___ __ ___ 6~ tYpt /Z / '/ Policy fl.s.rch WORKING PAPER$ Debt and Intrnational Finance __.L Inte.national Economics Department The World Bank October 1993 WPS 1211 Savings-lnvestment Correlations and Capital Mobility in Developing Countries Nlandu Mamingi Many developing countries are financially inttgrated in the long run and several show evidence of capital mobility in the short run. Savings-investment correlations ere lower for middle- income than for lower-income countries. The Policy Resetrch Working Psais disseminate the findings of work in progress and encaurage thcexchange ofideas am mg Bank stafT and all others interested in development issues. These papers, distributed by the Rcsearch Advisory Staff, carry the names of the authors, rcflect nlytheirviews, and should beused and citedaccordingly. The findingp,inteapreutions, and conclusions ame the authors'own They should not be attributed ta the World Bank, its Board of Directors, its management, or any of its member countries. Pollay Ro"roh and rIntwntional Finance WPS 1211 This paper - a product of the Debt and Intemational Finance Division, Intcrnational Economics Department - is part of a largei effort in the department to study the effects of extemal financing on developing countries. Copies of the paper are available free from the W'orld Bank, 1818 H Street NW, Washington, DC 20433. Please contact Rose Vo, room S8-042, extension 31047 (October 1993,28 pages). Mamingi estimates savings and investment More important, the estimates from this correlations for 58 developing countries to assess robust estimation technique indicate that sav- the capital mobility (in the Feldstein-Horioka ings-investment corrclations arc lower for sensc) in these countries. middle-income than for lower-income countries. Using a new estimation technique (fIully Mamingi also provides cvidence of capital modified ordinary least squares) - which mobility for several of these countrics in the simultaneously corrects for serial correlation, short run. endogeneity, and sample bias (asymptotically) - Mamingi finds that many developing coun- tries are financially integrated in the long run. The Policy Research Working Paper Sc,ies disseminates the findings of work under way in the B ank. An objectiv c of the scri es is to get these findings out quickly, even if presentations are less than fully polished. The findings, interpretations, and conclusions in these papers do not necessarily represent official Bank policy. Produced by the Policy Rcsearch Dissemination Center Savings-investment Correlations and Capital Mobility in Developing Countries by Nlandu Mamingi Debt and International Finance Division International Economics Department The World Bank Table of Contents 1. Introduction 1 2. The Feldstein- Horioka Regression: OLS Estimates 4 3. Unit Root and Cointegration Properties of the Data a 4. Fully Modified OLS 7 5. Causality and Short-run Capital Mobility 9 6. Summary and Concluding Remarks 12 Aopendix 1: Derivation of FMOLS Estimators 14 Appendix 2: Tables 15 1. List of Countries 15 2. Regression Results: OLS Estimates 16 3. OLS Estimates with a Time Trend 18 4. Capital Mobility with OLS Estimates 19 5. FMOLS Results 20 6. Capital Mobility with FMOLS Results 21 7.Correlations: saving-investment estimates and country size 22 8. Granger Causality from the ECMs 23 9. Short-Run OLS Estimates 24 10. Capital Mobility in the Short-Run 25 References 26 1. Introduction The finding by teldstein and Horioka (1980) that national saving Ind domestic investment are highly positively correIted has generated a lot of debate among economists on the extent to which the "capital immobility implication" attributed to this correlation implies lack of financial openness. In fact, despite disagreements over the implications of this finding, the puzzle itself has been replicated in a number of subsequent studies using mainly cross section data from OECD or EC countries (see, for examples, Pena'i and Dooley (1984) I phy (1984), Feldstein and Bachetta (1990), Bayoumi (1990), and Tessr (1991)). Explanations of the puzzle (which have yet to meet the consensus of the protagonists) include analysis of the impacts of sample bias, endogeneity of saving, capital controls and.o: fiscal policy, productivity shock and lack of integration of goods markets, and country size (see Tesar (1991) for a discussion of the d:fferent arguments). The few studies cn the experiences of developing countries reveal that the magnitude of the coefficient measuring the degree of capital mobility is lower (see Wong (1990) and Montiel (1993)). Wong utilizes a cross-section approach to analyze saving-investment correlations for a sample of 45 countries over the period 1975-1981. Although the correlation (0.08) that he found is lower than those in previous studies, his finding is, nevertheless, very sensitive to influential observations. Thus, after dropping 5 countries, the correlation becomes 0.613. Montiel adopts a I would like t) thank Ronald Johannes, Stijn Claessens, Punam Chuhan, Vikrarn Nehru and Yonas 3iru for valuible comments. 2 time series approachi with various tests for capital mobility (strength of saving-investment correlations, size ot gross capital flows, uncovered interest rate parity and behavior of domestic consumption cver time). He is able to show that many developing countries experience sapital mobility. This paper reexamines the evidence on capital mobility on the basis of saving- investment correlations u-,ing annual time series data from 58 developing countlies for the period from 1970 to 1990 with special attention to the issues of serial correlation, endogeneity of saving, and sample bias. A time series approaciA is adopted here because the cross-section approach utilized in most of the studies on saving-investment correlations is flawed ill many respects (see Gundlach and Sinn, 1992, p. 818). First, results from cross-section models are hard to interpret, at least in this type of exercise. Indeed, as capital mobility estimates are derived at a particular point in time, the key question of how much of an increase in saving truly ends up as domestic investment' becomes difficult to answer. Further, the use of long-term averages of savings and investment ratios leads to an upward bias in capital mobility correlations. Second, there is no guarantee that capital mobility estimates for different countries are effectively equal, something cross-section models imply. in fact, for reasons such as capital controls and differences In country size, one would expect capital mobility estimates to vary across countries. rhird, since the saving-investment correlation is primarily a long-run relationship, a cointegration (long-run relationship between variables) approach is a more appropriate methodology. A study of capital mobility is important because different degrees of capital mobility hold different policy implications. In the event of perfect capital mobility, one should expect: (a) monetary policy to be ineffective in influencing the prices of domestic financial assets; and (b) expansionary fiscal policy to be ineffective for purposes of demand management. Whereas complete capital immobility, which implies that domestic investment is entirely financed by 1This is pointed out by Gundlach and Sinn, 1992, p.618. 3 domestic or national saving, should give rise to an active role for monetary and fiscal policies (see Montiel (1993) for further details). This paper contributes to the literature in three ways. First, tne study shows that an omission of a significant time trend in the Feldstein-Horioka regression can, in some cases, either significantly change the capital mobility estimate (i.e., Nepal and Venezuela) or alter the extent of the puzzle (i.e., Burundi, India and Venezuela). This is a well-known variable omission problem, which unfortunately hais been overlooked in the literature on saving-investment correlations. Second, the application of cointegration and error correction models enable us to obtain long-run and short-run estimates of capital mobility. Third, the use of a robust estimation technique (such as the fully modified ordinary least squares (FMOLS) of Phillips and Hatisen (1990)) can, under some conditions, attenuate the extent of tho) puzzle. The examples of India, Thailand and Paraguay are cases in point. More imponantly, contrary to the finding of previous papers, the FMOLS estimates do indicate that saving-investment correlations for middle-income countries are as a whole lower than those for low-income countries. Overall, this study finds that the evidence of high correlations between saving and investment is largely absent in developing countries. The paper is organizea as follows. Section 2 reports and interprets the ordinary least squares (OLS) estimates from the Feldstein-Horioka regression for developing countries. Section 3 examines the unit root and cointegration properties of the data. Section 4 reexamines th Oasic regression with fully modified ordinary least squares. Section 5 develops causality analys to shed light on the issue of endogeneity of saving. It also investigates short-run capital mobility through the error correction models. Section 6 summarizes the main findings. 4 2. The Feldstein-Horioka Regression: OLS Estimates The objective of this section is to estimate and to interpret the basic Feldstein-Horioka regression. Annual data on the gross national saving - GDP ratio (St) and the gross domestic investment - GDP ratio (It) for 58 developing countries (see Table 1) over the period 1970 to 1990 are used. The data are obtained from the World Bank's World Tables 1991 and 1992 (update). The basic Feldsteiri - Horioka regression is as follows: lt = c, + b.S, + e, (1 ) where the variables are defined as above, c is a constant term and et is the error term. According to Feldstein and Horioka (1980), the coeffirient b measureb the degree of capital mobility and takes values from zero (perfect capital mobility) to one (complete capital immobility). The OLS estimates from equation (1) are presented in Table 2. The results show that the following countries experience capital mobility at least at the 5 percent level of significance: Brazil, Colombia, Costa Rica, Gambia, Israel, Kenya, Madagascar, Ma,aysia, Malta, Mauritania, Morocco, Rwanda, Sierra Leone, a,,d Togo. Capital is immobile in the following countries: B'irundi, Fiji, Ghana, Guatemala, India, Honduras, Malawi, Nepal, Niger, Sri Lanka, Thailand, Philippines, Tunisia and Venezuela. Other countries of the sample are in an intermediate position (intermediate degree of financial openness). In terms of the coefficient size, sixteen countries nave coefficients greater than 0.60 (the benchmark for developed countries established by authors who dealt with OECD and EC countries). It is well known, however, that the estimate of b in equation (1) can suffer from a variable omission bias. A time trend variable can be expected to be the most important omitted 5 variable. In fact, I' Is quite possible that the time trend captures most omitted variables. Thus, equation (1) is modified as follows: I = c+hbS' +dT +-el (2) where It and St are defined as above and T is the time trend, If equation (2) is true, then the estimate of b in equation (1) is biased.2 Surprisingly, a deterministic trend is significant in the regression specification for thirty- two countries. Table 3 reports the OLS estimates. The comparison of Tables 2 and 3 for the above thirty.two countries indicates that for several countries the coefficient b is significantly changed or the conclusion about the degree of capital mobility is substantially altered. This is the case for Co'e d'lvoire, India, Morocco, Pakistan, Sri Lanka, Trinidad and Tobago, Venezuela and Zambia. As reported ii-, Table 4, the OLS estimates of equations (1) and (2) imply that 21 countries are characterized by perfect capital mobility. The countries are: Colombia, Costa Rica, C6te d'lvoire, Gambia, Israel, Kenya, Lesotho, Madagascar, V.-'aysia, Malta, Mauritania, Pakistaii, Peru, Rwanda, Sierra Leone, Sri Lanka, Ugarnda, Togo, Trinidad and Tobago, Venezuela and Zambia. Counitries which lack capital mobil,ty are: Fiji, Guatemala, Honduras, Malawi, Niger, Philippines, Thailand and Tunisia, The other countries in our sample display imperfect capital mobility. Equations (1) and (2) are subject to several econometric problems. First, as is often the case with OLS results from time series data, there is autocorrelation in the error term which introduces bias in the sampling variances and makes the estimates inefficient. In short, the t statistics are unreliable. Secrond, the savings variable may well be endogenous; implyinig inconsistent estimates. Third, the small size of the sample introduces a sample bias Last but 2The bias is equal to the "true coefficient of the omitted variable times the regression coefficient of the excluded variable on the included variable" (Maddala 1977, p 156) 8 not least, results are meaningless If savings and Investment are Integrated of order one (or have different degrees of integration) and their linear combination (that is, et) is not stationary. Because uf Kts seriousness, the last problem is investigated first in the next section. 3. Unit Root and Cointegrat in Properties of the Data The objective of this section Is to study the unit root (non stationaritv of the univariate series) and cointegration (long-run relationship between integrated variables) properties of the data to determine whether or not regressions (1) and (2) are spurious. The (Augmented) Dickey-Fuller test reveals that savings and investment have a unit root (integrated of order one) for all developing countries except fur Kenya and Benin (results are available upon request). The unit root result is not clear cut for savings for Burundi, Gambia, Tunisia, C6te d'lvoire and Chile and for investment for Chile, Colombia, Costa Rica, Malta and -urkina Faso. As each of the two variables apparently contains a unit root for many countries, it is necessary to examine whether their linear combination is stationary, that is, whether the two variables are cointegrated. If the two variables are cointegrated then the usual statistical inference can proceed normally and the basic results obtaii,ed from (1) and (2) can be validated to some extent. As the usual tests for cointegration (cointegrated Durbin-Watson, Dickey-Fuller, Augmented Dickey-Fuller and Phillips-Ouliaris) have low power against many alternatives given the small sample size, the t statistic of the coefficient of the error correction term in either one of error correction models (see equations (3) and (4) in section 5) is utilized to test for cointegration. Specifically, cointegration is accepted if the t statistic of either a1 in equation (3) or a2 in equation (4) is significantly different from zero and negative. 7 Before reportin; the results on cointegration, it is worth emphasizing that cointegration is a desirahle property even in the presence of capital mobility contrary to the a,gument that some authors make about capital mobility implying 'he absence of cointegration (see, fo; example, Leachman (1991)). Indeed, as the error tern, in equation (1) represents the current account balance, the solvency property requires et to be bounded ( Montiel, 1993, p. 32). The following quote also reinforces this idea "It cannot be concluded, however, that a country is shut off from the international capital market if its current account balance is found to be integrajed of order zero, 1(0). A number of studies suggest that cver time both s.aving and investments rates are influenced by the same exogenous variables. In that case saving and investment rates could be cointsgrated and the current account balance would be i(rf) even if the country is linked to international capital market (Gundlach and Sinn, 1992, p. 618))." The results reported in Table 8 indicate that the t statistic is significant at the 10 and 5 per cent level and negative in equation (3); hence cointegration is accepted in all the relationships examined here (with the exceptions of Kenya and Benin whose variables in levels are already stationary). Although ;he results presented so far for equations (1) and (2) are acceptable, there are problems of serial autocorrelation, endogeneity of saving and sample v ias which need to be t-l-en care of. The next section deals with these issues using fully modified OLS technique. 4. Fully Modifiedr OLS Estimates The fully modified OLS of Phillips and Hansen (1990), (FMOLS) is utilized in equations (1) and/or (2). This technique corrects for endogeneity and serial correlation and asymptotically eliminates .he sample bias. In the literature on saving-investment correlations, the instrumental variable (IV) method has been comrrmonly used to solve the problem of endogeneity of saving. However, aside from endogeneity, there are problems of serial correlation and sample bias that need to be addressed 8 and the IV method does not solve them. In a cointeg .-'tion context. safnple bias may occur not only in small sample sizes but also in moderate or even large sample sizes (second-order sample bias). oe IV estimates ina\ be useful as first estimates of the FMOLS if the ratio signal to noise is low. Otherwise, the OLS estimates are ied as first estimates of the FMOLS. In this paper, the latter method is pursued (see appendix 1 for details on FMOLS) Table 5 reports the FMOLS estimates, Since the length of the la! truncation to be used in the estimation of the long-run covariance matrix is not clear cut, although PThillips and Hansen (1990) rely on the cross correlogram between the "innovations" and the "exogenoij:, variable" as well as on the correlogram of the innovations. tht final b estimate obtdined here is the mean of the b coefficients from different lag truncations. The standard error is also obtained analogously to the final b. Table G translates the results of the previous table in terms of capital mobility. These results are the correct long-rur, estimates of capital rnobility. As can be seen, only 11 out of 58 countries fail to demonstrate any degree of capital mobility. The comparison of Tables 4 and 6 shows that several countries have changed their status in terms of capital mobility. 8 hus, for example, Costa Rica, C6te d'lvoire, Malaysia, Malta, Lesotho, Togo, and Zambia are no longer in the category of "mobile capital". Instead, except for Zambia. there are now in the category of " imperfect mobile capital". Table 7 quantifies the relationship between saving-investment correlations and country size by way of simp!e correlations or Spearman's rank cor,elations. Contrary to previous findings, the correlation coefficients indicate that saving-investment FMOLS estimates and country size are negatively correlated. This is particularly true for a small sample of countries (eigh!). In other words, the larger the country, the lower is the correlation between saving and investment, hence, the mo,e mobile is capital in the country. To sum up, the FMOLS estimates show that a large number of countries do experience capital mobility in one form or another contrary to results obtained in earlier studies and there is 9 a negative relationship between saving-investment correlation estimates and country size. Policy implications can be directly inferred from the size of capi'al mobility estimates. In this respect, several papers explain them well (see Montiel (1993) in particular). Although this is not the place to duplicate these pipers, we can, nevertheless, emphasize that, among others, this finding means that in many third world countries monetary policy is ineffective in dictating the price of financial assets and fiscal policy is powerless in the sense that the crowding out effect on private investment does not occur. 5. Causality and Short-run Capital Mobility This section deals with causality analysis to shed light on endogeneity or exogeneity of saving in equations (1) or (2), and estimates short-run capital mobility. These two goals Pre pursued in the framework of error correction models (ECMs). Exogeneity or causality analysis is important to the extent that it can legitimize the use of instrumental variables techniques. Note that in the literature endogeneity of saving is taken as granted. In fact, the support for the IV approach is weakened if it is shown that saving is exogenous (the FMOLS estimates are not affected by this remark (see Phillips and Hansen (1990)). The study on short-run capital mobility is undertaken to show that capital mobility is also a short run phenomenon. An error correction model is either a vector autoregression or a dynamic model which contains both short and long run elements. In other words, in these models, the change in one variable is explained by the past equilibrium error, the present/or the past change of the other variable(s) and the past change of the explained variable The dynamic versson of 1he ECM derived from Hendry's approach to econometrics is of iruerest here to capture short-run capital mobility. The vector autoregression approach helps us conduct causality analysis 10 Causality and exogeneity are llnked. Briefly, In equations (3) and (4), St is said to be weakly exogenous with respect to the parameters of interest if cov(ut, u't) = 0, that is, basically the lagged error term and the past of It do not belong to equation (4). St is strongly exogenous if St is weakly exogenous and It does not Granger cause St. As can be seen causality is an Important component of exogeneity. In this paper, the emphasis is more on causality than on exogeneity. Causality in the Granger sense is utilized here. According to Granger (1969), a variable St does not Granger cause another variable It if the past of St does not help better predict It than does the past of t alone. Granger causality can be tested using either the usual vector autoregression or the error correction models. In fact, if variables are cointegrated, then Granger causality is adequately tested in the ECM framework. Precisely, for the two variables of interest, the following ECMs can be fitted: Alt = c + a, e, | + lagged(AI,, AS, ) + u, (3) AS, = c + a, e , + lagged(AI,, AS, ) + IJ, (4) where et-, is the lagged error correcting term from (1) or (2), A is the first difference operator, the u's are the error terms supposed to be white noise. The "Granger representation theorem" states that every cointegrated vector has a valid error correction model representation, that is, at least one of the a's in the above equations is different from zero and negative. Clearly, if a1 < 0 and significantly different from zero, then the ECM (3) is valid and Granger causality runs from St to It or precisely from the lagaed equilibrium error to It. 11 The results presented in Table 8 show that causality runs from savings to investment in all the countries under investigation as the t statistics indicate except for Burkina Faso, Chile, Colombia and Korea where a feedback seems to prevail. The unidirectional causality indicates that saving is likely to be an exogenous variable rather than an endogenous variable contrary to the current literature, with the exceptions of Burkina Faso, Chile, Colombia and Korea. In other words, the IV method can only be justified for these four countries. Thus, (1) and (2) are valiJ regressions and short-run capital mobility can be tested with the following: AI, =c-+a3e,_l +,0AS, +u, (5) AS, = c + ui, (6) where a3 < 0 and significantly different from zero, the us are white noise and j is the multiplier of impact which here captures short-run capital mobility. It is worth noting that in the real Hendry's methodology it is not the lagged error correcting term from (1) (or (2)) which is the lagged error correcting term but the one period lag of (It -c - b St)). Further, the lagged variables have not been added to preserve the degree of freedom. Table 9 presents the results of the inquiry. Accordingly, the following countries experience capital mobility in the short-run (see Table 10): Algeria, Brazil, Colombia, Costa Rica, Cote d'lvoire, Ecuador, El Salvador, Fiji, Gambia, Guatemala, Haiti, Israel, Jamaica, Korea, Lesotho, Malawi, Malaysia, Malta, Mauritius, Morocco, Nigeria, Pakistan, Rwanda, Sri Lanka, Paraguay, Thailand, Togo, Trinidad and Tobago, and UCanda. Capital immobility is registered by Egypt, Honduras, Niger and Mauritania. Five others countries experience imperfect capital mobility. 12 6. Summary and Concluding Remarks This paper estimates saving-investment correlations for 58 developing countries in order to assess the degree of capital mobility (in the Feldstein-Horioka sense) for these countries. The paper utilizes a time series approach and pays special attention to the problems of serial correlation, endogeneity of saving and sample bias. Using a new estimation technique (fully modified ordinary least squares'. ;tudv finds that many developing countries are financially integrated in the long-run. The f s of i)is robust estimation technique indicate that in the context of developing coun,, s. avings- investment correlations are in general lower for middle-income countries than for low-income countries. Further, using an error correction model a la Hendry, the paper also provides evidence of capital mobility in the short-run in several of these countries. Overall, our results indicate that saving-investment correlations are much lower for developing countries than those obtained by other studies using mainly OECD or EC data. The finding of low saving-investment correlations implies that financial assets in several developing countries are mobile, especially in the long-run. As these cointries are small open economies, expansionary fiscal policy is ineffective for purposes of demand management to the extent that private investment is not crowded out. Further, under a fixed exchange rate regime, monetary policy is ineffective in dictating the prices of domestic financial assets. Naturally, the extent of these macroeconomic policy implications largely depends on the degree of capital mobility of countries. Our results indicate that the above macroeconomic policy effects (i.e., ineffectiveness of fiscal policy) are more present in middle-income countries than in low-income countries. Two areas are relevant for further research. First, aid flows could be included in the basic model in order to test the robustness of results obtained here Second, the sensitivity of the negative relationship between saving-investment correlations and country size to change in country sample could be investigated. 13 Appendix I Derivation of FMOLS estimators (see Hansen and Phillips,1990, p.227- 234). Let us suppose the following: y, = c+bx, +U,1 XI = X, 1 +Ul2g where Yt and xt are the variables of interest and ii, = (u,,,u,1) is a vector of stationary disturbances with ergodic zero mean and finite positive covariance matrix (£). It is the possible correlation between the two components of ut which brings about endogeneity of xt . The long-run covariance matrix (Q) and other statistics are necessary to obtain FMOLS estimators: A = + A l t 10) [21 522 j = E(uOuO) A= EE(u0z4) k=l WI) 1 2 = a) I I - 0)1 2 r°22 AX (011= Yi *- 0 'AX U,f = U, °12 °12 22t U1 = U1 - 012 0 A, 22 1 . where w11.2 is the conditional variance of ut, given the change in the variable xt and f stands for FMOLS. 14 In fact, to obtain the FMOLS estimators of interest, A and n) are tentatively estimated as follows: 1 7 ' A = T- I Y., k-7i) I=k I T I T T-1 it, 7+ 0 (14 U, + 11E, Ek1 ) k-I (-k.l where the weights (kl -1-k/(/+l) are utilized to make the long-run covariance matrix posiJive definite and / is the lag truncation. The fully modified estimator a{, standard errors sf and t statistic tf are, respectively: -f af - (XX)-I[XXy' -m TA21] sf [X X), WiI1 /2 tif (al' -(a,') 1.sf where X represents all the right-hand side variables including the constant, af= (c, bf )'and 'm [ IJ Note that the FMOLS estimator is asymptotically equivalent to the maximum likelihood estimator applied to the whole system. 15 Appendix 2: Tables The data for the different tables are from the World Bank's World Tables 1991 and 1992 (update). Table 1: List of Countries Algeria Nepal Benin Niger Bruzil Nigeria Burkina Faso Pakistan Burundi Paraguay Cameroon Peru Central African Repub. Philippines Chile Rwanda Colombia Senegal Congo Sierra Leone Cdte d Ivoire Sri Lanka Costa Rica Thailand Dominican Republic Togo Ecuador Trinidad & Tobago Egypt [ unisia El Salvador Turkey Fiji Uganda Gabon Venezuela Gambia Zamnbia Ghana Zimbabwe Guatemala Haiti Honduras India Israel Jamaica Kenya Korea Republic Lesotho Madagascar Malawi Malaysia Mali Malta Mauritania Mauritius Mexico Morocco 16 Table 2: Regression Results: OLS Estimates (see eq.(1)) Country b R2 DW Country b ' DW Algeria 0.56 0.34 0.96 Benin 0.42 0.13 0.85 (0 17) (0.21) Brazil 0.09 -0l.30 0.63 Burkina 0.38 0.19 0.67 (0.09) (0.16) Burundi 0.75 0.22 0.56 Camer. 0.36 0.42 1.30 (0.29) (0.09) C. Af. R. 0.57 0.57 1.56 Chile 0.55 0.46 1.34 (0.11) (0.13) Colom. -0.08 -0.03 1.77 Congo 0.53 0.32 0.86 (0.11) (0.16) C. Rica 0.19 0.06 1.46 C6te Iv. 0.40 0.28 0.89 (0.17) (0.14) Dom. R. 0.35 0.23 1.27 Ecuador 0.51 0.27 1.55 (0.13) (0.17) Egypt 0.68 0.49 0.71 El Salv. 0.57 0.55 1.66 (0.15) (0.11) FiJi 0.75 0.26 0.57 Gabon 0.57 0.34 1.01 (0.26) (0.17) Gambia -0.14 0.01 0.46 Ghana 0.74 0.46 1.25 (0.29) (0.17) Guate. 0.89 0.6B) 1.95 Haiti 0.28 0.15 0.63 (0.16) (0.13) Hondu. 0.83 0.53 1.20 India 0.97 0.65 0.27 (0.13) (0.16) Israel -0.21 -0.03 0.19 Jamaica 0.74 0.46 1.38 (0.32) (0.17) Kenya 0.22 0.00 1.81 Korea 0.33 0.36 1.05 (0.22) (0,09) Lesotho -0.36 0.06 0.38 Madaga. 0.02 -O.5 0.64 (0.23) (0.23) Malawi 0.88 0.36 1.11 Malaysia 0.11 -0.05 0.45 (0.25) (0.32) 17 Table 2: Regression Results (continued) Mali 0.16 0.17 0.52 Malta 0.10 -0.00 0.98 (0.07) (0.10) Maurita. -0.03 -0.05 1.02 Mauritius 0.61 0.34 0.70 (0.22) (0.1 9) Mexico 0.28 0.10 0.61 Morocco -0.15 0.34 0.70 (0.15) (0.21) Nepal 1.05 0.72 0.44 Niger 0.97 0.71 1.14 (0.14) (0.14) Nigeria 0.61 0.52 1.43 Pakistan 0.20 0.44 0.74 (0.13) (0.06) Philippi. 1.24 0.66 1.10 Rwanda -0.06 -0.04 0.67 (0.20) (0°14) Senegal 0.35 0.48 0.63 S.Leone -0.04 -0.04 1.12 (0.08) (0.12) Sri Lanka 0.70 0.11 0.64 Thailand 0.74 0.41 1.03 (0.38) (0.19) Paraguay 0.53 0.19 0.44 Peru -0.007 0.05 0.77 (0.22) (0.005) Togo 0.13 -0.03 0.71 Trinidad 0.28 0.18 0.68 (0.21) (0.12) Tunisia 0.92 0.4. 0.97 Turkey 0.41 0.28 1.20 (0.23) (0.14) Uganda -0.24 0.11 0.20 Venez. 0.71 0.33 1.09 (0.13) (0.21) Zambia 0.49 0.53 1.98 Zim'bwe 0.54 0.43 0.88 (0.10) (0.14) Note: Equation (1) is of interest with the b estimates and the standard errors in parentheses as well as the adjusted R2 and the Durbin-Watson statistic for autocorrelation (DO. '8 Table 3: OLS Estimates with a Time Trend (see equation 2). Co'try b d j-- D.W Co'try b d DW Algeria 0.46 -0.36 0.45 1.09 Brazil 0.15 -0.19 0.11 0.90 (0.16) (0.16) (0.09) (0.10) Burun. 0.50 0.69 0.75 1.58 Chile 0.54 0.16 0.50 1.53 (0.17) (0.11) (0 12' (0 10) Colom. -0.16 0.07 0.00 2.00 Congo 0.66 -0.81 0.51 1.40 (0.13) (0.05) (0.15) (0.28) C.Rica -0.11 0.21 0.22 2.02 C6te I. 0.22 -0.54 0.47 0.70 (0.19) (0.09) (0.13) (0.19) Egypt 0.78 0.52 0.71 1.32 Fiji 0.81 -0.44 0.49 0.91 (0.12) (0.13) (0.22) (0.14) Gamb. -0.24 0.69 0.25 0.70 India 0.67 0.20 0.89 1.45 (0.25) (0.24) (0. 10) (0.03) Israel 0.03 *0.76 0.87 1.44 Leso. -0.01 2.67 0.86 1.46 (0.11) (0.06) (0.1 0) (0.26) Meda. 0.08 0.18 0.06 0.77 Malawi 0.73 -0.40 0.47 0.55 (0.22) (0. 10) (0.24) (0.17) Malay. -0.20 0.37 0.19 0.55 Mali 0.17 0.25 0.60 1.12 (0.33) (0. 18) (0.05) (0,05) Mexico 0.52 -0.20 0.21 0.71 Moro. -0.74 0.64 0.37 1.73 (0.19) (0.11) (0.23) (0.18) Nepal 0.60 0.52 0.96 2.49 Nigeria 0.54 -0.34 0.59 1.63 (0.06) (0.04) (0.12) (0.16) Pakis. 0.04 0.19 0.51 0.67 Parag. 0.52 0.31 0.38 0.59 (0.08) (0.09) (0.19) (0.12) Rwan. -0.01 0.34 0.44 1.32 Sene. 0.21 -0.31 0.73 1.11 (0.11) (0.08) (0.07) (0.07) Sierra -0.10 -0.19 0.15 1.55 Sri La. 0.37 0.39 0.23 0.59 (0.11) (0.08) (0.38) (0.19) Trinid. 0.09 -0.61 0.52 1.04 Ugan. -0.00 0.74 0.82 0.74 (0.10) (0.16) (0.06) (0.08) Vene. 0.20 -0.85 0.44 0.99 Zamb. 0.03 -1.19 0.67 1.79 (0.31) (0.40) (0.17) (0.39) Note: Equation (2) is of interest with the b and d estimates, the standard errors in parentheses, the adjusted R2 and the Durbin-Watson Statistic for autocorrelation (D.). 19 Table 4: Capital Mobility with OLS Estimates Mobile Mobile Intermediate Intermediate Immobile Colombia Togo Algeria Jamaica Fiji Costa Rica Trinidad & Tob. Benin Korea Guatemala C6te d'lvoire Uganda Burkina Faso Mali Malawi Gambia Venezuela Burundi Mauritius Niger Israel Zambia Cameroon Mexico Philippines Lesotho Sri Lanka Central Af. R. Nepal Tunisia Madagascar Kenya Chile Nigeria Honduras Malaysia Congo Paraguay Thailand Malta Dominican R. Senegal Mauritania Ecuador Zimbabwe Pakistan Egypt Brazil Peru El Salvador Ghana Rwanda Gabon Haiti Sierra Leone India Turkey Note: Mobile: b in Table 2 or 3 is statistically equal to zero at the 5% level; intermediate: b is statistically different from zero and one and 0 < b<1; immobile: b is statistically different from zero and not different from one. The result for Morocco is ambiguous ( see Table 3). We do not consider other negative coefficients (i.e., Colombia, Costa Rica, Gambia, Uganda and Malaysia) ambiguous because they are not significantly different from zero, at least at the 5% level. 20 Table 5: FMOLS Results (see Appendix I for detalls) Country b Country b Country b 1 Country b Alger0s* 0.40 Brazil* 0.33 3urkina 0.56 Burundi* 0.60 (0.21) (0.11) (0.18) (0.23) Camer. 0.32 C. Af. R. 0.58 Chile 0.72 Colom. -0.07 (0.12) (0.12) (0.12) (0.10) Congo 0.85 Costa R.* -0.56 C6te Iv.1 0.40 Dom. R. 0.40 (0.18) (0.14) (0.19) (0.13) Ecuador 0.55 Egypt* 0.62 El Selva. 0.64 Fiji 1.16 (0.17) (0.10) (0.10) (0.23) Gabon 0.67 Gambla* -0.17 Ghana 0.84 Gu'mala^ 1.09 (0.21) (0.35) (0.19) (0.15) Haiti 0.17 Honduras 0.84 India' 0,50 Israel* 0.20 (0.20) (0.17) (0.12) (0.12) Jamaica 0.94 Korea 0.10 Lesotho* 0 51 Mad'car* 0.,5 (0.19) (0.23) (0.21) (0.23) Malawi* 0.98 Malaysia -1.39 Mali* 0.09 Malta 0.24 (0.21) (0.45) (0.06) (0,12) Mau'nia* 0.25 Mau'tius^ 0.48 Mexico 0.29 Morocco -0.67 (0.27) (0 19) (0.19) (0.29) Nepal* 0.59 Niger 1.08 Nigeria* 0.64 Pakistan* -0.01 (0.18) (0.26) (0.10) (0,10) Parguay* 0.28 Peru 0.01 Philippi. 1.14 Rwanda* -0.11 (0.22) (0.01) (0.22) (0.12) Senegal* 0.36 Sierra L.* -0.29 Sri Lan.* 0.09 Thailand 0.33 (0.07) (0.11) (0.52) (0.15) Togo 0.53 Trinid.* 0.17 Tunisia 1.06 Turkey 0.28 (0.26) (0.1 0) (0.27) (0 14) L!ganda* -0.15 Venez'la* 0.35 Zambia 0.80 Zim'bwe 0.55 (0.14) (0.30) (O 21) (0.16) Note: The b estimates and their standard errors in parentheses are the mean estimates of different lag truncations (1 to 5) of the long run covariance matrices with the exceptions of Uganda (1 to 4 lags), Mauritania (1 to 3 lags) and Mauritius (1 to 4 lags). (*) means that a time trend is included. 21 Table 6: Capital Mobility with FMOLS Estimates Mobile Intermediate Intermediate Immobile Colombia Algeria Nepal Fiji Gambia Brazil Malta Guatemala Haiti Burundi Mauritius Honduras Israel Burkina Faso Togo Jamaica Korea C6te d'lvoire Senegal Malawi Madagascar Central Af. Rep. Thailand Niger Mali Cameroon Nigeria Tunisia Mauritania Congo Turkey Ghana Pakistan Chile Gabon Rwanda Dominican Republic Philippines Paraguay Ecuador Zambia Peru Egypt Sri Lanka El Salvador Mexico India Trinidad & Tobago Lesotho Uganda Venezuela Note: Mobile: b in Table 5 is statistically equal to zero at the 5% level; intermediate: b is statistically different from zero and one and U < b<1; immobile: b is statistically different fiom zero and not different from one. The results for Costa Rica, Malaysia, Morocco and Sierra Leone are ambiguous. 22 Table 7: Correlations between Saving-investment Estimates Lid Country Size (1987 US dollar GNP per capita) Type of Correlation FMOLS Estimates OLS Estimates Simple Correlation 1990 GNP for 34 -0.313** -0.188 Countries (-1.865) (.1.082) S!mple Correlation Average GNP (1970- -0.188 -0.1d4 1990) (-1.082) (-1.062) for 34 Countries Simple ('-elation 1970 GNP for 34 -0.090 -0.184 Countries (-0.531) (-1.062) Spearmai;'s Rank Average GNP (1970- -0.881* -0.381 Corielation 1990) (-4.561) (-1.009) for 8 countries Note. Sources: GNP: World Tables 1992 update. FOLS: Table 5. OLS :Table 2 or 3. GNP: 1987 US dollar GNP per capita. Cross section correlations between saving-investment estimates and country size are calculated. Countries with negative saving-investment estimates have been excluded, 1990 GNP: GNP per capita for 34 countries are collected for 1990. Average GNP: an average for GNP per capita over the period 1970-1990 is computed for each country. 1970 GNP: GNP per capita for 34 countries are collected for 1970. Coefficients are simple correlations or Spearman's rank correlations. The Spearman's rank correlation coefficient is obtained by ranking the observations in each series and by calculating the correlation between the ranks of the two series. This correlation is used here for very small sample size ( 8 countries). The eight countries are the following: Burkina Faso, India, Korea, Mauritius, Paraguay, Thailand, Togo, and Trinidad and Tobago. ( . ) are the I statistios t = (rA 7T2i) / I - r2 with n the sample size and r the correlation *-'oefficient. (*) and (") mean significant at the 1 % and 10% levels, respective'y. 23 Table 8: Granger Causality from the ECMs (see equation (3)) Country tal Country ta1 Country tai Country tal Algeria* -1.951 Brazil* -3.150 Burkina -2.934 Burundi* -1.909 Camer. -2.726 Centraf -1.420 Chile* -3.022 Colomb. *3.706 Congo -2.809 C.Rica^ -2.524 Cote Iv.* -2.683 Dom. R. -2.921 Ecuador -2.398 Egypt* -2.033 El Salv. -2.713 Fiji -2.609 Gabon -1.802 IGCambia* -1.765 Ghana -2.047 Guate.* -2.110 Haiti -1.563 Hondur. -2.746 India* -2.461 Israel* -3.513 Jamaica -1.848 Korea -2.477 Lesotho* -2.559 Madag.^ -2.266 Malawi -1.794 Malays.* -2.466 Mali* -1.832 Malta -1.813 Maurita.* -2.007 Mau'tius' -4.602 Mexico' -1.729 Morocco -2.462 Nepal* -2.433 Niger -2.409 Nigeria -5.136 Pakistan* -1.862 Parag.* -3.390 Peru -2.881 Philip. -4.226 Rwanda -3.167 Senegal* -3.914 S.Leone* -3.078 S.Lanka* -2.102 Thailand -2.244 Togo -2.179 Trinid.* -2.769 Tunisia -2.954 Turkey -2 903 Uganda* -2.245 Venez.' -2.476 Zambia -2.674 Zim'bwe -1.844 Note: The results of eq. (3) are reported here; those of eq. (4) are not significant with the exceptions of Burkina Faso, Chile and Korea. The t statistics of a1 have the following critical values: -1.341 and -1.753 at the 10 and 5 percent level of significance, respectively. Equation (3) is utilized with one lag of the chanCe in the variautes with the exceptions of Colombia (no lag), Cote d'lvoire (3 lags), Nepal (no lag), Pakistan (2 lags), Rwanda (no lag) and Thailand (no lag). Naturally, for the latter countries, the critical values are different. (*) means that the error correcting term comes from a rnodel with a time trend term. 24 Table 9: Short-run OLS Estimates (p) from equation (5) Country p Country p Country D Country p Algeria 0.09 Brazil* 0.02 Burkina 0.34 Burundi* 0.53 (0.18) (0.05) (0.13) (0.13) Camer. 0.33 C. Afr. R. 0.37 Chile* 0.48 Colombia -0.18 (0.13) (0.13) (0.12) (0.15) Congo 0.38 Costa R.* *0.20 C6te Iv.* 0.07 Dom. R. 0.19 (0.18) (0.18) (0.06) (0.09) Ecuador *0.14 Egypt* 1.09 El Salva. 0.21 Fiji 0.28 (0.23) (0.29) (0.14) (0. 1 8) Gabon 0.36 Gambia* -0.15 Ghana 0.33 Guate'la' 0.41 (0. 1) (0.12) (0.16) (0.27) Haitl 0.14 Honduras 0.73 India* 0.82 Israel' *0.13 (0.10) (0.24) (0.07) (0,09) Jamaica 0.21 Korea(-) -0.28 Lesotho* 0.12 Mad'car* -0.49 (0.17) (0.21) (0.10) (0.29) Malawi* 0.28 Malaysia *0.001 Mali* 0.14 Malta -0.12 (0.28) * (0.20) (0.04) (0.15) Maurita.' 0.70 Mauritius 0.13 Mexico* 0.51 Morocco 0.03 (0.35) (0.21) (0. 18) (0.22) Nepal* 0.89 Niger 0.78 Nigeria* 0.07 Pakistan* 0.06 (0.08) (0.14) (0.12) (0.07) Pa'guay^ 0.16 Peru 0.51 Philippi. 0.48 Rwanda* -0.04 (0.14) (0.17) (0.20) (0. 1 0) Senegal' 0.13 Sierra L.* 0.41 Sri Lan.* 0.01 Thailand 0.39 (0.08) (0. 10) (0 .26) (0.27) Togo 0.03 Trinid.T.* -0.09 Tunisia 0.31 Turkey 0.48 (0.15) (0.07) (0 .16) (0.20) Uganda* -0.02 Vene.-la -0.31 Zambia* -0.22 Zimbwe 0.61 (0.07) (0.27) (0.13) (0. 16) Note: (.): standard errors; (*): an error correcting term from (2) is of interest; (-) presence of two orror correction models. All the reg_ssions pass the tests oT autocorrelation, ARCH, heteroscedasticity and misspecification. One lag is included in eq. (5) for Madagascar, Togo. Peru, Honduras and Maurtius. 25 Table 10: Capital Mobility In the Short-Run Mobile Mobile Intermediate Immobile Algeria Pakistan Burundi Mauritania Brazil Rwanda Cameroon Niger Colombia Sri Lanka Central African Rep Egypt Costa Rica Paraguay Chile Honduras Cte d'lvoire Thailand Congo Ecuador Trinidad & Tobago Gabon El Sa'vador Togo Mali Fiji Uganda Mexico Gambia Mauritius Nepal Guatemala Senegal Haiti Sierra Leone Israel Tunisia Jamaica Turkey Korea Zimbabwe Lesotho Burkina Faso Malawi Dominican R. Malaysia Peru Malta Philippines Morocco India Nigeria Ghana Note: This table is derived from Table 8. The degree of mobility (j in eq. (5) ) is defined analogously to that in Table 6. The results for Madagascar, Zambia and Venezuela are ambiguous 26 REFERENCES Bayoumi, Tamim (1990), "Saving-Investment Correlations: Immobile Capital Government Policy, or Endogenous Behavior?" International Economic Journal, 37, 360-387. Charemza, W. Wojciech and Deadman, D.D. (1992), New Directions in Econometric Practce. General to Speciric Modelling, Edwar Elgar, Hants. Dooley, Michael, Frankel, Jeffrey and Mathieson, Donald (1987), "Intemational Ca,)ital Mobility: What Do Saving-investment Correlations Tell Us?" International Monetary Fund Staff Paoers, 34, 503-530. Feldstein, Martin and Horioka, Charles (1980), "Domestic Saving and International Capital Flows," Economic Journal, 90, 314-329. Feldstein, Martin and Bachetta, Phillipe (1991), "National Savings and International Investment," in Bernheim, B. Douglas and John Shoven, (eds.), National Saving and Economic Performance, University of Chicago Press, Chicago, 201-226. Golub, S. Stephen (1990), "Intemational Capital Mobility: Net versus Groiss Stocks and Flows," Jouumal of Intemational Money and Finance, 9, 424-439. Granger, W. J. Clive (1969), "Investigating Causal Relations by Econometric Models and Cross- Spectral Methods," Econometrca, 37, 24-36. Gund!ach, Erich and Sinn, Stefan (1992), "Unit Root Tests of the Current Account: Implication for International Capital Mobility," A2plied Economics, 24, 617-625. Hansen, Bruce and Phillips, C. S. Peter (1990), " Estimation and Inference in Models of Cointegration: A Sirr ulation Study," in Fomby, B. Thomas and George F. Rhodes, Jr. (eds), 27 Advances in Econometrics: Co-integration, Spurious Regressions, and Unit Roots, JAI Press, INC., 225-248. Harvey, C. Andrew (1988), Forecasting, Structural Time Series Models, and The Kalman Filter, Cambridge University Press, Cambridge. Leachman, L. Lori (1991), "Saving, Investment, and Capital Mobility among OECD Countries", ODei Economies Review, 2, 137-163. Maddala, S. Gandharrao (1977), Econometrics, McGraw-Hill Book Company, New York. Mamingi, Nlandu (1992), Essays on the Eftects of Misspecified Dynamics and Temporal Aggregation on Cointegrated Relationships, Ph.D. dissertation, State University of New York, Albany. Mamingi, Nlandu (1993), "The Effects of a Time Trend Omission on Cointegration: Some Empirical Examples," unpublished paper. Montiel, J. Peter (1993), "Capital Mobility in Developing Countries: Some Measurement Issues and Empirical Estimates," PRE Working Paper, No 1103, The World Bank (Februery), 1-57. Murphy, Robert (1984), "Capital Mobility and the Relationship between Saving and Investment in OECD countries," Journal of International Money and Finance, 3, 327-342. Penati, Alessandro and Dooley, Michael (1984), "Current Account Imbalances and Capital Formation in Industrial Countries, 1949-1981," International Monetary Fund Staff Papers, 31, 1-24. Phillips, C. B. Peter and Hansen, Bruce (1990), "Statistical Inference in Instrumental Variables Regression with 1(1) Processes," Review of Economic Studies, 57, 99-125. 28 Sachs. S. Jeffrey (1981), "Tne Current Account and Macroeconomic AdJustment in the 1970s," Brookinas Paoers on Economic Activity, 1, 200-269. Summers, Lawrence (1988), "Tax Policy and International Competitiveness," In Frenkel, Jeffrey (ed.), Intemational Aspects of Fiscal Policy, University of Chicago Press, Chicago. Tesar, Lindar (1991), "Savings, Investment, and International Capital Flows," Joumal of Intemational Economics, 31, 55-78. Wong, Y. David (1990), "What Do Saving-investment Correlations Tell Us about Capital Mobillty? " Joumal of Intemational Money and Finance, 9, 60-74. World Bank (1991), World Tables 1991, Washington, D.C. World Bank (1992), World Tables 1992 (uodate), Washington,D.C. Pollcy Research Working Paper Series Contact Title Author Date for paper WPS1 194 How Fast Has Chinese Industry Tom Rawski September 1993 E. Khine Grown? 37471 WPS1 195 The Enterprise Sector and Mark Schaffer September 1993 E. Khine Emergence of the Polish Fiscal 37471 Crisis, 1990-91 WPS1 196 Corporate Tax Structure and Jeffrey Bernstein September 1993 C. Jones Production Anwar Shah 37754 WPS1 197 Determinants of Inflation among Bruno Boccara September 1993 C. Jones Franc Zone Countries in Africa Shantayanan Devarajan 37754 WPS1 198 Enterprise Reform in China: The Natalie Lichtenstein September 1993 M. Rangarajan Evolving Legal Framework 81710 WPS1199 Public Pension Governance and Olivia Mitchell October 1'93 D. Evans Performance: Lessons for Developing 37496 Countries WPS1200 The Life-Cycle Distributional Jane Falkingham October 1993 D. Evans Consequences of Pay-As-You-Go Paul Johnson 37496 and Funded Pension Systems WPS1 201 Five Criteria for Choosing among Margaret E. Grosh October 1993 M. Ouintero Poverty Programs 37792 WPS1202 Privatization and Foreign Investment Frank Sader October 1993 Rose Vo in the Developing World, 1988-92 33722 WPS1203 Determinants of Value-Added Tax Zeljko Bogetic October 1993 F. Smith Revenue: A Cross-Section Analysis Fareed Hassan 36072 WPS1 204 Structural Adjustment, Economic Nisha Agrawal October 1993 K. Rivera Performance, and Aid Dependency Zafar Ahmed 34141 in Tanzania Michael Mered Roger Nord WPS1205 Wage and Employment Decisions Simon Commander October 1993 0. del Cid in the Russian Economy: An Analysis Leonid Liberman 36303 of Developments in 1992 Ruslan Yemtsov WPS1206 Empirical Perspectives on National Ishac Diwan October 1993 A. Yideru Index Funds Vihang Errunza 36067 Lemma W. Senbet WPS1207 Characteristics and Performance Bill H. Kinsey October 1993 H. Binswanger of Settlement Programs: A Review Hans P. Binswanger 31871 Pollcy Research Working Paper Series Contact Title Author Date for paper WPS1208 Primary School Achievement in Levi M. Nyagura October 1993 I. Conachy English and Mathematics in Abby Riddell 33669 Zimbabwe: A Multi-Level Analysis WPS1209 Should East Asia Go Regional? Arvind Panagariya October 1993 D. Ballantyne No, No, and Maybe 37947 WPS1210 The Taxation of Natural Resources: Robin Boadway October 1993 C. Jones Principles and Policy Issues Frank Flatters 37699 WPS1211 Savings-Investment Correlations Nlandu Mamingi October 1993 R. Vo and Capital Mobility in Developing Countries 31047 WPS1212 The Links between Economic Policy Ravi Kanbur October 1993 P. Attipoe and Research: Three Examples from 526-3003 Ghana and Some General Thoughts