CHAPTER IV EMPIRICAL RESULTS AND DISCUSSION

Transcription

1 CHAPTER IV EMPIRICAL RESULTS AND DISCUSSION 4.0 Introduction The present chapter is divided into eight sections. Section briefly discusses the trend and composition of Foreign Direct Investment (FDI) in India. Section discusses the empirical analyses including descriptive statistics, correlation coefficient and unit root test. The volatility series generated using different measures are discussed in section Section empirically examines the impact of exchange rate levels on FDI. Section examines the impact of exchange rate volatility on FDI. Section identifies the factors determining FDI in India. The results of cointegration through Autoregressive Distributed Lag (ARDL) model are discussed in section - 4.7, while the concluding remarks are given in section The present study is exclusively based on secondary data, collected from various issues of Monthly Bulletin, Reserve Bank of India (RBI) and Hand Book of Indian Economy , RBI. The study uses quarterly data collected during the period 1996: II to 2008: I. Data on FDI inflow are collected from Quarterly Balance of Payment, RBI. Gross Domestic Product (GDP) at factor cost (at constant price) for the base year , monthly data on bank lending interest rate, wholesale price index for the base year , export and import, Bombay Stock Exchange for the base year , trade based Real Effective Exchange Rate (RER) and Nominal Effective Exchange Rate (NER) for the base year and Bilateral exchange rate between US$ and Indian Rupee (BIEX) are collected from Hand Book of Indian Economy , RBI. 79

2 For the empirical analysis GDP is considered as a proxy for market size. Bank Prime Lending Rate (LR) is considered as a proxy for cost of investment. Monthly data for whole sale price index (base year ) is collected and converted into quarterly basis for measuring inflation rate (INFL). Openness of trade (OPEN) is measured by adding export and import and divided by GDP, as a proxy for trade barriers. Wealth effect (SM) is measured using Bombay Stock Exchange index (base year =100). Lag of FDI are used to represent the agglomeration effect in the model. 4.1 Trend and Composition of FDI One of the main objectives of the liberalization process initiated in the early 1990 s was to attract foreign capital in the form of FDI. It was believed that FDI would help to increase the nation s foreign exchange reserve, improve the balance of payments position, open up foreign markets for India, increase productivity and competitiveness of Indian companies. Due to liberalisation regime and growing foreign investors confidence, the non-dept creating flows like Foreign Investment (both FDI and Foreign Portfolio Investment (FPI)) inflows in India shows a stable increasing trend compare to debt creating flows. Figure - 1 displays the share of debt and non-debt creating capital flows during the period to

3 FIGURE - 1 SHARE OF DEBT AND NON-DEBT CREATING CAPITAL FLOWS DEBT AND NON-DEBT CREATING CAPITAL FLOW (in Per cent) Non debt creating debt creating YEAR Figure - 1 shows that in , the non-debt creating flows was about 1.49 per cent, which increased to the peak of per cent in , but later slipped to per cent in The figure indicates that there was a shift in the components of capital flows from debt to non-debt creating flows during the liberalization regime. The motivation of foreign investor s interest to invest in India can be viewed from the flow of FDI approvals and actuals during the period 1991 to 2008, which is shown in figure - 2. It is shows that the shares of actual flows to FDI approvals ranges between 17 to 70 per cent. However, in recent years the gap between actual and approval flows are closing up, as the approvals take some time to fructify into actual flows. 81

4 FIGURE - 2 FDI APPROVALS AND ACTUAL FLOWS IN INDIA FOREIGN INVESTMENT (RS. IN CRS) 60,000 50,000 40,000 30,000 20,000 10,000 0 Approval Actual YEAR Liberalization of FDI policy has affected the magnitude and pattern of Foreign Investment (FI) inflows during the period , which is illustrated by figure - 3. FDI, FPI & FI (in US $ Million) 70,000 60,000 50,000 40,000 30,000 20,000 10, ,000 FIGURE - 3 FOREIGN INVESTMENT INFLOW IN INDIA FDI FPI FI YEAR The figure shows that Foreign Portfolio Investment (FPI) exceeded FDI flows in the early years of liberalisation (i.e., ), but from 1997 onwards the inflows shifted in favour of FDI. Thus, FDI inflows have increased in importance and are becoming the single most important component of total Foreign Investment inflows in India. Table 4.1 presents the components of FDI in India during to

5 TABLE COMPONENTS OF FOREIGN DIRECT INVESTMENT IN INDIA (US$ Million) Sl. No. Components Equity (100.00) (100.00) (59.57) (66.80) (54.90) (51.57) (62.44) (66.68) (74.65) (78.19) (79.58) a. Government (SIA/FIPB) (51.16) (58.26) (36.14) (36.23) (18.25) (21.47) (17.55) (12.57) (9.76) (6.69) (13.36) b. RBI (0.00) (7.88) (11.27) (12.51) (14.68) (12.36) (20.79) (24.92) (32.39) (49.85) (51.16) c. NRI (48.84) (33.35) (1.66) (0.57) d. Acquisition of shares (0.51) (8.98) (14.37) (18.19) (17.01) (15.37) (24.34) (28.43) (14.98) (13.17) e. Equity capital of unincorporated Bodies (1.51) (3.12) (3.77) (0.74) (8.73) (4.85) (4.06) (6.67) (1.89) 2. Re-invested Earnings (33.51) (26.84) (36.41) (33.78) (31.47) (30.80) (23.06) (20.86) (18.27) 3. Other Capital# (6.92) (6.36) (8.70) (14.65) (6.10) (2.52) (2.29) (0.95) (2.15) Direct Investment (1+2+3) (100) (100) (100) (100) (100) (100) (100) (100) (100) (100) (100.00) Source: Hand Book of Indian Economy 2008, RBI. Note: Parentheses show percentages; # - Data pertain to inter-company debt transactions of FDI entities; SIA/FIPB means Secretariat of Industrial Authorities/Foreign Investment Promotion Board; and NRI Non Resident Indians. 83

6 One of the interesting phenomena observed from the table is that, while the share of FDI flows from government route is declined from per cent in to per cent in , whereas, the share of automatic route (RBI route) increased from 7.88 per cent in to per cent in This shows that the largest share of total FDI was mainly through SIA/FIPB compare to RBI routes. This reflects the interest of Government of India in encouraging FDI through the automatic route. It is also observed from the table that the share of reinvested earnings contributing around 30 per cent of total FDI and maintains its level upto and then declined it to per cent in Thus the major share of FDI comes from equity and re-invested earnings. Table shows the trend and pattern of country-wise FDI inflows in India during the years 1991 to

7 TABLE COUNTRY-WISE FDI INFLOWS IN INDIA (Rupees in Crores) Countries UK (8.33) (3.19) (1.51) (13.51) (10.74) (3.62) (7.77) (19.44) (2.61) (3.04) Germany (3.68) (5.92) (2.48) (6.21) (4.72) (6.16) (1.34) (1.25) (2.50) (2.69) Netherlands (3.42) (3.98) (2.28) (5.67) (13.47) (8.45) (1.49) (6.01) (3.09) (3.00) France (2.53) (4.87) (2.95) (3.20) (2.33) (1.90) (0.36) (1.07) (0.70) (1.93) Switzerland (2.14) (0.42) (0.20) (2.35) (0.34) (2.76) (2.02) (0.61) (0.99) (0.59) EUROPE (20.10) (18.38) (9.40) (30.94) (31.60) (22.89) (12.98) (28.38) (9.90) (11.26) US (12.49) (16.75) (12.18) (16.16) (20.31) (20.22) (10.30) (7.59) (4.89) (5.45) Japan (5.02) (8.17) (4.79) (3.98) (4.58) (5.26) (2.56) (0.86) (2.35) (1.17) TRIAD (37.60) (43.30) (26.37) (51.09) (56.50) (48.36) (25.84) (36.82) (17.14) (17.87) Mauritius (13.15) (44.14) (62.35) (32.21) (26.06) (35.34) (40.58) (40.61) (48.99) (44.79) Singapore (2.23) (1.15) (1.81) (2.11) (1.03) (2.76) (4.94) (6.25) (14.55) (14.80) South Korea (0.76) (1.26) (0.10) (0.90) (1.50) (0.60) (1.82) (0.73) (0.55) (0.68) Others (46.25) (10.16) (9.37) (13.69) (14.91) (12.93) (26.82) (15.58) (14.31) (11.44) Total FDI (100.00) (100.00) (100.00) (100.00) (100.00) (100.00) (100.00) (100.00) (100.00) (100.00) Source: Government of India, Department of Scientific and Industrial Research, Ministry of Scientific and Technology, Foreign Collaborations; various RBI Annual Reports; and Economic Survey, various issues. Note: Parantheses show percentages. 85

8 It is apparent that FDI from Mauritius holds a major share since and retains it at per cent in The traditional sources of foreign investment are from the Triad countries, viz., Europe, United States of America (USA) and Japan contributing per cent of the foreign investment inflows during , which further increased to per cent in However, FDI flows from the Triad countries declined thereafter to per cent in The importance of US has fluctuated wildly, but on the whole has been rising especially since During , the relative share of foreign investment inflows from Japan was 5.05 per cent, which steadily declined to 1.17 per cent in Thus, while liberalisation seems to have attracted American enterprises, the Japanese multinational enterprises (MNEs) have not shown any discernible response to lndia's FDI and trade liberalisation. A number of new sources of FDI, such as South Korea, Singapore and other countries, have emerged significant in recent years. These countries together contributed 25 per cent of all FDI approvals in Sector-wise break-up of FDI during the year is shown in table

9 TABLE SECTOR-WISE FOREIGN INVESTMENT IN INIDA (Rs. in Crores) Sector-wise FDI Manufacturing (36.07) (7.96) (12.62) (28.95) (29.14) (39.83) (42.01) (17.63) (19.18) (21.05) Food and Dairy Products (10.00) (5.99) (3.93) (2.35) (4.38) (7.89) (4.41) (1.05) (0.76) (0.58) Electricity (11.79) (9.16) (11.15) (2.90) (6.16) (0.60) (2.08) (1.87) (4.27) (2.95) Construction (14.29) (11.76) (9.01) (5.69) (10.39) (13.13) (9.86) Trade, Hotels, Restaurants (2.35) (4.58) (0.95) (1.99) (1.64) (1.44) (1.51) Transport (20.00) (4.79) (5.03) (0.72) (1.37) (3.02) (0.18) (1.77) (4.20) (1.77) Financing, Insurance, Realestate and Business Services (1.43) (19.03) (2.09) (13.45) (14.09) (15.65) (10.48) (18.92) (26.69) (27.83) Computer Services (2.86) (3.66) (16.02) (17.91) (11.35) (16.03) (22.92) (8.85) (5.33) (7.26) Educational, R&D Services (0.08) (0.08) (0.06) (0.07) (0.30) (0.30) (0.46) (0.80) (1.07) Total FDI (100.00) (100.00) (100.00) (100.00) (100.00) (100.00) (100.00) (100.00) (100.00) (100.00) Source: Foreign Collaborations, Department of Scientific and Industrial Research, Ministry of Scientific and Technology, Government of India; various RBI Annual Reports; and Economic Survey, various issues. Note: Parentheses show percentages. 87

10 The table provides the list of items which India favours for FDI inflows under its reform policies. The top sectors which received the largest amount of FDI inflow during include manufacturing (36.07%), food and dairy products (10.00%), electricity (11.79 %) and transport (20.00 %). However, during , the trend changed in favour of finance, insurance, real estate and business services (13.45 %), computer services (17.91 %), and manufacturing (28.95 %). Electricity, transport, finance, food and dairy products, which were important sectors attracting FDI in the early nineties, recorded a down trend in the latter half of the nineties. Services, like hotel and restaurants, real estate, and computer services also registered an increasing trend in the latter half of the nineties. The inflow of FDI into computers increased from 6.0 per cent in to per cent in , which declined to 7.26 per cent in In sum, there have been significant changes in the pattern and composition of FDI inflows, with few clear trends over the decade as whole. Table presents the key macroeconomic indicators of India during the period to The table reveals that the ratios of export and import to GDP have grown steadily from 5.8 and 8.8 per cent in to 15.1 and 25.5 per cent in respectively. Similarly, the ratios of foreign investment to GDP, export, import and foreign exchange reserve have grown favourably from 0.1, 0.8, 0.7 and 1.5 per cent in to 5.2, 38.8, 25.63, per cent in respectively. The share of foreign investment to Gross Capital Formation (GCF) remained stable at about four per cent during to , after which it rose to per cent in Due to Asian Crisis shocks in , all the ratios in general declined during the period, but later recorded an increasing trend. 88

11 TABLE KEY MACRO ECONOMIC INDICATORS Year Export /GDP Import /GDP FI /GDP FI /Export FI /Import FI /FER FI /GCF Source: Hand Book of Indian Economy , RBI, India. Note: FI Foreign Investment; FER Foreign Exchange Reserve; GDP Gross Domestic Product; and GCF Gross Capital Formation. Figure - 4 display the relationship between log of FDI and real effective exchange rate in India during 1996Q2 to 2008Q1. The figure depicts downward movements in the real exchange rate appear to be followed by subsequent increase in FDI inflows during the period. 89

12 FIGURE - 4 MOVEMENTS OF FOREIGN DIRECT INVESTMENT AND REAL EFFECTIVE EXCHANGE RATE IN INDIA LOG OF FOREIGN DIRECT INVESTMENT LFDI LRER LOG OF REAL EFFECTIVE EXCHANGE RATE YEAR 4.2 Empirical Results This section discusses the descriptive statistics of variables included in the FDI function, their correlation coefficient and unit root test results included in the analysis of FDI function. Table shows descriptive statistics of all the variables. It is observed from the table that the series of FDI, SM and BIEX are not normally distributed. This is confirmed by J-B statistics, which rejects the null hypothesis of normal distribution at one per cent level of significance. 90

13 TABLE DESCRIPTIVE STATISTICS Sl. No. Variables μ Max. Min. σ m 3 m 4 J-B 1. FDI * 2. GDP OPEN SM * 5. INFL LR RER NER BIEX * Note: μ mean; Max. maximum; Min. minimum σ - standard deviation; m 3 skewness; m 4 - kurtosis; J-B - Jaqua-Bera test for normality respectively; and * indicates one per cent significance level. Pearson correlation coefficient of the variables included in the FDI functions are presented in table Looking at the analysed results it is apparent that there is no any serious multicollinearity problem in the series. 91

14 TABLE CORELATION COEFFICIENT Variables FDI GDP OPEN SM INFL LR RER NER BIEX FDI 1 GDP OPEN SM INFL LR RER NER BIEX

15 4.2.1 Unit root test For the preliminary analysis, the presence of unit roots for all the variables in the mean equation are tested by applying Augmented Dickey Fuller (ADF) and Phillips- Perron (PP) tests and are presented in table TABLE UNIT ROOT TESTS Sl. No. Series ADF PP Levels 1 st Difference Levels 1 st Difference 1. FDI C * * C&T * * 2. GDP C * * C&T * * * 3. OPEN C * * C&T * * * 4. SM C * * C&T * * 5. INFL C * ** * C&T * 6.005* ** * 6. LR C * * C&T * * 7. RER C * * C&T ** * * 8. NER C * ** C&T * ** 9. BIEX C * * C&T * * Note: * and ** indicate significant at one and five per cent levels respectively; and C constant; and C&T - constant and trend, respectively. 93

16 The table reveals that the selected series are non-stationary at levels, except for INFL and RER with constant and trend. However, all selected series are stationary at their first difference, hence, the series are integrated of order one. 4.3 Results for Measures of Exchange Rate Volatility Exchange rate volatility is directly unobservable. Hence to overcome this problem, observations are generated by using conditional volatility framework and a few other methods by selecting three exchange rate series namely, trade based real effective exchange rate (RER), nominal effective exchange rate (NER) for the base year , and bilateral exchange rate (BIEX) between US$ and Indian Rupee. The study uses monthly data for the period April 1996 to January 2008, later converted into quarterly series Test for the presence of ARCH effect Having Known that series (RER, NER and BIEX) are stationary in table - 4.7, AR (1) (i.e., autoregressive with lag one) model has been run to test for the presence of autoregressive conditional hetroskedasticity (ARCH) effect in the series. Rejection of the null hypothesis of no ARCH effect indicates that the series varies over time. When fitting ARCH equations, Lagrange Multiplier (LM) and F-tests have been used to test the null hypothesis of no ARCH effect. The results of ARCH-Langrange Multiplier (LM) tests are presented in table

17 TABLE ARCH-LM TEST RESULTS Sl. No. Series F-Statistic n*r 2 1. RER (0.4796) (0.476) 2. NER (0.394) (0.390) 3. BIEX (0.001) (0.002) Note: Parentheses show probability value; and n*r 2 indicate sample size (n) multiplied by goodness of fit (R 2 ). The probability values lower than 0.05 indicate that the null hypothesis is rejected at five per cent level of significance implying that volatility varies over time. As can be seen in table - 4.8, the presence of ARCH effect is exhibited in the case of BIEX. In the case of RER and NER, no ARCH effect has detected, and hence there is no need to apply the ARCH approach. Therefore, the measure of volatility for RER and NER can be obtained using moving average standard deviation (MASD) and Hodrick and Prescott (HP) method Measure of volatility based on ARCH process series. Table presents ordinary least square (OLS) models and ARCH tests for BIEX 95

18 TABLE OLS MODEL AND ARCH TESTS A. OLS: BIEX t = BIEX t-1 + e t (0.680) (4.91)* B. ARCH test: m 3 = m 4 = J-B test = * F-stat = 10.56* Q(5) = 13.6* Q 2 (5) = 11.72* ARCH-LM (1) = 9.959* Note: * denotes significance at one per cent level. Standard errors are in parentheses; m 3 and m 4 are the sample Skewness and Kurtosis statistics based on the standardized residuals; J-B is Jaqua-Bera test for Normal distribution; and the Lagrange multiplier test for ARCH-LM is sample size (n) multlipy by goodness of fit (R 2 ). This test statistics is distributed as chi-square, with degrees of freedom equal to the number of lag terms in the model; and the statistics Q(5) and Q 2 (5) are the Ljung-Box tests based on the fifth order serial correlation in the residuals and squared normalized residuals. The table indicates that BIEX series is positively skewed (0.48) and leptokurtic (7.553). Further, the Jaqua-Bera (J-B) test rejects the null hypothesis of normal distribution for BIEX with high level of confidence (128.83), thus confirming the nonnormality of the BIEX series. To test the ARCH effect, the study uses AR model with the optimum lag one based on Akaike Information Criteria (AIC) which gives the estimated residuals. The Ljung-Box Q-statistic 1 for the residuals as well as squared residuals shows significant autocorrelation coefficients up to lag length five. Further, the ARCH-LM test shows the presence of ARCH effect, as the F-statistics is statistically significant at one per 1 The Ljung-Box (1979) Q statistic at lag k is a test statistic for the null hypothesis that there is no autocorrelation up to order k and is computed as: Q LB = T( T + 2) k 2 j j= 1 T T J where, T j is the j-th autocorrelation and T is the number of observations. Under the null hypothesis Q is asymptotically distributed as a 2 χ, with degrees of freedom equal to the number of observations. 96

19 cent level (see table - 4.9). After the confirmation of ARCH effect, ARCH (1) model has been estimated and its result is presented in table TABLE ARCH MODEL Details Co-efficient Standard Error z-statistic A. Conditional Mean Constant BIEX t B. Conditional Variance Constant 4.46E E ε 2 t C. Diagnostic Test Log Likelihood = DW = 2.12 ARCH-LM (1) = 0.007[0.93] AIC = Q(5) = 3.50 Q 2 (2) = 3.58 Note: * and ** denote one and five per cent level of significance respectively; and D.W. - Durbin and Watson statistics, The Lagrange multiplier test for ARCH-LM is sample size (n) multlipy by goodness of fit (R 2 ). This test statistics is distributed as chi-square, with degrees of freedom equal to the number of lag terms in the model; and the statistics Q(5) and Q 2 (2) are the Ljung-Box tests based on the fourth and second order serial correlation in the residuals and squared normalized residuals. Using the ARCH-LM test, the estimation of the ARCH (1) model confirms the absence of ARCH effects. Therefore, it is not necessary to proceed with the generalised ARCH (GARCH) model. However, the result confirms volatility in BIEX series, when conditional variances series are obtained from ARCH (1) model. Figure - 5 displays the levels of BIEX and volatile series of the BIEX data, where LBIEX exhibits bilateral exchange rate series in log form, and BIEXVAR exhibits the volatility series obtained from the ARCH (1) model for BIEX series. 97

20 FIGURE - 5 LAG OF BILATERAL EXCHANGE RATE AND ITS VOLATILITY SERIES LBIEX BIEXVAR 1996Q2 1997Q2 1998Q2 1999Q2 2000Q2 2001Q2 2002Q2 2003Q2 2004Q2 2005Q2 2006Q2 2007Q2 1996Q2 1997Q3 1998Q4 2000Q1 2001Q2 2002Q3 2003Q4 2005Q1 2006Q2 2007Q Measure of volatility based on MASD and HP method The RER and NER series are used to generate volatility through MASD and HP methods. In the first stage, optimum lag for MASD method is computed to estimate the volatility series, followed by computing volatility series using HP method by separating cyclical component from the long-term trend component. Further, nonnested model has been run to select optimum measure of volatility between MASD and HP methods Optimum lag selection for MASD Table shows the results of optimal lag selection for MASD through AIC by following distributed lag model based on equation (3.2). 98

21 TABLE MASD OPTIMUM LAG SELECTION BASED ON AIC THROUGH DISTRIBUTED LAG MODEL Sl. No. LAG RER NER The table shows that at lag four, the values are a minimum for RER and for NER respectively. Therefore, RER and NER at lag four can be chosen for MASD method to obtained volatile series. In the study, volatility series has also been computed using HP method. Figure - 6 shows the different volatility series generated through MASD and HP models for both RER and NER. The volatile series obtained through HP are named as RERVHP and NERVHP, whereas the volatile series obtained through MASD are named as RERVSD and NERVSD respectively. 99

22 FIGURE - 6 LAG OF RER AND NER LEVELS AND THEIR VOLATILITY Q2 1997Q2 1998Q2 1999Q2 2000Q2 2001Q2 2002Q2 2003Q2 2004Q2 2005Q2 2006Q2 2007Q2 LNERTB Q2 1997Q2 1998Q2 1999Q2 2000Q2 2001Q2 2002Q2 2003Q2 2004Q2 2005Q2 2006Q2 2007Q2 LRERTB Q2 1997Q2 1998Q2 1999Q2 2000Q2 2001Q2 2002Q2 2003Q2 2004Q2 2005Q2 2006Q2 2007Q2 RERVSD Q2 1997Q2 1998Q2 1999Q2 2000Q2 2001Q2 2002Q2 2003Q2 2004Q2 2005Q2 2006Q2 2007Q2 NERVSD Q2 1997Q2 1998Q2 1999Q2 2000Q2 2001Q2 2002Q2 2003Q2 2004Q2 2005Q2 2006Q2 2007Q2 NERVHP Q2 1997Q3 1998Q4 2000Q1 2001Q2 2002Q3 2003Q4 2005Q1 2006Q2 2007Q3 RERVHP

23 RERVHP and NERVHP show periodical up and down-swings (see Figure - 6). Whereas, the volatility measures through MASD for RERVSD and NERVSD do not seem to display increasing or decreasing behaviour like RERVHP and NERVHP, with the fluctuations largely maintained within band Identifying optimum measure of volatility through non-nested model The results of non-nested testing carried out to select the most appropriate measure of volatility between MASD and HP for RER and NER series are reported in table Omitting detailed estimates from each reduction stage, the summary model statistics for the progressively reduced sequence for Model1 to Model2 and Model3 to Model4 are also reported in the table. TABLE COMPARATIVE MODEL EVALUATION RESULTS MODEL SELECTION CRITERIA FROM NON-NESTED TESTING A. Model Statistics Model Series T k d.f. RSS Sigma AIC Model - 1 RERVSD Model - 2 RERVHP Model - 3 NERVSD Model - 4 NERVHP B. Model Reduction Test AIC Model - 1 vs. Model - 2 = Model - 3 vs. Model - 4 = Note: T shows number of observations; k number of variables; d.f. indicates degrees of freedom; RSS Residual Sum of Squares; and AIC Akaike information criteria. 101

24 The results show that the optimal measure of volatility for both RER and NER derived from MASD consistently dominate. It can be seen that AIC strongly favours the Models 1 and 4 primarily due to their parsimony and the highest degree of freedom. This is also confirmed via AIC>0, implying that Model - 1 is preferred to Model - 2 and Model - 3 is preferred to Model - 4. Overall, the study employs ARCH method for BIEX series and MASD for RER and NER series for measuring volatility. Thus, the present study would consider two different measures of volatility for three different exchange rate series. The unit root test result for these series are given in table TABLE UNIT ROOT TESTS Sl. No. Variables ADF PP Levels 1 st Difference Levels 1 st Difference 1. RERVSD C * * C&T ** * ** * 2. NERVSD C ** * * C&T ** * * 3. BIEXVAR C * * C&T * * Note: * and ** indicate significance at one and five per cent levels respectively; C constant; and C&T - constant and trend respectively. The result shows that there is a mixture of integrated to order of one and zero, (i.e., I(1) and I(0)) of underlying regressors. For instance, RERVSD are stationary at levels in both ADF and PP test. NERVSD seem to be stationary at levels in ADF test, but not in PP test. However, all the series are stationary at first difference (i.e., integrated of order one I(1)). 102

25 4.4 The Impact of Exchange Rate Levels on FDI This section presents the results of the estimated coefficients of real exchange rate through distributed lag model to examine the impact of exchange rate levels on FDI in India. The estimated results using equation (3.1) is presented in table TABLE REGRESSION RESULTS Variables Model-1 Coefficients t - value Constant FDI t *** RER t R F-stat 2.895** Durbin Watson stat AIC Note: ** and *** - indicate five and ten per cent levels significance respectively. Having confirmed that both FDI and RER are stationary at order one (table - 4.7), the estimated model captured dynamic effect of FDI and RER by including one lag based on AIC. The result reveals that the coefficient of real exchange rate level is negative and statistically insignificant. Thus, the result finds a weak evidence of negative association between real exchange rate levels and FDI in India. 4.5 The Impact of Exchange Rate Uncertainty on FDI This section examines the impact of exchange rate uncertainty on FDI in India by employing distributed lag model based on equation (3.2). The estimated results are presented in table

26 TABLE REGRESSION RESULTS Variables Model - 2 RERVSD Model - 3 NERVSD Model - 4 BIEXVAR Constant (1.447) (1.578) (1.962)** VOL t (1.771)*** (2.154)** (2.283)** VOL t (0.604) (0.068) VOL t (2.688)* (2.074)** FDI t (1.345) (1.478) (2.373)** R Adj. R F-statistic 3.54** 3.05** 3.94** Durbin Watson stat Note: ** and *** - indicate five and ten per cent levels significance respectively; and brackets show t- value. In the table, volatility measured in Model - 2 and Model - 3 are from MASD, whereas in Model - 4 it is measured using ARCH method. The results indicate that in all models, the coefficient of exchange rate volatility is negative and significant at five and ten per cent levels respectively. Thus, the result confirms the hypothesis that increases in exchange rate volatility would reduce the inflow of FDI in India. This result matches those of some previous studies, which found that volatility in exchange rate decreases FDI flows (see Itagaki (1981); Cushman (1985); Goldberg and Kolstad (1995); Sung and Lapan (2000); and Blonigen (2005)). The negative coefficient of exchange rate uncertainty confirms the prediction of the theoretical model, that a multinational enterprise (MNE) will ignore better business opportunities in the host country due to exchange rate volatility. Thus, exchange rate volatility might emerge as an important determinant for risk averse investor, when the foreign investor decides to invest in India. 104

27 4.6 Factors Determining FDI in India In addition to exchange rate levels and its volatility, the study also attempts to examine the influence of other factors, such as market size, inflation rate, lending interest rate, trade openness, agglomeration effect and wealth effect, in determining FDI inflows in India. By following equation (3.3), table shows the estimated results with three different measures of volatility. TABLE REGRESSION RESULTS: FDI FUNCTION Independent Variables Model - 5 RERVSD Model - 6 NERVSD Model - 7 BIEXVAR Constant (0.439) (0.852) (1.439) RER t (0.251) (0.215) (0.429) VOL t (1.237) (1.316) (0.056) VOL t (2.057)** (2.077)** VOL t (2.825)* (1.961)** FDI t (2.776)** (1.861)*** (2.077)** GDP t (2.432)** (1.810)*** (0.742) OPEN t (0.952) (0.255) (0.213) INFL t (1.980)** (0.118) (0.361) LR t (1.084) (1.179) (0.809) SM t (2.050)** (0.119) (0.875) R Adjusted R F-statistic 2.443** 2.652** 2.94** Durbin Watson stat Note:*, ** and *** indicate one, five and ten per cent levels of significance respectively; and Brackets show t-value. 105

28 The results reveal that the coefficient of real exchange rate is negative and insignificant. Whereas, the coefficients of exchange rate volatility emerge negative and significant in all the models. Only lags of FDI and GDP are positive and significant at five and ten per cent levels respectively. In the case of inflation rate and lending interest rate, the coefficients are with expected negative sign but statistically insignificant. The coefficient of openness of trade shows ambiguous sign, which is insignificant. Wealth effect seems to be positive and significant at five per cent level in Model - 5, but positive and significant in Model - 6 and Model - 7. The results from Model - 5 to 6 show weak evidence of effects of inflation rate, lending interest rate, wealth effect, openness of trade and real exchange rate in determining FDI in India. This may be due to the small sample size used for the analysis, which may lead to degrees of freedom problem. Hence, to overcome the difficulties, Autoregressive Distributed Lag (ARDL) model would be appropriate for examining the long-run relationships using cointegration techniques. The purpose is to find out whether exchange rate levels and its volatility have any short-run and longrun effects on FDI, besides identifying the factors determining foreign country decision to invest in India. 4.7 Cointegration Test based on ARDL Approach Prior to the testing of cointegration, the test of order of integration for each variable using ADF and PP tests have been conducted in tables and Even though the ARDL framework does not require pre-testing of variables to be done, unit root test would convince whether or not the ARDL model should be used. The results in table and 4.13 show that there is a mixture of integrated to the order of one and zero (i.e., I (1) and I (0)) of underlying regressors. For instance, INFL and 106

29 RERVSD are stationary at levels in both ADF and PP test. RER and NERVSD seem to be stationary at levels (i.e., integrated of order zero I (0)) in ADF test, but not in PP test. In the case of PP test, GDP and OPEN are stationary at I (0). Thus, it can be seen that mixed results prevail across the ADF and PP tests). These results indicate that unit root test reached by ADF and PP tests are robust in this study. Given the fact that INFL, RERSD, NERVSD and RER are stationary at level, while the rest are stationary at first difference, long-run equilibrium relationship may be investigated by using the bounds test for cointegration within ARDL modeling approach Bounds test Table presents F-statistic of the ARDL model. By considering RERVSD, NERVSD and BIEXARCH as a proxy for exchange rate volatility, the model has been run with constant and no trend, and with constant and trend. Critical values of the Bounds F-statistic are also given in the table. In the first stage of the ARDL procedure, the order of lags for equation (3.10) has been obtained from unrestricted Vector Autoregressive (VAR) by means of AIC. 107

30 Model Measure of Volatility TABLE BOUNDS TEST F-statistic 1. RER_MASD C C&T NER_MASD C C&T BIEX_ARCH C C&T Critical Value Bounds of the F-Statistic# C(III) C(V) Note: C - constant; C&T - constant and trend; and. # - the critical value bounds are from Table C(III) and C(V) in Pesaran, Shin and Smith (2001, p. 300). The results show that when tested for the joint significance of lagged level variables with an intercept and no trend, all F-statistics are significant at two lags. The calculated F-statistics is greater than the upper bound critical value at five per cent level of significance, indicating that there is long-run relationship among the variables. When is incorporated trend into the Model - 2, 4 and 6, all F-statistics again emerge significant at five per cent level of significance with two lags. Thus, the null hypothesis of no cointegration cannot be accepted, implying that there is cointegration relationship among the variables in each of the models. In the next phase, long-run coefficient and the associate error-correction model have been estimated. Here, a major decision emerges regarding the choice of lag lengths that need to be used in the error-correction model. Akaike information criteria (AIC) methods have been employed for the purpose. These methods have been 108

31 widely used in time-series analysis to determine the appropriate length for model selection. In theory, as a user of this information criteria for model selection guide, one should choose to select the model with the smallest information criterion to determine the appropriate model, i.e., choose the model that has the smallest standard error with the highest order of ARDL. The point estimates of both AIC and Schwarz information criterian (SIC) models are similar, but the estimated standard errors obtained using the model selected by AIC are much smaller and also give a much higher order ARDL. Therefore, AIC method is employed in the second stage to select lag order for all the variables Long-run test results Having discovered a long-run relationship under F-statistic, the ARDL procedure is applied to estimate its coefficients and draw inference about the values. The longrun results with three different measures of volatility based on AIC with appropriate ARDL models are reported in table The long-run coefficients for inflow of FDI into India with and without trend have been presented in the table. 109

32 TABLE LONG RUN RESULTS Model ARDL GDP OPEN SM INFL LR RER VOL C T Measure of Volatility RERVSD 1. (2,1,0,2,0,0,1,2) * (1.98)** (2.10)* (3.38) (2.15)** (0.73) (0.71) (2.31)* (1.05) 2. (2,2,0,2,1,0,0,1,0) (2.56)** (0.15) (0.05) (2.63)* (0.48) (0.64) (1.39) (0.57) (0.53) Measure of Volatility NERVSD 3. (2,1,0,2,0,0,1,0) (2.28)** (1.97)** (1.32) (4.10)* (0.78) (0.39) (2.55)* (1.94) 4. (2,1,0,2,0,1,1,0) (1.805) (0.46) (0.93) (1.86) (0.50) (0.58) (1.88) (0.76) (0.46) Measure of Volatility BIEXVAR 5. (2,1,0,2,0,1,1,1) (1.96)** (1.82) (2.36)** (3.29)* (0.03) (1.17) (1.97)** (1.19) 6. (2,2,0,2,1,0,1,1) (2.07)** (2.28)** (3.11)* (1.71) (0.06) (1.23) (2.53)* (1.01) (1.28) Note: *, ** and *** indicate one, five and ten per cent level significance respectively; and Parentheses show t-values. 110

33 The coefficients of trade openness, inflation and exchange rate volatility emerge negative and statistically significant at one per cent and five per cent levels respectively, when Model - 1, 3 and 5 are estimated without trend. Whereas, the coefficients of market size (GDP) and wealth effect (SM) carry a positive sign and are significant at five per cent level respectively. Both RER and BPLR have the expected signs, but are statistically insignificant. On the other hand, when trend term is incorporated into the Model - 2, 4 and 6, the results show that GDP has a positive sign and is significant. RER and lending interest rate still have a negative effect, but emerge statistically insignificant. Inflation rate and exchange rate volatility are negative and significant only in Model - 2 for the former and Model - 6 for the latter. The result also shows that stock market wealth carries positive sign as expected, and is statistically significant only in Model - 6 with trend Error correction model results Table , 4.20 and 4.21 present the estimates of error correction models (ECM). 111

34 TABLE ERROR CORRECTION MODEL Independent Model - 1 Model 2 Variables Coefficient t-value Coefficient t-value FDI t ** ** FDI t ** GDP t ** ** GDP t ** ** GDP t ** OPEN * SM t SM t ** SM t ** ** INFL t ** INFL t LR t ** RER RER t ** - - RERVSD * ** RERVSD t ** ** RERVSD t ** - - ECM t ** ** C ** ** T ** Adjusted R F-statistic 26.8* 19.01* Durbin Watson stat Note: *, ** and *** indicate one, five and ten per cent level significance respectively 112

35 TABLE ERROR CORRECTION MODEL Independent Model 3 Model 4 Variables Coefficient t-value Coefficient t-value FDI t ** * FDI t ** GDP t ** GDP t ** * OPEN t ** SM t SM t SM t ** ** INFL t * LR LR t RER t RER t NERVSD t * ** ECM t ** ** Constant Trend Adjusted R F-statistic 20.26* 24.79* Durbin Watson stat Note: *, ** and *** indicate one, five and ten per cent level significance respectively 113

36 TABLE ERROR CORRECTION MODEL Independent Model 5 Model 6 Variables Coefficient t-value Coefficient t-value FDI t ** ** FDI t ** GDP t ** GDP t ** * GDP t ** OPEN t SM t SM t ** SM t ** ** INFL t ** INFL t LR t LR t RER t * * RER t ** BIEXVAR t * BIEXVAR t * ** ECMt ** ** C ** ** T ** Adjusted R F-statistic 20.36* 19.41* D.W Note: *, ** and *** indicate one, five and ten per cent level significance respectively. It is apparent from the tables that the lagged error-correction terms carry negative sign and statistically significant at five per cent level in all the models (Model-1 to 6). This further supports the cointegration results obtained by using F-test at the first stage (see table ). Since the prime focus of the present investigation is with respect to the exchange rate variable, different measures of volatility have been used as shown in table The results show estimated coefficients of real exchange rate to be negative and statistically insignificant in all the models. This implies that a depreciation of Indian Rupee encourages inflows of FDI into the country. For instance, other factors 114

37 remaining constant, a one per cent depreciation of Indian Rupee on average would increase FDI by 0.27 per cent to 0.52 per cent. This result is supported by earlier studies, like Cushman (1988); Froot and Stein (1991); Klein and Rosengren (1994); Goldberg and Kolstad (1995); Blonigen (1997); Goldberg and Klein (1997); Amuedo- Dorantes and Pozo (2001); and Kiyota, et. al. (2004). Similar results are also found in error correction mechanism (ECM) tables , table and table , except in Model-1 and 5, where the real exchange rate levels are negative and significant at five per cent level at lag one. More in line with earlier works, evidence of the detrimental impact of exchange rate volatility on FDI has been found in the form of negative and significant coefficients in the case of all measures of volatility, viz., RERVSD, NERVSD and BIEXAR. There is only limited evidence of the same effect when trend is included in the NERVSD model. This result is confirmed by some of the previous studies, which found that volatility in exchange rate decreases FDI flows (see Itagaki (1981); Cushman (1985); Goldberg and Kolstad (1995); Sung and Lapan (2000); and Blonigen (2005)). The negative and significant coefficient of exchange rate volatility is indicative of a negative impact of exchange rate uncertainty on FDI flows into the country. The negative coefficient of exchange rate uncertainty confirms the prediction of the Goldberg and Kolstad model presented in section - 3.2, that a MNEs will ignore better business opportunities in the host country if it finds high variability in exchange rate. FDI investors lack the capability to hedge (in order to reduce risk of exchange rate variability) in the long-run. Therefore, exchange rate volatility might emerge as an important determinant for risk averse investor (Benessy-Quere, et. al. 2001). Hence, 115

38 stabilisation of currency is also an important policy factor that needs to be considered as a measure to promote FDI. GDP, included to capture the effects of size of the market in the estimated Models - 1 to 6, show positive sign and significance at five per cent level in all the models, excepting Model 4, which includes trend and volatility measures in terms of nominal exchange rate. GDP incorporates several dimensions of the economy, but the estimated relationship in the FDI model suggests the importance of a growing domestic market, besides increasing and more efficient utilisation of natural resources, and exploitation of economies of scale (Chakrabarti 2001). These help to promote sustainable economic growth and development that create domestic capacity to maximise benefits from FDI, which in turn tends to strengthen the domestic economy through greater competitiveness, expanded domestic and export markets, and optimal financial resource allocations. In the case of trade openness, the results emerge negative and significant. This implies that India s trade openness decreases inflow of FDI. That is, as the foreign country participates more in FDI, trade tends to decrease. This can be considered as an evidence of the prevalence of horizontal FDI, where firms penetrate foreign markets through FDI rather than trade. This implies FDI inflows through horizontal linkages. Sun, et. al. (2002) also found similar negative impact due to the existence of greater competition in the domestic market. Agglomeration effects are clearly evident from all the ECM tables (i.e., 4.19, 4.20, 4.21), where the lags of FDI are positive and significant at five per cent respectively. This indicates that foreign investors are attracted to countries with an existing concentration of other foreign investors. In this case, investment by larger number of 116

39 existing foreign investors is seen as a good signal of favourable condition to invest in India with less uncertainty. Supportive to the present study, Wheeler and Moody (1992), Amirahrnadi and Wu (1994), Head, Rise and Swenson (1995), Head and Rise (1996), Cheng and Kwang (2000), Cheng and Kwan (2000) and He (2003) had also found evidence of agglomeration effect in their studies. This result is also consistent with the evidence found for countries like Italy (Bronzini 2004) and the U.S. (Head, et. al. 1995). The negative association between inflation rate and FDI shows that foreign investors are risk-averse. The result indicates that one per cent increase in inflation rate will cause FDI to decrease, on an average, by 0.11 per cent to 0.23 per cent. This implies that higher inflation rate may lead to a reduction in FDI inflows to India. It is because investors will not risk expected profits from investment. The prices of goods and services in the host country would increase due to inflation, resulting in higher costs leading to higher factor prices. Thus, inflation in the host country tends to negatively impact investment decision of multinational corporations (MNCs), especially of those which plan to raise capital in the domestic market. Like exchange rate volatility, high inflation rate is considered a reflection of poor macroeconomic conditions. Therefore, to encourage investment, it is necessary to stabilise inflation rate in India. Policies that control inflation rate tend to reduce risk and uncertainty of investment, and improve business environment that attract FDI. Studies, such as Schneider and Frey (1985), Garibaldi, et. al. (2001), Trevino, et. al. (2002), Rodrik (1990 and 1996), and Fischer and Sahay (1996 and 2000), have confirmed the negative relationship between inflation and FDI inflows. 117