Policy, Research, and External Affairs WORKING PAPERS Pubilc Economics Country Economics Department The World Bank December 1990 WPS 556 Taxing Choices in Deficit Reduction John Baffes and Anwar Shah To control their deficits, Brazil, Mexico, and Pakistan should try to raise revenues and curtail spending simultaneously. In Argentina and Chile, the first priority should be to control public spending. The Policy.Research, and Extmal Affairs Complex distrbutes PRE Wodidng Ppers to disseninatethefindings of work in progrmss and to enourage the exchange of ideas among Bank staff and aU other interested in developincit issues. These papes carry the names of the authors, rdlect only thear views, and should be used and cited accordingly. The findings, inteprttions, and conclusions are the authors own. They shold not be attributed to the Wcrld Bank, its Board of Directors, its ranagement, or any of its mnember countries. Policy, Research, and Extern I Affairs WPS 556 This paper - a product of the Public Economics Division, Country Economics Department - is part of a larger effort in PRE to investigate the causes and consequences of macroeconomic imbalances. Copies are available free from the World Bank, 1818 H Street NW, Washington DC 20433. Please contact Ann Bhalla, room NIO-059, extension 37699 (24 pages). Baffes and Shah use the cointegration approach revenues and spending to control the deficit and to determine whether deficits are more effec- that spending and taxes tend to feed each other in tively controlled by raising taxes or controlling those countries. expenditures - or both. They use long-term historical time series data for Argentina, Brazil, In Argentina and Chile, they found the Chile, Mexico, and Pakistan. deficit to be explosive - and caused by spend- ing. There was no empirical evidence of efforts Many studies have examined causality in the to adjust revenues to control the deficit. relationships between taxes and spending in developed countries. Some have found evidence They recommend that to control the deficit that higher spending tends to lead to higher Brazil, Mexico, and Pakistan should try to raise taxes. Some have found that higher taxes lead to revenues and curtail spending simultaneously. more spending. Some find that causality runs In Argentina and Chile, however, the first both ways. priority should be to control spending. Baffes and Shah find that Brazil, Mexico, and Pakistan have continuously tried to align The PRE Working Paper Series disseminates the findings of work under way in the Bank's Policy, Research, and External AffairsComplex. Anobjective oftheseries is to getthese findings outquickly, 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 PRE Dissemination Center TAXING CHOICES IN DEFICIT REDUCTION by John Baffes and Anwar Shah Table of Contents ,. Introduction ...... 1 II. Theoretical Considerations . . 2 III. Data, Estimation, and Results . . 7 i. Cointegration. 7 ii. Causality .9 iiI. Variance Decompositions ............ ............ 10 iv. Impulse Responses .......... 11 IV. Summary and Conclusions .... .. .... 12 Endnotes ... 14 Tables ................................. 16 Figures ................................. 20 References ...................... 23 The views expressed in this paper are those of the authors alone and should not be attributed to the World Bank. The authors are grateful to Bela Balassa, Johannes Linn, Martha de Melo, Zmarak Shalizi, Javad K. Shirazi, and Vinaya Swaroop for helpful comments. TAXING CHOICES IN DEFICIT REDUCTION I. Introduction Government deficits especially in developing countries are rising at an alarming pace. Further, it is believed that chronic deficits often discourage economic growth, and adversely affect other macroeconomic aggregates. Controlling deficits involves raising taxes, or reducing expenditures. Raising taxes has adverse effects on the private sector and on economic growth in general. Reducing expenditures is also a difficult task because It involves long-run commitments. Further, if public spending is primarily devoted to development of basic Infrastructure, as in many developing countries, then avenues for reducing spending might be quite limited (see Shah, 1990). In either case however, the problem Is that raising taxes may induce higher spending or reducing spending may induce lower taxes, without necessarily affecting the deficit. The latter result obtains if expenditure reduction results in unacceptably low standards of public services and thereby unleash strong anti-public sector sentiment creating political pressures to lower taxes. This paper focuses on qualitative and quantitative effects of spending on revenues and vice-versa for five developing countries. While many studies have examined relationships between taxes and spending In developed countries from a causality point of view, no common agreement exists as to the direction of the causality. For example, Anderson, Wallace and Warner (1986) and Von Furstenberg, Green, and Jeong (1985) have found evidence that higher spending tends to lead taxes. Manage and Marlow (1986) and Ahiakpor and Amirkhalkhali (1989), on the other hand, have found that causality runs the opposite direction. In a more recent study, Miller and Russek (1990) found bidirectional causality between expendltures and revenues for the U.S. (their study Included federal, state, and local level data).- 1 The most Important element which differentiates the present study from previous ones is that it carries out a formal test of whether governments make consistent attempts to align revenues with spending. In doing so we are also able to r.ake inferences as to whether non-tax revenues play an important role in determining the level of the deficit.2 Further, this Is the first study of its kind for developing nations. The specific objectives of this study are: (a) to test whether a long- run relationship between revenues and expenditures exists; (b) If such a relationship exists, what is the direction of the causality; and (c) quantify those causality effects by estimating the error correction representation and subsequently calculating variance decompositions and impulse responses. The paper is structured in the following manner. The next section discusses the theoretical model and the concepts which are required for the development of the test. Section III describes the data, the estimation procedure, and the empirical results. The last section presents conclusions along with some policy implications and directions for future research. II. Theoretical Considerations Consider a government whose objective Is to maximize welfare by choosing the level of public goods and services to be consumed. The instantaneous indirect welfare function, V(p,E), is defined as: (1) V(p,E) a Max (W(z): px = E), x where x denotes che vector of goods and services, p represents the exogenously determined price vector, and E denotes expenditures. W(x) Is a twice continuously differentiable concave welfare function. In order to finance expenditures the government uses revenues, denoted as T. In general it would be expected that T = E so that the government solely covers expenditures 2 through taxation. In most cases however the governments do not operate on a balanced budget schedule so that T * E, and hence, (2) Dt = T - Er. The natural question arising at this moment Is whether Dt represents short-run deviations from zero or whether It consistently deviates from zero even in the long run.3 An intuitive way to test whether revenues and expenditures drift apert In the short run only, would be to form the regression (3) Tt = 90 + gIEt +Ct where 10 and 3 denote parameters to be estimated. Then test Ho $0 = C' and 1li = 1 against HI: o0 * O and P I 1, where acceptance of Ho would Imply that the government has been making consistent attempts to equate revenues to spending, at least in the long run. Notice that if H Is true, (3) collapses to (2) where t = Dt, which means that the government finances expenditures entirely through revenues. This test however presents some shortcomings. First, the test is very restrictive in the sense that it Is rather unlikely for the governments to run balanced budgets on an annual basis, so H Is likely to be rejected. Second, this procedure fails to take into consideration certain properties of time series variables which sometimes may Invalidate standard regression results, namely that the variables being considered are stationary. To circumvent those shortcomings we take the alternatlive of testing wnether Tt and Et are cointegrated. Notice that cointegration not only takes Into account stationarity properties of the variables being considered, but also It examines whether T and Et move together in the long run, allowing for short- run deviations.4 Cointegration requires that all variables are of the same order of 3 integra.lon. Let ; first give an intuitive-explanation of the concept of the order of Integration. If a series has a finite mean and variance It is called integrated of order zero, and is denoted as I(O). If the series needs to be differenced once to become I(O), it is then called integrated of order one, I(1). In general a series that is required to be differenced d times to become I(O) is called I(d). If two series are I(d) and there exists a linear combination of those series which is I(b) with b < d, then the series are to be cointegrated, denoted as CI(d,d-b). For practical purposes we generally consider I(1) series since most economic variables become I(O) after being differenced once and hence the cointegrated system is CI(1,1). Ir what follows, the terms stationarity (or stationarity in levels) and I(O) will be used interchangeably. First, the order of integration of the variables under consideration must be determined. Three prominent procedures to determinr tile order of Integration are: (a) Dickey-Fuller (DF), (b) augmented Dickey-Fuller (ADF), and (c) Durbin-Watson (DW). The'DF test is based on the regression: AXt = A + 1lXt + et, where Xt denotes the variable of interest and a denotes the difference operator; p and $ denote parameters to be estimated. The null hypothesis (H ) is: X is not I(O). Ho is rejected if the estimate of t3 is negative and significantly different from zero. The ADF test is based on: AXt =i+ 3X + E r 7 AX + c , where xr is selected so that e is white t-I 1=1 I tI t . noise; I, $, and V denote parameters to be estimated as before. Again Ho is rejected if $ is negative and significantly different from zero. Finally, the DW test is based on the Durbin-Watson statistic of: Xt = I + £t. Low DW t. ~~t statistic indicates that X is not I(O). t If the variables of interest are all say I(1), to test for cointegration we regress one variable on the other and then test whether the estimated residual is I(O). In other words we estimate (3) and then test for 4 stationarity of ce. Notice that since one of the primary objectives of this study Is to test whether the government m.kses consistent attempts to equate revenues with spending, we first test for stationarity of Dt which implicitly imposes the balanced budget constraint as a long-run restriction.5 Therefore, testing for coir.tegration between Et and Tt assuming that a long-run relationship between revenues and spending exists, becomes a simple unit root test In the univariate process (Engle and Yoo, 1987). In the case that Dt is not I(O) we proceed to estimate (3). Notice that since the cointegration parameter is unique in the bivariate case, if we find cointegration by restricting 10 = 0 and $ = 1 then the regression in (3) should produce tl.e same outcome .6 As a second step we test to see whether there exists causality between expenditures and revenues (Granger, 1969). Notice that if the lndividual series are I(1) (which Is the case as it will be shown in the next section) we have to take differences to induce stationarity. Hence, we estimate the following relationship: i. T (4a) AT + AEt = + E aIATt-I + E $t Et i + Ut. ~~~~~ t (4b) AE~t + ATt v + EIAEt + E AT t + V, where 1, of, a .t, v, al, 1l, I, and S denote parameters to be estimated; 'r denotes the number of lags which is not necessarily tne same for all variables. ut and vt are assumed to be mutually uncorrelated white noise processes. If $0 = a = 0 and some 's and 8 's have non-zero values, (4a-4b) implies a simple causal relation with feedback (i.e. simple bidirectional causality). If a0 0 O and 60 0 O and some Si's and 6I's have non-zero values, then (4a-4b) implies instantaneous bidirectional causality. Finally, unidirectional causality is implied if the above relations hold for 5 one equation only. In terms of Tt and Et. simple causality means that past Et only affects Tt, while instantaneous causality means that both past and current Et affect Tt and vice-versa. Such causal relationships can be detected with conventional F-tests. One other Important element we consider in this study Is the relation between cointegration and causality. If the variables being considered are cointegrat_d, there exists causality in at least one direction (Granger, 1986). Further, an additional implication of cointegration is that if there exists cointegration then the system can be represented by an error correction mechanism (ECM).7 The implication of such representation is that we quantify the causality effects by constructing the Vector Autoregressive (VAR) representation of the bivariate system as defined by Tt and Et and also incorporate the long-run relationship as follows (for a complete characterization of VAR processes see Sims (1980)): T T (5a) ATt = v 0Dt_ + E (Xi AT + 1fE +,t- to (5b) AE = 8 D + F XAT + E 8AE + V where Dt 1 is the lagged level of deficit. The remaining variables and parameters are defined in (4a-4b). It Is interesting to notice that if we replace AEt and ATt by Dt In the causality regressions (4a-4b) we arrive at the ECM representation; so cointegration unifies ECM and conventional. causality models. (Sa-Sb) is sometimes called restricted VAR, where the restriction is the residual from the cointegration regression (in this case the observed deficit or the error term of (3)). The advantage of the ECM as opposed to the unrestricted VAR is that by including the deiicit in the equations we retain information in levels, without distorting the stationarity properties of the variables involved in the system, since, because of 6 cointegrati.on Dt_1 Is I(0). From the estimated VAR system we can then calculate Impulse responses and variance decompositions as means of quantification of the causality effects. III. Data, Estimation, and Results Data In the current study Include total government expenditures and total revenues for the countrien of Argentina, Brazil, Chile, Mexico, and Pakistan. The data series for Argentina cover the 1904-1983 time period and were obtained from unpublished World Bank Tables. Data for Brazil cover the 1905-1983 period and were obtained from Estatisticas Historicas do Brasil. Data for Mexico cover the 1895-1984 time period and were obtained from Estatisticas Historicas de Mexico. Data for Pakistan cover the 1947-1989 period and were obtained from various publications of the Central Statistical Office, Government of Pakistan. Finally, data for Chile cover the 1960-1985 time period and were obtained from Banco Central De Chile. Because of some missing observations, the data set for Argentina and Mexico contain 71 and 78 observations respectively. All series were adjusted by the respective GDP deflators in view of the high inflation rates experienced by the Latin Arerican countle3 Included in this study. The .ATS package was used to estimate the models. The remainder of this section, which is divided in four parts, will discuss and analyze results regarding cointegration (Table 1), causality (Table 2), as well as variance decompositions (Table 3) and impulse responses (Figures 1, 2, and 3). (1) Cointegration The first step regarding cointegration is the determination of the order of integration of revenues and spending. Table 1 reports such results. In all 7 cases both revenues and spending were found to be nonstationary and hence the first differences had to be considered. Stationarity tests for the first differences Indicated that both variables In all countries were I(1). The second step was to determine the order of integration of D . For Argentina all three tests indicated that Dt is nonstationary (D% level of significance). This means that the Argentinean government ha- not been able to equate revenues and spending In the long run. The next step was to test for cointegration by considering the unrestricted version of the cointegration regression (i.e. relationship (3)). All three tests indicated that there exists strong evidence of cointegration. (The cointegration regression - not reported here - gave a coefficient of 0.65). To summa.ize, cointegration tests for Argentina indicate that although revenues and spending move together In the long run there Is an unexplained component in spending since revenues account for 65% of spending only. Cointegration tests for Brazil showed evidence of cointegration. In particular the DW and ADF test indicate mild evidence of cointegration while the DF test Indicates strong evidence. In what follows we consider that revenues and spending in Brazil are cointegrated. For Chile, all three cointegration tests showed that revenues and spending are not cointegrated (5% level of significance). The conclusion was the same even when relationship (3) was estimated (estimates not reported here). The implication of such result is that there exist a nonstationary component (such as seigniorage or borrowing) that causes explosiveness of the deficit which is not being taken into censideration. For Mexico, the results are the same as in Brazil. Notice that all three tests supported stationarity of the deficit at the 1% level of significance. For Pakistan the results are almost Identical. The only differencc is that the results hold for the 5% level of significance. Notice that the ADF test 8 for Pakistan indicates that Et and Tt are I(2). However, the deficit Is still I(O), thus supporting cointegration. (ii) Causality Table 2 reports causality results. In order to conserve space we do not report parameter estimates and other statistics of the system. To facilitate comparability with other studies we report results for all three types of causality (i.e. simple causality, instantaneous causality, and causality with the cointegration restriction Imposed). For all three countries for which cointegration was found instantaneous causality runs both directions (i.e. revenues cause spending and spending causes revenues). Notice that the finding of at least one direction causality is consistent wlth the existence of cointegration between revenues and spending foi those three countries. On the other hand, while simple causality runs from spending to revenues for Pakistan no simple causality in any direc.tion was found for Brazil and Mexico. When the colntegration restriction is imposed, Mexico exhibits bidirectional causality. In Brazil, revenues cause spending while the opposite is true for Pakistan. Since cointegration was not found for Argentina and Chile, the possibility that the deficit causes (or is caused by) expenditures and/or revenues was examined. Thus, we tested all possible causal relations among T, E, and D. Specifically, the results for Argentina show that Instantaneous causality runs in all directions (i.e. among deficit, revenues, and spending). On the other hand, simple causality runs from spending to revenues and from deficit to revenues. It is of interest to notice that while simple causality was found in a few cases only, instantaneous ca"isality was found in almost any case examined. The implication of this is that decisions to reduce/increase spending are being made simultaneously with decisions to Increase/reduce revenues. 9 Finally, it should be noticed that the qualitative nature of the causality results did not change wnen other lag structures were considered. Single exceptien to this constitutes Pakistan, in which case ihen more than two lags were considered no causality was found in any direction. (iii) Variance Decompositions Variance decompositions exhibit the contribution of each source of innovation to the variance of the k-year ahead forecast error for each of the variables. Stated otherwise, variance decompositions refer to a breakdown of the change in the value of the variable in a given year arising from changes in the same variable as well as other variables in previous years. Table 3 gives estimates of variance decompositions for Brazil, Mexico, and Pakistan. As it was mentioned earlier, for Chile and Argentina T and E do not fully describe the deficit so we did not form the ECM representation. The results can be summarized as follows. 95% of the variation of expenditures in Mexico is accounted by past expenditures, while only 5% is accounted by past revenues. This pattern seems to be consistent through the whole period examined. The same picture is presented when we consider revenues, i.e. 96% of the variation in revenues is accounted by past expenditures while 4% only is accounted by past revenues. In Brazil, changes in expenditures account for most of the variation in future expenditures as in the case of Mexico. On the other hand, changes in revenues and expenditures equally account for the variance of revenues. Again this pattern is consistent through the whole period examined. In Pakistan, changes in expenditures account for most of the variation in future expenditures. On the other hand, changes in revenues account for abcut 75% of the variation in future revenues throughout the period examined. To summarize, it appears that in most cases expenditures account for most of the variation in both future expenditures and future revenues. 10 (iv) Impulse Responses Impulse responses give the dynamic response of each variable to policy changes affecting this variable as well as of the ocher variables included In the system. In other words, impulse responses describe whether a shock of one variable has a persistent or transitory effect on the other variables as well as on the variable itself. Figure 1 depicts impulse responses for Brazil. In general all four impulse responses presert similar picture. A one-standard deviation shock on spending induces more spending as well as more revenues in the first period while after the third period both spending and revenues return to the pre-shock level. The same picture is presented when we consider the own-effect of revenues. On the other hand revenue shock has no effect on spending in the first period, while after a negative effect in the second period spending returns to Its pre-shock level. Figure 2 glves impulse responses for Mexico. The effect of a shock of spending on both spending and revenues as well as the effect of revenue shock on revenues are similar: after an increase In the first period the variables tend to return to the pre-shock levels following an oscillatory process. Revenue shock on spending however has no effect in the first period, while after a negative effect in the second period it oscillates before it returns to its pre-shock level. Figure 3 gives impulse responses for Pakistan. Spending shock has the same effect on both revenues and spending. I.e. there is a large positive effect in the first period. After that, the variables have a tendency to return to the pre-shock levels. To some extent spending responds the same way to revenues shock. Comparing Pakistan with Brazil and Mexico however we observe that in the Pakistani case the shocks tend to be persistent, that is, it takes longer for the variables to return to the pre-shock level. This result is consistent with the fact that for Brazil no simple causality but 11 strong instantaneous causality was found while both types of causality were found for Pakistan. To summarize, it seems that both Brazil and Mexico present the same picture in the sense that the shocks have short run effects only. On the contrary, in Pakistan the shocks have persistent effects. IV. Summary and Conclusions In this paper an attempt was made to: (a) determine whether governments have continuously attempted to align revenues or spending to control the deficit. (This was done by testing whether there exists cointegration between revenues and spending); (b) test for causality between taxes and spending; and (c) quantify the causality effects by (f) estimating an error correction model and (ii) calcuiating variance decompositions and impulse responses. The tests were carried out for the countries of Argentina, Brazil, Chile, Mexico, and Pakistan. The availability of long data series primarily determined the selection of those countries. The results can be summarized as follows: The governments of Brazil, Mexico, and Pakistan seem to have successfully aligned revenues and spending as means of controlling the deficit over the time period examined while a similar deduction for Argentina and Chile could not be made. For Brazil, Mexico, and Pakistan strong instantaneous causality runs both directions. In Argentina and Chile deficit was found to cause and be caused by expenditures. Impulse responses for Mexico and Brazil were found to have short-run effects only while for Pakistan the effects were more persistent. In terms of variance decompositions it was found that variations In both revenues and spending are explained in most part by past spending. The above results suggest that to control the deficit, Brazil, Mexico, and Pakistan should attempt to raise revenues and curtail expenditures simultaneously. In 12 Argentina and Chile, on the other hand, controlling public expenditures should be the first priority. An important qualification Is in order here. The results stated earlier, are based on long term relationships that may differ from the present situation. A case in point is Chile, that has succeeded in eliminating the budget deficit since the end of the period of observations in our sample. In contrast, the deficit has increasedl in Argentina and Brazil in more recent years. 13 ENDNOTES Other studies Include Furstenberg, Green, and Jeong (1986); Ram (1988a, 1988b); Ahsan, Kwan, and Sahni (1989); Holtz-Eakin, Newey, and Rosen (1989); Miller and Russek (1990). In particular Miller and Russek tested for causality by imposing the cointegration restriction as defined in (3) of this study. The present study makes an explicit distinction between the cointegration restriction defined in (2) and the cointegration restriction defined In (3). 2 An Important element of non-tax revenue, especially in Latin America, is seigniorage (i.e. inflation tax). See Kiguel and Neumeyer (1989) for a treatment of seigniorage in Argentina. 3 To be precise, the government is constrained to run balanced budget In present value terms. So the constraint to test would be, 7 o(1+8) t lT, - co (1+8) Et = 0, where a denotes the discount rate. We chose to test the balanced budget constraint associated with (2) because It is convenient to form the ECM representation and subsequently carry out the causality tests. See Hamilton and Flavin (1986) for a test regarding the balanced budget restriction In present value terms. 4 Ihe theory of cointegration Is discussed extensively in Engle and Granger (1987), Hendry (1986), and Granger (1986). 5 Several studies have considered theoretical restrictions In testing cointegration. For example, Campbell and Shiller (1987) tested the term structure of the interest rate; Corbae and Oullaris (1988) tested the Purchasing Power Parity. Other studies of similar nature include Hall (1986) and Ambler (1989). 14 6 It is true that the cointegratlon regression gives an excellent estimate of the cointegratlon parameter. However this Is the case In large samples only. In small samples the regression does not guarantee that the cointegration parameter will be found. The equivalence between cointegration and ECM comes directly from Granger representation theorem (Engle and Granger, p. 225). The ECM type of models were first Introduced by Phillips (1957). 15 TABLE 1: Stationarity Tests DW DF ADF ARGENTINA (71 observations) Et 0.075 -0.396 -0.511 AE 1.939 -8.030 -5.786 Tt 0.105 -1.171 -0.946 AT 2.581 -11.429 -6.199 Dt 0.352 -1.975 -2.047 BRAZIL (78 observations) Et 0.038 0.861 1.926 AEt 2.463 -11.179 -7.816 T 0.028 1.430 2.608 t ATt 2.481 -11.165 -7.170 Dt 0.625 -3.621 -2.977 CHILE (26 observations) Et 0.318 -1.292 *-1.362 AEt 1.979 -4.638 -3.224 T 0.114 -0.867 -1.008 ATt 1.402 -3.418 -3.196 D t 0.636 -2.046 -2.470 MEXICO (78 observations) E t 0.053 0.381 0.206 AEt 1.856 -7.994 -4.889 Tt 0.089 -0.647 -0.388 ATt 2.262 -9.817 -3.174 Dt 0.963 -4.091 -8.289 continued 16 PAKISTAN (42 observations) Et 0.024 4.018 3.442 AE 1.358 -4.420 -2.872 t T 0.023 6.066 4.744 t AT 1.047 -2.921 -1.027 t. D 0.783 -3.081 -3.257 t. NOTE: Et and Tt denote spending and revenues; Dt denotes deficit; A represents the difference operator (i.e. AEt = Et - Et t and AT = T - T ). The critical values for Argentina, Brazil, and Mexico are: DW = 0.259 (5%) and 0.376 (1%) (from Sargan and Bhargava (1983), Table 1); DF and ADF = -2.89 (5%) and -3.59 (1%) (from Fuller (1976), TA statistic, Table 8.5.2). The critical vOlues for Pakistan are: DW = 0.493 (5%) and 0.705 (1%) (from Sargan and Bhargava (1983), Table 1); DF and ADF = -2.93 (5%) and -3.58 (1%) (from Fuller (1976), T statistic, Table 8.5.2). The critical values for Chile are: DW = 0.770 (5%) and 1.081 (1%) (from Sargan and Bhargava (1983), Table 1); DF and ADF = -3.00 (5%) and -3.75 (1%) (from Fuller (1976), TA statistic, Table 8.5.2.). These statistics are based on Monte Carlo experiments made on 100, 50, and 25 observations, respectively. The number of lags in the ADF test was determined by the Akaike information criterion. 17 TABLE 2: Causality Tests ARGENTINA CHILE I II I II T does not cause E 0.07 6.83 0.18 0.46 E does not cause T 6.47 10.64 0.05 0.39 T does not cause D 0.10 28.68 0.91 1.27 D does not cause T 6.47 34.58 0.05 0.83 D does not cause E 0.07 5.49 0.18 40.02 * * E does not cause D 0.10 5.51 0.91 41.74 BRAZIL I II III T does not cause E 0.08 32.67 4.54 E does not cause T 0.06 32.65 0.28 MEXICO I II III T does not cause E 0.01 660.41 13.65 E does not cause T 0.15 661.78 6.30 PAKIST.' I II III T does not cause E 2.65 4.05 2.34 E does not cause T 3.25 4.38 3.52 NOTE: I means simple causality, II means instantaneous causality, while III means causality with the cointegration restriction imposed. One star (*) Indicates rejection of no causality at the 5% level of signlflcance while two stars (**) Indicate rejection of no causality at the 1% level. All tests were carried out with one lag. 18 TABLE 3: Proportion of Forecast Error Variance K Periods Ahead Produced by Each Innovation. Ixinovation In: MEXICO BRAZIL PAKISTAN Error in K E T E T E T E 1 1.000 .000 1.000 .000 1.000 .000 2 .946 .054 .966 034 .955 .045 3 .952 .048 .961 .039 .939 .061 4 .949 .051 .961 .039 .935 .065 5 .950 .050 .961 .039 .934 .066 6 .950 .050 .961 .039 .933 .067 7 .950 .050 .961 .039 .933 .067 8 .950 .050 .961 .039 .933 .067 9 .950 .050 .961 .039 .933 .067 10 .950 .050 .961 .039 .933 .067 T 1 .966 .U34 .496 .504 .207 .793 2 .954 .046 .494 .506 .250 .750 3 .954 .046 .494 .506 .262 .738 4 .953 .047 .494 .506 .265 .735 5 .952 .048 .494 .506 .266 .734 6 .952 .048 .494 .506 .267 .733 7 .952 .048 .494 .506 .268 .732 8 .952 .048 .494 .506 .267 .733 9 .952 .048 .494 .506 .267 .733 10 .952 .048 .494 .506 .267 .733 NOTE: K indicates years. E and T denote spending and revenues respectively. One lag was used to estimate the ECM model. No major differences in the results were observed when other lag lengths were considered. 19 SPENDING SHOCK/SPENDING RESPONCE SPENDING SHOCK/REVENUE RESPONSE 8- 4 3 4 2 21 -2 -2 1 2 3 4 5 e 7 8 9 10 1 2 3 4 5 6 7 8 9 10 N REVENUE SHOCK/SPENDING RESPONSE REVENUE SHOCK/REVENUE RESPONSE 5 5 4 0 3 2 -6 -10~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ - 10~~~~~~~~~~~~~~~~~~~~~- -15 -2 1 2 3 4 5 8 7 8 9 10 1 2 3 4 5 6 7 8 9 10 FIGURE 1: Brazil SPENDING SHOCK/SPENDING RESPONSE SPENDING SHOCK/REVENUE RESPONSE 30 -40 20 30 20 10 10 0 0 -10-1 - 10 -20 -20 -30~~~~~~~~~~~~~~~~~~2 _;30 -30l 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 e 7 8 9 10 N} REVENUE SHOCK/SPENDING RFSPONSE REVENUE SHOCK/REVENUE RESPONSE 4 8 26 0 ~~~~~~~~~~~~~~~~~~~~~4 2 -2 -4 -2 -4 -8 -6 -10 -8 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 FIGURE 2: Mexico SPENDING SHOCK/SPENDING RESPONSE SPENDING SHOCK/REVENUE RESPONSE 26 10 20 8 16 6 10 4 5 2 1 3 4 6 6 8 9 10 1 2 3 4 S 7 8 9 10 N REVENUE SHOCK/SPENDING RESPONSE REVENUE SHOCK/REVENUE RESPONSE 20 S 15 4 3 10 2 0- 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 FIGURE 3: Pakistan REFERENCES Ahiakpor, J.C.W. and S. Amirkhalkhali. "On the Difficulty of Eliminating Deficits with Higher Taxes." Southern Economic Journal. 56(1989):24-31. Ahsan, S.M., A.C.C. Kwan, and B.S. Sahni. "Causality between Government Consumption Expenditure and National Income: OECD Countries." Public Finance. 44(1989):205-224. Ambler, S. "Does Money Matter In Canada? Evidence from a Vector Error Correction Model." Review of Economics and Statistics. LXXX(1988): 651-658. Anderson, W., MOS. Wallace, and J.T. Warner. "Government Spending and Taxation: What Causes What?" Southern Economic Journal. 52(1986): 630-639. Banco Central De Chile. Indicatores Economicos y Sociales: 1960-1985. Direccion de Estudios, Chile, 1985. Campbell, J.Y. and R.J. Shiller. "Cointegration Tests of Present Value Models." Journal of Political Economy. 95(1987):1061-1088. Central Statistical Ozfice, Government of Pakistan. 25 Years of Pakistan In Statistics: 1947-1972. Karachi, 1972. Also supplementary issues. Corbae, D. and S. Ouliaris. "Cointegration and Tests of Purchasing Power Parity." Review of Economics and Statistics. LXX(1988):508-511. Engle, R.F. and C.W.J. Granger. "Co-Integration and Error Correction: Representation, Estimation and Testing." Econometrica. 55(1987): 251-276. Engle, R.F. and B.S. Yoo. "Forecasting and Testing in Co-integrated Systems." Journal of Econometrics. 35(1987):143-159. Estatisticas Historicas do Brasil. Series Estatisticas Retrospectivas, Vol 3. Fundacao Instituto Brasileiro de Geografia e Estatistica, 1987. Estadisticas Hist;ricas de Mexico. Tomo I, II. Instituto National De Estadisticas Geografia e Informatica. 1986. Fuller, W.A. Introduction to Statistical Time Series. John Wiley & Sons, New York, 1976. Granger, C.W. "Investigating Causal Relations by Econometric Models and Cross-Spectral Methods." Econometrica. 37(1969):424-438. Granger, C.W. "Developments in the Study of Cointegrated Economic Variables." Oxford Bulletin of Economics and Statistics. 48(1986):213-228. Hall, S.G. "An Application of the Granger & Engle Two-Step Estimation Procedure to United Kingdom Aggregate Data." Oxford Bulletin of Economics and Statistics. 48(1986y:229-240. 23 Hamilton, J.D. and M.A. Flavrin. "On the Limitations of Government Borrowing: A Framework for Empirical Testing." American Economic Review. 76(1986): 808-819. Hendry, D. "Econometric Modelling with Cointegrated Variables: An Overview." Oxford Bulletin of Economics and Statistics. 48(1986):201-212. Holtz-Eakin, D., W. Newey, and H.S. Rosen. "The Revenues-Expenditures Nexus: Evidence from Local Government Data." International Economic Review. 30(1989):415-429. Kiguel N.A. and P.A. Neumeyer. "Inflation and Seigniorage In Argentina." PPR Working Paper. Macroeconomic Adjustment and Growth Division, Country Economics Department, The Woirld Bank, WPS 289, October 1989. Manage, N. and M.L. Marlow. "The Causal Relation between Federal Expenditures and Receipts." Southern Economic Journal. 52(1986).617-629. Miller, S.M. and F.S. Russek. "Co-Integration and Error-Correction Models: The Temporal Causality between Government Taxes and Spending." Southern Economic Journal. 57(1990):221-229. Phillips, A.W. "Stabilization Policy and the Time-Forms of Lagged Responses." Economic Journal. 67(1957):265-277. Ram, R. "Additional Evidence on Causality between Government Revenue and Government Expenditure." Southern Economic Journal. 54(1988a):763-769. Ram, R. "A Multicountry Perspective on Causality Between Government Revenue and Government Expenditure." Public Fiz- nce. 43(1988b):261-269. Sargan, J.D. and A. Bhargava. "Testing Residuals from Least Squares Regression for Being Generated by the Gaussian Random Walk." Econometrica. 51(1983):153-174. Shah, A. "Dynamics of Public Infrastructure and Private Sector Productivity." Paper presented at the 6th World Congress, Econometric Society, Barcelona, Spain, 1990. Sims, C. "Macroeconomics and Reality." Econometrica. 48(1980):1-49. Von Furstenberg, G.M., R.J. Green, and J-H. Jeong. "Have Taxes Led Government Expenditures? The United States as a Test Case." Journal of Public Policy. 5(1985):321-348. Von Furstenberg, G.M., R.J. Green, and J-H. Jeong. "Tax and Spend, Or Spend and Tax." Review of Economics and Statistics. LXVIII(1986):179-188. 24 PRE Working Paper Series Contact Aor Da for pape WPS540 Venture Capital Operations and Silvia Sagari Their Potential Role in LDC Markets Gabriela Guidotti WPS541 Pricing Average Price Options for the Stijn Claessens November 1990 S. King-Watson 1990 Mexican and Venezuelan Sweder van Wijnbergen 31047 Recapture Clauses WPS542 The Metals Price Boom of 1987-89: Boum-Jong Choe November 1990 S. Lipscomb The Role of Supply Disruptions and 33718 Stock Changes WPS543 Development Assistance Gone Donald B. Keesing November 1990 S. Fallon Wrong: Why Support Services Have Andrew Singer 37947 Failed to Expand Exports WPS544 How Support Services Can Expand Donald B. Keesing November 1990 S. Fallon Manufactured Exports: New Andrew Singer 37947 Methods of Assistance WPS545 Health and Development: What Nancy Birdsall November 1990 L. Mitchell Can Research Contribute? 38589 WPS546 The Transition to Export-Led Growth Stephan Haggard November 1990 E. Khine in South Korea, 1954-66 Byung-Kook Kim 39361 Chung-in Moon WPS547 Does High Technology Matter? An Andrea Boltho November 1990 M. Hileman Application to United States Regional Robert King 31284 Growth WPS548 Deposit Insurance in Developing Samuel H. Talley November 1990 M. Pomeroy Countries Ignacio Mas 37666 WPS549 Intertemporal Substitution in a Patricio Arrau December 1990 S. King-Watson Monetary Framework: Evidence 31047 from Chile and Mexioo WPS550 Firms' Responses to Relative Price John L Newman December 1990 A. Murphy Changes in C6te d'lvoire: The Victor Lavy 33750 Implications for Export Subsidies Raoul Salomon and Devaluations Philippe de Vreyer WPS551 Australia's Antidumping Experience Gary Banks December 1990 N. Artis 37947 WPS552 Selected World Bank Poverty Nancy Gillespie December 1990 M. Abiera Studies: A Summary of Approaches, 31262 Coverage, and Findings WPS553 Money, Inflation, and Deficit in Egypt Marcelo Giugale December 1990 V. Israel Hinh T. Dinh 36097 l PRE Working Paper Series Contact b Author 12am. for glaAer WPS554 Korea's Labor Markets Under Dipak Mazumdar December 1990 M. Schreier Structural Adjustment 36432 WPS555 The Macroeconomics of Price Reform Simon Commander December 1990 0. del Cid in Socialist Countries: A Dynamic Fabrizio Coricelli 39050 Framework WPS556 Taxing Choices in Deficit Reduction John Baffes December 1990 A. BhalIa Anwar Shah 37699 WPS557 The New Fiscal Federalism in Brazil Anwar Shah December 1990 A. Bhalla 37699 WPS558 Alternative Instruments for Kenneth M. Kletzer December 1990 J. Carroll Smoothing the Consumption of David M. Newbery 33715 Primary Commodity Exporters Brian D. Wright