between the nancial analysts' and market valuations. Financial analysts are important information intermediaries in capital markets. They provide information that investors value, as demonstrated in a substantial body of research. Like other market participants, they learn from the market valuation. Examining the interactions between analysts and the
market is useful in understanding how the nancial industry operates.
There is an abundant amount of literature on the relationship between stock market prices and analyst forecasts or recommendations. Barber, Lehavy, Mcnichols, and Trueman (2001) and Jegadeesh, Kim, Krische, and Lee (2004) show that trading according to
analysts' consensus recommendations yields signicant returns. Chung and Kryzanowski (2001) nd that investor demands are
related to the number of analysts following. Diether, Malloy, and Scherbina (2002) investigate the relationship between earnings forecast dispersion and subsequent stock returns. Elgers, Lo, and Pfeiffer (2003), Clement and Tse (2003), Gleason and Lee (2003), and
Corresponding author.
E-mail addresses: rubylai@teamail.ltu.edu.tw (Y.-W. Laih), hnlai@cc.ncu.edu.tw (H.-N. Lai), liica@yuntech.edu.tw (C.-A. Li).

http://dx.doi.org/10.1016/j.iref.2014.10.004
1059-0560/ 2014 Elsevier Inc. All rights reserved.

236

Y.-W. Laih et al. / International Review of Economics and Finance 35 (2015) 235248

Frankel, Kothari, and Weber (2006) examine the market's response to earnings forecast revisions. These research papers often examine the market reaction in rather short time periods; Altnkl and Hansen (2009) and Altnkl, Balashov, and Hansen (2009) even
use intradaily windows. Our work differs with them in that we use a cointegration model to see whether the market can still learn
from the analysts in the long term.
Another similar line of research is to examine dividend-based equity valuation models. Early works are primarily developed for
valuing aggregate market indices (Bakshi & Chen, 1996, 1997; Bekaert & Grenadier, 2000; Campbell & Shiller, 2001; Chen, 2012).
There is also a substantial body of literature measuring whether changes in earnings can predict future stock returns at the aggregate
level (Ang & Bekaert, 2007; Kryzanowski & Mohsni, 2013; Lamont, 1998; Lee, Myers, & Swaminathan, 1999). Asset pricing research
has focused largely on the portfolio level, but not on stock valuation per se. Only a few papers look at valuing individual stocks
(Bakshi & Chen, 2005; Vuolteenaho, 2002; Wei & Yang, 2012). Our work, on the other hand, focuses on the relationship between
rm-level market value and analyst forecast earnings rather than realized earnings.
We study value discovery for 736 U.S. rms with available data since 1983. We use Lee et al. (1999) multi-stage residual income
model (hereafter LMS) to estimate the analyst valuation. The model is capable of reecting a rm's fundamentals through analysts'
earnings forecasts.1 We then model the relationship between analyst and market valuations as a cointegrated system. Our approach
herein does not assert that price be equal to value at all times, but rather that they should converge to the intrinsic value over the long
term. Transaction costs, market friction, information asymmetry, and psychological bias will break the long-run equilibrium between
these two valuations, and the short-run dynamic between the price and value depends on the distinct level of costs from market
imperfections.
The traditional linear vector error-correction model (VECM) is often applied to estimate a smooth time series, but it is unable to
consider the asymmetric adjustment of the long-term relationship among the variables in the model. In this study, we try to apply
the linear VECM to sample rms whenever possible. For those samples which does not pass the linear cointegration test propose
by Johansen (1988), we employ the Hansen and Seo (2002) two-regime threshold vector error-correction model (TVECM) with
the intercept term as the threshold variable, in which there could be an asymmetric adjustment to long-run equilibrium, depending
on whether the deviation from the equilibrium exceeds a critical threshold. There are two advantages from using the TVECM in our
study. First, the model isolates the periods during which the two valuations diverge, which may indicate the existence of asset
price bubbles. For all the rms tted with TVECM in our sample, one of the regimes converges and the other diverges. Second,
Enders and Siklos (2001) argue that the powers of unit-root and co-integration tests will be low if asymmetric effects exist in the
underlying time series, and so it is necessary to consider other models when the traditional VECM does not t the data well; see
also Balke and Fomby (1997) and Enders and Granger (1998). Although TVECM may not be the best model, it serves as an alternative
to detecting the properties that cannot be done by the traditional models.
Our value discovery analysis leads to several interesting ndings. Firstly, convergent linear cointegration relationship exists between analyst and market valuations in the majority (415 out of 736) of rms. TVECM is applied to another 37 rms. Thus, the
speed of the adjustment to the long-run equilibrium values is symmetric for most of the rms. For the long-run cointegrating relationship, the two valuations do not move closely together. Over-extrapolation or over-condence often exists in the market valuation relative to the analyst valuation for this kind of rms, which is consistent with the ndings in Frankel and Lee (1998), Ali, Hwang, and
Trombley (2003) and other studies that the analyst valuation predicts future returns. Secondly, according to the sizes of the adjustment coefcients from VECM, analyst valuations, rather than prices, do most of the adjustment in bringing the deviation back toward
the long-run equilibrium level. The rm-level stock return generally predicts valuation changes. However, short-run Granger causality
results show that investor and analyst valuations disconnect in most of the regimes, that is, changes in market valuations do not predict changes in analyst valuation and vice versa.
Thirdly, following common factor weights' methodology in Gonzalo and Granger (1995), we nd that market valuations have marginal information dominance in contributing to a rm's intrinsic value, with the median of common factor weights being 54.3%, and
that of analysts is 44.10%. In an efcient market, the price fully reects available information, so the theoretical common factor weight
of market valuation is 100% and analysts always follow the market. By contrast, Gervais and Odean (2001) and Seru, Shumway, and
Stoffman (2010) suggest that individual investors have difculty learning from their experiences, and if they learn, this is a slow process. Our ndings are more consistent with the behavioral than the efcient market argument. Using quarterly data, we still nd that
the contributions by analysts are substantial for some rms, which indicates that it may take a very long time for the market to adjust
to the analyst valuation.
Lastly, we examine the common factor weights of analyst valuations in a cross-sectional analysis. The relative importance of analyst valuation may be determined by several factors. First, if there is more information coming out of the market, then analysts may
play less important roles in leading the market. We nd that this is indeed the case: analysts' common factor weight is inversely related to market information proxies such as rm sizes and turnover ratio. Second, we would like to know whether the more uncertain
the environment for the rm, the more analysts lead the market. The results show that it is the other way round; analysts lead the
market for those rms with low idiosyncratic risk, low return-on-asset volatility and low payout-yield variability. Third, for the
rms with large difference between analyst and market valuations, or for the rms for which the associations between earnings
and prices are weak, the common factor weights of analyst valuations tend to be large.

1
Lee et al. (1999) show that the ratio of the average V/P, where V is the rm's intrinsic value constructed by the residual income model, and P is the stock price, predicts future returns. However, they do not consider the possibility that stock prices may affect analyst forecast revisions. Subsequent papers using the LMS technique
include D'Mello and Shroff (2000), Dong, Hirshleifer, Richardson, and Teoh (2006), Lin, Chou, and Cheng (2011), and Ma, Whidbee, and Zhang (2011).

Y.-W. Laih et al. / International Review of Economics and Finance 35 (2015) 235248

237

The rest of this paper is organized as follows. Section 2 explains the corporate evaluation process and introduces the econometric
models. Section 3 describes the data, discussing the analyst valuation calculated by the LMS model with time-varying and rmspecic cost of equity. Section 4 discusses the empirical results and Section 5 concludes this paper.
2. Corporate value discovery process
2.1. The model
This section introduces a model of the corporate valuation process, motivated by the methodology of Pascual, Pascual-Fuster, and
Climent (2006) that distinguishes two sources of information asymmetries on cross-listing securities. Consider the assessment of a
rm's valuation by two different parties: nancial analysts (F) and investors (I). The following structural model describes the corporate valuation process of the two parties:
F

vt vt1 zt vt1 t t t ;


i
i
i
vt vt1 t t J t ;

i F; I;

i i

At st vt :

In Eq. (1), the efcient intrinsic value vt of a rm consists of its value in the previous period vt 1, and the innovation zt. The innovation includes two uncorrelated zero-mean stochastic processes it, i = F, I, which represent value-related innovations updating party
i's information set, and the other innovation t that may be partly conveyed to the two parties. Eq. (2) describes party i's expectation,
which consists of the lagged intrinsic value vt 1, the party's private information ti, and part of the common innovation (nt Jti),
where Jti is a zero-mean, homoscedastic, and uncorrelated process allowing the possibility for the parties to imperfectly receive the
public information.
In Eq. (3), Ait describes the party's evaluation for the rm's value. The rst term sit captures the behavioral bias component that
masks the valuation vit, while sit N 1 and sit b 1 imply over- and under-extrapolation, respectively. If sFt /sIt is close to one, then both parties
are subject to similar behavioral bias.
The term Ait is a non-stationary process. The non-stationary component vt 1 is common to all assessments. Since the assessments
are made for the same rm, they are not expected to drift far apart from each other, and the difference between them should be stationary. Generally, investors make buysell decisions according to all available information including the analyst reports. Analysts in
turn revise their forecasts based on rms' fundamental news and the value-related information embedded in the market trade activity. Therefore, AFt and AIt co-integrate with the theoretical cointegration vector (1,); that is, AFt + AIt denes a stationary process.
We use the stock price as a proxy for market valuation and the implicit value from the residual income model as a proxy for analyst
valuation. We examine the response of the marketanalyst valuation nexus to infer the rm's intrinsic value.
2.2. Cointegration, threshold error correction, and common factor weights
To estimate the cointegrating relationship between AFt and AIt, we adopt both the Johansen (1988, 1991) linear vector error correction model (VECM) and the Hansen and Seo (2002) two-regime threshold vector error correction model (TVECM). As the convergence may only occur when the spread between analyst and market valuations is less than the arbitrage costs, the use of TVECM
could be potentially more meaningful in characterizing their valuation dynamics.
The baseline empirical model for the co-integrated analyst and market valuations is
F

At
I
At

c0
I
c0



a1
wt1
b1

a2
b2



A
I
A

ut
I
ut

!
;

where c0i is a constant and wt 1 = c1 + AFt 1 + AIt 1 is the normalized cointegration term, reecting the long-run equilibrium.
According to Eq. (3), depends on sFt /sIt. Here, Ai is a k by one vector of Ait j Ait j 1, j = 1, 2, , k 1, and a's and b's are one
by k coefcient vectors associated with AF and AI.
The speed of adjustment parameter i is the response of party i to a divergence from the true (but unobservable) intrinsic
value. Following the price discovery hypotheses, we dene the two-way value adjustment to mean that both error correction
coefcients i are statistically signicantly different from zero, so that innovations in the intrinsic value of a rm are impounded
into both valuations. On the other hand, one-way value discovery occurs entirely in Ai if its adjustment coefcient i is not signicantly different from zero while the other j is. For example, if I is signicantly different from zero, then it implies any deviation of w from the long-run equilibrium at time t 1 will lead to change in AI at t. If, at the same time, F is not signicant,
then it implies the major price adjustment only takes place in AI but not AF, so that there is only one-way value discovery from F
to I. uit's are uncorrelated white noises.

238

Y.-W. Laih et al. / International Review of Economics and Finance 35 (2015) 235248

When there is structural change with regard to the informational environment of a rm, the speeds of adjustments for value
discovery may no longer be the same. To account for this possibility, model (4) may be expanded to be a more general one, i.e. a
TVECM:

At
I
At

8


F
>
a1
>
w
>
>
I
t1
<
b1



 0
>
> 0F
a1
>
>
: 0I wt1 b 0
1

!

F
F
ut
A

if wt1
I
I
A
ut
!
:



0F
0
F
ut
a2
A

if wt1 N
I

0I
b2
A
ut

a2
b2



In the model, is the threshold parameter, which is analogous to the arbitrage costs in the price-discovery literature. The model
allows an asymmetric adjustment to the long-run equilibrium, in which the switch between the two regimes depends on the deviation from the equilibrium, wt 1, exceeding the critical threshold . It is also possible that the long-run equilibrium is attainable in one
regime but not in the other.
In order to assess the relative contribution of each valuation toward the corporate value discovery process, we utilize Gonzalo and
Granger (1995) common factor weights with the modied permanent-transitory decomposition of Gonzalo and Ng (2001) to measure the relative importance between these two valuations. We transform Eq. (4) or (5) into a vector moving average (VMA) and
its integrated form. The Gonzalo and Granger permanent-transitory (PT) decomposition has the form:
At f t Gt ;

where At is the vector of analyst and market valuations, is a loading matrix; Gt is a vector of I(0) transitory components that does not
Granger cause ft in the long run. ft = At is a vector of I(1), the permanent component. Specically:



F
F I
At
f t At
I
At

!
;

which is a linear combination of cointegrated variables, where the coefcient vector determines the weights of variables. Namely,
the weights of AF and AI are F/( I + F) and I/( I + F), which, following Baillie et al. (2002), is estimated by I/( I F) and
F/( I F), respectively; see Appendix A for details. A variable with a greater weight moves more closely with the common
long-run stochastic trend, so that it contributes more to the common factor. Under the TVECM, the weights are computed separately for wt 1 and wt 1 N in a similar way.
3. Data
3.1. Data descriptions
Financial data on the sampled rms are gleaned from Compustat les, stock prices are from CRSP, and analysts' earnings forecasts
are from I/B/E/S Detailed History les. The limitation by the timing of nancial reports compels us to use quarterly-frequency data.
For a rm to be included in the sample, there must be data available for at least 30 quarters between 1983Q1 and 2010Q2.2 To mitigate
the effects of outliers, we trim the observations in which the estimated costs of equity are in the extreme top and bottom one percentile (Barth, Beaver, Hand, & Landsman, 2004; D'Mello & Shroff, 2000). 736 rms are left in the sample.
3.2. Analyst and market valuations
To acquire synchronous assessments from analyst and market valuations in each quarter, we form matched data sets that minimize the span between these two valuations. We collect the average analysts' earnings per share forecasts on the third Thursday of
each quarterly month, when the I/B/E/S update the database, and use the stock prices at the same day to measure the market valuation, AI, for each sample rm. To avoid stale estimates, only those issued within 60 days before the third Thursday of each quarterly
month, are used to compute the average forecasts.3 We use the residual income model to estimate implicit value.4 The analyst valuation of a rm, AF, at time t is dened as
F
At

Bt


F
r t1
r t Bt
1 rt


F
r t2
r t Bt1
1 r t 2


F
r t3
r t Bt2
1 rt 2 r t

2
In the original sample, there are 3470 rms with at least 10 quarters of observations. Because at least 12 parameters have to be estimated under the traditional
VECM in Eq. (4), we exclude the rms with less than 30 observations. The demand for big sample size inevitably limits our investigation to those rms with more analyst
coverage.
3
We require the announce date, the activation date, or the review date be at or before the sample day, but not 60 days earlier than that day.
4
See Lee et al. (1999) and D'Mello and Shroff (2000) for a more detailed review of this method.

Y.-W. Laih et al. / International Review of Economics and Finance 35 (2015) 235248

239

where Bt is the adjusted book value per share available at time t, rtF + i is forecasted return of equity (ROE) for time t + i, and rt is
the cost of equity.5 Specically, Bt is calculated by the adjusted book value per share at t 1 period,6 plus earnings per share in period
t,7 and then minus total payout, including dividends and net repurchases.8 The nite horizon estimate for each rm is computed
by treating the average of the last four quarterly earnings as perpetuity, and the terminal value is restricted to be nonnegative
(D'Mello & Shroff, 2000; Elliott, Koeter-Kant, & Warr, 2007).
The forecasted ROE at time t + i is computed as follows:
F

r ti

F
eti
;
F
Bti1

where BFt + i 1 is the forecasted book value and eFt + i is the mean forecasted EPS at time t + i before extraordinary and special items.9
We compute future book value per share using the adjusted current book value per share, the long-term dividend payout ratio, and
earnings forecasts as Bt + i = Bt + i 1 + (1 Kt)eFt + i, where Kt is the long-term dividend payout ratio at time t, dened as the average
of the last two years' dividend payout ratios. We exclude share repurchases due to the practical problems associated with determining
the likelihood of their recurrence in future periods. Following Lee et al. (1999), if rms experience negative earnings, we divide the
dividends paid by 0.06 multiplied by total assets to derive an estimate of the dividend payout ratio. Payout ratios of less than zero
(greater than one) are assigned a value of zero (one).
A good valuation model incorporates time-varying and rm-specic properties corresponding to riskiness of future cash ow to
investors. We thus use the Capital Asset Pricing Model (CAPM) to estimate rt, a rm's cost-of-equity, as:



r t r f r m r f t ;

10

where t = cov(rt, rm)/var(rm) is obtained by running a regression of daily returns of rm rt on the CRSP value-weighted index returns
(with dividends) over a period of 500 days ending on date t 1 relative to each quarter date t. Here, (rm rf) is the average quarterly
return difference between market return rm and risk-free rate rf, with rm computed by the average quarterly market return during
1983Q12010Q2, and rf, as suggested by Lee et al. (1999), takes the 3-month T-bill rate. The CRSP risk premium during 19832010
is 0.01743. Thus, we use analysts' earnings per share forecasts from the I/B/E/S le released quarterly, time-varying cost-of-equity
rt, and the latest nancial statement data, in order to estimate each rm's analyst valuation AFt every quarter.
3.3. Estimations
To estimate the cointegration relationship between AF and AI, we follow the standard procedures by rstly assessing the stationarity of these two valuations. The augmented Dickey and Fuller (1981, ADF) and Phillips and Perron (1988, PP) unit root tests show
that the null hypothesis that AF and AI contain unit root cannot be rejected while the null hypothesis that AF and AI contain a unit
root is rejected for all sample rms. Next, we test for the presence of a linear cointegration relationship using Johansen's (1988)
method. The test allows linear cointegration with different specications in the intercept of Eq. (4):
Mode A :
Mode B :
Mode C :

c0 0; c1 0
i
c0 0; c1 unconstrained
i
c0 ; c1 unconstrained:

11

A linear cointegration relationship in a mode is said to be convergent if (a) the Johansen test demonstrates that these two valuation
are cointegrated with one cointegration relationship, and (b) the absolute values of the long-term error-correction coefcient F and
I, and the absolute value of lagged coefcients, a1 and b2, are all less than one, otherwise it is divergent. If convergent relationships
exist in more than one mode for a rm, the LjungBox test is used to choose the mode which produces the error terms that are closest
to white noise. 415 of the 736 sample rms show convergence in the cointegration test, 118 rms show divergent linear relationship,
and 203 rms do not have linear relationship.
To explore the possible asymmetric dynamic adjustment between the two variables for the 203 rms, we estimate the TVECM and
the threshold effect is examined by the Hansen and Seo's (2002) SupLM test. The TVECM ts well in 37 rms, and the rest of 166 rms
show no threshold cointegration relationship. All of the 37 rms have one converged and one diverged regimes: 19 rms converge in
5
Although I/B/E/S updates the database each month, rms issue nancial statements in a quarterly basis. To ensure that nancial results are close to earnings forecasts, we only use the forecasts in the quarterly months, when the nancial statements from the previous quarters have been released.
6
The adjusted book value is the common equity (Compustat item #Q59) minus treasury stock dollar amount (item #Q98). The number is then divided by common
shares outstanding (#Q61) minus treasury stock number of common shares (TSTKNQ). Quarterly data of TSTKNQ (Compustat item #Q87) were not available until 1999,
and hence annual data (#A87) are used before that time.
7
Following Skinner (2008), it is the earnings before extraordinary and special items (Compustat #18 0.6 #17) divided by the number of common shares outstanding (#Q61 TSTKNQ), assuming an effective tax rate of 40%.
8
As stock repurchases are economically signicant after 1980 (Skinner, 2008), we dene the total payout as the increase in Cash Dividends (Compustat #Q89) plus
the increase in Treasury Stock Amount (#Q98), divided by the number of common shares outstanding (Compustat #Q61 Compustat TSTKNQ).
9
If no forecasted EPS are available in that quarter, we use the forecasted long-term growth rate (Ltg) in the I/B/E/S le to compute a one-quarter-ahead EPS forecast. If
Ltg is not available, we replace EPS in four prior quarters with one-quarter-ahead EPS forecast value.

240

Y.-W. Laih et al. / International Review of Economics and Finance 35 (2015) 235248

the wt 1 regime, and 18 rms converge in the wt 1 N regime. It seems the TVECM is effective in separating converged and
diverged regimes.
3.4. Summary statistics
We now focus on 452 rms which shows either converged linear relationship in VECM, or the converged regimes in TVECM.
Table 1 reports the characteristics of this sample. Both the market value and the traded shares are positively skewed, indicating
that some of our sample rms are of extremely large sizes and more actively traded in the market. The mean ROE of our sample
rms is lower than those of ROA and the cost of capital, and the median of ROE is higher than those of ROA and the cost of capital,
which means that there are some rms of which the ROE's are very small. Although our study period covers the latest subprime mortgage crisis, both AF and AI are still positively skewed. On average, the analyst valuations are smaller than the market valuations for individual rms.
Fig. 1 plots the median of AI AF in each quarter by the black line and the interquartile range by the gray band. The difference is
positive in general, reaches its rst peak in the rst quarter of 1998, falls down to the lowest point in the rst quarter of 2003, rises to
the highest point in the third quarter of 2007, and falls during and after the recent nancial crisis. The interquartile range is the biggest
in the rst half of year 2000, and is also very high in the third quarter of 2007. Fig. 2 shows the average discount rate rt, which appears
to be falling until the third quarter of 2003. Then it rises to the recent peak in the second half of 2007 and falls again.
4. Empirical results
4.1. Vector error correction models
Table 2 reports the main results from the cointegration and VECM analysis. Panel A of Table 2 shows that the linear VECM ts 415
rms and the TVECM suits 37 rms. The majority of the linear VECM cases belong to Mode B, in which there is no intercept in the equations and non-zero intercept may exist in the cointegrating vector. A possible explanation for the intercept and threshold is the divergences between the two valuations due to information asymmetry and psychological bias. Panel B of Table 2 shows the lag structure of
the VECM. We use the AIC/SBC criterion to determine the lag lengths in the model.
Panel C of Table 2 presents descriptive statistics for the estimated constants, cointegrating vectors and thresholds. The constant
term, c1, which does not exist in the theoretical cointegrating relationship AF + AI, implies imperfections in the valuation process,
presumably due to obstacles in information collections and transmissions. Although the mean and median values of c1 are small,
the large standard deviations indicate that there are extreme cases in which the adjustment to the long-run equilibrium is seriously
impeded. Regarding the cointegrating vector, measures the ratio of AF and AI in the long-run equilibrium. The market valuation and
analyst valuation do not move closely together, and the corresponding mean estimates of are between 0.277 and 0.752, indicating that the law of one price does not hold between the two valuations. It implies that the long-run impact of a new shock could
have different permanent impacts on two assessments. To contrast the a priori cointegration vector (1, sFt /sIt), derived from the corporate value discovery process discussed in Section 2, the estimated 's imply that over-extrapolation or over-condence exists in the
market valuation relative to the analyst valuation for this kind of rms.
The mean and median estimates of the absolute value of in TVECM are respectively 2.316 and 0.656, with variations smaller
than those of c0 in the linear model. For the 19 rms that converge at the region wt 1 , the mean and median of are respectively
10.20 and 6.25. For the 18 rms that converge at wt 1 N , the mean and median of are respectively 32.31 and 17.08. The majority of the observations fall in converged regimes, and AF and AI diverge only when the difference between them is very large.
Table 1
Summary statistics.
This table presents summary statistics for the characteristics of the 452 rms which show either converged linear cointegration relationship in VECM or converged
cointegration regime in TVECM over the same period. The average quarterly Market Value of the sample rm is presented in millions of US dollars. Traded Shares
(in millions) represent the average quarterly number of shares traded for rms. Return on Assets, Return on Equity, Ltd to Total Capital (total long-term debt divided
by invested capital) and Total Debt to Capital are calculated using average quarterly data. AI is the quarterly market valuation. AF is the quarterly analyst valuation. Cost
of Capital of each rm-quarter is estimated by the CAPM. The systematic risk, Beta, is run with daily returns of each rm and returns on the CRSP value-weighted index
returns (with dividends) over a period of 500 days relative to the end of the quarterly day.
Firm variables

Mean

Std

Q1

Median

Q3

Market Value
Traded Shares
Return on Assets (%)
Return on Equity (%)
MB ratio
Ltd to Total Capital (%)
Total Debt to Capital (%)
AI
AF
Cost of Capital (%)
Beta

3016.66
34.92
3.15
1.83
2.77
30.97
40.63
21.44
13.28
3.70
0.99

7714.88
177.12
3.75
23.13
36.49
33.12
49.95
19.29
16.68
14.09
0.51

309.63
3.77
1.01
0.99
1.23
7.42
16.15
9.47
4.49
2.32
0.63

849.05
10.17
3.02
3.19
1.88
27.76
39.90
17.00
10.15
2.93
0.91

2463.06
26.28
4.93
4.81
2.99
46.14
60.66
28.00
18.62
3.65
1.27

Y.-W. Laih et al. / International Review of Economics and Finance 35 (2015) 235248

241

Fig. 1. AI AF. This gure plots the cross-sectional median of AI AF in each quarter by the black line and the interquartile range by the gray band, where AI is the market
closing price at the third Thursday of each quarterly month, and AF is constructed by the residual income model in Eq. (8) using analysts' earnings forecasts at the same
day.

Table 3 provides the joint tests of the long-run value discovery and short-run Granger causality relationship between AF and AI for
the 452 linear convergent regimes. A regime presents two-way value discovery if both of the error-correction coefcients, F and I
(or F and I), are signicantly different from zero. A regime offers only the analyst value discovery if I is signicantly different from
zero and F is not that is, only the market valuation is adjusted to the analyst valuation and there is no signicant long-run adjustment in the other way round. Similarly, a regime presents only the market value discovery if F is signicantly different from zero and
I is not, and there is no value discovery if neither F nor I is signicantly different from zero. On the other hand, the signicance of
any of the short-run coefcients, a2,j (a2,j), indicates that the market valuation Granger causes the analyst valuation, and the significance of b1, j (b1,j) indicates that the analyst valuation Granger causes the market valuation, and there is two-way Granger causality if
at least one of the a2,j and one of the b1,j are signicant.

Fig. 2. Average discount rates. This gure plots the cross-sectional average discount rate. For each rm, the discount rate is rt = rf + (rm rf)t, where t = cov(rt, rm)/
var(rm) is obtained by running a regression of daily returns of rm rt on the CRSP value-weighted index returns (with dividends) over a period of 500 days ending on
date t 1 relative to each quarter date t.

242

Y.-W. Laih et al. / International Review of Economics and Finance 35 (2015) 235248

Table 2
VECM results.
Panel A of the table presents the model specication based on the Johansen (1988) test of linear cointegration and Hansen and Seo (2002) SupLM test of the coefcients
of the error correction term in the two regimes having the same value. Mode A represents the model estimated on the restrictions that ci0 = 0 and c1 = 0, Mode B represents the model that only ci0 = 0 is constrained, and there is no restriction in Mode C. Panel B presents the descriptive statistics for the number of lag length in
Eqs. (4) and (5). Panel C presents the cross-sectional descriptive statistics of the cointegration coefcients and threshold parameters given by Eqs. (4) and (5).
Panel A: Model specication
# of rms

Model
Linear VECM
Mode A
Mode B
Mode C
Two-regime VECM
Total

415
65
304
46
37
452

Panel B: Number of lags

Liner VECM
Two-regime VECM

Min

Mean

Medium

Max

1
1

1.612
1.342

1
1

6
6

Panel C: Cointegrating vectors and threshold

Mean

Std

Q1

Median

Q3

Linear VECM
c1
Mode B
Mode C

Mode A
Mode B
Mode C

1.025
0.834
0.496
0.277
0.537
0.532

84.366
33.583
6.753
1.391
7.805
2.555

7.486
7.886
1.215
0.818
1.420
1.284

0.153
0.379
0.558
0.502
0.530
0.711

12.164
12.010
0.100
0.264
0.055
0.126

Two-regime VECM

0.752
2.316

0.968
19.543

1.054
6.588

0.556
0.656

0.297
10.232

The test results reveal that there is no short-run Granger causality in the majority of the regimes (319 out of 452), followed by
analyst valuation Granger causes market valuation (83 rms). It means that short-term interactions between analyst and market valuations do not occur very often. However, most of the rms exhibit long-run value discovery. The biggest category is of two-way
discovery, which accounts for about 60% of the sample, and it is followed by the one-way AI discovery. The one-way discovery for
AF accounts for one quarter of the sample, which suggests that the adjustments of the market toward analyst valuations do not
often occur.
4.2. Value discovery results
Panel A of Table 4 presents cross-sectional descriptive statistics of the estimated error correction vector from the VECM given by
Eqs. (4) and (5). The error correction vectors provide the information on the adjustment of each valuation series to the deviation from
the equilibrium in the previous period. At least one of the valuations must respond to the deviation in order to prevent bubble opportunities. For instance, if the market valuation is too low relative to the analyst valuation, that is, wt 1 N 0, then the market valuation
will rise and the analyst valuation will fall in the following period to restore the equilibrium. Thus, the theoretic expected signs of F
Table 3
Long-run value discovery and short-run Granger causality.
This table presents the test results of the long-run value discovery and short-run Granger causality test. The value discovery tests for signicant deviations from zero for
the adjustment parameter i. There is one-way value discovery for AF if F is not signicantly different from zero whereas I is. There is one-way discovery for AI if I is
not signicantly different from zero whereas F is. There is two-way discovery if both F and I are signicantly different from zero. The Granger causality (GC) tests the
j) is signicantly different from
signicance of any of the short-run coefcients. The market valuation Granger causes the analyst valuation if at least one of the a2, j (a2,
zero, and the analyst valuation Granger causes the market valuation if at least one of the b1, j (b1, j) is signicant. Two-way Granger causality means at least one of the a2, j
j) are signicant. No GC means none of the coefcients is signicant.
(a2, j) and at least one of the b1, j (b1,

2-way GC
AI GC AF
AF GC AI
No GC
Sum

Two-way discovery

AF only discovery

AI only discovery

Sum

6
20
47
203
276

1
3
8
61
73

6
14
28
55
103

13
37
83
319
452

Y.-W. Laih et al. / International Review of Economics and Finance 35 (2015) 235248

243

Table 4
Adjustment coefcients and the common factor weights.
Panel A of the table presents cross-sectional descriptive statistics of the adjustment coefcients of the error correction item given by Eqs. (4) and (5). Panel B presents
descriptive statistics of the common factor weights in value discovery calculated using the GonzaloGranger method. The weights of AF and AI are respectively given by
Eqs. (A.2) and (A.3).
Panel A: Adjustment coefcients

F
I

Mean

Std

Q1

Median

Q3

0.244
0.040

0.308
0.375

0.450
0.168

0.180
0.046

0.002
0.284

Panel B: Common factor weights in value discovery

Weights of

Mean

Std

Q1

Median

Q3

AF
AI

0.481
0.506

0.347
0.345

0.215
0.214

0.441
0.543

0.749
0.775

and I will be positive and negative, respectively, and their absolute values show the magnitude of the response. The median of
adjustment coefcients for the market valuation, I, is 0.046, and the median correction in the analyst valuation, F, of 0.180.
Two adjustment coefcients with opposite signs indicate that the analyst and market valuation nexus for these rms has a normal
pattern of valuation adjustment toward the equilibrium. In addition, the sizes of the adjustment coefcients show that AF adjusts at
a faster rate than AI. This suggests that the market valuation has higher value discovery ability.
According to the signicance of the adjustment coefcients, we subsequently use Eqs. (A.2) and (A.3) in Appendix A to calculate
the common factor weights between analyst and market valuations in the value discovery process, which are listed in Panel B of
Table 4. The mean (median) contributions of AI and AF are 0.506 (0.543) and 0.481 (0.441), respectively. The market valuation has
richer information content than the analyst valuation in corporate value discovery and plays a more important role than the analyst
valuation in determining the equilibrium value. The results provide further evidence that new information tends to be incorporated
into market valuation, while the analyst valuation follows behind.
It has to be emphasized that the overall evidence does not entirely support efcient market hypothesis. While it is true that the
common factor weight of the market valuation is bigger, the analyst valuation should not play any role in long-run price discovery
in an efcient market. In such a market, even if analysts add new information, the market will absorb it and fully reects the value
in the short term, so that it is possible that the analyst valuation Granger causes the market valuation. In the long term, however, analysts should not lead the market. The fact that the median common factor weight of analyst can be as high as 0.441 indicates that the
market is slow to adjust to the analyst valuation, and this result is consistent with the behavior story of Gervais and Odean (2001) and
Seru et al. (2010) that investors are slow to learn from their experiences.
4.3. The cross-section of analysts' common factor weights
The size of our sample, 452 rms in total, enables us to run cross sectional regressions to examine factors that affect the share of the
analyst valuation contribution to value discovery. The dependent variable in the regressions is CFW, which is dened as the logarithm
of the common factor weight of AF in Eq. (A.2) for each rm. Since the common factor weight lies between 0 and 1, the transformed
variable lies between minus innity and 0.
We start with the following regression to analyze possible factors that can impede or strengthen AF in value discovery10:
C FW b0 b1 ln MV b2 MB b3 IR b4 VS b5 AccRsq
b6 TR b7 SD8ROA b8 SDTPY b9 DCR :

12

Here, ln MV in Eq. (12) is the logarithm of the market capitalization. Existing research has identied many possible impediments to
business valuation. Bhushan (1989) and Lang and Lundholm (1996) suggest that the demand for analyst research and information
produced about a security increases in rm size. However, if information is abundant in big rms, the contributions of analysts
may become marginal. Next, MB is the market-to-book ratio, which reects the rm's growth and investment opportunities
(Frankel et al., 2006). Growth rms have more unrecorded, intangible assets, whose valuation depends heavily on future protability.
Forecasting future prospects requires expertise and the collection of data beyond the nancial statements.
The term IR is the rm's idiosyncratic risk. Pontiff (2006) demonstrates that it dampens arbitrage activity and represents components of the fundamental value of the arbitrage position uncorrelated with market index returns or any other hedge portfolio returns
available to the arbitrageur. Thus, idiosyncratic risk should impede information transmission in valuation. Following Gagnon and
Karolyi (2010), we estimate each rm with a regression model of the quarterly return difference on contemporaneous market
index returns:
Rt r m ;
10

Appendix B provides detailed descriptions of the independent variables in the cross-sectional analysis.

13

244

Y.-W. Laih et al. / International Review of Economics and Finance 35 (2015) 235248

Table 5
Correlation matrix.
This table reports Pearson correlation coefcients between variables used in cross-sectional regressions. CFW is the natural logarithm of the common factor weight of
the analyst valuation. lnMV is the natural logarithm of the average quarterly market value (in millions of U.S. dollars) of the 452 rms. MB is the ratio of market-to-book.
IR is idiosyncratic risk. VS is the ratio of value spread. AccRsq is derived from the tted residuals from a pooled cross-sectional regression of prices on the book values of
shareholders' equity and earnings using data from the prior 5 years. TR is the turnover ratio. SD8ROA is the standard deviation of ROA over the prior 8 quarters. SDTPY is
the standard deviation of total payout yield. DCR is the debt to capital ratio. Three, two and one asterisks (*) denote the signicance of the correlation coefcient at the
0.01, 0.05, and 0.10 levels, respectively.

CFW
lnMV
MB
IR
VS
ACCRSQ
TR
SD8ROA
SDTPY
DCR

CFW

lnMV

MB

IR

VS

ACCRSQ

TR

SD8ROA

SDTPY

0.0097
0.0447
0.1513***
0.0092
0.0474
0.1870***
0.1021**
0.0905*
0.1300***

0.0041
0.5824***
0.1170**
0.0871*
0.1713***
0.0202
0.0867*
0.2116***

0.1940***
0.0582
0.0133
0.1490***
0.0376
0.0074
0.0126

0.2294***
0.1301***
0.6411***
0.0481
0.0577
0.2591***

0.0261
0.1390***
0.0208
0.0006
0.0960**

0.1620***
0.0014
0.0019
0.0858*

0.0606
0.0271
0.2257***

0.0143
0.0104

0.0753

where Rt is the daily return of market valuation, and rm are the daily S&P 500 total return index returns. The standard deviation of
the residuals of Eq. (13) is used as a proxy for the idiosyncratic risk of the value discovery. We then calculate an average standard deviation of the residuals for each rm during our study period.
The value spread between analyst and market valuations, VR, represents the dispersion between market valuation and theoretical
analyst valuation. We compute this measure in a similar spirit to the ratio of bidask spreads from Eun and Sabherwal (2003),
Grammig, Melvin, and Schlag (2005), and Korczak and Phylaktis (2010). Value spread is calculated as the absolute value of the difference between AF and AI and then divided by their average. It is calculated for a given rm each quarter, and then the quarterly observations are averaged over the sample period. The value spread captures differences in opinion among market participants. If analyst
forecasts are useful in correcting the market valuation, then their common factor weight will increase with the value spread.
Frankel et al. (2006) argue that the strength of the contemporaneous association between security prices and accounting numbers
is an indicator of the informativeness of accounting numbers. Therefore, we calculate AccRsq to measure the price earnings association.
Each year we run a pooled cross-sectional regression of prices on the book equities and earnings using the prior ve-year data. Each
rm's ve residuals are scaled by the price, squared, and averaged. This average is then subtracted from the corresponding average for
the entire population of rms to create AccRsq for the year. The ve-year window then rolls forward to calculate AccRsq for the next
year. Therefore, the larger the AccRsq, the better the earnings and book equities explain prices.
We use the turnover ratio TR to capture the (uninformed) liquidity traders that contribute to the overall trading volume in a
security. Holding the supply of shares constant, more liquidity trading makes the price process noisier, i.e., increase volatility
(Verrecchia, 1982; Bhushan, 1989). The greater the turnover ratio of the rm's stocks, the more liquidity traders (the less informed
traders) there are in the market, so turnover ratio should be positively correlated to the analyst valuation contribution to value discovery.11 On the other hand, liquid markets allows informed traders to trade without excess costs and facilitates price discovery,
which implies that turnover ratio should be negatively correlated to the analyst forecasts' common factor weight.
Frankel et al. (2006) argue that high return volatility indicates high uncertainty about a security's cash ow and thus presents
traders with an incentive to acquire information, which would mitigate the uncertainty. Return volatility can also result from information asymmetry between management and outside investors. Therefore, return volatility spurs the analyst valuation contribution
to value discovery. We calculate the measure (SD8ROA) by the standard deviation of ROA over the prior 8 quarters. According to
Skinner (2008), we dene ROA as operating income before depreciation (Compustat #13) divided by lagged total assets (Compustat
#6). This measure captures the idea that there are smaller forecast errors in low return volatility rms.
Larraina and Yogo (2008) nd that the variation in net payout yield, the ratio of net payout to asset value, is mostly driven by
movements in expected cash ow growth, instead of movements in discount rates. Chen and Zhao (2009) note that cash ow
news dominates at the rm level but discount rate news dominates at the aggregate level. Because our study focuses on equity
valuation, we use the standard deviation of total payout yield, or the ratio of total payout to net value (SDTPY). Both SD8ROA and
SDTPY measure rms' uncertainty from different dimensions.
Avramov, Chordia, Jostova, and Philipov (2009) nd that earnings dispersion is related to future returns through default risks. AlAttar, Hussain, and Zuo (2008) further observe that the explanatory power of accounting information for returns declines when rms
face bankruptcy risks. Therefore, we include debt-to-capital ratio (DCR) as a proxy for nancial distress risk in the model to see how
the analyst forecasts' common factor weight changes with this risk.
Table 5 presents correlation coefcients between variables used in cross-sectional regressions. Analyst valuation in value discovery
is signicantly correlated with debt-to-capital ratio and inversely correlated with idiosyncratic risk, turnover ratio, SD8ROA and SDTPY.

11
Hribar and McInnis (2012) suggest that market sentiments may affect all the participants therein, including nancial analysts, and that analyst forecasts are relatively more optimistic for hard-to-value stocks when sentiment is high. If turnover ratio is a proxy for investor sentiment, then Hribar and McInnis' argument implies
that turnover ratio is negatively correlated with analysts' common factor weight.

Y.-W. Laih et al. / International Review of Economics and Finance 35 (2015) 235248

245

Table 6
Determinants of common factor weight in the analyst valuation.
This table presents the cross-section means of coefcients, standard deviations (in parentheses), and adjusted R2s from the regression models. The dependent variable is
the natural logistic transformation of the common factor weight coefcient of the analyst valuation. Variable denitions appear in B. The sample includes 452 rms
during the period 1983Q12010Q2. The t-test is based on White heteroscedasticity consistent standard errors. Three, two and one asterisks (*) denote the signicance
of the correlation coefcient at the 0.01, 0.05, and 0.10 levels, respectively.
Model

(i)

1.0115
(0.8468)
0.1459
(0.1232)
0.0087
(0.0269)

0.2703***
(0.0585)
0.0086
(0.0285)

0.0797**
(0.0358)
0.0004***
(0.0001)
2.6454***
(0.7503)
0.0182***
(0.0014)
0.0002***
(0.0001)
0.0122**
(0.0058)
0.0500

0.0674*
(0.0363)
0.0004***
(0.0001)
2.9917***
(0.6961)
0.0179***
(0.0014)
0.0002***
(0.0000)
0.0108*
(0.0058)
0.0494

lnMV
MB

(ii)

IR
VS
ACCRSQ
TR
SD8ROA
SDTPY
DCR
R2-adjusted

(iii)

(iv)

1.0089**
(0.4259)

1.6714
(1.3149)
0.4137***
(0.1496)
0.0003
(0.0251)
70.9446***
(19.2638)
0.1112***
(0.0338)
0.0003***
(0.0001)

0.0062
(0.0298)
49.5316***
(15.6165)
0.0981***
(0.0353)
0.0003***
(0.0000)

0.0194***
(0.0012)
0.0002***
(0.0000)

0.0321

0.0179***
(0.0011)
0.0002***
(0.0000)
0.0129**
(0.0058)
0.0512

Idiosyncratic risk is highly correlated with turnover ratio (correlation coefcient of 0.6411) and inversely correlated with rm size
(correlation of 0.5824) and debt-to-capital ratio (correlation of 0.2591).
We next analyze the determinants of the analyst valuation in value discovery, as shown in Table 6. The various specications aim
to minimize the potential problem of multicollinearity. Value spread and debt-to-capital ratio are positively correlated with the common factor weight of analyst valuation, and rm size, idiosyncratic risk, AccRsq, turnover ratio, return volatility and total payout yield
variability negatively inuence the common factor weight. The signs of rm size and turnover are as expected, which indicate that
analysts lead the market in rms with less information or less trading.
The negative signs of volatility variables, such as idiosyncratic risk, return volatility and total payout yield variability, suggest the
weakness of analyst forecasts, that is, analysts play a more important role in value discovery when the environment is relatively
certain. The positive signs of value spread, on the other hand, indicate that analyst forecasts are useful to lead the market for the
rms of which the two different valuations are far apart. The positive signs of DCR indicate that analysts tend to lead the market
when nancial distress risks are higher. If the association between accounting numbers and returns are weak when rms face bankruptcy risks, as Al-Attar et al. (2008) suggest, then it implies that analyst leads the market when the earnings are not related to prices.
The negative sign of AccRsq is puzzling. It suggests that analyst forecasts lead more in the rms of which the priceearning relationship is weak. If AccRsq is interpreted as a certainty variable, then this result implies that analysts lead the market more in an uncertain environment, contrary to the results of other volatility variables. Moreover, if the contemporary relationship between prices
and earnings is weak, it is even puzzling why analyst forecasts lead market valuation for those rms. Perhaps earning management
weakens the association between prices and earnings. Perhaps it is the bubble contained in the price that weakens the association.
Both conjectures are possible explanations for the analysts to lead the market, and are consistent with the result that the value spread
is positively correlated with the analysts' common factor weight.12
5. Conclusions
This paper models information transmission between market valuation and theoretical analyst valuation. We employ the traditional linear vector error correction model as well as a threshold cointegration model proposed by Hansen and Seo (2002) to capture
the asymmetric adjustment process to the long-run equilibrium relationship. Two-way value discovery exists in 276 out of 452 of the
rms with convergent relationships, and analyst valuation leads market valuation in 73 rms, which indicates that analysts' forecasts
contribute to value discovery substantially. Although linear cointegration exists for the analyst and market valuations in the majority
of stocks, two valuations do not move closely together. Over-extrapolation or over-condence often exists in the market participants
related to the analysts for our samples. In terms of the short-run relationship, the Wald tests reveal that the majority of the regimes do
not exhibit Granger causality between analyst and market valuations, implying that short-run information transmission is weak between the two valuations.
12
However, the coefcients of AccRsq lose their signicance if we re-run the regression by trimming the extreme 1% (ve rms) of the observations. It seems that the
results of AccRsq are not very robust.

246

Y.-W. Laih et al. / International Review of Economics and Finance 35 (2015) 235248

We then estimate common factor weights that indicate the proportion of the permanent component of a rm's value adjustments
attributable to the market valuation and the theoretical analyst valuation. We nd that market valuations have information dominance in discovering a rm's value with common factor weights averaging 54.3%. It implies that analysts' contributions are substantial, and apparently at odds with the efcient market hypothesis.
The determinants of the analyst valuation in value discovery are mainly attributed to less market information (low market value
and turnover ratio), low uncertainty (idiosyncratic risk, return and payout yield variability), low earning and price association (high
value spread, low price-earning association, and high debt-to-capital ratio). In other words, market valuation leads analyst valuation
when the market is liquid and when information is abundant, so that analysts' contributions are marginal. Market valuation is also
prevailing for the rms with much uncertainty, implying that the analysts are less capable of coping with a changing environment.
Lastly, analyst valuation leads the market valuation for the rms that the price-earning association of the two is weak, and whether
the phenomenon is due to earning management or asset bubble is worth further investigations.
Acknowledgment
We are very grateful for the inspiring comments and suggestions made by an anonymous referee and the seminar participants
at National Central University, Tamkang University, National Tsing Hua University, National Chengchi University, and Chung Yuan
Christian University.
Appendix A. Calculating common factor weights
The common factor weights of AF and AI in the GonzaloGranger decomposition are F/( F + I) and I/( F + I), respectively. Let
= ( F, I) be the vector of the adjustment coefcients and = (F, I) be the cointegrating factor,13 and then according to Gonzalo
and Granger (1995),
 0
1 0

A:1

where = (F , I ) and = (F, I) are vectors such that = = 0. Gonzalo and Granger (1995) estimate by the
1
eigenvector associated with the smallest eigenvalue of matrix X = X1
uu XuvXvv Xvu, where X is a variancecovariance matrix for residual
vectors from the ordinary least squares regression; subscripts u and v denote the level and rst-difference regression, respectively.
Gonzalo and Ng (2001) claim that Gonzalo and Granger's method is numerically less precise in small samples. They show that the
eigenvector associated with the smallest eigenvalue of matrix is a better estimator of . In particular, when = (1, 1), one may
choose = (1, 1), so that the weights of AF and AI are F /(F + I ) and I /(F + I ), respectively. Since = 0, Baillie et al.
(2002) present that weights of AF and AI equal I/( I F) and F/ I F, respectively.
In fact, Baillie et al.'s result is not limited to the case in which = (1, 1). To see this, note that I = FF / I and I =
F F
/I. Then Eq. (A.1) can be written as
 0
1 0

h



i


F
F F
I
F
F F
I 0 1
F
F F
I
; =
; =
; =
!


 FF
F
F F
I
F F
F F

; = =
I I
!


 F F I I
F
F F
I
F F

; = =
I I
!
I I
F I




:
;

F F I I F F F I I F
As a result, the weights of AF and AI are






F
I
F
I
I
F
= = ;

A:2





I
I
F
F
I
F
= =

A:3

and

respectively.
13

In Section 2.2 we standardize the cointegrating factor wt 1 = c1 + AFt + AIt such that = I/F.

Y.-W. Laih et al. / International Review of Economics and Finance 35 (2015) 235248

247

Appendix B. Denitions of rm-level variables

Variable

Denition

lnMV
Market-to-book (MB)

The natural logarithm of the average quarterly market value in millions of U.S. dollars.
The ratio of Market-to-book. The quarterly ratios are averaged over the sample period for a given rm. Market-to-book is
market value of equity (Compustat #25 Compustat #199) divided by book value of common equity (Compustat #60).
The standard deviation of the residuals that are obtained by regressing returns from the pairs, against the returns on the
S&P 500 index. Following Gagnon and Karolyi (2010), we estimate each rm's regression model of the quarterly return
difference on contemporaneous market index returns:

Idiosyncratic risk (IR)

Rt r m ;

Value spread (VS)

Priceearnings association
(AccRsq)

Turnover ratio (TR)

Return volatility (SD8ROA)
Total payout yield variability
(SDTPY)
Debt-to-capital ratio (DTC)

where Rt is the return of market valuation, and rm is the S&P 500 total return index returns. The standard deviation of the
residuals from the regression is used as a proxy for the idiosyncratic risk of the value discovery.
The ratio of value spread. Value spread is calculated as the absolute value of the difference between AF and AI divided by
their average. It is calculated for a given rm each quarter, and then the quarterly observations are averaged over the
sample period.
Each year we run a pooled cross-sectional regression of prices on the book equities and earnings using the prior ve-year
data. Each rm's ve residuals are scaled by the price, squared, and averaged. This average is then subtracted from the
corresponding average for the entire population of rms to create AccRsq for the year. The ve-year window then rolls
forwards to calculate AccRsq for the next year.
The ratio of the average quarterly trading volume (Compustat #Q18) to the number of common shares outstanding
adjusted Treasury Stock over the sample period.
The standard deviation of ROA over the prior 8 quarters. This measure captures the idea that there are smaller forecast
errors in the low earnings volatility rms.
The standard deviation of total payout yield. The payout yield is the ratio of total payout to net value.
Total debt divided by invested capital.

References
Al-Attar, A., Hussain, S., & Zuo, L.Y. (2008). Earnings quality, bankruptcy risk and future cash flows. Accounting and Business Research, 38(1), 520.
Ali, A., Hwang, L.S., & Trombley, M.A. (2003, April). Residual-income-based valuation predicts future stock returns: Evidence on mispricing vs. risk explanations.
Accounting Review, 78(2), 377396.
Altnkl, O., Balashov, V.S., & Hansen, R.S. (2009, September). Evidence that analysts are not important information-intermediaries. Working paper. Tulane University.
Altnkl, O., & Hansen, R.S. (2009, October). On the information role of stock recommendation revisions. Journal of Accounting and Economics, 48(1), 1736.
Ang, A., & Bekaert, G. (2007, May). Stock return predictability: Is it there? Review of Financial Studies, 20(3), 651707.
Avramov, D., Chordia, T., Jostova, G., & Philipov, A. (2009). Dispersion in analysts' earnings forecasts and credit rating. Journal of Financial Economics, 91, 83101.
Baillie, R.T., Booth, G.G., Tse, Y., & Zabotina, T. (2002, July). Price discovery and common factor models. Journal of Financial Markets, 5(3), 309321.
Bakshi, G.S., & Chen, Z. (1996, Spring). Inflation, asset prices and the term structure of interest rates in monetary economies. Review of Financial Studies, 9(1), 241275.
Bakshi, G.S., & Chen, Z. (1997, April). An alternative valuation model for contingent claims. Journal of Financial Economics, 44(1), 123165.
Bakshi, G., & Chen, Z. (2005, May). Stock valuation in dynamic economies. Journal of Financial Markets, 8(2), 111151.
Balke, N.S., & Fomby, T.B. (1997, August). Threshold cointegration. International Economic Review, 38(3), 627645.
Barber, B.M., Lehavy, R., Mcnichols, M., & Trueman, B. (2001, April). Can investors profit from the prophets? Security analyst recommendations and stock returns.
Journal of Finance, 56(2), 531563.
Barth, M.E., Beaver, W.H., Hand, J.R., & Landsman, W.R. (2004, September). Accruals, accounting-based valuation models, and the prediction of equity values. Working
paper. Stanford University.
Bekaert, G., Grenadier, S.R., 2000. Stock and bond pricing in an affine equilibrium. Mimeo. Columbia University.
Bhushan, R. (1989, July). Firm characteristics and analyst following. Journal of Accounting and Economics, 11(23), 255274.
Campbell, J.Y., & Shiller, R.J. (2001, April). Valuation ratios and the long-run stock market outlook: An update. NBER working paper 8221. Cambridge University.
Chen, S. (2012, April). The predictability of aggregate Japanese stock returns: Implications of dividend yield. International Review of Economics and Finance, 22(1),
284304.
Chen, L., & Zhao, X. (2009, April). Return decomposition. The Review of Financial Studies, 22(12), 52135249.
Chung, R., & Kryzanowski, L. (2001, Summer). Tests of investor cognizance using earnings forecasts of North American analysts. International Review of Economics and
Finance, 10(2), 187204.
Clement, M.B., & Tse, S.Y. (2003, January). Do investors respond to analysts' forecast revisions as if forecast accuracy is all that matters? The Accounting Review, 78(1),
227249.
D'Mello, R., & Shroff, P.K. (2000, October). Equity undervaluation and decisions related to repurchase tender offers: An empirical investigation. Journal of Finance, 55(5),
23992425.
Dickey, D.A., & Fuller, W.A. (1981, July). Likelihood ratio statistic for autoregressive time series with a unit root. Econometrica, 49(4), 10571072.
Diether, K.B., Malloy, C.J., & Scherbina, A. (2002, October). Differences of opinion and the cross section of stock returns. Journal of Finance, 57(5), 21132141.
Dong, M., Hirshleifer, D., Richardson, S., & Teoh, S.H. (2006, April). Does investor misvaluation drive the takeover market? Journal of Finance, 61(2), 725762.
Elgers, P.T., Lo, M.H., & Pfeiffer, R.J., Jr. (2003, MayJune). Analysts vs. investors weightings of accruals in forecasting annual earnings. Journal of Accounting and Public
Policy, 22(3), 255280.
Elliott, W.B., Koeter-Kant, J., & Warr, R.S. (2007, March). A valuation-based test of market timing. Journal of Corporate Finance, 13(1), 112128.
Enders, W., & Granger, C.W.J. (1998, July). Unit-root tests and asymmetric adjustment with an example using the term structure of interest rates. Journal of Business &
Economic Statistics, 16(3), 304311.
Enders, W., & Siklos, P.L. (2001, April). Cointegration and threshold adjustment. Journal of Business & Economic Statistics, 19(2), 166176.
Eun, C.S., & Sabherwal, S. (2003, April). Cross-border listings and price discovery: Evidence from U.S.-listed Canadian stocks. Journal of Finance, 58(2), 549575.
Frankel, R., Kothari, S., & Weber, J. (2006, April). Determinants of the informativeness of analyst research. Journal of Accounting and Economics, 41(12), 2954.
Frankel, R., & Lee, C.M. (1998, June). Accounting valuation, market expectation, and cross-sectional stock returns. Journal of Accounting and Economics, 25(3), 283319.
Gagnon, L., & Karolyi, G.A. (2010, July). Multimarket trading and arbitrage. Journal of Financial Economics, 97(1), 5380.

248

Y.-W. Laih et al. / International Review of Economics and Finance 35 (2015) 235248

Gervais, S., & Odean, T. (2001, January). Learning to be overconfident. Review of Financial Studies, 14(1), 127.
Gleason, C.A., & Lee, C.M.C. (2003, January). Analyst forecast revisions and market price discovery. Accounting Review, 78(1), 193225.
Gonzalo, J., & Granger, C. (1995, January). Estimation of common long-memory components in cointegrated systems. Journal of Business and Economic Statistics, 13(1),
2735.
Gonzalo, J., & Ng, S. (2001, October). A systematic framework for analyzing the dynamic effects of permanent and transitory shocks. Journal of Economic Dynamics and
Control, 25(10), 15271546.
Grammig, J., Melvin, M., & Schlag, C. (2005, January). Internationally cross-listed stock prices during overlapping trading hours: Price discovery and exchange rate effects. Journal of Empirical Finance, 12(1), 139164.
Grossman, S.J., & Stiglitz, J.E. (1980, June). On the impossibility of informationally efficient markets. The American Economic Review, 70(3), 393407.
Hansen, B.E., & Seo, B. (2002, October). Testing for two-regime threshold cointegration in vector error-correction models. Journal of Econometrics, 110(2), 293318.
Hribar, P., & McInnis, J.M. (2012, Februar). Investor sentiment and analysts' earnings forecast errors. Management Science, 58(2), 293307.
Jegadeesh, N., Kim, J., Krische, S.D., & Lee, C.M. (2004, June). Analyzing the analysts: When do recommendations add value? Journal of Finance, 59(3), 10831124.
Johansen, S. (1988, JuneSeptember). Statistical analysis of cointegration vectors. Journal of Economic Dynamics and Control, 12(23), 231254.
Johansen, S. (1991, November). Estimation and hypothesis testing of cointegration vectors in Gaussian vector autoregressive models. Econometrica, 59(6), 15511580.
Korczak, P., & Phylaktis, K. (2010, September). Related securities and price discovery: Evidence from NYSE-listed non-U.S. stocks. Journal of Empirical Finance, 17(4),
566584.
Kryzanowski, L., & Mohsni, S. (2013, January). Growth of aggregate corporate earnings and cash-flows: Persistence and determinants. International Review of Economics
and Finance, 25(1), 1323.
Lamont, O. (1998, October). Earnings and expected returns. Journal of Finance, 53, 15631587.
Lang, M.H., & Lundholm, R.J. (1996, September). Corporate disclosure policy and analyst behavior. The Accounting Review, 71(4), 467492.
Larraina, B., & Yogo, M. (2008, January). Does firm value move too much to be justified by subsequent changes in cash flow? Journal of Financial Economics, 87(1),
200226.
Lee, C.M.C., Myers, J., & Swaminathan, B. (1999, October). What is the intrinsic value of the Dow? Journal of Finance, 54(5), 16931741.
Lin, H.C., Chou, T.K., & Cheng, J.C. (2011, October). Does market misvaluation drive post-acquisition underperformance in stock deals? International Review of Economics
and Finance, 20(4), 690706.
Ma, Q., Whidbee, D.A., & Zhang, A.W. (2011, February). Value, valuation, and the long-run performance of merged firms. Journal of Corporate Finance, 17(1), 117.
Pascual, R., Pascual-Fuster, B., & Climent, F. (2006, May). Cross-listing, price discovery and the informativeness of the trading process. Journal of Financial Markets, 9(2),
144161.
Phillips, P.C., & Perron, P. (1988). Testing for a unit root in time series regression. Biometrika, 75(2), 335346.
Pontiff, J. (2006, October). Costly arbitrage and the myth of idiosyncratic risk. Journal of Accounting and Economics, 42(12), 3552.
Seru, A., Shumway, T., & Stoffman, N. (2010, February). Learning by trading. Review of Financial Studies, 23(2), 705739.
Skinner, D.J. (2008, March). The evolving relation between earnings, dividends, and stock repurchases. Journal of Financial Economics, 87(3), 582609.
Verrecchia, R.E. (1982, November). Information acquisition in a noisy rational expectations economy. Econometrica, 50(6), 14151430.
Vuolteenaho, T. (2002, February). What drives firm-level stock returns. Journal of Finance, 57(1), 233264.
Wei, P., & Yang, X. (2012, October). Do investors value REITs and non-REITs differently? International Review of Economics and Finance, 24(4), 295302.