-
Electronic copy available at:
http://ssrn.com/abstract=2229290
Day of the Week Effect on
Gold Returns
Research Paper
Presented by: Sunita Arora Assistant Professor
Department of Commerce Government College for Women
Rohtak
Under the Guidance of: Dr. Narender Kumar Professor
Department of Commerce M D University
Rohtak
-
Electronic copy available at:
http://ssrn.com/abstract=2229290
1
Day of the Week Effect on Gold Returns
Abstract:
Day of the week effect has vastly been investigated for stock
markets, but commodity markets have got less attention in this
regard. Present study is an attempt to investigate the day of the
week effect on gold futures return. For the purpose of study,
opening and closing prices of gold futures contract, traded on
Indias largest commodity exchange i.e. Multi Commodity Exchange of
India Limited (MCX), have been analysed with the help of dummy
variables. Daily return from close to close prices was calculated
and was further decomposed into trading period return and non
trading period return. Gold is traded on MCX six days a week, from
Mondays through Saturdays; therefore dummies for all the six days
have been applied. Period of study spans from 1st April 2006 to
31st March 2012. Data has been analysed by applying linear
regression of dummy variables on return. But after applying linear
regression, ARCH effect was found to be left. To account for ARCH
effect, GARCH(1,1) model was applied. Extreme value analysis was
also carried on with the help of chi-square test. Presence of day
of the week effect was found in all the three types of return
series, but the impact was more during non trading period than
during trading period. Results of the study may be useful for
commodity market players in timing their trading. Key Words:
Commodity market, Day of the week effect, Trading Period Return,
Non-Trading Period Return, Heteroskesdasticity. JEL Classification:
G14. Introduction: Efficient market hypothesis implies that asset
prices should move randomly and there
should not be a pattern, on the basis of which market
participants are able to earn
abnormal returns. This implies that the return for an asset in
an efficient market should
not be predictable and there should be no seasonality in the
returns of an asset i.e. return
for different periods e.g. different periods of a day, different
days of a week, different
weeks of a month and different months of a year should not be
different. If return for
different trading periods is same, return generating process is
said to be following trading
time model and if return is same for different time periods,
whether tading or non-trading,
it is said to be following calendar time model. The present
study is an attempt to
-
2
investigate the applicability of trading time model or in other
words, day of the week
effect, in commodity futures market in India and for the
purpose, precious metals
segment has been chosen for the study and metal selected is
gold.
Gold is traded 24 h a day and has been an important precious
metal for many millennia
and almost all the gold ever mined is still in existence. Demand
for gold arises from
consumers in the form of jewellery, dental fillings and other
uses; from industry as one of
the most ductile metals and as an excellent conductor of heat
and electricity; and from
central banks and investors. Gold also plays an important role
as a store of value
especially in times of political and economic uncertainty.
(Aggarwal, R. and Lucey, B.M.
2007: 218)
Multi Commodity Exchange of India Limited started operations in
November 2003 and
holds a market share of over 85% of the Indian Commodity futures
market.1 It offers
trading in different segments like bullion, metals, energy, oil
and oilseeds, plantations,
pulses, spices etc. Gold is the highest traded metal on Multi
Commodity Exchange of
India Limited (MCX). Following table shows the value of gold
traded on MCX from
2003 to 2011.
Table I Volume of Gold traded on MCX (` in lakhs)
Year Gold 2003 13364 2004 4012886 2005 17920890 2006 91844784
2007 74786092 2008 184054386 2009 207797608 2010 248477853 2011
384270982
Source:
http://www.mcxindia.com/SitePages/HistoricalDataForVolume.aspx,
accessed on 05 January 2012. Amount rounded off to nearest lakh
rupees.
1http://www.mcxindia.com/aboutus/aboutus.htm#top.accessedon21062012.
-
3
Objective of the Study: There are many studies that investigated
day of the week effect
on stock market returns of the developed countries and some
attempts have been made to
investigate day of the week effect on the returns of stock
markets of emerging countries.
But commodity markets got less attention in this regard. There
are very few studies
which investigated the day of the week effect on the commodity
markets of emerging
countries. This stimulated us to study the day of the week
effect on the return of highly
traded commodity futures contract in India i.e. gold
contract.
Review of Previous Studies: Review of the studies already done
in the field is necessary
to understand the methodology to be applied for the purpose.
Review of some studies
consulted by us is:
French, K.R. 1980, analysed closing prices of S&P portfolio
from 1953 through 1977, for
the purpose of testing- whether the return generating process
follows trading time
hypothesis or calendar time hypothesis. Period of study was
divided into five sub periods
and the results of the study show that return for Monday was
negative and lower than the
average return for any other day for each of the five sub
periods. Neither the trading time
nor the calendar time model was found to be an accurate
description of return generating
process.
Ball, C.A., Torous, W.N. and Tschoegl, A.E. 1982, analysed
morning and afternoon
fixing prices of gold in London from January 2, 1975 to June 30,
1979 and generated four
series of return- morning to morning change, afternoon to
afternoon change, overnight
change and the within day change. Results of the study show that
overnight changes are
less variable than within the day changes, daily variances are
not equal, weekend
-
4
variances are not very different from daily variances. With
regard to return variances,
trading time hypothesis was found to be better fit on data than
a calendar time hypothesis.
Ma, C.K. 1986, analysed daily gold afternoon fixings from
January 1972 to June 1985 by
dividing the period of study into two parts before October 1981
and after October 1981,
on the basis of implementation of same day settlement procedure.
The study concluded
that high Wednesday effect before 1981 was due to settlement
effect; and the return
generating process in the gold market is more consistent with
the calendar time
hypothesis. Thus a higher than usual return on the day following
a non trading period
may be expected in a bull market and a lower return in a bear
market period.
Dyl, E.A. and Maberly, E.D. 1986, analysed daily opening and
closing prices of S&P 500
from June 1, 1982 to May 17, 1985 and studied separately trading
and non trading period
returns. Study did not observe the presence of day of the week
effect for close to close
return but observed the same for the non trading period.
Gay, G.D. and Kim, T.H. 1987, analysed data on Commodity
Research Bureau Index
from September 1956 to March 1985 and divided the period of
study into two sub
periods, on the basis of enactment of Economic Recovery Act,
1981 which ends any
motivation for tax loss selling in the market. The study
observed significant returns for
Wednesdays and Fridays, highest Friday return than all other
days return with lowest
variance and for Mondays, lowest and negative return with
highest variance before 1981.
The pattern disappeared during the post 1981 time period.
Agrawal, A. 1994, studied indices data for 18 countries. Time
period for 12 countries out
of these 18 countries spans from 1971 to 1987 and for the
remaining 6 countries time
period varied. Results for USA were reported from other studies
for comparison. The
-
5
study observed that Wednesday and Friday returns were large and
significantly positive
in most of the countries. For Monday, seven countries exhibit
significant negative return.
Strong negative Tuesday effect was found in several countries
examined.
Kamath, R.R., Chakornpipat, R. and Chatrath, A. 1998, analysed
closing prices index of
stock exchange of Thailand and its ten industry classified
indices from January 1980
through December 1994 and observed negative Monday and positive
Friday returns.
Berument, H. and Kiyamaz, H. 2001, analysed closing prices of
S&P 500 from January
1973 to October 1997. Highest return was observed on Wednesdays
and lowest return
was observed on Mondays; highest volatility was observed on
Fridays and lowest on
Wednesdays.
Sarma, S.N. 2004, analysed daily return generated by SENSEX,
NATEX AND BSE 200,
from January 1, 1996 to August 10, 2002. Results of the study
show that there is
seasonality in Indian stock markets returns pattern and a
trading strategy of buying on
Mondays and selling on Fridays may be designed by the investors
of SENSEX; and for
other two indices a passive buy and hold strategy was found to
be more effective.
Kaur, H. 2004, studied day of the week effect in return as well
as in variance by including
dummies in both the equations of GARCH model for a period from
1993 to 2003. The
study was divided into two sub periods, one before the
introduction of the rolling
settlement and the other after the introduction of rolling
settlement by Bombay Stock
Exchange and National Stock Exchange. The results of the study
indicate the presence of
the day of the week effect in returns as well as in volatility
with Mondays, Tuesdays and
Fridays having lower return and Wednesdays having higher return;
and lower volatility
was observed for Tuesday.
-
6
Aly, H., Mehdian, S. and Perry, M.J. 2004, analysed closing
prices of Egyptian stock
market index from April 26, 1998 to June 6, 2001. The study
observed Monday positive
and significant return but not significantly different from
other days of the week.
Variance for Monday was found to be significantly higher than
the variance for rest of
the week.
Lucey, B.M. and Tully, E. 2006, analysed cash and futures series
of gold and silver,
traded on COMEX from January 1982 to November 2002. For
unconditional mean,
variance; seasonality in the mean, if present, was found to be
weak and not statistically
significant; and seasonality was stronger for variance than for
the mean. For conditional
mean and variance in gold and silver cash, negative Monday
effect was observed but the
same was not observed for gold and silver futures market.
Singhal, A. and Bahure, V. 2009, analysed daily closing values
of BSE SENSEX, BSE
200 and Nifty form April 2003 to April 2008 and observed lower
Monday and higher
Friday return than the return of rest of the week. The same was
existent even after some
adjustments related to investors expectations were made by the
researchers.
Rahman, M.L. 2009, analysed daily closing prices of 3 indices of
Dhaka Stock Exchange
from 04-09-2005 to 08-10-2008 and observed Thursday (last
trading day of the week at
Dhaka Stock Exchange during the period of study) mean return for
all the three indices
was positive and significant; whereas for Sundays and Mondays,
mean return was found
to be negative and significant.
Puja, P. 2010, tested the presence of day of the week effect in
aggregate indices, BSE 100
Index, BSE 500 Index, BSE-Sensitive Index, S&P CNX 500 and
Nifty, for a period from
January 1, 1990 to November 30, 2004. The results of the study
show that average return
-
7
for the Mondays, Tuesdays and Thursdays- returns for Nifty,
S&P CNX 500 are negative
and significant; and for Fridays- Nifty, S&P CNX 500 and BSE
500 have significant
negative return.
Suman and Chahal, S.S. 2011 examined the day of the week effect
on BSE Sensex from
January 1, 1999 to May 31, 2010. No day of the week effect is
observed on returns of
BSE Sensex and same is found in volatility, the highest
volatility being occurring on
Mondays.
On the basis of above literature review it can be said that no
set theory can be developed
for the presence of day of the week effect in stock as well in
commodity markets and a lot
of research is needed in this area. Present study is an attempt
in this regard.
Data for the Study: For the purpose of present study, daily
closing prices of near month
futures contract for gold, traded on Multi Commodity Exchange of
India Limited, have
been collected from the website of the exchange
(www.mcxindia.com). Near month
futures contract has been considered on the assumption that it
is the most traded futures
contract for any underlying asset. Following- French. 1980,
Ball, et al. 1982, Dyl, et al.
1994, Kamath, et al. 1998, we exclude the return for the days
following holidays during
the week. The data has been analysed with the help of MS Excel
2007, SPSS 16 and
Eviews-6.
Period of Study: Period of study spans from 1 April 2006 to 31
March 2012.
Analysis of Data: Data has been analysed by applying following
statistical and
econometric techniques:
Return: Series of daily closing prices has been converted in to
continuous daily
percentage return series by applying the following formula:
-
8
R ln P ln P 100
WhereR is the return for day t, lnis the natural log, Pt and
Pt-1 are the closing prices for
day t and its previous trading day. Daily return has further
been decomposed into trading
period return and non trading period return with the help of
following formulae:
Trading Period Return = lnclose ln 100
Where ln is the natural log, close is the closing price for day
t and is the open
price for day t.
Non trading Period Return = ln ln 100
Where ln is the natural log, is the open price for day t and is
the close
price for day t-1. Thus for non trading period return, the
impact of period from previous
close to present open is considered.
Basic Statistics: Before applying any statistical or econometric
technique on the data,
basic statistics of the data should be studied for the purpose
of knowing statistical
properties of the data. In the present study basic statistics
for data have been studied and
the results are shown in Table II, Table III and Table IV.
Stationarity of Data: Basic requirement for analyzing any time
series is that the data
must be stationary. Because all the three return series in the
present study are time series,
so these have been tested for stationarity on the basis of
graphical presentation as well as
with the help of the Augmented Dickey-Fuller (ADF) test.
The Augmented Dickey-Fuller (ADF) test consists of estimating
the following regression
(Gujarati, D.N. and Sangeetha. 2007: 817).
Y t Y
Y
-
9
If on the basis of this test, a time series is found to be non
stationary, it is transformed, to
make it stationary before applying econometric techniques on
it.
Day of the Week Effect: Day of the week effect on gold return
has been studied by
introducing dummies for different trading days of the week in
the regression equation. As
trading on Multi Commodity Exchange of India Limited is carried
on six days a week,
from Mondays through Saturdays, dummies for all the six days
were introduced. First of
all ordinary least square regression was applied with the help
of following equation:
Where 1...6 are coefficients to be estimated and M, Tu, W, Th, F
and S are dummies for
the trading days of the week. M=1 if the day is a Monday and 0
otherwise. Same process
is followed for other trading days. The hypothesis to be tested
is:
Ordinary least square equation assumes the existence of constant
variance, but if the
variance is time varying results of the equation may be
misleading. So the results of the
ordinary least square equation were tested for ARCH effect left
in the residuals. If the
same was found to be there, GARCH (1.1) model was applied on the
data to account for
ARCH effect. In GARCH (1,1) model, the return equation described
above was followed
but variance equation was:
Where is previous period error term, also known as ARCH term and
h is
previous period variance, known as GARCH term.
Extreme Value Analysis: To investigate the effect of the day of
the week on extreme
returns, 10% minimum values and 10% maximum values of return
were considered and
days for the same were observed. Non-parametric Chi-square test
was applied to test
-
10
whether there is any significant difference in days of the week
for the same. This
procedure was applied only for daily return series.
Empirical Results: Table II, Table III and Table IV show the
Descriptive Statistics for
daily return series, trading period return series and
non-trading period return series. 5%
trimmed mean has been calculated with the help of SPSS 16. 5%
trimmed mean is the
simple average for the 95% values- after deleting extreme 2.5%
both from lower area and
from upper area. Thus for 5% trimmed mean outliers are excluded
from both the sides.
Table II Descriptive Statistics for Daily Return
Statistic Overall Return
Monday Return
Tuesday return
Wednesday Return
Thursday Return
Friday Return
Saturday Return
Mean 0.067429 0.015691 0.075405 0.073069 -0.001750 0.117426
0.126942 Trimmed Mean 0.084588 0.067804 0.092849 0.098615 -0.007256
0.128034 0.103208 Median 0.081086 0.026450 0.067692 0.114561
0.050355 0.113148 0.087474 Std. Dev. 1.109455 1.124558 1.170264
1.254630 1.190188 1.209932 0.511579 Skewness -0.321837 -0.907212
-0.471834 0.015007 0.172630 -0.455469 0.233846 Kurtosis 8.787535
6.122517 7.245348 10.76070 5.183417 8.237026 21.57338
Jarque-Bera 2547.479 160.3109 240.3589 770.4339 62.30303
352.0258 4185.418 Probability 0.000000 0.000000 0.000000 0.000000
0.000000 0.000000 0.000000
Observations 1803 295 305 307 306 299 291
Table III Descriptive Statistics for Trading Period Return
Statistic Overall Return
Monday Trading Period Return
Tuesday Trading Period Return
Wednesday Trading Period Return
Thursday Trading Period Return
Friday Trading Period Return
Saturday Trading Period Return
Mean 0.050414 -0.020089 0.069356 0.051941 -0.014466 0.125498
0.091359 Trimmed Mean 0.066473 0.015394 0.097867 0.070358 -0.011160
0.136461 0.068886 Median 0.081429 0.011242 0.102084 0.106070
0.081803 0.168955 0.063495 Std. Dev. 1.048100 1.040023 1.141528
1.168860 1.134154 1.154031 0.431462 Skewness -0.300378 -0.584469
-0.640804 0.250309 -0.087367 -0.545324 2.062100 Kurtosis 9.879979
6.553224 7.449919 13.39583 5.326086 9.350240 16.23641
Jarque-Bera 3585.075 171.9827 272.5212 1385.643 69.37542
517.2085 2338.575 Probability 0.000000 0.000000 0.000000 0.000000
0.000000 0.000000 0.000000
Observations 1804 295 305 307 306 299 292
Table IV Descriptive Statistics for Non-Trading Period
Statistic Overall Return
Monday Non Trading Period
Tuesday Non Trading Period
Wednesday Non Trading
Period
Thursday Non Trading
Period
Friday Non Trading Period
Saturday Non Trading
Period
-
11
Return Return Return Return Return Return Mean 0.016795 0.035780
0.006049 0.021128 0.012717 -0.008071 0.034081 Trimmed Mean 0.017546
0.031310 0.007735 0.015931 0.007147 0.003008 0.038739 Median
0.031182 0.053478 0.022449 0.032614 0.021927 0.008975 0.038665
Maximum 2.185756 1.724572 1.640995 2.185756 1.088430 2.109134
1.906272 Minimum -3.645309 -2.964293 -1.523890 -2.157625 -1.036991
-1.893307 -3.645309 Std. Dev. 0.331065 0.386029 0.287064 0.387774
0.252017 0.289693 0.361159 Skewness -1.080972 -1.485405 -0.124122
-0.137791 0.408767 -0.327630 -3.226559 Kurtosis 26.60410 23.33778
12.30846 18.36202 7.165566 19.32522 46.17567
Jarque-Bera 42207.31 5192.628 1101.927 3019.698 229.7588
3325.653 23107.60 Probability 0.000000 0.000000 0.000000 0.000000
0.000000 0.000000 0.000000
Observations 1803 295 305 307 306 299 291 The main observations
from the results depicted in above tables are-
Value of trimmed mean has improved than mean in almost all the
cases except for
Thursday daily return, Saturday daily return, Saturday trading
period return,
Monday non trading period, Wednesday non trading period and
Thursday non
trading period. Thus all these exceptional periods have more
negative extreme
values than other periods have.
As far as skewness is related some periods have positive
skewness and some have
negative skewness. Main observation about skewness is that
skewness for
Saturday daily return has a positive value of 0.233846 whereas
its trading period
return has a positive value of 2.062100 and non trading period
return has a
negative value of 3.226559, it is an exceptional observation for
which no reasons
could be found by the researchers of the present study.
All the distributions are non-normal as the null hypothesis of
normal distribution
is rejected by Jarque-Bera test ( value for all the cases is
0)
Stationarity of Data: The results of OLS regressions will be
spurious if the dependent
variable is non-stationary (Pandey, I.M. 2002: 4). In the
present study, dependent variable
is daily gold return, gold return for trading period and gold
return for non trading period.
-
12
So we checked the stationarity of all the three return series
graphically and the results
have been presented with the help of Graph I, Graph II and Graph
III.
Graph I Daily Return
Graph II Trading Period Return
Graph III Non Trading Period Return
-
13
It is clear from the above graphs that trading period return is
more clustered than non
trading period return and non trading period return has more
spikes than trading period
return. But as far as stationarity is concerned, all the return
series seem to be stationary.
Stationarity is further been checked by the Augmented
Dickey-Fuller (ADF) test.
Table V
Result of the Augmented Dickey-Fuller Test Null Hypothesis:
Return Series have a unit root Exogenous: Constant Lag Length: 0
(Automatic based on SIC, MAXLAG=24)
Augmented Dickey-Fuller test statistic t-Statistic Prob.* Daily
Return -43.80035 0.0001 Trading Period Return -42.70118 0.0000 Non
Trading Period Return -36.71938 0.0000 *MacKinnon (1996) one-sided
-values. Results of graphical presentation are confirmed by the
results of the Augmented Dickey-
Fuller test. Null hypothesis of having unit root by the series
have been rejected in case of
all the three series and it can be said that all the three
return series are stationarity and the
required econometric models/techniques can be applied on the
series.
Day of the Week Effect on Return Series:
-
14
Results of the ordinary least square equation with dummies for 6
days of the week, as
discussed in the methodology, have been applied on daily return,
trading period return
and non trading period return and the results are presented in
Table VI, Table VII and
Table VIII.
Table VI Results of Ordinary Least Square Equation with Dummy
Variables on Daily Return
Dependent Variable: DAILY RETURN Method: Least Squares Sample
(adjusted): 2 1804 Included observations: 1803 after adjustments
Variable Coefficient Std. Error t-Statistic Prob. MONDAY 0.015691
0.064626 0.242796 0.8082TUESDAY 0.075405 0.063557 1.186412
0.2356WEDNESDAY 0.073069 0.063350 1.153424 0.2489THURSDAY -0.001750
0.063453 -0.027573 0.9780FRIDAY 0.117426 0.064192 1.829304
0.0675SATURDAY 0.126942 0.065068 1.950901 0.0512Durbin-Watson stat
2.059987 Obs*R-squared
46.52858 0.0000
Table VII Results of Ordinary Least Square Equation with Dummy
Variables on Trading Period Return
Dependent Variable: TRADING PERIOD RETURN Method: Least Squares
Sample: 1 1804 Included observations: 1804 Variable Coefficient
Std. Error t-Statistic Prob. MONDAY -0.020089 0.061030 -0.329167
0.7421TUESDAY 0.069356 0.060021 1.155527 0.2480WEDNESDAY 0.051941
0.059825 0.868218 0.3854THURSDAY -0.014466 0.059923 -0.241416
0.8093FRIDAY 0.125498 0.060620 2.070230 0.0386SATURDAY 0.091359
0.061342 1.489331 0.1366Durbin-Watson stat 2.010375 Obs*R-squared
33.54269 0.0000
Table VIII
Results of Ordinary Least Square Equation with Dummy Variables
on Non Trading Period Return Dependent Variable: Non Trading Period
Return Method: Least Squares Date: 12/20/12 Time: 20:51 Sample
(adjusted): 3 1804 Included observations: 1802 after adjustments
Variable Coefficient Std. Error t-Statistic Prob.
-
15
MONDAY 0.031003 0.019128 1.620819 0.1052TUESDAY 0.000653
0.018790 0.034756 0.9723WEDNESDAY 0.019022 0.018712 1.016584
0.3095THURSDAY 0.010374 0.018743 0.553493 0.5800FRIDAY -0.009807
0.018960 -0.517268 0.6050SATURDAY 0.034996 0.019217 1.821112
0.0688
Non Trading Period Return (-1) 0.144435 0.023345 6.187105
0.0000Durbin-Watson stat 2.009072 Obs*R-squared 9.706437 0.0018
It is clear from the Tables VI, VII and VII that for daily
return, Friday and Saturday are
significant days, respectively at 6.75% and 5.12% level of
significance; for trading
period return significant day is only Friday; and for
non-trading period return, Saturday is
significant at 6.88% level of significance.
Durbin Watson stat is the test for the presence of serial
correlation in residuals. This stat
should be near to two, so that there is no serial correlation in
residuals. We found this stat
near to two in case of daily return and for the return for
trading period; but it was
1.710703 in case of non trading period return, so we tried to
include volume and open
interest as exogenous variable, but the value of Durbin Watson
stat was not improved.
Then we tried the one period lagged value of non trading period
return and get the Durbin
Watson stat near to two. Thus the problem of serial correlation
was solved by adding one
period lagged value of non trading period return as exogenous
variable. In all the three
cases value of Obs*R-squared, which is the test for
heteroskesdasticity in residuals, is
significant, thus indicating the presence of heteroskesdasticity
and also indicating the
need for applying the model of ARCH family. For this purpose we
applied vastly applied
GARCH (1,1) model.
Table IX Results of GARCH (1,1) Model with Dummy Variables in
Daily Return Equation
Dependent Variable: DAILY RETURN Method: ML - ARCH (Marquardt) -
Normal distribution
-
16
Sample (adjusted): 2 1804 Included observations: 1803 after
adjustments Convergence achieved after 87 iterations Presample
variance: backcast (parameter = 0.7) GARCH = C(7) +
C(8)*RESID(-1)^2 + C(9)*GARCH(-1) Variable Coefficient Std. Error
z-Statistic Prob. MONDAY 0.010191 0.057318 0.177802 0.8589TUESDAY
0.101268 0.044831 2.258879 0.0239WEDNESDAY 0.055899 0.035984
1.553437 0.1203THURSDAY -0.005852 0.045093 -0.129780 0.8967FRIDAY
0.079437 0.046444 1.710368 0.0872SATURDAY 0.077948 0.119096
0.654495 0.5128 Variance Equation C 0.011293 0.001965 5.745980
0.0000RESID(-1)^2 0.053784 0.006091 8.829623 0.0000GARCH(-1)
0.936935 0.006047 154.9423 0.0000
Table X
Results of GARCH (1,1) Model with Dummy Variables in Trading
Period Return Equation Dependent Variable: TRADING PERIOD RETURN
Method: ML - ARCH (Marquardt) - Normal distribution Sample: 1 1804
Included observations: 1804 Convergence achieved after 76
iterations Presample variance: backcast (parameter = 0.7) GARCH =
C(7) + C(8)*RESID(-1)^2 + C(9)*GARCH(-1) Variable Coefficient Std.
Error z-Statistic Prob. MONDAY -0.022111 0.051695 -0.427720
0.6689TUESDAY 0.108090 0.039164 2.759934 0.0058WEDNESDAY 0.039982
0.034118 1.171859 0.2413THURSDAY -0.011069 0.040357 -0.274282
0.7839FRIDAY 0.099143 0.041501 2.388910 0.0169SATURDAY 0.062691
0.104806 0.598164 0.5497 Variance Equation C 0.009615 0.001579
6.089796 0.0000RESID(-1)^2 0.062532 0.006109 10.23690
0.0000GARCH(-1) 0.930495 0.005651 164.6511 0.0000
Table XI Results of GARCH (1,1) Model with Dummy Variables in
Non Trading Period Return Equation
Dependent Variable: Non Trading Period Return Method: ML - ARCH
(Marquardt) - Normal distribution Sample (adjusted): 3 1804
Included observations: 1802 after adjustments Convergence achieved
after 35 iterations Presample variance: backcast (parameter = 0.7)
GARCH = C(8) + C(9)*RESID(-1)^2 + C(10)*GARCH(-1) Variable
Coefficient Std. Error z-Statistic Prob.
-
17
MONDAY 0.013703 0.007657 1.789679 0.0735TUESDAY -0.000837
0.008241 -0.101580 0.9191WEDNESDAY -0.001659 0.007046 -0.235438
0.8139THURSDAY 0.011841 0.009431 1.255457 0.2093FRIDAY 0.012321
0.005683 2.168150 0.0301SATURDAY 0.029269 0.005301 5.521213
0.0000Non Trading Period Return(-1) -0.091117 0.018611 -4.895889
0.0000 Variance Equation C 0.010691 0.000770 13.87952
0.0000RESID(-1)^2 0.942720 0.032904 28.65054 0.0000GARCH(-1)
0.386123 0.015935 24.23079 0.0000
After accounting for heteroskesdasticity, observations related
to Tables IX, X and XI are:
For all the three return series, GARCH as well as ARCH term are
significant indicating that volatility in case of gold return is
time varying and persistent.
Daily return is negative on Thursday but insignificant and for
other days it is positive; significant return is found only for
Tuesday and Friday at 2.39% and
8.72% respectively.
Trading period return is negative for Monday and Thursday, but
not significant for both the days. Significant positive return
occurs on Tuesday and Friday.
Non trading period return is negative on Tuesday and Wednesday,
though not significant. Significant positive return occurs on
Monday, Friday and Saturday at
7.35%, 3.01% and 0% level of significance. One period lagged
return for non
trading period has also significant impact on the return of non
trading period.
Extreme Value Analysis: For analysing the impact of day of the
week on extreme
values, day of the week for 180 extreme positive daily returns
as well as 180 extreme
negative returns were observed and non-parametric chi- square
test was applied. Results
for this exercise are presented in table XII.
Table XII Days for Extreme Values
-
18
Day (Frequency) Positive Negative Monday 28 29 Tuesday 42 37
Wednesday 29 35 Thursday 37 45 Friday 36 30 Saturday 8 4 2 23.933
32.533 0.0000 0.0000
For stock market, negative return is the indication of
declaration of bad news and positive
return is the indication of declaration of good news. Though
same may not hold in
commodity return, but to some extant impact of the day of the
week, on extreme positive
and extreme negative can be observed. In the present study it
has been done with the help
of applying non-parametric chi square test. On observing the
results given in Table XII, it
is clear on the first instance that extreme positive and extreme
negative returns are
observed on Saturdays in minimum cases. Result of chi-square
also confirm the day of
the week effect on extreme returns.
Conclusions and Discussion:
Present study is an attempt to investigate the day of the week
effect on gold return of near
month futures contract, traded on biggest Indian commodity
exchange, Multi Commodity
Exchange of India Limited (MCX). For the purpose of study,
opening and closing prices
have been analysed with the help of dummy variables.
Heteroskesdasticity was found to
be there in residuals of all the series after applying ordinary
least square equation, and the
same was accounted for by applying GARCH (1,1) model. Results of
the study indicate
that to some extent there is an impact of day of the week on
daily return. When the daily
return was decomposed to trading period and non trading period
return, more impact of
day of the week was found during non trading period than during
trading period. Same
observation was made by Dyl, et al. 1986 for S&P 500. Here
it is important to mention
-
19
that gold is traded on MCX from 10.00 a.m. to 11.30 p.m. from
Mondays through
Fridays; and from 10.00 a.m. to 2.00 p.m. on Saturdays; and as
discussed in methodology
part that for any day, impact of previous close to present open
is considered in the study
as non trading period return. During the non trading period of
MCX, markets of US are
open, so one of the probable reasons for the impact during non
trading period may be the
spill over from US commodity market; and other reasons may be
impact of volume and
open interest rather than the impact of a particular day of the
week. The empirical
findings of some future research may through some light on the
impact of these factors.
For designing a strategy of buying and selling, transaction cost
is also to be considered by
the market players, but they can use the results of the present
study for timing their
tradings.
Reference:
Agrawal, A. 1994. Anomalies or illusions? Evidence from Stock
markets in eighteen countries. Journal of
International Money and Finance, 13: 083-106.
Aggarwal, R. and Lucey, B.M. 2007. Psychological barriers in
gold prices? Review of Financial
Economics, 16: 217-230.
Aly, H., Mehdian, S. and Perry, M.J. 2004. An Analysis of
Day-of-the-Week Effects in the Egyptian Stock
Market. International Journal of Business, 9(3): 301-308.
Ball, C.A., Torous, W.N. and Tschoegl, A.E. 1982. Gold and the
Weekend Effect. The Journal of
Futures Markets,2(2): 175-182.
Berument, H. and Kiymaz, H. 2001. The Day of the Week Effect on
Stock Market Volatility. Journal of
Economics and Finance, 25(2): 181-193.
Dyl, E.A. and Maberly, E.D. 1986. The Daily Distribution of
Changes in the Price of Stock Index Futures.
The Journal of Futures Markets, 6(4): 513-521.
French, K.R. 1980. Stock Returns and the Weekend Effect. Journal
of Financial Economics, 8:55-69.
-
20
Gay, G.D. and Kim, T.H. 1987. An Investigation into Seasonality
in the Futures Market. The Journal of
Futures Markets,7(2): 169-181.
Gujarati, D.N. and Sangeetha. 2007. Basic Econometrics. New
Delhi: Tata McGraw-Hill Publishing
Company Limited.
Kamath, R.R., Chakornpipat, R. and Chatrath, A. 1998. Return
Distributions and the Day-of-the-Week
Effects in the Stock Exchange of Thailand. Journal of Economics
and Finance, 22(2-3): 97-106.
Kaur, H. 2004. Time Varying Volatility in the Indian Stock
Market. Vikalpa, 29(4): 25-42.
Lucey, B.M. and Tully, E. 2006. Seasonality, risk and return in
daily COMEX gold and silver data 1982-
2002. Applied Financial Economics, 16: 319-333.
Ma, C.K. 1986. A Further Investigation of the Day-of-the-Week
Effect in the Gold Market. The Journal of
Futures Markets, 6(3): 409-419.
Pandey, I.M. 2002. Retrieved from
http://www.iimahd.ernet.in/assets/snippets/workingpaperpdf/2002-09-
08IMPandey.pdf on 05-12-2012.
Puja, P. 2010. Days-of-the-Week-Effect and Stock Return
Volatility: Theory and Empirical Evidence.
Advances in Management, 3(4):14-21.
Rahman, M.L. 2009. Stock Market Anomaly: Day of the Week Effect
in Dhaka Stock Exchange.
International Journal of Business and Management,
4(5):193-206.
Sarma, S.N. 2004. Stock Market Seasonality in an Emerging
Market. Vikalpa, 29(3): 35-41.
Singhal, A. and Bahure, V. 2009.Weekend Effect of Stock Returns
in the Indian Market. Great Lakes
Herald, 3(1): 12-22.
Suman, and Chahal, S.S. 2011. Day of the Week Effect on Stock
Market Return and Volatility of India.
HSB Research Review, 2(2): 54-59.