Impact of Put Warrant Introductions on the Time-Varying Volatility and
Trading Properties of the Underlying Stocks
Yi-Chen Wang
*
Associate Professor, Department of Financial Operations
National Kaohsiung First University of Science and Technology
Yuan-Hung Hsu Ku
Assistant Professor, Department of Financial Operations
National Kaohsiung First University of Science and Technology
Shih-Kuo Yeh
Full Professor, Department of Finance
National Chung Hsing University
*Corresponding author: all correspondence please contact by following ways:
Tel: 886-7-6011000~3115
Fax: 886-7-6011039
Email: yjwang@ccms.nkfust.edu.tw
0

Impact of Put Warrant Introductions on the Time-Varying Volatility and
Trading Properties of Underlying Stocks
Yi-Chen Wang, Yuan-Hung Hsu Ku, and Shih-Kuo Yeh
ABSTRACT
We adopt Engle’s (2002) Multivariate Dynamic Conditional Correlation Generalize
Autoregressive Conditional Heteroscedasticity (DCC-GARCH) model to understand whether or not stock price, trading volume, and future volatility on Taiwan’s security market are more volatile after the issuance of third-party put warrants. Prior to forming a profitable trading strategy, we must study the impact of the interactions between trading properties of underlying assets after the issue of third-party put warrant listings.
Empirical results show that uncertainty over time-varying conditional future volatility is significantly increased after the introduction of put warrants. The addition of third-party put warrants causes investors to have difficulty in estimating future volatility of underlying security, controlling portfolio risk, and in estimating portfolio’s risk-return profile. Moreover, using the trivariate and bivariate DCC-GARCH frameworks, the trading properties of securities are more closely related to each other after the put warrants have been issued. Therefore, it is much easier for investors to forecast stock returns by observing the changing directions in trading volume and volatility because the relationships between them are much stronger. This also increases the investors’ effectiveness in forming a profitable trading strategy and in forming a diversifiable portfolio.
KEY WORDS: Third-Party Put Warrant, multivariate DCC-GARCH model,
Time-varying volatility, dynamic correlations
1

INTRODUCTION
In complete market, the trading properties of the security market are indicated by stock price, trading volume, and trading volatility and will not be affected by the introduction of warrants since existing financial assets wholly expand the universe of investment opportunity. When a market is incomplete, investment opportunity sets can be expanded more efficiently by adding new assets into the feasible set chosen by the investors. Investors may form a different trading strategy from the situation where new assets are not considered in the portfolio. In other words, the properties of stock price, trading volume, and volatility of underlying security will be altered by the issuance of warrants. When a market is incomplete, warrants can be considered non-redundant assets.
Taiwan authorities allowed security houses to issue third-party warrants to
Taiwan’s market in 1997 in order to provide new hedging instruments to stockholders.
In 1997 two restrictions were imposed on security houses: the first being the construction of a hedging portfolio. Regulation requires security houses to construct a hedge position against warrant issuance; therefore, they must hold a long/short position on underlying assets against their issuance of third-party call/put warrant. The second restriction is that security houses are restricted to short sell underlying stocks for any reason. This restriction prevents security houses from constructing a short position as a hedge portfolio against their own put warrant issuance. Due to this reason, only third-party call warrants were introduced to Taiwan since 1997. In 2003, authorities agreed that security houses may short sell underlying stocks for hedging purpose only with a restriction that they can only borrow underlying stocks from small stockholders.
The Securities Borrowing and Lending (SBL) center, which opened in June 2003,
2

allows security houses to borrow underlying stocks from small stockholders.
Third-party put warrants were officially introduced to Taiwan’s market in July 2003.
The introduction of put warrants was more than six years after call warrants. Thus we are asking the question: Will the introduction of third-party put warrants have a significant impact on the price, trading volume, and volatility of underlying stocks?
We address two issues regarding the introductory effect of put warrants. The first issue is whether or not the trading properties of underlying securities are more or less volatile after the listing day of put warrants. Several researches deal with the volatility issue of warrant issuance. Draper, Mak, and Tang (2001) study the Hong Kong derivative warrant market and find that the volatility of underlying stocks is less affected by the warrant listing. Aitken and Segara (2005) examine the Australian warrant market and find that the price volatility of underlying stocks is significantly higher after warrant listing dates. Based on their findings, we will investigate the changing in time-varying volatility of stock prices, trading volumes, and future volatility for the before- and after-put-warrant period in Taiwan. The second issue we will discuss is how trading strategies are initiated by observing the issuance of put warrants. Before initiating a trading strategy, investors must collect information for trading assets. Knowing the interactions between stock prices, trading volumes, and volatility would help investors realize the price behavior of underlying stock after put warrant issuance, and would also allow investors to make a profitable strategy by longing or shorting underlying stocks.
In order to realize the interactions of stock price, trading volume, and volatility, we briefly divide them into three parts. The first part deals with the relationship between volatility and stock price, the second part with the relationship between stock price and
3

trading volume, and the third introduces the relationship between trading volume and volatility. By knowing the three parts of the interrelationship of trading properties, investors may initiate a profitable trading strategy and generate positive profit.
Several studies including French, Schwert and Stambaugh (1987), Campbell and
Hentschel (1992), Bekaert and Wu (2000), Chiang and Yang (2005)) examine the interaction between volatility and stock price through the time-varying risk premium hypothesis and all support this hypothesis. The hypothesis suggests that if volatility is expected to rise and change over time, then required returns are negatively related to a time-varying expected volatility. In other words, the time-varying risk premium effect suggests a negative relationship between price and volatility.
Recent researches have studied the relationship between stock price and trading volume. Under sequential arrival hypothesis
1
and the mixture of distribution hypothesis 2 , the trading volume is always positively correlated with price change.
McMillan and Speight (2002) report a positive dynamic relationship between volume and returns in the UK futures market. Chiu, Lee, Lin and Chen (2005) also report a positive interaction between stock returns and trading volumes.
Recent studies show evidence that trading volume contains valuable information for predicting future volatility (Lamoureux and Lastrapes (1990), and Blume, Easley and O’Hara (1994)). Karpoff (1987) finds a positive relationship between trading volume and volatility based on differences in opinion hypothesis. Suominen (2001) constructs a theoretical model to show that traders use past periods’ trading volume to
1 Through sequential arrival hypothesis, trading in financial assets is mainly induced by the arrival of new information as investors subsequently revise the market expectations. The relationship between volume and return reveals valuable information.
2 Mixture of distribution hypothesis assumes that the distributions of information flows are not identical for all securities. Different distributions generate predictable across-security differences in the joint distribution of returns and volume.
4

estimate the availability of private information and use it to adjust their trading strategies in equilibrium. Under the Suominen (2001) model, the expected trading volume can be either positively or negatively correlated with expected volatility.
Many studies show that implied volatilities provide more efficiency in forecasting future volatility (Skiadopoulos (2004), Giot (2005), Corrado and Miller (2005), and
Jiang and Tian (2005)). Also, an increase in implied volatility enhances the diversification benefits (Connolly, Stivers and Sun (2005)), so we therefore use implied volatility instead of price volatility in our study.
We adopt a more flexible structure than the Engle’s (2002) dynamic GARCH model by adding the dummy reflecting the put warrant issuing effect. We first use the bivariate dynamic model to realize the concise dynamic correlation between stock return, trading volume, and implied volatility of underlying stocks. These bivariate dynamic correlations can tell us the precise interrelationship between two of the three factors, but it cannot tell us the overall interaction among all three factors. Therefore, we further conduct a trivariate dynamic model to estimate the overall time-varying conditional volatility and correlation among stock price, trading volume, and volatility.
Our study contributes to finance in several ways for the following reasons. Firstly, although Draper, Mak and Tang (2001) and Aitken and Segara (2005) conducted their studies on the warrant listing effect, our study emphasizes the put warrant issuance effect. We compare the trading properties of the underlying stocks in Taiwan for the after-call-before-put warrant period and after-put warrant period. Secondly, we realized that implied volatility contains more information for forecasting future volatility. In so doing, our analysis focuses on implied volatility rather than on return or price volatility. Valuable information regarding Taiwan’s stock markets will be gained by
5

studying implied volatility and subsequently used in creating profitable and diversifiable trading strategies.
Thirdly, we apply the multivariate GARCH models rather than univariate models to describe the time-varying behaviors of trading assets. We further use bivariate and trivariate dynamic models to analyze the dynamic interactions between trading properties.
The empirical results suggest several conclusions. Firstly, the estimated future volatility of stock return and trading volumes decreases insignificantly after the put warrant issuance date. In contrast, the uncertainty of future volatility significantly increases. Adding third-party put warrants into the security market causes investors’ to have difficulty in estimating future volatility of underlying security. This, in turn, leads to difficulty in controlling portfolio risk and in estimating a portfolio’s risk-return profile. Secondly, the time-varying correlations among trading properties are more likely be closed to one
3
by the introduction of put warrants. Thus, investors can easily form a profitable trading strategy by observing the time-varying correlation coefficients since they induce the explanatory power of the trading properties. Thirdly, the bivariate DCC-GARCH estimators reveal that the interactions between trading properties become closer after the put issues.
Consistent with previous studies, the time-varying dynamic conditional correlation between trading volume and implied volatility; and between implied volatility and return is significantly negative. In contrast, the time-varying dynamic conditional correlation between trading volume and stock return is not conclusive. Although listing
3 If the correlation among stock price, trading volume and implied volatility is negative on before-put-warrant period, the coefficient is more negative to one on the after-put-warrant period. If the correlation among stock price, trading volume and implied volatility is positive on before-put-warrant period, the coefficient is more positive to one on the after-put-warrant period
6

of third-party put warrant increases the uncertainty in estimating future volatility of underlying security, it also increases the interactions of security trading properties. The tight interactions within security trading properties increase the probability of engaging a profitable trading strategy. This also increases the probability of generating positive profits.
The remainder of this study is organized with the next two sections describing the empirical methodology and sampling data used in this study. The section following these presents the empirical findings of the DCC-GARCH model and our study ends with concluding remarks.
METHODOLOGY
DCC-GARCH Model
In this section, we first present Engle’s (2002) multivariate dynamic conditional correlation GARCH (DCC-GARCH) model, which estimates conditional correlation coefficients simultaneously with the conditional variance-covariance matrix. By allowing conditional correlations to vary over time, his specification is viewed as a generalization of the Constant Conditional Correlation model (CCC model, Bollerslev
(1990)). To illustrate the dynamic conditional correlation model for our purposes, let x t
be a 3×1 vector containing the return, volume, and implied volatility series in a conditional mean equation as: x t
 μ t
 ε t
, where ε t
Ω t

1
~ N

0, Η t

(1) where
μ t

E
 x t
Ω t

1

is the conditional expectation of x given the past information t
Ω t

1
, and ε t
is a vector of errors in the autoregression AR(1), are assumed to be conditional multivariate normally distributed, with means of zero and
7

variance-covariance matrix Η t

{ h ij
} . Under the assumption that the return, volume and implied volatility series x are determined by the information set available at time t t-1, the model may be estimated using maximum likelihood methods, subject to the requirement that the conditional covariance matrix, Η t
, be positive definite for all values of ε t
in the sample. We also assume that μ t
has the following formation as:
μ i,t
 Φ
0
 Φ
1 x i,t

1
,
 i (2)
Φ
1
measures the ARCH effect in data series. In the traditional multivariate GARCH framework, the conditional variance-covariance matrix can be written as:
Η t

G t
R t
G t
where G t
 diag
  it
(3) h it
is the estimated conditional variance from the individual standard univariate
GARCH(1,1) models in the following manner, h it
  i
  i

2 i , t

1
  i h i , t

1

i (4)
R is the time-varying conditional correlation coefficient matrix. According to the t specification in equation (4), each market’s variance is modeled as a function of the constant, the square of last period’s own residuals
 i
2
, t

1
, and its lagged conditional variance h i , t

1
. After the above basic construction, the dynamic correlation coefficient matrix of the DCC model can be denoted further:
R t

 diag
  t


1
2
Q t
 diag
  t


1
2
Q t

  ij , t
 diag
  t
 
1
2
 diag


1 q
11 , t
,
1 q
22 , t
,
1 q
33 , t


(5)
8

In order to standardize the residual error term, Engle sets z t

G t

1 ε t
where
G is a 3×3 diagonal matrix of conditional standard deviations. t z t
is the standardized residuals vector with mean zero and variance one. Engle also suggests estimating the following time-varying correlation process
 ij , t
 q ij , t q ii , t q jj , t where q ij , t




1 ij

 a a
 z
 i , t

1 z b
  j , t

1 ij

 a
 ij z i , t

1 z q ij , t

1 j , t

1
 b
  ij q ij , t

1

(6) the time-varying correlation coefficients in DCC-GARCH model can be divided into two parts. The first part indicated in the right hand side of equation (6)
 ij represents the unconditional expectation of the cross product z it z jt
, i.e. the unconditional correlation coefficient. The second part indicated on the right hand side of equation (6) a z i , t

1 z j , t

1
 b q ij , t

1
shows the conditional time-varying covariance.
Compare the traditional GARCH (1,1) model in equation (4) with DCC-GARCH model in equation (6), we can show that the DCC-GARCH model standardized the residual error term into standard normal distribution, and the constant term in
DCC-GARCH model represents the unconditional correlation between error term, other than the CCC constant correlation setting (Bollerslev (1990).
Additionally, DCC-GARCH model contributes to the parameters estimation process in two parts. The first is that the conditional correlation defined in the
DCC-GARCH can be modeled individually as a GARCH process, and the second is that the unconditional expectations

of the residual errors can be estimated ij
9

separately by historical data.
DCC-GARCH Model with Put Warrant Issuance Effect
GARCH models are well accepted in related fields because they capture many stylized facts such as volatility clustering and thick-tailed returns. However, since the conditional variance is a function of the magnitudes of the lagged error terms it involves the estimation of a set of parameters. Those parameters are assumed to be constant over the sample period. In this sense, a flexible estimation structure on the conditional volatility and correlation will be incorporated into models in order to capture the changing in trading properties after the issuance of third-party warrants.
Our sampling period starts from the third-party call warrant issuing day, and ends on the third-party put warrant closing day. In order to capture the put warrant issuing effect, we use a dummy variable ( I ) in equation (7) to represent the trading days for the after-call-before-put warrant and after-put warrant periods.
After adding the put warrant issuing effect into the DCC-GARCH model, the estimated conditional variance h it
from GARCH(1,1) is rewritten as: h it
  i
  i
 i
2
, t

1
  i h i , t

1
  i
I t
 t
*

i (7) t
*
represents the put warrant issue day, and I t
 t
*
denotes a dummy variable of put issuing effect. I t
 t
*
is equal to 1 if t
 t
*
, which represents the trading period after-put warrant issuance, and , I t
 t
*
is equal to zero if t
 t
*
, which represents the trading period after-call-before-put warrant issuance.
We use the same concepts to introduce a put warrant issuing effect into the
10

conditional correlation as well as the conditional variance process. Therefore, we also specify the following time-varying correlation with put warrant issuing effects process as
 ij , t
 q ij , t q ii , t q jj , t where q ij , t


1
 
I t
 t
*
 
 ij
 a
 z i , t

1 z j , t

1
  ij
 q ij , t

1
  ij
 
(8) t
*
also represents the put warrants issuance day, and indicator I also denotes a dummy variable indicating the put warrant issuing day. The coefficient

is used to capture the changing property on conditional covariance and conditional correlation.
If the market’s completeness can be improved by the introduction of third-party put warrants, the interdependencies between trading volume, stock price, and volatility will be more connected. We then expect to observe a tighter interrelationship between those trading properties. Therefore, the coefficient

is expected to be positive if third-party put warrants are introduced to the market.
DATA
Sampling Criterions
We compare the difference of volatility on stock returns, trading volume, and implied volatility between the before-put warrant and after-put warrant periods. In order to describe the put warrant issuing effect, we set up four sampling standards which are described as follows: issuing number of put warrants, the continuity of data, trading days before and after put warrants issuance, and the duration for put warrants. Firstly, as of October 2005, there were a total of 74 outstanding put warrants. The number of put
11

warrant issuances in Taiwan has been 42, 8, and 24 for the years 2003, 2004, and 2005, respectively. In October 2005, 74 put warrants were outstanding, but only 43 companies were denoted as the underlying securities for put warrants. This demonstrates that different security houses may issue put warrants on the same stock. For put issues in
2005, 19 out of 24 were introduced after July. Since there were less than 100 sampling days between July and October, the put issuance sample from 2005 has been omitted in order to get a better estimating property from the GARCH model. In 2004, there were only 8 put-warrant issues representing less than 10% of total outstanding issues. Due to this low number of issues, we have not included the results from 2004 in our sampling target. The year 2003 was the first year of put warrant issuance and almost 60% (42 of
70) of put warrants were introduced. In order to capture the introducing effect of put warrants, we have chosen 2003 to be our sampling period.
Our second sampling criterion is the continuity of data. Since our purpose is to test if there is a put warrant issuance effect on volatility of trading properties and the interactions between them, our data will only cover the days on which put warrants were outstanding. For example, as of October 2005, there were a total of four third-party put warrants issued on the Taiwan Semiconductor Corporation (TSMC). The third issue started on 14 January 2004 and ended on 9 July 2004, but the fourth issue did not start until 8 July 2005. This means that the TSMC stock cannot be thought of as a put warrant’s underlying stock from 9 July 2004 to 8 July 2005. Due to the TSMC issue, our time series data will omit the trading days between 9 July 2004 and 8 July 2005, and treat the observations on 9 July 2004 and 8 July 2005 as a continuous time series. The data property could be distorted by this kind of data treatment; therefore we omit the issues where put warrant issuance is not continuous.
12

The exclusion of trading days before-put and after-put warrant issuance that are far from approximately equivalent are represented as the third sampling standard.
Trading days before-put warrant are calculated from the first issuance date of the call warrant on the underlying stock to the first issuance date of the put warrant. It is important to note that trading days before call warrant listing are excluded from the data.
This is done to eliminate the effects of introducing call warrants. DCC-GARCH models use the information from sampling data to provide a better estimation. If there are 200 time series observations in a sample, DCC-GARCH will generate 200 time-varying volatilities for the sample. The 200 th
conditional volatility is generated from whole data from day one to day 200. If the number of trading days before put issue is larger than the days after put warrant issuance, the impact on days before-put warrant issuance may dominate the impact on days after-put warrant issuance. Therefore, we choose the issues where trading days before-put and after-put warrant issuance is approximately equivalent. Using the CMC Magnetics Corporation (CMC) issues as an example, the trading dates before-put and after-put warrant issuance are 1381 and 281, respectively.
The number of trading days before-put warrant is much larger than the after-put warrant period; therefore, we omit this kind of underlying stock.
The fourth sampling criterion is duration of put warrants. For time series studies, the larger the data set, the more accurately the GARCH estimator is. Therefore, the put issues whose duration is less than six months are not included in the sample in order to generate reliable empirical results.
Our study uses daily stock returns, trading volumes, and implied volatility to investigate the issuing effect of put warrants. Data is gathered from the Taiwan
Economic Journal (TEJ).
13

<INSERT TABLE 1 HERE>
Table 1 shows that eight put warrants are selected. These represent four underlying stocks with put issues that satisfy our selection criteria. It is shown that there is more than one put warrant outstanding at the same trading day on the underlying stock. As we can see from Table 1, the put warrant issues around the sampling period for CSC (China Steel Corporation), Acer, SYNNEX, and Foxconn is 4,2,1,1, respectively. In one trading day, there may be several warrants outstanding on the market, but since we only need one data point for one trading day, we establish another criterion for choosing the sampling data. If there are four put issues outstanding on the same day, the implied volatility of each put warrant is different in size owing to their trading frequency. For avoiding the liquidity risk implicit in the warrant market, we choose the sampling data with the largest trading warrant on that sampling day. The choice of the most liquid put avoids the liquidity risk implicit in warrant return and volatility, therefore, the implicit volatility from the most liquid put warrant is selected according to the liquidity concern.
EMPIRICAL FINDINGS
Volatility Effect on Put Warrant Introduction
Table 2 summarizes the return, trading volume, and implied volatility statistics for the underlying stocks of third-party put warrants. As reported in Table 2, almost all the autocorrelations of level residuals and squared residuals (shown in the fourth to first
14

column from the right denoted as Ljung-Box Q statistics) are significant at the 5% level for all series, suggesting autoregressive conditional heteroscedasticity (ARCH) residuals exist in all the series. From this finding, the heteroskedastic pattern exists for these stocks and it is reasonable to apply the GARCH model to Taiwan’s stock market.
<INSERT TABLE 2 HERE>
As we can see from table 2, the Jarque-Bera coefficients are significant at the 1% level. The Jarque-Bera test for normality indicates that the stock return, trading volume, and implied volatility changes are generally not normally distributed. Stock return has less kurtosis or skewness. In contrast, trading volume and implied volatility have excess kurtosis and skewness. Our methodology accounts for both the autocorrelation and non-normality observed in the series.
From Table 3, we observed ARCH residuals in all time series. Therefore, an AR(1) framework is attempting to capture the autocorrelation effect in the mean equation. The coefficient Φ
1
in table 3 reveals the autoregressive effect in mean equation parameters for return, trading volume, and implied volatility. Trading volume is significantly negative correlated to prior trading volume, while implied volatility is significantly positive correlated to prior implied volatility. After considering the autocorrelation effect with our empirical framework, the Ljung-Box Q statistics (Q
2
(8) and Q
2
(24) in table 3 are no longer significant at the 5% level for all series suggesting that ARCH residuals are eliminated by considering AR(1) process.
<INSERT TABLE 3 HERE>
15

The coefficient Φ
0
in table 3 reveals the unconditional mean of stock return, trading volume and implied volatility. The unconditional mean of trading volume is significantly positive at the 1% level for all sampling stocks. In addition, the unconditional mean of implied volatility is significantly positive for CSC, Acer, and
SYNNEX, while the unconditional mean of implied volatility is significantly negative for Foxconn.
By incorporating non-normality properties into time series data, we adopt the multivariate GARCH model to investigate the changes in time-varying conditional volatility in the event of put warrant issuance. Coefficient ω in Table 4 represents the unconditional variance in the multivariate GARCH (1,1) volatility equation. Parameter
η represents the proxy of put issuance effect. As we can see from Table 4, the implied volatility’s η is significantly positive at the 5% level for all samples. For example, the conditional variance on implied volatility after put warrant issuance for the CSC stock is
0.0399, while the unconditional variance on implied volatility for the CSC stock is
0.0001. The issuance of third-party put warrants on the CSC stock causing implied volatility increases by 398
4
times.
<INSERT TABLE 4 HERE>
On the other hand, we find no significant evidence on return volatility and trading volume volatility upon the introduction of put warrants. Connolly, Stivers and Sun
(2005) report that an increase in implied volatility enhances the diversification benefits.
4 From table 4, the unconditional and conditional variance on implied volatility for CSC is 0.0001 and
0.0399, respectively. The percentage change in variance on implied volatility is (0.0399-0.0001)/
0.0001=39800%
16

In contrast, our empirical results show an increase in the uncertainty of implied volatility. An increase in the uncertainty of implied volatility will not lead to diversification benefits, but to difficulty in forecasting future volatility of stocks. The difficulty in forecasting future volatility of stocks causes investors to be unable to form a profitable and diversifiable portfolio. We conclude that the issuance of third-party put warrants reduce the forecasting power in future volatility, while forecasting power in stock return and trading volume is not significantly affected.
Correlation Estimation Results
The corresponding interrelationship between return, trading volume, and implied volatility is investigated by the bivariate GARCH framework. The bivariate framework will be conducted to see whether corresponding correlations between return, trading volume, and implied volatility are significantly increased after put warrant issuance.
Through this kind of study, the formation of a profitable trading strategy regarding a put warrant issuance event can be realized. We briefly conduct three bivariate
DCC-GARCH models to realize the interrelationship between each of the trading properties. The first model investigates the interactions between stock return and trading volume whereas the second deals with the relationship between trading volume and implied volatility. The third and final model talks about the the interdependencies between implied volatility and stock return.
Bivariate Correlation between return and trading volume
We first investigate the interrelationships between stock return and trading volume. In
Table 5, the coefficient b in the first row of each sampling stock shows that the conditional correlation between return and trading volume are significantly positive for
17

SYNNEX and Foxconn at the 1% level. It is consistent with the sequential arrival and mixture of the distribution hypotheses. In contrast, stock return is significantly negative with regard to the trading volume for CSC and Acer at the 1% level.
Therefore, our empirical results suggest no conclusive findings on the relationship between return and trading volume.
<INSERT TABLE 5 HERE>
On the other hand, the coefficient

in Table 5 shows an increase in conditional correlation between stock return and trading volume for the period of the put warrant listing. The result tells us that the interdependence between stock return and trading volume is closely related. In addition, the explanatory power of stock return to trading volume increased after the issuance of put warrant.
In order to realize the timing pattern in conditional correlations between stock return and trading volume, Figure 1 is plotted and shows the time varying relationship between return and trading volume. From Figure 1, the time-varying correlation coefficients are more volatile after the put listing day for all sampling stocks. However, the time-varying conditional correlations for Acer, as described in table 5, become volatile on the six months later after the put warrant issuance day. The

coefficient for Acer is 0.9707, which is much smaller than the

coefficient for the other three stocks. Although the dynamic correlations between stock return and trading volume is significantly higher for the period of put warrant listing, the significantly increased but slightly in size of correlations is observed.
18

<INSERT FIGURE 1 HERE>
Bivariate Correlation between return and implied volatility
Next, we deal with the interdependencies between stock return and implied volatility. A great number of studies show that the leverage effect hypothesis and time-varying risk premium hypothesis are used to explain why volatility changes asymmetrically due to different events. In general, bad news tends to produce more volatility than good news according to the leverage effect hypothesis. This causes a decrease in stock price and an increase in stock volatility. However, the time-varying risk premium hypothesis suggests that if volatility is expected to change over time, equity return will be related to a time-varying expected volatility and a decrease in stock volatility leads to an increase in stock price. French, Schwert and Stambaugh (1987), Campbell and
Henstschel (1992), Bekaert and Wu (2000) and Chan, Cheng and Lung (2005) all report empirical evidence to support the time varying risk premium hypothesis.
The coefficient b in Table 5 shows that stock return is significantly negative related with implied volatility for all sampling stocks at the 1% level. Our result is consistent with previous studies. Coefficient

in Table 5 (row two of each sampling stock) also shows that the changes in the interactions between stock return and implied volatility are significantly positive for all stocks. We can conclude that stock returns are more closely related to implied volatility after the put issuing day. If investors expect an increase in implied volatility for put issues, they will realize that the decrease in stock return is largely caused by the issuance of a put warrant.
Figure 2 also plots the time varying interrelationship between stock return and implied volatility. Panel A describes the dynamic correlation coefficients for CSC and
19

are more volatile right after the issuing date of a put warrant. Panel C and D in Figure
2 indicate the dynamic correlations for Foxconn and SYNNEX, respectively. From
Panel C, the coefficients for Foxconn are volatile and more negative after the put warrant issuing day. Panel D shows that the dynamic correlations for SYNNEX become more negative since the put warrant issuance day. As described in Panel B, the time varying correlation coefficients plotted for Acer cannot be distinguished very well after the issuance of put warrant for the same reason that the

coefficient for Acer is only 0.924.
<INSERT FIGURE 2 HERE>
Bivariate Correlation between trading volume and implied volatility
Then, we study the interrelationships between trading volume and implied volatility.
Several studies mention that trading volume contains information about market makers’ belief regarding stock markets. Karpoff (1987) reports that stock return volatility and contemporaneous trading volume are positively correlated. Schwert
(1989) reports that changes in trading volume are positively related to change in stock return volatility. Lamoureux and Lastrapes (1990) find that trading volume contains relevant information for predicting future volatility. However, Alkeback and Hagelin
(1998) report the reverse result. They argue that a decrease in volatility in the stock increases its trading volume because reduced volatility decreases the risk of the market maker for inventory.
In Table 5, the correlation coefficient b shows that the relationship between volume and implied volatility is significantly negative at the 1% level. We argue that
20

an increase in volatility after put warrant issuance is followed by a decrease in trading volume. This is consistent with Alkeback and Hagelin (1998). The coefficient

in
Table 5 (row three of each sampling stock) reveals that the changes in the interactions between trading volume and implied volatility are significantly positive for all stocks.
This result shows that trading volume is more closely related to implied volatility after the put issuing day. An increase expected in volatility after the put warrant issuance is followed by a large decrease in trading volume.
Plots in Figure 3 show the time varying interrelation for trading volume and implied volatility. Panel A in Figure 3 indicate that the dynamic correlations between trading volume and implied volatility are dramatically volatile for CSC after the put issuance day. However, the correlation coefficients for Foxconn and SYNNEX shown in Panel C and Panel D become more volatile after put issuance. We also observe the lagged volatile effect on the time varying relationship between trading volume and implied volatility for Acer.
<INSERT FIGURE 3 HERE>
By learning that implied volatility is significantly negatively correlated with trading volume and stock return, an increase in implied volatility will induce a decrease in trading volume and stock return. By the bivariate correlation estimation results, we realize that the interdependencies between trading properties are more closely related to each other. Knowing the close interactions between trading properties, the forecasting power by investors in the changing directions and sizes in trading properties is increased after the put warrant issuance day. Therefore, investors may
21

form a different trading strategy and gain the positive profit by knowing the information on correlations.
Trivariate correlations
Using the bivariate DCC-GARCH framework, we precisely realize the interdependencies between stock return, trading volume, and implied volatility. We also expect that the correlations between each trading properties are increased after the put warrant issuance day. A unique indicator representing the interrelationship among the three trading properties cannot be available by the bivariate framework. Therefore, we conduct the trivariate GARCH model to understand whether or not the overall correlation among the three trading properties is increased in the event of a put warrant issuance.
The coefficient b in Table 6 shows that the conditional correlations among stock return, trading volume, and implied volatility are significantly positive for Acer,
SYNNEX, and Foxconn. The conditional correlations among these three are significantly negative for CSC. The introductory effect on put warrant issuance is revealed by the coefficient
δ
. Coefficient
δ
in Table 6 shows that the changes in correlation among three trading properties are significantly increased after put warrant introduction. The positive significance sign of
δ indicates that all-in-one correlation among return, trading volume, and implied volatility is increased by the introduction of put warrants. The interdependence among trading properties is tightly related, which induces the better ability in forecasting the direction and size in correlation, and also enhances the probability of diversification.
22

<INSERT TABLE 6 HERE>
CONCLUDING REMARKS
In order to understand whether or not stock price, trading volume, and future volatility of underlying securities in Taiwan’s security market are more volatile after the issuance of third-party put warrants, we adopt Engle’s (2002) Multivariate Dynamic Conditional
Correlation Generalize Autoregressive Conditional Heteroscedasticity (DCC-GARCH) model to examine the time-varying volatilities of these trading properties. Prior to forming a profitable and diversifiable trading strategy, we need to know the impact of the interactions between trading properties of underlying assets. Also, we must understand the overall interdependence among those trading properties on the issue of third-party put warrant listing. The empirical results show that uncertainty of time-varying conditional future volatility is significantly increased after the introduction of put warrants. This means that adding third-party put warrants into the security market cause investors difficulty in estimating future volatility of underlying security. This also causes difficulty in controlling portfolio risk and in estimating portfolio’s risk-return profile.
Moreover, using the trivariate DCC-GARCH framework, the interactions between each two of the trading properties are more closely related to each other after the put warrants issuing day. The bivariate interrelationship between trading properties shows that it is much easier for investors to forecast changing directions in stock returns by trading volume and volatility for the reason that the relationships between each other are much stronger. By understanding the higher correlation between trading
23

properties, the investors may initiate a profitable and diversifiable portfolio to gain positive profits. By conducting the trivariate DCC-GARCH model, we produce an overall correlation among three trading properties. The overall correlation is increased after the issuing day of a put warrant and this leads to a higher probability in forming a diversifiable portfolio.
24

References
Aitken and Segara, 2005, Impact of warrant introductions on the behaviour of underlying0 stocks: Australian evidence, Accounting and Finance ,
Vol.45, iss. 1, 127
Alkeback and Hagelin, 1998, The impact of warrant introductions on the underlying stocks, with a comparison to stock options, The Journal of Futures Markets , Vol. 18, iss. 3, 307
Bekaert and Wu, 2000, Asymmetric volatility and risk in equity markets, The Review of
Financial Studies , Vol. 13, 1
Blume, Easley and O’Hara, 1994, Market statistics and technical analysis: the role of volume, Journal of Finance , Vol. 49, 153
Campbell and Hentschel, 1992, No news is good news: An asymmetric model of changing volatility in stock returns, Journal of Financial Economics , Vol. 31, 281
Chan, Cheng and Lung, 2005, Asymmetric volatility and trading activity in index futures options, The Financial Review , vol. 40, 381
Chaudhury and Elfakhani, 1997, Listing of put options: Is there any volatility effect?
Review of Financial Economics , Vol.6, iss. 1, 57
Chiang and Yang, 2005, International asset excess returns and multivariate conditional volatilities, Review of Quantitative Finance and Accounting , Vol. 24, iss. 3, 295
Chiu, Lee, Lin and Chen, 2005, Studies on the Effect of Trading Volume and Return
Volatility on Call Warrants and Underlying Stocks in Taiwan, Quarterly Journal of
Business and Economics , Vol.44, iss. 1, 29
Connolly, Stivers and Sun, 2005, Stock market uncertainty and the stock-bond return
25

relation, Journal of Financial and Quantitative Analysis , Vol. 40, iss. 1, 161
Corrado, and Miller Jr., 2005, The forecast quality of CBOE implied volatility indexes,
The Journal of Futures Market , Vol. 25, iss. 4, 339
Draper, Mak and Tang, 2001, The derivative warrant market in Hong Kong:
Relationships with underlying assets, Journal of Derivatives , Vol.8, iss. 4, 72
Engle, 2002, Dynamic conditional correlation: a simple class of multivariate generalized autoregressive conditional heteroskedasticity models, Journal of
Business and Economic Statistics, Vol. 20, 339
French, Schwert and Stambaugh, 1987, Expected stock returns and volatility, Journal of Financial Economics , Vol. 19, 3
Giot, 2005, Implied volatility indexes and daily value at risk models, Journal of
Derivatives , Vol. 12, iss. 4, 54
Jiang and Tian, 2005, The model-free implied volatility and its information content,
The Review of Financial Studies , Vol. 18, iss. 4, 1305
Karpoff, 1987, A theory of trading volume, Journal of Finance , Vol. 41, 1069
Kumar, Sarin and Shastri, 1998, The impact of options trading on the market quality of the underlying security: An empirical analysis, The Journal of Finance ,
Vol.53, iss. 2, 717
Lamoureux and Lastrapes, 1990, Heteroscedasticity in stock return data: Volume versus GARCH effects, Journal of Finance , Vol. 45, 221
McMillan and Speight, 2002, Return-volume dynamics in UK futures, Applied
Financial Economics , Vol.12, Iss. 10, 707
Peter, 1998, Extreme events from the return-volume process: A discretization approach for complexity reduction, Applied financial economics , Vol.8, iss. 3, 267
26

Pierre, 1998, The impact of option introduction on the conditional return distribution of underlying securities, The financial review , Vol. 33, iss. 1, 105
Sarwar, 2005, The informational role of option trading volume in equity index option markets, Review of Quantitative Finance and Accounting, Vol. 24, iss. 2, 159
Schwert, 1989, Why does stock market volatility change over time?, Journal of
Finance , Vol. 44, 1115
Skiadopoulos, 2004, The Greek implied volatility index: construction and properties,
Applied Financial Economics , Vol. 14, iss. 16, 1187
Suominen, 2001, Trading volume and information revelation in stock markets, Journal of financial and quantitative analysi s, Vol.36, iss. 4, 545
Watt, Yadav and Draper, 1992, The Impact of Option Listing on Underlying Stock
Returns: The UK Evidence, Journal of business finance and accounting ,
Vol.19, iss. 4, 485
27

Table 1: Data description for selected underlying stock on put warrant
Underlying Stocks on Put Warrant
Number of put warrant issuance
Starting date of put warrants
Ending date of put warrants
Duration of put warrants (years)
Trading dates before put warrants
Trading dates after put warrants
CSC
4
7/9/2003
6/17/2004
0.94
221
228
Acer
2
SYNNEX Foxconn
1 1
7/30/2003 8/21/2003 11/26/2003
6/1/2004 8/18/2004 7/22/2004
0.84
223
205
0.99
172
250
0.66
252
166
28

Table 2: Basic distribution statistics
CSC
Return
Volume
Implied
Volatility
Ace
Return
Volume
Implied
Volatility
Return
Volume
SYNNEX
Implied
Volatility
Mean
0.0016
0.1491
0.6198
0.0010
0.1792
0.6322
0.0001
0.3276
0.5945
Standard deviation
0.0204
0.6828
0.3699
0.0261
1.0943
0.3130
0.0259
1.4989
0.2086
Skewness
0.5623***
2.1667***
6.0846***
-0.0302
9.6123***
4.2091***
-0.2932**
7.0091***
8.2634***
Kurtosis
1.7381***
6.3956***
44.7565***
2.2727***
138.6544***
22.3324***
2.5020***
78.3466***
103.7156***
Jarque-Bera
80.3582***
1119.0266***
91.1035***
Q(8) Q(24) Q 2 (8) Q 2 (24)
10039.2123*** 1232.5385*** 2119.1518*** 852.1886*** 1019.7934***
115.5707***
20.0259**
37.6302***
43.6094***
51.3949***
31.6196*** 67.0375***
5.3445 12.8680
40335.57107*** 276.8727*** 300.6225*** 142.3258*** 148.1924***
8.0124
345355.5079*** 17.6792**
13.5740*
110857.3187*** 19.4742**
39.0044**
25.6474
23.3301
43.1574***
14.3388
0.0852
25.6198
0.4014
32.7603*** 47.9865***
0.2612 0.7650
193026.2659*** 264.9687*** 454.8883*** 69.6668*** 70.0791***
Return
Volume
Foxconn
Implied
Volatility
0.0003
0.1589
0.5897
0.0214
0.6829
0.3544
0.0212
2.1068***
1.4847***
7.4531***
38.2396***
1270.6040***
17.4321**
45.2800***
-16.2120*** 307.2460*** 1654491.4239*** 30.1813***
36.1684**
57.8222***
36.4508**
8.1022
4.2763
0.3988
15.1829
13.1333
0.7979
***, ** and * represent significance at the 1% , 5 % and 10% levels, respectively.
Q(8), Q(24), Q 2 (8) and Q 2 (24) are the Ljung-Box tests for the 8 th and 24 th order serial correlation of standardized residuals and standardized squared residuals, respectively.
29

Table 3: The conditional mean equation in AR(1) process
The conditional mean equation is expressed as a autoregression (AR1) process:
 t
 Φ
0
 Φ
1 x t

1
Φ
0
Φ
1
Q(8)
Return 0.0015 (1.5692)
(2)
Q(24) Q 2 (8)
-0.1207 (-2.1247)** 11.9651 30.2436 4.6769
CSC
Volume 0.1421 (5.1186)*** -0.2169 (-7.2250)*** 18.4011** 33.7250* 10.4750
Implied Volatility
Return
0.0237 (3.0961)*** 0.9580 (75.8934)*** 7.6751 33.2803* 10.4750
0.0008 (0.6917) -0.0730 (-1.3695) 2.7821 2.90097 2.51290
0.2554 (10.3961)*** -0.6460 (-15.4798)*** 12.9097 34.4358* 2.4001 ACER Volume
SYNNEX
Implied Volatility
Return
Volume
Implied Volatility
0.0450 (4.3686)*** 0.9127 (44.0631)*** 0.7855
0.0003 (0.3389) -0.0002 (-0.0040) 10.8750
2.9939
17.1763
0.7855
9.1094
0.4909 (8.9861)*** -0.3235 (-4.4108)*** 6.0575 27.1755 0.5849
0.0276 (3.3185)*** 0.9490 (66.2193)*** 10.2952 24.7848 0.5849
0.0003 (0.3212) 0.06354 (1.1835) 14.6277* 3.68349 3.7596 Return
Foxconn Volume 0.2040 (6.23926)*** -0.2717 (-4.7122)*** 15.2247* 33.6468* 1.8672
Implied Volatility -0.0190 (-3.7145)*** 1.0189 (111.9217)*** 5.8722
*** , ** and * represent significance at the 1% , 5 % and 10% levels, respectively.
24.3911 1.8672
Q 2 (24)
4.82917
18.5676
18.5676
4.2986
16.7108
2.9939
20.8254
2.3880
2.3880
13.6754
9.6231
9.6231
30

Table 4: Time-varying conditional volatility on DCC-GARCH model with put warrant issuance concern
Adding the put warrant issuing effect into DCC-GARCH model, the estimated conditional variance h it
from GARCH(1,1) is rewritten as: h it
  i
  i
 i
2
, t

1
  i h i , t

1
  i
I t
 t
*

i (7) where t
*
represents the issue day of put warrants, and warrant issuance, and , I t
 t
*
is equal to zero if
I t
 t
*
denotes a dummy variable of put issuing effect. I t
 t
*
is equal to 1 if t
 t
*
, which represents the trading period after-call-before-put warrant issuance. t
 t
*
, which represents the trading period after-put
ω α β η
Return 0.0001 (1.4774) 0.1203 (1.9111)* 0.7726 (6.2640)*** -0.0000 (-1.3762)
CSC
Volume
Implied Volatility
0.0126 (718.0482)*** -0.0356
0.0001 (1.9554)* 0.9982
(-17.0002)***
(4.0460)***
1.0066
0.4104
(583.0667)***
(10.5329)***
-0.0001
0.0399
(-0.1888)
(5.4697)***
ACER
Return
Volume
0.0001 (1.7711)*
0.2156 (6.2191)***
0.0004 (3.9545)***
0.1004
2.0213
(2.2419)**
(8.2795)***
Implied Volatility 0.4882 (3.8527)***
Return
SYNNEX Volume
Implied Volatility
0.0001 (2.3073)**
0.4952 (5.4191)***
0.0000 (1.4161)
0.2505
1.8648
2.7672
(3.3299)***
(7.4673)***
(8.7353)***
Return 0.0001
0.5897
(1.6687)*
(1.8492)*
0.1156 (1.9183)**
Foxconn Volume -0.0100 (-0.6346)
Implied Volatility 0.0005 (3.7377)*** 9.14848 (12.5798)***
*** , ** and * represent significance at the 1% , 5 % and 10% levels, respectively.
0.8101
-0.0023
0.1876
0.5794
-0.0088
0.2906
0.6224
-0.3320
0.0107
(10.2925)***
(-1.4070)
(2.3698)**
(4.5686)***
(-0.5387)
(8.8608)***
(3.4811)***
(-0.4609)
(2.9675)***
-0.0000
-0.0035
0.0435
-0.0000
0.4669
0.0008
-0.0000
-0.0481
0.0007
(-1.3976)
(-0.0664)
(6.8111)***
(-1.1858)
(2.1239)**
(3.7931)***
(-0.6129)
(-0.5806)
(2.2723)**
31

Table 5: Brivariate time-varying conditional correlations on DCC-GARCH model with put warrant issuance concern
The time-varying correlation with put warrant issuing effects are processed as follows: q ij , t
 ij , t



1
 q ij , t , where
 q ii , t
I t
 q t
* jj ,
 t 
 ij
 a
 z i , t

1 z j , t

1
  ij q ij , t

1
  ij
 
(8)
 ij
represents the unconditional expectation of the cross product of residual error terms stock return and implied volatility. While indicator z it z jt
, in other words,
I denotes the dummy of put warrant issuance day, and
 ij
is the unconditional correlation coefficient among trading volume,

is used to capture the changing property on conditional variance and conditional correlation. a b
δ
CSC
Acer
Return-Volume
Return-Implied Volatility
Volume-Implied Volatility
Return-Volume
Return-Implied Volatility
Volume-Implied Volatility
0.0767
0.0378
0.06697
-0.0819
0.0117
0.0137
(12.2908)***
(2642152.2049)***
(180.9825)***
(-55.6217)***
(9576979.3675)***
(238.5793)***
SYNNEX
Foxconn
Return-Volume
Return-Implied Volatility
Volume-Implied Volatility
Return-Volume
Return-Implied Volatility
Volume-Implied Volatility
-0.0190
-0.0055
0.0246
0.1047
-0.0053
0.02293
(-225.0253)***
(-227768.6640)***
(764491.1473)***
(78250632.9017)***
(-21.04698)***
(25.304534)***
*** , ** and * represent significance at the 1% , 5 % and 10% levels, respectively.
-0.0119
-0.2663
-0.9597
-0.4530
-0.7614
-0.5323
0.0778
-0.0026
-0.0080
0.1137
-0.3667
-0.3258
(-44.5100)***
(-79.9938)***
(-116.7019)***
(-135.0020)***
(-118.9910)***
(-626.5010)***
(4.6684)***
(-7136.2310)***
(-4972390.8997)***
(43681475.9694)***
(-75.3078)***
(-1074.6943)***
7.7230
2.7141
3.0434
0.9707
0.9240
0.8918
59.4703
11.5036
21.0936
4.4029
1.6493
1.3283
(41.9131)***
(62010085.2129)***
(22.1156)***
(2334.5700)***
(5509520.8793)***
(81767.4021)***
(4.5388)***
(10298.6122)***
(4525520.3288)***
(52219209.2804)***
(45.8357)***
(681.1252)***
32

Table 6: Trivariate time-varying conditional correlations on DCC-GARCH model with put warrant issuance concern
We specify the following time-varying correlation with put warrant issuing effects process as q ij , t
 ij , t



1
  q ij , t q ii , t
I t
 q t
* jj ,  t


, where ij
 a
 z i , t

1 z j , t

1
  ij
  q ij , t

1
  ij
 
(8)
 ij
represents the unconditional expectation of the cross product of residual error terms stock return and implied volatility. While indicator z it z jt
, in other words,
I denotes the dummy of put warrant issuance day, and
 ij
is the unconditional correlation coefficient among trading volume,

is used to capture the changing property on conditional variance and conditional correlation.
α β η
CSC
Return
Volume
Implied Volatility
Return
ACER Volume
Implied Volatility
Return
SYNNEX Volume
Implied Volatility
0.0120
-0.0092
0.0385
(209.5321)***
(-174.1385)***
(3011.10)***
-0.0131
0.0017
0.8717
(-438.8760)***
(19.0534)***
(165.723)***
2.0010
2.7133
0.2915
(354.1202)***
(1394.4440)***
(767.9200)***
Foxconn
Return
Volume 0.0348 (12.4760)*** 0.8594
Implied Volatility
*** , ** and * represent significance at the 1% , 5 % and 10% levels, respectively.
(8.8729)*** 10.4620 (11.4476)***
33

Figure 1: Brivariate time-varying conditional correlation coefficient between return and trading volume
0.5
0.4
0.3
0.2
0.7
0.6
0.1
0
2002/8/19 2002/10/19 2002/12/19 2003/2/19 2003/4/19 2003/6/19 2003/8/19 2003/10/19 2003/12/19 2004/2/19 2004/4/19
0.5
0.4
0.3
0.2
0.1
0
2002/9/5
1
0.9
0.8
0.7
0.6
2002/11/5 2003/1/5 2003/3/5 2003/5/5 2003/7/5 2003/9/5 2003/11/5 2004/1/5 2004/3/5 2004/5/5
Panel A: CSC
1
0.8
0.6
0.4
0.2
0
2002/11/4
-0.2
2003/1/4 2003/3/4 2003/5/4 2003/7/4 2003/9/4 2003/11/4 2004/1/4 2004/3/4 2004/5/4 2004/7/4
-0.4
-0.6
-0.8
Panel C: Foxnnon
Panel B: Acer
0.05
0
2002/12/11
-0.05
-0.1
0.3
0.25
0.2
0.15
0.1
0.45
0.4
0.35
2003/2/11 2003/4/11 2003/6/11 2003/8/11 2003/10/11 2003/12/11 2004/2/11 2004/4/11 2004/6/11 2004/8/11
Panel D: SYNNEX
34

-0.13
-0.15
-0.17
Panel A: CSC
-0.3
-0.4
-0.5
-0.6
-0.7
-0.8
-0.9
0
-0.1
-0.2
Figure 2: Brivariate time-varying conditional correlation coefficient between return and implied volatility
-0.05
2002/8/19 2002/10/19 2002/12/19 2003/2/19 2003/4/19 2003/6/19 2003/8/19 2003/10/19 2003/12/19 2004/2/19 2004/4/19
-0.07
0
2002/9/5 2002/11/5 2003/1/5 2003/3/5 2003/5/5 2003/7/5 2003/9/5 2003/11/5 2004/1/5 2004/3/5 2004/5/5
-0.1
-0.09
-0.11
-0.2
-0.3
Panel C: Foxnnon
-0.4
-0.5
-0.6
Panel B: Acer
-0.2
-0.25
-0.3
-0.35
0
2002/12/11 2003/2/11 2003/4/11 2003/6/11 2003/8/11 2003/10/11 2003/12/11 2004/2/11 2004/4/11 2004/6/11 2004/8/11
-0.05
-0.1
-0.15
Panel D: SYNNEX
35

Figure 3: Brivariate time-varying conditional correlation coefficient between trading volume and implied volatility
0.4
-0.4
-0.6
-0.8
-1
0.2
0
2002/8/19 2002/10/19 2002/12/19 2003/2/19 2003/4/19 2003/6/19 2003/8/19 2003/10/19 2003/12/19 2004/2/19 2004/4/19
-0.2
1.2
1
0.8
0.6
0.4
0.2
0
2002/9/5 2002/11/5
-0.2
2003/1/5 2003/3/5 2003/5/5 2003/7/5 2003/9/5 2003/11/5 2004/1/5 2004/3/5 2004/5/5
Panel A: CSC (Chinese Steel Corporation)
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
2002/11/4
-0.1
2003/1/4 2003/3/4 2003/5/4 2003/7/4 2003/9/4 2003/11/4 2004/1/4 2004/3/4 2004/5/4 2004/7/4
-0.2
-0.3
Panel B: Acer
0.4
0.3
0.2
0.1
0.6
0.5
0
2002/12/11
-0.1
2003/2/11 2003/4/11 2003/6/11 2003/8/11 2003/10/11 2003/12/11 2004/2/11 2004/4/11 2004/6/11 2004/8/11
-0.2
-0.3
-0.4
Panel C: Foxnnon Panel D: SYNNEX
36