Dividend, investment and the direction of causality

Transcription

1 Working Paper 2/2011 Dividend, investment and the direction of causality P S Sanju P S Nirmala M Ramachandran DEPARTMENT OF ECONOMICS PONDICHERRY UNIVERSITY March 2011 system28 [Type the company name] 1/11/2011

2 Working Paper No. 2 (2011) Department of Economics Dividend, investment and the direction of causality P S Sanju P S Nirmala Ph. D. Scholars, Department of Management Studies, Pondicherry University, Puducherry M Ramachandran Professor, Department of Economics, Pondicherry University, Puducherry Abstract In this study, we address an econometric issue which has so far been neglected by the empirical studies on separation principle. The earlier studies largely applied Granger causality test by differencing the data if they are integrated time series. Such approach produces specification bias if integrated variables in level are cointegrated and thus, ignoring the long run dynamics among the variables. To circumvent this problem, we test for cointegration between investments and dividends and estimate a dynamic panel vector error correction model. Annual time series data over the period 1995 to 2008 for various sectors chosen on the basis of the available sectoral indices of National Stock Exchange (NSE) are considered for empirical analysis. The empirical evidence derived from group mean and lambda-pearson tests seem to indicate that there is long run causal link between investments and dividends and therefore, firms decisions regarding dividend payout and investments are inseparable. Keywords: Dividends, Panel Unit root, panel Cointegration, FMOLS 1. Introduction The theoretical work pioneered by Miller and Modigliani (1961) have demonstrated that dividend payout of a firm cannot have any contemporaneous impact on its share price or on the value of the firm in an economy characterized by perfect capital market, rational behaviour and perfect certainty 1. Under such conditions, firms in 1 Perfect capital market is characterized by an equal and costless access to information about price and other details of shares, with no brokerage fees, no transfer taxes or other transaction costs and no tax differentials between dividends and capital gains. Rational behaviour implies that investors prefer more to less wealth and are indifferent to whether they receive income in the form of dividends or capital gains. By perfect certainty it means that investors are assured of the future investment program of the firm and its future profits (Miller and Modigliani, 1961). principle decide the optimal level of investment regardless of how it is financed or in other words dividend payout of a firm is independent of its investment decision. This proposition is popularly known as separation principle in the finance literature. The issue that dividends and investments are neither complementing nor competing with each other turned out to be an empirical curiosity and the first empirical verification of this proposition was conducted by Fama (1974). Consistent with the theoretical prediction, the empirical evidence indicated that there is no period-by-period association between dividend payout and investments of individual American firms. Subsequently, there were several empirical studies that supported this proposition: McDonald et al. (1975) for French firms; Morgan and Saint- 1

3 Pierre (1978) for Canadian firms; Smirlock and Marshall (1983), Pruitt and Gitman (1991) and Elston (1996) for US firms; and Partington (1985) for Australian firms. On the contrary, Dhrymes and Kurz (1967) and several others have argued that internal funds are cheaper sources of financing when there are capital market imperfections; hence, firms prefer to use internal funds to finance additional investments than issuing new securities. Under such circumstances, there is a trade-off between dividends and investments with limited amount of internal funds. Moreover, there are empirical studies that report evidences to support the claim that dividends and investments are interdependent. For instance, Dhrymes and Kurz (1967) and McCabe (1979) found strong evidence in favour of the hypothesis that investments and dividend decisions of US firms are jointly determined. Similar empirical result is also reported by Chiarella et al. (1992) for Australian firms. The empirical studies during the last two decades largely applied causality tests to examine the separation principle. For instance, Mougoué and Mukherjee (1994) have reported evidence of bi-directional causality between investments and dividend decisions for US manufacturing firms. Similarly, evidence of bi-directional causality between investments and dividends of American firms is reported by DeFusco et al. (2007). Mougoué (2008) has applied both linear and nonlinear causality tests to detect the direction of causality between dividends and investment decisions of American firms. While the linear causality tests present evidence in support of the separation principle, nonlinear causality tests present evidence in favour of strong bi-directional causality between dividends and investment decisions. However, Louton and Domian (1995) found unidirectional causality from dividends to investments for approximately 33% of the American firms. To our knowledge, the recent study by Bhaduri and Durai (2006) is the only empirical exercise pertaining to India that investigates the separation principle. The study applied panel Granger causality test and found that dividends and investment decisions of Indian manufacturing firms are jointly determined. The direction of causality seems to be conditional upon the growth level of the firms. The causality runs from dividends to investments for firms with higher growth whereas there is bi-directional causality for firms which grow at a slow pace. The empirical studies largely depended on causality tests either using time series or panel data to examine the separation principle. There are few crucial econometric issues which are ignored by the existing empirical studies that investigate the issue under consideration. The studies document that the data used in econometric estimations are nonstationary stochastic process and hence, define a vector autoregression to conduct causality test among variables which are differenced sufficient number of times to ensure stationarity. This, however, generates a specification bias if variables in their levels are cointegrated. Under such situation, ideally we need to test for cointegration and if variables are cointegrated, we must construct an error correction mechanism to test for causality. Further, studies that examine cointegration between dividends and investments do not proceed to estimate the error correction mechanism and thus ignore the long run dynamics among the variables 2. We account for these issues in econometric estimations, which makes this study differ from the earlier empirical works. Firstly, we use panel data to examine the separation principle and apply panel 2 There is an attempt by Mougoué (2008) who has tested for cointegration using time series data on investment and dividend. However, they could not construct an error correction mechanism since investment and dividend are not cointegrated. 2

4 cointegration test proposed by Pedroni (1999, 2004) to test for cointegration between dividends and investments. Further, we apply principles of FMOLS to estimate cointegrating vector as it accounts for heterogeneity across individual members of the panel. Secondly, the long run dynamics between dividends and investments is captured through an error correction mechanism if variables of interest are cointegrated. The rest of the paper is organized as follows: Section II deals with the econometric methodology; empirical results are discussed in Section III and concluding remarks are presented in Section IV. 2. The Econometric Methodology 2.1 Panel cointegration test In this study, we use the econometric method proposed by Pedroni (1999, 2004) which is meant for testing cointegration among a set of variables using panel data. This test is an extension of the Engle and Granger (1987) two step residual based procedure for testing the null hypothesis of no cointegration in the case of heterogeneous panels. The major advantage of this test is that it allows for individual member specific fixed effects, deterministic trends and slope coefficients. The approach proposed by Pedroni (1999) involves the following steps: In our case, a panel regression of investment (I) on dividend (D) of the following form is estimated (1), where t = 1,..., T; i = 1,..., N; T is number of observations; N is number of cross-sectional units; is slope coefficient of ith member; is the member specific intercept; is member specific trend coefficient. The investment and dividend in Equation 1 are integrated of the same order and said to be cointegrated if is a stationary process; hence, testing for cointegration between investment and dividend involves testing for stationarity of. The stationarity of the residuals from Equation 1 can be tested by estimating the following auxiliary regression: (2) The null hypothesis =1 implies that has unit root; hence, accepting the null hypothesis of no cointegration. In order to test the null hypothesis, Pedroni (1999) proposes two different sets of statistics, namely, the within-dimension statistics and the between-dimension statistics. Withindimension statistics are also known as panel cointegration statistics and betweendimension statistics as group mean panel cointegration statistics. There are seven test statistics of which, Panel Variance, Panel Rho, Panel PP and Panel ADF statistic are withindimension statistics, while Group Rho, Group PP and Group ADF statistics are between dimension statistics. Although the null hypothesis is the same, the alternative hypothesis is, however, different for the two sets of statistics. The null hypothesis relating to within dimension statistics is defined as =1 for all i against the alternative of for all i. The alternative hypothesis implies that there is cointegration among the variables of all the members of the panel. The null hypothesis pertaining to between dimension statistics is defined as = 1 for all i against the alternative of for all i. In this case, unlike within dimension statistics, a common value for is not assumed. Thus, the alternative hypothesis implies that cointegration exists for atleast one individual member of the panel. The between dimension statistics, therefore, allows to model an additional source of potential heterogeneity across individual members of the panel. 2.2 Fully modified ordinary least squares The application of OLS method to obtain the cointegrating vector from a panel leads to 3

5 biased estimates due to endogeneity problem. However, the FMOLS method of Pedroni (2000) accounts for heterogeneity across individual members of the panel, corrects for serially correlated errors and resolves the endogeneity problems; hence, the estimates are unbiased. The FMOLS produces two types of estimators, viz., pooled panel estimator and group mean panel estimator. The former is based on within dimension of the panel whereas the latter is based on between dimension of the panel. In the case of pooled panel estimator, the null hypothesis is defined as H0: = for all i against the alternative of H1: = for all i, where is the hypothesized common value for under the null and is some alternative value for which is also common to all members of the panel. In the case of group mean panel estimator, the null hypothesis is defined as H0: = for all i against the alternative of H1: for all i, where are not necessarily constrained to be homogeneous across different members of the panel. Thus the group mean FMOLS estimator provides greater flexibility by allowing heterogeneity of the cointegrating parameters. 2.3 Panel vector error correction model If each element of a vector of variables is individually I (1) and a linear combination of them is I (0), then the Granger Representation theorem (Engle and Granger, 1987) justifies the existence of a dynamic error correction representation among the variables. Canning and Pedroni (2008) construct the dynamic panel error correction model to identify the direction of long-run causality following a two-step procedure. Firstly, the estimates of the cointegrating parameters are obtained using the principles of FMOLS. Secondly, the disequilibrium errors are estimated from the cointegrating relationship to define the error correction model. For instance, the error correction model for investment and dividend can be defined as (3) where is first difference operator; and are intercept terms; and are speed of adjustment coefficients; is disequilibrium error in the previous period; s are the slope coefficients; and and are white noise error terms. Using the error correction model, the long run causal relationship between investment and dividend can be inferred. If the variables of interest are cointegrated, then according to Granger representation theorem, at least one of the two variables adjusts to correct the disequilibrium in the system; implying that either or should be statistically significant. Testing the null hypothesis that = 0 and = 0 helps identify the long-run causality between dividends and investments. For instance, acceptance of the null hypothesis that = 0 indicate innovations in dividends have no long-run causal effect on investments. Similarly, the null hypothesis that = 0 implies innovations in investments have no long-run causal impact on dividend. To test for the existence of long-run causal relationship, Canning and Pedroni (2008) construct a group mean based test and a lambda-pearson test. These tests are useful in determining the direction of long-run effects in heterogeneous panels, but not its magnitude. The group mean panel estimate for is computed as, hence, the group mean based test averages the individual. The null hypothesis therefore implies that the long-run causal effect is zero on average for the panel. The null hypothesis that is tested using the test statistics computed as, where is the individual t-statistic for the null hypothesis that = 0. The Lambda-Pearson panel test uses the probability values associated with each 4

6 individual t-statistic ( and takes the form. The statistic is distributed as a with 2N degrees of freedom under the null hypothesis of no long run causal effect for the panel and the null is rejected in favour of the alternative on the right tail of the distribution. Using the same principles, we can obtain the group mean and lambda-pearson statistics for. 3. The Empirical Results The empirical investigation of causality between dividends and investments has been carried out using panel data consisting of annual time series over the period 1995 to 2008 and cross section data pertaining to eight sectors chosen on the basis of the available sector wise stock indices listed in NSE. The number of sectors and the corresponding number of firms are chosen based on certain criteria: (i) the firm s stock should have been listed in any one of the NSE sectoral indices; (ii) continuous annual time series data on dividend and investment for the selected sample period must be available; and (iii) each sector should have minimum of nine firms that satisfy criteria (i) and (ii). Such screening yielded a final sample 3 of eight sectoral indices with a total of 110 firms, the details of which are given in Table 1. Dividend is measured as the amount of annual dividend paid out by the firm while investment is measured as the sum of annual change in gross land and building, plant and machinery, transport and communication equipment, furniture, social amenities and other fixed assets. The firm level data on dividend and investment are collected from Prowess database provided by the Center for Monitoring Indian Economy. 3 Complete list of the selected firms under various sectoral indices can be availed from the authors upon request. Table 1 S. No. of NSE sectoral index No. firms 1 CNX IT Index 11 2 CNX Bank Index 12 3 CNX FMCG Index 09 4 CNX PSU Bank Index 12 5 CNX MNC Index 09 6 CNX PSE Index 20 7 CNX Service Sector Index 24 8 CNXInfrastructure Index 13 Total 110 Notes: IT Information Technology; FMCG Fast Moving Consumer Goods; PSU Public Sector Units; MNC Multinational Corporations. Testing for cointegration involves examination of the unit root properties of the data, because the variables must be integrated of the same order. The present study uses three alternative methods, which are meant for testing unit roots in heterogeneous panel, proposed by Im, Pesaran and Shin (IPS) (2003), Fisher type Augmented Dicky-Fuller (Fisher-ADF) and Phillip-Perron (Fisher-PP) tests proposed by Maddala and Wu (1999) and Choi (2001) respectively. These tests are chosen especially for the reason that they accommodate individual member specific unit root process unlike other panel unit root tests that assume common unit root process. The specifications concerning IPS test will have either individual intercept or individual intercept and trend whereas Fisher-ADF and Fisher-PP tests consist of three specifications: individual intercept; individual intercept and trend; and none. The statistics of the chosen unit root tests to test the null hypothesis that dividend is nonstationary in level are reported in Table 2. The statistics, excepting Fisher-PP test with respect to the specification that include member specific intercept and trend, fail to reject the null. This implies that the data on dividends of all the chosen sectors are nonstationary. However, the results reported in Table 3 indicate that dividend in first 5

7 difference are stationary with an exception that the IPS test fails to reject the null in the case of Bank sector. Thus, the unit root test statistics presented in Tables 2 and 3 indicate that data on dividend pertaining to all the sectors under consideration follow I (1) process. The results regarding investment in level are presented in Table 4. The IPS test fails to reject the null for all the sectors, excepting the specification that includes individual intercept and trend with respect to Bank and MNC indices. The IPS test thus indicates that investment in level is nonstationary. The Fisher-ADF test yields mixed results. In the case of sectors PSE, Service and Infrastructure, the test fails to reject the null under all the specifications. However, in case of remaining sectors, some specifications fail to reject the null, while others reject the null. This implies that investment in level can be considered as nonstationary. Fisher-PP test, however, gives contradictory results as compared to other two test results. Based on the results of IPS and Fisher-ADF tests, investment in level is nonstationary. The results regarding investment in first difference are reported in Table 5. All the three tests reject the null for all the sectors; hence, investment data for all the sectors seem to follow I (1) process. Table 2. Panel unit root test (dividend in level)[ Null hypothesis H 0: series has unit root] Dividend in Level Sectoral Index Fisher ADF test Fisher PP test IPS test S1 S2 S3 S1 S2 S3 S2 S3 IT (0.72) (0.93) (0.27) (0.73) (0.91) (0.94) Bank (0.70) (0.86) FMCG (0.83) (0.65) (0.99) PSU Bank (0.36) (0.06) (0.88) MNC (0.99) PSE (0.76) (0.53) Service (0.98) (0.81) Infrastructure (0.97) (0.98) Notes: Numbers in parentheses are p-values. S1 refers to the specification without intercept and trend; S2 refers to the specification with intercept; S3 refers to the specification with intercept and trend. 6

8 Table 3. Panel unit root test (dividend in first difference) [Null hypothesis H 0: series has unit root] Dividend in first difference Sectoral Index Fisher ADF test Fisher PP test IPS test S1 S2 S3 S1 S2 S3 S2 S3 IT Bank (0.04) (0.26) (0.31) FMCG PSU Bank MNC (0.02) PSE Service Infrastructure Notes: Numbers in parentheses are p-values. S1 refers to the specification without intercept and trend; S2 refers to the specification with intercept; S3 refers to the specification with intercept and trend. Table 4. Panel unit root test (investment in level) [Null hypothesis H 0: series has unit root] Investment in level Sectoral Index Fisher ADF test Fisher PP test IPS test S1 S2 S3 S1 S2 S3 S2 S3 IT (0.05) (0.12) (0.03) (0.10) (0.78) (0.06) Bank (0.04) (0.05) (0.07) (0.26) (0.04) FMCG (0.02) (0.08) (0.55) (0.44) PSU Bank (0.46) (0.04) (0.03) (0.04) (0.13) (0.18) MNC (0.04) (0.30) PSE (0.30) (0.34) (0.17) (0.08) (0.18) Service (0.27) (0.52) (0.28) (0.94) 0.34 (0.63) Infrastructure (0.06) (0.20) (0.74) (0.36) 1.10 (0.87) Notes: Numbers in parentheses are p-values. S1 refers to the specification without intercept and trend; S2 refers to the specification with intercept; S3 refers to the specification with intercept and trend. 7

9 Table 5. Panel unit root test (investment in first difference) [Null hypothesis H 0: series has unit root] Investment in first difference Sectoral Index Fisher ADF test Fisher PP test IPS test S1 S2 S3 S1 S2 S3 S2 S3 IT Bank FMCG PSU Bank MNC PSE Service Infrastructure Notes: Numbers in parentheses are p-values. S1 refers to the specification without intercept and trend; S2 refers to the specification with intercept; S3 refers to the specification with intercept and trend. Thus, the evidence regarding unit root largely indicate that both investments and dividends are integrated series of order one; hence, we can proceed to test for cointegration between these two variables. In this regard, three alternative specifications of Equation 1 is estimated; imposing the restriction that = = 0; 0 & = 0 and 0 & 0. The null hypothesis that there is no cointegration between investments and dividends directly means that the regression error ( ) has a unit root. Tables 6 through 13 report various test statistics for the sectors under consideration. For the purpose of interpretation, we consider the Panel-ADF and the Group-ADF test statistics to test for cointegration between the variables 4. 4 Joyeux and Ripple (2004) state that the group-adf test as well as the panel-adf test performs better in case of time dimension of panel being less than 100. In similar direction, Lee and Chang (2008) also point out that the panel-adf and group- ADF statistics have better small sample properties as compared to other statistics. Table 6. Panel cointegration test results: CNX IT index (Null hypothesis H 0: no cointegration) Test Statistics S1 S2 S3 Within Dimension Statistics Panel-v (0.08) Panel Rho (0.10) Panel PP Panel ADF Between Dimension Statistics Group Rho (0.26) Group PP Group ADF Notes: Nos. in parentheses are p-values. S1 refers to the specification without intercept and trend; S2 refers to the specification with intercept; S3 refers to the specification with intercept and trend. 8

10 Table 7 Panel cointegration test results: CNX Bank index (Null hypothesis H 0: no cointegration) Test Statistics S1 S2 S3 Within Dimension Statistics Panel-v Panel Rho (0.06) Panel PP Panel ADF Between Dimension Statistics Group Rho (0.02) 0.61 (0.73) Group PP Group ADF Notes: Nos. in parentheses are p-values. S1 refers to the specification without intercept and trend; S2 refers to the specification with intercept; S3 refers to the specification with intercept and trend. Table 8. Panel cointegration test results: CNX FMCG index (Null hypothesis H 0: no cointegration) Test Statistics S1 S2 S3 Within Dimension Statistics Panel-v (0.62) (0.95) Panel Rho (0.39) Panel PP Panel ADF Between Dimension Statistics Group Rho (0.08) (0.12) 0.56 (0.71) Group PP Group ADF Notes: Nos. in parentheses are p-values. S1 refers to the specification without intercept and trend; S2 refers to the specification with intercept; S3 refers to the specification with intercept and trend. Table 9. Panel cointegration test results: CNX PSU Bank index (Null hypothesis H 0: no cointegration) Test Statistics S1 S2 S3 Within Dimension Statistics Panel-v (0.63) Panel Rho Panel PP Panel ADF Between Dimension Statistics Group Rho (0.02) (0.39) Group PP Group ADF Notes: Nos. in parentheses are p-values. S1 refers to the specification without intercept and trend; S2 refers to the specification with intercept; S3 refers to the specification with intercept and trend. Table 10. Panel cointegration test results: CNX MNC index (Null hypothesis H 0: no cointegration) Test Statistics S1 S2 S3 Within Dimension Statistics Panel-v (0.27) (0.98) Panel Rho (0.20) Panel PP Panel ADF Between Dimension Statistics Group Rho (0.22) Group PP Group ADF Notes: Nos. in parentheses are p-values. S1 refers to the specification without intercept and trend; S2 refers to the specification with intercept; S3 refers to the specification with intercept and trend. 9

11 Table 11. Panel cointegration test results: CNX PSE index (Null hypothesis H 0: no cointegration) Test Statistics S1 S2 S3 Within Dimension Statistics Panel-v 0.27 (0.39) (0.99) Panel Rho (0.02) Panel PP Panel ADF Between Dimension Statistics Group Rho (0.08) (0.16) 1.09 (0.86) Group PP Group ADF Notes: Nos. in parentheses are p-values. S1 refers to the specification without intercept and trend; S2 refers to the specification with intercept; S3 refers to the specification with intercept and trend. Table 12. Panel cointegration test results: CNX Service sector index (Null hypothesis H 0: no cointegration) Test Statistics S1 S2 S3 Within Dimension Statistics Panel-v (0.03) Panel Rho (0.02) Panel PP Panel ADF Between Dimension Statistics Group Rho (0.62) Group PP Group ADF Notes: Nos. in parentheses are p-values. S1 refers to the specification without intercept and trend; S2 refers to the specification with intercept; S3 refers to the specification with intercept and trend. Table 13. Panel cointegration test results: CNX Infra..index (Null hypothesis H 0: no cointegration) Test Statistics S1 S2 S3 Within Dimension Statistics Panel-v 1.33 (0.09) 0.30 (0.38) (0.99) Panel Rho (0.35) Panel PP Panel ADF Between Dimension Statistics Group Rho (0.03) (0.13) 0.53 (0.70) Group PP Group ADF Notes: Nos. in parentheses are p-values. S1 refers to the specification without intercept and trend; S2 refers to the specification with intercept; S3 refers to the specification with intercept and trend. The panel-adf and group-adf statistics in Table 6 indicate that the null of no cointegration between investments and dividends is rejected in the case of IT sector. Similarly, both panel-adf and group-adf statistics presented in Tables 7 through 13, for sectors Bank, FMCG, PSU Bank, MNC, PSE, Service and Infrastructure respectively show that the null hypothesis can be rejected. In addition, the Panel-PP and the Group-PP test statistics also reject the null of no cointegration between investments and dividends for all the sectors. Thus, all the test statistics consistently confirm the existence of long-run equilibrium relationship between investments and dividends. Since there is cointegration between investments and dividends, it is worth examining the long run cointegrating parameter of the model. However, application of OLS method to cointegrated panel produces biased and inconsistent estiamtes. In this regard, the Panel group 10

12 mean FMOLS method of Pedroni (2000) is more suitable as it not only provides consistent estimates of the cointegrating parameters but also accounts for endogeneity of the regressors and serial correlation. Therefore, we estimate the cointegrating relationship using the principles of Pedroni (2000) and the estimates of cointegrating parameters and the t-statistics are presented in Table 14 for all the sectors 5. The results indicate that the cointegrating coefficient is positive and statistically significant at conventional level for all the sectors under consideration. This implies that dividend has a significant positive influence on investment in the long-run. However, the long run response of investment to dividend varies across various sectors. For instance, dividend seems to have larger influence on investment in the case of FMCG sector and least impact on investment in the case of PSE sector. Table 14. Panel group FMOLS results NSE Sectoral Index Coefficient CNX IT 3.18 (6.97)*** CNX BANK 2.10 (10.37)*** CNX FMCG 5.04 (1.92)* CNX PSU BANK 1.50 (10.21)*** CNX MNC 1.13 (6.42) *** CNX PSE 0.40 (3.90)*** CNX SERVICE SECTOR 1.41 (10.87)*** CNX INFRASTRUCTURE 0.98 (5.98)*** Notes: Numbers in parentheses are t-values. *** denotes significance at 1% level and * denotes significance at 10% level Since the variables are cointegrated, the next step is to construct an error correction model to infer the direction of causality between investments and dividends. We estimate the error correction model 6 of the form given in Equation 3. The results of the group mean based test and lambda-pearson based test for all the sectors are produced in 5 The estimates are obtained using the RATS code provided by Peter Pedroni. 6 This has been estimated in Win RATS 6.0 using the RATS code provided by Peter Pedroni. Table The group mean based test fails to reject the null of no long-run causal effect of dividend on investment for all the sectors, while the lambda-pearson test rejects the null for all the sectors. For the null of no long-run causal effect of investment on dividend, the group mean based test fails to reject the null for all the sectors, excepting IT and Bank. However, the lambda-pearson based test rejects the null for all the sectors excepting PSE. Such contradictory results of the group mean based test and the lambda-pearson based test can occur when the values of individual s are significantly positive for some fraction of the panel and significantly negative for the remaining portion of the panel. However, it can be concluded that a long-run causal effect is present, even though it is positive for some members of the panel and negative for others (Canning and Pedroni, 2008). These results imply that the long-run causal effect between investment and dividend are zero on average for all the sectors excepting IT and Bank. In case of these two sectors, long run causal effect of dividend on investment is zero on average while the reverse causal effect is nonzero on average. Further, as indicated by the lambda-pearson test results, bi-directional causality between investment and dividend exists for all the sectors under consideration, excepting PSE. In case of PSE sector, the causality runs from dividend to investment. In sum, the econometric evidences indicate that firms decisions regarding dividend payout and investments are inseparable. 7 Before conducting the panel estimation, it is worth investigating whether the member specific parameters are homogenous. In this context, Canning and Pedroni (2008) has used Wald test to examine the homogeneity of the parameters across the members of the panel. However, the reliability of the test is likely to be poor if the time dimension of the sample is short; hence, given the short time sample used in the present study, we have not attempted to test for homogeneity of the parameters. 11

13 Table 15. Panel long-run causality test results (Null Hypothesis (H 0): No long-run causal effect for the panel) NSE Sectoral Index : D i,t I i,t : I i,t D i,t Estimate Test p-value Estimate Test p-value CNX IT Group Mean (0.05) Lambda-Pearson CNX BANK Group Mean (0.10) Lambda-Pearson CNX FMCG Group Mean (0.23) Lambda-Pearson CNX PSU BANK Group Mean (0.17) Lambda-Pearson CNX MNC Group Mean (0.34) Lambda-Pearson CNX PSE Group Mean (0.48) Lambda-Pearson (0.40) CNX SERVICE SECTOR Group Mean (0.29) Lambda-Pearson CNX INFRASTRUCTURE Group Mean (0.57) Lambda-Pearson Conclusion In this paper, we address an important econometric issue which has been neglected by the empirical studies which investigate the theoretical proposition that investments and dividend decisions of firms are separable. The earlier studies largely applied Granger causality test using differenced time series data on investments and dividends if they are integrated time series. Such approach produces specification bias if integrated variables in their level are cointegrated; thus, ignoring the long run dynamics among the variables. To circumvent this problem, we test for cointegration between investments and dividends and estimate a dynamic panel vector error correction model. The empirical investigation of separation hypothesis has been carried out using panel data consisting of annual time series over the period 1995 to 2008 and cross section data pertaining to eight sectors chosen on the basis of the available sector wise stock indices listed in NSE. The empirical results derived from group-mean based test and lambda- Pearson based test indicated that there is long-run causal link between investments and dividends for all the sectors, excepting PSE for which the causality runs from dividends to investments. Consistent with the inference drawn by the earlier empirical studies in the Indian context, the present study also support the view that firms decisions regarding dividend payout and investments are inseparable. 12

14 References Bhaduri, S. N. and Durai, S. R. S. (2006) Empirical Relationship between the Dividend and Investment decision: Do Emerging Market Firms Behave Differently?, Applied Financial Economics Letters, 2, Canning, D. and Pedroni, P. (2008) Infrastructure, Long-run Economic Growth and Causality Tests for Cointegrated Panels, The Manchester School, 76, Chiarella, C., Pham, T. M., Sim, A. B. and Tan, M. M. L. (1992) The Interaction of the Financing and Investment Decisions: Preliminary Results in the Australian Context, Asia Pacific Journal of Management, 9, Choi, I. (2001) Unit root tests for panel data, Journal of International Money and Finance, 20, DeFusco, R. A., Dunham, L. M. and Geppert, J. M. (2007) The Dynamic Relation among Investment, Earnings and Dividends. Available at ecognition/pdfs/lee_dunham.pdf (accessed 18 December 2009). Dhrymes, P. J. and Kurz, M. (1967) Investment, Dividend, and External Finance Behavior of Firms. Available at (accessed 23 December 2009). Elston, J. A. (1996) Dividend Policy and Investment: Theory and Evidence from US Panel Data, Managerial and Decision Economics, 17, Engle, R. F. and Granger, C. W. J. (1987) Co- Integration and Error Correction: Representation, Estimation, and Testing, Econometrica, 55, Fama, E. F. (1974) The Empirical Relationships between the Dividend and Investment Decisions of Firms, The American Economic Review, 64, Im, K., Pesaran, M. H., and Shin, Y. (2003) Testing for Unit Roots in Heterogeneous Panels, Journal of Econometrics, 115, Joyeux, R. and Ripple, R. D. (2004) The Evaluation of Standard of Living and the role of Household Electricity Consumption: A Panel Cointegration Analysis. Available at 004/SOL.pdf (accessed 5 January 2010). Lee, C. and Chang, C. (2008) Tourism development and economic growth: A closer look at panels, Tourism Management, 29, Louton, D. A. and Domian, D. L. (1995) Dividends and Investment: Further Empirical Evidence, Quarterly Journal of Business and Economics, 34, Maddala, G. S. and Wu, S. (1999) A Comparative Study of Unit Root Tests with Panel Data and a New Simple Test, Oxford Bulletin of Economics and Statistics, 61, McCabe, G. M. (1979) The Empirical Relationship Between Investment and Financing: A New Look, The Journal of Financial and Quantitative Analysis, 14, McDonald, J. G., Jacquillat, B. and Nussenbaum, M. (1975) Dividend, Investment and Financing Decisions: Empirical Evidence on French Firms, The Journal of Financial and Quantitative Analysis, 10, Miller, M. H. and Modigliani, F. (1961) Dividend policy, growth, and the valuation of shares, The Journal of Business, 34, Morgan, I. and Saint-Pierre, J. (1978) Dividend and Investment Decisions of Canadian Firms, The Canadian Journal of Economics, 11, Mougoué, M. (2008) An Empirical reexamination of the Dividend-Investment relation, Quantitative Finance, 8, Mougoue, M. and Mukherjee, T. K. (1994) An Investigation into the Causality among firms' Dividend, Investment, and 13

15 Financing Decisions. Available at (accessed 23 December 2009). Partington, G. H. (1985) Dividend Policy and its Relationship to Investment and Financing Policies: Empirical Evidence, Journal of Business Finance & Accounting, 12, Pedroni, P. (1999) Critical Values for Cointegration Tests in Heterogeneous Panels with Multiple Regressors, Oxford Bulletin of Economics and Statistics, 61, Pedroni, P. (2000) Fully Modified OLS for Heterogeneous Cointegrated Panels, Advances in Econometrics, 15, Pedroni, P. (2004) Panel Cointegration: Asymptotic and Finite Sample Properties of Pooled Time Series Tests with an Application to the PPP Hypothesis, Econometric Theory, 20, Pruitt, S. W. and Gitman, L. J. (1991) The Interactions between the Investment, Financing, and Dividend Decisions of Major U.S. Firms, The Financial Review, 26, Smirlock, M. and Marshall. W. (1983) An Examination of the Empirical Relationship between the Dividend and Investment Decisions: A Note, The Journal of Finance, 38,

16 Working Paper Series 2/2011 DEPARTMENT OF ECONOMICS PONDICHERRY UNIVERSITY RV NAGAR, KALAPET PUDUCHERRY