The International Journal of Accounting

39 (2004) 197 217

The decision relevance and timeliness of accounting

earnings in Saudi Arabia
Mohammed Al-Sehali a, Nasser Spear b,*
b

a
King Saud University, Saudi Arabia
Department of Accounting and Business Information System, University of Melbourne, Melbourne,
Victoria 3010, Australia

Abstract
This study examines the decision relevance and timeliness of accounting earnings in Saudi Arabia
during the 1995 1999 sample period. The empirical results suggest that the publication of
accounting earnings does not cause significant revision to the market assessment of future cash flows
of Saudi firms. On the other hand, it appears that the publication of accounting earnings leads
individual investors to revise their security holdings. However, this evidence is limited to cases
where firms reported profit. The empirical results further suggest that earnings are timely in terms of
their association with security returns and that increasing the measurement interval significantly
improves this association. The tests also show that positive and negative earnings have differential
implications for the timeliness of accounting earnings. Further tests show that this evidence is not
consistent with the loss liquidation argument [J. Account. Econ. 20 (1995) 125] and, potentially, may
reflect the lack of tax incentives to liquidate investments in loss firms. Finally, the results show that
Saudi managers do not incorporate economic losses into accounting earnings on a timely basis which
may reflect reduced market demand for accounting information, low levels of public debt, low
expected litigation costs, and weak monitoring by analysts and other stakeholders.
D 2004 Published by University of Illinois. All rights reserved.
Keywords: Decision relevance; Timeliness; Accounting earnings; Saudi Arabia

1. Introduction
The last three decades witnessed the emergence of a research paradigm that tests the
relation between alternative capital markets metrics (e.g., security prices, security

* Corresponding author.
E-mail address: nasser@unimelb.edu.au (N. Spear).
0020-7063/$30.00 D 2004 Published by University of Illinois. All rights reserved.
doi:10.1016/j.intacc.2004.02.004

198

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

returns, and trading volume) and accounting data (such as earnings, book value, among
others). The aim of these tests is to provide insights into the relevance and reliability
of accounting data for members of the investment community and, perhaps, for
standard setters. They test the decision relevance and timeliness of accounting
information.
The majority of research on the decision relevance and timeliness of accounting data
uses data from the United States, the United Kingdom, and the Australian capital markets.
Recently, there has been a growing interest in alternative national accounting models. For
example, Ball, Kothari, and Robin (2000) examine the effect of international institutional
factors on two properties of accounting earnings: timeliness and conservatism. Their
sample includes firms from common-law countries (Australia, Canada, UK, and United
States) and code-law countries (France, Germany, and Japan). Similarly, Ball, Robin, and
Wu (2000) conduct a similar study using a sample of firms from Hong Kong, Malaysia,
Singapore, and Thailand. In addition, some studies examine the relevance of accounting
information in other emerging markets. For example, Chen, Chen, and Su (1999)
investigate the relevance of accounting earnings in China, Choi and Choe (1998)
investigate the effects of annual earnings announcements on investors trading behavior
in the Korean stock market.
To date, the extant literature lacks significant empirical evidence on the current role of
accounting information in security valuation in the Saudi security market, despite its status
as one of the largest (by market capitalization) among emerging markets.1 This paper aims
at addressing this gap by investigating the decision relevance and timeliness of accounting
information in Saudi Arabia during the 1995 1999 sample period. Furthermore, following
Hayn (1995), we assess the differential implications of positive versus negative earnings
on the timeliness of accounting earnings. Finally, we examine the extent to which
accounting earnings asymmetrically incorporates economic losses relative to economic
gain (i.e., the impact of accounting conservatism as defined by Basu, 1997) in Saudi
Arabia.
We argue that the role of accounting information in security valuation warrants an
empirical investigation due to factors that impact both the demand placed on accounting
information and the supply of accounting information in Saudi Arabia. The empirical
evidence gathered by this study helps to understand the role of accounting information in
security valuation in Saudi Arabia and possibly have policy implications for accounting
standard setters in Saudi Arabia.
With respect to the demand placed on accounting information, the government of
Saudi Arabia has recently outlined a number of initiatives that aim to foster a greater
local and international private sector involvement in its economic development activities,
and it has opened the door for more involvement by non-Saudi nationals to invest in its
capital markets. These initiatives are likely to increase the demand placed on accounting
information by current and perspective investors. On the other hand, many large, listed

1
Notable exceptions include the work of Al-Bogami, Green, and Power (1997), who used an event study to
investigate the market reaction to quarterly earnings announcements of 39 listed Saudi firms during the 1987
1991 sample period.

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

199

Saudi companies are partially owned by the Saudi Government. The high concentration
of government ownership for these companies is likely to mean higher levels of
information asymmetry.2 Furthermore, some listed Saudi companies are family controlled; one family often holds a controlling interest in various public and private
companies. Some of these companies are tied together by cross holdings and personal
relationships. This form of ownership may lead to direct access to insider information by
government officials while the public continues to face information asymmetry. This
setting, in turn, creates alternative forms of contracting relationships and is likely to
reduce the demand for public disclosure and the demand for timely and transparent
accounting information.
With respect to the supply of accounting information, in contrast to many
developed economies where the provision of accounting information is governed
by a national set of Generally Accepted Accounting Principles (GAAP), the Saudi
firms have traditionally adopted Anglo-American GAAPs (U.S. GAAP, U.K. GAAP,
and more recently, International Accounting Standards GAAP). While the establishment of a domestic standard-setting body in 1992, the Saudi Organization for
Certified Public Accountants (SOCPA) has helped to reduce reporting diversity.
SOCPA is still very much in its early stages, having released only 16 accounting
standards. Both decision relevance and timeliness of accounting information in Saudi
Arabia are likely to be influenced by the degree of alignment between the underpinnings for Anglo-American type of GAAP and the principles and doctrines adopted
by the Saudi investment community. In addition, the low expected cost of litigation
in Saudi Arabia, as well as the lack of a public debt market (due to Islamic
tradition), suggests that Saudi managers and auditors may have low incentives to
produce timely and transparent financial reports. That is, from a supply point of
view, accounting information may have a lower impact on security prices in Saudi
Arabia than in other countries.
The next section provides a brief overview of key Saudi institutional factors. Section 3
discusses the research design. Section 4 discusses the research findings. Section 5
concludes the study.

2. Key Saudi institutional factors

The Kingdom of Saudi Arabia is an Arabic Islamic monarchy, headed by the King,
who also serves as the countrys Prime Minister. The King exercises his authority
through a body referred to as the Council of Ministers. Two other councils advise the
King and Council of Ministers. The first council is the Consultative Council (Majlis

2
Traditionally, information asymmetry has been interpreted as differences between managers and owners of
firms. That is, managers possess private information about the firm and its earnings that shareholders do not have.
In the context of this paper, we argue that information asymmetry in Saudi Arabia exists between investors
themselves (government vs. nongovernment). Indeed, these two groups differ in their information resources, their
investment horizons, and their investment strategy.

200

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

Alshoura), which includes academics, businesspeople, government members, and

religious scholars. The second council is the Council of the Assembly of Senior
Religious Scholars (Majlis Kibar al-Ulama), a body that is charged with the
responsibility of ensuring that Saudi Arabia is governed in conformity with Islamic
law and teaching.
Islamic teaching plays the major role in the formulation and development of the legal
system in Saudi Arabia. The Quran, the holy book of Islam, is the most important
source of legislation, followed by the Sunna, the teachings of the Prophet Mohammed.
As a matter of Islamic law, if an authoritative legal statement from the Quran or the
Sunna (collectively referred to as Sharia) can be cited, it is binding and supersedes all
other sources of legislation (Ernst & Young International, 1998, pp. 56 58). Other
sources of law in Saudi Arabia include Royal and Ministerial Decrees and Departmental
Circulars.
2.1. The Saudi economic and financial system
The Saudi economy, the largest in the Middle East, has been developed through a series
of 5-year plans. These plans have been formulated to pursue economic diversification
through the development of private sector activities as well as greater economic
development from the private sector. The governments commitment to foster greater
private sector involvement in its economic development activities suggests that the private
sector will play a leading role in financing these development activities, creating a greater
demand for relevant and reliable financial information to facilitate effective decision
making. It also suggests that financial markets in Saudi Arabia will be more active and
move toward sophisticated trading arrangements.
Saudi Arabia does not have a physical stock market location; the exchange of stocks
is conducted through a computerized network. The Saudi stock market is the eighth
largest stock market by market capitalisation in the developing world, and it is the largest
stock market in the Arab world, accounting for 63% of the volume of shares on all
Arabian stock exchanges. Generally speaking, only Saudi nationals may own or,
otherwise, deal in shares listed on the Saudi exchange. However, nationals of other
Gulf countries (GCC) may also own shares in certain joint stock companies. In 2000, the
government permitted non-Saudi nationals who reside in Saudi Arabia to invest in the
Saudi stock market through existing local funds managed by domestic commercial
banks. As a result of these governmental initiatives, there are currently 132 investment
funds managed by commercial banks.
2.2. The Saudi accounting profession
The accounting profession in Saudi Arabia is young, but maturing, as it undergoes
continuous development. During its infancy, there was no comprehensive, authoritative
support. This condition changed with the passage of the first Law of Certified Accountants
in 1992. Article 19 of the Law established the first authorized professional association of
accountants, the SOCPA. SOCPA is responsible for regulating the accounting profession
and its practices. The regulatory sources of law governing the accounting profession

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

201

include the Companies Act, the Zakat and Income Tax Act,3 the Ministry of Commerce
(1986a,b) decision, and SOCPA regulations.
SOCPA introduced a professional qualification for Saudi nationals, modeled after the
certified public accountants designation in the United States. The Ministry of Commerce
license qualified individuals and associations of qualified individuals to practice. At of the
end of 1997, there were over 300 licensed practicing accountants in Saudi Arabia. The
accounting profession is far weaker in Saudi Arabia than that in the West, but the
accountants and their corporate clients face lower expected litigation cost. While the
Ministry of Commerce and SOCPA host a committee that investigates any wrongdoing in
the profession, the procedures and decisions of this committee are not disclosed to the
public. To date, not a single audit firm has been sued, even in cases of corporate failure.
Since its inception, SOCPA has issued 16 accounting standards. Accordingly, Saudi
companies do not yet have a unified set of GAAP, and Saudi companies are still informally
guided by U.S., UK and/or IAS standards.

3. Research design
3.1. The decision relevance of accounting earnings
We assess the decision relevance of accounting earnings in Saudi Arabia in two ways.
First, we utilize the research design of Ball and Brown (1968) to investigate the association
between the sign of security returns and the sign of annual earnings of Saudi firms
surrounding and during the release week of annual reports.4 If earnings provide a summary
performance measure, we should observe positive (negative) security returns for profitable
(losing) firms. We construct a wealth index to trace the value of US$ 1 invested (in equal
amounts) in all securities at the end of Week 52 (i.e., 52 weeks prior to the week of the
annual report) and held to Week + 13 (i.e., 13 weeks after the week of the annual report).
In other words:
WIw

1 XY
1 Retj;w
N j TP

3
Zakat is a religious wealth tax levied on Saudi and other GCC nationals and their wholly owned companies
and that proportion of mixed-ownership companies that is owned by Saudi or other GCC nationals. Zakat is a
fixed-rate tax of 2.5% on capital that is not invested in fixed assets or long-term investments. Income tax, on the
other hand, is levied on companies wholly owned by foreign nationals and the proportion of mixed-ownership
companies that is owned by foreign nationals. Income tax is levied on taxable income, which is calculated as
gross income less all expenses that are necessarily incurred in earning income and satisfy the deductibility rules.
The current rate of corporate income tax in Saudi Arabia varies from 25% to 45%, with no tax-free threshold.
4
Note that Saudi firms do not announce earnings ahead of the annual report release date. Note also that Saudi
investors may have access to private information about firm performance before the release week of the annual
report (a research in-progress by one of the authors of this paper suggests that information leakage is present in
the Saudi market). For these reasons, we examine the market movement surrounding and including the annual
report release date.

202

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

where WIw is the wealth index during week w; Retj,w is the raw rate of return for security j
during week w; and N is the number of securities.
Similar with Ball and Brown (1968),5 the wealth index is then plotted for portfolios
constructed for (1) all firms and years, (2) all firms and years in which the reported
earnings level figure is positive, and (3) all firms and years in which the reported
earnings level figure is negative. Chi-square statistics are computed for a 2  2
classification of firms by the sign of earnings levels and the sign of raw returns for
each week.
Second, we use the research design of Beaver (1968) to investigate the presence of
any abnormal trading volumes surrounding and during the release week of the annual
report. Benchmarked against trading volume during nonreporting weeks, we investigate the presence of abnormal trading volume movements during (and surrounding)
the annual report release date. Similar to Beaver, a weekly average of daily percentage
of shares traded is calculated for each firm for each week in the report period as
follows:

vj;w


number of shares of firm j traded during week w

number of shares outstanding of firm j traded during week w
1
number of trading days during week w

The weekly volume is divided by the number of shares outstanding to remove any bias
caused by the presence of firms with a large number of shares outstanding. The
percentage of shares traded per week is divided by the number of trading days to adjust
for nontrading days such as public holidays. The normal trading week in Saudi Arabia is
equal to 5.5 days (five full days and one half day on Thursdays). The report release period
is defined as the 7-week period surrounding and including the release week (3 weeks
before the release date, the report release week, and 3 weeks after the report release
week). We compute the average volume across all observations for each week during the
report period, benchmarked against the average volume during the nonreport period
(Week 52 4, Week + 4 + 13).
3.2. The timeliness of accounting earnings in Saudi Arabia
We use the research design of Easton and Harris (1991) to investigate the extent to
which accounting earnings incorporate current period economic income, as proxied by
security returns. We assess timeliness by examining the contemporaneous relationship
between the earnings levels, earnings changes, and security returns of Saudi firms over
1-, 2-, and 5-year intervals. Following Easton and Harris and Easton, Ohlson, and

5
Our index differs from the Ball and Brown (1968) index in that it does not control for market-wide factors.
However, the use of an index identical to that of Ball and Brown does not change our findings.

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

203

Harris (1992), the following regression is estimated for the pooled cross-sectional and
time-series sample, as well as for each of the intervals noted above:
Reti;t a b1 Ei;t =Pi;t1 b2 DEi;t =Pi;t1 ei;t

where E is the earnings level per share; DE is the earnings change per share Ret raw
annual security return; and P the price per share of firm i at time t.
Similar with Easton and Harris (1991), we assess timeliness by examining the
explanatory power of the model (adjusted R2) in Eq. (3). Inferences about the impact of
the length of the measurement interval on timeliness are drawn by examining the change in
the explanatory power (adjusted R2) of Eq. (3) as the measurement interval increases.
3.3. The differential implications of positive and negative earnings on the timeliness of
accounting earnings
We use the research design of Hayn (1995) to investigate the differential implications of
positive and negative earnings on the relationship between earnings and the annual
security returns of Saudi firms. We divide the pooled sample into four groups. The first
group includes observations where the level of earnings and the change in earnings are
both positive. The second group includes observations where the level of earnings is
positive but the change in earnings is negative. The third group includes observations
where the level of earnings and the change in earnings are both negative. The final group
includes observations where the level of earnings is negative but the change in earnings is
positive. The regression model outlined in Eq. (3) is then estimated for each group.
Inferences about the role of the sign of reported earnings on the association between
security returns and earnings can be drawn from comparing the adjusted R2 of the pooled
model for each group for annual as well as longer measurement intervals (2 and 5 years).
3.4. Accounting conservatism in Saudi Arabia
We examine accounting conservatism in Saudi Arabia by utilizing the research design
of Basu (1997), in which security returns proxy for economic income. Conservatism is
defined as the extent to which current period accounting earnings asymmetrically
incorporate economic losses relative to economic gain. Following Basu, we estimate a
linear regression of accounting earnings on security returns as follows:
Ei;t =Pi;t1 a b1 RDi;t b2 Reti;t b3 Reti;t RDi;t ei;t

where RD is a dummy variable equaling one if Ret is negative and zero otherwise.
Following Ball, Robin et al. (2000), we control for market-wide return by deducting the
sample mean return in fiscal year t for calculating Ret. The coefficient b2 measures the
contemporaneous sensitivity of accounting earnings to positive security returns (economic
gain). The coefficient b3 measures the incremental sensitivity of accounting earnings to
contemporaneous negative returns (economic loss).

204

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

3.5. Variable definitions

Reti,t
Vi,t
Ei,t
DEi,t

The raw security returns for firm i, adjusted for stock dividend, stock split, and capitalization changes
compounded over the time period t, where t is either 1 week, 1 year, or 5 years.
The traded volume for security i adjusted for stock split over the contemporaneous week, t.
Earnings per share for firm i for period t excluding extraordinary items, discontinued operations, and
zakat and tax obligations scaled by beginning-of-period price.6
Change in earnings per share for firm i for period t. This variable proxies for unexpected earnings for
firm i for period t.

3.6. Sample selection procedures

This study uses data from the period 1995 1999. The sample is drawn from the total
population of firms listed on the Saudi stock market during the full 5-year period. Of the
72 firms listed, 10 were excluded because they were suspended from trading on the Saudi
Arabian stock market, pending satisfaction of certain operating and financial criteria
imposed by the Saudi stock market. Of the remaining 62 firms, 52 firms satisfy our
selection criteria:
(1) Completing weekly share prices series for the 1995 1999 sample period;
(2) completing financial statements data for at least 4 years from the 1995 1999 sample
period;
(3) completing dividend and stock-split data for the 1995 1999 sample period; and
(4) by having clearly identifiable earnings announcement (annual report release) dates for
each yearly observation.
The final sample represents 72% of the total population by number and represents
96.45% of the total population by market capitalization. The final data set includes 256
annual earnings announcements relating to 52 Saudi Arabian firms for the sample period
1995 1999.
3.7. Data sources
Due to the lack of any form of electronic market or financial database in Saudi Arabia
at the time we conducted this research, all of the required data were hand collected.
Weekly high, low, and closing stock prices and the trading volume for each security were
6
Zakat and tax obligations are excluded from the measurement of earnings for four distinct reasons. First, the
sample consists of two groups of firms: (1) fully Saudi-owned firms that are required to pay only zakat, (2) partly
Saudi-owned firms that are required to pay both zakat and income tax. Thus, ignoring zakat and tax obligations
ensures a measure of consistency. Second, because the zakat obligation represents a religious duty, some firms
believe that they should be in charge of paying only the minimum amount of zakat required by law. Accordingly,
many firms may engage in active zakat management practice. Third, certain firms are entitled to special tax
concessions for certain periods. Finally, flexibility inherent in the Saudi zakat and tax code generally leads to
disputes between companies and ZITD regarding the amount due for zakat and tax, which suggests that it is
difficult to ascertain a firms final tax liability.

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

205

obtained from Saudi Arabian Monetary Agency. Dividend data were obtained from SSRC.
Stock-split data were obtained from the Saudi Stock Market Review publications issued by
Bakheet Financial Advisors. Financial accounting data were manually obtained from the
relevant annual reports, hard copies of which were housed in the library of SOCPA or the
library of the Institute of Public Administration in Riyadh, Saudi Arabia. Data for a few
missing annual reports were obtained from the microfilm archives of three leading Saudi
Arabian newspapers: Al-Riyadh, Al-Jazeerah, and Okaz.
There is no database that contains earnings-announcement dates for firms listed on the
Saudi stock market. Instead, all firms listed on the Saudi stock exchange are required by
law to publish their complete financial statements in at least two Saudi newspapers within
3 months following the fiscal year-end. Accordingly, earnings-announcement dates were
collected by hand from the microfilm archives of the financial sections of these newspapers. The archives of Saudi newspapers were carefully checked to determine the exact
earnings-announcement date. This is defined as the date on which the complete financial
statements of a firm first appeared in any Saudi newspaper.

4. Empirical results
4.1. Descriptive statistics
Table 1 presents the summary statistics for the main variables of interest. All accounting
variables are stated on a per-share basis and scaled by beginning-of-period price. As can be
seen in Table 1, the mean (median) value for annual earnings per share as a percentage of
beginning share price is 4.8% (5.3%). Almost 25% of the firms reported annual losses.
Two-year earnings are lower than twice annual earnings, and 5-year earnings are greater
than five times annual earnings. Because the variables are scaled by beginning-of-period

Table 1
Descriptive statistics
Variable

Number of
observations

Mean

S.E. of
the mean

S.D.

Lower
quartile

Median

Upper
quartile

Percent
negative

(A) Annual pooled observations

E
254
0.048
RET
254
0.046

0.008
0.019

0.130
0.299

 0.003
 0.168

0.053
 0.022

0.103
0.243

25.09
53.14

(B) Two-yearly pooled observations

E
100
0.079
RET
100
0.178

0.021
0.043

0.209
0.435

 0.039
 0.173

0.095
0.163

0.186
0.420

28.00
36.00

(C) Five-yearly pooled observations

E
48
0.258
RET
48
0.296

0.053
0.120

0.364
0.831

 0.012
 0.300

0.285
0.140

0.543
0.920

23.07
30.76

E: earnings per share excluding extraordinary items, discontinued operations, and zakat and tax obligations scaled
by beginning-of-period price.
RET: annual security returns.

206

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

price, average reported values will tend to increase disproportionately over longer intervals
due to the reinvestment of earnings and the influence of negative values.
The mean (median) value for annual security returns is 4.6% (  2.2%) and is
significantly different from zero. Interestingly, while over 53% of the observations had
negative annual security returns, this percentage is quite similar with those reported in U.S.
studies. Nevertheless, aggregate security returns over longer intervals are positive, and the
proportion of negative return realizations drop significantly.
Table 2 presents the time lag between the end of the fiscal year and the week of the
release of the annual report (earnings-announcement week) for the sample firms. The
average lag period in weeks is 11.50, 10.73, 10.63, 10.80, and 7.10 during 1995, 1996,
1997, 1998, and 1999, respectively. While it is not clear why there has been a sharp drop
in the average lag period during 1999, the average time lag in each sample year is below
the 3-month rule imposed by SOCPA. By the end of Week 4, 13% of the firms released
their annual report, and by the end of Week 13, 70% of the firms released their annual
report.
As can be seen in Table 3, the most frequent firm year-end in Saudi Arabia is
December, used by approximately 83% of the sample. During the sample period, no firm
studied changed its year-end. The most frequent annual report release month is March,
which accounted for 27% of the sample, and the months of January, February, and March,
combined, accounted for 63%.
4.2. The decision relevance of accounting earnings in Saudi Arabia
Fig. 1 plots the average weekly wealth indices for three portfolios constructed from all
firms and all years. The first portfolio includes all 256 observations. The positive portfolio
includes 191 observations, all of which had positive reported earnings. The negative
portfolio includes 65 observations, all of which had negative reported earnings.
Table 2
Number of weeks between fiscal year-end and annual report release date
Number of weeks

4 or less than 4
5
6
7
8
9
10
11
12
13
14
15
More than 15
Total

Number of releases
1995

1996

1997

1998

1999

Total

Cumulative %

5
3
1
2
1
1
5
2
4
6
3
5
14
52

4
5
2
4
0
3
3
5
4
5
3
4
10
52

7
1
1
4
3
1
5
4
4
7
5
0
10
52

1
4
5
2
2
1
6
7
7
1
5
1
10
52

16
10
7
1
1
1
3
0
0
2
0
3
4
48

33
23
16
13
7
7
22
18
19
21
16
13
48
256

13
22
28
33
36
39
47
54
62
70
76
81
100

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

207

Table 3
Distribution of financial statement and announcements dates
Month

Firm year-end
Number

January
February
March
April
May
June
July
August
September
October
November
December
Total

2
4
2
1

43
52

Times earnings reports were announced in each month

%

Number

Cumulative %

0.00
0.00
0.04
0.00
0.08
0.00
0.04
0.00
0.02
0.00
0.00
0.83
100

39
54
68
34
14
9
12
12
5
1
4
2
254

15
21
27
13
6
4
5
5
2
0
2
1

15
37
63
77
82
86
91
95
97
98
99
100

Fig. 1 shows a marked positive association between the signs of the reported earnings
and of the wealth index. The weekly chi-square statistics for the 2  2 classification by the
signs of the reported annual earnings and of wealth index show that it is unlikely that there
is no relationship between the two signs in the majority of the weeks up to that of annualreport announcement. These statistics are especially strong for the total sample portfolio
and for the positive-earnings portfolio. The chi-square statistics for the negative-earnings
portfolio are only significant during 32 weekly periods.

Fig. 1. Wealth index for various portfolios (earnings levels).

208

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

Consistent with the U.S. evidence reported in Ball and Brown (1968) and the
Australian evidence reported in Brown (1970), we infer from Fig. 1 that most of the
information contained in reported earnings is anticipated by the Saudi market before
the annual report is released because the release of the actual earnings number does not
appear to cause any unusual jumps in the wealth index during the announcement week.
The upward and downward drifts begin at least 52 weeks before the annual report is
released. This evidence suggests not only that the Saudi market begins to anticipate
reported earnings 52 weeks before the announcement week but also that it continues to do
so with increasing success. Consistent with international evidence, the upward drift for the
positive earnings portfolio continues for approximately 10 weeks after the announcement
Table 4
Average wealth index statistics for total sample portfolio, positive earnings portfolio, and negative earnings
portfolio
Weeks

Average
wealth index

Average
wealth index
(negative)

Average
wealth index
(positive)

Weeks

Average
wealth index

Average
wealth index
(negative)

Average
wealth index
(positive)

 52
 51
 50
 49
 48
 47
 46
 45
 44
 43
 42
 41
 40
 39
 38
 37
 36
 35
 34
 33
 32
 31
 30
 29
 28
 27
 26
 25
 24
 23
 22
 21
 20

1.0018
0.9991
1.0031
0.9950
0.9955
0.9970
0.9939
0.9988
1.0048
1.0120
1.0150
1.0147
1.0149
1.0147
1.0133
1.0119
1.0094
1.0055
1.0024
1.0039
1.0083
1.0058
1.0130
1.0159
1.0218
1.0256
1.0303
1.0350
1.0396
1.0455
1.0495
1.0516
1.0482

1.0000
1.0000
1.0030
0.9948
0.9957
0.9921
0.9869
0.9842
0.9861
0.9834
0.9824
0.9776
0.9744
0.9708
0.9691
0.9679
0.9654
0.9567
0.9516
0.9534
0.9561
0.9562
0.9676
0.9708
0.9726
0.9726
0.9653
0.9663
0.9678
0.9748
0.9850
0.9829
0.9713

1.0020
0.9982
1.0031
0.9950
0.9954
0.9984
0.9960
1.0033
1.0106
1.0210
1.0252
1.0262
1.0277
1.0286
1.0273
1.0259
1.0234
1.0210
1.0186
1.0199
1.0251
1.0215
1.0276
1.0308
1.0381
1.0430
1.0516
1.0579
1.0635
1.0690
1.0709
1.0746
1.0739

 19
 18
 17
 16
 15
 14
 13
 12
 11
 10
9
8
7
6
5
4
3
2
1
0
1
2
3
4
5
6
7
8
9
10
11
12
13

1.0491
1.0474
1.0470
1.0476
1.0501
1.0443
1.0439
1.0429
1.0498
1.0500
1.0535
1.0546
1.0536
1.0551
1.0588
1.0615
1.0607
1.0606
1.0648
1.0717
1.0727
1.0780
1.0855
1.0802
1.0743
1.0837
1.0884
1.0865
1.0932
1.0821
1.0675
1.0553
1.0563

0.9666
0.9477
0.9485
0.9471
0.9472
0.9432
0.9311
0.9356
0.9232
0.9167
0.9171
0.9135
0.9117
0.9119
0.9118
0.9077
0.9026
0.8999
0.8985
0.9068
0.9112
0.9319
0.9298
0.9257
0.9164
0.9282
0.9135
0.9066
0.9080
0.9061
0.9022
0.9091
0.9146

1.0768
1.0806
1.0799
1.0812
1.0844
1.0780
1.0798
1.0803
1.0920
1.0945
1.0990
1.1017
1.1010
1.1029
1.1078
1.1128
1.1134
1.1142
1.1203
1.1266
1.1256
1.1282
1.1365
1.1314
1.1269
1.1349
1.1454
1.1462
1.1576
1.1414
1.1263
1.1157
1.1371

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

209

week. Surprisingly, there does not appear to be any postearnings announcement drift for
the negative-earnings portfolio. If anything, it appears that the market goes through a
period of correction for potential overreaction.
Table 4 presents the weekly wealth index values for three portfolios shown in Fig. 1.
The average annual wealth index value for the total sample increased by 7.17% during the
52 weeks leading to the annual report release week. The average annual wealth index
value for the positive (negative) earnings portfolio increased (decreased) by 12.03%
(9.32%) during the 52 weeks leading to the annual report release date.
Only 11% (1%) of the increase (decrease) in the wealth index for the positive (negative)
earnings portfolio took place during the 4 weeks leading to the annual report release date.
Consistent with international evidence in the United States and Australia, this finding
suggests that the release of the annual earnings reports is not providing decision-relevant
information to Saudi market participants because most of the information contained in
earnings have trickled into the market throughout the year.
Fig. 2 shows the average trading volume across all observations for each week during
the 7-week report period (3 weeks before and 3 weeks after the report release week, as well
as the report release week itself).
The average trading volume during the nonreport release period is 0.156%, and the
average volume during the week of the release of the annual report is 0.186%, which is
19.23% larger than that during the nonreport release period and is the largest during the
reporting period. This finding clearly shows that individual Saudi investors do shift their
portfolio positions at the time of the annual report release week.
The above analysis is further extended by examining the trading volume of profitreporting firms separately from that of loss-reporting firms. Effectively, the calculation of
relative volume (V) is replicated for each group. The average trading volume of profit
(loss) firms during the reporting period is benchmarked on the average trading volume of
profit- (loss-) reporting firms during nonreporting periods.
Fig. 3 shows the average volume across all profit-reporting observations for each week
during the 7-week report period (3 weeks before and 3 weeks after the report release week,

Fig. 2. Volume analysis (all firms).

210

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

Fig. 3. Volume analysis (profit firms).

as well as the report release week itself). The average trading volume for profit-reporting
firms during the nonreport release period is 0.13%, which is lower than that for the total
sample. The average volume during the week of the release of the annual report of profit
firms is 0.175%, which is also lower than that for the total sample, but is 34.6% larger than
that during the nonreport release period and is the largest during the reporting period.
Consistent with the evidence reported for the total sample, this finding clearly shows that
individual Saudi investors do shift their portfolio positions at the time of the annual report
release week of profit-reporting firms.
Fig. 4 shows the average volume across all loss-reporting observations for each week
during the 7-week report period (3 weeks before and 3 weeks after the report release week,
as well as the report release week itself).
The average trading volume for loss-reporting firms during the nonreport release period is
0.23%, which is significantly higher than that for the total sample and for the profit group.
The average volume during the week of the release of the annual report of loss-reporting

Fig. 4. Volume analysis (loss firms).

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

211

firms is 0.22%, which is higher than that for the total sample and for the profit group (mostly
because they represent some of the largest Saudi companies), but is 5% lower than that
during the nonreport release period. Interestingly, all of the weekly trading volume averages
during the reporting period are lower than that during the nonreport release period. This
evidence stands in contrast to that reported for the total sample and for the profit group and
clearly shows that individual Saudi investors do not shift their portfolio positions at the time
of the annual report release week of loss-reporting firms. In the spirit of Hayn (1995), this
finding may suggest that Saudi investors perceive reported losses by Saudi firms to be
temporary. However, because there are no tax implications for losses on investments in
shares by Saudi investors, the reported evidence is more consistent with the view that Saudi
investors do not have any incentive to liquidate their holdings in loss firms.
In sum, it appears that the publication of accounting earnings in newspapers does not
provide decision-relevant information that will cause significant revision to the market
assessment of the future cash flows of Saudi firms. On the other hand, it appears that the
publication of accounting earnings provides decision-relevant information that leads
individual investors to revise their security holdings. However, this evidence is limited
to cases where firms reported profit. It appears that reported losses do not lead Saudi
investors to revise their security holdings, mostly because of the lack of any tax incentive
for holders of investments in loss firms.
4.3. The timeliness of accounting earnings in Saudi Arabia
The left-hand side of Table 5 presents the empirical results for the regression outlined in
Eq. (3) for the total sample. To ensure that the inferences made are not affected by any
inefficiency caused by heteroscedasticity, all t statistics are calculated after correcting for
the heteroscedasticity in the manner described by White (1980). The regression using the
pooled sample yields an estimated coefficient for earnings levels of 1.31 (significant at the
1% level) and earnings changes of 0.302 (insignificant at conventional levels). The yearly
Table 5
Regression of annual security returns on earnings and earnings changes
Reti;t a b1 Ei;t =Pi;t1 b2 DEi;t =Pi;t1 ei;t
Total sample
Year

Pooled
.028
1996
.011
1997
.197***
1998  .191***
1999
.141***

Total sample excluding electricity companies

b1

b2

Adjusted N
R2

1.310***
0.302
.226
1.350***
0.174
.380
1.067***  1.105*
.226
0.124
0.563  .016
2.128***  0.250
.485

202
50
51
51
47

Year

Pooled  .030
1996  .021
1997
.116***
1998  .270***
1999
.104**

E: earnings per share scaled by beginning-of-period price.

DE: changes in earnings per share scaled by beginning-of-period price.
Ret: annual security returns.
* White-based significant at the 10% level.
** White-based significant at the 5% level.
*** White-based significant at the 1% level.

b1

b2

1.939***
0.156
1.600***
0.166
1.822***  1.090**
1.419***
0.173
2.771***  0.634

Adjusted N
R2
.345
.408
.458
.352
.485

186
46
47
47
43

212

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

coefficients for earnings levels are also significant in every year, except 1998. On the other
hand, none of the yearly coefficients for earnings changes is significant at the 5% level.
The adjusted R2 statistics are quite high in every year, except 1998.
The right hand side of Table 5 presents the empirical results for the same regression after
excluding all Saudi electricity companies. These observations are excluded for two reasons.
First, during December 1998, the Saudi government outlined a plan to merge all electricity
companies. The announcement of this plan positively affected the raw security returns of
these firms during 1998 and 1999. Second, the Saudi government has consistently provided
generous subsidies to all electricity companies and, especially, during periods of reported
losses. During 1998, all of the electricity companies reported significant losses, yet their raw
security returns were positive (even after adjusting for market returns), possibly reflecting a
view that the reported losses will be covered by governmental subsidies.
The adjusted R2 from the pooled regression is 34.5% compared with the adjusted R2 of
22.6% from the equivalent regression for the total sample. For the year-by-year
regressions, the adjusted R2 statistics are a great deal higher than those for the total
sample, including 1998. The regression using the pooled sample yields an estimated
coefficient for earnings levels of 1.94 (significant at the 1% level) and earnings changes of
0.156 (insignificant at conventional levels).7
We examine whether the timeliness of accounting earnings is affected by the degree of
governmental ownership. We split our sample into two groups: the first group (Group A)
includes 18 firms where governmental ownership equals or exceeds 20% and the second
group (Group B) includes 34 firms where governmental ownership is below 20%. Group
A includes all electricity companies as well as certain banks and most cement companies,
among others. The adjusted R2 for the regression outlined in Eq. (3) for Group A is 12%
and the adjusted R2 for the same regression for Group B is 39.2%. This evidence is
consistent with our view that the degree of timeliness of accounting earnings is driven by
the proportion of ownership by the Saudi government.
Table 6 presents the empirical results for the regression of contemporaneous returns on
earnings levels and changes, as outlined in Eq. (3), for 2- and 5-year pooled data for the
total and for the reduced samples (after excluding the electricity group). Given the
observed influence of the electricity group on the reported findings of the annual
regressions, the discussion of the estimates reported in Table 6 focuses solely on those
for the reduced sample.
Table 6 shows that the adjusted R2 statistic obtained from the pooled regression for the
2-year return interval is 63%. For the two samples of 2-year return intervals, the average
value of the adjusted R2 statistics is 59.3%. These adjusted R2 statistics are significantly
larger than that obtained for the pooled annual regression (34.5%) and the mean adjusted
R2 for the individual annual regressions (42.57%). Clearly, these statistics suggest that
doubling the measurement interval significantly improves the association between
earnings levels and security returns. Table 6 further indicates that the adjusted R2 obtained
from the pooled earnings level regression for the 5-year return interval is 73.7%, which is
7
We conduct the same empirical test after controlling the effect of the scale variable {1/Pj,t  1} by including
it as an additional explanatory variable. In fact, the findings do not reveal any major differences compared with
the original tests.

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

213

Table 6
Regression of security returns on earnings and earnings changes (2- and 5-year analysis)
Reti;t a b1 Ei;t =Pi;t1 b2 DEi;t =Pi;t1 ei;t
Total sample
Year

2- Year
.073**
pooled
1996 1997 .175***
1998 1999 .031
5 Year
 .141

Total sample excluding electricity companies

b1

b2

Adjusted N Year
R2

1.306*** 0.368 .394

1.319*** 0.209 .462
0.721** 1.986* .234
1.728*** 1.381 .575

99 2 Yearly  .056
pooled
51 1996 1997 .050
47 1998 1999  .171***
47 5-yearly
 .379***

b1
2.047***

b2

Adjusted N
R2
0.333 .629

91

1.839***  0.029 .645

2.492***  0.129 .542
2.237*** 1.402 .737

47
43
43

E: earnings per share scaled by beginning-of-period price.

DE: changes in earnings per share scaled by beginning-of-period price.
Ret: annual security returns.
* White-based significant at the 10% level.
** White-based significant at the 5% level.
*** White-based significant at the 1% level.

significantly larger than the adjusted R2 obtained for the pooled 2-year earnings level
regression (63%).8
Clearly, these statistics suggest that increasing the measurement interval significantly
improves the association between earnings and security returns of Saudi firms. This
evidence is consistent with the proposition of Easton et al. (1992) that errors in
aggregate earnings are likely to become less important for longer periods of aggregation.
All of the reported adjusted R2 statistics are comparable with those reported in the
United States (see, e.g., Easton & Harris, 1991; Easton et al., 1992) and suggest that
reported earnings by Saudi firms are timely in that they reflect current period economic
income as proxied for by security returns. The closeness of the reported statistics for the
individual coefficients and for the adjusted R2 to those reported in the Unied States
indicates that Saudi investors do not appear to be discounting the reported earnings figures.
4.4. The differential implications of positive and negative earnings on the relationship
between earnings and security returns of Saudi firms
Table 7 presents a breakdown of the annual frequency and percentage of loss-reporting
firms in Saudi Arabia during the sample period 1995 1999. It is clear that the incidence of

8
We check for the impact of dirty surplus items in the owner-equity section of Saudi companies by
reviewing all the hard copies of the financial statements of our sample during the period 1995 1999. Indeed, we
find that the owner-equity section of Saudi firms does not include dirty surplus items that can violate the clean
surplus assumption. In other words, Saudi firms account for dirty surplus items in their income statements. This
finding reveals that net income (earnings after zakat and income tax) is identical to the comprehensive income of
all our sample firms. Furthermore, we conduct the same empirical tests after substituting the earning before
Zakat and income tax with comprehensive income to examine the affect of using comprehensive income as
the earnings variables. The empirical results were identical to these reported using earnings before Zakat and
income tax as the proxy for earnings.

214

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

Table 7
Frequency of losses
Year

No. of Firms

No. of loss

% of loss

1995
1996
1997
1998
1999
All years

52
52
52
52
48
256

12
14
10
14
14
64

23.00
26.90
19.20
26.90
29.10
25.00

loss is quite common in Saudi Arabia in that it ranges from 19.2% in 1997 to 29.1%
in 1999.
As can be seen in Table 8, the adjusted R2 statistic for the earnings level and the
earnings change model for all firms is 22.6%. When loss-reporting firms are excluded, the
adjusted R2 statistic increases from 27% to 28.7%. The adjusted R2 statistic based
exclusively on loss-reporting firms is only 3.5%. This pattern is consistent with that
reported in a U.S. context (see, e.g., Hayn, 1995). Hayn (1995) notes that losses are likely
to be considered temporary because shareholders can always liquidate their investments in
the firm rather than suffer from indefinite losses. This argument assumes that losses are
likely to be recurring and/or have tax consequences. Hayn uses a time series analysis to
show a constant decline in adjusted R2 statistics as the frequency of losses for a given firm
increases. Given the limited data in the Saudi context, time series analysis is not feasible.
Accordingly, we examine this issue by observing the adjusted R2 statistics for four
different subgroups constructed on the basis of the sign of reported earnings as well as the
sign of the reported change in earnings.
As can be seen in Table 8, in firms that report a profit as well as a positive increase in
earnings, the adjusted R2 statistic increases to 29.6%. For firms that report a profit
accompanied by a decrease in earnings, the adjusted R2 statistic drops to 20.1%.
Interestingly, for firms that report a loss and a decrease in earnings, the adjusted R2
statistics increases to 30.9%. While our tests differ in form from those conducted by Hayn
Table 8
Adjusted R2 statistics based on regression of security returns on earnings and earnings changes (profit and loss
analysis)
Reti;t a b1 Ei;t =Pi;t1 b2 DEi;t =Pi;t1 ei;t
Reported earnings

Changes in earnings

Adjusted R2 (%)

Profit
Profit
Profit
Loss
Loss
Loss
All observations

Positive and negative

Positive only
Negative only
Positive and negative
Positive only
Negative only

152
88
64
50
19
31
202

28.70
29.60
20.10
3.50
0.00
30.90
22.60

E: earnings per share scaled by beginning-of-period price.

DE: changes in earnings per share scaled by beginning-of-period price.
Ret: annual security returns.

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

215

(1995), this evidence may reflect the lack of tax incentive argument noted earlier rather
than the liquidation option argument as noted by Hayn.

4.5. Conservatism in accounting earnings in Saudi Arabia

We define conservatism in the sense of Basu (1997) as the extent to which current
period accounting earnings asymmetrically incorporate economic losses, relative to
economic gain. As can be seen in Table 9, the b2 slopes for the total sample and for
the reduced sample are significant for the annual, 2- and 5-yearly samples, suggesting
that accounting earnings in Saudi Arabia exhibit a high sensitivity to economic income.
In contrast, the incremental negative return slopes b3 are consistently insignificant,
suggesting that accounting earnings in Saudi Arabia exhibit low or no sensitivity to
economic loss.
The lack of income conservatism in Saudi Arabia is consistent with results reported for
other code-law countries (see Ball, Kothari et al., 2000; Ball, Robin et al., 2000). Thus,
despite adopting Anglo-American GAAPs, Saudi firms do not appear to be fully adhering
to them in practice. In fact, similar with the reporting system in Japan, one-time accounting
write-offs are rare in Saudi Arabia. In addition, most Saudi firms do not report any pension
liabilities or postretirement liabilities. Furthermore, the banking sector is known for not
incorporating economic losses into reported income.
Similar with Ball, Robin et al. (2000), we interpret these results on the grounds that
Saudi managers and auditors may have a low incentive to incorporate economic losses into
accounting earnings due to a reduced market demand for accounting information, low
levels of public debt, low expected costs arising from stockholder and creditor litigation,
and weak monitoring by analysts and other stakeholders.

Table 9
Contemporaneous association between earnings and returns measuring the extent to which current period
accounting earnings asymmetrically incorporates economic losses, relative to economic gain (annual, 2-, and 5yearly intervals)
Ei;t =Pi;t1 a b1 RDi;t b2 Reti;t b3 Reti;t RDi;t ei;t
Total sample

Total sample, excluding electricity companies

Year

b2

b3

Adjusted
R2

Year

b2

Pooled
annual
Pooled
2-yearly
5 Yearly

0.197***

 0.060

.173

253

0.160***

0.399***

 0.133

.407

99

0.437***

 0.121

.583

47

Pooled
Annual
Pooled
2-Yearly
5-yearly

b3

Adjusted
R2

0.024

.239

233

0.289***

0.029

.621

91

0.329***

 0.013

.713

43

E: earnings per share scaled by beginning-of-period price.

Ret: security returns less mean sample return.
RD: a dummy variable equaling one if Ret is negative and zero otherwise.
*** White-based significant at the 1% level.

216

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

5. Concluding remarks
Motivated by the work of Ball, Kothari et al. (2000), Ball, Robin et al. (2000), this
study provides empirical evidence on the decision relevance and timeliness of earnings in
Saudi Arabia. We assess the ability of accounting earnings to convey decision-relevant
information by examining the impact of earnings announcements on security returns and
the trading volume of Saudi firms. We assess the timeliness of accounting earnings by
examining the contemporaneous association between earnings levels, earnings changes,
and security returns of Saudi firms over 1-, 2-, and 5-year intervals. The results of the
empirical tests suggest that the publication of accounting earnings in Saudi Arabia is not
decision relevant, but that earnings figures are timely in terms of their association with
security returns over 1-, 2-, and 5-year intervals. The tests also show that positive and
negative earnings have differential implications for the timeliness of accounting earnings.
However, further tests show that this evidence is not consistent with the loss-liquidation
argument (Hayn, 1995) and may potentially reflect the lack of tax incentives to liquidate
investments in loss firms. Additional tests show that accounting earnings exhibit low
sensitivity to economic losses consistent with the view that Saudi managers and auditors
have low incentives to incorporate economic losses into accounting earnings in a timely
manner.

Acknowledgements
We thank the Editor (Rashad Abdel Khlaik) and three anonymous referees for their
valuable and substantive comments on an earlier draft of this paper. We also thank Peter
Easton, Doug Hanna, Frank Selto, Richard Lee, Matthew Pinnuck, Julian Yeo, and the
workshop participants at the University of Melbourne for their valuable input on earlier
drafts.

References
Al-Bogami, S.A., Green, C.D., & Power, D.M. (1997). The share price response to interim financial reports in
less-developed countries: The case of the Kingdom of Saudi Arabia. Accounting Research (Arabic), 1(5),
465 504.
Ball, R., & Brown, P. (1968, Autumn). An empirical evaluation of accounting income numbers. Journal of
Accounting Research, 159 178.
Ball, R., Kothari, S.P., & Robin, A. (2000). The effect of institutional factors on properties of accounting
earnings: International evidence. Journal of Accounting and Economics, 29, 1 51.
Ball, R., Robin, A., & Wu, J. (2000). Incentives versus standards: Properties of accounting income in four East
Asian countries, and implications for acceptance of IAS. Working Paper, University of Rochester.
Basu, S. (1997). The conservatism principle and the asymmetric timeliness of earnings. Journal of Accounting
and Economics, 24, 3 37.
Beaver, W. (1968). The information content of annual earnings announcements. Journal of Accounting
Research, (Supplement), 67 92.
Brown, P. (1970). The impact of the annual net profit report on the stock market. The Australian
Accountant, (July), 277 283.

M. Al-Sehali, N. Spear / International Journal of Accounting 39 (2004) 197217

217

Chen, C.J., Chen, P.S., & Su, X. (1999). Is accounting information value relevant in the emerging Chinese stock
market? Working Paper, Accounting Research Network.
Choi, J., & Choe, C. (1998). Explanatory factors for trading volume responses to annual earnings announcements: Evidence from the Korean stock market. Pacific-Basin Finance Journal, 6, 193 212.
Easton, P., & Harris, T. (1991). Earnings as explanatory variable for returns. Journal of Accounting Research, 29,
19 36.
Easton, P., Ohlson, J., & Harris, T. (1992). Aggregate accounting earnings can explain most of security returns.
Journal of Accounting and Economics, 15, 119 142.
Ernst, & Young International (1998). Doing business in Saudi Arabia. Riyadh: Whinney Murray and Co.
Hayn, C. (1995). The information content of losses. Journal of Accounting and Economics, 20, 125 153.
Ministry of Commerce (1986a). Accounting objectives and concepts and general standards of presentation and
disclosure. Riyadh, Saudi Arabia: Author.
Ministry of Commerce (1986b). Auditing standards. Riyadh, Saudi Arabia: Author.
White, H. (1980). A heteroskedasticity consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica, 48, 817 838.