Abstract: Dynaic Interaction aong Mutual Fund Flows, Stock Market Return and Volatility M.Thenozhi Professor, Departent of Manageent Studies, IIT Madras, Chennai-600 036 t_iit@yahoo.co and Manish Kuar Research Scholar, Departent of Manageent Studies, IIT Madras, Chennai-600 036 anishkuar_iit@yahoo.co.in This study has exained the dynaic interaction between utual fund flows and security returns and between utual fund flows and volatility. The results based on the conteporaneous relationship using daily data suggest that a positive relationship exist between stock arket returns and utual fund flows easured as stock purchases and sales. This positive concurrent relationship continues to exist even after controlling for volue. The analysis of causal relationship between utual fund flows and arket returns show that utual fund out flows (sales) are significantly affected by return in the equity arket, however, the latter is not significantly influenced by variation in these flows which suggests negative feedback trading behavior in the Indian arket. The results show that a strong positive relationship exists between stock arket volatility and utual fund flows easured as stock purchases and sales. This positive concurrent relationship continues to exist even after controlling for volue. The analysis on the direction of relationship between volatility and utual fund flows using the VAR approach suggests that arket volatility is positively related to lag flow, and that shock in flow has a positive ipact on arket volatility. The results provide evidence that the relationship is stable even after including these exogenous variables such as volue and arket fundaental variables such exchange rates, dividend and short ter interest rates in the odel. Increase in the aggregate inflows and outflows are associated with ore volatile arket. Introduction The cash flows into utual funds have generally been strongly correlated with arket returns and this relationship reflects the oentu trading or feedback trading hypothesis (Davidson and Dutia (989), Delong et al. (990), Hendricks et al. (993), Warther (995), and Zheng (999)). The hypothesis suggests that a shock to security returns leads to a change in utual inflows, which in turn leads to a further change in security returns. It is often stated that utual fund flows cause security returns to rise and fall and one possible reason attributed for this is the price pressure hypothesis (Harris et al. 986; Shleifer, 986). Price pressure theory suggests that increased inflows into equity
utual funds stiulate a greater deand by individuals to hold stock, and this causes share prices to increase while the inforation revelation hypothesis (Lee et al., (99) and Warther, (995)) suggests that if utual fund investors possess inforation or if they trade in the sae direction as another group of investors who possess inforation, then their trades will reveal or be associated with new inforation. Under this scenario, if utual fund investors are well infored their trades will be a signal to buy stocks and the arket in this case will not be responding to fund flows because of price pressure, but rather react efficiently to new inforation. However, if utual fund investors are unsophisticated and have a poor track record (noise traders), then the signal would be to sell stocks. Though, utual fund flows and stock returns have a high positive correlation as cited in literature, it does not necessarily ean that the forer causes the latter and vice versa. Potter (996) in his study used Granger causality tests to exaine the lead lag relationship between returns and fund flows for several categories of equity funds. The results show the evidence that security returns are useful to predict flows into aggressive growth funds but not into incoe funds. Soe studies (Warther (995), Reelona et al. (997)) have failed to find evidence that utual fund flows are affected by lagged security prices and security prices in one period are affected by utual fund flows in previous periods. Soe recent studies (Fortune (998), Christos et al. (2005), Natalie and Parwada (2007) show that feedback exists, i.e.,security returns do affect future fund flows, and soe fund flows do affect future security returns. However, it has also been found that fundaentals of fir influences fund flows than the stock returns (Cha and Lee (200)) and stock volatility influences flow of funds (Goetzann and Massa (2003)). Overall, the evidence on causal relationship between stock returns and utual fund flows is ixed and there is need for a odel which reveals the interaction aong stock returns stock volatility and flow of funds controlling for arket fundaentals. Moreover, ost of the studies (Warther (995), Fortune (998), Mosebach and Najand (999) etc) have used onthly data to test the causality. A few studies (Edelen and Warner (200)) have used daily data but for a short period of tie i.e. one and half year. The use of daily data and that too for a longer period is of paraount iportance for the results to be ore inforative to investor s. Hence, this study distinguishes itself fro prior work in several ways. Firstly, the study extends earlier studies by placing additional ephasis on role of arket fundaentals and risk in exaining the relationship between the utual fund flows and stock arket returns. Secondly, the study considers daily data for the utual fund flows and stock arket index. Thirdly, the study focuses on the Indian capital arket, naely the National Stock Exchange (NSE), where the interaction between utual funds flows and security returns is intense and the actions of the institutional investors have significant effect both on the behavior of the investors and on the prices of the securities listed in the NSE. Thus, the overall objective is to exaine whether the inforation on utual fund flows can be used to predict the changes in arket returns and volatility. Moreover, this study will also enhance the understanding of the dynaics in Indian arkets by analyzing the
conteporaneous and causal relationships between arket returns and utual fund flows and between arket volatility and utual fund flows. Specifically the ajor focus of this study is to exaine the conteporaneous and causal relationship between: (). Mutual fund flows and arket returns, (2). Mutual fund flows and arket returns in presence of control variables i.e. trading volue and in presence of arket fundaentals (3). Mutual fund flows and arket return volatility and (4). Mutual fund flows and arket return volatility in presence of control variables i.e. trading volue and in presence of arket fundaentals The reaining portion of this paper is organized as follows. Section 2 suarizes the findings fro literature. The data, saple period and the ethodology used for exaining the interaction aong utual fund flows, arket returns and arket volatility is explained in section 3. The epirical results of the study are discussed in section 4. Finally, section 5 suarizes the findings and brings out the iplications of the study. 2. Literature review: Previous studies that have exained the relationship between utual fund flows and stock arket returns have focused on the developed stock arkets i.e. the US arket. Warther (995) pioneered the study of security returns and aggregate utual fund cash flows. He exained the correlation between net inflows and security returns using onthly data for the period January 984 June 993. The net oney inflows were decoposed into expected and unexpected coponents. Expected fund flows were estiated by regressing current flows on past flows, and unexpected fund flows were derived as the residual fro the expected flow regression. The results provide the evidence that aggregate security returns are highly correlated with concurrent unexpected cash flows into utual funds but unrelated to concurrent expected flows. His result supports the popular belief that fund inflows and returns are positively related. However, the results reject both sides of a feedback trading odel, which eans that security returns neither lag nor lead utual fund flows. Reelona et al., (997) used a siilar ethodology to Warther s (995) to exaine the effects of arket returns on aggregate fund flows. However, the study iproves upon the work of Warther (995) in several ways. First of all, they included returns on other securities not held by the fund, as well as own returns, as deterinants of unexpected flows. Further, their regression of unexpected flows into a utual fund group on own returns and other returns was estiated using Instruental Variables rather than Ordinary Least Squares. The result shows that unexpected equity fund flows were not affected by either conteporaneous or lagged stock returns, while the bond fund flows were affected by conteporaneous bond returns, but not by lagged bond returns. However, the fact that fund flows and returns have a high positive correlation does not necessarily ean that the forer causes the latter and vice versa. Potter (996) used Granger causality tests to investigate the lead lag relationship between returns and fund flows for several categories of equity funds. The result provides the evidence that stock
returns can be used to predict the flows into aggressive growth funds, but the sae does not apply in the case of incoe funds. Moreover, the result also rejects the hypothesis that the fund flows in the four fund groups lead the security returns. Fortune (998) used VAR odels with seven variables and onthly data for the period January 984 through Deceber 996 to exaine the relationship between fund flows and returns. The result provided evidence of positive correlation between fund flows and conteporaneous returns. However, the results show that feedback do exists. Security returns do affect future fund flows and soe fund flows do affect future security returns. Overall, the evidence on causal relationship between stock returns and utual fund flows is ixed. The results of Fortune (998) are in strong contrast with the conclusions of Warther, Potter, and Reelona et al. that flows do not appear to be affected by past security returns. Potter and Schneeweis (998) in their study ade an attept to investigate the factors which affect aggregate utual fund flows. They found copeting investent classes to be econoically and statistically significant explainers of aggregate utual fund flows. The results also show that factors ipacting flows to riskier groups differ fro the factors deterining flows to less risky categories aong equity sub-categories. Moreover, the epirical results provide the evidence that security returns are useful in predicting flows into aggressive growth funds and growth funds. However, the results reject the hypothesis that equity fund flows lead security returns. Edwards and Zhang (998) eployed Granger causality test and instruental variable analysis to exaine the relationship between aggregate onthly utual fund flows and stock and bond onthly returns. The result shows that with one exception, flows into stock and bond funds do not affect either stock or bond returns. However, the agnitude of flows into both stock and bond funds are significantly affected by stock and bond returns. Mosebach and Najand (999) applied Engle and Granger error correction odel, followed by a state space procedure to exaine the long run equilibriu relation between the net flow of funds into equity utual funds and the S&P 500 index using onthly data fro January 984 to July 998. The results provide evidence of causal relation between the net inflow of funds and the stock arket. The result shows that the net flow of funds invested in the stock arket is influenced by the level of the stock arket in the previous onth. The result also shows that a current strong equity arket encourages ore investent in the arket. This iplies that the causality between the level of the stock arket and flow of funds into the arket is bi-directional. Edelen and Warner (200) exained the relation between stock arket returns and aggregate flows into US equity utual funds using high frequency daily data for the period 2 February 998 30 June 999. Their ajor findings are as follows. First, aggregate utual fund flow is correlated with concurrent arket returns at a daily frequency. This concurrent relation suggests that funds flow and institutional trading affect returns. Second, the results provide liited epirical evidence that utual fund
flow causes security prices to rise and fall (Warther (995)).Third, the results also find a very strong association between funds flow and the previous days return. This association indicates funds flow reacting to returns or to the inforation driving returns ainly with a one-day lag, but that investors generally require an overnight period to react. Papadaou and Siriopoulos (2002) used siilar ethodology to Warther s (995) to exaine the effect of arket returns on aggregate fund flows using onthly data fro the Greek equity utual fund investing spanning January 998 to March 2002. The result shows that there is sall positive concurrent relation between unexpected net flows and arket returns, which the author attributed to inforation revelation. The results also suggest soe evidence that utual fund flows cause prices to raise and to fall. The author finally concludes that there is low correlation between fund flows and returns. Goetzann and Massa (2003) exained the relationship between daily index fund flows and asset prices. The results indicate a strong conteporaneous correlation between fund inflows and S&P arket returns. The other objective of the study was to exaine shocks to prices originated by deand flows into index funds (typically "liquidity trading" types of shocks). The results provide the evidence that the arket reacts to daily deand, while only negative reactions appear due to past returns, and that the investors' behavior appears to be ainly otivated by risk aversion instead of return. Alexakis et al. (2005) exained the interaction between utual fund flows and stock returns in Greece. The statistical evidence derived fro the error correction odel indicates that there is bidirectional causality between utual fund flows and stock returns and cointegration results show that utual funds flow cause stock returns to rise or fall. Thus, inflows and outflows of cash in equity funds see to cause higher and lower stock returns in Greek stock arket. Oh and Parwada (2007) analysed relations between stock arket returns and utual fund flows in Korea. The results show that there is significant positive correlation between returns and both purchases and sales but a significant negative correlation is observed in the case of net flows. Tests on the direction of causality suggest that it is predoinantly returns that contain inforation on flows, although flows easured as stock purchases ay also contain inforation about returns. Thus, the review of literature shows that the theoretical literature has suggested several alternative otivations for the trading behavior of utual funds. The earlier literature tried to address the following fundaental questions: () Is institutional trading related to changes in stock prices? (2) Does institutional trading cause stock returns or do institutions siply follow oveents in stock prices? Most of the studies (Warther (995), Potter (996), and Reelona et al. (997), Potter and Schneeweis (998)) show that fund flows do not appear to be affected by past security returns. Soe studies (Mosebach and Najand (999), Alexakis et al. (2005) provide evidence of bi-directional causality between utual fund flows and stock returns, few studies (Edelen and Warner (200), Papadaou and Siriopoulos (2002) have shown liited evidence of utual fund
causing stock arket to rise and fall. The epirical results on the dynaic relation between utual fund trading and stock returns are also ixed. The key probles associated with the previous studies are as follows. First, ost of the previous epirical studies have focused priarily on the conteporaneous relation between stock returns and utual fund flows. The error distribution of the returns series does not exhibit constant variance. The assuption of constant variance over a tie period for the return series is not appropriate. Engle and Patton (200) in his study described the three stylized facts about volatility. First, volatility exhibits persistence. Periods of high and low volatility tend to be clustered. Second, volatility tends to be ean reverting. In other words, there is a noral level of volatility to which volatility eventually returns. Third, innovations ay have an asyetric ipact on volatility positive deviations fro the ean have ore (or less) of an ipact on volatility than negative deviations. There is a need to exaine the conteporaneous relationship between utual fund flows and stock returns after taking heteroscedasticity into account. ARCH/GARCH class of odel incorporates heteroscedasticity in a sensible way and they can be extended to include other effects on conditional variance. Second, ost have the studies have exained the conteporaneous or dynaic relationship between returns and utual fund flows. They have not included volatility in the analysis along with returns and utual fund flows. It is possible that the dynaic relationship between arket return and utual fund flows ay be affected by volatility effects associated with inforation flow and in part because volatility is a key ingredient of the risk-return tradeoff that pereates odern financial theories (Lee and Rui (2002)). In the last decade, volatility in the stock arket has received considerable attention fro investors, regulators and acadeicians and is especially closely onitored by derivatives traders since the derivative contracts is dependent upon the volatility of the underlying asset. Does utual fund flows affect arket volatility? If so, what is the direction of the relationship between utual fund flows and arket volatility? Warther (998) in his study asks the siilar question, whether utual fund flows have any ipact on the arket stability, but he does not indicate any epirical evidence on utual fund flows and volatility relationship. The epirical evidence about the relationship between arket volatility and utual fund flows sees to be absent. Third, ost of the studies (Karpoff (987), Gallant et al. (992), Blue et al. (994)) on volue and price changes have provided the evidence of positive relationship between volue and price change. However, Elden (999) indicated the positive relation between gross flow and trading volue. Hence, if utual fund flows is viewed as replaceent of trading volue, then the relationship between utual fund flows and return and utual fund flows and volatility would lead to siilar results because of the outcoe of the trading volue and price change relation. Therefore, there is need to exaine the relationship between returns and fund flows and volatility and fund flows after controlling for trading volue. Earlier studies have not included control variable i.e. trading volue while exaining the relationship between utual fund flows and stock arket returns.
Moreover, ost of the studies in this area have been carried in the well-developed financial arkets, usually the U.S. arkets. As Khorana et al. (2005) point out, there has been relatively little research perfored on utual funds outside the U.S. When copared to developed arkets, eerging arkets are considerably saller and less liquid. This dearth of liquidity can play an iportant role in deterining the relationship between stock returns and utual fund flows; it can potentially alter the previous findings for the developed arkets. Nowadays, any international investent bankers and brokerage firs have ajor stakes in overseas arkets. Harvey (995) found eerging arket returns are ore likely to be influenced by local inforation than developed arkets; in fact, eerging arket returns are generally ore predictable than developed arket returns. Indian stock arkets have received relatively little attention until recently. Now there is ore interest and research on Indian arket data due to the country s rapid growth and potential opportunities for investors. Since the establishent of National Stock Exchange (NSE), the financial arkets in this Asian country have attracted considerable global investents. National Stock Exchange of India Liited, started in 994 and within a short span of year becae the largest exchange in India in ters of volues transacted. Trading volues in the equity segent have grown rapidly with average daily turnover increasing fro Rs.7 crores during 994-95 to Rs.6,253 crores during 2005-06. In India, the 990s have seen unprecedented growth in utual funds. Before liberalization (99-992) the size of utual fund industry was just Rs., 000 crores. It rose to Rs. 4,00 crores in 99, and subsequently touched a new height of Rs. 72,000 crore in year 998. Since then, total assets under anageent has been increasing exponentially and thus revealing the efficiency of growth in the utual fund industry in India. The total assets under anageent of Mutual Fund industry rose by 9.45% fro Rs.309953.04 crores to 339232.46 crores in Noveber, 2006 as published by Association of Mutual Funds of India (AMFI). The Indian utual funds industry has been growing at a very healthy pace of 6.68 per cent for the past eight years and it is expected that the trend will continue. Thus, with the Indian stock arkets rallying to newer heights, utual funds in India are also rallying in ters of total assets under anageent in the tune with the arket. Given this background, the present study overcoes the drawback identified in the earlier study by exaining the conteporaneous as well as the dynaic (causal) relation between utual fund flows and return. The study also exaines the conteporaneous as well as the dynaic (causal) relations between utual fund flows and volatility of the S&P CNX Nifty Index of the National Stock Exchange of India. This study iproves upon previous studies in several aspects. First, the study exaines the relationship between return and utual fund flows in eerging arkets like India after taking heteroscedasticity into account. Here, the variance is conditioned on prior error ters, thus it allows the variance to change over tie. Most of the forer studies (Warther (995), Potter (996), and Reelona et al. (997), Edwards and Zhang (998), Potter and Schneeweis (998), Papadaou and Siriopoulos (2002), Mosebach and
Najand (999)) used onthly data, to exaine the relationship between utual fund flows and stock returns. Using the daily data of S&P CNX Nifty Index, the study exaines conteporaneous and causal relations not only between utual fund flows and arket returns but also between utual fund flows and volatility of returns. Second, iportant point that distinguishes this study fro the existing literature is ethodology adopted to investigate the dynaic relationship between variables of interest. The study exaines the dynaic relationship between returns and utual fund flows and volatility and utual fund flows using the Vector Autoregression (VAR) odel. Moreover, the study utilizes the exponential generalized autoregressive conditional heteroscedasticity (EGARCH) odel to easure return volatility. The proposed EGARCH odel accounts for the tie varying volatility process with asyetric responses to both positive and negative price changes. The study also uses a control variable i.e. trading volue to check whether the conteporaneous and causal relationship is still significant after controlling for trading volue. As an iproveent on the linear causality tests and to analyze whether utual fund flows affect arket returns and volatility in the presence of arket fundaentals, tests for the effect of utual fund flows in the presence of variables such as dividends, exchange rates and the interest rate is also been perfored in the spirit of Cha and Lee (200). The usefulness of including arket fundaentals lies in the fact that, should causality in this context only be in the direction of stock returns to flows, and not otherwise, then this would prove that it is only arket returns that drive utual fund flows. Thus, this study differs significantly, for it use of appropriate econoetric techniques, the uses of control variables and daily data for the eerging arket of India, where the interaction between utual funds flows and security returns is intense and the actions of the institutional investors (either right or wrong) have a wide effect both on the behavior of the less infored investors and on the prices of the securities of the National Stock Exchange of India Liited. 3. Data and Methodology The data set coprises daily arket index of S&P CNX Nifty Index of National Stock Exchange of India Liited. The series span the period fro st January 200 to 20 th April 2008. The daily stock index returns are continuously copounded rate of return, coputed as the first difference of the natural logarith of the daily stock index value. Given the price level P, P 2,..., P t, the return at tie t is fored by: R t = ln(p t /P t- ). In any given day, the utual fund flows in stock arket can be aggregated and suarized into two basic easures: sale and purchase, and a corresponding overall easure of net (total purchase total sales). Hence, this study uses three utual fund variables naely standardized purchase flow denoted as MFP; standardized sales flow denoted as MFS; standardized net flow denoted as MFN coputed as the difference between total purchases and total sales volues. All flows (Sales, Purchases and Net) are noralized by the trailing 90-day oving average of the S&P CNX Nifty arket capitalization to control for arket and fund growth as per Warther (995) and Oh and Parwada (2005). The sapling period is fro 4 th February 2000 to 20 th April 2007.
In order to study the relationship between utual fund flows and volatility, the daily arket volatility estiate is needed. Volatility is unobservable, hence in this study, the conditional return variance (volatility) of the S&P CNX Nifty Index is estiated using the EGARCH (,) odel proposed by Nelson (99). The EGARCH odel accounts for the tie varying volatility process with asyetric responses to both positive and negative price changes. The vector autoregression (VAR) odel for causality tests assues that the tie series under investigation are stationary. In order to test the stationarity of the arket returns, utual fund purchase, utual fund sales, utual fund net, dividend, exchange rate and MIBOR, the study eploys Augented Dickey and Fuller (ADF) test and the Phillip and Perrons (PP) test. a) Augented Dickey-Fuller Regression Yt () = ρ0 + ρyt ( ) + Yt ( i) + εt () i= b) Phillips-Perron Regression Yt = α0 + αy ( t ) + υt (2) The difference between the two unit root tests lies in their treatent of any nuisance serial correlation. The PP test tends to be ore robust to a wide range of serial correlation and tie dependent heteroscedasticity (Lee and Rui (2002)). The testing for stationarity is forulated in the statistical hypothesis testing fraework as a test of the null hypothesis H0: Series is non-stationary, against the alternative H: Series is stationary. Methodology for exaining Conteporaneous Relationship The error distribution of the stock returns series does not exhibit constant variance. The assuption of constant variance over a tie period for the return series is not appropriate. The ARCH/GARCH classes of odel (Engle (982), Bollerslev (986)) have shown their superiority not only in odeling heterscedasticity of financial tie series but can also be extended to include other effects on conditional variance. The estiate of the return volatility is obtained using an EGARCH odel (Nelson (99)). The exponential version of GARCH (EGARCH) to easure return volatility is used for several reasons. The EGARCH odel has several advantages over the GARCH odel. GARCH odel does not take into account the asyetry and non-linearity in the conditional variance. Moreover, the GARCH odel iposes positive constraints on the estiated paraeters. EGARCH odel iposes no positive constraints on the estiated paraeters and it takes care for asyetry in asset return volatility, thereby avoiding possible isspecification in the volatility process. In addition, EGARCH allows for a general probability density function (i.e., Generalized Error Distribution, GED), which nests the noral distribution along with several other possible densities. The EGARCH odel expresses the conditional variance of a given tie series as a non-linear function of its own past values and the past values of standardized innovations.
In order to test whether the conteporaneous relationship between utual fund flows and arket returns still exists after controlling for heteroscedasticity, the following EGARCH (,) odel is estiated. R = α + α F + ε (3) t 0 t t ε ε logσ = φ + ϕlogσ + γ + ψ + ξ (4) 2 2 t t t t σt σt and R = α + ε (5) t 2 t ε ε logσ = φ + ϕ logσ + γ + ψ + λf + ξ (6) 2 2 t t t 2 2 t 2 2 t σt σt Another iportant issue is the relationship between trading volue and price changes. Most of the epirical studies on volue-return and volue volatility have focused priarily on the conteporaneous relation between price changes and volue in three fors of the epirical relationship: a positive relationship between volue and stock returns (Epps (975), Rogalski (978)), a positive relationship between volue and absolute returns (Sirlock and Starks (988)) and an unrestricted V-shaped relationship between volue and return (Karpoff (987), Gallant et al. (992), Blue et al. (994)). Edelen (999) in his study provide the evidence of positive relation between gross flow (a half of the su of inflow and outflow) and trading volue. However, there is no positive relationship between net flow (inflow inus outflow) and trading volue. Hence, if utual fund flow is erely a substitute for trading volue, then the utual fund flows and returns and utual fund flows and volatility relationship will be spurious consequence of trading-return and trading volatility relationship. In order to test the conteporaneous relationship between return-utual fund flows and volatility-utual fund flows, trading volue is included in the equation 5 and 8. The trading volue, V t, is easured as ln(tv t /TV t- ) where TV, TV 2 TV t is the daily trading volue. The relationship between utual fund flows and stock arket returns and utual fund flows and arket volatility after including trading volue will be analyzed, by using the equation given below: R = α + α F + α V + ε (7) t 0 t 2 t t ε ε logσ = φ + ϕlogσ + γ + ψ + ξ (8) 2 2 t t t t σt σt and R = α + ε (9) t 3 t
ε ε logσ = φ + ϕ logσ + γ + ψ + λf + κv + ξ (0) 2 2 t t t 2 2 t 2 2 t t σt σt The equation 3 represents the EGARCH odel with utual fund flows (purchase, sales and net) in the ean equation. The equations 3 & 4 will be used to test whether conteporaneous relationship between utual fund flows and returns exists or not. The equation 6 represent the EGARCH odel with utual fund flows (purchase, sales and net) in the variance equation. The equations 5 & 6 will be used to test whether conteporaneous relationship between utual fund flows and volatility exists or not. The other objective of the study is to exaine the relationship between price and utual fund flows, after controlling for volue hence trading volue is included into the conditional ean and variance equation. The equation 7 & 8 will used to exaine the conteporaneous relation between returns and fund flows after controlling for volue, while equation 9 & 0 will used to exaine the conteporaneous relation between volatility and fund flows after controlling for volue The paraeters of the above equations (3 to 0) are estiated by axiu likelihood ethod. The left hand side of equation 4, 6, 8 and 0 is the log of the conditional variance. This iplies that the leverage effect is exponential, rather than quadratic. The exponential nature of the EGARCH ensures that the conditional variance is always positive even if the paraeter values are negative, thus there is no need for paraeter restrictions to ipose nonnegativity. ψ i captures the asyetric effect. The presence of leverage effects can be tested by the hypothesis that ψ i < 0. The ipact is asyetric if ψ 0.Table 2 to 5 reports the results of the estiated EGARCH (,) odel. i Methodology for exaining Dynaic Relationship The earlier section ainly ephasizes on the conteporaneous relationship between arket returns and utual fund flows and conditional volatility and utual fund flows. This section presents the Granger causality ethod to exaining the dynaic (causal) relationship. In bivariate case, the presence of Granger causality is tested by investigating whether the past of one tie series iproves the predictability of the present and future of another tie series. The study uses vector autoregression (VAR) odel to exaine the presence of linear Granger causality. The benefit of VAR odels is that they account for linear inter-teporal dynaics between variables, without iposing a priori restrictions of a particular odel. A VAR odel including S&P CNX Nifty stock index returns and utual fund flows can be expressed as:
R = α + β R + χ F + ε t i t i t RF i= i= and () F = η + µ R + π F + ε t i t i t FR i= i= (2) Siilarly a VAR odel including index returns volatility and utual fund flows can be expressed as: n h = φ + Ω h + F + ε t i t i t hf i= i= and n (3) n F = ϕ + ηh + ψ F + ε t i t i t Fh i= i= n (4) where R t, Ft and ht represent stock index returns, utual fund flows (purchase, sales and net) and conditional volatility, ε RF, ε FR and n denote autoregressive lag lengths. i.causality in Presence of Volue, ε hf and ε Fh are orthogonal error ters and As an iproveent on the causality tests and to analyze whether utual fund flows affect arket returns and volatility in the presence of trading volue, tests for the effect of utual fund flows in the presence of trading volue has also been perfored. The usefulness of including trading volue lies in the fact that, should causality in this context only be in the direction of stock returns to flows and volatility to flows, and not otherwise, then this would prove that it is only arket returns that drive utual fund flows. The following regression equations incorporating trading volue are used: A VAR odel including S&P CNX Nifty stock index returns and utual fund flows can be expressed as: R = α + β R + χ F + δv + ε (5) t i t i t t RFV i= i= and F = η + µ R + π F + γv + ε (6) t i t i t t FRV i= i= Siilarly a VAR odel including index returns volatility and utual fund flows can be expressed as: n h = φ + Ω h + F + νv + ε n (7) t i t i t t hfv i= i= and
n F = ϕ + ηh + ψ F + θv + ε n (8) t i t i t t FhV i= i= ii.causality in Presence of Market Fundaentals As an iproveent on the causality tests and to analyze whether utual fund flows affect arket returns and volatility in the presence of arket fundaentals, tests for the effect of utual fund flows in the presence of variables such as dividends, exchange rates and the interest rate is also been perfored in the spirit of Cha and Lee (200). The usefulness of including arket fundaentals lies in the fact that, these variables reflect the short run variations in the fundaentals of the Indian econoy, have been used together with the equity arket-related variables to see whether or not utual fund investors take into account their expectations about the state of the Indian econoy. The other reasons for including arket fundaentals lies in the fact that, should causality in this context only be in the direction of stock returns to flows, or volatility to flows and not otherwise, then this would prove that it is only arket returns that drive utual fund flows. A VAR odel including S&P CNX Nifty stock index returns and utual fund flows can be expressed as: (9) R = α + β R + χ F + Div + φ Axrate + ρ Dint+ ε t i t i t RF i= i= (20) F = η + µ R + π F + Div + φ Axrate + ρ Dint+ ε t i t i t 2 2 2 FR i= i= Siilarly a VAR odel including index returns volatility and utual fund flows can be expressed as: (2) h = α + βv + χ F + Div + φ Axrate + ρ Dint+ ε t i t i t hf i= i= (22) F = η + µ V + π F + Div+ φ Axrate+ ρ Dint+ ε t i t i t 2 2 2 Fh i= i= Within the context of this VAR odel, linear Granger causality restrictions can be defined as follows: If the null hypothesis that χ s jointly equal zero is rejected, it is argued that utual fund flows (purchase, sales and net) Granger causes returns. Siilarly, if the null hypothesis that µ s jointly equal zero is rejected, returns Granger cause utual fund flows. If both of the null hypotheses ar e rejected, a bi-directional Granger causality, or a feedback relation, is said to exist between variables. Siilar null hypothesis can be defined for the VAR odel including the index returns volatility and utual fund flows. Different test statistics have been proposed to test for linear Granger causality restrictions. To test for strict Granger
causality for pairs of { R t, F t ) and ( F t, h t ) in this linear fraework, a standard joint test (F-test) is used to deterine whether lagged value of one tie series has significant linear predictive power for current value of another series. 5. Results The Augented Dickey Fuller test and Philip Perrons test statistics as given in Table indicate that the all series are stationary as the absolute value of statistics is greater than the critical value. Conteporaneous Relationship Most of the studies (Warther (995, 998), Goetzann and Massa (998), Elden and Warner (200) etc) have reported that utual fund flows and stock arket prices tend to ove together over tie. In order to gain insight into the relationship between utual fund flows and arket returns the study estiates the EGARCH odel. Table 2 and 3 presents the estiation of EGARCH odel. In Table 2, EGARCH odel is estiated with fund flows in the ean equation. It is observed that arket returns are highly related with purchase, sales and net fund flows respectively. Since α coefficient reflects the relationship between concurrent returns and flow and is significant at % level. Thus, our results using longer tie period is consistent with the conteporaneous relationship presented in Warther (995, 998), Goetzann and Massa (998), Elden and Warner (200). In order to get a ore direct evidence of the relationship aong fund flow, arket returns and volatility, the study reconstructs EGARCH (,) odel. The daily utual fund flows (purchase, sales and net) are included in EGARCH (,) odel as exogenous variable, respectively. The epirical results are shown in Panel B of Table 3. It can be inferred that the relationship between concurrent volatility and utual fund flows is significantly positive. This iplies that an increased fund flow is associated with increased arket volatility. The relationship between concurrent returns and flow is significant at % level. Another issue that catches our attention is the well docuented relationship between price changes and trading volue (Karpoff (987), Gallant et al. (992), Blue et al. (994)). Edelen (999). Hence, this study includes trading volue as a repressor in the ean and variance equation of EGARCH (,) odel, to check whether fund flows coefficient is still statistically significant after controlling for trading volue, thus getting ore consistent and unbiased results. The results are presented in Table 4 and 5. The results in Table 4 suggests that flows are still strongly positively related to arket returns after controlling for trading volue. Moreover, results in the Table 5 suggests that fund flows are positively related to volatility after controlling for trading volue.
The evidence using daily data suggests that there exists a strong positive relationship between conteporaneous fund flows and arket returns and fund flows and volatility. The ajor findings using daily data is a strong positive concurrent relationship between stock arket returns and utual funds purchase, sales and net. The positive relationship between returns and fund flows and between volatility and fund flows ay be because of two reasons. One, the individual investors ay play an iportant part in the arket by buying (redeeing) utual funds shares when arket is up (down). Uninfored (or less infored) investors face ore difficulty in interpreting the price signals. Moreover, uninfored investors tend to revise their beliefs ore frequently, thus resulting in slower disappearance of price fluctuations fro their trading than those fro their infored counterparts after the new public inforation. Thus, uninfored investors would ore likely overreact to fundaental price oveents, which would lead to increased price volatility. Second, the trading strategies of utual fund anagers would also exert significant influence on the arket. Thus, the findings of concurrent positive relationship between returns and fund flows and between volatility and fund flows can be explained in a unified fraework of the ipact of individual investors and utual fund anagers. Dynaic Relationship Mutual fund flows and arket returns have a high positive correlation does not necessarily ean that the forer causes the latter and vice versa since there could be other explanations for this phenoenon. In order to gain ore insights into the relationship between utual fund flows and arket returns, the study eploys various VAR based test. Relationship between Fund Flows and Market Index Returns Granger causality test in the VAR frae work is perfored between returns and utual fund flows defined as purchase, sales and net. The correct lag length is deterined by using Schwarz inforation criteria. The results are reported in Table 6 whose coluns designate the dependent, or caused, variables and whose rows define the independent, or causing variables. The utual fund flows defined as Purchases, Sales and Net, fails to show any significant ipact on the arket return. The R 2 is low, however, at about 2%, which iplies that flows capacity to explain the arket return is only arginal. The arket return is positive and significant by its past two lags for all fund flows, while fund flows is significantly influenced by its past lags. This result iplies that an increase or decrease in utual fund flows tends to spur other utual fund investors to act in the sae direction. There is significant positive correlation between returns and Sales fund flows but a significant negative correlation is observed in the case of Net fund flows. It can be concluded that returns have a strong effect on utual fund outflows and net. This finding suggests that at an aggregate level, negative feedback trading is indicated, which is
inconsistent with the U.S. utual fund findings (Edelen and Warner, 200), but siilar to Japanese institutions (Ki and Nofsinger, 2005) and Korean arket (Natalie and Parwada, 2005). Moreover, the epirical evidence also suggests that utual fund sales and net can be predicted by the daily lagged flows and by the lagged S&P CNX Nifty Index returns. Direction of Relationship: The results of Granger causality test based on VAR fraework is reported in Table 7. The value of the Chi-square statistic suggests, whether the causing variable Grangercauses the caused variable or not. This is the test of the joint hypothesis that all coefficients on the causing variables (rows) in regressions with the caused variables (coluns) as dependent variables are zero. The significance level associated with each Chi-square statistic is the probability that a value of Chi-square equal to or greater than the observed saple value would occur by chance. A significance level of 0.05 or less indicates that Granger-causation exists; if the significance level exceeds 0.05, any effect of the causing variable observed in the data is attributed to chance. The hypothesis that returns does not Granger-cause flow is rejected for sales and net but accepted for purchase at high levels of statistical significance (Chi-Square statistics on Sales and Net are significant at the % level). The results suggest that, arket returns ay contain inforation about utual fund sales in Indian equity utual funds. However, the result accepts the hypothesis that flow does not Granger-cause return for Purchases, sales and net. The result of Granger causality corroborates the negative feedback trading hypothesis in Indian arket. Low past security returns otivates utual fund investors to involve in a feedback trading process, by redeeing utual fund shares. This evidence is exained further in the Granger causality tests designed to detect causal relationship between utual fund flows and arket returns after incorporating volue and arket fundaentals. The results reporting the direction of the equity fund flowsstock arket returns relationship in the presence of volue are presented in Table 8 and 9. Granger causality tests fro equity fund flows to stock arket returns are perfored by incorporating Volue as exogenous variable in the VAR fraework. The volue is appropriately odified to eet the stationarity condition. The results in Table 9 show that the null hypothesis that fund flows do not Granger-cause equity arket returns in the presence of volue is not rejected. This result is consistent with that of Table 7 in that equity fund flows do not affect stock arket returns directly in the presence of volue. Moreover, the results in Table 9 also show that the null hypothesis that arket returns do not Granger-cause equity fund flows in the presence of volue is rejected for sales and net but accepted for purchase. Again the results are consistent with that of Table 7. The results reporting the direction of the equity fund flows-stock and arket returns relationship in the presence of arket fundaentals variables such as dividends, exchange rates and MIBOR rates are presented in Table 0 and.
The findings in Table reject the hypothesis that equity fund flows do not Grangercause arket returns. In Table the hypothesis that returns do not Granger-cause flows in the presence of arket fundaentals is ore robustly rejected for Sales and Net but accepted for purchase. The fund flows-arket returns relationship in the presence of volue and arket fundaentals consistently suggests negative feedback trading by utual fund investors. In suary, stock arket returns contain additional inforation about equity fund flow(sales and net) while equity fund flows do not contain any additional inforation about arket returns iplies that equity fund flows ay be responding to changes in arket returns. The results of our study can be suarized as follows. We find that a positive relationship exists between stock arket returns and utual fund flows, and stock arket volatility and utual fund flows. The tests on the direction of causality suggest that it is predoinantly returns that contain inforation on flows. Dynaic Relationship between Mutual Fund Flows and Volatility: There are two school of thoughts through which the fund flow and return volatility are related. First school of thought suggests that the cash inflows (or outflows) into (or out of) funds at the individual fund level over a short period of tie (e.g., at daily frequency). This cash flow ight be related to past fund perforance. Fund anagers who follow positive feed back strategies rely on past perforance of stock to predict future returns. They buy securities when the arket goes up and sell it when the arket goes down thus, pushing security prices away fro their fundaental values. Other fund anagers ay follow negative feedback (contrarian) strategies, and their trades drive security prices toward their fundaental values. Since positive (negative) feedback strategies increase (decrease) short-ter volatility, the extent to which flow-induced trades depend on past return is iportant (Cao et al. 2008). A second school of thought (Black (986) and Lee et al. (99)) conclude that noise traders cause wide swings away fro fundaentals, and that investor sentient and noise traders are an iportant factor in the overall arket oveent. It is a coon belief that utual fund investors are the least infored investors. Thus, it is reasonable to use utual fund flow as a proxy for uninfored investor sentient. Reports in the popular press clai that fund flow is a good indicator of retail investor sentient, and this sentient is often irrational. To the extent that investor sentient is iportant in the arket place and aggregate flow is a good proxy of the sentient, flow into (or out of) utual funds will be related to arket-wide returns and volatility (Cao et al. 2008). In our bi-variate VAR, the coefficients of arket volatility and fund flows with one day lag characterize the relationship between volatility and fund flow. Table 2 reports coefficient estiates of the bivariate VAR odels. The lag length is selected based on the SBIC criteria.
Looking first at the bivariate odel, the results suggest that arket volatility is positively related to purchase (inflows) at lags suggesting that flow has a positive ipact on subsequent arket volatility. Moreover, arket volatility is also positively related to the sales (outflow). These results indicate that, holding everything else constant, purchase and sales are associated with higher volatility on the next day. The results in Table 2 also suggests that utual fund flows (purchase and sales) are negatively related to previous day volatility, which suggests that utual fund investors ight tie arket volatility. The study also exaines the Granger causality between volatility and flow based on the VAR results. The other objective of the study is to investigate the dynaic relation between utual fund flow and arket volatility, hence, the focus is to test () whether flow Granger-causes arket volatility and (2) weather volatility Granger-causes flow. Table 3 presents the Chi-Square test results and corresponding p-values. The results reject the null hypothesis that utual fund flow (purchase and sales) does not Granger-cause arket volatility at the 5% significance level for both VAR specifications. The chi-square statistics for the null Purchase does not granger cause Volatility and Sales does not granger cause Volatility are 22.6 and.8 respectively and their corresponding p-values are 0 and 0.0479. These results support our earlier finding that flow has a significant ipact on volatility and vice verse. Karpoff (987) in his study concludes that trading volue is related to volatility. This correlation between volue and volatility raises the concern of whether the volatility-flow dynaic is a spurious anifestation of the volatility-volue dynaic. In order to rule out this issue, the study includes volue as a exogenous variable in the VAR odel. The results in Table 4 and 5 present coefficient estiates and causal relationship based on the VAR odel. The results provide the evidence that the fund flow-volatility dynaic still holds, even after controlling for volue. Market volatility is positively related to fund flows (purchase and sales) at the first lags, with t-statistics of 4.4562 and.73068, respectively. Moreover, the results also corroborate our earlier findings that there is bi-directional causality between fund flows and volatility. Hence, it can be concluded that the ipact of fund flow on volatility is not a spurious anifestation of the ipact of overall volue on volatility. This result is consistent with the study of Cao et al. (2008). In order to gain further insights about the relationship between volatility and fund flows, the study includes arket fundaentals variables in the VAR fraework. Table 6 and 7, presents the results of the bivariate causality test between volatility and fund flows with fundaental variables as exogenous variable. Thisanalysis suggests that the fund flow-volatility dynaics still hold even after controlling for arket fundaentals.
Market volatility is positively related to fund flows (purchase and sales) at the first lags, with t-statistics of 4.25 and.7 respectively. Conclusion: This study has exained the dynaic interaction between utual fund flows and security returns and between utual fund flows and volatility in an eerging capital arket, naely the India. The results based on the conteporaneous relationship using daily data suggest that a positive relationship exist between stock arket returns and utual fund flows easured as stock purchases and sales. This positive concurrent relationship continues to exist even after controlling for volue. In order to investigate the causal relationship between utual fund flows and arket returns, Granger causality test has been perfored in the VAR fraework. The statistical evidence suggests that utual fund out flows (sales) are significantly affected by return in the equity arket, however, the latter is not significantly influenced by variation in these flows. Investors in the Indian arket extrapolate trends in stock price changes, and thus, after soe price decrease, they anticipate further dip in stock prices and hence sell shares. Such actions, when taken by a large nuber of investors, would suggest that stock prices will continue to decline in future. Therefore, investor s expectations lead the to sell utual fund units after a decrease in stock prices, respectively (Alexakis et al. 2005). This suggests the negative feedback trading behavior in the Indian arket. Mutual fund anagers selling decisions get affected by the arket returns. The results are inconsistent with the U.S. utual fund findings (Edelen and Warner, 200), but siilar to Japanese institutions (Ki and Nofsinger, 2005) and Korean arket (Natalie and Parwada, 2005). The other objective of the study is to exaine the dynaic relationship between aggregate utual fund flow and arket-wide volatility. The results based on the conteporaneous relationship using daily data suggest that a strong positive relationship exists between stock arket volatility and utual fund flows easured as stock purchases and sales. This positive concurrent relationship continues to exist even after controlling for volue. The results are consistent with the results of Oh and Parwada (2005) for the Korean arket. Further analysis on the direction of relationship between volatility and utual fund flows using the VAR approach suggests that arket volatility is positively related to lag flow, and that shock in flow has a positive ipact on arket volatility. The study also attepted to uncover the dynaic relationship between utual fund flows and volatility using exogenous variable such as volue and arket fundaental variables such exchange rates, dividend and short ter interest rates. The results provide evidence that the relationship is stable even after including these exogenous variables in the odel. Increase in the aggregate inflows and outflows are associated with ore volatile arket. Overall, the study using daily data enabled us to conduct ore rigorous test and shed light on the iportance of the relationship between utual fund flows and arket returns and utual fund flows and volatility. The results of the study can be incorporated in the
trading odel used by practitioners and investors. Regulators can consider the bidirectional influence of fund flows and volatility of stock arket and the fund flows changes in the arket returns affects fund flows in their policy decisions. Further Variable Purchase Sales Net Coefficient Z-statistic Coefficient Z-statistic Coefficient Z-statistic α 0-0.002528-4.5870* -0.000935 -.48680 0.00207 4.23497* α 6.03404 8.22632* 8.508474 3.73089* 7.933399 7.265736* research on the utual fund industry, possibly on an individual utual fund level and the copleentary use of event studies, ay help in iproving our understanding of the relationship between utual fund flows and stock returns. Table. Unit Root Test Augented Dickey Fuller Test Philip Perron Test Series Statistic Critical Value Statistic Critical Value Index Returns -3.24544-3.433767-38.24342-3.433765 Purchase -7.06475-3.433775-34.06603-3.433765 Sales -8.95293-3.43377-33.54594-3.433765 Net -5.648-3.43377-38.89456-3.433765 Dividend Yield -7.56443-3.433773-43.6545-3.433765 MIBOR -7.3888-3.433793-69.49705-3.433765 Exchange Rate -40.236-3.433765-40.27702-3.433765 Table 2: Conteporaneous relationship between Mutual Fund flows and Market Returns