Journal of Econometrics 101 (2001) 1}23

Tests for the error component model in the

presence of local misspeci"cation
Anil K. Bera *, Walter Sosa-Escudero
, Mann Yoon
Department of Economics, University of Illinois, 1206 S. Sixth Street, Champaign, IL 61820, USA

Department of Economics, National University of La Plata, Calle 48 No. 555, Of. 516,
(1900) La Plata, Argentina
Department of Economics and Statistics, California State University at Los Angeles,
5151 State University Drive, Los Angeles, CA 90032, USA
Received 1 February 1998; received in revised form 1 March 2000; accepted 5 July 2000

Abstract
It is well known that most of the standard speci"cation tests are not valid when the
alternative hypothesis is misspeci"ed. This is particularly true in the error component
model, when one tests for either random e!ects or serial correlation without taking
account of the presence of the other e!ect. In this paper we study the size and power of
the standard Rao's score tests analytically and by simulation when the data are contaminated by local misspeci"cation. These tests are adversely a!ected under misspeci"cation. We suggest simple procedures to test for random e!ects (or serial correlation) in the
presence of local serial correlation (or random e!ects), and these tests require ordinary
least-squares residuals only. Our Monte Carlo results demonstrate that the suggested
tests have good "nite sample properties for local misspeci"cation, and in some cases even
for far distant misspeci"cation. Our tests are also capable of detecting the right direction
of the departure from the null hypothesis. We also provide some empirical illustrations to
highlight the usefulness of our tests.  2001 Elsevier Science S.A. All rights reserved.
JEL classixcation: C12; C23; C52
Keywords: Error component model; Testing; Random e!ects; Serial correlation; Local
misspeci"cation

* Corresponding author. Tel.: #1-217-333-4596; fax: #1-217-244-6678.

E-mail address: anil@"sher.econ.uiuc.edu (A.K. Bera).
0304-4076/01/$ - see front matter  2001 Elsevier Science S.A. All rights reserved.
PII: S 0 3 0 4 - 4 0 7 6 ( 0 0 ) 0 0 0 7 1 - 3

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

1. Introduction
The random error component model introduced by Balestra and Nerlove
(1966) was extended by Lillard and Willis (1978) to include serial correlation in
the remainder disturbance term. Such an extension, however, raises questions
about the validity of the existing speci"cation tests such as the Rao's (1948) score
(RS) test for random e!ects assuming no serial correlation as derived in Breusch
and Pagan (1980). In a similar way doubts could be raised about tests for serial
correlation derived assuming no random e!ects. Baltagi and Li (1991) proposed
a RS test that jointly tests for serial correlation and random e!ects. One
problem with the joint test is that, if the null hypothesis is rejected, it is not
possible to infer whether the misspeci"cation is due to serial correlation or to
random e!ects. Also, as we will discuss later, because of higher degrees of
freedom the joint test will not be optimal if the departure from the null occurs
only in one direction. More recently, Baltagi and Li (1995) derived RS statistics
for testing serial correlation assuming "xed/individual e!ects. These tests require maximum-likelihood estimation of individual e!ects parameters.
For a long time econometricians have been aware of the problems that arise
when the alternative hypothesis used to construct a test deviates from the
data-generating process (DGP). As emphasized by Haavelmo (1944, pp. 65}66),
in testing any economic relations, speci"cation of a given "xed set of possible
alternatives, called the priori admissible hypothesis, X, is of fundamental
importance. Misspeci"cation of the priori admissible hypotheses was termed as
type-III error by Bera and Yoon (1993). Welsh (1996, p. 119) also pointed out
a similar concept in the statistics literature. Typically, the alternative hypothesis
may be misspeci"ed in three di!erent ways. In the "rst one, which we shall call
&complete misspeci"cation', the set of assumed alternatives, X, and the DGP,
X, say, are mutually exclusive. This happens, for instance, if one tests for serial
independence when the DGP has random individual e!ects but no serial
dependence. The second case occurs when the alternative is underspeci"ed in
that it is a subset of a more general model representing the DGP, i.e., XLX.
This happens, for example, when both serial correlation and individual e!ects
are present, but are tested separately (one at a time). The last case is &overtesting'
which results from overspeci"cation, that is, when XMX. This can happen
when, say, Baltagi and Li (1991) joint test for serial correlation and random
individual e!ects is used when only one e!ect is present. (For a detailed
discussion of the concepts of undertesting and overtesting, see Bera and Jarque,
1982.) In this paper, we study analytically the asymptotic e!ects of misspeci"cations on the one-directional and joint tests for serial dependence and random
individual e!ects. These results compliment the simulation results of Baltagi and
Li (1995). Then, applying the modi"ed RS test developed by Bera and Yoon
(1993), we derive a test for random e!ects (serial correlation) in the presence of
serial correlation (random e!ects). Our tests can be easily implemented using

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

ordinary least-squares (OLS) residuals from the standard linear model for panel
data. Our testing strategy is close to that of Hillier (1991) in the sense that we try
to partition an overall rejection region to obtain evidence about the direction (or
directions) in which the model needs revision.
The plan of the paper is as follows. In the next section we review a general
theory of the distribution and adjustment of the standard RS statistic in the
presence of local misspeci"cation. In Section 3, the general results are specialized
to the error component model. In Section 4, we present two empirical illustrations. Section 5 reports the results of an extensive Monte Carlo study. These
results, along with the empirical examples, clearly demonstrate the inappropriateness of one-directional tests in identifying the speci"c source of misspeci"cation(s), and highlight the usefulness of our adjusted tests. Section 6 provides
some concluding remarks.

2. E4ects of misspeci5cation and a general approach to testing in the presence of

a nuisance parameter
Consider a general statistical model represented by the log-likelihood
(c, t,
). Here, the parameters t and
are taken as scalars to conform with
our error component model, but in general they could be vectors. Suppose an
investigator sets
"
and tests H : t"t using the log-likelihood function



(c, t)"(c, t,
), where
and t are known values. The RS statistic for




testing H in (c, t) will be denoted by RS . Let us also denote h"(c, t,
)


R
and hI "(c, t ,
 ), where c is the maximum-likelihood estimator (MLE) of
 
c when t"t and
"
. The score vector and the information matrix are


de"ned, respectively, as
*(h)
d (h)"
?
*a

for a"c, t,

and

J
A
1 *(h)
J(h)"!E
" J
RA
n *h*h
J
(A

J
J
J

AR
R
(R

J
A(
J
,
R(
J
(

where n denotes the sample size. If (c, t) were the true model, then it is well

known that under H : t"t ,


1
" s (0),
RS " d (hI )J\ (hI )d (hI ) P
R n R
RA
R


A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

" denotes convergence in distribution and J (h),J "J !

where P
RA
RA
R
J J\J . Under H : t"t #m/(n,
RA A AR


" s (j ),
RS P
(1)
R
 
where the noncentrality parameter j is given by j ,j (m)"mJ m. Given



RA
this set-up, asymptotically the test will have correct size and will be locally
optimal. Now suppose that the true log-likelihood function is (c,
) so that

the alternative (c, t) becomes completely misspeci"ed. Using a sequence of

local values
"
#d/(n, Davidson and MacKinnon (1987) and Saikkonen

(1989) obtained the asymptotic distribution of RS under (c,
) as
R

" s (j ),
RS P
R
 

(2)

where the non-centrality parameter j

is given by j ,j (d)"



dJ
J\ J
d with J
"J !J J\J . Due to this noncentrality
(RA RA R(A
R(A
R(
RA A A(
parameter, RS will have power in the model (c, t,
) even when t"t ; and,
R

therefore, the test will have incorrect size. Notice that the crucial quantity is
J
which can be interpreted as the partial covariance between d and d after
R(A
R
(
eliminating the e!ect of d on d and d . If J
"0, then the local presence of
A
R
(
R(A
the parameter
has no e!ect on RS .
R
Turning now to the case of underspecixcation, let the true model be represented by the log-likelihood (c, t,
). The alternative (c, t) is now underspeci
"ed with respect to the nuisance parameter
, leading to the problem of
undertesting. In order to derive the asymptotic distribution of RS under the
R
true model (c, t,
), we again consider the local departures
"
#d/(n

together with t"t #m/(n. It can be shown that (see Bera and Yoon, 1991)

" s (j ),
RS P
R
 

(3)

where
d#J m)
j ,j (m, d)"(dJ
#mJ )J\ (J


(RA
RA RA R(A
RA
"j (m)#j (d)#2mJ
d.


R(A
Using this result, we can compare the asymptotic local power of the underspeci"ed test with that of the optimal test. It turns out that the contaminated
noncentrality parameter j (m, d) may actually increase or decrease the power

depending on the con"guration of the term mJ
d.
R(A
The problem of overtesting occurs when multi-directional joint tests are
applied based on an overstated alternative model. Suppose we apply a joint test
for testing hypothesis of the form H : t"t and
"
using the alternative



model (c, t,
). Let RS be the joint RS test statistic for H . To "nd the
R(


A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

asymptotic distribution of RS under overspeci"cation, i.e., when the DGP is

R(
represented by the log-likelihood either (c, t) or (c,
), let us consider the


following result, which could be obtained from (1) by replacing t with [t,
].
Assuming correct speci"cation, i.e., under the true model represented by
(c, t,
) with t"t #m/(n and
"
#d/(n,


" s (j ),
RS P
R(
 

(4)

where

 

J
J
m
R(A
j ,j (m, d)"[m d] RA
.


J
J
d
(RA
(A
Using this fact, we can easily "nd the asymptotic distribution of the overspeci"ed test. Consider testing H : t"t and
"
in (c, t,
) where (c, t)




represents the true model. Under (c, t) with t"t #m/(n, we obtain by


setting d"0 in (4)
" s (j ),
RS P
R(
 

(5)

where j ,j (m)"mJ m.


RA
Note that the noncentrality parameter j (m) of the overspeci"ed test RS is

R(
identical to j (m) of the optimal test RS in (1). Although j "j , some loss of

R


power is to be expected, as shown in Das Gupta and Perlman (1974), due to the
higher degrees of freedom of the joint test RS .
R(
Using result (2), Bera and Yoon (1993) suggested a modi"cation to RS so that
R
the resulting test is valid in the local presence of
. The modi"ed statistic is given by
1
RSH" [d (hI )!J
(hI )J\ (hI )d (hI )][J (hI )!J
(hI )J\ (hI )J
(hI )]\
R n R
R(A
(A
(
RA
R(A
(A
(RA
[d (hI )!J
(hI )J\ (hI )d (hI )].
(6)
R
R(A
(A
(
This new test essentially adjusts the mean and variance of the standard RS .
R
Bera and Yoon (1993) proved that under t"t and
"
#d/(n RSH has


R
a central s distribution. Thus, RSH has the same asymptotic null distribution as

R
that of RS based on the correct speci"cation, thereby producing an asympR
totically correct size test under locally misspeci"ed model. Bera and Yoon (1993)
further showed that for local misspeci"cation the adjusted test is asymptotically
equivalent to Neyman's C(a) test and, therefore, shares the optimality properties
of the C(a) test. There is, however, a price to be paid for all these bene"ts. Under
the local alternatives t"t #m/(n

" s (j ),
RSHP
R
 

(7)

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

where j ,j (m)"m(J !J
J\ J
)m. Note that j !j *0, where


RA
R(A (A (RA


j is given in (1). Result (7) is valid both in the presence or absence of the local

misspeci"cation
"
#d/(n, since the asymptotic distribution of RSH is

R
una!ected by the local departure of
from
. Therefore, RSH will be less

R
powerful than RS when there is no misspeci"cation. The quantity
R
j "j !j "mJ
J\ J
m
(8)



R(A (A (RA
can be regarded as the premium we pay for the validity of RSH under local
R
misspeci"cation. Two other observations regarding RSH are also worth noting.
R
First, RSH requires estimation only under the joint null, namely t"t and
R


"
. Given the full speci"cation of the model (c, t,
) it is, of course,

possible to derive a RS test for t"t in the presence of
. However, that

requires MLE of
which could be di$cult to obtain in some cases. Second,
when J
"0, RSH"RS . In practice this is a very simple condition to check.
R(A
R
R
As mentioned earlier, if this condition is true, RS is an asymptotically valid test
R
in the local presence of
.

3. Tests for error component model

We consider the following one-way error component model introduced by
Lillard and Willis (1978), which combines random individual e!ects and "rstorder autocorrelation in the disturbance term
y "x b#u , i"1, 2, 2, N, t"1, 2, 2, ,
GR
GR
GR
u "k #l ,
GR
G
GR
l "ol
#e , "o"(1,
(9)
GR
GR\
GR
where b is a (k;1) vector of parameters including the intercept,
k &IIDN(0, p) is a random individual component, and e &IIDN(0, p). The
G
I
GR
C
k and l
are assumed to be independent of each other with
G
GR
l &N(0, p/(1!o)). N and denote the number of individual units and the
G
C
number of time periods, respectively. For the validity of the tests discussed here,
we need to assume that the regularity conditions of Andereson and Hsiao (1982)
are satis"ed. Also, testing for p involves the issue of the parameter being at the
I
boundary. Although for the nonregular problem of testing at the boundary
value, both the likelihood ratio and Wald test statistics do not have their usual
asymptotic chi-squared distribution, the RS test statistic does [see, e.g., Bera
et al., 1998].
Let us set h"(c, t,
)"(p, p, o). Consider the problem of testing for the
C I
existence of the random e!ects (H : t"0) in the presence of serial correlation

(
O0). To derive our RSH, which will now be denoted as RSH, we note that it is
R
I
su$cient to consider the scores and the information matrix evaluated at

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

h "(c , t ,
)"(p, 0, 0) because of the block-diagonality of the informa
  
C
tion matrix involving the b and h parameters. These quantities have been
derived in Baltagi and Li (1991):
*
N
uu
"d "!
#
,
A
*p
2p 2p
C
C
C
*
N
u(I e e )u
2 2
"d ,d "!
1! ,
,
I
R
*p
2p
uu
I
C
*k
uu
\ ,
"d ,d "N
M
(
*o
uu



(10)

where I is an identity matrix of dimension N, e is a vector of ones of

,
2
dimension , u"(u ,2, u ,2, u ,2, u ) and u is an (N;1) vector

2
,
,2
\
containing u
. To simplify notation, here the score for the parameter p is
GR\
I
denoted as d . We will continue to follow this convention for the elements of the
I
information matrix and for expressing our test statistics. Denoting
J"(N)\E(!*/*h*h) evaluated at h , we have


1
J"
1

2p
C
2(!1)p
C
0

This implies that

2(!1)p
C .

2(!1)p
C

!1
J
"J
"
,
IMA
R(A
p
C
!1
J "J "
,
IA
RA
2p
C
!1
J "J "
,
MA
(A

(11)

where c stands for the parameter p. Since J

'0, indicating the asymptotic
C
IMA
positive correlation between the scores d and d , the one-directional test for the
I
M
random e!ects reported in Breusch and Pagan (1980) is not valid asymptotically
in the presence of serial correlation. For this case our RSH can be easily
I
constructed, from Eq. (6), as
N(A#2B)
RSH"
,
I 2(!1)(1!(2/))

(12)

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

where A and B denote, as in Baltagi and Li (1991),

u (I e e )u
2 2
A"1! ,
u u
and
u u
B" \ .
u u
Note that u are the OLS residuals from the standard linear model
y "x b#u without the random e!ects and serial correlation. Also notice
GR
GR
GR
that A and B are closely related to the estimates of the scores d and d ,
I
M
respectively. It is easy to see that the RSH adjusts the conventional RS statistic
I
given in Breusch and Pagan (1980), i.e.,
NA
RS "
,
I 2(!1)

(13)

by correcting the mean and variance of the score d for its asymptotic correlaI
tion with d .
M
To see the behavior of RS let us "rst consider the case of complete misspeci"I
cation, i.e., p"0 but oO0. Using (2) and (11), the noncentrality parameter of
I
RS for this case is
I
!1
j (o)"dJ
J\ J
d"2o
,
(14)

MIA IA IMA

where for simplicity we use o in place of d. In this case, the use or RS will lead to
I
rejection of the null hypothesis p"0 too often. For local departures RSH will
I
I
not have this drawback when oO0 since under p "0, RSH will have a central
I
I
s distribution. Let us now consider the underspeci"cation situation i.e., when
we have both p'0 and oO0, and we use RS to test H : p "0. From (1), (3)
I
I
 I
and (11), we see that the change in the noncentrality parameter of RS due to
I
nonzero o is given by
j (m, d)!j (m)"j (o)#2mJ
d"o



IMA

2(!1)
!1
#2p o
I

p
C

2(!1) o po
"
# I ,
(15)

p
C
where we use p in place of m. From (15), it is easy to see that when o'0, the
I
presence of autocorrelation will add power to RS ; but when o(0 it can
I
loose power if the individual e!ect is very high and p is low. In this situation,
C
the noncentrality parameter of RSH is not a!ected. From (7) and (11), the
I

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

noncentrality parameter of RSH under p '0 and oO0, can be written as

I
I
j ,j (p)"p(J !J
J\ J
)

 I
I IA
IMA MA MIA
!1 (!1)
"p
!
I 2p
p !1
C
C
1
p
1
" I (!1) ! ,
(16)
p
2
C
which does not depend on o. There is, however, a cost in applying RSH when o is
I
indeed zero. From (8) the cost is

p !1
.
(17)
j ,j (p)"pJ
J\ J
" I

 I
I IMA MA MIA p
C
Note that this cost is present only under p '0. That is, there is a cost only in
I
terms of the power of RSH; the size is una!ected. Later we will provide an
I
interesting interpretation of this cost of RSH in terms of the behavior of the
I
unadjusted test RS under p '0.
M
I
As mentioned before, Baltagi and Li (1995) derived a RS test for serial
correlation in the presence of random individual e!ects. Naturally, the test
requires MLE of p. Our procedure gives a simple test for serial correlation in
I
the random e!ects model. In this situation RSH is obtained simply by switching
M
p and o to yield
I
N(B#(A/))
RSH"
.
(18)
M (!1)(1!(2/))
If we assume that the random e!ects are absent throughout, then RSH in (18)
M
reduces to
NB
RS "
.
M
!1

(19)

This conventional RS statistic (19) is also given in Baltagi and Li (1991).

As we have done for RS , we can also study the performance of RS under
I
M
various misspeci"cations. When there is complete misspeci"cation, i.e., when
o"0 but p '0, the noncentrality parameter of RS is
I
M
p !1
j (p )"mJ
J\ J
m" I
,
(20)
 I
IMA MA MIA
p
C
where we have used p in place of m. Therefore, RS will reject H : o"0 too
I
M

often when p '0. Similarly, when there is underspeci"cation, i.e, oO0 with
I
p'0, the change in the noncentrality parameter due to the presence of the
I

10

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

random e!ect, is
j (m, d)!j (d)"j (p )#2dJ
m


 I
MIA
!1 p p
I I #2o .
"
(21)
p p
C C
Therefore, we have an increase in (or a possible loss of ) power when o'0 (or
o(0). The noncentrality parameter of RSH will not be a!ected at all under
M
p'0. On the other hand, we do, however, pay a penalty when p "0 and we
I
I
use the adjusted test RSH. The penalty is
M
!1
j (o)"oJ
J\ J
"2o
.
(22)

MIA IA IMA


Due to this factor the power of RSH will be somewhat less than that of RS when
M
M
p is indeed zero; the size of RSH, however, remains una!ected. It is very
I
M
interesting to note that
j (o)"j (o),


given in (14). Similarly, from (17) and (20)

(23)

j (p )"j (p).

(24)
 I
 I
An implication of (23) is that the cost of using RSH when p "0 is the same as
M
I
the cost of using RS when oO0. Similarly, (24) implies that the loss in the
I
noncentrality parameter of RSH when o"0 is equal to the unwanted gain in the
I
noncentrality parameter of RS when p '0. We will explain these seemingly
M
I
unintuitive phenomena after we "nd a relationship among the four statistics,
RSH, RS , RSH, and RS . It should be noted that the equalities of Eqs. (23) and
I
I
M
M
(24) are not speci"c for the error component model, and they hold in general.
This can be seen by comparing j (d) below (2) with j in Eq. (8), where t swaps


position with
and m is replaced by d.
Baltagi and Li (1991, 1995) derived a joint RS test for serial correlation and
random individual e!ects which is given by
N
RS "
[A#4AB#2B].
IM 2(!1)(!2)

(25)

Under the joint null p "o"0, RS is asymptotically distributed as s . Use of

I
IM

this will result in a loss of power compared with the proper one-directional tests
when only one of the two forms of misspeci"cation is present, as we noted while
discussing (5). For example, when o"0 and p'0, the noncentrality paraI
meter of both RS and RS is (see (1) and (5)).
I
IM
p !1
j (p )"p J " I
.
(26)
 I
I IA p
2
C

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

11

Since for RS and RS we will use, respectively, s and s critical values, RS

I
IM


IM
will be less powerful. An interesting result follows from (12), (13), (18), (19) and
(25), namely,
RS "RSH#RS "RS #RSH,
(27)
IM
I
M
I
M
i.e., the two directional RS test for p and o can be decomposed into the sum of
I
the adjusted one-directional test of one type of alternative and the unadjusted
form for the other one. Using (27) we can easily explain some of our earlier
observations. First, consider the identities in (23) and (24). From (27), we have
RS !RSH"RS !RSH.
(28)
M
M
I
I
Let us consider the case of p"0 and oO0. Then the left-hand side of (28)
I
represents the &penalty' of using RSH (instead of RS ) while the right-hand side
M
M
amounts to the &cost' of using RS . Eq. (28) implies that these penalty and cost
I
should be the same, as noted in (23). A reverse argument explains (24). Secondly,
the local presence of o (or p ) has no e!ect on RSH (or RSH); therefore, from (5)
I
I
M
and (27), we can clearly see why the noncentrality parameter of RS will be
IM
equal to that of RS (or RS ) when p"0 (or o"0).
M
I
I
So far we have considered only two-sided tests for H : p"0. Since p*0, it
 I
I
is natural to consider one-sided tests, and it is expected that it will lead to more
powerful tests. Within our framework, it is easy to construct appropriate
one-sided tests by taking the signed square root of our earlier two-sided
statistics, RS and RSH. We will denote these one-sided test statistics as RSO
I
I
I
and RSOH, and they are given by
I
N
A
RSO "!
I
2(!1)

and

N
(A!2B).
RSOH"!
I
2(!1)(1!2/)
The negative sign is due to the fact that */*p"!(N/2p)A and the
I
C
one-sided tests are based on this score function or its adjustment. Under
H : p"0, the adjusted test RSOH will be asymptotically distributed as N(0, 1).
 I
I
The unadjusted RSO will be asymptotically normal but with a nonzero mean
I
(2(!1)/o when oO0 as can be seen from (14). The statistic RSOH was
I
"rst suggested by Honda (1985) and its "nite sample properties have been
investigated by Baltagi et al. (1992). Similar one-sided versions for RS and RSH
M
M
can also be used. However, in practice, the direction of serial correlation is rarely
known for sure for the one-directional tests to be more powerful. It is easy to see
that the one directional tests will not satisfy the equality in (28). In our empirical
illustrations below and in the Monte Carlo study we also use these one-sided
tests and study their comparative "nite sample performance.

12

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

4. Empirical illustrations
In this section we present two empirical examples that illustrate the usefulness
of the proposed tests. The "rst is based on a data set used by Greene (1983,
2000). The equation to be estimated is a simple, log-linear cost function:
ln C "b #b ln R #u ,
GR


GR
GR
where R is measured as output of "rm i in year t in millions of kilowatt-hours,
GR
and C is the total generation cost in millions of dollars, i"1, 2, 2, 6, and
GR
t"1, 2, 3, 4. The second example is based on the well-known Grunfeld (1958)
and Grunfeld and Griliches (1960) investment data set for "ve US manufacturing "rms measured over 20 years which is frequently used to illustrate panel
issues. It has been used in the illustration of misspeci"cation tests in the errorcomponent model in Baltagi et al. (1992), and in recent books such as those by
Baltagi (1995, p. 20) and Greene (2000, p. 592). The equation to be estimated is
a panel model of "rm investment using the real value of the "rm and the real
value of capital stock as explanatory variables:
I "b #b F #b C #u ,
GR

 GR
 GR
GR
where I denotes real gross investment for "rm i in period t, F is the real value
GR
GR
of the "rm and C is the real value of the capital stock, i"1, 2, 2, 5, and
GR
t"1, 2, 2, 20.
We estimated the parameters of both models by OLS and implemented the
following seven tests based on OLS residuals: the Breusch}Pagan test for
random e!ects (RS ), the proposed modi"ed version (RSH), the LM serial
I
I
correlation test (RS ), the corresponding modi"ed version (RSH), the joint test
M
M
for serial correlation and random e!ects (RS ), and the two one-sided tests for
IM
random e!ects (RSO and RSOH). The test statistics for both examples are
I
I
presented in Table 1; the p-values are given in parentheses.
All of the test statistics were computed individually, and the equality in (27) is
satis"ed for both data sets. In the example based on Greene's data the unmodi"ed tests for serial correlation (RS ) and for random e!ects (RS to some extent,
M
I
and RSO quite strongly) reject the respective null hypothesis of no serial
I
correlation and no random e!ects, and the omnibus test rejects the joint null.
But our modi"ed tests suggest that in this example the problem seems to be
serial correlation rather the presence of both e!ects. For Grunfeld's data,
applications of our modi"ed tests point to the presence of the other e!ect. The
unmodi"ed tests soundly reject their corresponding null hypotheses. The modi"ed versions of the random e!ect tests (RSH and RSOH) also reject the null but
I
I
the modi"ed serial correlation test (RSH) barely rejects the null at the 5%
M
signi"cance level. It is interesting to note the substantial reduction of the
autocorrelation test statistic, from 73.351 to 3.712. So in this example the
misspeci"cation can be thought to come from the presence of random e!ects

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

13

Table 1
Empirical illustration. Tests for random e!ects and serial correlation
Data
Greene
Grunfeld

RS
I

RSH
I

RS
M

RSH
M

RS
IM

RSO
I

RSOH
I

5.872
(0.015)

0.269
(0.604)

15.569
(0.000)

9.966
(0.002)

15.838
(0.000)

2.423
(0.007)

0.518
(0.3020)

453.822
(0.000)

384.183
(0.000)

73.351
(0.000)

3.712
(0.054)

457.535
(0.000)

21.303
(0.000)

19.605
(0.000)

p-values are given in parenthesis.

rather than serial correlation. As expected, the joint test statistic is highly
signi"cant.
In spite of the small sample size of the data sets, these examples seem to
illustrate clearly the main points of the paper: the proposed modi"ed versions of
the test are more informative than a test for serial correlation or random e!ect
that ignores the presence of the other e!ect. In the "rst case, serial correlation
spuriously induces rejection of the no-random e!ects hypothesis, and in the
second case the opposite happens: the presence of a random e!ect induces
rejection of the no-serial correlation hypothesis. The joint test RS rejects the
IM
joint null but is not informative about the direction of the misspeci"cation.
RS provides a correct measure of the joint e!ects of individual component
IM
and serial correlation. The main problem is how to decompose this measure
to get an idea about the true departure(s). From a practical standpoint if
RS "RS #RS does not hold, that should be an indication of the presence
IM
I
M
of an interaction between random e!ects and serial correlation; and the unadjusted statistics RS and RS will be contaminated by the presence of other
I
M
departures. For example, for the Grunfeld data
RS #RS !RS "RS !RSH"RS !RSH"69.638.
I
M
IM
I
I
M
M
This provides a measure of the interaction between p and o, and is also equal to
I
the correction needed for each unadjusted test.
It is important to emphasize that the implementation of the modi"ed tests is
based solely on OLS residuals. It could be argued that a more e$cient test
procedure could be based on the estimation of a general model that allows for
both serial correlation and random e!ects, and then the tests of the hypotheses
of no-serial correlation and no-random e!ects as restrictions on this general
model (either jointly or individually) could be carried out. But this would require
the maximization of a likelihood function whose computational tractability is
substantially more involved than computing simple OLS residuals. Hsiao (1986,
p. 55) commented that the `computation of the MLE is very complicateda. For
more on the estimation issues of the error component model with serial correlation see Baltagi (1995, pp. 18}19), Majunder and King (1999) and Phillips (1999).

14

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

5. Monte Carlo results

In this section we present the results of a Monte Carlo study to investigate the
"nite sample behavior of the tests. To facilitate comparison with existing results
we follow a structure similar to the one adopted by Baltagi et al. (1992) and
Baltagi and Li (1995).
The model was set as a special case of (9):
y "a#bx #u , i"1, 2, 2, N, t"1, 2, 2, ,
GR
GR
GR
u "k #v ,
GR
G
GR
v "ov
#e , "o"(1,
GR
GR\
GR
where a"5 and b"0.5. The independent variable x was generated following
GR
Nerlove (1971):
x "0.1t#0.5x
#u ,
GR
GR\
GR
where u has the uniform distribution on [!0.5, 0.5]. Initial values were
GR
chosen as in Baltagi et al. (1992). Let p, p, p and p represent the variances of
I T
C
u , k , v and e , respectively, and let q"p /p, which represents the `strengtha
GR G GR
GR
I
of the random e!ects. Here, p"p#p, and we set p"20.q and o were
I
T
allowed to take seven di!erent values (0, 0.05, 0.1, 0.2, 0.4, 0.6, 0.8), and three
di!erent sample sizes (N, ) were considered: (25, 10), (25, 20) and (50, 10). Since
for each i, v follows an AR(1) process, p"p/(1!o). Then, according to this
GR
T
C
structure, the random e!ect term and the innovation were generated as
k &IIDN(0, 20(1!q)),
G
e &IIDN(0, 20(1!q)(1!o)).
GR
For each sample size the model described above was generated 1000 times
under di!erent parameter settings. Therefore, the maximum standard errors of
the estimates of the size and powers would be (0.5(1!0.5)/1000K0.015. In
each replication the parameters of the model were estimated using OLS, and
seven test statistics, namely, RS , RSH, RS , RSH, RS , RSO and RSOH were
I
I
M
M
IM
I
I
computed. The tables and graphs are based on the nominal size of 0.05. Our
simulation study was quite extensive; we carried out experiments for all possible
parameter combinations for the three sample sizes. We present here only
a portion of our extensive tables and graphs; the rest is available from the
authors upon request.
Calculated statistics under q"o"0 were used to estimate the empirical sizes
of the tests and to study the closeness of their distributions to s through Q}Q
plots and the Kolmogorov}Smirnov test. From Table 2 we note that both RS
I
and RSH have similar empirical sizes, but these are below the nominal size 0.05
I
for N"25, "10 and N"50, "10. The results for RS , RSH, RS are not
M
M
IM

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

15

Table 2
Empirical size of tests. (nominal size"0.05)
Tests
(N, )

RS
I

RSH
I

RS
M

RSH
M

RS
IM

RSO
I

RSOH
I

(25, 10)
(25, 20)
(50, 10)

0.047
0.050
0.043

0.048
0.051
0.040

0.087
0.060
0.065

0.072
0.056
0.062

0.062
0.057
0.059

0.045
0.052
0.046

0.051
0.058
0.053

good. All of them reject the null too frequently, but the empirical sizes improve
as we increase N or . Comparing the performances of RS and RSH, we notice
M
M
that RSH has somewhat better size properties. As expected, the one-sided tests
M
RSO and RSOH have larger empirical sizes than their two-sided counterparts.
I
I
Overall, except for a couple of cases, the size performance of all tests are within
one standard errors of the nominal size 0.05.
The results of Table 2 are consistent with the Q}Q plots in Fig. 1 for
N"25, "10. To save space "gures for the other two combinations of (N, )
are not included. We also do not present the "gures for the joint and one-sided
tests, since they resemble those reported for the other tests. From the plots note
that the empirical distributions of the test statistics diverge from that of the s at

the right tail parts. For RS and RSH the points are below the 453 line,
I
I
particularly for the high values, and that leads to sizes being below 0.05 as we
just noted from Table 2. However, the number of points (out of 1000) that are far
away from the 453 line at the tail parts are not many. For RS and RSH we
M
M
observe a higher degree of departure from the 453 line in the opposite direction,
and this leads to much higher sizes of the tests. Results from the Kolmogorov}Smirnov test, not reported here, accept the null hypothesis of the
overall distribution being the same as s for the "rst "ve, and standard normal
for the last two statistics. For the true sizes of the tests, however, it is only the tail
part, not the overall distribution, that matters.
Let us now turn into the performance of tests in terms of power. For N"25
and "10, the estimated rejection probabilities of the tests are reported in
Table 3, and are also illustrated in Figs. 2(a)}(d). The results for q"o"0.08 are
not reported since in most cases the rejection probabilities were one or very
close to one. Moreover, our adjusted tests are designed for locally misspeci"ed
alternatives close to q"o"0.0, and the main objective of our Monte Carlo
study is to investigate the performance of our suggested tests in the neighborhood of q"o"0.0. Let us "rst concentrate on RS , RSH, RSO and RSOH
I
I
I
I
which are designed to test the null hypothesis H : p"0. When o"0, RS and
 I
I
RSO are, respectively, the two- and one-sided optimal tests. This is clearly
I
evident looking at all the rows in Table 3 with o"0; RSO has the highest
I

16

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

Fig. 1. Q}Q plots. Sample size (25, 10).

powers among all the tests and RS just trails behind it. The power of RSH is less
I
I
than that of RS when o"0. The losses in power are, however, not very large, as
I
can also be seen from Fig. 2(a). When q exceeds 0.2 (or p exceeds 4, since we set
I
p"20q) both tests have power equal to 1. The amount of loss in using RSH
I
I
when o"0 was characterized by (17) in terms of the decrease in the noncentrality parameter. That loss increases with q(p ). However, the overall power of RSH
I
I
is guided by the noncentrality parameter in (16):
p
p !1
j (p )" I (!1)! I
,
 I
2p
p
C
C
where the second term is the amount of penalty in using RSH when o"0, and it
I
is given in (17). Since the "rst term dominates, the relative value of the loss is
negligible. While RSH and RSOH do not sustain much loss in power when o"0,
I
I
we notice some problems in RS and RSO when p"0 but oO0. RS and
I
I
I
I
RSO reject H : p"0 too frequently. For example, when q"0 (i.e., p"0)
I
 I
I
and o"0.4, RS and RSO have rejection probabilities 0.847 and 0.888, respecI
I
tively. For other values of o the proportion of rejections of p"0 (when it is
I

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

17

Table 3
Estimated rejection probabilities of di!erent tests. Sample size: N"25; "10
q

RS
I

RSH
I

RS
M

RSH
M

RS
IM

RSO
I

RSOH
I

0.00
0.00
0.00
0.00
0.00
0.00

0.00
0.05
0.10
0.20
0.40
0.60

0.047
0.053
0.123
0.322
0.847
0.998

0.048
0.050
0.080
0.158
0.325
0.776

0.087
0.143
0.381
0.869
1.000
1.000

0.072
0.141
0.333
0.788
0.999
1.000

0.062
0.122
0.342
0.818
1.000
1.000

0.045
0.085
0.187
0.416
0.888
0.998

0.051
0.039
0.061
0.128
0.354
0.804

0.05
0.05
0.05
0.05
0.05
0.05

0.00
0.05
0.10
0.20
0.40
0.60

0.344
0.442
0.514
0.734
0.955
0.998

0.298
0.301
0.296
0.364
0.576
0.867

0.153
0.351
0.598
0.949
1.000
1.000

0.072
0.118
0.326
0.789
1.000
1.000

0.308
0.423
0.605
0.932
1.000
1.000

0.435
0.530
0.591
0.776
0.971
1.000

0.373
0.402
0.359
0.428
0.641
0.890

0.10
0.10
0.10
0.10
0.10
0.10

0.00
0.05
0.10
0.20
0.40
0.60

0.752
0.759
0.830
0.907
0.988
1.000

0.691
0.630
0.644
0.648
0.790
0.933

0.371
0.563
0.792
0.990
1.000
1.000

0.047
0.123
0.301
0.794
0.999
1.000

0.702
0.728
0.852
0.980
1.000
1.000

0.808
0.818
0.876
0.937
0.991
1.000

0.760
0.707
0.723
0.710
0.830
0.949

0.20
0.20
0.20
0.20
0.20
0.20

0.00
0.05
0.10
0.20
0.40
0.60

0.983
0.977
0.987
0.991
0.999
1.000

0.968
0.962
0.967
0.942
0.954
0.990

0.802
0.906
0.966
0.997
1.000
1.000

0.042
0.139
0.300
0.785
0.999
1.000

0.977
0.981
0.988
0.998
1.000
1.000

0.988
0.984
0.992
0.994
0.999
1.000

0.982
0.973
0.975
0.958
0.964
0.992

0.40
0.40
0.40
0.40
0.40
0.40

0.00
0.05
0.10
0.20
0.40
0.60

1.000
0.999
1.000
1.000
1.000
1.000

1.000
0.999
1.000
1.000
1.000
1.000

0.995
0.999
0.999
1.000
1.000
1.000

0.045
0.125
0.321
0.774
0.998
1.000

1.000
0.999
1.000
1.000
1.000
1.000

1.000
1.000
1.000
1.000
1.000
1.000

1.000
0.999
1.000
1.000
1.000
1.000

0.60
0.60
0.60
0.60
0.60
0.60

0.00
0.05
0.10
0.20
0.40
0.60

1.000
1.000
1.000
1.000
1.000
1.000

1.000
1.000
1.000
1.000
1.000
1.000

1.000
1.000
1.000
1.000
1.000
1.000

0.045
0.156
0.311
0.739
0.998
1.000

1.000
1.000
1.000
1.000
1.000
1.000

1.000
1.000
1.000
1.000
1.000
1.000

1.000
1.000
1.000
1.000
1.000
1.000

true) for RS can be seen in Fig. 2(b). As we discussed in Section 3, this unwanted
I
rejection probabilities is due to the noncentrality parameter j (o) in (14), which

is `purelya a function of the degree of departure of o from zero. RSH and RSOH
I
I
also have some unwanted rejection probabilities but the problem is less severe.

18

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

Fig. 2. Tests for random e!ects. Sample size (25, 10).

For the above case of q"0 and o"0.4 the rejection probabilities for RSH and
I
RSOH are, respectively, 0.325 and 0.354. Fig. 2(b) gives the power of RSH when
I
I
q"0 for di!erent values of o. As we mentioned earlier, RSH and RSOH are
I
I
designed to be robust only under local misspeci"cation, i.e, for low values of o.
From that point of view, they do a very good job } their performances
deteriorate only when o takes high values. Now by directly comparing the oneand two-sided tests for H : p"0, we note that the former has higher rejection
 I
probabilities except for a few cases when q"0.0. For these cases the score
*/*p takes large negative values and that leads to acceptance of H when
I

one-sided tests are used and rejection of H when we use the two-sided test.

Note that RSO "sign(RS rejects H if RSO '1.645 while using RS
I
I

I
I
rejection occurs if RS '3.84 which exceeds 1.645.
I
From Table 3 and Fig. 2(c), we note that when q'0, an increase in o('0)
enhances the rejection probabilities of RS . For example, when q"0.05 the
I
rejection probabilities of RS for o"0.0 and 0.2 are, respectively, 0.344 and
I
0.734. This can be explained using the expression (15), which gives the changes in
the noncentrality parameter of RS due to o. From (16) we see that the
I
noncentrality parameter of RSH does not depend on o. This result is, of course,
I

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

19

Table 4
Estimated rejection probabilities of di!erent tests for negative o
q

RS
I

RSH
I

RS
M

RSH
M

RS
MI

Sample size: N"25; "10

0.00
!0.05
0.00
!0.10
0.00
!0.20
0.00
!0.40
0.05
!0.05
0.05
!0.10
0.05
!0.20
0.05
!0.40

0.039
0.044
0.162
0.573
0.254
0.202
0.097
0.039

0.031
0.019
0.016
0.048
0.289
0.340
0.369
0.679

0.173
0.396
0.902
1.000
0.097
0.184
0.680
0.997

0.170
0.346
0.857
1.000
0.130
0.314
0.830
1.000

0.118
0.285
0.833
1.000
0.269
0.365
0.770
1.000

Sample size: N"25; "20

0.00
!0.05
0.00
!0.10
0.00
!0.20
0.00
!0.40
0.05
!0.05
0.05
!0.10
0.05
!0.20
0.05
!0.40

0.041
0.049
0.136
0.610
0.652
0.613
0.507
0.303

0.025
0.025
0.010
0.018
0.707
0.758
0.829
0.963

0.247
0.640
0.999
1.000
0.090
0.244
0.882
1.000

0.217
0.600
0.999
1.000
0.200
0.557
0.987
1.000

0.168
0.520
0.992
1.000
0.665
0.806
0.992
1.000

valid only asymptotically and for local departures of o from zero. Fig. 2(d) shows
that there is some uniform gain in rejection probabilities of RSH only when
I
o"0.4. For smaller values of o, the rejection probabilities sometimes even
decrease but are always close to values for the case o"0.
As we indicated earlier there could be some loss of power of RS when o(0.
I
We performed a small-scale experiment for this case, results of which are
reported in Table 4. First note that when q"0, an increase in the absolute value
of o leads to an increase in the size of RS . For example, when N"25, "10
I
and q"0, the rejection frequencies for o"0 and !0.4 are, respectively, 0.047
and 0.573. This is due to the noncentrality parameter (14) which is a function of
o. When q'0 (p'0), the changes in the noncentrality parameter could be
I
negative, and there could be a substantial loss in power of RS . For instance, for
I
the above (25, 10) sample size combinations, and q"0.05, the powers of RS ,
I
for o"0.0 and !0.4 are, respectively, 0.344 and 0.039. RSH does not su!er
I
from these detrimental e!ects as we see from Table 4. Its size remains small for
all o(0, and power even increases as the absolute value of o becomes larger.
In a similar way, we can explain the behavior of RS and RSH using Table 3
M
M
and Figs. 3(a)}(d). From Table 3 we note that, as expected, when p "0, RS
I
M
has the highest powers among all the tests. The powers of RSH are very close to
M
those of RS . Therefore, the premium we pay for the wider validity of RSH is
M
M
minimal.

20

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

Fig. 3. Tests for serial correlation. Sample size (25, 10).

The real bene"t of RSH is noticed when o"0 but q'0; the performance of
M
RSH is quite remarkable, as can be seen from Fig. 3(b). RS rejects H : o"0 too
M
M

often, whereas, quite correctly, RSH does not reject H so often. For example,
M

when q"0.2 and o"0, the rejection proportions for RS and RSH are 0.802
M
M
and 0.042, respectively. Even when we increase q to 0.6, the rejection proportion
for RSH is only 0.045, whereas RS rejects 100% of the time. In a way, RSH is
M
M
M
doing more than it is designed to do, that is, not rejecting o"0 when o is indeed
zero even for large values of q.
From Fig. 3(c), we observe that the power of RS is strongly a!ected by the
M
presence of random e!ects, while there is virtually no e!ect on the power of RSH
M
as seen from Fig. 3(d) even for large values of q. This performance of RSH is
M
exceptionally good. For negative values of o in Table 4, we see that the presence
of q has a less detrimental e!ect on RSH. For example, when o"!0.10, the
M
rejection probabilities of RS are 0.396 and 0.184 for q"0.0 and 0.05, respecM
tively; for the same situations, the powers of RSH are, respectively, 0.346 and
M
0.314.
Comparing the performance of RSH and RSH, we see that the former is even
M
I
more `robusta in the presence of q, both in terms of size and power, than is the

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

21

latter in the presence of serial correlation. To see this from a theoretical point of
view, let us consider (17) and (22), which are, respectively, the penalties of using
RSH and RSH. From (17), p /p(!1)/, the penalty in using RSH, also depends
I
M
I C
I
on o through p"20(1!q)(1!o), while (22), 2o(!1)/, is a function of
C
o only and is of smaller magnitude in terms of .
Finally, we discuss brie#y the performance of the joint statistic RS in the
IM
light of our results (4) and (5). This test is optimal when p '0 and oO0. As we
I
can see from Table 3, in this situation RS has the highest power most of the
IM
time. However, when the departure from p "0, o"0 is one-directional (say,
I
p'0, o"0), RS and RS have the same non-centrality parameter (see (26)).
I
I
IM
Since RS and RS use the s and s tests, respectively, there will be a loss of
IM
M


power in using RS . For example, when q"0.10 and o"0, the powers for RS
IM
I
and RS are 0.752 and 0.702, respectively. Similarly, when q"0, o"0.2, the
IM
power of RS and RS are, respectively, 0.869 and 0.818. These results are
M
IM
consistent with those of Baltagi and Li (1995). Although RS has overall good
IM
power, it cannot help to identify the exact source of misspeci"cation when there
is only a one-directional departure.
The qualitative performance of all the tests do not change when we increase
the sample sizes to N"25, "20, and N"50, "10 and they further
illustrate the usefulness of our modi"ed tests. These results are not presented but
are available from the authors upon request.

6. Conclusions
In this paper we have proposed some simple tests, based on OLS residuals for
random e!ects in the presence of serial correlation, and for serial correlation
allowing for the presence of random e!ects. These tests are obtained by adjusting the existing test procedures. We have investigated the "nite sample size and
power performance of these and some of the available tests through a Monte
Carlo study. We have also provided some empirical examples. The Monte Carlo
study, along with the examples, clearly show the usefulness of our procedures to
identify the exact source(s) of misspeci"cation. One drawback of our methodology is that we allow for only local misspeci"cation. For nonlocal departures,
e$cient tests could be obtained after estimating full model(s) by maximum
likelihood; that, however, will loose the simplicity of our tests using only OLS
residuals.

Acknowledgements
We would like to thank the co-editor A. Ronald Gallant, an associate editor
and two anonymous referees for many pertinent comments that helped us to

22

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

improve the paper. Thanks are also due to Miki Naoko for her help in
preparing the manuscript. An earlier version of this paper was presented at
Texas A&M University, March 1996; the Midwest Econometric Group
Meeting, the University of Wisconsin at Madison, November 1996; the
Economics seminar at University of San Andres, Argentina, November 1997;
and the Annual Meeting of the Argentine Association of Political Economy,
Bahia Blanca, Argentina. We wish to thank the participants and Badi Baltagi for
helpful comments and discussion. However, we retain responsibility for any
remaining errors.

References
Andereson, T.W., Hsiao, C., 1982. Formulation and estimation of dynamic models using panel data.
Journal of Econometrics 18, 47}82.
Balestra, P., Nerlove, M., 1966. Pooling cross-section and time-series data in the estimation of
a dynamic model: the demand for natural gas. Econometrica 34, 585}612.
Baltagi, B., 1995. Econometric Analysis of Panel Data. Wiley, New York.
Baltagi, B., Chang, Y., Li, Q., 1992. Monte Carlo results on several new and existing tests for the
error component model. Journal of Econometrics 54, 95}120.
Baltagi, B., Li, Q., 1991. A joint test for serial correlation and random individual e!ects. Statistics
and Probability Letters 11, 277}280.
Baltagi, B., Li, Q., 1995. Testing AR(1) against MA(1) disturbances in an error component model.
Journal of Econometrics 68, 133}151.
Bera, A.K., Jarque, C.M., 1982. Model speci"cation tests: a simultaneous approach. Journal of
Econometrics 20, 59}82.
Bera, A.K., Yoon, M.J., 1991. Speci"cation testing with misspeci"ed alternatives. Bureau of Economic and Business Research Faculty Working Paper 91-0123, University of Illinois, presented
at the Econometric Society Winter Meeting, Washington, DC, December 1990.
Bera, A.K., Yoon, M.J., 1993. Speci"cation testing with locally misspeci"ed alternatives. Econometric Theory 9, 649}658.
Bera, A.K., Ra, S.S., Sarkar, N., 1998. Hypothesis testing for some nonregular cases in econometrics.
In: Chakravorty, S., Coondoo, D., Mukherjee, R. (Eds.), Econometrics: Theory and Practice.
Allied Publishers, New Delhi, pp. 319}351.
Breusch, T.S., Pagan, A.R., 1980. The Lagrange multiplier test and its applications to model
speci"cation in econometrics. Review of Economic Studies 47, 239}253.
Das Gupta, S., Perlman, M.P., 1974. Power of the noncentral F-test: E!ect of additional variate on
Hotteling's -test. Journal of the American Statistical Association 69, 174}180.
Davidson, R., MacKinnon, J.G., 1987. Implicit alternatives and local power of test statistics.
Econometrica 55, 1305}1329.
Greene, W., 1983. Simultaneous estimation of factor substitution, economies of scale and nonneutral technical change.. In: Dogramaci, A. (Ed.), Econometric Analyses of Productivity.
Kluwer-Nijo!, Boston, pp. 121}144.
Greene, W., 2000. Econometric Analysis, 4th Edition. Prentice-Hall, Englewood Cli!s, NJ.
Grunfeld, Y., 1958. The determinants of corporate investment. Unpublished Ph.D. Thesis, Department of Economics, University of Chicago.
Grunfeld, Y., Griliches, Z., 1960. Is aggregation necessarily bad? Review of Economics and Statistics
42, 1}13.
Haavelmo, T., 1944. The probability approach in econometrics. Econometrica(Suppl.) 12, 1}115.

A.K. Bera et al. / Journal of Econometrics 101 (2001) 1}23

23

Hillier, G.H., 1991. On multiple diagnostic procedures for the linear model. Journal of Econometrics
47, 47}66.
Honda, Y., 1985. Testing the error component model with non-normal disturbances. Review of
Economic Studies 52, 681}690.
Hsiao, C., 1986. Analysis of Panel Data. Cambridge University Press, Cambridge.
Lillard, L.A., Willis, R.J., 1978. Dynamic aspects of earning mobility. Econometrica 46, 985}1012.
Majunder, A.K., King, M.L., 1999. Estimation and testing of a regression with a serially correlated
error component. Working paper, Department of Econometrics and Business Statistics, Monash
University, Australia.
Nerlove, M., 1971. Further evidence of the estimation of dynamic economic relations from a timeseries of cross-sections. Econometrica 39, 359}382.
Phillips, R.F., 1999. Estimation of mixed-e!ects and error components models with an AR(1)
component: some new likelihood maximization procedures and Monte Carlo evidence. Working
paper, Department of Economics, George Washington University.
Rao, C.R., 1948. Large sample tests of statistical hypothesis concerning several parameters with
applications to problems of estimation. Proceedings of the Cambridge Philosophical Society 44,
50}57.
Saikkonen, P., 1989. Asymptotic relative e$ciency of the classical test statistics under misspeci"cation. Journal of Econometrics 42, 351}369.
Welsh, A.H., 1996. Aspects of Statistical Inference. Wiley, New York.