International Research Journal of Finance and Economics

Comparative Assessment of Feltham–OhlsonSign-Oriented & Traditional Models

Department of Accounting, Shahid Chamran University of Ahwaz, IranMaster Accounting
Abstract
In this paper the efficiency of the four adjusted versions of Feltham-Ohlson(1995)models were studied using panel data (during 1998 to 2007) of firms listed on Tehran SockExchange. The first and third 留學生畢業dissertationversions (traditional models) are similar to Ohlson (1995) andFeltham-Ohlson(1995) models, respectively and the difference is that the other informationvariable has been ignored due to non access to analysts’ forecast. The second and fourthversions (sign-oriented models) attempt to improve traditional models by the segregation ofabnormal earnings in terms of their sign. The research results show that sign-orientedmodels have better performance in the prediction of abnormal earnings compared totraditional model. Using forecasted abnormal earnings in all of the valuation functionsindicate that none of the studied models can appropriately forecast firm’s value. Also, theresults suggest the superiority of Feltham–Ohlson based to the Ohlson based models bothin respect of abnormal earnings forecast and regarding studied firms’ valuation.
Keywords: Linear information models, sign of abnormal earnings, Firm valuation, Clean surplus accounting

1. Introduction
Many factors involve in the economic growth of the society. One of the major elements of thecountries’ economic growth is the existence of advanced financial and monetary markets. Capitalmarket is an element of financial markets. Investors provide the financial resources required for capitalmarket. They attempt to invest their financial resources in firms providing the highest return for them.Investors make investment decisions on the basis of received information.Information has two fundamental roles in capital market. Its first role is to make stock balancedprices affecting on the better assignment of the financial resources and increased firms’ decisionsefficiency. The second role is to enable persons to exchange their current and future consumptionclaims in different conditions and to access the favorable consumption patterns and, also, provide theirshare in social risks (Belkaoui, 2004).International Research Journal of Finance and Economics - Issue 36 (2010) 60Accounting is one of the information resources used by investors. Accounting is a data systemwhich provides investors with information on the firm’s financial status, its operational status as wellas information related to company ability in making cash flow, for decision making. To date various
researches have been conducted to identify that which category of this information plays the major role
in investors’ decision in different countries. In present research, since the required information are#p#分頁標題#e#
extracted from balance sheet and income statement, the objective is to review the usefulness of the first
two information categories in models which consider the time value of money.In section 2 the theoretical background and previous literatures have been explained. Section 3deals with the used models and methodology of research. Section 4 involves in the description ofresearch findings and, finally, the conclusion and research suggestions are given in section 5.
2. Theoretical Background and Literature Review
2.1. Theoretical Background
Since when the market-based accounting researches were commenced and developed, one of the mostimportant issues attracted the attention of accounting researches has been to study the feasibility ofmaking link between accounting data and stock prices to thereby prove the accounting claim on itscontribution to the decision making of the users of information gained from accounting process. Todate various researches have been made to predict stock prices, any of which considers one or morevariable as being more effective on stock prices changes than others and, therefore, they attempted toreview the effect of the given variables by controlling other variables. In such a condition, someresearches try to submit models regarding time value of money. These models mainly have steadytheoretic principles. Discounted Cash Flow, Discounted Dividends and Discounted Abnormal Earningsmodels are of this category.
Using the clean surplus relationship and applying discounted dividend model, Ohlson (1995)
submitted a model studied and tested by many researchers to date, and, some of them have
occasionally found better results through its adjustment. Ohlson model (1992) comprises of two parts;
in first part, with using linear information models, abnormal earnings of each period is introduced as
appropriate basis for the forecast of next period’s abnormal earnings and in the second part it benefits
from discounting the estimated abnormal earnings of the future periods for calculating the firm’s value.
2.1.1. Stock Valuation Models Based on Linear Information Models
The linear information models were initially presented by Ohlson (1995) and Feltham-Ohlson (1995).
In fact, the linear models are the dynamic models of information describing the time-series behavior of
abnormal earnings. Dechow et al. (1999) stress that the real outcome of Ohlson (1995) and Feltham
and Ohlson (1995) is that the linear models which they presented make a relationship between current
information and firm’s inherent value.
Studies of Ohlson (1995) and Feltham-Ohlson (1995) have become the major reference of
researches regarding linear information models. The main contribution of these models is to provide a
steady theoretic framework for stock valuation based on basic accounting variables (earning and book
http://www.400190.com/dissertation_writing/#p#分頁標題#e#value). In addition, these models allow any other kind of information to involve in firm’s value
forecast. Whereas, the results of previous studies dealing with the survey of these models do not
confirm their usefulness in market prices forecast. The results suggest that the values calculated by
these models are lower than market value. We continue with the explanation on the principles of these
models.
Discounted dividends models define the firm’s value as the present value of the future payable
dividends as follows:
E d V t (1)
61 International Research Journal of Finance and Economics - Issue 36 (2010)
Where: V t : Market value of the firm's equity at time t
Et [dt+i]: Estimated dividends at time t+i
r: Discount rate or cost of capital
The concept of clean surplus is confirmed on this rule that the retained earnings is restricted to
the registration of earnings and dividends within the period. Hence, the relationship between the book
value of the stakeholders’ equity and the earnings and dividends can be stated as follows:
bt = bt−1+ X t − dt (2)
Where: bt: Book value of equity at time t
Xt: Accounting profits of year t
The book value of the Stakeholders’ equity at time t-1 multiplied by the capital cost rate is
considered as the firm’s normal earnings. So, the earnings of time t minus the firm’s normal earnings
can be defined below as abnormal earnings:
X X t rb t
a
t = − −1 (3)
Where: X a
t : Abnormal earnings of year t
With the combination of equations 2 and 3,equation 4 can be written as follows:
d X ( r )bt bt
a
t = t + 1+ −1− (4)
Using the above equation and substitution of dt+i in Equation 1, one can reach valuation model
based on abnormal earnings or residual income:
( ) Σ
+
= +
∞
=
+
1 1
( ]
i
t t r
V b E X i
a
t t i (5)
Abnormal profit or residual profit valuation model indicate that firm’s value equals to the
stakeholders’ equity book value plus the current value of future expected abnormal profits (G.A.D.
Preinrich, 1983; E.O. Edwards and P.W. Bell, 1961; and K.V. Peasnell, 1982). It is the one of the
interesting features of this model that the firm’s value is not affected by accounting methods.
2.1.2. Ohlson (1995) Linear Information Model
Ohlson assumes that the time-series behavior of abnormal earnings is as follows:
+1 11 1 +1 = + + t t
a
t
a
t x ω x V ε (6)
Where:
+1 2 +1 = + t t t V γV ε
a
xt : abnormal earnings of year t
Vt : Other information variable at time t
11 ω : Persistence of abnormal earnings ( 0 1 11 < ω < )
After prediction of abnormal earnings, Ohlson applied them in his valuation model as follows:#p#分頁標題#e#
t
a
t t t V bV x V 1 2 = +α +α (7)
Where:
(1 )(1 )
, 1
1 11
2
11
11
1 ω γ
α
ω
ω
α
+ − + −
+
=
+ −
=
r r
r
r
2.1.3. Feltham and Ohlson (1995) Linear Information Model
Feltham and Ohlson (1995) have incorporated the accounting conservatism into their model. They have
segregated operational and financial assets. Accounting conservatism systematically causes lower
valuation of operational assets and has no effect on financial assets. Thus, one of the conservative
effects is the reduced normal earnings criterion. To take this conservatism into account, Feltham and
Ohlson present the following model:
+1 11 12 1 1 +1 = + + + t t t
a
t
a
t x ω x ω bV V ε (8)
International Research Journal of Finance and Economics - Issue 36 (2010) 62
Where:
+1 22 2 2 +1 = + + t t t t bV ω bV V ε
1 1 1 3 +1 = + t t t V γ V ε
2 2 2 4 +1 = + t t t V γ V ε
V2t ,V1t : other information variables at time t
11 ω : Persistence of abnormal earnings ( 0 1 11 < ω < )
12 ω : Conservatism parameter ( 0 12 ω > )
22 ω : Growth in the book value of equity parameter
1 2 γ ,γ : Persistences of the other information variables ( , 1 1 2 o < γ γ < )
The valuation formula of this model is as follows:
t t t
a
t t t V bV x bV V V1 2 1 1 2 2 = +α +α +β +β (9)
Where:
(1 )(1 )
& (1 )
1 11 22
12
2
11
11
1 ω ω
ω
α
ω
ω
α
+ − + −
+
=
+ −
=
r r
r
r
(1 )(1 )(1 )
& (1 )
(1 )(1 )
1
11 22 2
12
2
11 1
1 ω ω γ
β
ω γ
β
+ − + − + −
+
=
+ − + −
+
=
r r r
r w
r r
r
Since the segregation of operational assets from financial ones is not so simple, some
researchers have applied this model assuming that all firm’s assets are of operational kind (J.N. Myers,
1999; K. Lo and T. Lys, 2000; K. Ota, 2002). This idea have also used in present research.
2.2. Literature Review
To date many researches have been taken regarding the efficiency test of linear information models,
some of which have applied them after making some adjustments on some approaches of relevant
models. A brief summary on the most important researches will be given below.
Khodadadi et al. (2005) have reviewed seven linear information models with using 15 years#p#分頁標題#e#
time-series data in 21 firms listed on Tehran Stock Exchange. Their research results show that linear
information model No. 1, the same model presented by Ohlson (1995) which does not consider the
other information variable, has the best function in the forecast of abnormal earnings.
C.M.C Lee, et al. have studied whether traditional criteria (based on dividend, earnings and
ratio of book value to market value) can predict stock efficacy better or a criterion based on Ohlson
model. They conclude that Ohlson model is more successful in the forecast of stock returns than other
methods.
http://www.400190.com/dissertation_writing/Making a study on the previous researches having been tested Ohlson and Feltham- Ohlson
models, K. Lo and T. Lys (2000) concluded that there are empirically many methods to improve the
theoretic framework of these models and their validity test.
Dechow et al. (1999) emphasize that linear information models should make a relationship
between current information and firm’s value in order to achieve Ohlson and Feltham-Ohlson models
considered results.
J. Hand and W. Landsman (1998) found that when the other information variable is ignored in
Ohlson model, the dividend find direct relation with market value of stakeholders’ equity whereas it is
expected that this relation is reverse.
R. Frankel and C.M.C Lee (1998) once tested the Ohlson model with employing historical
information of profit and once more with applying profits predicted by analysts and, thus, they
concluded that using the analysts’ forecast would increase the model validity.
T.L. Stober (1996) employed Feltham-Ohlson model in the forecast of abnormal earnings in US
and with using a combination of cross-section and time series data in a 10 years period. Bauman (1999)
repeated Stober’s (1996) research but he used time-series data during a 15 years period of time. In two
63 International Research Journal of Finance and Economics - Issue 36 (2010)
above mentioned researches though the values of coefficients are as expected, their significance level
has not been referred to.
McCrae and Nilsson (2001) have tested Ohlson model in the forecast of abnormal earnings and
firm’s value in Sweden. They used cross-section data in their research and proved the efficiency of this
model.
Study of Choi, Ohanlon and Pope (2001) is similar to research made by McCrae and Nilsson
(2001) but they tested US stock exchange market. They, also, proved the applicability of Ohlson
model.
Moreover, Callen and Morel (2001) tested Ohlson model with using time-series data in a 25-
year period of time and with ignoring the other information variable in the United States and they also
proved the validity of this model.
When testing Ohlson model in Athena stock exchange, G.A. Karathanasis and S.N. Spilioti#p#分頁標題#e#
(2005) state that in order to overcome the problems caused by statistical methods such as autocorrelation,
co-linearity among independent variables and heterogeneous dispersion in employing
Ohlson model it is necessary to use a combination of these two types of data (panel data) instead of
solely use of cross-section or time series data. In their research, they tested a combination of cross
section- time series data in a 5 years period of time and confirmed the capability of model application.
Koji Ota (2002) reviewed the validity of Ohlson (1995) linear information model in Japan and
tried to improve its function through ignoring other information (Vt). Koji Ota believes that Myers
(1999), J. Hand &W. Landsman (1999) and M. Barth et al. (1999) have attempted to estimate Vt with
using other accounting information whereas in his research it has been tried to focus on autocorrelation
of error factor through ignoring Vt. The findings of this research, in general, confirm the
validity of Ohlson model (1995). And, the ignoring of auto-correlation from error factor obtained due
to deletion of other information or Vt is followed by model improvement.
R.G. Graham and Raymond King (2000) have reviewed the relationship between the stock
prices and the accounting earnings and also book values in 6 Asian countries (Indonesia, Korea,
Malaysia, Philippines, Taiwan and Thailand). Their analysis is based on abnormal earnings model
which expresses the firm’s value within the framework of book value and abnormal earnings. The main
objective of Graham and King was to review the possible effect of various accounting methods on the
explaining ability of abnormal earnings model. In their point of view, such researches can be very
useful and guiding in respect of settling accounting international standards.
By distinguishing between the firms with negative and positive abnormal earnings, Giner and
Iniguez (2006) concluded that this work improves Ohlson (1995) and Feltham–Ohlson (1995) models
performance. This suggestion is the main motivation for performing the present research.
3. Models and Methodology of Research
3.1. Research Population and Sampling Method
The statistical population considered in this research is firms listed on Tehran Stock Exchange until
March 20, 2007. Firms accepted thereafter have not been included in statistical population. With
considering the following conditions, statistical population was modified and all firms having such
conditions were studied.
1) The availability of data on their financial statements during the interval of 1998-2007.
2) The availability of data on their monthly prices during the interval of 1998-2007.
3) Having a fiscal year not ending to 20th March.
4) Being a non insurance or investment firm.
5) Having a nonnegative book value during the studied years.#p#分頁標題#e#
6) Being an active firm in Tehran Stock Exchange (firms whose stocks are purchased and sold
during the research period).
International Research Journal of Finance and Economics - Issue 36 (2010) 64
3.2. Sampling of Statistical Population
The sampling of the present research has been made in an objective stepwise manner, so that in each
step the firms not having each of the above conditions were extracted and, finally, all remained firms
were selected to conduct a test. Table 1 explains this procedure in a better manner.
Table 1: Sampling procedure
Description Qty
Active Firms whose accounting data during the time interval of research is accessible. 112
Deducted: firms whose monthly price data in this time interval have not been accessible. (37)
Deducted: firms whose fiscal year not ending to 20th March. (16)
Deducted: firms having a negative book value during the time interval of research. (1)
Total sum of qualified firms selected 58
3.3. Representation of Research Models and Assumptions
This research has two main assumptions each having been tested within the framework of four models
and the obtained results are compared with each other. These assumptions are as follows:
A There is a significant relationship between abnormal earnings of each period and
abnormal earnings of prior period
Hereafter we will refer to this as the existing forecast relation (EFR) hypothesis. This
assumption in second and fourth models (Ohlson and Feltham-Ohlson sign-oriented models,
respectively), with considering the sign of abnormal earnings (according to Giner and Iniguez, 2006),
and in first and third models (Ohlson and Feltham-Ohlson traditional models, respectively), without
considering their sign and, also, as a regression relation were tested by the least squared errors method.
It is noteworthy that in all studied models the of other information variable was ignored due to non
access to required data, and, the non existence of the first order serial correlation in error factor was
also reviewed through the formation of a regression relation between error measures of each year with
the prior year.
To estimate abnormal earnings we first obtain the cost of capital (r) for each company based on
the CAPM model.^ In consistency with the earnings figure, we measure the cost of capital after
taxes: (1 ).[ ( ) ] j,t t t j,t t r = − tax rf + β Rmkt − rf where:
t tax : Effective rate of tax in year t1
t rf : Free of risk return rate in year t
j,t β : Systematic risk for firm j at time t
t (Rmkt − rf ) : Risk premium at time t
In model No. 1 of this research, the abnormal earnings are not segregated in terms of sign and
are underestimated within the framework of below regression model.
, +1 11 , , +1 = + j t
a
j t
a
j t x ω x ε (10)#p#分頁標題#e#
Where:
a
j t x , : abnormal earnings of firm j at time t
11 ω : Persistence of abnormal earnings
Giner and Iniguez (2006) state that non segregation of abnormal earnings in terms of sign lead
to the utilization of negative abnormal earnings as the same as the positive ones in calculations. While
the investors’ interpretation of negative abnormal earnings differs from the positive ones because of
two following reasons:
1 It was computed each year as the median of the ratio: ,tax profit/profit before taxes, for all firms in the sample.
65 International Research Journal of Finance and Economics - Issue 36 (2010)
1. Investors do not generally consider the negative abnormal earnings in their calculations
because they hope to solve the current problems of firm and the future profitability in their
calculations.
2. Loss making firms during winding-up are often valuated under liquidation option and,
therefore, the least return of investment in firms’ share is fixed and the current losses have no
effect thereon.
On the same basis, they present their model No. 2 (the second model of this research) as
follows:
, 1 11 , 11 , , +1
+
+ = + + j t
a
j t t
a
j t
a
j t x ω x ω D x ε (11)
Where:
a
j t x , : abnormal earnings of firm j at time t
11 ω : Persistence of negative abnormal earnings
+ + 11 11 ω ω : Persistence of positive abnormal earnings
j t D , : Dummy variable for firm j at time t( 1 , = j t D if 0 , a >
j t x and 0 , = j t D otherwise
They state that since the negative abnormal earnings are temporary, the coefficient of these
earnings should be significant and varies from 1 (close to zero) while it is expected that this coefficient
should be, close to 1 for positive abnormal earnings. Therefore, the segregation of abnormal earnings
in terms of their sign will improve the results.
In the third model of this research, it has been paid to Feltham-Ohlson model and, as the same
as the first model, the sing of abnormal earnings has not been taken into account. The relations relevant
to this model are as follows.
, +1 11 , 12 , 1 , +1 = + + j t j t
a
j t
a
j t x ω x ω bV ε (12)
Where:
, +1 22 , 2 , +1 = + j t j t j t bV ω bV ε
a
j t x , : Abnormal earnings of firm j at time t
11 ω : Persistence of abnormal earnings
12 ω : Conservatism parameter( 0 12 w > )
22 ω : Growth rate in the book value of equity parameter
The fourth model is the same as the third model having been adjusted by the sign of abnormal
earnings and is expressed as follows.
, 1 11 , 11 , 12 , 12 , , +1
+ +
+ = + + + + j t j t j t
a
j t t#p#分頁標題#e#
a
j t
a
j t x ω x ω D x ω bV ω bV ε (13)
Whrere:
, +1 22 , 2 , +1 = + j t j t j t bV ω bV ε
a
j t x , : Abnormal earnings of firm j at time t
11 ω : Persistence of negative abnormal earnings
+ + 11 11 ω ω : Persistence of positive abnormal earnings
12 ω : Conservatism parameter for firms with negative abnormal earnings( 0 12 w > )
+ + 12 12 ω ω : Conservatism parameter for firms with positive abnormal earnings ( 0 12 12 ω + ω + > )
22 ω : Growth rate in the book value of equity parameter
At last, the capability of these four models in forecasting the abnormal earnings is compared.
B Calculated values obtained from discounted abnormal earnings of the future periods can
appropriately express the firm’s value in present time
This assumption hereafter will be referred to as the existing valuation relation (EVR)
hypothesis. According to Ohlson (1995) model, the firm’s value at any time can be calculated from the
International Research Journal of Finance and Economics - Issue 36 (2010) 66
total sum of stakeholders’ equity at that time and the current value of the future forecasted abnormal
earnings as follows.
Σ
+
= +
∞
=
+
1
1
t (1 )
t
a
t
t t r
V bv X (14)
As previously explained, Ohlson (1995) and Feltham-Ohlson (1995) convert this relation to a
linear relation by means of mathematical functions. in this respect and as per models submitted by
Giner and Iniguez (2006) the valuation functions of the four models considered in this research are
expressed as follows.
Model No Valuation function Where:
Model 1 V bv X a
t = t +α t
1 11
11
ω
ω
α
+ −
=
r
Model 2 V bv X a
t = t +α t
1 ( )
( )
11 11
11 11
r D
D
t
t
+
+
+ − +
+
=
ω ω
ω ω
α
Model 3 t
a
t t t V bV x bV 1 2 = +α +α (1 )(1 )
& (1 )
1 11 22
12
2
11
11
1 ω ω
ω
α
ω
ω
α
+ − + −
+
=
+ −
=
r r
r
r
Model 4 t
a
t t t V bV x bV 1 2 = +α +α
(1 [ ] )(1 )
(1 )( )
&
1 ( )
( )
12 12 22
12 12
2
11 11
11 11
1 ω ω ω
ω ω
α
ω ω
ω ω
α
+ − + + −
+ +
=#p#分頁標題#e#
+ − +
+
= +
+
+
+
r D r
r D
r D
D
t
t
t
t
In order to test this assumption, the estimated values (V) are calculated on the basis of valuation
models and coefficients assigned to each model and they are compared with real values (P) in the
following manner.
1. T test is used to review if the means of the estimated and real values are equal.
2. Wilcoxon signed-ranks test is used to review if the medians of the estimated and real values
are equal.
Also, as a result of using panel data method, tests special to this analysis method is applied
which will be explained later.
Zaranejad and Anvari (2005) state that panel data are referred to a set of data based on which
the observations are studied by a large number of cross-section variables (N), mostly selected by
random, during the specified periods of time (T). In this case, N*T statistical data are called panel data.
Since the panel data include both approaches of time-series and cross sectional data, employing
appropriate statistical explaining models to describe the features of those variables is more complex
than models used in time-series or cross sectional data. In recent years, the panel data method has been
used in many applied researches. The advantage of using this method is increased statistical power of
coefficients compared to separated analysis of statistical data as cross-section or time-series. In this
method, with a common consideration on the changes of variables in any section and at any time, all
available data are used and, therefore, the error of observations becomes lower. Although the range of
statistics is wide during the analyzing of statistical cross-sectional data, much more information is used
to survey panel data. As a result, with an increase in the range of statistics and information, the degree
of freedom is increased and the estimations made on statistical population will have much more
efficiency (as quoted by Gujarati, 2003). The other advantage is that using this method would lead to
the observation of some effects not being simply seen in time-series and cross-sectional data.
Therefore, panel data are better and more suitable for the dynamic study of changes.
Of course, it cannot be said that there is no problem while modeling with panel data. They state
that there are different methods to test panel data and special tests to select each of these methods.
These methods are Pooled Data Method, Fixed Effect Method, Random Effect Method, and SUR
Method; the selection between Random Effect Method and Fixed Effect Method is made according to
Hausman Test, the selection between Pooled Data Method and Fixed Effect Method according to
Chow Test, and the selection between Pooled Data Method and Random Effect Method according to
Beusch-Pagan LM Test, respectively.#p#分頁標題#e#
67 International Research Journal of Finance and Economics - Issue 36 (2010)
According to Zaranejad and Anvari (2005), the first condition to use a Random Effect Model is
that the variables should have been selected randomly. Thus, since in this research the variables have
not been selected randomly from different variables this method can be of no use. The Fixed Effect
Model also incorporates the effects of other variables in the intercept of regression equation so that the
relevant regression equation should have intercept, and, since the forecasting model under test in
present research, according to Ohlson (1995) and Feltham-Ohlson (1995), lacks intercept, this method
cannot be applied as well. Moreover, SUR Method may be used in cases where the length of timeseries
is more than the number of cross sections and as this condition has not been included in present
research, this method cannot be utilized. Therefore, Pooled Data Method was employed in this
research.
There are special tests to survey the stationary of variables in panel data methods; tests such as
Levin and Lin Test, Im, Pesaran and Shin Test, ADF Test, Fisher Test, Breitung Test and Hadri Test,
some of which have been explained by Zaranejad and Anvari (2005). Breitung Test was utilized to
review the stationary of abnormal earnings in this research.
4. Findings of Research
Firstly, in Table 2, a summary on the descriptive statistics of the research variables has been presented
and, then, it is paid to the review of research findings.
Table 2: Descriptive statistics (accounting data in million euros)
Variable N Mean Median Minimum Maximum Skewness Kurtosis
Abnormal earnings 580 22.3678 8.5589 0.31 136.17 2.608 6.945
Earnings 580 1.2115 0.7638 0.4 3.34 1.749 2.635
Lagged book value(bvj,t-1) 580 1.6514 0.8343 0.36 4.44 1.022 -0.226
Cost of equity rate 580 -17.1214 -5.1604 -92.24 0.2 -2.48 6.695
Free of risk return rate 11 0.1732 0.17 0.15 0.2 0.318 -1.552
Systematic risk(βj,t) 580 1.1655 0.3607 0 6.22 2.48 6.695
Market Value of equity 580 17.5731 11.8515 6.06 53.45 1.956 4.426
Fixed conversion rate(1 Euro=13000 Rials) is used to convert accounting data into Euro
4.1. Stationary Test
In panel and time-series data, the required condition for testing data is the dynamics of research
variables. So, firstly, to be assured of the trueness of the results gained in next steps, the stationary of
the variables of abnormal earnings and book value were tested within the framework of panel data with
using Breitung Test. The null hypothesis of this test has been designed on the basis of the lack of
stationary; the results of this test have been submitted in Table 3.
Table 3: The Results of Abnormal Earnings and Book Values stationary test
Variables Statistic sig
Abnormal earnings -6.55 0.000
Book values -2.77 0.0028#p#分頁標題#e#
According to the contents of this table, the null hypothesis was failed in a significant level of
1% and the opposite assumption is supported. Thus, the abnormal earnings and book values have
stationary.
International Research Journal of Finance and Economics - Issue 36 (2010) 68
4.2. Assumption on the Existing Forecasting Relationship
For testing this hypothesis in all models, the information related to the years of 1998-2002 was used in
step 1 for the first estimation period and in each of the next periods it has been added one year to the
estimation period in order to use the obtained information for each estimation in next steps to estimate
the firm’s value at the end of the last year of the relevant estimation period. The results associated with
this test have been displayed in Table 4. It should be noted that as a result of existing heterogeneity
among periods, EGLS method by weighing on the periods was applied in this research.
As specified in this table, the obtained coefficients are mostly significant in the levels of 1% to
10% and confirm the existing forecast relationship and models application. The noteworthy point of
this table is that since 2005, concurrent with changes in the party governing the country as well as the
problems due to generated mental environment leading to stagnancy in stock market, the coefficients
values and the models forecast power were subjected to considerable fall.
4.3. The Comparison of the Four Models Results Regarding the EFR Hypothesis
Taking a little consideration on the results shown in Table 4, we find that using this results may cause
errors arisen when comparing traditional and sign-oriented models because of the high similarity of
four models’ results. For the same purpose in this research other criteria were used. The criteria utilized
for comparing the four studied models are the Sum of Squared Residuals, Akaike and Schwarz
Criterions that in all three cases the lower criterion indicated the superior model. Based on these
criteria, the comparison results of four models during six estimation periods have been given in Table
5.
As specified in this table, the Sum of Squared Residuals for the third model in five periods and
for the fourth model in one period out of six estimation periods is lower than other models.
Consequently, with considering the Sum of Squared Residuals criterion, the third model is superior to
other models. Akaike criterion in all periods indicates the superiority of the fourth model. At last,
Schwarz criterion also introduces the fourth model as the superior model in three periods out of six
estimation periods. Finally, with taking into account all aspects above associated with the EFR
hypothesis, the fourth model can be titled as the superior model. Also, though the sign of conservatism
variable in the third as well as the fourth model – for firms having negative abnormal earnings- is#p#分頁標題#e#
unlike to the prior expectation, but Feltham-Ohlson (1995) based models have generally acted more
successfully than Ohlson (1995) based ones.
4.4. Temporary Persistence Test of Negative Abnormal Earnings
As previously explained, according to Giner and Iniguez (2006) investors assume the negative
abnormal earnings as more temporary than the positive ones. Their two reasons for this claim have
been given before. In order to study this issue in sign-oriented models of this research (second and
fourth models) Wald coefficients test was used. Since according to theory, the abnormal earnings
coefficient should be between zero and one, the more this coefficient differs from 1 (close to zero) the
more temporary will be the relevant abnormal earnings. Therefore, the null hypothesis of Wald test is
confirmed on these coefficients equaling to 1 and it is expected that this assumption is failed for
negative abnormal earnings and is confirmed for the positive type. This test results have been shown in
Table 6.
69 International Research Journal of Finance and Economics - Issue 36 (2010)
Table 4: The results of EFR hypothesis in four models
Estimation Model 1 Model 2 Model 3 Model 4
period 11 ω AR2
11 ω +
11 ω AR2 11 ω 12 ω AR2 11 ω +
11 ω 12 ω +
12 ω AR2
+
22 ω
1998-2002 0.716097* 0.80 0.885741* -0.187931** 0.80 0.2005* -32.12591* 0.73 0.652453* -0.126849 -11.00133* 16.7147* 0.77 1.253698*
1998-2003 0.720112* 0.78 0.861372* -0.171273* 0.78 0.135527* -32.27204* 0.70 0.623132* -0.120399 -10.29964* 16.57176* 0.75 1.264519*
1998-2004 0.721884* 0.74 0.872184* -0.182181* 0.74 0.125879* -32.98580* 0.62 0.5706* -0.091790 -13.00296* 19.87825* 0.71 1.245715*
1998-2005 0.016118* 0.54 0.016102* 0.091433** 0.51 -0.101156* -35.69544* 0.49 -0.092718* 0.329638* -32.77021* 44.58863* 0.57 1.239737*
1998-2006 0.016113* 0.51 0.016097* 0.091541** 0.48 -0.083822* -30.44098* 0.43 -0.0293* 0.493369* -11.18442* 17.12382* 0.57 1.245292*
1998-2007 0.0116150* 0.48 0.0116314* 0.555364* 0.52 0.038908* 6.7988** 0.73 -0.0966* 0.136475 -34.1754* 41.97902* 0.55 1.236819*
**: significant at 5%; *: significant at 1%
Table 5: Comparison of Research Models in Respect of EFR hypothesis
Estimation Sum of Squared Residuals Akaike criterion Schwarz criterion
period Model1 Model 2 Model 3 Model4 Model1 Model 2 Model 3 Model4 Model1 Model 2 Model 3 Model4
1998-2002 1.42E+16 1.38E+16 7.78E+15* 1.33E+16 33.45458 33.44088 33.78357 33.42573* 33.46944* 33.4706 33.81329 34.48515
1998-2003 1.63E+16 1.57E+16 9.15E+15* 1.53E+16 33.57183 33.55164 33.75925 33.53872* 33.58449 33.57695* 33.78456 33.58934
1998-2004 1.08E+18 1.06E+18 9.33E+17* 1.02E+18 34.6913 34.6714 34.82401 34.65403* 34.70237 34.69354* 34.84615 34.69831
1998-2005 1.19E+18 1.19E+18 9.46E+17* 9.75E+17 34.67778 34.67297 34.656 34.44694* 34.68764 34.69271 34.67574 34.48641*#p#分頁標題#e#
1998-2006 1.2E+18 1.2E+18 1.03E+18* 1.14E+18 34.818 34.81378 34.99906 34.77017* 34.82692 34.83162 35.01691 34.80586*
1998-2007 1.2E+18 1.2E+18 1.25E+18 4.38E+17* 34.80467 34.72798 34.72636 34.62259* 34.81283 34.74429 34.74267 34.65521*
*-indicating superior model
International Research Journal of Finance and Economics - Issue 36 (2010) 70
Table 6: Persistence Tests of Negative and positive Abnormal Earnings
Negative abnormal earnings persistence test
(H0:ω11 = 1)
Positive abnormal earnings persistence test
(H0: 1 11 11 Estimation ω +ω + = )
period
F statistic Sig F statistic Sig
1998-2002 2.27 0.0133 150.027 0.000
1998-2003 6.6 0.0107 156.43 0.000
1998-2004 5.62 0.0183 156.57 0.000
1998-2005 163.808 0.000 615.55 0.000
1998-2006 163.510 0.000 615.77 0.000
1998-2007 383.037 0.000 11.31 0.001
The contents of this table show that though the negative abnormal earnings are regarded
temporary- contrary to our expectation- they show a higher stability as compared to the positive type.
Perhaps, this is because of the lacking conditions assumed by Giner and Iniguez (2006) in Iran. So that,
firstly, there is no liquidation option in Iran and, secondly, the firms’ critics in Iran is mainly due to
boycott imposed on this country and there is no hope to solve it in a short time.
4.5. Review of the EVR Hypothesis
To survey this hypothesis, with using the coefficients obtained by the formation of regression relation
of each model in previous steps, and, also, with using simplified valuation functions special for each
model (as presented in section 3), it has been paid to the estimation of firm’s value each year and, then,
all estimated values related to the years from 2002 to 2007 were compared with the relevant market
prices with using the following tests.
3. T test is used to review if the means of the estimated and real values are equal.
4. Wilcoxon signed-ranks test is used to review if medians of the estimated and real values are
equal.
It is obvious that if the model could estimate the prices correctly, the results of two above tests
should indicate the equality of average and mean of market prices (P) and estimated values (V). In both
tests above, the null hypothesis is confirmed on the equity of means and/or medians. And, hence, it is
necessary to confirm the null hypothesis for both tests to support the second assumption. Table 7,
display the results obtained from T test and Wilcoxon signed-ranks test in four models.
Table 7: The Results of means and medians Equality Test
T test(H0: V P μ = μ ) Wilcoxon signed rank test(H0:medv=medp) Models
T statistic Sig Z statistic sig
1 7.414 0.000 -15.372 0.000
2 7.409 0.000 -15.618 0.000
3 0.748 0.455 -9.057 0.000
4 6.949 0.000 -14.686 0.000
The contents of this table, in most models fail the null hypothesis in the 1% significant level.#p#分頁標題#e#
So, it can be concluded that there is significant discrepancy between means and medians of market
prices and estimated values in all models (except the third model for which only the means equality
assumption-and not medians equality assumption-has been confirmed). Therefore, the EVR hypothesis
in all four studied models is failed. But, it is required to compare the valuation power of four models
studied in this research to find the superior model. For the same purpose, the comparison of the
average absolute value of the valuation errors of four models is used. Results show that traditional
models cause lower errors in the firm’s valuation and, thus, have a better function compared to signoriented
models. Second column in Table 8 illustrates this comparison totally for all estimation
periods. This table indicates the occurrence of more valuation errors when using sign-oriented models.
71 International Research Journal of Finance and Economics - Issue 36 (2010)
Moreover, the results of this table show that the third model cause less valuation errors compared to the
first model(for the aim of comparing traditional models).
In most researches involving in the review of the application of Ohlson (1995) and Feltham-
Ohlson (1995) linear models, the estimated values are usually lower than real values in most cases. In
this research, as shown in the last three columns of table 8, first model in 93%, second model in 95%,
third model in 78%, and fourth model in 96% of observations, respectively, have underestimated the
firm’s value.
Table 8: Comparative survey on the Valuation Power of Four Models
Models Average of Absolute
valuation errors
Total number of
observations Observations with V>P Observations with V<P
1 696207 348 24 324
2 731786 348 17 331
3 335740 348 75 273
4 778770 348 15 333
In view of all cases presented above, though the EVR hypothesis has not been supported in all
models, the statistical results regarding this assumption, unlike the EFR hypothesis, shows that
traditional models have a better function compared to sign-oriented models. But, with considering the
non-ability of any of the four models in firm’s valuation, and the better performance of sign-oriented
models in forecasting abnormal earnings one cannot certainly introduce a superior model in this
respect.
5. Results and Applications of Research
5.1. Results of Research
In this research, the applicability of two adjusted versions of Ohlson model (1995) and two adjusted
versions of Feltham-Ohlson model (1995) was studied to forecast abnormal earnings and firm’s value
within the framework of two studied assumptions. The first hypothesis involves in the confirmation of
the four models considered to forecast the abnormal earnings of each period by the abnormal earnings#p#分頁標題#e#
of previous period and the second one pays to the favorability of these models to estimate the firm’s
value with using discounted abnormal earnings of previous periods. These models are segregated into
two general categories of traditional and sign-oriented. The only difference of these two categories is
that the design of the sign-oriented models (second and fourth models) is so that it may be possible to
obtain different coefficients for positive and negative abnormal earnings. The reason is that according
to Giner and Iniguez (2006) the negative abnormal earnings, unlike the positive ones, have temporary
persistence and, as a result, a coefficient close to zero. Thereby, the segregation of earnings in term of
their sign, and the calculation of a separate coefficient would improve the models’ efficiency in
forecast and valuation.
The research results indicate the confirmation of models’ efficiency in the prediction of
abnormal earnings in four studied modes. This results conform to the most researches conducted
outside of Iran, i.e. Ohlson (1995), Ota (2002), McCrae and Nilsson (2001), Choi et al.(2001) and
Callen & morel(2001) and conflict with the results of research made by Callen et al. (2002). And, also,
comply with the results of research made by Khodadadi et al. (2005) involving in the review of linear
information models in Iranian firms. Nevertheless, similar to the result of research made by Myers
(1999), none of these models have succeeded to forecast the value of firms listed on Tehran Stock
Exchange. The ground of this non achievement is not the weakness of these models, because stock
prices may have no scientific and theoretic backup in Tehran Stock Exchange. Since the studied
models’ values in most cases are underestimated (similar to the results of researches made by Myers
(1999), Dechow et al. (1998) and Frankel and Lee (1998)) the possibility of the existence of price
International Research Journal of Finance and Economics - Issue 36 (2010) 72
bubble in Tehran Stock Exchange Organization is not far-reaching. The study of above-mentioned
cases, in turn, requires more empirical researches.
Also, results obtained from the comparison of four studied models of this research indicate the
superiority of sign-oriented models to traditional models to forecast abnormal earnings. This result is a
support to the claim of Giner and Iniguez (2006), On the other hand, though the results show a better
function of traditional models in the firm valuation process, the determination of the superior model is
not simply possible in this respect because of non-ability of all four models in the estimation of firm’s
value. Making decision in this respect requires more researches to be made in which it will be paid to
the comparison of traditional and sign-oriented models while making linear models agreeable with#p#分頁標題#e#
inflation conditions of Iran. In such a condition, perhaps will using adjusted book values based on
inflation rate and using them in valuation models be resulted in the better function of sign-oriented,
models compared to the traditional models in the valuation, and support the relevant theories.
5-2- Application of Research Results
The research results will be applicable for the following real and legal persons:
A Investors: it is expected that the research results can indicate a model describing the investors’
stock theoretic value. The results help them making better decisions and, thereby, cause
increased investment return.
B Firms’ Managers: the firms’ stock price in financial markets is focused by most managers.
Since this is one of their assessment factors in shareholder’s viewpoint, achieving a model
segregating the stock unsubstantial price from the prices having theoretical backup will provide
a better criterion for the assessment made by shareholders and managers defending of their
operation.
C Tehran Stock Exchange Organization: Determination of scientific and correct values to prevent
the formation of price bubble and, consequently, sudden stagnancy in market is important to the
Stock Exchange Organization. Therefore, it is predicted that this research testing the theoretic
model to calculate the price in Iran may be utilized by the said organization.
73 International Research Journal of Finance and Economics - Issue 36 (2010)
References
[1] Barth, M. Elliott, J. and Finn, M. (1999). ‘Market rewards associated with patterns of
increasing earnings’. Journal of Accounting Research, 37 (autumn): 387-413.
[2] Bauman, M. P.(1999).‘ An empirical investigation of conservatism in book value measurement,
Managerial Finance’. 25 (12):42-54.
[3] Belkaoui, A.(2004).‘ Accounting Theory’. Thomson.
[4] Callen, J.L. Livnat, J. and Segal, D.(2002).‘ Accounting Restatements: Are they always bad
news?’. Working paper, University of Toronto.
[5] Callen, J.L. and Morel, M. (2001).‘ Linear accounting valuation when abnormal earnings are
AR(2)’. Review of Quantitative Finance and Accounting, 16:191-203.
[6] Choi, Y. O'Hanlon, J. and Pope, P. F. (2001).‘ Linear information models in residual incomebased
valuation: a development of the Dechow, Hutton and Sloan empirical approach’.
Working paper, Lancaster University.
[7] Dechow, P. Kothari, S. P. and Watts, R. L. (1998).‘ The relation between earnings and cash
flows’. Journal of Accounting and Economics: 133–168.
[8] Dechow, P. M. Hutton, A.P. and Sloan, R.G.(1999).‘ An empirical assessment of the residual
income valuation model’. Journal of Accounting and Economics, 26:1-34.
[9] Edwards, E.O. and Bell, P.W. (1961).‘ The Theory and Measurement of Business Income’.#p#分頁標題#e#
University of California Press: Berkeley.
[10] Feltham, G. A. and Ohlson, J.A. (1995).‘Valuation and clean surplus accounting for operating
and financial activities’. Contemporary Accounting Research, 11 (2):689-731.
[11] Frankel, R. and Lee, C.M.C.(1998).‘accounting valuation, market expectation, and crosssectional
returns’. Journal of Accounting and Economics, 25:283-319.
[12] Giner, B. and Iniguez, R.(2006).‘An Empirical Assessment of the Feltham-Ohlson Models
considering the sign of abnormal earnings’. Accounting and Business Research, 36(3):169-190.
[13] Graham R. and king, R. (2000).‘ The Relation of Firm Market Values with Book Values and
留學生dissertation網Residual Accounting Earnings in Six Asian Countries’. Working paper, Oregon State
University.
[14] Habibi, H. Sajadi, S.H. and Farazmand, H.(2006).‘ Factors effective on income management in
firms listed on Tehran Stock Exchange’. Master’s thesis, Faculty of Economics and Social
Sciences, Shahid Chamran University of Ahwaz.
[15] Hand, J. and Landsman, W. (1998).‘ Testing the Ohlson Model: v or not v. That Is the
Question’. Working Paper, University of North Carolina-Chapel Hill.
[16] Hand, J. and Landesman, W. (1999).‘ The Pricing of Dividends in Equity Valuation’. Working
paper, University of North California.
[17] Karathanasis, G.A and Spilioti, S.N. (2005).‘An Empirical Application of The Clean Surplus
Valuation Models: the case of the Athens Stock Exchange, Applied Financial Economics
,15:1031-1036.
[18] Khodadadi, V. Dastgir, M. Norvash, I. and Momeni, M. (2005).‘ Design of Linear Information
Model in Tehran Stock Exchange:Expansion of Ohlson model’. PhD thesis, Faculty of
Management, Tehran University.
[19] Lee, C. M. C. Myers, J. and Swaminathan, B. (1999).‘ What is the Intrinsic Value of the
Dow?’. Journal of Finance, 54(4):1693 - 1742.
[20] Lo, K. and Lys, T. (2000).‘ The Ohlson model: contribution to valuation theory, limitations,
and empirical applications’. Journal of Accounting, Auditing and Finance, 15 (3):337-370.
[21] McCrae, M. and Nilsson, H. (2001).‘ The explanatory and predictive power of different
specifications of the Ohlson (1995) valuation models’. The European Accounting Review, 10
(2):315-341.
[22] Myers, J. N. (1999).‘ Implementing residual income valuation with linear information
dynamics’. The Accounting Review,74:1-28.
International Research Journal of Finance and Economics - Issue 36 (2010) 74
[23] Ohlson, J. A. (1995).‘ Earnings, book values, and dividends in equity valuation’. Contemporary
Accounting Research, 11 (2):661-687.
[24] Ota, K. (2002).‘A test of the Ohlson (1995) model: empirical evidence from Japan’. The#p#分頁標題#e#
International Journal of Accounting, 37 (2):157-182.
[25] Peasnell, K.V. (1982).‘ Some formal connections between economic values and yields and
accounting numbers’. Journal of Business Finance and Accounting, 9:361-381.
[26] Preinrich, G.A.D. (1983).‘ Annual survey of economic theory: the theory of depreciation’.
Econometrica, 6:219-241.
[27] Stober, T. L. (1996).‘ Do prices behave as if accounting book values are conservative? Crosssectional
tests of the Feltham-Ohlson (1995) valuation model’. Working paper, University of
Notre Dame.
[28] Zaranejad, M. and Anvari E. (2005).‘ Application of Panel Data in Econometrics’. Faculty of
http://www.400190.com/dissertation_writing/Economics and Social Sciences, Shahid Chamran University of Ahwaz publications, journal of
economic researches, Vol. 2, No. 4:21-51.