wPS ;-514 POLICY RESEARCH WORKING PAPER 2514 Intersectoral Dynamics The frequent recommen- dation to exclude the oil and Economic Growth sector from economic analysis in Ecuador may be short-sighted, as adverse shocks to the oil industry are likely to affect Norbert M. Fiess other sectors through the Dorte Verner financial and public sectors, with which the oil sector has many links. There are also significant long-run relationships between the agricultural, industrial, and service sectors. The World Bank Latin America and the Caribbean Region Economic Policy Sector Unit January 2001 I POLICY RESEARCH WORKING PAPER 2514 Summary findings Fiess and Verner analyze sectoral growth in Ecuador improved understanding of intersectoral dynamics may using multivariate cointegration analysis. They find facilitate the implementation of policy aimed at significant long-run relationships between the increasing economic growth in Ecuador. agricultural, industrial, and service sectors. Moreover, There appears to be no direct link between the oil they are able to derive dynamic sector models that sector and the non-oil industrial sectors. But strong combine the short-run links between the three sectors evidence supports cointegration between the oil industry with long-run dynamics. and financial services as well as between the oil industry When they disaggregate the three sectors into their and public services. This means, among other things, that intrasectoral components, they discover many interesting the oil sector cannot be excluded from intersectoral relationships that contribute to a better understanding of growth analysis, because an adverse shock to the oil inter- and intrasectoral dynamics in the context of industry is likely to affect other sectors through the Ecuadorian economic growth. financial sector, the public sector, or both. Their findings suggest that more attention should be paid to interdependencies in sectoral growth, since an This paper-a product of the Economic Policy Sector Unit, Latin America and the Caribbean Region-is part of a larger effort in the region to investigate intersectoral growth dynamics. Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Anne Pillay, room 18-154, telephone 202-458-8046, fax 202- 522-2119, email address apillay@worldbank.org. Policy Research Working Papers are also posted on the Web at www.worldbank.org/research/workingpapers. The authors may be contacted at nfiess@worldbank.org or dverner@worldbank.org. January 2001. (29 pages) The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carty the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Produced by the Policy Research Dissemination Center Intersectoral Dynamics and Economic Growth in Ecuador Norbert M. Fiess Dorte Verner World Bank 1818 H Street Washington, D.C. 20433 U.S.A. nfiess@worldbank.org dvemer@worldbank.org The authors would like to thank Ana Lucia Armijos, Paul Beckernan, Hugh Blunch, Eliana Cardoso, David Yuravlivker and an anonymous referee for helpful comments and suggestions. 1. Introduction The revival of growth theory during the past 15 years has also led to an expansion in empirical work on economic growth over the last decade. While the main focus in the empirical growth literature is on the determinants of aggregate economic growth, less emphasis has so far been placed on sectoral economic growth. The sectoral growth literature builds mainly on the dual economy model originating in Lewis (1954), Fei and Ranis (1961) and Sen (1966). This model seeks to explain economic growth by emphasising the roles of agriculture and industry and the interplay between them.' The dual economy model views the agricultural sector as the basis of an emerging economy, a generator of the capital necessary for take-off towards the second stage of economic development: industrialisation. Once industrialisation has taken place, the agricultural sector becomes gradually a mere appendage to the economic system, with no internal economic integration and a low degree of intersector linkages. The dual economy literature generally rules out two major issues about the later stages. First, the literature denies that agriculture may be an important growth-promoting factor. Second, it rules out feedback mechanisms between agriculture and industry. Recent developments in the sectoral growth literature dispute this view of the dual economy model. Mellor and Lele (1970), Mellor (1972), Johonston and Kilby (1975) argue that a virtuous cycle between agricultural intensification and non-agricultural activity could emerge on the basis of production and consumption linkages. An increased demand of farmers for inputs such as machinery and machinery repair can stimulate non- agricultural activity through backward linkages. Non-agricultural activity could be stimulated by agriculture at the same time via forward linkages such as the requirement to process agricultural products through spinning, milling or canning. Gopinath, Roe and Shane (1996) analyse the possible link between agriculture and food processing and find that productivity gains in agriculture feed back into the food processing industry, where they lead to cheaper inputs. Lower priced inputs lead in turn to increased derived demand for primary agricultural products, thus partly mitigating the price decline. The two sectors evolve interdependently over time, contrary to what the dual economy model predicts. 1 For references to studies that build upon the framework along the lines of the classical dualistic framework, see Blunch and Verner (1999). 2 Blunch and Vemer (1999) present evidence from a sectoral growth analysis in three African countries and find long-run relationships and short-run causality between the industrial, agricultural and service sectors. The empirical evidence of high interdependence between agriculture and other economic sectors is interesting, since agriculture is generally assigned a low degree of sector interdependence and thus rarely seen as a key sector for economic development (Chenery and Watanabe (1958) and Hirschman (1961)). The agricultural sector with its exports of bananas, shrimps cocoa and has traditionally played an important part in the economic development of Ecuador. In the 1990s cut flowers also became an important export commodity. The recent experience with the trade liberalisation, implemented at the end of the 1980s, indicates further that Ecuador's export diversification has been dominated by processed goods which are intimately related to natural resources such as marine products or raw or processed food products and not by industrial exports (Michaely, 1999). Since this development in particular underlines the high degree of interdependence between agricultural and industrial output for the Ecuadorian economy, it is the aim of this paper to quantify these intersector dynamics. The emphasis in this paper is mostly on identifying intersector growth dynamics using advanced data analysis techniques and less on explaining the determinants of growth. Identifying main lines of causality and intersector linkages can help policy makers to obtain a better understanding of the economic growth process in Ecuador and to formulate more effective development strategies. It also provides useful information for future economic modeling of intersector growth. The remainder of this paper is organised as follows. Section 2 presents a brief country profile of Ecuador and attempts to place the present analysis into an historic context. Section 3 describes the data and the econometric methodology. Section 4 presents the empirical findings and, finally, section 5 presents the main conclusions. 3 2. A country profile of Ecuador The nationalisation of the oil industry in 1972 and the oil boom in the 1970s turned Ecuador from a poor, primary-export dependent economy into a middle-income country with a stock of wealth in the form of oil reserves. The industrialisation strategy of the 1970s was highly protective in nature and led to a capital-intensive industry, which produced inefficiently when compared internationally. As a result, most capital goods, for the purpose of investment, as well as most intermediate goods were imported, while the domestic capital goods production remained small and concentrated in low-technology intensive processes. In the years prior to the debt crisis Ecuador imported more than two-thirds of all installed machinery (see Hentschel, 1994). The outbreak of the debt crises in 1982, which halted international capital flows to most developing countries, brought for Ecuador a high degree of macroeconomic instability, which persisted through the majority of the 1980s. The economy was further disrupted by a major earthquake in 1987, which destroyed the national oil pipeline and halted oil exports for 5 months. The end of the 1980s brought a change in the development model towards export- diversification and trade liberalisation, with the result of a rapid and comprehensive trade liberalisation between 1989 and 1992, the adoption of the common external tariff of the Andean Group and the opening of the frontier with Colombia. The late 1980s and early 1990s also saw a substantial reduction in public consumption, the elimination of many implicit and explicit state subsidies and a liberalisation of interest rates (see Marconi and Samaniego, 1995). Even though Ecuador's economy is highly concentrated, with oil, bananas and shrimps representing the major export commodities, the recent trade liberalisation brought a slight change in the structure of Ecuador's exports. While the three major export commodities accounted for 85% of aggregate exports in the late 1980s and early 1992, their combined share dropped to 70% in 1996-97 (Michaely, 1999). According to Michaely (1999), the export diversification of Ecuador has been dominated by processed goods which are intimately related to natural resources such as marine products or raw or 4 processed food products and not by industrial exports.2 This underlines the general importance of the agricultural sector for the development of other sectors and as a potential source of growth in Ecuador. Table 1: Contribution to GDP in percentage share of total GDP 1965 1970 1973 1980 1990 1998 Agriculture 25.78 24.97 18.09 14.36 17.67 17.28 Industry 22.30 24.19 39.22 33.80 31.74 32.96 - Oil 19.38 10.21 11.81 13.52 - Manufacturing 15.23 17.17 14.11 18.16 15.45 15.48 - Electricity 0.59 0.76 0.67 0.76 1.53 1.40 - Construction 6.49 6.26 5.07 4.68 2.94 2.56 Services 47.64 50.33 40.14 50.02 48.96 48.67 - Commerce 14.83 14.81 12.39 14.66 13.35 13.52 - Transport 4.00 5.36 4.53 6.11 6.17 6.25 - Financial 1.72 2.46 2.22 3.88 2.36 3.55 - Other (non-governmental) 18.96 18.15 14.27 16.08 18.25 18.53 - Government 8.12 9.55 6.72 9.29 8.82 6.81 Source: Banco del Ecuador 2~~~~~~~~~~~~~~ 2 Table A6 in the appendix lends further evidence to this point and shows that exports are mainly primary or semi-industrialized. 5 Figure 1: Sectoral development in GDP (1965 - 98) 24. Agriculture - : .... ..~~~~~~~~~~~~~~~~~~~~~~~. ...... 24 23. 1965 1970 1975 1980 1985 1990 1995 2000 25 - Industry . . .. 24, 1965 1970 1975 1980 1985 1990 1995 2000 25.5 Services 25 24.< ---------.| 1965 1970 1975 1980 1985 1990 1995 2000 Source: Banco del Ecuador 3. Data and Methodology Data Description The data used in this study consists of quarterly data for real GDP in the industrial (Ind), agricultural (Agr) and service (Ser) sectors from 1965 to 1998. The data was provided by the Banco del Ecuador. The three series are depicted in log-levels in Figure 1 and show an increase over the whole sample. The industrial GDP series is marked by one large jump, caused by the rapid increase in industrial output after the foundation of a national oil industry in 1972. In March 1987 industrial GDP fell temporarily due to a major earthquake, which destroyed the export oil pipeline and halted oil exports for 5 months. The drop in the agricultural GDP series in 1983 is caused by the adverse impact of the natural phenomena of El Ninlo. Table 1 presents the contribution of the different sectors in percentage shares of total GDP at various points in time. When comparing the contribution of the sectors to the total GDP across time, it appears that the weight of the agricultural sector declined from 25.8% in 1965 to 17.3% in 1998, while the industrial sector managed to increase its 6 share in the same time from 22.3% in 1965 to 33% in 1998. Once we disaggregate the industrial sector, we find that the increasing weight of the industrial sector can be largely attributed to the oil industry. The share of the manufacturing sector appears to have remained largely constant, averaging around 15% of total GDP, while construction effectively reduced its share in total GDP from 6.5% to 2.6%. The aggregated service sector managed to keep a constant share of total GDP of just below 50%. However, the disaggregation of the service sector shows a different picture. While the weight of public sector services declined from 8.1 % in 1965 to 6.8% in 1998, the financial service sector managed to more than double its contribution to total GDP from 1.7% in 1965 to 3.6% in 1998. The transport sector also steadily increased its share in total GDP from 4.0% in 1965 to 6,3% in 1998. Methodology All time series were log-transformed and tested for unit roots. Based on the augmented Dickey-Fuller (ADF) unit root test all series appear I(1) in levels and I(O) in first differences. See Table Al in the appendix for a summary of the unit root tests. Since nonstationary variables might cointegrate to form a stable long-run relationship, we use the multivariate Johansen approach (1988) to explore possible cointegration relationships in the data.3 We intend to interpret cointegration as evidence for interdependence between the different sectors and propose to explore the dynamics and linkages between the sectors further by estimating dynamic models which incorporate short- as well as long-run information. 3 Appendix I provides a brief review of the multivariate Johansen (1988) approach. 7 4. Empirical Findings Section 4.1 present the results of a cointegration analysis over the whole sample (1965:Q1 to 1998:Q4) using quarterly real GDP data provided by the Banco de Ecuador. To test the robustness of our findings, we additionally investigate two different subsamples. Section 4.2 concentrates on the period from 1965:Q1 to 1989:Q4, section 4.3 focuses on the period from 1990:Q1 to 1998:Q4. Section 4.4 estimates a dynamic short- run sector growth model based on the results of section 4.3. Section 4.5 disaggregates the industrial, agricultural and service sectors into their components and presents the evidence of bivariate cointegration tests between different intrasector components. Appendix 2 repeats parts of the analysis of section 4 for an alternative data set, using annual real GDP data from the World Bank Latin American and Caribbean Regional Database. 4.1 Evidence of Cointegration Our sectoral growth VAR model includes a constant in the cointegration space and 4 lags of each of the variables industrial, agricultural and service sector GDP. This is sufficient to produce random errors.4'5 4 The deterministic components of the VAR were defined according to the rank test based on the so-called Pantula principle (see Johansen and Juselius (1992). According to the Pantula principle three different model specifications (no linear trends in the levels of the data (Model 2), linear trends in the levels of the data (Model 3) and time-trend in the cointegration space (Model 4)) are estimated and the results are presented from the most restrictive alternative (i.e. r = 0, and Model 2) through to the least restrictive alternative (, i.e. r = n- 1, and Model 4). The critical values for Model 2, Model 3 and Model 4 correspond to Table 1*, Table I and Table 2* in Osterwald-Lenum (1992). The test procedure according to the Pantula principle is then to move through Table Al from the most restrictive model and at each stage to compare the trace test statistics to its critical value and only stop the first time the null hypothesis is not rejected. In our case, this is for r = I and Model 2, The rank test thus suggests the inclusion of a constant in the cointegration space. 5 The model specification is presented in the appendix in Table A3. The diagnostics on the residuals of the system show the absence of autocorrelation but indicate some non-normality. Since Cheung and Lai (1993) have shown that the trace-test is robust to both skewness and excess kurtosis, we decided to estimate the model with this specification. 8 Table 2:6 Null Alternative Lag: 4 95% 90% Hypothesis Hypothesis With Constant Critical Critical Value Ho: rank = r Value i trace test r=0 r > 0 45.84 35.10 31.88 r < I r> 1 13.48 20.17 17.79 r <2 r>2 4.02 9.10 7.50 Rejection at the 5% level of significance Source: Authors' calculations. The estimates of trace test statistics, X ua, which test the hypothesis of less than or equal to r cointegrating vectors are reported in Table 2. The number of cointegrating vectors is determined by starting at the top of Table 2 and moving down until Ho cannot be rejected. Since the trace-test statistics for the null hypothesis of no-cointegration ( X tra,e = 45.84) exceeds its 95%-critical value of 35.10, but the null hypothesis of rT I cannot be rejected, there appears to be evidence for one cointegrating relationship between the industrial, the agricultural and the service sectors.7 The Ecuadorian economy has been subjected to frequent and substantial external (e.g., oil boom, debt crisis and natural disasters) and internal shocks (e.g., changes in the development model) during the 1965 to 1998 period. Therefore, empirical evidence in favour of one stable long-run relationship describing the sectoral growth dynamics of Ecuador would be a surprising result. To address the issue of stability of the cointegration relationship over time, we perform a recursive cointegration analysis, where the trace-statistics for the hypothesis of less than one cointegration vector is estimated for different sample periods. Operationally, the data from 1965:Ql to 1982:Ql (roughly the first half of the sample) is used as a base period for the calculation of the first test statistic and the sample 6 This result comes about by starting at the top of Table I and moving downwards until Ho cannot be rejected. As this is the case in the second row, the analysis maintains the Ho of zero cointegrating vectors, this implies the existence of exactly one cointegrating vector in the data. The World Bank Latin American and Caribbean Regional Database contains data on annual real GDP from 1965-98 for Ecuador. Since this data base could have been used for our cointegration analysis, we perforn additionally a cointegration analysis using this data set (see Appendix 2). This allows us to contrast the results of a cointegration analysis based on annual data with the results of a cointegration analysis using a higher frequency. Since cointegration is a long-run property, the frequency of the data should not matter and we would expect to find similar results. As Appendix 2 shows, this appears to be the case. 9 size is then successively increased by one observation at a time until the end of the sample. The corresponding trace-statistics are plotted in Figure 2. The graph is scaled such that unity corresponds to the 10% level of significance. As can be seen from Figure 2, the trace-test statistic rejects the hypothesis that the rank is null only from the late 1980s onwards. Thus, only from approximately 1990 does there seem to be a common stochastic trend between the three sectors. Prior to that there appears to be no evidence for cointegration between the three sectors. To investigate this apparent break in the sample further, we split the sample into the period prior to and post 1990 and perform two separate cointegration analyses. Figure 2: Recursive trace test for cointegration between agricultural, industrial and service sector GDP 1965 - 1998) The Trace tets 1 4~~~~~~~~~~~~~~ 1 2 10 _ _ _ _ _0 __ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ __''' 09 007 4.2 The sample from 1965:01 to 1989:04 When estimating a VAR from 1965:04 to 1989:04, non-normality is strongly rejected for the industry GDP series. An analysis of the residuals indicates that the non-normality is due to two large jumps in the series. These may be caused by the rapid increase in industrial output after the foundation of a national oil industry in 1972 and by the aftermath of a major earthquake in March 1987, which destroyed the export oil pipeline and halted oil exports for 5 months. The rapid increase of industrial output after 1972 as well as the 1987 earthquake represent large outliers and could bias the outcome of our 10 cointegration analysis and explain the lack of cointegration prior to 1990. We therefore report the findings for four different subsamples and test for cointegration between the three sectors inclusive and exclusive of the oil industry. The four different subsamples are 1965:Q1 to 1986:Q4 1965:Q1 to 1989:Q4, 1973:Q1 to 1986:Q4, 1973:Q1 to 1989:Q4 and the findings of the individual cointegration tests are reported in column III (inclusive oil) and column IV (exclusive oil) in Table 3. The findings of the different tests reveal an unanimous picture. The findings are unaffected if the oil industry is included or excluded from the analysis. And even if we exclude the time period prior to the foundation of the national oil industry or post the 1987 earthquake, we fail to establish a cointegration relationship in the data prior to 1990. Table 3: Cointegration tests for different subsamples Null Alternative X , " 95% 90% Hypothesis Hypothesis (excl. oil) Critical Critical Value Value - II III IV V VI 1965Q1 - 1989Q4 trace test r = 0 r > 0 32.38 25.87 35.10 31.88 r < 1 r> 1 9.90 12.33 20.17 17.79 r < 2 r>2 2.70 4.24 9.10 7.50 1973Q1 - 1989Q4 trace test r = 0 r > 0 34.36 27.64 35.10 31.88 r I r> 1 16.61 15.27 20.17 17.79 r s 2 r>2 3.79 3.91 9.10 7.50 1965Q1 - 1986Q4 tr ace test r=0 r>0 25.97 26.20 35.10 31.88 r s I r> 1 10.13 11.31 20.17 17.79 r s 2 r>2 2.99 3.21 9.10 7.50 1973QI - 1986Q4 X trace test r = 0 r > 0 30.20 31.76 35.10 31.88 r s I r> 1 14.24 12.76 20.17 17.79 r s 2 r>2 4.72 3.82 9.10 7.50 Source: Authors' calculations. 11 4.3 The sample from 1990:01 to 1998:04 Since the recursive cointegration analysis from 1965 to 1998 indicates at least one cointegration relationship from 1990 onwards, we focus in this section on the period from 1990:Ql to 1998:Q4. When re-examining our model specification, we find that a lag length of 2 is now sufficient to produce random errors. The model mis-specification tests are presented in Table A4 in the appendix. The X ,test indicates one significant cointegrating vector. Table 4 Null Alternative Lag: 2 95% 90% Hypothesis Hypothesis With Constant Critical Value Critical Value X trace test r=O r>O 41.81 35.10 31.88 r s I r> 1 17.38 20.17 17.79 r s 2 r>2 2.89 9.10 7.50 Rejection at the 5% level of significance Source: Authors' calculations. Normalising the cointegration vector on the 3rd element, yields the following estimates for ,B (Table 5) and a (Table 6): Table 5 Agr 1.000 Ind 0.077 Ser -0.722 Constant -2.588 Source: Authors' calculations. Table 6 a t-statistics AAgr -0.176 -3.997 A Ind -0.182 -2.723 A Ser -0.138 -3.374 Note: A indicates a variable in first differences. Source: Authors' calculations. The column of,f is the cointegrating parameter vector or, in other words,8 spans the cointegration space. The coefficients of a can be interpreted as adjustment coefficients measuring the relative importance of a deviation from equilibrium on a given 12 endogeneous variable. Since AAgr, AInd and ASer all have significant adjustment coefficients, all three variables adjust to a disturbance in the cointegration relationship. Since it is now common practice to try to identify the cointegration space, we impose restrictions on the cointegration vector to see if one of the three variables can be excluded from the cointegration space. From the point of view of a dual economy model (Lewis (1954), Fei and Ranis (1961) and Sen (1966)) such a test might seem important. As mentioned earlier, the dual economy model rules out a long-run relationship between agricultural and industrial outputs. Thus, only if Agr and Ind are both part of the cointegration space, are we able to support a long-run relationship between the agricultural and the industrial sectors and take it as evidence against the dual economy model. If we cannot exclude one of the two series from the cointegration space, the resulting cointegration relationship indicates a long-run relationship between the agricultural or the industrial sectors and the service sector, but not between the agricultural and industrial sectors. This would then not necessarily indicate a violation of the dual economy model. The results of the different hypotheses tests are summarised in Table 7. Since a joint test of long-run exclusion and weak-exogeneity is rejected for Agr, Ind and Ser, all three variables are needed to form the long-run relationship. None of the sectors can therefore be excluded from the cointegration relationship or even be treated as weakly exogenous to the system of equations. 13 Table 7 Agr Ser Ind Const. LR-test p-value HI 0 1 X *(2) = 10.17 0.01 H2 1 0 X (2) = 12.27 0.00 H3 1 0 X (2)= 6.21 0.04 H4 I I . 0 2(1)= 0.90 0.34 Note: Table 7 summarises the findings of different hypotheses tests on the coefficients of a and 6B. A 0 indicates that the coefficient of a variable, i, has been restricted to zero and is equivalent to a test of long- run exclusion, a I indicates the variable used for normalisation and a * indicates that a variable has been left unrestricted. All tests are joint tests for long-run exclusion and weak exogeneity, i.e., a, = A, = 0 . All tests are likelihood ratio (LR) tests which are distributed as zx, conditional upon the rank and the number of restrictions imposed. Source: Authors' calculations. Since H4 cannot be rejected, which says that the constant can be excluded from the cointegration space and, when normalising on service sector GDP, Ser, the long-run reduces to: Ser = 0.747*Agr + 0.339*Ind 4.4 A Dynamic Short-run Growth model To combine short-run and long-run information for the three sectors in a growth model we estimate a parsimonious dynamic model, which contains the cointegration relationship and up to 1 lag of Ind, Agr and Ser in first differences.8 Following Hendry's general-to-specific system reduction approach all insignificant variables are removed from the system based on F-tests and the resulting system is then estimated by full information maximum likelihood (FIML) to fiurther improve its robustness. The resulting model is presented in Table 8. The final model passes the Hendry-Mizon LR test of over-identifying restrictions, X2 (3) = 3.4835 [p=0.3229], and therefore represents a valid reduction of the initial system. All variables that appeared to be strongly significant in the long-run, also retain - 8 A specification of I lag is adequate because the two lags in levels used for the cointegration analysis correspond to one lag in first-order differences. 14 with the exception of A Indt-l - their significance in the short-term. Industrial growth only seems to have a direct positive impact on service sector growth. Table 8: Dynamic Short-Run Sector Growth Model (FIML estimation) A Agr, A Indt A Sert A Agrtl. A Indt-l A Ser,-1 CIt., AAgrt -1 -0.33381 0.34612 (-1.874) (4.471) A Indt -1 0.94599 -0.72683 0.19976 (3.805) (-2.272) (1.497) ASert -1 0.24592 0.16285* -0.53421 0.18986 (1.679) (2.005) (2.908) (-2.814) Diagnostics 2 Single Equation 3 F,(2,30) Xn,o,n (2) F,ch(1,30) AAgr 0.01264 0.736 1.841 3.615 A Ind 0.02155 1.585 1.668 1.139 A Ser 0.01201 4.007 2.272 1.037 Vector Analysis Fa (18,71) = 1.4329 nonn (6) = 6.3 53 F, (84,78) = 0.8523 * Rejection at a 5 percent level of significance. Note: Values in parentheses are t-statistics. A indicates a variable in first differences, subscripts denotes the time period, i.e., t: current period, t-1: lagged one period, CI denotes the restricted cointegration relation. Source: Authors' calculations The agricultural sector seems to play a major role in determining growth in the other two sectors. The strong positive growth effect of agriculture on the industrial sector is of particular interest since it indicates direct Granger causality from agriculture to industry. An explanation for the direct linkage between the agricultural and industrial sector could be provided by the fact that the marine product and processed food industry, which depends directly on agriculture and fishing, managed to increase their overall export shares substantially over recent years. While there appears to be a direct impact of agricultural growth on industrial growth, the industrial sector affects the growth of the agricultural sector only indirectly via the error correction term and via the growth equation of the service sector, ASer,1.. Growth in the agricultural sector also seems to positively affect growth in the service sector, possibly indicating ani increase in commerce with agricultural produces. 15 The service sector seems also to have an important impact on the growth in the other two sectors. But, its impact on growth in the industrial as well as the agricultural sectors is negative. Analysing GDP growth by sector allowed us to recover important intersector dynamics. The dynamic structure of our model is fairly simple and highlights the main lines of causality. However, we have to keep in mind that the industrial, agricultural and service sectors represent themselves as aggregates. There may well exist a much more complex dynamic structure at the intrasector level that might be diffused or even eliminated through sector aggregation. In the next section we, therefore, present the evidence of bivariate cointegration tests between the different intrasector components as defined in Table 1. This might enable us to gain further insight into the inter and intrasector growth dynamics of Ecuador and provide us with important information for further model building. 4.5 Intersector and Intrasector Dynamics In order to explore inter- as well as intrasector dynamics, we perform a recursive cointegration analysis between different sector components. This allows us to establish evidence for cointegration while at the same time addressing the issue of stability of the cointegration relationship. We exclude the time period from 1965 to 1972, prior to the oil nationalisation, from the analysis and use the period from 1973 to 1982 as the base period for our recursive cointegration analysis. The graphical plots of the trace-statistics for the different bivariate cointegration tests are reported in Figures 1 to 9 in the appendix. To save space we only report a selection of the recursive cointegration tests.9 The findings widely confirm the multivariate cointegration analysis of the last section. We find that over the full sample the agricultural sector cointegrates with manufacturing, commerce, transport and public services. The fact that the agricultural sector cointegrates directly with most other sectors is interesting, since agriculture is generally assigned a low degree of forward and backward linkages and thus rarely seen as 9 The other graphs are available from the authors upon request. 16 a key sector for economic development (Chenery and Watanabe (1958) and Hirschman (1961)). However, the recursive analysis also reveals that the trace test for cointegration is not rejected in all investigated subsamples. In line with the earlier results, a stable long- run relationship seems only to form from around 1990 onwards. Figure 3: Recursive Trace test for cointegration between agricultural and manufacturing sector GDP (1973 - 1998) The Trace tbst. I so zo? I 00~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~( o75 Source: Authors' calculations. Of interest is also the development of the trace test for cointegration between the agricultural sector and public services. The null hypothesis of no cointegration is rejected from 1987 onwards which is much earlier than for the other cases. For the manufacturing sector, the picture is even more interesting. While the graphical plots of the trace tests again seem to point to more stability between the sector components towards the end of the sample, the trace test for cointegration between the manufacturing and public sectors is rejected over the full sample, indicating a stable long- run relationship between these two variables. The fact that there also appears to be a stable cointegration relationship between oil and public services is perhaps not such a surprising finding, given the importance of the oil revenues for the economy of Ecuador. 17 Since there appears to be a general tendency for more stability towards the end of the sample, we also perform bivariate cointegration analyses for 1990:Ql to 1998:Q4. The results of these tests are summarised in Table 9. The identified cointegration relationships seem to broadly support the findings of the recursive cointegration analysis over the larger sample and further indicate that the agricultural sector is well interlinked with the other sector components. Another interesting result of Table 9 is that there appears to be no direct link between the oil sector and the non-oil industrial sector. There is however, strong evidence for cointegration between the oil industry and financial services as well as between the oil industry and public services. Since financial services and public services are well interlinked with most other sector components, this underlines that we cannot exclude the oil sector from an intersector growth analysis since an adverse shock to the oil industry might affect the other sectors via the financial and/or the public sector. Table 9: 1990Ql - 1998Q4 Oil Agri. Manu. Constr. Transp. Com. Finance Agriculture 14.71 Manufacturing 15.35 27.71** Construction 10.21 14.61 23.20** Transport 28.77** 19.66* 44.61** 14.59 Commerce 16.32 28.11** 26.01** 10.67 26.75** Finance 20.70** 22.73** 18.80* 13.16 19.97** 18.26* Public 34.96** *29.56** 36.95** 14.75 32.30** 33.12** 40.60** Services Note: Significance at the 10% level is indicated with * and significance at the 5% level with **. A significant test statistic indicates that the null of no-cointegration can be rejected. Critical value at 90%/ level: 17.85, critical value at 95% level: 19.96 Source: Authors' calculations. 18 Further research will have to combine the different sector components within a multivariate analysis to explore the different inter- and intrasector dynamics to a fuller extent.I 5. Conclusion This paper explores whether the experience of Ecuador since 1965 supports the dual economy model based on an empirical analysis of the sectoral components in growth in GDP. While we find evidence for a long-run relationship between the different sectors from 1965 to 1998, the relationship only proves stable from the end of the 1980s onwards. Our findings point to a large degree of interdependence in sectoral growth. Moreover, we identify the agricultural sector as a major driving force in sectoral growth in Ecuador. We take the latter point as evidence against the basic dual economy model, which implies that a long-run relation cannot exist between agricultural and industrial output. When discussing future agricultural development in Ecuador, it might be therefore useful to keep Chile's experience in mind, which demonstrates the importance of high value added agricultural activities for economic development. One other interesting finding of our study is the impact of the oil industry on the recent economic development of Ecuador. While there appears to be no direct link between the oil sector and the non-oil industrial sector, the oil industry cointegrates directly with financial, as well as public services. Since financial and public services cointegrate directly with most other sector components, we cannot rule out important indirect links between the oil industry and other sectors. Thus, the often advocated practice to a priori exclude the oil sector from economic analyses may be too short- sighted. Furthermore, a multivariate cointegration analysis between all sector components of an economy would allow us to directly map the inter- and intrasector growth dynamics of an economy. This is left for future research. '0 As a first step in this direction, in Table A5 we present the results of a multivariate cointegration analysis which combines all 7 sector components. At the I % level of significance we find up to 3 significant cointegration vectors, indicating a fairly complex dynamic structure between the intrasector components. 19 Appendix 1: The Multivariate Cointegration Analysis of Johansen The Johansen procedure allows us to test for cointegration in a multivariate system. Starting from an unrestricted vector autoregressive model (VAR), the hypothesis of cointegration is formulated as a hypothesis of reduced rank of the long run impact matrix 1I (Johansen, 1988, Johansen and Juselius, 1990). The VAR is generated by the vector Zt, which defines the potential endogenous variables of the model. Taking first differences of the variables, the VAR can be transformed into an error correction model Azt = UAZt-1+...+Fk-IAzt-k=l +Zt-k +wDt +t, Et -m(,Y) where the estimates of r, = -(I-A1 -... -A ,),(i=1,...,k-1) describe the short run dynamics to changes in Zt and rI = -(I - Al-...-A,) captures the long run adjustments and D contains deterministic terms. Cointegration occurs in the case of reduced rank of I. Only if the rank is reduced (r0 224.75** 177.20 165.58 159.48 r s I r> 1 160.82** 143.09 131.70 126.58 rs 2 r>2 111.84** 111.01 102.14 97.18 r s3 r> 3 75.72* 84.45 76.07 71.86 r < 4 r>4 50.29 60.16 53.12 49.65 r < 5 r>5 31.26 41.07 34.91 32.00 r 6 r>6 14.33 24.60 19.96 17.82 r <7 r>7 1.71 12.97 9.24 7.52 ** Significant at the 1% level, * significant at the 5% level. Source: Authors' calculations. Fig. Al: Recursive trace test for cointegration between agricultural and commercial service sector GDP (1973 - 1998) Th. Tnc tw zoo 'I '-I7_>AD /V~~~/ Fig. A2: Recursive trace test for cointegration between agricultural and transportation service sector GDP (1973 - 1998) Th. TM" Um 2 110~ ~ ~ ~~~~~~~~~~~2 A I'~~~~~~~~~~~~_ '- "''' *25~~~~~~~~~~~~~~~2 Fig. A3: Recursive trace test for cointegration between manufacturing and financial service sector GDP (1973 - 1998) Th.T T- - . 015 \ ' 0,01 Fig. A4: Recursive trace test for cointegration between manufacturing and commercial service sector GDP (1973 - 1998) M. Tins ,t _ 0B5~ ~ ~ ~~~s 1 ,; , _ _ ' 'S 1588 '0 18, 09 Fig. A5: Recursive trace test for cointegration between manufacturing and transportation service sector GDP (1973 - 1998) The T- tb 0890Z ' ' 8. ',8 ' 8 1'9' ' '1 . i '' '88'' ' o ' ' ' 198 i2 24 Fig. A6: Recursive trace test for cointegration between agricultural and public service sector GDP (1973 - 1998) Th. Tne %.t Fig. A7: Recursive trace test for cointegration between manufacturing and public service sector GDP (1973 - 1998) The TmNlfts ,,,~~~~ i no 00 Oo S . ,'5._ '.4 o. no Fig. A8: Recursive trace test for cointegration between oil and public service sector GDP (1973 - 1998) The T." 52b IN 25 Appendix 3: The World Bank Latinamnerican and Caribbean Regional Database contains data on annual real GDP from 1965-98 for Ecuador. Since this data base could have been. alternatively used for our cointegration analysis, we also perform a cointegration analysis using this data base. This allows us to contrast the results of a cointegration analysis based on annual data with the results of a cointegration analysis based on quarterly data. Since cointegration is a long-run property, the frequency of the data should not matter and we would expect to find similar results. The results of the cointegration analysis with annual sectoral GDP data from the World Bank Latinamerican and Caribbean Regional Database indicate one cointegration relationship and thus confirm the results of our cointegration analysis based on the quarterly data base of the Banco del Ecuador. Table A6: Null Alternative Lag: 3 95% 90% Hypothesis Hypothesis With Constant Critical Critical Value Value x tace test r 5 0 r>0 38.49* 35.10 31.88 r S I r> 1 14.66 20.17 17.79 r < 2 r>2 4.83 9.10 7.50 Note: Estimates are based on a VAR specification with a constant in the cointegration space and a lag- length of 3. This model specification was found to be sufficient to produce random errors Source: Authors' calculations. 26 Table A7: ECUADOR: MERCHANDISE TRADE 1990 1992 1994 1996 1997 1998 Merchandise exports (FOB):* 2724 3102 3843 4900 5264 4203 Primary and semi-processed 2617 2924 3476 4373 4708 3663 goods: Oil and oil derivatives: 1418 1345 1305 1776 1557 923 Crude oil 1268 1260 1185 1521 1412 789 Oil derivatives 150 86 120 255 146 134 Bananas and plantains 471 683 708 973 1327 1070 Coffee and coffee products: 130 82 414 160 121 105 Coffee 104 61 366 129 92 72 Processed coffee 26 21 48 30 30 33 Shrimp 340 542 551 631 886 872 Cacao and cacao products: 131 75 102 164 132 47 Cacao 75 36 66 91 60 19 Processed cacao 56 39 35 73 72 28 Fish and sea products: 88 107 187 291 307 351 Tuna 13 30 21 59 69 61 Fish 34 26 52 26 30 22 Fishmeal 9 7 10 54 23 13 Other processed sea products 32 44 105 152 185 255 Hemp 8. 7 11 15 15 13 Wood 0 8 20 29 38 23 Natural flowers 14 30 59 105 131 162 Other primary products 17 44 119 230 194 97 Manufactured products: 107 177 366 527 556 540 Chemicals and pharrnaceuticals 12 17 32 46 51 57 Metal manufactures 14 34 119 109 142 130 Hats 8 6 8 5 5 4 Textile manufactures 6 19 41 52 61 52 Other manufactured products 68 101 166 315 296 298 Merchandise imports (FOB): 1647 1977 3209 3571 4520 5110 Consumption goods: 160 321 715 779 948 1080 Consumer durables 97 138 304 459 563 660 Consumer non-durables 63 183 411 319 385 420 Fuels and lubricants 69 75 78 122 379 273 Intermediate goods: 860 817 1157 1586 1796 1991 Agricultural 73 97 114 219 246 247 Industrial 707 652 957 1221 1393 1572 Construction materials 80 68 86 145 157 171 Capital goods: 554 761 1259 1083 1396 1766 Agricultural 24 20 31 34 43 51 Industrial 341 440 596 698 918 1108 Transport 189 301 632 351 435 607 Other imports 4 2 0 1 1 1 Source: Informacion Estadistica Mensual, Septiembre 30 de 1999, Banco de Ecuador. *: In thousands of dollars. 27 Bibliography Blunch, N.-H.; Vemer, D. (1999): Sector Growth and the dual economy model: evidence from Cote d'Ivoire, Ghana and Zimbabwe, Policy Research Working Paper No. 2175, The World Bank. Chenery, HI.B.; Watanabe, T. (1958): International comparisons of the structure of production, Econometrica. Cheung, Y.-W., Lai, K. S. (1993): Finite sample sizes of Johansen's likelihood ratio tests for cointegration, Oxford Bulletin of Economics and Statistics, 55,3, 313-328. Gopinath, M.; Roe, T.L; Shane, M.D. (1996): Competitiveness of U.S. Food processing: benefits from primary agriculture, American Journal of Agricultural Economics, 78, 4, 1044-55. Hendry, D.F.; Mizon, G.E. (1993): Evaluating dynamic econometric models by encompassing the VAR, in: P.C.B. Phillips (ed.): Models, methods and applications of econometrics: Essays in honor of A.R. Bergstrom (Blackwell: Oxford, 1993). Hentschel, J. (1994): Trade and Growth in Ecuador: A partial equilibrium analysis, Policy Research Working Paper No. 1352, The World Bank. Hirschman, A.O.; (1959): The strategy of economic development, New Haven, Conn.: Yale University Press. Johansen, S. (1988): Statistical analysis of cointegration vectors, Journal of Economic Dynamic and Control, 12, 231-254. Johansen, S.; Juselius, K. (1990): Maximum likelihood estimation and inference on cointegration. With applications to the demand of money, Oxford Bulletin of Economics and Statistics, 52, p.169-210. Johansen, S.; Juselius, K. (1992): Testing structural hypothesis in a multivariate cointegration analysis of PPP and the UIP for UK, Journal of Econometrics, 53, p.169- 209. Johnston, B,; Kilby, P. (1975): Agriculture and structural transformation: Economic strategies in late developing countries, London: Oxford Press. Krueger, A.; Schiff, M.; Valdes, A. (1992): The political economy of agricultural pricing policy: a synthesis of the economics of developing countries. Baltimore, Md: John Hopkins University Press. Lewis, W.A.; (1954): Economic development with limited supplies of labour, Manchester School of Social Science, 22, 139-91. 28 Marconi, S.; Samaniego, p. (1995): Las fuentes del crecimiento economico: una perspectiva a partir de la demanda, Notas Technicas 19, Banco Central del Ecuador. Martin, W.; Mitra, D. (1998): Productivity growth and convergence in agriculture and manufacturing, Development Research Group, World Bank., and Dept. of Economics, Florida International University, Washington., D.C. and Miami., Fla. Mellor, J.; Lele, U (1973): Growth linkages of the new food grain technologies, Indian Journal ofAgricultural Economics, 18 (1). Mellor, J. (1976): The new economics of growth: A strategy for India and the developing World, Ithaca, N.Y.: Cornell University Press. Michaely, M. (1999): The Andean Group: Ecuador's trade and commercial policies, mimeo. Osterwald-Lenum, M. (1992): A note with quantiles of the asymptotic distribution of the maximum likelihood cointegration rank test statistics, Oxford Bulletin of Economics and Statistics, 54, 3, 461- 473. 29 Policy Research Working Paper Series Contact Title Author Date for pape- WPS2493 Annuity Markets in Comparative Estelle James November 2000 A. Yaptenco Perspective: Do Consumers Get Dimitri Vittas 31823 Their Money's Worth? WPS2494 The Relevance of Index Funds for Ajay Shah November 2000 A. Yaptenco Pension Investment in Equities Kshama Fernandes 31823 WPS2495 The Australian Annuity Market David M. Knox November 2000 A. Yaptenco 31823 WPS2496 Perspectives on the Sources of Somik V. Lall November 2000 R. Yazigi Heterogeneity in Indian Industry G. Chris Rodrigo 37176 WPS2497 State Policies and Women's Monica Das Gupta November 2000 M. Das Gupia Autonomy in China, India, and the Sunhwa Lee 31983 Republic of Korea, 1950-2000: Patricia Uberoi Lessons from Contrasting Danning Wang Experiences Uhong Wang Xiaodan Zhang WPS2498 Sustaining Economic Welfare: Kirk Hamilton November 2000 L. Rivera Estimating Changes in Per Capita 82819 Wealth WPS2499 The Treatment of Non-Essential Isidro Soloaga December 2000 L.Tabada Inputs in a CobbtDougias Technology; 36896 An Application to Mexican Rural Household-Level Data WPS2500 Investigating Corruption Canice Prendergast December 2000 H. Sladovich 37698 WPS2501 Anti-Corruption Policies and Jeff Huther December 2000 A. Santos Programs: A Framework for Anwar Shah 31675 Evaluation WPS2502 Implications of the Currency Crisis Masahiro Kawai December 2000 A. Azarcon- for Exchange Rate Arrangements Shigeru Akiyama 36049 in Emerging East Asia WPS2503 Proposed Strategy for a Regional Masahiro Kawai December 2000 A. Azarcori Exchange Rate Arrangement in Shinji Takagi 36049 Post-Crisis East Asia WPS2504 Social Capital and the Quality of Stephen Knack December 2000 P. Sintirn-sc aoG sY Government: Evidence from the 38526 United States Policy Research Working Paper Series Contact Title Author Date for paper WPS2505 Family Altruism and Incentives Roberta Gatti December 2000 R. Bonfie'd 31248 WPS2506 Ethnicity and Wage Determination Abigail Barr December 2000 R. Bonfield in Ghana Abena Oduro 31248 WPS2507 Public Expenditures and Gunnar S. Eskeland December 2000 H. Sladovich Environmental Protection: When Is 37698 the Cost of Funds Irrelevant? WPS2508 Sources of Financial Assistance for Mattias Lundberg December 2000 V. Soukhanov Households Suffering an Adult Mead Over 35271 Death in Kagera, Tanzania Phare Mujinja WPS2509 How Tax Policy and Incentives Jacques Morisset December 2000 N. Busjeet Affect Foreign Direct Investment: Neda Pirnia 33997 A Review WPS2510 Environmental Protection and Gunnar S. Eskeland December 2000 H. Sladovich Optimal Taxation 37698 WPS2511 Monetary Policy under Flexible Pierre-Richard Agenor December 2000 M. Gosiengfiao Exchange Rates: An Introduction to 33363 Inflation Targeting WPS2512 Quantifying the Impact of Technical Keith E. Maskus December 2000 L. Tabada Barriers to Trade: A Framework for John S. Wilson 36896 Analysis Tsunehiro Otsuki WPS2513 Do State Holding Companies Facilitate Michel Kerf December 2000 M. Leon Private Participation in the Water Sector? 36151 Evidence from C6te d'lvoire, the Gambia, Guinea, and Senegal