Delhi Business Review X Vol. 9, No.

1 (January - June 2008)

21
POR POR POR POR PORTFOLIO OPTIMISA TFOLIO OPTIMISA TFOLIO OPTIMISA TFOLIO OPTIMISA TFOLIO OPTIMISATIO TIO TIO TIO TION IN THE INDIAN STOCK N IN THE INDIAN STOCK N IN THE INDIAN STOCK N IN THE INDIAN STOCK N IN THE INDIAN STOCK
MARKET INDUSTR MARKET INDUSTR MARKET INDUSTR MARKET INDUSTR MARKET INDUSTRY SECTOR AN Y SECTOR AN Y SECTOR AN Y SECTOR AN Y SECTOR ANAL AL AL AL ALY YY YYSIS SIS SIS SIS SIS
R RR RRak ak ak ak akesh Gupta* esh Gupta* esh Gupta* esh Gupta* esh Gupta*
Parikshit K. Basu** Parikshit K. Basu** Parikshit K. Basu** Parikshit K. Basu** Parikshit K. Basu**
IVERSIFICATION always reduces non-systematic risk within a portfolio to a certain extent.
At the same time, selection of individual items or industry sector influences returns. Optimum
portfolio selection within a capital market is primarily based on the best risk-return trade-off
among the industry sectors. Literature suggests that much of market volatility can be attributed to
substantial increase in sector specific and sub-sector specific risks. Performance of the economy influences
industry sector returns differently and changes over time periods. Thus, changing pattern of
correlations between sectors is vital for portfolio optimization purpose. The present study has estimated
the dynamics of correlations of stock market returns between industry sectors in India using
Asymmetric DCC GARCH model and tested efficient portfolios that generates returns above the market
average. Use of Asymmetric DCC GARCH model helps in capturing the dynamics of correlations and
as such a better estimate of the expected future correlations resulting in a portfolio that reflects future
outcome more closely. If the risk adjusted returns of the industry selected portfolio is more than the
index returns, we could argue that there is value in industry selection and accurate measurement of
the correlations help generate these excess returns. Using daily market data for the period April 1997
to April 2007 on a sample of 10 industry sectors selected randomly indicates that investors can
substantially improve their reward to risk as compared with the market returns. Sharpe ratio of the
optimised portfolio improves to 0.994 (for optimised portfolio) from 0.527 (for S&P Nifty index).
Key Words: Portfolio Management, Portfolio Optimsation, Efficient Portfolios.
Introduction
Background
Diversification always reduces non-systematic risk within a financial assets portfolio to a certain
extent. Any textbook dealing with the area of investments includes basic analysis on portfolio
diversification. Common objective of financial investors is to achieve an optimal risk-return
combination. It can be achieved either by maximising return with an accepted level of risk or by
minimising risk with an acceptable rate of return. Diversification impacts risk component of the
portfolio in particular. It implies a spread of investments and allows a middle road through the
highs and lows of market performance. In other words, diversification allows an opportunity for
investments to grow with minimum volatility. Securities behave differently from one another within
the same market based on its own performance, industry/sector conditions, national and international
factors and so on. Literature suggests that much of market volatility can be attributed to substantial
increase in sector specific and sub-sector specific risks (Black et al. 2002).
D
* Lecturer, School of Commerce, Faculty of Business and Informatics, Central Queensland University, Bruce Highway, Rockhhampton,
QLD 4702, Australia.
** Senior Lecturer, School of Marketing & Management, Charles Sturt University, Bathurst, NSW 2795 Australia.
Rakesh Gupta and Parikshit K. Basu
22
In the last three decades, a large number of countries had initiated financial reform process to open
up their economies and integrated into the global economy. India is one of the late entrants the
reform process officially started in 1991 only. The Indian stock market is possibly one of the oldest
in Asia, but remained at a small scale and largely outside the global integration process until the
late 1980s (Wong et al. 2005). The major stock market in India located in Mumbai (formerly known
as Bombay) has always played the dominant role in the equity market in India. It has been
traditionally governed by brokers leading to conflict of interest situation between the interest of
common investors and those of brokers/owners of stock exchanges (Datar and Basu 2004). Reforms
in equity market in India commenced slightly earlier than the overall reforms - in mid-1980s. With
the establishment of National Stock Exchange (NSE), a new institutional structure was introduced
in India that could ensure smooth functioning of market through a combination of new technology
and efficient market design. The Securities Exchange Board of India (SEBI) was set up as a market
regulator with statutory powers to control and supervise operations of all participants in the capital
market viz. stock exchanges, stock brokers, mutual funds and rating agencies. The development of
debt market is another significant development, which has been facilitated by deregulation of
administered interest rates. Opening of stock exchange trading to Foreign Institutional Investors
(FIIs) and permission of raising funds from international market through equity linked instruments
have introduced a degree of competition to domestic exchanges and other market participants.
Coverage of stock exchanges in India as primary and secondary markets is expanding steadily.
NSE and BSE ranked 3
rd
and 6
th
among the stock exchanges in the world in 2006 in terms of
number of transactions (GOI 2007).
Interestingly, stability in prices for the Bombay Stock Exchange (BSE) in India was considered to
be an important feature. During the period 1987 to 1994, average annual price fluctuations of
ordinary shares on BSE were 25.1% as compared with London Stock Exchange (22%) and the New
York Stock Exchange (23.9%) (Poshakwale 1996). Regarding efficiency and level of global integration
of the Indian stock market, findings are diversified and often contradictory. An early empirical
study based on 11 years data 1963-73 even found evidence of random walk in BSE stocks that could
be compared with behaviour of stock prices in the markets like LSE and NYSE (Sharma and
Kennedy 1977). Wong et al. (2004) observed that the Indian stock market is integrated with mature
markets and sensitive to their dynamics in the long run. In the short-run, the US and Japan
markets influenced the Indian market but not vice versa. In particular, the Indian stock markets
appear to have integrated more with the global markets since late 1990s due to lower barriers on
foreign investments and growing presence of Indian firms in the international markets (Banerjee
and Sankar 2006). However, more recent data covering the period 1991-2006 for two major equity
markets in India (BSE and NSE) found no evidence of random walk model (Gupta and Basu 2007).
Thus, evidences are contradictory. In any case, with or without explicit validity of random walk
model, benefits of diversification cannot be denied. There is no study on benefits of diversification of
the Indian stock market.
Performance of the economy influences industry sector returns differently and changes over time
periods. Studies found strong evidence of relationships between performances of the Indian Stock
Market and macroeconomic variables such as inflation, domestic output growth and macroeconomic
management (Naka et al. 1999). After experiencing very moderate growth till 1980s India largely
liberalised its economy in the early 1990s. The reforms undertaken by the Indian government in
1991 have resulted in an increase in the economic growth. In the last three years (2004-05 to 2006-
07), average annual real GDP growth rate in India was about 8.4%. However, sectoral growth
rates varied significantly during the period. Services contributed more than two-thirds of this GDP
growth and the entire residual contribution came from industry. As a result, in 2006-07, the share
of agriculture in GDP declined to 18.5% and the share of industry and services improved to 26.4%
Delhi Business Review X Vol. 9, No. 1 (January - June 2008)
23
(GOI 2007). Within the services sector, IT and IT-enabled activities, trade, hotels, transport and
communication services performed much better than the others. Within the industry sector,
manufacturing, textile products, chemicals, non-electrical equipments etc. performed better than
the average in recent years (GOI 2007). Accordingly, stocks from different sectors and sub-sectors
also performed differently during the period and need to be viewed with proper care. Each sector is
unique in its own way and so are the companies operating in each sector. All securities behave
differently from one another even within a sector. Still there exist some commonalities within
each sector. Because investments in each sector react differently to market conditions and other
factors, in general it may be worthwhile to maintain a sector diversified portfolio in order to balance
out the ups and downs and to optimise the portfolio. Again, to obtain diversification benefits
correlation between the two sets assets must be less than perfect and should be considered in a
dynamic setting. Thus, changing pattern of correlations between sectors is vital for portfolio
optimization purpose.
Research Objectives
The present study has made an attempt to estimate the dynamics of correlations of stock market
returns between industry sectors in India using Asymmetric DCC GARCH model and tested efficient
portfolios that generates returns above the market average. Use of Asymmetric DCC GARCH
model helps in capturing the dynamics of correlations and as such a better estimate of the expected
future correlations resulting in a portfolio that reflects future outcome more closely. If the risk
adjusted returns of the industry selected portfolio is more than the index returns, we could argue
that there is value in industry selection and accurate measurement of the correlations help generate
these excess returns.
Research Methodology
DCC Model
In this study we estimate the time varying correlations using the Asymmetric Dynamic Conditional
Correlation (DCC) model of Cappielo, Engle and Sheppard (2006). This model is an introduction of
asymmetric term in original DCC model of Engle (2002) as modified by Sheppard (2002) as a
general model. The conditional correlation between two random variables r
1
and r
2
that have mean
zero can be written as:
one. of variance a and mean zero has that
e disturbanc ed standardis a is
where 2, 1, i , ) ( h Let
) ( ) (
) (
t i, , ,
2
, 1 1
t i,
2
, 2 1
2
, 1 1
, 2 , 1 1
, 12

= = =
=

for h t andri r E
r E r E
r r E
t i t i t t
t t t t
t t t
t
(1)
Substituting the above into equation (1) we get:
) (
) ( ) (
) (
, 2 , 1 1
2
, 2 1
2
, 1 1
, 2 , 1 1
, 12 t t t
t t t t
t t t
t
E
E E
E

= =
(2)
Using GARCH(1,1) specification, the covariance between the random variables can be written as:
) ( ) (
12 1 , 12 12 1 , 2 1 , 1 12 , 12
+ + =
t t t t
q q (3)
Rakesh Gupta and Parikshit K. Basu
24
The unconditional expectation of the cross product is while for the variances

12
= 1
The correlation estimator is:
t t
t
t
q q
q
, 22 , 11
, 12
, 12

(4)
This model is mean reverting if + < 1. The matrix version of this model is written as:
Q
1
= S(1 ) + (
t1

t1
) + Q
t1
(5)
Where S is the unconditional correlation matrix of the disturbance terms and
t t
q Q
, 2 , 1
= .The log
likelihood for this estimator can be written as:
) log log 2 ) 2 log( (
2
1
1
1
t t t t
T
t
t
R R D n L

=
+ + + =
(6)
Where { }
t i t
h diag D
,
= and R
t
is the time varying correlation matrix.
As this model does not allow for asymmetries and asset specific news impact parameter, the modified
model Cappiello, Engle and Sheppard (2006) for incorporating the asymmetrical effect and asset
specific news impact is:
G n n G B Q B A A G N G B Q B A Q A Q Q
t t t t t t 1 1 1 1 1
) (

+ + + = (7)
Where A, B and G are diagonal parameter matrixes, n
t
= I[
t
< 0]o
t
(with o indicating Hadamard
product), [ ]
t t
n n E N = . For Q and
N
expectations are infeasible and are replaced with sample
analogues,

T
t
t t
T
1
1
and

T
t
t t
n n T
1
1
, respectively. ] [ ] [
,
*
,
*
t ii t ii t
q q Q = = is a diagonal
matrix with the square root of the i
th
diagonal element of Q
t
on its i
th
diagonal position. In this paper
we only look for the asymmetrical effects and not the asset specific news impacts.
Efficient Portfolios
The efficient frontier is defined as the set of portfolios that exhibit the minimum amount of risk for
a given level of return or the highest return for a given level of risk and lies above the global
minimum variance portfolio. Elton, Gruber and Padberg (1976) show one is able to use a simple
decision criterion to reach a optimal solution to the portfolio problem by assuming that a risk-free
asset exists and either the single index model adequately describes the variance-covariance structure,
or that a good estimate of pair wise correlations is a single figure. This simple criterion not only
allows one to determine which stocks to include but how much to invest in each.
Delhi Business Review X Vol. 9, No. 1 (January - June 2008)
25
The first approach utilises the single-index model to construct optimal portfolios. Where returns
are determined as follows:

of variance and zero of mean with error term random the is
R in change a given R in change expected the measures hat constant t a is
e performanc s market' the of t independen is that i security on return the is
index market on the return the is R
i security on return the is R where
2
m i
m
i
i

i
i
i
i m i i i
R R + + =
Assuming that short selling is possible, the task would be to find the unconstrained vector of
relative weights for each security so that the Sharpe ratio is maximised. That is:
portfolio on the return the of deviation standard the is
portfolio on the return mean the is R where
R R
ratio, Sharpe the
maximise o security t each on s ' X weights, relative the find To
p
f
p
i
p
p

=
(8)
Given that
( )
(10)
(9)

R X R X
and
R R X R R
1 1 1
2 2 2 2 2 2 2
2
1 1
i
i i i
2
1
f
p
i f
p
(
(

+ + =
|
.
|

\
|
=
=

= = =
= =
=
N
i
N
i
N
j
i m j i j i m i i p
N
i
N
i
p
N
i
i
X X X X
E

These equations are substituted into the Sharpe ratio equation and in order to maximise the Sharpe
ratio it is necessary to take the derivative of the Sharpe ratio with respect to each
i
X and set it
equal to zero. The derivation yields the amount of the portfolio that should be invested
0
i
X in any
security as:
Rakesh Gupta and Parikshit K. Basu
26
(11)
1
C where
) (
) (
1
2
2
2
1
2
2
0
1
2
0
2
0
0

=
=
=
+

=


=
N
i
j
m
N
i
j
f
j
m
N
i
i f
j
i f
j
i
j
j
j
j
R R
C R R
C R R
X

Thus by applying the above equation then one is able to determine the respective weightings for
each security within the portfolio and to find the optimal portfolios risk and return measures. That
is, the risk and returns are obtained by substituting the respective weights found for each security
into the returns and variance formula given in (9) and (10) respectively
1
.
Data Requirement
For this study we use monthly returns of the National Stock Exchange (NSE) and the monthly
returns of the different sector indices for the period April 1997 to April 2007. In order to calculate
the volatility of the respective index, we use daily prices to calculate the daily returns and the daily
average volatility of each market index returns. We calculate monthly volatility (Volatility
m
=
Daily volatility X vn, where m represents period and n number of trading days in the period) of
each market on the basis of actual number of trading days in the month for the emerging market.
We use DataStream for index values of the respective equity indexes. Indexes included in the study
are; Abrasives, Air conditioners, Automobiles 2 wheeler, Aluminium, Construction, Durables,
Refinery, Software, Synthetic Textiles and Trading.
Table 1 lists sector returns and their summary statistics for the 10 market sectors. Mean daily
returns of the market sectors included in this study varies from 0.5% for Durables to 1.6% for
Software sector. There are major differences in the minimum and maximum daily returns for the
sectors and as such major differences in the variances.
Analysis of results
The unconditional correlations between sectors for the sample period are presented in Table 2.
Lowest correlations were observed between Abrasives and Software, 0.167 for the sample period
and highest between Aircondioners and Durables, 0.472.
1 The model is for no restrictions on short sale, the standard optimization problem can be written as:

=

= =
+ =
N
i
N
i k
k
k i k i k i i
N
i
i P
X X X Min
1 1
,
2
1
2 2

subject to the constraint:
1
1
=

=
N
i
i
X
if the short selling is not
allowed the additional constraint will be of non-negative weights, expressed as 0

X
1

1. Similar minimum or maximum weight

restrictions are imposed when introducing restrictions of minimum investments into Australian index and or maximum restrictions
into emerging market indexes.
Delhi Business Review X Vol. 9, No. 1 (January - June 2008)
27
The focus of this paper is to use conditional correlation estimates (as against unconditional ones) as
being more accurate. These correlations are estimated using Asymmetric Dynamic Conditional
Correlation Model (ADCC GARCH) as discussed earlier in the methodology section. Table 3 presents
the conditional correlations calculated using ADCC GARCH model. Lowest correlations of 0.1221
are between Abrasives and Aluminium and highest of 0.5747 between Construction and Refinery.
Direct comparison of unconditional correlations from table 2 and conditional correlations in table 3
is not meaningful as the argument in favour of conditional correlations lies in theoretical strengths
of Asymmetric DCC GARCH model which is expected to produce better estimates for correlations
as the model allows for the correlations to change over time.
In the next step we look at the mean returns and the Sharpe ratios for the NSE index. These have
been used to compare the portfolio returns to assess if the portfolio constructed using the model
performs better than the broad based index. Table 4 includes the portfolio mean returns for National
Stock Exchange (NSE) of India. We will compare the returns of the optimal portfolio using sector
indexes for the market and conditional correlations estimated using ADCC GARCH model. India
only portfolio has a mean return of 18.91% with a standard deviation\ of 24.48% and Sharpe ratio
of 0.52 based on a risk free rate of 6% and a 50% probability of achieving the target mean returns.
Table 4: S & P Nifty Index
Risk free return 6%
Mean annual returns 18.91%
Standard deviation 24.48%
Sharpe ratio 0.52
Probability of achieving mean returns 50.00%
The main focus of the paper is the use of computationally efficient model
2
for estimating the
correlations. As such we use conditional correlations in optimisation model to construct optimised
Table 1: Summary Statistics
Sector Observations Mean Skewness Variance Min Max
Abrasives 2608 0.0011 1.841 0.0005 -0.170 0.358
Air conditioners 2608 0.0013 0.272 0.0005 -0.142 0.125
Aluminium 2608 0.0005 0.028 0.0005 -0.127 0.108
Automobiles 2608 0.0007 0.154 0.0003 -0.118 0.097
Construction 2608 0.0013 0.259 0.0006 -0.171 0.199
Durables 2608 0.0005 2.922 0.0006 -0.144 0.130
Refinery 2608 0.0006 -0.023 0.0005 -0.152 0.110
Software 2608 0.0016 -0.059 0.0007 -0.198 0.160
Synthetic Textiles 2608 0.0006 0.230 0.0004 -0.127 0.115
Trading 2608 0.0010 0.207 0.0005 -0.118 0.178
2 Results for portfolios with unconditional correlations are not presented here but can be reqested from the authors.
R
a
k
e
s
h

G
u
p
t
a

a
n
d

P
a
r
i
k
s
h
i
t

K
.

B
a
s
u
2
8
Table 2: Unconditional Correlations
Abrasives Aircon Alum Autos2 Construc Durables Refinery Softwa SyntTex Trading
Abrasives 1.000
Aircon 0.219 1.000
Alum 0.205 0.331 1.000
Autos2 0.189 0.339 0.392 1.000
Construc 0.264 0.354 0.376 0.378 1.000
Durables 0.232 0.472 0.399 0.414 0.426 1.000
Refinery 0.232 0.377 0.437 0.386 0.441 0.443 1.000
Softwa 0.167 0.357 0.299 0.329 0.360 0.446 0.323 1.000
SyntTex 0.251 0.442 0.426 0.398 0.436 0.532 0.429 0.357 1.000
Trading 0.224 0.309 0.291 0.285 0.363 0.378 0.353 0.332 0.357 1.000
Table 3: Conditional Correlations (ADCC GARCH)
Abrasives Aircon Alum Autos2 Construc Durables Refinery Softwa SyntTex Trading
Abrasives 1
Aircon 0.2213 1
Alum 0.1221 0.2851 1
Autos2 0.2097 0.3823 0.2680 1
Construc 0.3550 0.3328 0.4662 0.4763 1
Durables 0.2181 0.3583 0.3455 0.4321 0.4460 1
Refinery 0.2352 0.4069 0.4081 0.4394 0.5747 0.4542 1
Softwa 0.2042 0.3756 0.3597 0.5338 0.4380 0.5653 0.5116 1
SyntTex 0.2330 0.4169 0.3054 0.4537 0.4025 0.3419 0.4169 0.3332 1
Trading 0.2369 0.3733 0.3239 0.4148 0.4782 0.4520 0.3764 0.3671 0.3479 1
Delhi Business Review X Vol. 9, No. 1 (January - June 2008)
29
portfolio and compare the returns of this portfolio with the S & P Nifty index. Proportion invested
into each market sector, annualised returns and standard deviation of the portfolio are presented
in Table 6 and the estimated Sharpe ratios are presented in Table 7.
Table 6: Indian portfolio of different market sectors using ADCC GARCH correlations
Weight Returns % Standard Deviation %
Abrasives 12.89 30.59 38.54
Airconditioning 8.05 34.44 36.93
Aluminium -9.29 13.60 36.25
Automobiles(2) -3.21 20.19 29.72
Construction 6.60 33.81 41.82
Durables -6.79 13.34 41.47
Refinery -9.94 16.91 36.22
Software 13.40 43.52 44.99
Synthetic Textiles -4.45 17.98 33.90
Trading -7.27 28.13 37.68
Portfolio 32.69 26.58
Table 7: Comparison of Sharpe Ratios
Portfolio Mean Return Sharpe Ratio* Probability
S & P Nifty 18.91% 0.527 50.00%
Optimal portfolio 32.69% 0.994 69.54
3
%
* Proxy for Risk free rate is average money market rates for April 2007, acquired from Reserve bank of
India (6.0%).
Comparisons of Sharpe ratio (Table 7) indicate that the optimised portfolio using ADCC GARCH
correlations is a superior portfolio as compared with the S & P Nifty index
4
. The mean return of
the optimised portfolio is 32.69% which is more than double than that of NSE. Standard deviations
of these portfolios are very similar, 24.48% for NSE and 37.68%. Results clearly indicate the expected
returns of the optimised portfolio can be improved without significantly compromising on the risk.
Using Sharpe ratio for objectively comparing the returns we see that Sharpe ratio improves from
0.527 to 0.994. This is a significant improvement from the NSE returns.
Conclusions
The importance of industry selection has been tested in this study using a sample of 10 industry sectors
for equity markets in India. It can further be argued that if industry selection is important so the
correlations within the industry pairs are important for portfolio optimization purpose so as to enhance
3 This is the probability of achieving the target return of 18.91% in both cases.
4 S & P Nifty is a capitalisation weighted index of 50 stocks representing 21 industry sectors from the National Stock Exchange,
India
Rakesh Gupta and Parikshit K. Basu
30
portfolio returns. If correlations within the industry are important, than an accurate estimate of
correlations becomes evident and Theoretically Asymmetric DCC GARCH estimates should provide us
with a better estimate of correlations and the results indicate correlations do change over time. On
theoretical grounds this study recommends using Asymmetric DCC GARCH model for estimates of
correlations. By using Asymmetric DCC GARCH model for estimating correlations and using these
correlations in portfolio optimization process we can enhance portfolio returns of the domestically
diversified portfolio.
References
Banerjee, S. and Sankar, S. (2006), Measuring Market Integration in Indian Stock Market, Finance India, Vol.20, No.3,
pp.895-912.
Black, A.J., Buckland, R., and Fraser, P. (2002), Efficient Portfolio Diversification: Changing UK Stock Market Sector and
Sub-Sector Volatilities, 1968-2000, Managerial Finance, Vol.28, No.9, pp.26-43.
Cappiello, L., Engle, R., and Sheppard, K. (2006), Asymmetric Dynamics in the Correlations of Global Equity and Bond
Returns, Journal of Financial econometrics, Advance Access published online on September 25, 2006.
Datar, M.K. and Basu, P.K. (2004), Financial Sector Reforms in India: Some Institutional Imbalances, Conference Volume,
Academy of World Business, Marketing and Management Development Conference, Gold Coast, Australia, July.
Elton, E.J., Gruber, M.J., and Padberg, M.W. (1976), Simple Criteria for Optimal Portfolio Selection, Journal of Finance,
Vol.31, No.5, pp.1341-1357.
Engle, R. (2002), Dynamic Conditional Correlation A Simple Class of Multivariate GARCH models, Journal of Business
and Economic Statistics, Vol.20, pp.339-350.
Government of India (2007), Economic Survey 2006-07, New Delhi: Ministry of Finance, http://indiabudget.nic.in/es2006-
07/esmain.htm (viewed in September 2007).
Gupta, Rakesh and Basu, P.K. (2007), Weak Form Efficiency in Indian Stock Markets, International Business and
Economics Research Journal, Vol.6, No.3, pp.57-64.
Naka, A., Mukherjee, T.K. and Tufte, D.R. (1999), Macroeconomic Variables and the Undervalued Indian Stock Markets,
Financial Management Association Meeting, Orlando, October, http://business.uno.edu/econ/staff/naka.html (viewed in
October 2007.
Poshakwale, S. (1996), Evidence on Weak form Efficiency and Day of the Week Effect in the Indian Stock Market, Finance
India, Vol.10, No.3, pp.605-616.
Sharma, J.L. and Kennedy, R.E. (1977), A Comparative Analysis of Stock Price Behaviour on the Bombay, London and
New York Stock Exchanges, Journal of Financial and Quantitative Analysis, September.
Wong, Wing-Keung, Agarwal, A., and Du, Jun (2005), Financial Integration for India Stock Market, A Fractional Cointegration
Approach, Departmental Working Papers wp0501, National University of Singapore, Department of Economics, http://
nt2.fas.nus.edu.sg/ecs/pub/wp/wp0501.pdf (viewed in October 2007).