Empirical Analysis of the Impact of Foreign Institutional Investment on the Indian Stock Market Volatility during World Financial Crisis 2008-09

Capital inflows play a substantial role in developing countries. It used to increase accumulation and rate of investments to create conditions for more intensive economic growth. Capital inflows are necessary for macroeconomic stability as capital inflows affect a wide range of macro economic variables such as exchange rates, interest rates, foreign exchange reserves, domestic monetary conditions as well as saving and investments. Capital inflows, however, are not without risk. The main risk posed by large and volatile capital inflows is that they may resulted in crisis and destabilize macroeconomic management. Given the role of FII flews and its associated risks, the main purpose of this paper was to investigate the cointegration and causality between the Indian stock market and foreign institutional investment (FII) In India during world financial turmoil of 2008. The cointegration and causal relationship using Engle-Granger (1987), Johansen (1991, 1995a) and Granger (1969) methodologies were investigated .The study found that BSE500 stock index and FII series are cointegrated and causality between them is bilateral.

All that has been done is to take one of the variables, X1,t , and to rearrange (2.68) to make it the subject. It could also be said that the equation has been normalized on X 1,t . But viewed another way, (2) is just a regression equation where t z is a disturbance term. These disturbances would have some very undesirable properties: in general t z , will not be stationary and is autocorrelated if all of the X i are I (1). As a further illustration, consider the following regression model containing variables y t , x 2t , x 3t which are all I(1) For the estimated model, the SRF would be written 4 Taking everything except the residuals to the LHS 5 Again, the residuals when expressed in this way can be considered a linear combination of the variables. Typically, this linear combination of I(1) variables will itself be I(1), but it would obviously be desirable to obtain residuals that are I(0). Under what circumstances will this be the case? The answer is that a linear combination of I(1) variables will be I(0), in other words stationary, if the variables are cointegrated. 2.1.1. Definition of Cointegration (Engle And Granger, 1987) Let wt be a k ×1 vector of variables, then the components of wt are integrated of order (d, b) if: (1) All components of wt are I(d) (2) There is at least one vector of coefficients α such that ~I( ) In practice, many financial variables contain one unit root, and are thus I(1), so that the remainder of this section will restrict analysis to the case where d = b = 1. In this context, a set of variables is defined as cointegrated if a linear combination of them is stationary. Many time series are non-stationary but 'move together' over time -that is, there exist some influences on the series (for example, market forces), which imply that the two series are bound by some relationship in the long run. A cointegrating relationship may also be seen as a long-term or equilibrium phenomenon, since it is possible that cointegrating variables may deviate from their relationship in the short run, but their association would return in the long run (Brooks, 2008).

Methods of Parameter Estimation in Cointegrated Systems
There are (at least) three methods that could be used: Engle--Granger, Engle-Yoo and Johansen. The first and third of these will be considered in some detail in this study.

The Engle-Granger 2-step method
This is a single equation technique, which is conducted as follows: Step 1 In the first step one should make sure that all the individual variables are I (1). Then the cointegrating regression using OLS should be estimated. In this step any inferences on the coefficient estimates in this regression is impossible -all that can be done is to estimate the parameter values. Then saving the residuals of the cointegrating regression, ˆt u is required. Finally, one should test these residuals to ensure that they are I (0). If they are I (0), proceed to Step 2; if they are I(1), estimate a model containing only first differences.
Step 2 In this step the step 1 residuals should be used as one variable in the error correction model, e.g.
Where . The stationary, linear combination of nonstationary variables is also known as the cointegrating vector. In this case, the cointegrating vector would be Additionally, any linear transformation of the cointegrating vector will also be a cointegrating vector. It is now valid to perform inferences in the second-stage regression, i.e. concerning the parameters β1 and β2 (provided that there are no other forms of misspecification, of course), since all variables in this regression are stationary. Suppose that the following specification had been estimated as a potential cointegrating regression What if instead the following equation was estimated?
If it is found that 1~I (0) t u , does this imply automatically that 2~I (0) t u ? The answer in theory is 'yes', but in practice different conclusions may be reached in finite samples. Also, if there is an error in the model specification at stage 1, this will be carried through to the cointegration test at stage 2, as a consequence of the sequential nature of the computation of the cointegration test statistic.
(3) It is not possible to perform any hypothesis tests about the actual cointegrating relationship estimated at stage 1.

Testing For and Estimating Cointegrating Systems Using the Johansen Technique Based On Vars
Suppose that a set of g variables (g ≥ 2) are under consideration that are I(1) and which are thought may be cointegrated. A VAR with k lags containing these variables could be set up: The test for cointegration between the ys is calculated by looking at the rank of the  matrix via its eigenvalues.
The rank of a matrix is equal to the number of its characteristic roots (eigenvalues) that are different from zero. The eigenvalues, denoted i  are put in ascending order 1 2 ..... g

     
If the s  are roots, in this context they must be less than 1 in absolute value and positive, and 1  will be the largest (i.e. the closest to one), while g  will be the smallest (i.e. the closest to zero). If the variables are not cointegrated, the rank of  will not be significantly different from zero, so 0 where r is the number of cointegrating vectors under the null hypothesis and ˆi  is the estimated value for the ith ordered eigenvalue from the  matrix. Intuitively, the larger isˆi  , the more large and negative will beln(1 ) i   and hence the larger will be the test statistic. Each eigenvalue will have associated with it a different cointegrating vector, which will be eigenvectors. A significantly non-zero eigenvalue indicates a significant cointegrating vector. trace  is a joint test where the null is that the number of cointegrating vectors is less than or equal to r against an unspecified or general alternative that there are more than r. It starts with p eigenvalues, and then successively the largest is removed. trace  = 0 when all the i  = 0, for i = 1, . . . , g. max  conducts separate tests on each eigenvalue, and has as its null hypothesis that the number of cointegrating vectors is r against an alternative of r + 1. Johansen and Juselius (1990) provide critical values for the two statistics. The distribution of the test statistics is non-standard, and the critical values depend on the value of g − r , the number of non-stationary components and whether constants are included in each of the equations. Intercepts can be included either in the cointegrating vectors themselves or as additional terms in the VAR. The latter is equivalent to including a trend in the data generating processes for the levels of the series. If the test statistic is greater than the critical value from Johansen's tables, reject the null hypothesis that there are r cointegrating vectors in favor of the alternative that there are r + 1 (for trace  ) or more than r (for max  The matrix β gives the cointegrating vectors, while α gives the amount of each cointegrating vector entering each equation of the VECM, also known as the 'adjustment parameters'. Finally, it must be noted that the above description is not exactly how the Johansen procedure works, but is an intuitive approximation to it (Brooks,2008).

Granger Causality
2.4.1. Two-Variable Models According to (Idris and Cheong, 2004), Correlation, does not necessarily imply causation in any meaningful sense of the word. The econometric graveyard is full of magnificent correlations that are simply spurious or meaningless (Eviews 3). As for the efficient-market hypothesis (EMH), there may not necessarily be a real association between market efficiency and co-integration. Granger (1988) argues that co-integration between two prices imply an inefficient market as the error correction model indicates that at least one of the prices is predictable. Therefore, the Granger-type causality procedure (Granger, 1969 and1988) was applied to determine the direction of causation among the Y and X series. The causality procedure was conducted based on a bivariate system [x, y].
Let t X , t Y , be two stationary time series with zero means. The simple causal model is In above equations m can equal infinity but in practice, of course, due to the finite length of the available data, m will be assumed finite and shorter than the given time series. The definition of causality given above implies that t Y is causing t X provided some j  is not zero. Similarly t X is causing t Y if some c i is not zero. If both of these events occur, there is said to be a feedback relationship between t X and t Y . The more general model with instantaneous causality is If the variables are such that this kind of representation is needed, then instantaneous causality is occurring and a knowledge of t Y will improve the "prediction" or goodness of fit of the first equation for t X (Granger,1969).
The null hypothesis is that X does not Granger cause Y.

H0: No causal relationship from X to Y Ha: X Granger-causes Y
This hypothesis would be rejected if the coefficients of the lagged X were jointly significant (different from zero). The null hypothesis for equation (2) is that "Y does not Granger cause X." This hypothesis would be rejected if the coefficients of the lagged Y were jointly significant. If both of these null hypotheses are rejected, then a bidirectional relationship (X ↔Y) is said to exist between the two variables. The causality patterns can be unidirectional causality, (Granger,1969) 3. Review of literature Mukherjee et al(2002) ,in their paper explored the relationship of foreign institutional investment (FII) flows to the Indian equity market with its possible covariates based on a daily data-set for the period January 1999 to May 2002. The set of possible covariates considered comprises two types of variables. The first type includes variables reflecting daily market return and its volatility in domestic and international equity markets as well as measures of co-movement of returns in these markets (viz., relevant betas). The second type of variables, on the other hand, are essentially macroeconomic ones like exchange rate, short-term interest rate and index of industrial production (IIP)-viz., variables that are likely to affect foreign investors' expectation about return in Indian equity market. Their results show that: (1) FII flows to and from the Indian market tend to be caused by return in the domestic equity market and not the other way round; (2) returns in the Indian equity market is indeed an important (and perhaps the single most important) factor that influences FII flows into the country; (3) while FII sale and FII net inflow are significantly affected by the performance of the Indian equity market, FII purchase is not responsive to this market performance; (4) FII investors do not seem to use Indian equity market for the purpose of diversification of their investment; (5) return from exchange rate variation and fundamentals of the Indian economy may have influence on FII decisions, but such influence does not seem to be strong, and; finally, (6) daily FII flows are highly auto-correlated and this auto-correlation could not be accounted for by the all or some of the covariates considered in their study. They stated that, policy implications of the findings are that a move towards a more liberalized regime in the emerging market economies should be accompanied by further improvements in the regulatory system of the financial sector. their results additionally suggested that in the case of India (and other countries having thin and shallow equity markets) the prime focus should be on regaining investors' confidence in the equity market so as to strengthen the domestic investor base of the market. Once this is achieved, a built-in cushion against possible destabilizing effects of sudden reversal of foreign inflows might develop. Only then would it be possible to reap fully the benefits of capital market integration. Bose and Coondoo (2004), examined the impact of the FII policy reforms on FII portfolio flows to the Indian stock markets. They tried to assess the impact on FII flows of several policy revisions related to FII investment during the period January 1999 to January 2004, through a multivariate GARCH regression model. Using techniques of time series intervention analyses they incorporated the effect of each individual policy intervention in a model that includes the two most important covariates of FII flows to India, namely stock market (BSE) returns and past FII flows. The range of policies considered encompasses liberalization policies as well as restrictive ones taken to assure stability of flows. Their results strongly suggested that liberalization policies have had the desired expansionary effect and have either increased the mean level of FII inflows and/or the sensitivity of these flows to a change in BSE return and/or the inertia of these flows. On the other hand, interestingly, the restrictive measures aimed at achieving greater control over FII flows also do not show any significant negative impact on the net inflows; we find that these policies mostly render FII investments more sensitive to the domestic market returns and raise the inertia of the FII flows. Anokye and Tweneboah (2008), Using multivariate cointegration and error correction model, examined the impact of Foreign Direct Investment (FDI) on the stock market development in Ghana. Their results indicated that there exists a long run relationship between FDI, nominal exchange rate and stock market development in Ghana. They found that a shock to FDI significantly influence the development of stock market in Ghana. Their results had several policy implications. First, it supports the policy maker's decision to slash restriction for the non-resident investors for listed companies. This would attract major investors to other sectors of the economy to bring the needed growth in the exchange market and the economy as a whole. Second, policy makers should devise strategies to increase the FDI stock (retain FDI) and offer incentives for long investing and listing on the stock market. Sumanjeet (2009) studied Foreign Capital Flows into India .he stated that existing studies reveals that the huge surge in international capital flows since early 1990s has created unprecedented opportunities for the developing countries like India to achieve accelerated economic growth. International financial institutions routinely advise developing countries to adopt policy regimes that encourage capital inflows. Since the introduction of the reform process in the early 1990s, India has witnessed a significant increase in capital inflows. The size of net capital inflows to India increased from US $ 7.1 billion in 1990-91 to US $ 108.0 billion in 2007-08. Today, India has one of the highest net capital inflows among the EMEs of Asia. Capital inflows, however, not an unmitigated blessing. The main danger posed by large and volatile capital inflows is that they may destabilize macroeconomic management. As evident, the intensified pressures due to large and volatile capital flows in India in the recent period in an atmosphere of global uncertainties has posed new challenges for monetary and exchange rate management. He concluded that undertaking more economic reforms is not easy but has to be done: the government can either manage the process or competitive forces will bring it upon us in a lopsided manner. The ball is in the government's court. Countries that permit free capital flows must choose between the stability provided by fixed exchange rates and the flexibility afforded by an independent monetary policy. Bansal And Pasricha (2009), studied the impact of market opening to FIIs, on Indian stock market behavior. India announced its policy regarding the opening of stock market to FIIs for investment in equity and related instruments on 14th September 1992. Using stock market data related to Bombay Stock Exchange, for both before and after the FIIs policy announcement day, they conducted an empirical examination to assess the impact of the market opening on the returns and volatility of stock return. they found that while there is no significant changes in the Indian stock market average returns, volatility is significantly reduced after India unlocked its stock market to foreign investors.

Data
The monthly data of BSE500 and FII series used as proxy for the Indian stock market and FII series.

Cointegration
In this part of study we evaluated the relationship between BSE500 stock index and FII series. One of the commonly used tools for evaluation relation between 2 variables is correlation. To test this relation between BSE500 and FII we calculated the correlation between BSE500 movements and FII series for equity, debt and their total. The results are presented in table 2.
As results show there is a high and significant correlation between BSE500 movements and FII flows in all cases. According to (Watson and Teelucksingh, 2002) estimation and hypothesis testing based on OLS is justified only if the two variables involved are I (0). Since both BSE500 and FII series in the level form are I (1), a fairly reasonable expectation is that any linear combination of these two variables, such as t e , would also be I (1). This violates the basic assumptions for OLS estimation and if we insist on applying OLS, we are likely to establish nothing more than spurious correlation, i.e. a correlation that does not establish any causal relationship between the two variables. Therefore, whether the above correlations are true should be evaluated comprehensively before any implication.
A tempting solution to this apparent problem is to fit the regression using the first differences of both variables. But in a seminal paper, Davidson, Hendry, Srba and Yeo (1978) argue that such an approach would ignore valuable information about the "long run". They propose instead an approach that combined the two and incorporates the short run dynamics implied by the first differences as well as the static or long run relationship between the undifferenced values which enter the relationship as an "error correction mechanism" (ECM). Engle and Granger (1987) show that the solution proposed by Davidson et al. (1978) is possible if and only if the variables involved in the relationship are cointegrated.
To test cointegration we test whether there is any cointegration and causality between movement in the BSE500 stock Index and FII series. First we used the Engle-Granger methodology to see if there is cointegration between two series. Before testing the residuals we made sure of unit root in both series. The results are reported in the tables 3.
As tables show both series are I (1) i.e. non-stationary in the level with intercept.
To test the cointegration between two series we estimate the following equation

18
Where B is BSE500 index and F is FII series. Step 1 of the Engle-Granger two-step procedure requires that we fit the cointegrating regression (by OLS) and test the residuals for unit roots (we have already established that B and F are I(1)). We estimated the cointegrating regression by OLS the results are reported in table 4. If the residuals are stationary, then we must reject the null of no cointegration. If they are nonstationary (the null hypothesis), then they do not cointegrated. If the null of no cointegration based on ADF test is rejected, the next step in the Engle-Granger procedure is to estimate the short-run dynamics. Using the Engle-Granger methodology we test the null, the results are presented in table 5. Since the test statistic (−4.18) is more negative than the critical values at the 1% level, the null hypothesis of a unit root in the test regression residuals is strongly rejected. We thus conclude that the two series are cointegrated. This means that an error correction model (ECM) can be estimated, as there is a linear combination of the BSE500 and FII series that would be stationary. The ECM would be the appropriate model rather than a model in pure first difference form because it would enable us to capture the long-run relationship between the series as well as the short-run one. Therefore, we estimate an error correction model by running the following regression Therefore, the next step in the Engle-Granger procedure is to estimate the short-run dynamics in an equation system such as equations above, with: The results are reported in the tables 6 and 7. 1  is estimated as -0.135103. It has correctly signed and it is significant. This is enough to confirm that BSE500 and FII are cointegrated and that the ECM form estimated here is a valid representation of the model. 2  is estimated as 0.646659 but it is not significant. A test of the residuals of these equations verifies that they are white noise. This is further evidence of the cointegratability of BSE500 and FII. Finally, according to above results about 13.5% of disequilibrium "corrected" each month.
Since we have only two variables and Engle-Granger approach can estimate only up to one cointegrating relationship between two variables then in our study, there can be at most one cointegrating relationship since there are only two variables in the model it is enough to our purpose here. Nevertheless, we examine the issue of cointegration within the Johansen VAR framework to confirm The correlation between two variables. The results are as follows: The first step in the Johansen is to establish whether the variables are cointegrated and, indeed, whether there may be more than one cointegrating vector (a maximum of two is possible here). We use assumption 2 (no deterministic trend in the data, intercept in the CE) and an underlying VAR with two lags. The results are displayed in table 8.
The first thing to do in the Johansen procedure is to test the null of r = 0 against the alternative r ≥ 1. As results show the Johansen test statistics show rejection for the null hypothesis of no cointegrating vectors under both the trace and maximal eigenvalue forms of the test. In the case of the trace, the null of no cointegrating vectors is rejected since the test statistic of 25.68 is greater than the 1% critical value of 20.04. Therefore, there is at least one cointegrating vector.
The next step is to test the null of r = 1 against the alternative of r ≥ 2. Moving on to test the null of at most 1 cointegrating vectors, the trace statistic is 0.58 while the 5% and 1% critical values are 3.76 and 6.65 respectively, so the null is not rejected at both level of significance. Thus, we conclude that there is exactly one cointegrating vector.
The normalized cointegrating vector is estimated as (including the constant term): ˆ( 1 8.967 )

   
The corresponding cointegrating regression deduced from normalization is: where the right-hand side of this equation is in the form . For purposes of comparison with OLS, it is perhaps better to write this result in the following more conventional format: This is somewhat different from the OLS result (which would have been the cointegrating vector obtained by application of the EG two-step procedure) although the difference is not dramatic.
We turn now to the estimation of the ECM model. The results obtained are shown in appendix 1. .At the top of table the (normalized) cointegrating vector is displayed and, below, the ECMs involving t B  and t F  as "dependent" variables are shown. On the right-hand side of each equation appears the cointegrating regression (CointEq1) and the coefficient attached to it is the "adjustment parameter". Here the adjustment coefficient associated with the ΔB t equation is negative (-0.125841) and it is also significant (t-statistic = (2.11526). This is sufficient to reject any "no cointegration" hypothesis.
values of Y and then to see whether adding lagged values of X can improve the explanation. Y is said to be Granger-caused by X if X helps in the prediction of Y, or equivalently if the coefficients on the lagged X's are statistically significant. Note that two-way causation is frequently the case; X Granger causes Y and Y Granger causes X. It is important to note that the statement "X Granger causes Y" does not imply that Y is the effect or the result of X. Granger causality measures precedence and information content but does not by itself indicate causality in the more common use of the term.
Using the granger causality framework we test the null hypothesis that FII does not Granger-cause BSE and that BSE does not Granger-cause FII in the. To test these hypotheses we estimated two following bivariate regression models respectively: Where: In order to test the above hypotheses the usual Wald F-statistic test is utilized, which has the following form: Where: RSS U = the sum of squared residuals from the complete (unrestricted) equation RSS R = the sum of squared residuals from the equation under the assumption that a set of variables is redundant, when the restrictions are imposed, (restricted equation) T = the sample size q = the lag length We picked a lag length, based on our assumption about the time period over which FII could help predict the BSE500 index. Our assumption was that BSE500 affected by FII outflows from the beginning of 2008 to Dec 2008.We used the SC information criterion to select an appropriate lag length in our analysis. SC criterion indicates lag order selected by the criterion is 12.All other criterion also confirm this selection. Given the selected lag length we tested the estimated models to test the hypothesis regard Granger causality between the variables under question. The results are reported in the table 9.
As a decision rule, if F calc  F crit then reject H 0  lagged FII terms are significant,  FII does Granger-cause BSE500.
According to the results, we do reject both null hypotheses. Econometrics results reveal that causality is bilateral. In other words, all coefficients are significant. The evidence of causal linkage between two variables implies that since each of this variables are cointegrated the predictability of each can enhanced significantly by utilizing information on other variable.

Conclusions and policy recommendations
Decline of about 60 percent in the index and a wiping off of about USD1.3 trillion in market capitalization since January 2008 when the Sensex had peaked at about 21,000 which was primarily due to the withdrawal of about USD 12 billion from the market by foreign portfolio investors between September and December 2008 (Kumar, 2009) and effects of this on local investors responses, was a disaster for the Indian stock market. This paper examined the cointegration and causal relationship between BSE500 and FII series in Indian economy. Since estimation and hypothesis testing based on OLS is justified only if the two variables involved are I (0) and since both BSE500 and FII series in the level form are I (1), a fairly reasonable expectation is that any linear combination of these two variables, such as t e , would also be I (1). This violates the basic assumptions for OLS estimation and if we insist on applying OLS, we are likely to establish nothing more than spurious correlation, i.e. a correlation that does not establish any causal relationship between the two variables. Therefore, whether the above correlations are true should be evaluated comprehensively before any implications (Watson and Teelucksingh, 2002).
To overcome this problem we test whether there is any cointegration and causality between movement in the BSE500 stock Index and FII series. Using two popular econometrics methods our results confirmed the cointegration between BSE500 and FII series. Further, econometrics results reveal that causality between BSE500 and FII is bilateral. The evidence of causal linkage between two variables implies that since each of this variables are cointegrated the predictability of each can enhanced significantly by utilizing information on other variable. Given the risk of capital inflows, it is recommended that price limits and volume quotas be used relative to the status of both the economy and Indian stock market trading cycles. When the economy is in recovery, these tools need to be loosened to attract investors and reduce the investors' pessimism about the market. However, in an extreme boom these tools need to be applied with greater intensity to reduce excess optimism that could develop into speculative behavior in the market. In edition to curb the negative impact of market volatility it is better to apply only lower level price limits and volume quotas, rather than both upper and lower limits in a time of panic or when the markets start to fragment. Without any doubt FII play a great positive role in the Indian economy. But as world financial crisis of 2008 showed it is a potential risk as well, a risk with low frequency but high severity which classified as catastrophic risk in the literature. This type of risk is speculative risk and is not insurable. Therefore it is necessary to cover this risk to prevent future potential crisis. Given the impossibility of insuring this type of risk by direct and indirect insurance system, it is recommended that Indian financial system policy makers consider a self insurance (hedging) mechanism through establishing a captive and build up funds there based on Maximum Possible Loss(in this case capital outflows) to prevent same future probable crisis.