A. Previous Empirical Work

Initial research on foreign portfolio investment highlighted the phenomenon of home bias, whereby despite the advantages of international diversification, portfolios were found to be heavily skewed towards domestic securities (French and Poterba, 1991). Over time, the degree of home bias in equity portfolios has declined. In keeping with this trend, the literature now focuses on explaining cross-border equity flows, rather than explaining their absence. The main questions addressed in the literature are: (i) what is the association between portfolio inflows and pull factors such as domestic market returns; (ii) what is the association between portfolio inflows and push factors such as foreign market returns; and (iii) what is the relative importance of pull versus push factors in determining portfolio inflows.

The literature generally shows a positive contemporaneous relationship between portfolio inflows and domestic stock market returns (Bohn and Tesar, 1996). However, this association need not necessarily imply causality or “return chasing” by foreign portfolio investors. The observed relationship might reflect exogenous changes in investor preferences that cause funds to flow into the host market, and hence bid up prices (Brennan and Cao, 1997). A positive contemporaneous correlation between inflows and domestic returns may thus be entirely spurious. Nevertheless, using daily data for Asian economies, Richards (2002) finds portfolio inflows to be positively associated with the lagged return in the domestic stock market. This is consistent with foreign investors being positive feedback traders. Griffin et al. (2002) also establish this relationship for several Asian countries using daily data. However, they do not find evidence of positive feedback trading for India, a point to which we return below.⁹

There is little consensus on the relationship between portfolio inflows and external or source market returns. For example, Bohn and Tesar (1996) and Brennan and Cao (1997) do not find any significant association between the purchase of foreign equities by U.S. investors and U.S. equity returns. Again the studies using daily data are more definitive. Both Richards (2002) and Griffin et al. (2002) find a positive and significant association between daily portfolio flows to Asian economies and lagged U.S. stock returns.

Results also vary across papers regarding the general question of whether external or domestic factors are more important in influencing portfolio flows.¹⁰ Calvo et al. (1993) find that global interest rates and business cycle conditions have been the most important determinants of inflows into Latin America. By contrast, Chuhan et al. (1998) find that domestic factors are at least as important as external factors in explaining flows to Asian countries. For transition economies, Garibaldi et al. (2002) find that both domestic and external factors, including past inflation, credit ratings, exchange rate regime, the level of reserves, world interest rates, and improvements in securities market infrastructure, are significantly associated with portfolio investment.¹¹

Regional factors have also been considered in the literature. An early paper to recognize their importance was Buckberg (1996), who proposed that portfolio flows to an emerging market be thought of as a two stage process. In the first stage, the total allocation to emerging markets is determined, and in the second stage, the allocation to each individual market is made. This decision-making process would imply that returns on emerging markets as a whole would affect the allocation to an individual country. Buckberg’s paper finds empirical support for this hypothesis. More recently, Richards (2002) and Griffin et al. (2002) include regional equity returns in their analysis of daily portfolio flows to individual emerging markets, and find the association to be positive. These results are consistent with Hernandez et al. (2001), who show that capital flows (much like currency crises) occur in clusters and that there is contagion in private capital flows to developing countries.

Studies on portfolio flows to India include Samal (1997), Pal (1998), and Chakrabarti (2002). The first two authors find the comovement between FII flows and the Bombay Stock Exchange (BSE) index to be high. Since this could be due to both series being 1(1) processes and possibly cointegrated, little can be learned about their short-term movements from the high correlation coefficients. Chakrabarti (2002) conducts a systematic analysis of the determinants of FII investment into India. The paper finds that the contemporaneous domestic stock market return has a positive and significant coefficient in the regressions for FII flows. Chakrabarti also finds that the beta of the Indian market is negatively associated with FII inflows up until the Asian crisis, but the relationship disappears thereafter.¹²

However, Chakrabarti’s analysis has a number of limitations. First, the paper does not include either foreign interest rates or factors specific to emerging markets, even though these have been found to be important in the literature. Second, the relationship between the BSE return and FII flows is estimated using contemporaneous values. As Chakrabarti himself recognizes, this implies the possibility of endogeneity bias and limits the ability to draw inferences about causality. In any case, Griffin et al. (2002) find some evidence that lagged Indian stock market returns are negatively associated with daily FII flows, and this is consistent with our results presented below.

B. Theoretical Framework and Suggested Regressors

Following Lewis (1999), FII investment can be modeled as a portfolio allocation choice. An international investor is assumed to choose portfolio shares for foreign and Indian equities (S^F and S^I, respectively) so as to maximize a simple mean-variance objective function:

where V₁ 0 and V₂ < 0, i.e., the objective function V is increasing in the expected value of the portfolio, W, and decreasing in its variance. E_t(.) is the expectations operator conditional upon information available at time t, Var(.) is the variance operator, and

where the returns on foreign and Indian equities are denoted by r^F and r¹, respectively.

The first order condition used to solve (1) will include the following partial derivatives of with respect to S^I:

where ${\sigma }_{IF}^2$ is the covariance between Indian and foreign returns and ${\sigma }_I^2$ is the variance of Indian returns. With homogeneous investors, the equilibrium obtained by solving (1) is the Capital Asset Pricing Model (Merton, 1982).

In the case of India, this theoretical model should not be taken too literally. As noted, limits still exist on FII holdings of individual Indian stocks. More importantly, domestic investors (the major players in the Indian market) are not permitted to hold foreign equities. It is therefore not plausible to think that S^I is chosen optimally, or that the actual share of international equity investment going to India constitutes an equilibrium allocation. Thus, rather than solve for the optimal S^I, we simply use (2) and (3) to gain some intuition about the likely determinants of FII flows.

Inspection of the r.h.s. of (2) reveals that increasing S^Ican raise the expected value of the portfolio if ${{E}}_{{t}}{{r}}_{{t}+1}^{{I}}>{{E}}_{{t}}{{r}}_{{t}+1}^{{F}}$, i.e., the Indian market promises higher returns. Moreover, FII inflows should be increasing in the expected return on the domestic market. They should also be negatively related to the expected foreign return, although this relationship may not be so straightforward. The mean-variance framework is a one-period model, and in a richer (multi- period) model, a positive correlation between FII inflows and foreign returns might be expected as a result of a wealth effect. This could occur, for example, through U.S. investors, made wealthier by rising U.S. stocks, buying more non-U.S. assets so as to rebalance their portfolios.¹³ As noted above, recent studies using daily data find just such a positive association between portfolio flows and U.S. returns. Thus, a priori, we regard the effect of the return on foreign (developed country) stocks on FII flows to India as ambiguous.

Even if the condition ${{E}}_{{t}}{{r}}_{{t}+1}^{{I}}>{{E}}_{{t}}{{r}}_{{t}+1}^{{F}}$ does not hold in (2), increasing S^I may still be desirable if it reduces the overall variance of the portfolio. Equation (3) shows that this is most likely to be the case—i.e., the r.h.s. of (3) is most likely to be negative—when either the variance of the Indian market is small relative to that of the foreign market, or when σ_IF is small (or even negative), which implies that there are diversification gains from investing in India. Thus the variances and covariances of returns on Indian and foreign stocks are also likely to influence FII inflows.

If the mean-variance framework is extended to include a safe asset, then global interest rates become an additional determinant of FII inflows. Since fixed income investments in developed countries are an alternative to equity investment, we would in general expect global interest rates to negatively affect portfolio inflows. As noted, this is a common result in the empirical literature.

Splitting the return on foreign assets into the return on developed market assets, r^FD, and the return on emerging markets other than India, r^FE, and denoting global interest rates by R, the above discussion suggests that portfolio inflows into the Indian market, F^I, are likely to be some function:

with the expected signs as noted.¹⁴

Equation (4) provides a basis to select explanatory variables for the regressions. The difficulty with this equation, however, is that expected returns are not observable. The challenge therefore is to find variables that might plausibly condition expected returns:

Global variables: One factor that is likely to affect expected stock returns in developed countries (the source of investment for India) is industrial production, since a pickup in economic activity in these countries is likely to increase future profits. Since current returns may be a good predictor of future returns, we also include the current return on U.S. stock markets as a potential explanatory variable. Portfolio flows to India are small in global terms, so we can include contemporaneous U.S. returns in the regressions without having to worry about endogeneity. We also include various global interest rate variables in the regressions, including LIBOR.

Regional variables: To capture the likely determinants of expected returns in emerging markets other than India, we include the return on the Morgan Stanley Capital International (MSCI) emerging market index,¹⁵ a dummy for currency crises in major emerging markets, and industrial production growth in emerging markets.

Domestic variables: As factors conditioning expectations of Indian equity returns, we include Indian industrial production, exchange rate depreciation, sovereign credit ratings, and dummies to capture political events such as a war or government resigning mid-term. Since higher expected Indian returns should boost FII flows, a positive outlook for growth and profitability, a strong external position, and a stable political environment are all likely to be FII positive. We also include the liquidity on the Bombay Stock Exchange (BSE) as a possible positive influence on expected returns. This is because the more liquid is the market, the easier it will be to close positions and the lower is likely to be the bid-ask spread.

Current domestic stock returns are likely to condition expectations of future returns. However, because of potential endogeneity we include the return on BSE with a one-month lag in the regressions. This also applies to the exchange rate depreciation variable. For the sake of consistency, we also include the returns on emerging market stocks and NASDAQ and LIBOR with a lag, even though for endogeneity reasons this is not strictly necessary.

Other variables. We also run variants of the regression which include as explanatory variables the variance and covariances of returns on the BSE, the NASDAQ and the MSCI emerging market index. Finally, we use dummy variables to account for seasonal (and budget) effects and a dummy for relaxation in the rules for FII investment in India. Construction of these variables is explained in Box 2.

A. Data

There are three different sources of data for FII flows: RBI, Securities and Exchange Board of India (SEBI), and IFS. The RBI has monthly data on FII equity flows for the entire period up to October 2001. SEBI has annual data on equity and debt flows for the entire period, but daily and monthly data are available only for the past two years. From IFS, quarterly data on equity flows are available from 1993:1–2000:4. However, the data from different sources do not match, probably because of different exchange rates being used, and due to some definitional differences, such as the RBI’s data being balance of payments data.

Thus, data from alternative series cannot be combined to construct a complete time series of FII flows. To use the longest time series available, we use the RBI’s monthly data on equity flows for the period 1992:9–2001:10. However, data on some of the other variables were not available for the first few months of this period and the usable data are from 1993:3-2001. The rest of the data have been drawn from various sources including RBI and IFS (see Box 2 for details).

In the literature, researchers have used several alternative forms of the dependent variable. In addition to using absolute values (Claessens, Dooley, and Warner, 1998), portfolio flows have been expressed relative to stock market capitalization (Froot, O’Connell, and Seasholes, 1998; Richards, 2002); as a percent of GDP (Hernandez, Mellado, and Valdes, 2001); and as a lagged moving-average (Brennan and Cao, 1997). We conduct the econometric analysis for FII equity flows in millions of U.S. dollars (FIIDOL); and FII equity flows as a percentage of market capitalization on the BSE (FIICAP). The regression model fits better for the FIICAP variable—R² is higher when this dependent variable is used—but otherwise the results are broadly similar irrespective of which dependent variable is used. Table 3 summarizes the basic statistics of the dependent and independent variables.

B. Regression Framework

The basic regression framework used in the analysis is given below.

Where FII_t is the FII equity flow in month t; c is the constant term; LDV; is the lagged dependent variable with ith lag (we use a lagged dependent variable with one lag in the regressions), thus LDV_1t= FII_t-1; X_j is the jth global variable; Y_k is the kth emerging market variable; Z_m is the mth domestic factor; and ε_t is the normally distributed error term with zero mean and constant variance. T is the number of observations in the regressions.

A. Tests for Unit Roots and Autocorrelation

As a first step, the time series properties of the dependent and independent variables are analyzed by estimating the following equation for each variable:

and testing for the null hypothesis, ρ=1, against the alternative hypothesis ρ≠1. The results are presented in Table 5. Since most of the series are in percentage terms, the series are found to be I(0) and the null hypothesis of a unit root is rejected in favor of the alternative hypothesis.¹⁶

Table 5.

India: Unit Root Tests of Dependent and Independent Variables ^1/

^1/The Phillips-Perron test allows for serial correlation and heteroscedasticity in the error term. The p value is the probability with which the null hypothesis of unit root can be accepted.

Table 5.

India: Unit Root Tests of Dependent and Independent Variables ^1/

	Augmented Dickey-Fuller	P Value	Phillips-Perron	P Value
FIIDOL	−2.5	0.12	−61.4	0.00
FIICAP	−2.4	0.13	−68	0.00
BSEDY	−4.0	0.00	−110.4	0.00
BSEY	−3.9	0.00	−115.9	0.00
EXCC	−4.4	0.00	−81.2	0.00
EMFY	−5.3	0.00	−86.8	0.00
NASDAQY	−5.5	0.00	−97.1	0.00
DOWY	−5.9	0.00	−107.7	0.00
SPY	−5.2	0.00	−103.6	0.00
LIBOR	−2.6	0.09	−6.8	0.28
FFUND	−3.2	0.01	−7.1	0.27
TB	−2.7	0.07	−6.9	0.28
LIQ	−1.6	0.47	−6.4	0.31
βED	−3.1	0.02	−11.1	0.10
βIE	−5.7	0.00	−20.4	0.01
IPINDC	−2.0	0.27	−11.3	0.10
IPEMC	−4.3	0.00	−6.7	0.30
IPC	−1.7	0.44	−9.0	0.17

^1/The Phillips-Perron test allows for serial correlation and heteroscedasticity in the error term. The p value is the probability with which the null hypothesis of unit root can be accepted.

View Table

Table 5.

India: Unit Root Tests of Dependent and Independent Variables ^1/

	Augmented Dickey-Fuller	P Value	Phillips-Perron	P Value
FIIDOL	−2.5	0.12	−61.4	0.00
FIICAP	−2.4	0.13	−68	0.00
BSEDY	−4.0	0.00	−110.4	0.00
BSEY	−3.9	0.00	−115.9	0.00
EXCC	−4.4	0.00	−81.2	0.00
EMFY	−5.3	0.00	−86.8	0.00
NASDAQY	−5.5	0.00	−97.1	0.00
DOWY	−5.9	0.00	−107.7	0.00
SPY	−5.2	0.00	−103.6	0.00
LIBOR	−2.6	0.09	−6.8	0.28
FFUND	−3.2	0.01	−7.1	0.27
TB	−2.7	0.07	−6.9	0.28
LIQ	−1.6	0.47	−6.4	0.31
βED	−3.1	0.02	−11.1	0.10
βIE	−5.7	0.00	−20.4	0.01
IPINDC	−2.0	0.27	−11.3	0.10
IPEMC	−4.3	0.00	−6.7	0.30
IPC	−1.7	0.44	−9.0	0.17

^1/The Phillips-Perron test allows for serial correlation and heteroscedasticity in the error term. The p value is the probability with which the null hypothesis of unit root can be accepted.

We test for autocorrelation in the regressions using the Durbin h test (since we have a lagged dependent variable in our regressions, the Durbin Watson test is likely to give biased estimates). On the basis of the results, we cannot reject the null hypothesis of no autocorrelation. Thus the OLS estimates are unbiased and efficient. If the error terms were serially correlated, OLS estimates with the lagged dependent variable would have been biased and inefficient.

B. Correlations Between Dependent and Independent Variables

Bivariate correlation coefficients between the dependent and various independent variables are presented in Table 6. The alternative dependent variables, FIIDOL and FIICAP, are highly correlated, so we expect regression results to be similar for these variables. The only right-hand side variable with which the dependent variable is significantly correlated is the domestic stock market yield. This strong contemporaneous correlation between equity yields and FII flows could be due to the simultaneity between these variables.¹⁷

Table 6.

India: Correlation Coefficients Between Dependent and Independent Variables ^1/

^1/A coefficient larger than .37 is significantly different from zero at 10 percent.

Table 6.

India: Correlation Coefficients Between Dependent and Independent Variables ^1/

	FIIDOL	FIIDOL(-l)	FIIDOL(l)	FIICAP	FnCAP(-l)	FnCAP(l)	BSEDY	BSEY	EXCC	LIBOR	NASDAQY	EMFY	RATING	BUDG	GOVT	CONT	ASIA2
FIIDOL	1.00
FIIDOL(-l)	0.35	1.00
FIIDOL(l)	0.34	0.31	1.00
FIICAP	0.96	0.34	0.34	1.00
FIICAP(-l)	0.34	0.96	0.32	0.38	1.00
FHCAP(l)	0.34	0.32	0.96	0.37	0.35	1.00
BSBDY	0.34	−0.04	0.02	0.39	0.03	0.02	1.00
BSEY	0.33	−0.05	−0.02	0.38	0.02	−0 02	0.99	1.00
EXCC	−0.17	−0.08	−0.25	−0.18	−0.11	−0.25	−0.28	−0.13	1.00
LIBOR	−0.11	−0.07	−0.11	−0.25	−0.20	−0.26	−0.21	−0.19	0.14	1.00
NASDAQY	0.09	−0.29	0.01	0.06	−0.26	−0.03	0.39	0 39	−0.07	0.01	1.00
EMFY	0.19	−0.09	0.24	0.22	−0.04	0.26	0.51	0.49	−0.19	−0.18	0.63	1.00
RATING	−0.27	−0.25	−0.22	−0.31	−0.26	−0.23	−0.25	−0.23	0.24	0.03	−0.23	−0.22	1.00
BUDG	0.25	0.26	0.20	0.21	0.22	0.20	−0.11	−0.14	−0.17	0.05	−0.19	−0.14	−0.05	1.00
GOVT	0.01	0 01	0.01	0.02	0.01	0.01	−0.01	0.02	0.22	0.06	−0.04	0.03	0.12	0.10	1.00
CONT	−0.02	0.04	−0.02	−0.01	0.02	0.01	−0.11	−0.12	−0.08	0.0.1	−0.21	−0.31	0.12	0.16	−0.08	1.00
ASIA	−0.09	0.01	−0.12	−0.12	−0.02	−0.14	−0.16	−0.15	0.15	0.12	0.02	−0.19	0.23	−0.16	0.23	−0.08	1.00

^1/A coefficient larger than .37 is significantly different from zero at 10 percent.

View Table

Table 6.

India: Correlation Coefficients Between Dependent and Independent Variables ^1/

	FIIDOL	FIIDOL(-l)	FIIDOL(l)	FIICAP	FnCAP(-l)	FnCAP(l)	BSEDY	BSEY	EXCC	LIBOR	NASDAQY	EMFY	RATING	BUDG	GOVT	CONT	ASIA2
FIIDOL	1.00
FIIDOL(-l)	0.35	1.00
FIIDOL(l)	0.34	0.31	1.00
FIICAP	0.96	0.34	0.34	1.00
FIICAP(-l)	0.34	0.96	0.32	0.38	1.00
FHCAP(l)	0.34	0.32	0.96	0.37	0.35	1.00
BSBDY	0.34	−0.04	0.02	0.39	0.03	0.02	1.00
BSEY	0.33	−0.05	−0.02	0.38	0.02	−0 02	0.99	1.00
EXCC	−0.17	−0.08	−0.25	−0.18	−0.11	−0.25	−0.28	−0.13	1.00
LIBOR	−0.11	−0.07	−0.11	−0.25	−0.20	−0.26	−0.21	−0.19	0.14	1.00
NASDAQY	0.09	−0.29	0.01	0.06	−0.26	−0.03	0.39	0 39	−0.07	0.01	1.00
EMFY	0.19	−0.09	0.24	0.22	−0.04	0.26	0.51	0.49	−0.19	−0.18	0.63	1.00
RATING	−0.27	−0.25	−0.22	−0.31	−0.26	−0.23	−0.25	−0.23	0.24	0.03	−0.23	−0.22	1.00
BUDG	0.25	0.26	0.20	0.21	0.22	0.20	−0.11	−0.14	−0.17	0.05	−0.19	−0.14	−0.05	1.00
GOVT	0.01	0 01	0.01	0.02	0.01	0.01	−0.01	0.02	0.22	0.06	−0.04	0.03	0.12	0.10	1.00
CONT	−0.02	0.04	−0.02	−0.01	0.02	0.01	−0.11	−0.12	−0.08	0.0.1	−0.21	−0.31	0.12	0.16	−0.08	1.00
ASIA	−0.09	0.01	−0.12	−0.12	−0.02	−0.14	−0.16	−0.15	0.15	0.12	0.02	−0.19	0.23	−0.16	0.23	−0.08	1.00

^1/A coefficient larger than .37 is significantly different from zero at 10 percent.

Some of the independent variables are significantly correlated with each other. For example, stock market yields in domestic, emerging, and developed markets; exchange rate depreciation and domestic political events; and rating downgrades and domestic political events. This may give rise to multicollinearity. We address this issue by dropping the variables that are susceptible to multicollinearity and analyzing the effects on the coefficients of the remaining variables, and on R². Encouragingly, regression results for respective variables (not shown) do not differ much across these different specifications, indicating that multicollinearity is probably not a serious problem in our regressions.

C. Results from Multivariate Regressions

The multivariate regressions for FIICAP—FII flows as a percent of market capitalization— are reported in Table 7. The first variant includes a whole host of variables—a “kitchen sink” regression. Thereafter, variables with the least significant coefficients were dropped sequentially so as to arrive at the regression labeled “parsimonious.” Results for the “kitchen sink” and “parsimonious” specifications are similar, and show that a combination of global, regional, and domestic factors are important in the regressions.¹⁸ The third column shows the results when the seasonal dummy, Timel, is replaced by the dummy for budget months.

Table 7.

India: Regression Results

*, ** and *** indicate significance at 10 percent, 5 percent and 1 percent levels, respectively.

Table 7.

India: Regression Results

	“Kitchen sink” OLS		“Parsimonious” OLS		“Parsimonious” OLS with BUDG
	Coeff	t-ratio	Coeff	t-ratio	Coeff	t-ratio
C	0.159***	3.09	0.160***	3.17	0.158***	3.09
FIICAP(-l)	0.302***	3.31	0.312***	3.56	0.303***	3.34
BSEY(-1)	−0.004***	−3.05	−0.004***	−3.29	−0.004***	−2.93
EXCC(-l)	−0.01	−1.36	−0.009	−1.41	−0.012	−1.76
RATING(-l)	−0.071**	−2.47	−0.072***	−2.83	−0.070***	−2.69
POLT	0.003	0.10
BUDG	0.011	0.40			0.042**	2.08
EMFY(-l)	0.005**	2.57	0.005***	2.91	0.005***	3.04
CONT	−0.012	−0.36
ASIA	−0.009	−0.26
NASDAQY(-1)	−0.001	−0.86	−0.001	−0.98	−0.001	−0.91
LIBOR(-l)	−0.018*	−1.87	−0.018**	−1.95	−0.016*	−1.76
TIME1	0.048*	1.86	0.053***	2.77
R2;	.44		.43		.41
AdjR2	.36		.38		.36

*, ** and *** indicate significance at 10 percent, 5 percent and 1 percent levels, respectively.

View Table

Table 7.

India: Regression Results

	“Kitchen sink” OLS		“Parsimonious” OLS		“Parsimonious” OLS with BUDG
	Coeff	t-ratio	Coeff	t-ratio	Coeff	t-ratio
C	0.159***	3.09	0.160***	3.17	0.158***	3.09
FIICAP(-l)	0.302***	3.31	0.312***	3.56	0.303***	3.34
BSEY(-1)	−0.004***	−3.05	−0.004***	−3.29	−0.004***	−2.93
EXCC(-l)	−0.01	−1.36	−0.009	−1.41	−0.012	−1.76
RATING(-l)	−0.071**	−2.47	−0.072***	−2.83	−0.070***	−2.69
POLT	0.003	0.10
BUDG	0.011	0.40			0.042**	2.08
EMFY(-l)	0.005**	2.57	0.005***	2.91	0.005***	3.04
CONT	−0.012	−0.36
ASIA	−0.009	−0.26
NASDAQY(-1)	−0.001	−0.86	−0.001	−0.98	−0.001	−0.91
LIBOR(-l)	−0.018*	−1.87	−0.018**	−1.95	−0.016*	−1.76
TIME1	0.048*	1.86	0.053***	2.77
R2;	.44		.43		.41
AdjR2	.36		.38		.36

*, ** and *** indicate significance at 10 percent, 5 percent and 1 percent levels, respectively.

Coefficients for the following variables are found to be significant (signs in parentheses): the lagged dependent variable (positive), lagged domestic stock market return (negative), lagged exchange rate depreciation (negative), lagged rating downgrades (negative), beginning of the year effect (positive), lagged or contemporaneous emerging market yield (positive), and LIBOR (negative).

The following variables were not found to be significant in the regressions: liquidity in the domestic stock market; industrial production growth in India, in emerging markets, or in developed countries; and the dummy variable for the relaxation of rules governing FII investment in India. In order to maximize the degrees of freedom, these variables were not included in the regressions reported in the Tables.

Global variables: Global interest rates have a negative and significant effect on FII flows, consistent with the findings in the literature. The coefficient for LIBOR, in Table 7, is −0.018 (the coefficient ranges between −0.016—0.028 in different specifications), which implies that a 1 percentage point increase in LIBOR has been associated with a slowdown in FII flows of about 20 percent, as compared to its monthly average (0.093). The lagged return on NASDAQ also has a negative coefficient in the regressions but it is not significant.

Emerging market variables: The lagged return in emerging markets (EMFY) has a positive coefficient, which is significant at the 1 percent level. The coefficient is about 0.005, which implies that a 1 percentage point increase in emerging market stocks return has been associated with an increase in FII flows to India of the order of 5 percent of the monthly average. This result is obtained with lagged as well as contemporaneous returns, but is stronger with lagged returns. Since flows to emerging markets and emerging market returns would be positively correlated, the result indicates that when emerging markets as a group receives more flows, India is also likely to benefit (consistent with Griffin et al. (2002)). This result is consistent with a two stage allocation process as suggested by Buckberg (1996).

The positive association with EMFY also suggests that Indian markets may be susceptible to contagion from a general decline in returns in emerging stock markets, or a reversal of capital flows from emerging countries. To specifically look at the effect of contagion from currency crisis in other emerging markets, we included a dummy for crises in emerging markets. We also included separate dummies for crises in emerging markets other than Asia, and for crisis in Asia. The coefficients of these variables are found to be negative but insignificant.

Domestic variables: The lagged domestic stock market return (BSEY) is found to be negatively associated with FII flows, with the coefficient statistically significant at a one percent level. This finding is somewhat surprising, since as noted above, theory would suggest a positive relationship between portfolio flows and (expected) domestic returns. The coefficient is about −0.004, which implies that a 1 percentage point increase in the domestic stock market yield has been associated with a decrease in FII flows of about 4 percent of its monthly average. Thus, rather than evidence of positive feedback trading, we find that FII investors are bargain hunters (i.e., “buying on the dips”).¹⁹ Another interpretation is that global investors allocate a fixed share of their portfolio to India and this results in FIIs selling after the market rises and buying after the market falls.

As noted earlier, this finding is consistent with Griffin et al. (2002) who find the effect of lagged stock return on FII investment to be negative and somewhat significant for India, unlike in many other emerging markets. Nonetheless, in order to test for the robustness of this result, we estimate a VAR model using daily data for FII flows, BSE returns and forward exchange rates. Results (not reported here), also indicate the coefficient on the lagged stock market return in the FII flows equations to be negative. Impulse response functions suggest that a one unit standard deviation shock to stock returns results in lower FII in the period ahead.

Other domestic variables are also found to be significant. Depreciation of the Indian rupee in the previous month is found to adversely affect FII flows in many specifications. A depreciating exchange rate would imply a weaker external position and raise concerns about erosion of the dollar value of returns on the Indian market. An alternative measure of external weakness would be a combination of exchange rate depreciation and reserve loss, however, such a variable was found to be insignificant.

A significant negative relationship is also found between FII flows and credit rating downgrades.²⁰ The coefficient on the credit rating variable is −0.07, implying that a rating downgrade has been associated on average with almost a 70 percent slowdown in FII flows to India, as compared to its monthly average. However, this result may exaggerate the importance of rating downgrades. An alternative interpretation is that since rating downgrades are usually announced following adverse economic or political conditions, it is these domestic factors (including progress on structural reforms) that are the primary determinants. Even so, the coefficients on the domestic political event dummies—war and government resigning mid term—are negative, but insignificant.

The finding that Indian macro variables—such as local market returns and rating downgrades—are significantly associated with portfolio inflows can be interpreted as implying that FIIs allocate funds to India at least partly on a top-down basis. Nonetheless, this result is not entirely conclusive, since it would also be consistent with a world where decision-making was entirely bottom-up, but returns on individual stocks were correlated with macroeconomic variables.

Finally, FII flows are found to be statistically stronger during the first four months of the year. As noted earlier, the significance of the coefficient on a beginning-of-the year dummy could reflect global factors. In the Indian context, it could also be a reflection of reform announcements that are often made during or ahead of the budget. To dig deeper into this issue, separate dummies for the first four months of the year were included in the regressions. The coefficient for February was found to be positive and significant, while the coefficients for January, March and April, are positive but insignificant. Further, if the Timel variable is replaced with a dummy variable that takes a value one for the budget month and a month ahead of it, and zero otherwise, the coefficient is positive and significant. These findings are not definitive, but they do suggest that at least part of the beginning-of-the-year effect is India-specific.²¹

D. Quantifying the Importance of Domestic Versus External Factors

Significance of global, emerging market and domestic factors: FII flows are also regressed separately on domestic, regional, and global factors (Table 8). We find that domestic variables, regional variables, and global variables each exert significant influences on equity flows to India. This lends support to the hypothesis that portfolio investment in India is not independent of events in other emerging markets.

Table 8.

India: OLS Regression Results for Different Sets of Variables

*, ** and *** indicate significance at 10 percent, 5 percent and 1 percent levels respectively.

Table 8.

India: OLS Regression Results for Different Sets of Variables

	Full Sample Domestic Variables		Full Sample Regional Variables		Full Sample Global Variables
	Coeff	t-ratio	Coeff	t-ratio	Coeff	t-ratio
C	0.064***	4.18	0.49***	3.66	0.16***	2.91
FIICAP(-l)	0.337***	3.68	.32***	3.71	.31***	3.64
BSEY(-l)	−0.003**	−2.04
EXCC(-l)	−0.014**	−2.01
RATING(-l)	−0.077***	−2.86
EMFY(-l)			.003**	2.01
NASDAQY(-1)					−0.0003	.24
LIBOR(-1)					−0.021**	−2.17
TIME1	0.049**	2.43	0.049**	2.23	0.053**	2.53
R2;	.31		.26		.26
AdjR2	.27		.23		.23

*, ** and *** indicate significance at 10 percent, 5 percent and 1 percent levels respectively.

View Table

Table 8.

India: OLS Regression Results for Different Sets of Variables

	Full Sample Domestic Variables		Full Sample Regional Variables		Full Sample Global Variables
	Coeff	t-ratio	Coeff	t-ratio	Coeff	t-ratio
C	0.064***	4.18	0.49***	3.66	0.16***	2.91
FIICAP(-l)	0.337***	3.68	.32***	3.71	.31***	3.64
BSEY(-l)	−0.003**	−2.04
EXCC(-l)	−0.014**	−2.01
RATING(-l)	−0.077***	−2.86
EMFY(-l)			.003**	2.01
NASDAQY(-1)					−0.0003	.24
LIBOR(-1)					−0.021**	−2.17
TIME1	0.049**	2.43	0.049**	2.23	0.053**	2.53
R2;	.31		.26		.26
AdjR2	.27		.23		.23

*, ** and *** indicate significance at 10 percent, 5 percent and 1 percent levels respectively.

To further test for the significance of global, emerging market, and domestic factors, we conducted F tests for the joint significance of each set of variables in the regressions. In terms of equation (5), the null hypotheses are: all the λs are jointly zero; or all the γs are jointly equal to zero; or all the δs are equal to zero. The F tests overwhelmingly reject these null hypotheses.

In order to gauge the relative importance of domestic and external factors in determining FII flows to India, we compare the standardized coefficients of different sets of variables. The standardized coefficients correspond to those that would be obtained by estimating the regression model for normalized variables (with mean zero and standard deviation equal to one). The coefficients of different variables thus obtained indicate how much change in the dependent variable in standard deviation terms is induced by a one standard deviation change in the independent variable. An easy way to calculate these normalized coefficients is to multiply the original coefficients by the standard deviation of the respective independent variables, and divide by the standard deviation of the dependent variable.²²

Using this methodology, and adding up the absolute values of standardized coefficients of domestic and external variables, we find that the two sets of variables are about equally important in determining FII flows to India. Individually, the lagged domestic stock market return exerts the greatest influence on FII flows, followed by emerging market returns, credit rating downgrades, seasonal/budget effect, and LIBOR.

E. Testing for Importance of Variances and Covariances of Returns

In order to test for the relevance of the mean variance framework for India, we include the variances and covariances of stock market returns in our regressions. The coefficients on all variance variables are found to be insignificant (Table 9). The “betas”, calculated as the covariance between the returns in the concerned markets divided by the variance of the returns (see footnote 9), between emerging market and developed country stock returns, between Indian and emerging market stock returns, and between Indian and developed country stock returns, all have negative but insignificant coefficients.

Table 9.

India: Regression Results When Variances and Covariances Are Included

*, ** and *** indicate significance at 10 percent, 5 percent and percent levels, respectively.

Table 9.

India: Regression Results When Variances and Covariances Are Included

	I Coeff (t ratio)	II Coeff (t ratio)	III Coeff (t ratio)	IV Coeff (t ratio)	V Coeff (t ratio)	VI Coeff (t ratio)
C	0.201** 2.49	0.161*** 3.12	0.164*** 3.23	0.165*** 3.05	0.147** 2.1	0.159*** 2.83
FIICAP(-l)	0.301*** 3.38	0.315*** 3.47	0.301*** 3.41	0.334*** 3.59	0.349*** 3.80	0.346*** 3.81
BSEY(-l)	−0.004*** −3.18	−0.004*** −3.23	−0.004*** −3.27	−0.005*** −3.37	−0.005*** −3.35	−0.005*** −3.35
EXCC(-l)	−0.010 −1.43	−0.009 −1.4	−0.01 −1.45	−0.007 −1.02	−0.007 −1.01	−0.007 −1.00
RATING(-l)	−0.076*** −2.89	−0.072*** −2.75	−0.075*** −2.92	−0.040 −1.41	−0.035 −1.25	−0.036 −1.22
EMFY(-l)	0.005*** 2.95	0.005*** 2.90	0.005*** 2.96	0.005*** 2.82	0.005*** 2.76	0.005*** 2.76
NASDAQY(-l)	−0.001 −1.03	−0.001 −0.99	−0.001 −1.00	−0.001 −0.80	−0.001 −0.84	−0.001 −0.84
LIBOR(-l)	−0.023* −1.87	−0.018* −1.94	−0.016* −1.69	−0.017* −1.75	−0.017 −1.31	−0.019** −2.01
TIME1	0.052*** 2.72	0.053*** 2.75	0.053*** 2.80	0.055*** 2.74	0.054*** 2.69	0.053*** 2.68
VBSE	0.000 −0.65
VNAS		0.000 −0.15
VEMF			0.000 −0.91
βED				0.027 −0.61
βID					−0.006 −0.23
βIE						−0.003 −0.06
R2;	.44	.43	.44	.40	.39	.39
AdjR2	.38	.38	.38	.34	.33	.33

*, ** and *** indicate significance at 10 percent, 5 percent and percent levels, respectively.

View Table

Table 9.

India: Regression Results When Variances and Covariances Are Included

	I Coeff (t ratio)	II Coeff (t ratio)	III Coeff (t ratio)	IV Coeff (t ratio)	V Coeff (t ratio)	VI Coeff (t ratio)
C	0.201** 2.49	0.161*** 3.12	0.164*** 3.23	0.165*** 3.05	0.147** 2.1	0.159*** 2.83
FIICAP(-l)	0.301*** 3.38	0.315*** 3.47	0.301*** 3.41	0.334*** 3.59	0.349*** 3.80	0.346*** 3.81
BSEY(-l)	−0.004*** −3.18	−0.004*** −3.23	−0.004*** −3.27	−0.005*** −3.37	−0.005*** −3.35	−0.005*** −3.35
EXCC(-l)	−0.010 −1.43	−0.009 −1.4	−0.01 −1.45	−0.007 −1.02	−0.007 −1.01	−0.007 −1.00
RATING(-l)	−0.076*** −2.89	−0.072*** −2.75	−0.075*** −2.92	−0.040 −1.41	−0.035 −1.25	−0.036 −1.22
EMFY(-l)	0.005*** 2.95	0.005*** 2.90	0.005*** 2.96	0.005*** 2.82	0.005*** 2.76	0.005*** 2.76
NASDAQY(-l)	−0.001 −1.03	−0.001 −0.99	−0.001 −1.00	−0.001 −0.80	−0.001 −0.84	−0.001 −0.84
LIBOR(-l)	−0.023* −1.87	−0.018* −1.94	−0.016* −1.69	−0.017* −1.75	−0.017 −1.31	−0.019** −2.01
TIME1	0.052*** 2.72	0.053*** 2.75	0.053*** 2.80	0.055*** 2.74	0.054*** 2.69	0.053*** 2.68
VBSE	0.000 −0.65
VNAS		0.000 −0.15
VEMF			0.000 −0.91
βED				0.027 −0.61
βID					−0.006 −0.23
βIE						−0.003 −0.06
R2;	.44	.43	.44	.40	.39	.39
AdjR2	.38	.38	.38	.34	.33	.33

*, ** and *** indicate significance at 10 percent, 5 percent and percent levels, respectively.

F. Robustness Analysis

Alternative series for explanatory variables: For some of the explanatory variables, more than one series were available (BSE or the NSE for returns on domestic shares; Dow Jones Index, NASDAQ or Standard and Poor’s index for stock market returns in the source country; LIBOR, federal funds rate, or treasury bill rates for the external interest rate). The alternative series were highly correlated, so it seemed legitimate to use any one of them. Nonetheless, the robustness of results was checked using alternative series.²³

Lagged dependent variable: We checked the robustness of results to using different lag lengths of the dependent variable. If we include higher lags of the dependent variable in the regressions, the coefficient for the first lag becomes smaller and less significant, and coefficients for the higher lags are positive and insignificant. Results on other explanatory variables remain the same. An alternative specification is one in which a lagged dependent variable is not included. For such a specification, the Durbin-Watson test rejects the null hypothesis of no autocorrelation. We correct for it through the Cochrane-Orcutt two-step method (see Judge et al. (1985)). The results on individual variables mostly remain unchanged, as compared with the results of our benchmark case (except for exchange rate depreciation), as shown in Table 10. We also estimate the model using FIIDOL as the dependent variable. The results are broadly similar to the ones with FIICAP variable, though the R² for regressions using FIIDOL is somewhat smaller.

Table 10.

India: Regression Results for ARI and FIIDOL

*, ** and *** indicate significance at 10 percent, 5 percent and 1 percent levels, respectively.

Table 10.

India: Regression Results for ARI and FIIDOL

	ARI Regression		Regression for FIIDOL
	Coeff	t-ratio	Coeff	t-ratio
C	0.241***	3.32	139.12***	1.81
C	0.241***	3.32	139.12***	1.81
FIIDOL(-1)			0.30***	3.43
BSEY(-1)	−0.005***	−3.76	−5.80***	−2.77
EXCC(-1)	−0.005	−0.74	−15.69**	−2.56
RATING(-1)	−0.083***	−2.75	−96.03**	−2.36
EMFY(-1)	0.005***	2.97	6.83**	2.57
NASDAQY(-1)	−0.002	−1.42	−1.54	−0.69
LIBOR(-1)	−0.028**	−2.08	−8.81	−0.62
TIME1	0.076***	3.30	83.89***	2.83
RHO	0.299***	2.68
R2;	.41		.37
AdjR2	.36		.31

*, ** and *** indicate significance at 10 percent, 5 percent and 1 percent levels, respectively.

View Table

Table 10.

India: Regression Results for ARI and FIIDOL

	ARI Regression		Regression for FIIDOL
	Coeff	t-ratio	Coeff	t-ratio
C	0.241***	3.32	139.12***	1.81
C	0.241***	3.32	139.12***	1.81
FIIDOL(-1)			0.30***	3.43
BSEY(-1)	−0.005***	−3.76	−5.80***	−2.77
EXCC(-1)	−0.005	−0.74	−15.69**	−2.56
RATING(-1)	−0.083***	−2.75	−96.03**	−2.36
EMFY(-1)	0.005***	2.97	6.83**	2.57
NASDAQY(-1)	−0.002	−1.42	−1.54	−0.69
LIBOR(-1)	−0.028**	−2.08	−8.81	−0.62
TIME1	0.076***	3.30	83.89***	2.83
RHO	0.299***	2.68
R2;	.41		.37
AdjR2	.36		.31

*, ** and *** indicate significance at 10 percent, 5 percent and 1 percent levels, respectively.