Impact of Foreign Portfolio Flows on Stock Market Volatility -Evidence from Vietnam

Impact of Foreign Portfolio Flows on Stock Market Volatility -Evidence from Vietnam Linh Nguyen, PhD candidate, School of Accountancy, Queensland University of Technology (QUT), Queensland, Australia. Nhung Le, School of Business, International University (IU), Vietnam National University, Hochiminh City, Vietnam. Abstract The paper examines the impacts of daily foreign flows on Vietnamese stock market volatility in two different sub-periods from the end of 2005 to 2011. By combining three widely-used econometric tools which are Vector Autoregressive Models, Granger Causality Tests and Impulse Response Functions, we find several interesting facts. First, almost significant relationships between foreign flows and market volatility are short-lived since there are just two significant links found in the weekly dataset and both of them are at lag 1. Second, past foreign flows relate to volatility stronger in the bull market as compared with the bear one: foreign gross purchases, foreign gross sales and foreign net purchases have significant links with volatility and all (Granger) cause market volatility in the bull market. Third, it is shown that market volatility responds positively (10-day accumulated response) to a random shock to all three flows in the first period (or bull market) even though there are individually negative reactions in some days. Finally, opposite to the bull market, an increase in foreign net purchases reduces market volatility in the bear market (negative reaction), indicating that domestic investors are then more prudent in trading and react to changes in foreign net purchases more slowly. 872

1. Introduction Foreign portfolio flows are an inevitable part under capital market liberalization. As investors trade outside their countries, many concerns about their effects on local stock markets have been raised. One of such concerns is whether or not foreign flows cause higher volatility, especially in emerging equity markets which are usually "small and illiquid" (Pavabutr & Yan, 2007, p.346) in comparison with wealthy capital flows from foreign investors. As a result, this paper aims to investigate if foreign portfolio flows affect Vietnamese stock market volatility and how they impact the market volatility. In particular, our paper aims to address the following research question: Do foreign portfolio flows affect Vietnamese stock market volatility and how? The relationship between foreign portfolio flows and local stock market volatility has received increasing attention, especially in emerging countries. However, a review of prior literature shows different findings in different markets. Huang and Yang (2000) study ten developing countries to examine if market liberalization leads to more volatile local stock markets. Their results show the stock price becomes more volatile in South Korea, Mexico, and Turkey but less volatile in Argentina, Chile, Malaysia, and the Philippines. There is no significant pattern for the other markets including Taiwan, Thailand, and Brazil. Law and Ngah (2008) also find supporting evidence for the fall in Malaysian stock market volatility after liberalization. The authors examine the effect of equity market liberalization on volatility in Malaysian stock market from 1985 to 2006 using the EGARCH model. Law & Ngah (2008) divide their full sample into four sub-ones corresponding to different time periods: pre liberalization, post liberalization but before the 1997-98 financial crisis, postliberalization during the crisis, and post-liberalization after capital controls periods to analyze and compare different effects in different periods. Pavabutr and Yan (2007) examine the effects of both predictable and unpredictable foreign flows in daily and weekly stock return volatility in Thai market from 1995 to 2002. The unpredictable flows are found to have a significant impact on stock return volatility (both daily and weekly). The influence of the predictable flows is however negligible. On the other hand, Nguyen and Bellalah (2008) conduct research on seven emerging markets ( Argentina, Brazil, Chile, Colombia, Mexico, Malaysia and Thailand) from January 1985 to January 2003 and report an insignificant impact of market liberalization on return volatility (on average). It should be noted that stock return volatility is however lowered when the participation of the US investors becomes effective and important on emerging markets, and when emerging markets increase in size. (Nguyen & Bellalah, 2008, p. 396). Even though prior studies have focused on emerging markets 873

including those in Southeast Asia such as Thailand, Malaysia, there has been little research regarding the impacts of foreign flows on Vietnamese stock market volatility. This paper extends the relevant literature by yielding a number of new findings on the relationship between past foreign flows and stock market volatility in Vietnam. Its results of how foreign flows impact local stock market volatility are likely to affect how investors make their investment decisions and how the Vietnamese government can adjust their regulation to attract or limit foreigners trading. The paper finds a strong link between foreign portfolio flows and Vietnamese stock market volatility. The relationship is, however, different on different lags and very short-lived. When we analyze weekly dataset to examine a more longterm relation, most correlation coefficients become insignificant. Both dynamic relation and (Granger) causality between past foreign flows and market volatility are found. In the bull market (26/10/05 to 15/10/07), all three flows: foreign gross purchases, foreign gross sales and foreign net purchases (Granger) cause market volatility. Whereas, in the bear market (16/10/07 to 16/03/11), only foreign net purchases are demonstrated to (Granger) cause stock market volatility. What is more, the paper also investigates how market volatility responds to one-standard deviation shock to foreign flows and finds an interesting result that market volatility reacts positively to shock to all three flows in the bull market but responds negatively to shock to foreign net purchases in the bear market (10-day accumulated response). The rest of the paper is organized as follows. Section 2 describes data and methodology employed. Section 3 presents results and discussion about the impacts of foreign portfolio flows on stock market volatility. Results summary and conclusion are offered in Section 4. 2. Data Description and Methodology 2.1 Data Data are hand-collected directly from the Hochiminh City Stock Exchange website (HOSE), from October 26 th, 2005, right after the Decision No. 238/2005/QĐ-TTG of The Prime Minister Phan Van Khai that specifies 49 % of the total listed shares of a firm on Vietnamese stock market as the maximum room for foreign investors, except banking industry (30 %), started to be effective (effective on Sunday, October 23 rd, 2005). Data include the market index (VN-index), foreign gross purchases and gross sales of stocks, investment certificates both through dealing and put-through on a daily basis. 1 Those data are then processed properly to be employed in the proposed models. Daily aggregate gross purchases on day t (in VND billions) = foreign gross purchases (stocks and investment certificates) by dealings on day t + foreign gross purchases (stocks and investment certificates) by put-through on day t. 874

Daily aggregate gross sales on day t (in VND billions) = foreign gross sales (stocks and investment certificates) by dealings on day t + foreign gross sales (stocks and investment certificates) by put-through on day t. Daily aggregate net purchases(in VND billions) = Daily aggregate gross purchases - Daily aggregate gross sales. Daily volatility = squared daily market return (Pavabutr, 2004) 1 The paper mainly employs DAILY dataset. However, WEEKLY dataset is also investigated just to see if there is any long-term relationship between variables. Therefore, terms such as foreign gross purchases, gross sales, market volatility are all on a daily basis. It will be stated clearly in the case of weekly variables. Daily market return is calculated as follows: Where: R mt: Daily market returns on day t VNindex t : VN-index on day t VNindex t-1 : VN-index on day t-1 The full dataset covers from October 26, 2005 to March 16, 2011. However, the study breaks them into two sub-samples: the first is from October 26, 2005 to October 15, 2007 and the second is from October 16, 2007 to March 16, 2011 since the market index (VN-index) generally experienced a dramatic uptrend (the bull market) during the first period and a downtrend (the bear market) in the remaining time. 2.2 Methodology Three main tests are employed in the study: dynamic relation with unrestricted Bi-variate Vector Autoregressive Models (VAR), Granger Causality Tests, and shock response with generalized Impulse Response Functions (IRF). The testing procedures in this paper follow suggestions from Gujarati (2004). VARs are a widely used model to examine the dynamics of flows and volatility. For example, Pavabutr and Yan (2007) employ VARs as their main models to study the impacts of foreign flows on Thai market volatility. Similarly, Cao, Chang & Wang (2008) examine the dynamic link between mutual fund flows and market volatility by using VARs. Granger Causality Test is employed to test if there is a (Granger) causation running from foreign flows to market volatility while IRF is applied to help interpret the VARs results by examine how the dependent variable (stock price volatility) responds to shocks in the error terms (Gujarati, 2004). 875

The first step is to test variables to see if they satisfy the requirements of Vector Autoregressive Model which is the paper s main model. VAR requires all its variables to be stationary. Therefore, stationary testing must be conducted first. Two popular unit root tests: Augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) are employed to examine whether time series are stationary or not. If all variables are stationary at levels, i.e., I (0), then VAR models can be used. If not, i.e. integrated of order d>0: I (d), additional Johansen Cointegration Tests are needed. Co-integration tests 2 are applied to determine if there exists a long-term relationship between variables (Gurajati, 2004). If two variables (one is stationary and another is non-stationary at levels) are co-integrated then VAR are be still applicable. However, if two time series are both non-stationary at levels and co-integrated then Vector Error Correction Models (VECM) are employed instead. In case of no co-integration found, such pairs of variables are not be included in the analysis. 2 According to Hansen & Juselius (1995), Johansen Co-integration Tests can be conducted when some time series (variables) are stationary I(0) and other can be I(1). Next, Bi-variate Vector Autoregressive Models (unrestricted versions) are employed to examine the dynamic relation between each of three flows measures (foreign gross purchases, foreign gross sales and foreign net purchases) and market volatility on a daily basis. The focus of this study is the link between past foreign flows and current market volatility. While running the models, a caution should be taken in choosing lag-length because the results are very sensitive. Akaike Information Criteria (AIC) is applied to choose an appropriate laglength (Gujarati, 2004). Specifically, the lag-length which has the smallest value of AIC is applied. In VAR models, correlation coefficients are estimated easily (by Ordinary Least Squared method) and such parameters help to explain the relation or correlation between lagged values of foreign flows and local market volatility. F-statistic (to test the null hypothesis that all coefficients in one equation are jointly equal to zero) is also examined to see whether the model is fit or not. Nevertheless, it should be noted that even in the case F- statistic is greater than F-critical value (null hypothesis is rejected), it does not necessarily mean that there is a relationship between past foreign flows and market volatility. There are two types of coefficients: those of past flows and those of past volatility in one equation (see estimated VAR models below). Therefore, it can be the case that no coefficient of lagged values of foreign flows is significant, indicating no relation, but at least one coefficient of past volatility is statistically significant, which confirms past volatility relates to itself. That is the reason why Granger Causality Tests are employed to clarify the link. The estimated VAR model for the analysis of foreign flows and market volatility is as follows: 876

F t = + + + v 1t (2.1) = + + + v 2t (2.2) Where: F t: Aggregate foreign flows (total gross purchases, gross sales or net purchases) on day t* σ t : measure of daily market volatility on day t : Constant : Correlation coefficients k: maximum number of lags v 1t, v 2t: Zero-mean white noise disturbances * Three flows: Aggregate foreign gross purchases, aggregate gross sales and aggregate net purchases are examined individually. Third, Granger causality tests are used to figure out the causal relationship (in Granger sense) between variables. Generally speaking, the main purpose of this test is to examine if past foreign flows (Granger) cause market volatility by testing the joint hypothesis of equalto-zero coefficients of lagged foreign flows. VARs coefficients are argued to be interpreted more deeply by using Granger causality test. X is said to (Granger) cause Y if X helps in the prediction of Y, or put another way, if the coefficients on the lagged X are statistically significant. It should be noted that the statement X (Granger) causes Y does not imply Y is the result of X. Four possibilities are raised by Gurajati (2004): unidirectional causality from Y to X, unidirectional causality from X to Y, bilateral causality and independence. Fourth, Generalized Impulse Response Functions are employed to investigate how market volatility reacts to shock to foreign flows. This study employs generalized impulse functions to prevent the results from being sensitive to the variables' order. In addition, generalized impulse response functions are not performed for all pairs of variables but just for those which are demonstrated to have causal relationship (from Granger Causality Tests). Finally, so as to dip further into the relationship between foreign flows and market volatility, the same testing process is performed for weekly data to see if there is any longterm link (Only important results of weekly data tests are to be presented and compared with daily ones). Weekly data are processed from daily dataset. Weekly variables such as weekly foreign gross purchases, foreign gross sales are computed as the same manner as daily ones but they are sum of five consecutive trading days. In addition, weekly volatility is estimated as the squared root of = where N t is the number of trading days in week t and r it is the return of day i, which falls in week t, less the average return of week t (Pavabutr & Yan, 2007). The procedure is applied for the full sample and two sub-periods on a daily basis. 877

3. Impacts of Foreign Portfolio Flows on Vietnamese Stock Market Volatility 3.1 Dynamic Relation between Foreign Flows and Market Volatility 3.1.1 Foreign Gross Purchases and Market Volatility As can be seen from the Table 1, foreign daily gross purchases have a significant relationship with market volatility only in the first period (26/10/05 15/10/07) or the bull market. However, the direction of their correlation is mixed among lagged values. There is a negative link between yesterday foreign gross purchases( lag 1) and today market volatility, the relation, however, turns to be positive for the lag 2 but with a lower level of significance. Lag 5 shows a highly significant positive correlation, which is followed by negative relation between lag 6 and lag 7 with market volatility. Nevertheless, no significant relation is found in the second period (bear market) and full sample. Table 1: Bi-variate VAR of Daily Volatility with Daily Foreign Gross Purchases (Partly) FULL SAMPLE 1 st PERIOD 2 nd PERIOD 26/10/05 15/10/07 16/10/07-16/03/11 DAILY_VOLATILITY DAILY_VOLATILITY DAILY_VOLATILITY DAILY_GP(-1) -0.00028-0.00739** 0.00054 t-statistic [-0.18608] [-2.21099] [ 0.32269] DAILY_GP(-2) -0.00035 0.00774* -0.00249 t-statistic [-0.22388] [ 1.96773] [-1.45701] DAILY_GP(-3) 0.00093 0.00378-0.00065 t-statistic [ 0.58661] [ 0.95764] [-0.37355] DAILY_GP(-4) -0.00040-0.00640-0.00079 t-statistic [-0.25277] [-1.64350] [-0.45895] DAILY_GP(-5) 0.00214 0.01573*** -0.00102 t-statistic [ 1.35395] [ 4.06367] [-0.58870] DAILY_GP(-6) -0.00174-0.00960** -0.00113 t-statistic [-1.11428] [-2.40605] [-0.66044] DAILY_GP(-7) -0.00196-0.00797** -0.00044 t-statistic [-1.31261] [-1.98296] [-0.26467] DAILY_GP(-8) 0.00503 t-statistic [ 1.48884] F-statistic 19.07011 7.49162 14.82371 *, **, *** significance at 10%, 5% and 1% 878

3.1.2 Foreign Gross Sales and Market Volatility On the contrary, the relation between foreign gross sales and market volatility spreads over three samples as shown in Table 2. In the first period, the correlation between foreign gross sales and market volatility is very short-term, just at lag 1 (positive) and lag 2 (negative), which is opposite to that of gross purchases. However, the relations are very statistically significant. On the other hand, there is a negative link between the current market volatility and yesterday foreign gross sales in the second period (just at 10% level of significance). After that, no significant relation is found until the first day of the previous week (lag 5) whose coefficient is positive. The interesting point here is the opposite direction in the relation of yesterday foreign gross sales (lag 1) and market volatility in two nearly opposite sub-periods. The VAR results of full sample seem to be the combination of its sub-periods. Table 2: Bi-variate VAR of Daily Volatility with Daily Foreign Gross Sales (Partly) FULL SAMPLE 1 st PERIOD 2 nd PERIOD 26/10/05-15/10/07 16/10/07-16/03/11 DAILY_VOLATILITY DAILY_VOLATILITY DAILY_VOLATILITY DAILY_GS(-1) 9.2886E-06 0.01628*** -0.00424* t-statistic [ 0.00435] [ 3.64204] [-1.79951] DAILY_GS(-2) -0.00490** -0.01865*** -0.00080 t-statistic [-2.22258] [-3.99284] [-0.32729] DAILY_GS(-3) -0.00108-0.00030-0.00077 t-statistic [-0.48805] [-0.06433] [-0.31234] DAILY_GS(-4) -0.00298-0.00392-0.00303 t-statistic [-1.34377] [-0.83306] [-1.23074] DAILY_GS(-5) 0.00557** 0.00674 0.00495** t-statistic [ 2.51990] [ 1.44100] [ 2.01875] DAILY_GS(-6) 0.00109 0.00601 0.00173 t-statistic [ 0.49206] [ 1.28197] [ 0.73103] DAILY_GS(-7) 0.00178-0.00551 t-statistic [ 0.83334] [-1.21559] F statistic 19.78735 8.23626 17.53370 *, **, *** significance at 10%, 5% and 1% 879

3.1.3 Foreign Net Purchases and Market Volatility Table 3 demonstrates that past net purchases of foreign investors do have a strong link with market volatility, especially in the first period when Vietnamese market (HoSE) rocketed. Looking more closely at the first period where many strong relationships are found between foreign net purchases and market volatility. The most significant correlation (1% level of significance or 99% confidence level) is at lag 1, 2 and 6. There are also significant coefficients (however, just at 10% level of significance) at lag 5 and 8. It is discovered that yesterday net purchases of foreign investors negatively relates to current market volatility. Nevertheless, that relationship is reversal immediately: positive at lag 2. Surprisingly, estimated coefficients from VARs of foreign gross purchases and net purchases are rather similar (in terms of their direction). Both have a negative link with market volatility at lag 1 and 6, and a positive link at lag 2 and 5. That can be explained by the correlation coefficient between gross purchases and net purchases which is about 0.8 in the first period. Consequently, they have a similar relationship with market volatility. On the other hand, there is only one significant relationship (lag 5) in the second period, which demonstrates that the link between foreign net purchases and market volatility reduces remarkably over time. Table 3: Bi-variate VAR of Daily Volatility with Daily Foreign Net Purchases (Partly) FULL SAMPLE 1 st PERIOD 2 nd PERIOD 26/10/05-15/10/07 16/10/07-16/03/11 DAILY_VOLATILITY DAILY_VOLATILITY DAILY_VOLATILITY DAILY_NP(-1) -0.00080-0.01289*** 0.00289 t-statistic [-0.51422] [-4.28806] [ 1.58107] DAILY_NP(-2) 0.00169 0.01478*** -0.00254 t-statistic [ 1.03431] [ 4.24959] [-1.35697] DAILY_NP(-3) 0.00142 0.00161-1.76E-05 t-statistic [ 0.86809] [ 0.45230] [-0.00929] DAILY_NP(-4) 0.00083 0.00036 0.00086 t-statistic [ 0.50290] [ 0.10067] [ 0.45674] DAILY_NP(-5) -0.00078 0.00663* -0.00380** t-statistic [-0.47845] [ 1.83426] [-2.01749] DAILY_NP(-6) -0.00270* -0.00943*** -0.00228 t-statistic [-1.65239] [-2.64994] [-1.21580] 880

DAILY_NP(-7) -0.00284* -0.00417-0.00250 t-statistic [-1.82103] [-1.17247] [-1.36036] DAILY_NP(-8) 0.00568* t-statistic [ 1.86878] F-statistic 19.57218 8.54930 15.56736 *, **, *** significance at 10%, 5% and 1% With regard to the full period, the relation between past foreign net purchases and market volatility is delayed to lag 6 and 7 and both of them are negative (at 10% level of significance). Generally speaking, three measures of foreign flows show the strongest relation with market volatility in the first period (the bull market) even though the direction of such correlation changes from one lag to another. The significant coefficients last longer, beyond one week as the case of daily foreign gross purchases and net purchases in the first period. As a result, tests on weekly dataset are necessary to examine if the relationship is long-term one. However, results of weekly data mostly show that the correlation is just short-term. Actually, there are only two significant coefficients between weekly net purchases and gross sales (both at lag 1) and weekly volatility. 3.2 Granger Causality between Foreign Flows and Market Volatility After examining the dynamic relation between past foreign flows and market volatility with VAR coefficients, Granger Causality Tests are employed to see whether foreign flows (Granger) cause market volatility. Table 4 shows that daily foreign gross purchases (Granger) cause market volatility in the first period when their correlation is the strongest. The probability is small enough to reject the null hypothesis that gross sales do not (Granger) cause market volatility (at 1% level of significance). However, there is no (Granger) cause running from past foreign gross purchases to market volatility in the second and full period. Table 4: Granger Causality Tests of Foreign Gross Purchases and Market Volatility Pairwise Granger Causality Tests Null Hypothesis: F-Statistic Probability DAILY_GP does not Granger Cause DAILY_VOLATILITY (1 st period) 3.60141 0.00045 DAILY_GP does not Granger Cause DAILY_VOLATILITY(2 nd period) 1.13030 0.34165 DAILY_GP does not Granger Cause DAILY_VOLATILITY (Full) 0.80833 0.58033 881

Foreign gross sales and net purchases show a stronger (Granger) causal relationship in 2 out of 3 samples. As shown in Table 5, changes in past values of foreign gross sales are a source of changes in current market volatility in full period and in the first sub-period. The latter causation is, however, stronger (at 1% level of significance or 99% level of confidence) Table 5: Granger Causality Tests of Foreign Gross Sales and Market Volatility Pairwise Granger Causality Tests Null Hypothesis: F-Statistic Probability DAILY_GS does not Granger Cause DAILY_VOLATILITY( 1st period) 3.96727 0.00032 DAILY_GS does not Granger Cause DAILY_VOLATILITY ( 2 nd period) 1.64121 0.13276 DAILY_GS does not Granger Cause DAILY_VOLATILITY(Full Sample) 2.00599 0.05129 Net purchases also (Granger) cause market volatility in two sub-periods at 1 % and 5% significance level (Table 6). Table 6: Granger Causality Tests of Foreign Net Purchases and Market Volatility Pairwise Granger Causality Tests Null Hypothesis: F-Statistic Probability DAILY_NP does not Granger Cause DAILY_VOLATILITY( 1 st Period) 5.38856 1.68E-06 DAILY_NP does not Granger Cause DAILY_VOLATILITY( 2 nd Period) 2.32979 0.02337 DAILY_NP does not Granger Cause DAILY_VOLATILITY (Full) 1.64669 0.11831 One striking point in the above Granger Causality Tests is the unidirectional causality from foreign gross sales (full sample) and net purchases (second period) to market volatility. In other words, market volatility does not cause foreign gross sales and net purchases (in Granger sense). As a result, market volatility could be explained better by changes in previous foreign gross sales and net purchases. 3.3 Impulse Response Functions: How Does Market Volatility Respond to Shock to Foreign Flows? Figure 1 illustrates response and accumulated response of market volatility to one-time shock to foreign gross purchases in the first period (26/10/05-15/10/07) when causal relationship has been found by Granger Causality Tests. From a general point of view, the daily effect of foreign gross purchases on market volatility is very fluctuated and short-lived. The response reaches its peak on day 6 with increase by 0.85 units for one-standard deviation rise in daily foreign gross purchases. The accumulated response of market volatility is not 882

very significant in the early period. It is almost zero until day 6 and the 10-day accumulated response is approximately 1.187 units. However, in comparison with gross sales and net purchases in the first period, gross purchases has the strongest accumulated impact (positive) on market volatility. Figure 1: Response and Accumulated Response of Daily Market Volatility to Shocks in Daily Foreign Gross Purchases (1 st Period) Impacts of foreign gross sales on market volatility in the first sub-sample as well as in full sample are presented in Figure 2 and 3. As can be seen from the graphs, foreign gross sales affect market volatility stronger in the first sub-period than in the full sample, especially in early days. In the first period, an increase of one standard deviation in daily foreign gross sales leads to a rise of 0.22 units in the current market volatility (period 1). The number goes up to 0.95 units on day 2 but then falls down and fluctuates on the following days. After ten days, the accumulated response is just 0.87 units. Figure 2: Response and Accumulated Response of Daily Market Volatility to Shocks in Daily Foreign Gross Sales (1st Period) There is a difference in individual response of daily market volatility to shock to daily gross purchases and gross sales in the bull market at day 2 even though both of them have positive accumulated effects on market volatility over 10-day period. Particularly, an increase in foreign gross purchases helps decrease market volatility but an increase in foreign gross sales makes the market more volatile. In addition, those two daily reactions (positive) 883

are of highest ones in 10 periods. Such findings could indicate that domestic investors in Vietnamese stock market (HoSE) are much worried when foreign investors sell more but trade prudently in the case foreigners increase their purchases when the market is experiencing a sharp up-trend. On the other hand, reaction of market volatility to foreign gross sales is even weaker in full sample. Most responses are insignificant (figure 3). However, one interesting point is that the accumulated effect after 10 days is negative, roughly -0.36 units (it is not a big impact, though) Figure 3: Response and Accumulated Response of Daily Market Volatility to Shocks in Daily Foreign Gross Sales (Full Sample) The final IRFs are those applied for net purchases in both two sub-periods. Response and accumulated response of market volatility to shock to foreign net purchases shown in Figure 4 (first period) are rather similar to those in Figure 1 (gross purchases). The reaction fluctuates over time. There are two negative responses of -1 unit on day 2 and -0.33 unit on day 8 in the first period. During the first four days, the effect is negative, which means that an increase in foreign net purchases will help reduce market volatility. However, after that, the level of volatility goes up and after 10 days, one-standard deviation increase in foreign net purchases will make a total rise by 0.88% in market volatility (accumulated response over a 10- day period). It seems that domestic investors spend time watching foreign move first and then begin to trade. This effect is much higher than the case in Thai Stock Market in Pavabutr s (2004) study. Pavabutr (2004) examine Thai stock market from January, 1995 to May, 2002 (three sub-periods: pre-asian financial crisis, crisis and post-crisis) and find that the effect of shock to foreign net purchases on market volatility was most significant in the crisis period: 0.1 % increase in volatility level after the tenth day for one standard deviation shock. This difference could be explained as follows. Pavabutr (2004) find that during the crisis period, foreign investors in Thailand were net buyers. Similarly, in Vietnam, up to 83% of daily foreign net purchases were positive (gross purchases are greater than gross sales) in the first 884

period. Therefore, net purchases in two countries are mostly cash inflows in the examined period. Pavabutr (2004) argue that during the Asian crisis period, foreign net purchases did not cause more volatility since those were inflows (buying) and it was domestic selling that led to a more volatile market. In Vietnam, even though foreign investors were net buyers most of the time, it was when market rocketed. Domestic investors could have a tendency to follows foreign moves which pushed the demand as well as the market go up remarkably during this time. Besides, over the period from 26/10/05 to 15/10/07 ( the first period), Vietnamese investors seemed not having a lot of information when trading stocks. Most of them were likely to follow or imitate foreigners trading because they believed in sophisticated information possessed by foreign investors. Figure 4: Response and Accumulated Response of Daily Market Volatility to Shocks in Daily Foreign Net Purchases (1st Period) However, moving on to the second period (Figure 5), it seems that domestic investors gained more trading experiences. Most of the time, the response of market volatility is negative to an increase in foreign net purchases. The 10-day accumulated response of market volatility is -1.95 % which is totally opposite to the first one. Most individual responses of market volatility are also negative to an increase in foreign net purchases. Consequently, an increase in net purchases of foreign investors does not induce more volatility like in the bull market but reduce it in the bear market. Put another way, in the second period, the stock market is less volatile to an increase in foreign net purchases as compared to the first one, which could indicate that domestic investors are then more prudent in trading and react to changes to foreign net purchases more slowly. 885

Figure 5: Response and Accumulated Response of Daily Market Volatility to Shocks in Daily Foreign Net Purchases (2nd Period) 4. Conclusion As demonstrated in our paper, past foreign flows have a significant relationship and impacts on Vietnamese stock market volatility. Interestingly, the relationship between foreign flows and Vietnamese stock market volatility is short-lived. Many significant relationships and (Granger) causations are found in daily dataset but just a few are discovered on a weekly basis. Secondly, past foreign flows relate to volatility stronger in the bull market as compared with the bear one: foreign gross purchases, foreign gross sales and foreign net purchases have significant links with volatility and all (Granger) cause market volatility in the bull market. Thirdly, all three flows have a positive accumulated impact (10-day) on market volatility in the bull market. In other words, a rise in any of those flows makes market more volatile during the specified period. Finally, opposite to the bull market, an increase in foreign net purchases reduces market volatility in the bear one, which could indicate that domestic investors are now more prudent in trading and react to changes to foreign net purchases more slowly. References Cao, C., Chang, E. C., & Wang, Y. (2008). An empirical analysis of the dynamic relationship between mutual fund flow and market return volatility. Journal of Banking & Finance, 32(10), 2111-2123. Decision No. 238/2005/GĐ-TTG of The Prime Minister on Percentage of Participation of Foreign Parties in Securities Market of Vietnam. Hanoi, 29 September, 2005. Retrieved from http://vanban.chinhphu.vn/portal/page?_pageid=578,33345598&_dad=portal&_schema=por TAL&docid=14891 Gujarati, D. (2004). Basic Econometrics (4 th ed). McGraw-Hill. Hansen, H. & Juselius, K. (1995). Cats in Rats: Cointegration analysis of time series. Evanston : Estima. 886

Huang, B.N., & Yang, C.W. (2000). The impact of financial liberalization on stock price. Volatility in emerging markets. Journal of Comparative Economics, 28, 321-339. Law, S-H., Ngah, S.R.S.W. (2008). Does stock market liberalization cause higher volatility in the Bursa Malaysia? International Journal of Business and Society, 9(1), 19-36. Nguyen, D.K.,& Bellalah,M. (2008). Stock market liberalization, structural breaks and dynamic changes in emerging market volatility. Review of Accounting and Finance, 7(4), 396 411. Pavabutr, P. (2004). Foreign portfolio flows and emerging stock markets: lessons from Thailand. (Doctoral dissertation, the University of Texas, Austin, 2004) Pavabutr, P., & Yan, H. (2007). The Impact of Foreign Portfolio Flows on Emerging Market Volatility: Evidence from Thailand. Australian Journal of Management, 32(2), 345-368. 887