A cause and effect relationship between foreign institutional investment flows and stock market returns vietnam case study

Hence, the effect of foreign trading in HOSE should be considered and HOSE with its dominance can be used as representative for the whole stock market of Vietnam for investigating foreig

UNIVERSITY OF ECONOMICS INSTITUTE OF SOCIAL STUDIES

VIETNAM - NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

A CAUSE AND EFFECT RELATIONSHIP BETWEEN FOREIGN INSTITUTIONAL INVESTMENT FLOWS

AND STOCK MARKET RETURNS

VIETNAM CASE STUDY

By

NGUYEN XUAN PHAP

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

HO CHI MINH CITY, August 2012

UNIVERSITY OF ECONOMICS INSTITUTE OF SOCIAL STUDIES

VIETNAM - NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

A CAUSE AND EFFECT RELATIONSHIP BETWEEN FOREIGN INSTITUTIONAL INVESTMENT FLOWS

VIETNAM CASE STUDY

A thesis submitted in partial fulfilment of the requirements for the degree of

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

I owe so much my family and closed friends who gave me tremendous support in life and study Therefore, I take this chance to express my deep gratitude

to them all

Finally, I pride myself for working very hard to finish this thesis This seems fruit of my studying at VNP with many obstacles that I have overcome with determination and enthusiasm

ABSTRACT

Many previous studies has investigated the relationship between Foreign Institution Investment (FII) and stock market return in emerging markets of Asia such as Taiwan, Thailand, Korea, especially India with a lot of studies reseaching this relationship However, some of them investigated the relationship in mean, the other focused on volatility Wishing to have the combination research, with new approach in econometrics like VAR, multivariate GARCH, I carried out the investigation on this relationship that there has been no empirical study reseached in Vietnam market so far This thesis tried to investigate the relationship between FII and the stock market return in Vietnam represented by Ho Chi Minh Stock market (HOSE) in mean and volatility Data are used from year 2004-2011 including the two sub-periods of pre-crisis and during criris and found that FII caused the stock market return in all periods in mean; the stock market return caused FII only during crisis Regards to volatility, there was no long-term effect between the two series FII had short-term effect to the stock market return, especially during crisis

ABBREVIATIONS

BEKK Multivariate model proposed by Baba, Engle, Kraft and Kroner

TABLE OF CONTENTS

Chapter I : INTRODUCTION 1

1.1 Problem statement 6

1.2 Research objectives 7

1.3 Research scope 7

1.4 Research questions 7

1.5 Thesis structure 7

Chapter II : LITERATURE REVIEW 8

2.1 Theory background 8

2.2 Empirical studies 9

2.3 Conceptual frame work 15

Chapter III : DATA SOURCE, RESEARCH METHODOLOGY AND HYPOTHESIS 16

3.1 Data source & variables 16

3.2 Methodology 17

3.2.1 Unit Root Test 18

3.2.2 Granger causality test 19

3.2.3 VAR BGARCH-BEKK (1,1) model 20

3.3 Hypotheses 25

Chapter IV : FINDINGS AND DISCUSSIONS 27

4.1 Data and descriptive statistics 27

4.2 Analysis results 32

4.3 Hypothesis testing and findings 41

Chapter V : CONCLUSIONS, POLICY IMPLICATIONS AND LIMITATIONS 44

5.1 Conclusions 44

5.2 Policy implications 44

5.3 Limitations 47

References 49

Appendix 53

LIST OF TABLES

Table 2.1: Liturature review summary 11 Table 4.1: Descriptive statistics of FII flow and stock market return in three periods

of time 28

Table 4.2: The results of unit root testing for FII and stock market return in periods

considering intercept and trend in equation 33 Table 4.3: VAR Granger causality between FII and stock market return 35 Table 4.4: Bivariate GARCH(1,1)-BEKK model for FII and stock market return in HOSE 37

Table 4.5: Result of hypothesis testing 42

LIST OF FIGURES

Figure 1.1 Deposited shares by years 2 Figure 1.2: Percentage of deposited shares as proportion of registered shares per year 3 Figure 1.3: Listing Scale 2011 4 Figure 1.4: VN-Index and Net trading volume of FII as proportion of the whole market of HOSE 6 Figure 2.1: Causality between FII flows and stock market return 15 Figure 4.1a : The whole period graph with sub-period marks for monthly FII in HOSE 27 Figure 4.1b : The whole period graph with sub-period marks for monthly stock market return in HOSE 28 Figure 4.2a : The whole period graph with sub-period marks of monthly FII in HOSE varied by season 31 Figure 4.2b : The whole period graph with sub-period marks of monthly stock market return in HOSE varied by season 32

LIST OF APPENDICES

Appendix A: Unit root test 53

Appendix B: VAR Residual Normality Test 61

Appendix C: VAR Granger Causality 64

Appendix D : Ljung-Box/Correlogram test 77

Appendix E: Bivariate Garch model estimation from R-Plus 83

Chapter I : INTRODUCTION

Foreign Institutional Investment (FII) flows which are commonly known as international portfolio flows, refer to capital flows made by individual and institutional investors across national borders in creating an internationally diversified portfolio FII flows almost had not appeared until 1980’s But following

a strong trend towards globalization, widespread financial liberalization and implementation of financial market reforms in many countries all over the world had actually led FII flows during 1990’s According to Bekaert and Harvey (2000), FII as a proportion of a developing country's GDP increases remarkably with liberalization as through an integration of domestic financial markets with the global markets, free capital flows from countries which are rich to those are scare in capital in seeking of higher rate of return and increased productivity and efficiency

of capital at global level

Vietnam stock market overview

In order to develop the country, Vietnam has been required to restructure the economy to be more efficient and competitive For this purpose, Vietnam had demanded much more capital of investment Thus, securities market establishment

in Vietnam is necessary to mobilize capital for the economy development The Prime Minister signed Decree No 48/1998/ND-CP on July 10th 1998 about stock and securities market and Decision No.127/1998/QĐ-TTg on July 11th 1998 to establish the two securities trading centers in Hanoi and Ho Chi Minh City However, till July 20th 2000, the Ho Chi Minh City Securities Trading Center was officially opened and then was changed into Hochiminh Stock Exchange (HOSE)

on August 08th 2007 (Hochiminh Stock Exchange, 2011).The Hanoi Securities Trading Center (HASTC) was put into operation on 08th March 2005 and officially changed into Hanoi Stock Exchange (HNX) on 24th June 2009 for listing and

transaction only stocks of companies with charter capital less than VND 30 billions (Hanoi Stock Exchange, 2011)

Figure 1.1 below shows us the increasing trend of deposited shares by years

Figure 1.1 Deposited shares by years Sources: Vietnam Securities Depository (2012) Annual report Retrieved August

30, 2012, from http://vsd.vn/en/p40c44/annual-report.htm

Obviously, deposited shares remarkably increase in the three years from

2008 This also shows the corporation equitization process increasing during crisis

in according with government policy to reconstruct corporations in which the equitization is required for more effectiveness

Besides, we can see the trend of deposited shares as proportion of registered shares per year for the whole Vietnam stock market in Figure 1.2 as follows;

0 5000 10000

Figure 1.2: Percentage of deposited shares as proportion of registered shares per

Up to now, numbers of listing stocks in HOSE and HNX have been 353 and

394 respectively However, considering the size and trading volume, HOSE is much bigger than HNX In year of 2011, numbers of transacted shares in HOSE and HNX are 8,280,645,959 and 7,943,573,441 with the equivalent amount of VND 159,144 billion (Hochiminh Stock Exchange, 2012) and VND 95,847 billion (Hanoi Stock Exchange, 2012) respectively This means that the trading volume of HNX is around 60.2% proportion of HOSE While foreign trading in HOSE and HNX is VND 57,294 billion and VND 4,497 billion respectively equivalent to 36% and 4.7

% of total trading of each corresponding exchange market Moreover, in terms of

% DEPOSITED/REGISTERED SHARES PER YEAR

% DEPOSITED/REGISTERED SHARES PER YEAR

foreign trading activities in HNX are not really remarkable compared to those in HOSE and their effect to HNX is less important than to HOSE Hence, the effect of foreign trading in HOSE should be considered and HOSE with its dominance can

be used as representative for the whole stock market of Vietnam for investigating foreign trading activities and the causality between FII and stock market return

Actually, there are some kinds of securities transacted in HOSE such as stock, bond and fund certificate However, company stock has a big scale in the market Thus, in this study only stock transactions in HOSE can be used for calculation

Figure 1.3 shows the listing scale in HOSE at the end of 2011

Figure 1.3: Listing Scale 2011 Sources: Hochiminh Stock Exchange (2012) Listing summary Retrieved January

15, 2012, from http://www.hsx.vn/hsx_en/Modules/Statistic/QMNY.aspx

91,91

6,23 1,870,00

Stock Bond Fund Certifcate Others

FII in Vietnam

FII takes part in the Vietnam stock market in kinds of individual or institutional investor that usually has been known as foreign fund To date, the fund invests in Vietnam mainly in the form of investment company or investment fund which have been established by foreign investors in third countries offering tax advantage like the British Virgin Island or Bermuda and then open investment accounts in Vietnam as a foreign portfolio investor In the past three years, more than 8,100 institutional investors and individuals abroad have opened trading accounts in Vietnam In 2010, there were 289 organizations and 950 foreign individuals are granted trading codes, up 25.5% and 6.7% compared to 2009 ( Nguyen, 2011) At the end of 2011 accounts of foreign investors were 15,569 including 1,724 organizations and 13,845 foreign individuals (Vietnam Securities Depository, 2012) Number of investment funds operating in Vietnam is quite large, about more than 400 funds, while fund established within the country is too small, only 23 funds with small size of capital as well

FII has flown into Vietnam with big proportion but after the first half of the year 2008 this FII decreased in the context the financial crisis widespread all over the world Figure 1.4 below shows us the situation Moreover, it shows that FII varied a lot with a big difference between the two periods before and after the year

2008 Compared to the first period with high variation, FII in the second period varied more slightly However, in this period FII has made the peak and the base point of net trading volume in June and November of 2008 respectively We might explain this situation as profit making activity of foreign investors when the stock market return turned down over several consecutive months from October of year

2007 till June 2008 Hence, they increased buying in April, May and June then increased selling during months later, reaching a negative net trading volume from November of year 2008

Figure 1.4: VN-Index and Net trading volume of FII as proportion of the whole

of the host country Moreover, securities market of Vietnam is small and immature whereas the foreign funds are now so big, possibly to dominate the capital market Hence, there may be the potential of that FII capital flies out of Vietnam market, drive the prices down sharply thereby induce considerable instability in the Vietnam stock market

However, the issue of whether FII flow affects stock market return or in a contrary direction is still on controversy An analysis of the direction of causality to understand the possible effect of volatility of FII flow on the domestic financial

-20 -10 0 10 20 30 40

0 200

market and economy is important from the viewpoint of policy makers Nevertheless, this issue still remains unresolved

1.2 Research objectives

 The empirical study aims to determine the causal relationship

between FII flow and the national stock market return

 Examine the role of FII flow in the stock market of Vietnam

1.3 Research scope

This study just focuses on Ho Chi Minh Stock Exchange (HOSE) only to determine the relationship between FII flow and stock market return because HOSE can represent to Vietnam stock market due to the large size of HOSE in capital as well as bustling foreign trading activities

1.4 Research questions

 Does FII flow cause the stock return and/ or vice versa?

 Does FII flow volatility have short-term or long-term effect on the

stock market return volatility and/or vice versa?

1.5 Thesis structure

Accordingly, the remainder of the paper is organized as follows The next chapter deals with literature review while the third chapter shows the data source, research methodology and hypothesis Chapter four discusses the findings of the case study Finally, conclusions, policy implications and limitations are presented in chapter five

Chapter II : LITERATURE REVIEW

2.1 Theory background

The theory says that local investors comprehend about local financial market more than foreign investors, leading to a positive feedback trading by foreign portfolio investors A positive feedback trading strategy leads to buy or (sell) decisions following a rise (or fall) in stock prices thereby brings more FII inflows into the market Conversely, the behavior of selling (or buying) while the stock prices are rising (or falling) is considered as a negative feedback trading Accordingly, local investors who are owing to the advantage of information may trade in stocks in response to some new inside information and cause a price variation, and then this variation in price may lead to FII flows due to foreign investors’ positive or negative feedback trading In such cases, local investors are considered as information traders with insights about the firms of which stocks they would want to buy and foreign investors are considered as positive or negative feedback traders who observe and trade stocks following the trend of the market Besides, Bikhchandani and Sharma (2001) gave out a theory regarding positive or negative feedback of mutual funds in financial market: Behavior of investors who reconstruct their portfolio due to a sudden rise or a fall in share prices, should be called a positive or negative feedback behavior of buying at high then selling at low price or buying at low then selling at high price

The contrarian theory of flows affecting contemporaneous and future stock market returns coexists Froot, O'Connell and Seasholes (2001) showed that international portfolio flows led price changes The acceptable explanation for the reverse direction of causality from FII flows to stock market returns is the 'herding behavior' of foreign investors According to Park and Sabourian (2011), herding behavior is in which agents switch behavior from buying to selling or in the contrary direction, following the crowd These herders act against their own information and make their investment decisions based on the other traders’

decisions Information externalities have conditioned rational herding to occur Accordingly, rational herding happens when agents’ own information is less efficient or never revealed to public (Redding, 1996) because it is covered by information of other traders’ observed decisions Moreover, phenomenon of mutual investors who watch the other investors’ behavior for trading is considered herding behavior (Bikhchandani & Sharma, 2001)

2.2 Empirical studies

In 1990’s and 2000’s, several studies have researched the cause and effect relationship between FII flow and domestic stock market return but the results have

been mixed in nature Malarvizhi and Jaya (2009) studied the impact of FII flow to

the Indian stock market (National Stock Exchange NSE was used for representative) and found that there was a unidirectional causal relationship between FII and stock market return (the return of Niffty Index) where FII caused

the stock market return but not the reverse In contrast, Brennan and Cao (1997) found a positive correlation between FII flow and stock market return This positive

relationship is explained due to information disadvantage of foreign investors and then they trade as momentum traders Evidence from causality tests conducted by Mukherjee, Bose and Coondoo (2002) suggested that FII flows to and from the domestic market tended to be caused by returns in the domestic equity market but not in a contrary direction Gordon and Gupta (2003) indicated that lagged domestic stock market return had a negative effect to FII flow

Subsequently, by using daily data, Bose and Coondoo (2004) found light evidence of bidirectional causal relationship between return on the stock index of Bombay Security Exchange (BSE) and FII net inflow Chakraborty (2007) used monthly data obtained from BSE and employed pair-wise Granger causality test to investigate the causal relationship between FII flow and stock market return in India They have found the evidence of bidirectional causality between FII flow and stock market return, whereas FII flow is more caused by stock market return While

data on daily basis from NSE, they have also found the existence of bidirectional causal relationship between FII flow and stock return in India stock market, whereas FII flow is more driven by stock return Lai, Low and Shiu (2008) studied

in TSE1 , tried using pair-wise Granger causality test and found the same result that FII tends to be a momentum trader and conversely, they can predict the stock return

Jo (2002) employed a combination of Two Stage Least Squared and ARCH

models to show that FII flows induced a greater volatility in market compared to

domestic investors in Korean stock market Pavabutr (2004) used modified GARCH(1,1) to identify the impact of FII flow to stock return volatility and found the evidence of that foreign net flow has positive effect on market volatility Rai and Bhanumurthy (2004) used the monthly data for the period started from January

1993 to November 2002 to examine the effect of stock return, risk in the stock market and other real factors on FII flows into India To capture the asymmetric effects between the good and bad news which induce volatility in returns, in turn causing volatility in FII, they employed TARCH model The result shows that the stock return in India is the main driving force to FII and foreign investors react more strongly to bad news than to good news They also recommend that stabilization of stock market volatility would attract more FII However, the reverse causal direction running from FII to stock return has not found Chauhan and Garg (2010) used the daily data from April 1st, 2001 to December 31st, 2009 and employed Granger causality test, VAR model, Wald test, Impulse Response Function to identify the investment behavior of various institutional investors FII’s are found to be engaged in the positive feedback trading activities and causing volatility in Indian stock market

Briefly, we can summarize the relevant empirical studies as table 2.1 below;

1 Taiwan Security Exchange

Table 2.1: Liturature review summary

Brennan and

Cao (1997)

1989:1-1994:4 Quarterly

quarter t expressed as

a proportion of the average absolute level of net purchases over the previous four quarters)

-SMR3 (local Indices)

SMR causes FII

Daily

Stage Least Squares) ARCH model

-FII (Net purchase amount)

-SMR (KOSPI 4)

FII flows

induces a

greater volatility in markets compared to domestic investors Mukherjee,

Bose and

Coondoo

(2002)

1999:1-2002:5 Daily

Granger causality tests

- FII (net investment

as a proportion of the size of market

capitalization)

- SMR (BSE Sensex)

FII flows are caused by returns in the domestic equity market

and not in a contrary direction Gordon and

Gupta (2003)

1992:9-2001:10 Monthly

Impulse Response Function

-FII equity flows as a percentage of market capitalization on the BSE

- SMR (BSE Sensex) and lag series

Lagged domestic SMR has a negative effect to FII flow

Bose and

Coondoo

(2004)

1999:1-2004:1 Daily

- SMR (BSE Sensex) -Dummy variable (policy intervention)

Light evidence

of bidirectional causal relationship between return

on the stock index of BSE and FII net inflow Rai and

Bhanumurth

y (2004)

1993:1-2002:11 Monthly

(threshold ARCH) GARCH(1,1)

-FII(net purchase amount)

- SMR (BSE Sensex) -Inflation

-The SMR in India is the main driving force to FII -Stabilization

of stock market volatility would attract more FII The

reverse causal direction has not been found

Pavabutr

(2004)

1995:1-2002:5 Daily

Thailand Modified

GARCH(1,1)

-FII(net trading volume)

- SMR (TSE 50 Index, FB 25)

FII has positive effect

on market volatility

Chakraborty

(2007)

1997:4-2005:3 Monthly

Granger causality tests

-FII (net FII flows as

a proportion of the preceding month's BSE market capitalization)

- SMR (BSE Sensex)

Bi-causal relationship between FII flow and SMR, whereas FII flow is more caused

by SMR Babu and

Prabheesh

(2008)

2003-2007 Daily

approach, Granger causality test, Impulse Response Function

-FII (net equity purchase)

- SMR (S&P CNX Niffty of NSE)

The existence

of bidirectional causal relationship between FII flow and SMR, whereas FII flow are more driven

by SMR

5 local Mutual Fund

Lai, Low and

Shiu (2008)

1994:6-2002:1 Daily

Granger causality tests

-FII (ratio of net purchase to total shares outstanding)

- SMR (TSE)

SMR causes FII and vice versa

Malarvizhi

and

Jaya(2009)

1999:4-2009:3 Monthly

Granger causality tests

-FII(net purchase amount)

- SMR (S&P CNX Niffty of NSE)

Unidirectional causality from FII to SMR

Chauhan and

Garg (2010)

2001:4-2009:12 Daily

Granger causality test, Impulse Response function,covariance decomposition

-FII( ratio of gross buying to selling ) -MF5( ratio of gross buying to selling)

- SMR (S&P CNX Niffty of NSE)

FII’s are found

to be engaged

in the positive feedback trading activities and causing volatility in Indian stock market

2.3 Conceptual frame work

Figure 2.1: Causality between FII flows and stock market return

Stock market return

Test for causal relationship between FII flows and stock market returns in mean

and volatility

Foreign

Institutional

Investment (FII)

Chapter III : DATA SOURCE, RESEARCH METHODOLOGY AND

HYPOTHESIS

3.1 Data source & variables

A period of eight consecutive financial years ranging from 2004 to 2011 is selected for empirical study Besides data of the whole period, for more robust testing, we divide the range into two sub-periods: The 1st is from Jan 2004 to Dec

2007 and the 2nd one is from Jan 2008 to Dec 2011.It means that we test the causal relationship between FII flow and stock market return before and during financial crisis

Monthly net FII flows (i.e., gross purchases ─ gross sales by foreign investors) into the Vietnam equity market and monthly stock market return which are calculated data from the secondary data obtained from HOSE website such as gross purchases, gross sales by foreign investors, value of closing VN-Index However, the two proxies below are considered suitable choice for estimation to investigate the causality between FII flow and stock market return in Vietnam

 Monthly net FII flows as a proportion of the monthly's trading volume (in HOSE)

(gross purchases ─ gross sales by foreign investors)t / (trading volume)t

 Monthly return on VN-Index (R) : R t = (Pt – Pt-1)/Pt-1

t : Month

Pt : Value of closing VN-Index at the end of month t

As a benchmark of study of Chakraborty (2007) and Lai, Low and Shiu (2008), the ratio of monthly net FII flow to monthly’s trading volume is used as FII flow measure in this study Monthly’s trading volume is used to replace for monthly market capitalization value or value of outstanding shares because it is suitable choice for Vietnam case study when the percentage of deposited shares as proportion of registered shares is only seventy and the percentage of trading volume

is less than this number Therefore this proxy more really reflects the activity of foreign investors in HOSE

Monthly return on VN-Index is calculated as the ratio of the excess of index value in a specific month over the preceding month to preceding month’s VN-index value

3.2 Methodology

Descriptive statistics like maximum, minimum, mean, median, standard

deviation, skewness and kurtosis are shown for each of the two time series in order

to depict the trend over the sample period Then, in order to determine a relationship between FII flows and stock market returns, specially the direction of causality, a

Granger causality tests are conducted between monthly net FII flow as a proportion

of present month's trading volume in HOSE and monthly return on VN Index But,

to improve standard Granger causality tests we should perform Unit root test for

checking stationarity of variables and then test co-integration (if any) between them before conducting Granger causality tests

The Bivariate Generalized Autoregressive Conditional HeteroscedasticBGARCH will be employed in this study to investigate the relationship of stock return and net FII flow in volatility In this model the interactivity between the variables and the impact of cross lagged variables are examined

VAR approach with models of simultaneous equations is employed to identify the causal relationship between FII and stock market return in mean and volatility i.e VAR-Granger causality test and VAR- BGARCH test which are both used in this study showing the difference in methodology between my study and other previous studies in which only one moment was examined either a first moment concerning with series’ mean interaction or a second moment regarding to mutual effect on series’ variance Therefore, this is also my contribution in this research

3.2.1 Unit Root Test

This test is used to determine the stationarity of each time-series sample Augmented Dickey-Fuller (ADF) and Phillips Perron (PP) tests are usually employed The ADF model is given three possible forms as follows:

ΔY𝑡 = 𝛿Y𝑡−1+ ∑ αiΔY𝑡−𝑖

Yt is time -series variable to be tested

ut is white noise error term

Minimum AIC (Akaike Information Criteria) & SIC (Schwartz Bayesian Information Criterion) is used to decide numbers of lags k

H0: Unit Root, =0 (Data is not stationary and needs to be differenced)

Ha: Stationary, <0 (Data is stationary and no need of differencing)

The difference between the three regressions concerns the existence of the deterministic elements a0 and βt

Without intercept and trend, ADF model becomes (1)

With intercept, the ADF model becomes (2)

With intercept and trend, the ADF model becomes (3)

The Dickey-Fuller (DF) t-statistic is used for hypothesis rejection

The Phillips Perron test can be used as suitable alternative PP test seems to ADF with serial correlations corrections made in DF test statistic

Most of macroeconomics variables are not stationary Thus, when the unit root test is failed to reject, we first do co-integration test between the two series before conducting a causality test between them Co-integration implies that there is causal relationship only, not causal direction between the two series If the variables are co-integrated, we will apply the ECM (Error Correction Model) version for testing to prevent spurious regression (Mehrara, 2007)

However, the two series in this study are stationary already (passed to reject Unit root test) so we can use Standard Version of Granger Causality test for testing the potential causality

3.2.2 Granger causality test

Standard Granger causality test

This test is based on VAR approach The hypotheses are as follows;

H0: X doesn’t cause Y

Ha: X causes Y

X Granger-causes Y that means X can be used to predict Y in future

Because the lag length can effect to the results of Granger causality test,

considering AIC on determining number of lags is required

Yt, Xt : Variables to be tested

Yt-i : lagged terms of variable Y

0, 0 : Constants i.e., intercepts

i, j, ij: Slope coefficients of independent variables where coefficient associated with lagged terms of dependent variable demonstrate the inertia of dependent variable and the coefficient corresponding to lagged terms of exogenous variable reflects the sensitivity of endogenous variable to changes in exogenous variable

m,n,p,q: Autoregressive lag length

uyt, uxt,: Uncorrelated white noise errors

i, j: Numbers of lags

t: Period of time

The hypotheses as follows;

H0: j =δj= 0 for all j

Ha: j ≠0 and δj ≠0 for at least one j

We conclude X causes Y if j is statistically significant but δj is not Conversely, Y causes X But if both j and δj are significant, then bi-causality exists

3.2.3 VAR BGARCH-BEKK (1,1) model

In some cases when we don’t know whether a variable is an explanatory or endogenous variable VAR approach with models of simultaneous equations is suitably employed to identify which it is Therefore, to test the causality between the two series in mean, we would employ Granger causality test based on VAR structural model as explained above However, this test only considers the interaction of first movement between the two series Thereby, the generalized autoregressive conditional heteroscedastic (GARCH) process is included with the VAR structural equations to examine the volatility or the second moment of the two series to become a VAR-GARCH model Moreover, to verify the interaction of the second movement between many series we would employ multivariate VAR-MGARCH model For this study, only two series are examined Thus, bivariate form VAR-BGARCH is chosen The difference between VAR-GARCH and VAR-BGARCH models is that the second model includes cross lagged variables in

volatility equations while the first one does not Thus, the VAR-BGARCH model captures the interaction of volatility between the two series and cross coefficients of

variables indicate the magnitude of effect of the series to another and vice versa

Ahn and Lee (2006) have tested many different combinations of the GARCH models such as AR-GARCH, VAR-GARCH, AR-BGARCH and VAR-BGARCH and concluded that the last model VAR-BGARCH capture the interactivity between stock index performance and output growth better Volatility multivariate models are found in many papers Among the models considering time varying covariance, the restricted BEKK and DVEC are preferred These models are direct generalizations of the univariate GARCH model of Bollerslev (Bauwens, Laurent,

& Rombouts, 2006) DVEC (Diagonal Vector Arch model) has proposed by Bollerslev, Engle, and Wooldridge (1988) However, it may not produce a positive-definite covariance matrix Furthermore, the model does not allow for dynamic dependence between volatility series (Tsay, 2005) Therefore, to guarantee these constraints, Engle and Kroner (as cited in Brooks, Burke, & Persand, 2003) proposed the newer model in which they used quadratic formulation for the parameters to ensure the positive-definite covariance matrix has been known as

BEKK Baba, Engle, Kraft and Kroner ) model

Through above discussion, I decide to choose VAR- BGARCH-BEKK (1,1) model because it is the most suitable model for my study: It produces a positive-definite covariance matrix and we can interpret the dependence between series’ volatilities This model can be expressed as below

Consider a stochastic process xt vector of dimension nx1 The set of past market information available until time t-1 is denoted by It-1 and θ denotes a finite vector of parameters According to Bauwens, Laurent and Rombouts (2006), xt can be written

as follows:

xt = µ(θ) + 𝜀𝑡

Where 𝜀𝑡 is the nx1 vector of random error that represents the shock or

𝜀𝑡 |It-1 ~ N( 0,H t )

µ(θ) = E [ xt|It-1 ]

µ(θ) is conditional mean matrix of xt

H t is conditional covariance matrix of xt

Conditional mean is a random series’ expected value that is conditioned or influenced by information of other random series Conditional variance of random series, likewise, is influenced by information of other random series Hence, the mean is often a function of past value of other series and the conditional variance is

a function of past squared residuals of the conditional mean equation (Watsham, & Parramore, 1997)

The general mean equation of VAR- BGARCH (1,1) can be stated as belows;

X𝑡 = 𝛼𝑡+ Γ X𝑡−1+ 𝜀𝑡

Where Γ : nxn matrix for parameters

𝛼𝑡 : nx1 matrix for coefficients

According to Ahn and Lee (2006); Watsham and Parramore (1997), the general mean equation of VAR- BGARCH (1,1) for the two series as follows;

The general conditional covariance matrix which has been known as BEKK model can be written as follows;

𝐴𝑖 and 𝐵𝑗 are unrestricted h x h matrices

𝐶′𝐴′𝑖 and 𝐵′𝑗 are transposed matrices of 𝐶 𝐴𝑖 and 𝐵𝑗

m and n is the lags of ARCH (shock) and GARCH (volatility) respectively

Tsay (2005) said that by using method of symmetric parameterization of the model, Ht is certainly positive definite Moreover, the model also permits to dynamic relationship between the volatilities series However, the model still has some drawbacks For instance, the parameters in 𝐴𝑖 and 𝐵𝑗 cannot directly explain the effect of previous volatilities or shocks In addition, the number of parameters in model is h2(m + n) + h(h + 1)/2,which is recognized quickly increasing with h and the lags m and n Yet, most of them are not statistically significant

For the two series, the BEKK (1,1) model or the volatility equation of BGARCH-BEKK (1,1) can be stated as follows;

i = 1 for FII flow; i = 2 for stock market return

Cij : Coefficients

Aij , Bij : Coefficients indicate ARCH and GARCH effect respectively

𝜎𝑖,𝑡2 = 𝜎𝑖𝑖,𝑡 : Variance of residual at time t

𝜎𝑖𝑗,𝑡 : Covariance of residuals at time t

𝜎𝑖,𝑡 : Conditional standard deviation

εi,t: Ordinary residual

t: Time

t-1: One period lagged time

Thus, the model is the vector autoregressive moving average process (VARMA) that captures both the first and the second movement of the two series

In mean equation the parameters 𝜌12 or 𝜌21 indicate the mean effect of stock return

to FII flow or vice versa Meanwhile the squared of parameters B 12 or B 21 in the volatility equation do not directly interpret the volatility effect of stock market return to FII flow or in a contrary direction (Tsay, 2005) Thus, in order to identity the direct volatility effect of the two volatility series we should develop or carry out matrix multiplication Some significant coefficients can interpret the mutual effect

The Maximum likelihood estimates

Maximum likelihood is employed to simultaneously estimate the parameters

of the mean and the variance equations The log likelihood contribution for multivariate Garch model is given below;

l = -1/2 mlog(2π) -1/2 log(𝐻𝑡 ) -1/2𝜀′𝑡𝐻𝑡−1𝜀𝑡

Where m is the number of mean equations of the multivariate Garch model, and

εt is the vector of residuals of mean equations (Eviews 6 Users Guide) The function value of Maximum likelihood would be estimated It indicates the relationship between volatility series The higher value reveals that the strength of the relationship between volatility series are improved (Ahn, & Lee, 2006)

In case of normal residuals, we can choose Newton method like the Berndt–Hall–Hall–Hausman (BHHH) algorithm for parameter estimates However, if the residuals are not normally distributed, the estimates will be still consistent by using quasi-maximum log likelihood (QML) estimate, given the correctly specified mean and covariance equations (Eviews 6 Users Guide) Therefore, due to the non-normal

of the distribution of residuals6 obtained from mean equations in this study, the Broyden–Fletcher–Goldfarb–Shanno (BFGS) algorithm that is one of the most

6 Refer to Appendix at table B.1, B.2 & B.3

popular method of quasi –Newton approach in solving non-linear estimate problems is chosen Moreover, it has shown evidence of better performance in cases

of non-smooth optimizations

Ljung-Box Q-statistics

At last, to test the model whether it is specified or there is no serial correlation within errors, the Ljung-Box Q-statistics is employed The null hypothesis of this statistics test at lag p is that there is no serial correlation up to order p

Q-statistics displays the correlogram of the standardized residuals or squared standardized residuals Therefore, it can be used to test for serial correlation in the mean equation or squared residuals correlation in the variance equation and moreover to check the specification of these equations For well specified equation, the testing shows all Q-statistics insignificant The choice of lag p might affect to the performance of the Ljung-Box Q-statistics In reality, some values of p are usually used However, with the sample size T, many studies suggest that the rule of thumb is to chose p nearly ln(T) for better performance Moreover, checking autocorrelations at lags of multiples of the seasonality is needed in case of seasonal series (Tsay, 2005)

Test 2:

HO2: Monthly net FII flow does not Granger -cause monthly VN-Index return or lagged FII does not have statistically significant effect on current return (𝛿𝑗 = 0 with all j)

As the effect to volatility is to be tested, there are null hypotheses as follows;

Test 3:

HO3: Volatility of monthly VN-Index return does not have short-term effect on volatility of monthly net FII flow or 𝜀2,𝑡_12 does not have statistically significant effect on 𝜎11,𝑡

Test 4:

HO4: Volatility of monthly net FII flow does not have short-term effect on volatility

of monthly VN-Index return or 𝜀1,𝑡_12 does not have statistically significant effect

on 𝜎22,𝑡

Test 5:

HO5: Volatility of monthly VN-Index return does not have long-term effect on volatility of monthly net FII flow or 𝜎22,𝑡−1does not have statistically significant effect on 𝜎11,𝑡

Test 6:

HO6: Volatility of monthly net FII flow does not have long-term effect on volatility

of monthly VN-Index return or 𝜎11,𝑡−1 does not have statistically significant effect

on 𝜎22,𝑡

Chapter IV : FINDINGS AND DISCUSSIONS

4.1 Data and descriptive statistics

The empirical period in this paper is from January 2004 to December

2011 The sample data is collected as monthly data The total observations are ninety eight This period covers the two periods of before and during financial crisis occurred from 2008 To investigate if any significant changes in each period of time compared to the trend of the whole period, the two sub-periods are considered for robust testing the causality as mentioned above Data of series are plotted for the whole period with sub-period marks as figures 4.1a and 4.1b for their preliminary image

Figure 4.1a : The whole period graph with sub-period marks for monthly FII

in HOSE

-20 -10 0 10 20 30 40

Figure 4.1b : The whole period graph with sub-period marks for monthly stock

market return in HOSE These figures shows no drifts or random walk and series seem stationary To know if series are stationary indeed, the unit root test will be done later Now, we just refer to these figures and table 4.1 for descriptive statistics of FII flow and stock market return in periods of time

Table 4.1: Descriptive statistics of FII flow and stock market return in three periods

Refer to Table 4.1 we can see a very large fluctuation of FII in the whole period when it reached to the maximum of 37.17 and the minimum of -14.47 with standard deviation of 9.6 This fluctuation is larger than itself in each sub-period Figure 4.1a shows that net FII flow was always positive during the period from 2004 to 2007 It means that foreign investors have bought more and sold fewer shares In this period they took very high proportion of the overall market trading volume with the maximum of 37.17 and the minimum is 0.4 In contrast, in period of 2008-2011 the FII fluctuated too much with the big gap between the maximum of 34.43 and minimum of -14.47 The foreign investors did not increase buying shares like the previous period and they tended to sell them more Then, such different fluctuations of FII in each period made FII fluctuate too much in the whole period Testing the distribution of FII in the whole period and the two sub- periods shows that the Jarque-Bera (JB) of FII in the periods are 10.41; 4.52 and 56.48 with probability are 0.01; 0.10 and 0.00 respectively JB of FII in the whole period and the second sub-period is too high (10.41; 56.48 compared to 5.99 at 5% significant level) with very low probability so the null hypothesis of a normal distribution should be rejected Thereby, FII in the whole period and the second sub-second period are not normally distributed and contrarily FII in the first period is normally distributed The kurtosis of FII in second period is 7.12 exceed 3 too much so the distribution is too peaked relative to the normal (leptokurtic)

Now, we examine the stock market return in the periods It fluctuated too much in the whole period from maximum of 38.52 to minimum of -24.01with standard deviation of 11.88 Considering each sub-period, it also fluctuated much with rather high maximum (38.52 and 27.99) and very low minimum (-18.07 and -24.01) respectively in each period While the stock return was mostly positive in the first sub-period with the mean of 4.29 and median of 0.56, it was negative in most of months in second sub-period with negative mean (-1.38) and