Macroeconomics and agriculture in Tunisia

MACROECONOMICS AND AGRICULTURE IN TUNISIA Monia Ben Kaabia Departamento de Análisis Económico Universidad de Zaragoza Email: [email protected] José M. Gil CREDA-UPC-IRTA Edifici ESAB - Campus del Baix Llobregat Av. del Canal Olimpic s/n 08860-Castelldefels (Barcelona) Ph: ++34-935521210 Fax: ++34-935521121 e-mail: [email protected] Houssem E. Chebbi Sfax University (Tunisia) e-mail: [email protected] Paper prepared for presentation at the XIth Congress of the EAAE (European Association of Agricultural Economists), ’The Future of Rural Europe in the Global Agri-Food System’ Copenhagen, Denmark 24-27 August 2005 Copyright 2005 by Monia Ben Kaabia, José M. Gil and Houssem E. Chebbi. All rights reserved. Readers may make verbatim copies of this document for non-commercial purposes by any means, provided that this copyright notice appears on all such copies. 1 MACROECONOMICS AND AGRICULTURE IN TUNISIA Abstract This paper aims to analyse the impact of changes in the monetary policy and the exchange rate on agricultural supply, prices and exports. The methodology used is based on the multivariate cointegration approach. Ten variables are considered: interest and exchange rates, money supply, inflation, agricultural output and input prices, agricultural supply and exports, income and the rate of commercial openness. Sample period covers annual data from 1967 to 2002. Due to the short-sample period, two subsystems are considered. First, long-run relationships are identified in each subsystem. Second, both subsystems are merged in order to calculate the short-run dynamics. Results indicate that changes in macroeconomic variables have an effect on the agricultural sector but the reverse effect does not hold. Key words: Macroeconomic policy, agro-food sector, Tunisia, impulse-response functions JEL Classification: C32, N57, O31 1. Introduction The ongoing globalisation process in the world economy is a big challenge for Tunisia, a country which has suffered a complex process of structural economic reforms. The Adjustment Structural Program implemented in 1986 generated a new environment of economic success. All sectors of economy started to recover and exports dramatically increased being one of the main contributors to economic development. As an example, in the last five years the Tunisian GDP increased at a 5.5% annual rate while inflation was maintained around 3.5 %. The agro-food sector in Tunisia plays an important role in Tunisian economy. It generates around 14% of total GDP, employs 22% of total labour force and agro-food exports represent around 15% of total exports, although still depending to a great extent on weather conditions. Moreover, since 1986 the agricultural sector is undergoing a modernization process characterized by a progressive intensification and the use of technology. However, the agricultural production has not been able to meet the needs of an increasing population. In general, the Government favoured imports of raw materials and food, which have provoked a progressive deterioration of trade balance. The agricultural policy was, then, oriented into two directions: 1) to promote the production of agricultural products in which self-sufficiency was low, through the implementation of a subsidies program (food security); and 2) to encourage the production of food products in which Tunisia had traditionally had a competitive advantage (olive oil, fruits, vegetables, etc) to finance the agricultural trade deficit. In many cases, results from such policies were, to some extent, different from those expected as the effect of many macroeconomic variables (as a consequence of the Adjustment Structural Program) were not taken into account. Although not explicitly recognized, changes in the macroeconomic policy have become increasingly important for the agro-food sector, as Tunisian agriculture has become more capitalized and more dependent on international markets, thereby becoming more vulnerable to changes in interest rates, exchange rates and international growth rates. The aim of this paper is, precisely, to provide a methodological approach taking into account data limitations to explain the relationships between macroeconomic variables and the agricultural sector in Tunisia. Special attention is paid to the distinction between long-run structural relationships and shortrun dynamics. Up to our knowledge, this is the first attempt to analyse such relationships in Tunisia. The existing literature on Tunisia is quite descriptive focussing on the evolution of agricultural trade flows which are only explained by changes in the agricultural policy (Arfa, 1994; Allaya, 1995; and El Abassi, 1995, among others). Since the mid seventies, a number of theoretical and empirical studies have analysed the impact of macroeconomic variables on the relative performance of the agricultural sector (see In and Mount, 1994, for a literature review on this topic). In the early studies, macroeconomic variables (income, 2 interest rate, exports,...) were introduced as purely exogenous in agricultural sector models. The paper by Schuh (1974) could be considered as the starting point of a second group of studies emphasizing the role of exchange rate in explaining agricultural variable fluctuations (Chambers and Just, 1979, 1981; Longmire and Morey, 1983; and Batten and Belongia, 1986). However, these empirical investigations neglect not only the possible effect of exchange rate changes on other macroeconomic variables (which can influence agricultural prices and exports indirectly) but also the effects of other macroeconomic variables (such as interest rates) both on exchange rate and agricultural variables. In this context, Chambers (1984) develops a general equilibrium model in order to analyse the effect of macroeconomic variables on agricultural trade where the exchange rate, income, interest rate as well as usual agricultural variables are treated as endogenous. Finally, it is possible to identify a third group of papers dealing with the analysis of the dynamic linkages between monetary variables and the agricultural sector. The question of money neutrality in the agricultural sector, and the speed of price adjustments, has been considered of central importance for policy analysis (Bessler and Babula, 1987; Devadoss and Meyers, 1987; Taylor and Spriggs, 1989; Larue and Babula, 1994; Dorfman and Lastrapes, 1996, among others). Results from most of the above-mentioned studies substantially differ from each other, and, in many cases, they are even contradictory. There exist alternative explanations for such differences: samples are not homogeneous, the number of variables included differs as well as their treatment as endogenous or exogenous, and the different methodological approaches used. However, there seems to exist a consensus on the fact that models analysing macroeconomic linkages to the agricultural sector should include the more relevant macroeconomic variables of the country being analysed and should treat them as endogenous (Devadoss et al., 1987; Taylor and Spriggs, 1989; Denbaly and Torgerson, 1991; Thraen et al., 1992; In and Mount, 1994; Ben Kaabia and Gil, 2000; among others). Partly for this reason, most of the analyses on this topic have recently been conducted using Vector Autoregression (VAR) models. This is also the methodological approach we have followed in this paper although adapted to take into account data limitations and their stochastic properties. The paper is organized as follows. The data used in this study, their stochastic characteristics as well as the methodological approach are presented in Section 2. Long-run equilibrium relationships are analysed in section 3. The short-run dynamics is considered in section 4. Finally, some concluding remarks are outlined. 2. Data and methodological approach Since the Sims’ (1986) seminal paper, VAR models have been one of the most widely used analysis tools to analyse the dynamic relationships between macroeconomic and agricultural variables. In VAR models, all variables are considered endogenous and no zero/one restrictions are imposed on the variables in the system. However, recent developments in time series analysis have modified the econometric framework for analysing such relationships. The concepts of non-stationarity and cointegration have become very popular and have to be explicitly tested to properly specify an econometric model. In this new context, Johansen (1988) and Johansen and Juselius (1990, 1992 and 1994) provide an interesting methodology that allows the researcher to distinguish between the short and the long run. On the one hand, it is possible to identify the long-run structural relationships among a set of variables and how variables in the system adjust to deviations from such long-run equilibrium relationships. On the other hand, it is possible to calculate the impulse response functions in a similar way to that in the VAR models. This distinction is useful as economic restrictions are considered to be long-run in nature while it is also interesting, for the policy analysis, to know how the system adjusts to disequilibrium. In this paper we have followed this methodological approach although we have introduced some modifications in order to adapt it to data limitations. Availability of data is a major problem for economic modelling in Tunisia. It is difficult to find a large enough sample period for many economic variables. In this study 10 variables have been considered which collect the most important information in relation to macroeconomic variables and the agricultural sector (see the Appendix for 3 data sources and units of measurement): 1) Real Exchange rate (ER), defined as national currency (TND) per US dollar taking into account both the US and Tunisian consumer price indices; 2) Real money supply (M) (money supply (M2) divided by the consumer price index); 3) Interest rate (R), defined as the one-year money market interest rate; 4) Inflation (P) expressed as the Consumer Price Index in first differences; 5) Real Gross Domestic Product (GDP); 6) Real farm output prices (PP), calculated as nominal farm output prices divided by the Consumer Price Index); 7) Real farm input prices (IP)1, calculated as nominal farm input prices divided by the Consumer Price Index); 8) Real agricultural exports (AX), calculated as the nominal exports value divided by the consumer price index); 9) Agricultural output (AP), calculated as the value of the Tunisian Agricultural Output divided by the Consumer price Index; and 10) Rate of commercial openness (RCO) calculated by dividing the international trade flows (imports + exports) by the GDP. This variable provides an indication on how the Tunisian economy is inserted in the world trade. All variables are in logarithms, except for the interest rate and the inflation, which are in a percentage form and are divided by one hundred to make the estimated coefficients comparable with logarithmic changes. The sample period covers annual data from 1967 to 2002. Time series univariate properties have been examined by using unit root tests. As in small samples such tests have limited power (Blough, 1992), two alternative unit root tests developed by Elliot et al., (1996) and Ng and Perron (2001) as well as the stationary test from Kwiatkowski et al. (1992) (KPSS) have been applied. All tests indicated that all variables were I(1)2 Taking into account the number of variables, the number of observations available for each variable and that all variables are I(1), the methodological approach followed in this paper consist of the following steps: i) The ten-variable system is divided into two subsystems. The first one has been defined by including: the real money supply, the inflation, the GDP, the farm input and output prices and the interest rate. Furthermore, taking into account the characteristics of the Tunisian economy, we have considered the interest rate as purely exogenous. The second subsystem includes the following seven variables: the farm input and output prices, the agricultural exports, the agricultural production, the exchange rate, the interest rate and the rate of commercial openness. Within this subsystem also the interest rate as well as the rate of commercial openness are defined as purely exogenous3. ii) Under the assumption of exogeneity for certain variables, the multivariate cointegration procedure developed by Pesaran et al. (2000) is used to test for cointegration in both subsystems. Moreover, cointegration vectors are identified as long-run meaningful economic relationships. iii) Merging results from the two subsystems into a single system with the original 10 variables, impulse response functions are computed to analyse short-run dynamics and to test the exogeneity assumptions made in the first step. 3. Long-run analysis 3.1. Model specification and cointegration rank All variables in each subsystem were I(1) and, then, a Vector Error Correction Model has been specified for each subsystem. The methodology developed by Pesaran et al. (2000) is used to determine the cointegration rank. These authors modified the Johansen (1988) procedure to explicitly allow for the introduction of exogenous variables. The base-line econometric specification for multivariate cointegration is a VAR(p) representation of a k-dimensional time series vector Yt reparametrized as a Vector Error Correction Model (VECM): 1 Fertilizer prices are used as a proxy in this study. 5 Results are not shown due to space limitations. They are available upon request. 3 In a further sep in the modelling process, specific tests will be carried out to test for the exogeneity of the mentioned variables. 4 t 1t-1p+t-1p-1t-1 t te + Y Y + ... + Y + D = Y Π − ∆Γ ∆ Γ µ ∆ (1) where, Yt is a (kx1) column vector of variables; Dt is a vector of deterministic variables (intercepts, trend...) where µ is the matrix of parameters associated with Dt ; Γi are (k×k) matrices of short-run parameters (i=1,...,p-1), where p is the number of lags; Π is a (k×k) matrix of long-run parameters and et is the vector of disturbances niid(0,Σ). When exogenous variables are considered the vector Yt vector can be partitioned as Yt = (Zt,’, Xt ‘)’, where Zt, is a (mx1) vector of endogenous variables and Xt is a (nx1) vector of exogenous variables (n=k-m), which can be considered as the “long-run forcing” variables in the system, that is, changes in Xt have a direct influence on the variables Zt, while they are not affected either by the changes in the equilibrium relationships nor by past changes in Zt. This is equivalent to the notion that the set of variables Zt do not Granger-cause Xt.. According to the mentioned partition the error term et can be decomposed as follows: )e,e(e xtytt ′ ′ ′ = with covariance matriz given by:         ΩΩ ΩΩ =Ω xxxy yxyy (2) According to (3), the error terms of the endogenous variables (eyt) can be represented in terms of the ext as follows: txt 1 xxyxyt uee +ΩΩ= − (3) where, ut ∼ IN(0,Ωuu) and Ω, being u xy 1 xxyxyyuu ΩΩΩ−Ω= −t independent of ext . Substituting (3) in (1) and considering a similar partition for the other matrices, )´,´( xz ′ µ µ=µ , ,,),( xz ′ Π′ Π′ =Π ),( xizii ′ Γ′ Γ ′ =Γ , (i=1, 2,..., p-1), we get a conditional model for ∆Zt as a function of Yt-1, ∆Xt, ∆Yt-1, ∆Yt-2,... , which adopts the following expression:: ∑ − = −− +Π+∆Ψ+Λ∆+δ=∆ 1p 1i t1tx,yyitittt uYYXDZ t=1, 2, ..., T (4) being :δ x 1 xxzxz µΩΩ−µ= − 1 xxzx − ΩΩ=Λ xi 1 xxzxzii ΓΩΩ−Γ=Ψ − i=1, 2, ..., p-1 x 1 xxzxzx,zz ΠΩΩ−Π=Π − If variables in Xt are not cointegrated, that is, 0 x = Π (and, then, zx,zz Π = Π ), Pesaran et al.. (2000) show that the k-variable system defined in (1) can be decomposed in the following two subsystems:4 • Conditional subsystem: (5) ∑ − = −− +Π+∆Ψ+Λ∆+δ=∆ 1p 1i t1tzitittt uYYXDZ 4 Under such decomposition, variables in Xt are assumed to be weakly exogenous with respect to the cointegration space. Moreover, if variables in Zt don not Grange-cause Xt, then such variables are assumed to be strongly exogenous with respect to such cointegration space, that is, they would be only explained by their own past in the marginal subsystem. 5 • Marginal subsystem: (6) xtit 1p 1i xixt YX ε+∆Γ+µ=∆ − − = ∑ Taking into account equations (5) and (6), to test for cointegration is equivalent to test for the Rank (r) of the matrix Πz: [ ] rRank :H zr = Π r=0, ..., m (7) To test for the number of cointegrating vectors (r), Pesaran et al., (2000), following Johansen (1988), proposed two statistics: the trace statistic and the λmax statistic. If the hypothesis of cointegration is not rejected (0<r<m), Yt is said to be cointegrated in the sense that there exists a kxr matrix β such that (β'Yt-1) is stationary and, consequently, the cointegration relationships can be can formally expressed as Πz=αzβ’. This procedure has been applied to the two subsystems described in the last section. Both subsystems are estimated including two lags5 and a constant restricted to the cointegration space6. Multivariate tests for autocorrelation (Godfrey, 1988) and normality (Doornik and Hansen, 1994) have been carried out to check for model statistical adequacy before applying the reduced rank tests. Results indicated that both subsystems could be considered correctly specified7. Table 1 shows the results from cointegration tests in both subsystems. As can be observed, for the first subsystem (upper part of Table 1) results from the λ-max and the trace tests indicate that there exist two cointegration vectors among the six variables includes while for the second subsystem (lower part of Table 1) results differ depending on the level of significance (two and three cointegration vectors for 5 and 10% levels of significance, respectively). Table 1. Results from cointegration tests First subsystem Y’= {M, P, GDP, PP,IP,R}1 Critical values λ-max2 Critical values Trace2 H0: r Ha: p-r λ-max. Trace (10%) (5%) (10%) (5%) 0 5 46,88 124,76 34,99 37,48 82,17 86,58 1 4 37,72 77,88 29,01 31,48 59,07 62,75 2 3 20,92 38,15 22,98 25,54 39,12 42,40 3 2 11,37 19,23 16,74 18,88 22,76 25,23 4 1 7,85 7,85 10,50 12,45 10,50 12,45 Second subsystem Y’ = {PP, IP, AX, AP, ER, R, RCO}1 Critical values λ-max2 Critical values Trace2 H0: r Ha: p-r λ-max. Trace (10%) (5%) (10%) (5%) 0 5 42,84 123,36 37,81 40,57 92,93 97,57 1 4 34,91 80,15 32,00 34,69 67,83 72,15 2 3 28,42 47,06 26,08 28,49 45,89 49,43 3 2 16,17 23,85 19,67 21,92 27,58 30,46 4 1 7,68 7,68 13,21 15,27 13,21 15,27 1 See the Appendiz for variable definitions 2 Critical values are taken from Pesaran et al. (2000). Taking into account the relatively large dimension of the VECM and the small sample available, the outcome of the test procedure has to be interpreted with some caution. Several simulation studies show (Abadir et al., 1999; Gredenhoff and Jacobson, 2001; and Johansen and Juselius, 2000) that the asymptotic critical values may not be very close approximations in small samples. For that reason, we 5 A small-sample adjusted Likelihood Ratio statistic has been used considering a maximum lag of three periods taking into account the sample size. 6 Results from unit root tests indicated that almost all the variables were non-stationary with non zero-means. 7 Results from multivariate first-order autocorrelation tests were 21.14 and 23.85 for the first and the second subsystem, respectively, which were well below the critical value at the 5% level of significance ( ). Results from multivariate normality tests were 14.43 and 17.83, for the first and the second subsystem, respectively, which were well below the critical value at the 5% level of significance ( χ). 65.37 2 25 =χ 31.18 2 10 = 6 have also studied the roots of the companion matrix and the t-ratios of the αz parameters from the last cointegration vector (Juselius, 1995). For both subsystems, all the roots were inside the unit circle, indicating that all variables were I(1). Moreover, the eigenvalues of the companion matrix show that, for both subsystems, the first four roots were close to unity while the rest were quite small. In other words, we could not reject the null of two and three cointegrating vectors for the first and the second subsystems, respectively. Finally, all t-ratios of the αz parameters of the third cointegration vector for the first subsystem were not significant, while in the second subsystem, some of them were significant8. Thus, the first subsystem has been specified with two cointegrating vectors, whereas three cointegration vectors have been chosen for the second one.9 3.2 Long-run structural relationships Identifying economically interpretable relations is the primary aim of this analysis. However, Juselius (1994) argues "the interpretation of the unrestricted cointegration space is far from straightforward when there are more than one cointegrating vector". Moreover, Johansen and Juselius (1994) suggest that only sometimes the unrestricted cointegrating vectors, surprisingly, can be directly interpreted in terms of theoretical economic relationships. Thus, some restrictions are needed in order to obtain a structural representation of such relationships. First subsystem Taking into account the variables included in the model as well as the economic theory which relates those variables, the following hypothetical cointegration relations could be expected: i) A money demand equation in real terms in which the monetary aggregate is related to the inflation in Tunisia, the Gross Domestic Product and an opportunity cost represented by the interest rate: t1 1 t 1 Pt 1 Rt 1 GDPtt 1sys 1PRGDPRM:Y)( ε+µ+β+β+β= ′ β (8) It is expected that βGDP>0; βR <0 and βP<0. If βGDP=1 Equation (8) would be consistent with the Quantity Theory of Money, wheras βP= 0 would exclude inflation to play a role in the demand for money in Tunisia. ii) A price transmisión equation: t2 2 t 2 PPtt 1sys 2PPIP:Y)( ε+µ+β= ′ β (9) from which it is possible to test the homogeneity condition: 1 IP PP PRC IP PP IP = ∂ ∂ = β β − The two equations can be written more compactly as: t1t1sys Y ε = β ′ − ∼I(0) where: β (10)       − = ′*011000 **00**1 1sys In this paper, a two-step procedure is going to be used in order to check if (10) is supported by data. In the first step, each single restricted relation (8)-(9) is tested for stationarity leaving the other relations unrestricted. In other words, if restrictions imposed are compatible with a stationary relationship. The second step involves jointly considering the full identification of the two relationships. Juselius (1998) points out that this approach maximizes the chance of finding a correct full identification of long-run relations. Hypotheses related to the first step adopt the general form H0i: β=(Hiϕ,ω)10. In such an expression, restrictions to be tested are only placed in a single cointegration vector while the 8 Results are not presented due to space limitations but they are available from authors upon request. 9 The unrestricted cointegration space is not presented due to space limitations. Test carried out on the long-run parameters in β indicated that all of them were significant. 12 See Johansen and Juselius (1992) for a full description of the procedure to formulate and test such hypotheses. 7 remaining (r-1) vectors are considered unrestricted. Johansen and Juselius (1992) suggest that this test can be used when we wish to test if there exists some vector in the cointegration space that linearly combines the variables in a particular hypothesized stationary relationship. Several hypotheses have been considered and tested. The specification of such hypotheses, as well as main results found are shown in Table 2. With respect to the first relationship, three different hypotheses have been tested. In the first one ( ), it is tested that real money is cointegrated with interest rate, GDP and inflation, imposing also income homogeneity. Results from the Likelihood Ratio (LR) statistic indicate that the null cannot be rejected 1sys 01 H 11. In the second hypothesis ( ), an additional restriction is considered (β=0). This hypothesis is strongly rejected, which means that the monetary authority is not fixing the monetary policy taking into account an aggregate money stock. The third hypothesis ( ) is similar to the first one but excluding the income homogeneity. Also in this case, we fail to reject the null hypothesis. Finally, in relation to the second relationship, hypothesis ( ) tests for price homogeneity in the agricultural sector. The LR statistic is under the critical value suggesting that monetary policy has a neutral effect on the real food-based prices. This means that, in the long run, input prices and output prices react in the same way and magnitude to changes in money supply. 1sys 02 H 1 P 1sys 03 H 1sys 04 H Once it has been checked that each single equation is a cointegrated relationship, the second step consists of testing a full identification of the structural long-run relationships following Johansen and Juselius (1994). Taking into account results showed in the upper part of the Table 2, two hypotheses have been tested. The first one ( ) jointly tests the hypotheses and , whereas the second tests the hypotheses and . Only in the second case we fail to reject the null hypothesis (the LR statistic is 11.55, which is well under the critical value at the 1% level of significance (χ 1sys 05 H 1sys 04 1sys 01 H1sys 04 H 1sys 03 HH 2(5) = 15.09)), indicating that in Tunisia, the inflation plays a significant role in the demand for money and that agricultural prices satisfy the homogeneity condition. Second subsystem In the second subsystem, taking into account the variables included and results obtained in the first one in relation to the agricultural prices, the following hypothetical cointegration relations could be expected: i) As the agricultural prices are also included in the second subsystem, and in order to check for data consistency, the first cointegration relationship would attempt to relate agricultural prices under the homogeneity restriction: t1 1 t 1 PP tt 2sys 1PPIP:Y)( ε+µ+β= ′ β (11) ii) The second relationship is going to be associated with an agricultural export equation for Tunisia, which would depend on the exchange rate, farm output prices and the rate of commercial openness: t2 2 t 2 RCOt 2 ERt 2 PP tt 2sys 2RCOERPPAX:Y)( ε+µ+β+β+β= ′ β (12) iii) The last relationship is defined as an agricultural suply equation in which farm input and output prices, the interest rate and the rate of comercial openness are included as main potential determinants: t3 3 t 3 RCOt 3 R t 3 IP tt 2sys 3RCORIPAP:Y)( ε+µ+β+β+β= ′ β (13) Equations (11), (12) and (13) can be formulated in compact form as: 11 Several authors such as Reimers (1992) and Abadir et al. (1999) pointed out the tendency of likelihood ratio tests to over-reject in small samples when testing for the cointegration rank. Garratt et al. (1999) undertook a bootstrapping exercise to obtain critical values for testing the over-identification restrictions. The resulting critical values were higher than the asymptotic ones. This result would imply that the over-identification restrictions tested here are not rejected with higher p-values. 8 t1t 2sys Yε=β′− ∼I(0) where (14)          − =β′ ***010*0 **0*010* *0000011 2sys Table 2. Hypothesis restrictions tests on the cointegration vectors in the first subsystem a Hypotheses on a single cointegration vector Hypothesis formulation ),H(),(: 11 1sys ii Φϕ=Φβ=β 0 Η Statistic Critical Value (1%) : 1sis 01 H     − =β′******* **001*1 Yt 1sys 1t Y χ2(2) = 8,77 9,21 : 1sis 02 H     − =β′******* **00101 Yt 1sys 2t Y χ2(3) = 12,37 11,34 : 1sis 03 H      =β′******* **00**1 Yt 1sys 3t Y χ2(1) = 5,58 6,63 : 1sis 04 H     − =β′******* *011000 Yt 1sys 4t Y χ2(4) = 10,12 13,28 Hypotheses on the full system Hypothesis formulation )H,H(),(: 221121 1sys iϕϕ=ββ=β 0i H Statistic Critical value (1%) : 1sis 05 H      − − =β′*011000 **001*1 Yt 1sys 5t Y χ2(6) = 35,16 16,81            − = ′ 1000000 0100000 0000010 0000101 H1 and H      − = ′1000000 0011000 2 : 1sis 06 H      − =β′*011000 **00**1 Yt 1sys 6t Y χ2(5) = 11,55 15,09                 = ′ 1000000 0100000 0000100 0000010 0000001 H1 and H      − = ′1000000 0011000 2 a Y’= {M, P, GDP, PP,IP,R}. An * indicates that the coefficient is unrestricted. In order to test restrictions on the cointegration space, a similar approach to that mentioned for the first subsystem has been followed. However, in this case, as there are three cointegration vectors, one further step has been included. As a first step, we have carried out some tests on each individual long-run relationship leaving the rest unrestricted. The first hypothesis ( ), as mentioned above, only tries to guarantee consistency of data used. Thus, we have tested if agricultural prices homogeneity is stationary. Results from the LR test indicate that the null cannot be rejected, the same results as in the first subsystem. Three alternative hypotheses have been defined for the agricultural exports equation. The first one (H ), tests for a stationary relationship among agricultural exports, farm output prices and the exchange rate and the rate of commercial openness. The second one ( excludes the rate of commercial openness and includes the agricultural supply. Finally, the third one () excludes the rate of commercial openness without including any other variable. Results from the LR tests indicate that only the two first hypotheses are supported by the data. Finally, in relation to the agricultural supply equation, two alternative hypotheses have been considered. In the first one () agricultural output is defined as a function of farm input prices, the interst rate and the rate of 2sys 01 H 2sys 02 2sys 03 H) 2sys 04 H 2sys 05 H 9 The study has shown that changes in agricultural variables have no significant effects on macroeconomic variables. Only shocks in agricultural prices have an effect on inflation. The main source of responses of the agricultural sector (mainly agricultural output and exports) is changes in the monetary policy and, more precisely, on money supply, which is consistent with how monetary policy is instrumented in Tunisia. Agricultural prices responses to macroeconomic shocks are not very significant as an indication of the degree of government intervention in Tunisia. In the case of agricultural exports, responses are larger if they are generated by changes in the exchange rate than if they are generated by changes in output prices, which is an indication that macroeconomic variables have to take into account when designing the agricultural policy. In the same context, the agricultural supply is quite inelastic but it reacts more to changes in the capital cost than to changes in input or output prices. To conclude, it has to be said that results presented in this paper depend on the variables and sample period chosen. Further analysis, including other variables and an extended sample period, could be conducted in the future. REFERENCES Abadir, K.M., Hadri, K. and Tzavalis, E. (1999). The Influence of VAR Dimensions on Estimator Biases. Econometrica 67: 163-181. Allaya, M. (1995). Intercambios Agroalimentarios y Cooperación entre Europa y el Magreb. Revista Española de Economía Agraria. Vol. 172: 31-49. Arfa, L. (1994). Les Exportations Agroalimentaires de la Tunisie vers la C.E. : Des Performances Inégales. MEDIT, vol. 4: 4-12. Batten, D.S. and Belongia, M.T. (1986). Monetary Policy, Real Exchange Rates and U.S. Agricultural Export. American Journal of Agricultural Economics 68: 422-427. Ben Kaabia, M. and Gil, J.M. (2000). Estimation and inference in cointegrated demand systems: an application to Tunisian meat consumption. European Review of Agricultural Economics, 28(3): 449-471. Bessler, D.A. and Babula, R.A. (1987). Forecasting Wheat Export: Do Exchange Rate Matter?. Journal of Business and Economic Statistics 5: 397-406. Blough, S.R. (1992). The Relationship between Power and Level for Generic Unit Root Tests in Finite Samples. Journal of Applied Econometrics 7: 295-308. Chambers, R. G. (1984). Agricultural and Financial Market Interdependence in the Short Run. American Journal of Agricultural Economics 66: 12-24. Chambers, R. G. and Just, R. E. (1979). A Critique of Exchange Rate Treatment in Agricultural Trade Models. American Journal of Agricultural Economics 61: 249-257. Denbaly, M. and Torgerson, D. (1991). Macroeconomic Determinants of Relative Wheat Prices: Integrating the Short Run and Long Run. Journal of Agricultural Economics Research 44(2): 27-35. Devadoss, S. and Meyers, W. H. (1987). Relative Prices and Money: Further Results for the United States, American Journal of Agricultural Economics 69:838-842. Devadoss, S., Meyers, W. H. and Starleaf, D. R. (1987). Macrolinkages between the Farm and no Farm Sectors and the Impact of Monetary Policy Decisions. CARD Monograph 87-M1. Center for Agricultural and Rural Development. IOWA State University. Doornik, J. and Hansen, H. (1994). A Practical Test for Univariate and Multivariate Normality. Discussion paper, Nuffield College, Oxford. Dorfman, J. H. and Lastrapes, W. D. (1996). The Dynamic Responses of Crop and Livestock Prices to MoneySupply Shock: A Bayesian Analysis Using Long-Run Identifying Restrictions. American Journal of Agricultural Economics 78: 530-441. El Abassi, M. (1995). Ouverture et Compétitive de l'Agriculture Tunisienne. La Nouvelle Conjoncture Internationale et les Flux d’Échanges avec l'Europe. Le Cas des Produits Agricoles et Agroalimentaires. Les Agricultures Maghrébines à l'Aube de l'An 2000. Options Méditerranéennes. Série B-Etudes et Recherches, vol 14, 337-347. Elliott, G., Rothenberg, T. J. and Stock, J. H. (1996). Efficient Tests for an Autoregressive Unit Root, Econometrica 64, 813–836. Garrat, A., Lee, K., Pesaran, M.H. and Shin, Y. (1999). A long Run structural Macroeconomic Model of the UK. DAE Working Paper Nº 9812, University of Cambridge. Godfrey, L.G. (1988). Misspecification Test in Econometrics. Cambridge University Press. Gredenhoff, M. and Jacobson, T. (2001). Bootstrap Testing Linear Restrictions on Cointegrating Vectors. Journal of Business and Economic Statistics 19(1): 63-72. 16 17 In, F. and Mount, T. (1994). Dynamic Macroeconomic Linkages to the Agricultural Sector. Athenaeum Press Ltd, Newcastle upon Tyne. Johansen, S. (1988). Statistics Analysis of Cointegration Vectors. Journal of Economic Dynamics and Control 12: 231-254. Johansen, S. and Juselius, K. (1990). Maximum Likelihood Estimation and Inference on Cointegration-with Applications to the Demand for Money. Oxford Bulletin of Economics and Statistics 52: 169-210. (1992). Testing Structural Hypotheses in a Multivariate Cointegration Analysis of the PPP and the UIP for UK. Journal of Econometrics 53: 211-244. (1994). Identification of the Long-Run and the Short-Run Structure: An Application to the ISLM Model. Journal of Econometrics 63: 7-36. Juselius, K. (1994). On the Duality between Long-Run Relations and Common Trends in the I(1) Versus I(2) Model: An application to aggregate money holdings. Economic Reviews 13(2): 151-178. (1995). Do Purchasing Power Parity and Uncovered Interest Rate Parity Hold in the Long Run? An Example of Likelihood Inference in a Multivariate Time-Series Model. Journal of Econometrics 69: 211-240. (1998). A structured VAR for Denmark Under Changing Monetary Regimes. Journal of Business and Economic Statistics, 16(4): 400-411. (1999). Models and relations in Economics and Econometrics. Journal of Economic Methodology 6(2): 259290. Kwiatkowski, D., Phillips, P., Schmidt, P. and Shin, Y. (1992). Testing the Null Hypothesis of Stationarity Against the Alternative of Unit Root. Journal of Econometrics 54: 159-178. Larue, B. and Babula, R. A. (1994). Evolving Dynamic Relationships between the Money Supply and Food-Based Prices in Canada and the United States. Canadian Journal of Agricultural Economics 42: 159-176. Longmire, J. and Morey, A. (1983). Strong Dollar Dampens Demand for US Farm Exports. FAER Report nº 123, USDA, Washington DC. Lütkepohl, H. (1993). Introduction to Multiple Time Series, Spring Verlag, Berlin. Ng, S. and Perron, P. (2001). Lag Length Selection and the Construction of Unit Root Tests with Good Size and Power, Econometrica 69, 1519–1554. Pesaran, M.H. and Shin, Y. (1996). Cointegation and speed convergence to equilibrium. Journal of Econometric 71:117-143. Pesaran, M.H. and Shin, Y. (1998). Generalised Impulse Response Analysis in Lineal Multivariate Models. Economics Letters 58: 17-29. Pesaran, M.H., Shin, Y. and Smith, R.J. (2000). Structural Analysis of Vector Error Correction Models with Exogenous I(1) Variables. Journal of Econometrics 97(2): 293-343. Reimers, H.E. (1992). Comparisons of Tests for Multivariate Cointegration. Statistical Papers 33: 335-359. Schuh, G. E. (1974). The Exchange Rate and U.S. Agricultural. American Journal of Agricultural Economics 56:1-13. Sriram, S. S. (1999). Demand for M2 in an Emerging-Market Economy: An Error-Correction Model for Malaysia. Working Paper WP/99/173. International Monetary Fund. Washington. Taylor, J. S. and Spriggs (1989). Effects of the Monetary Macroeconomy on Canadian Agricultural Price. Canadian Journal of Agricultural Economics 22: 278-289. Thraen, C. S, Hwang, T. C. and Larson, D. W. (1992). Linking of U.S. Monetary Policy and Exchange Rates to World Soybean Markets. Agricultural Economics 6: 365-384. APPENDIX Variable Symbol Source Units Exchange rate ER International Monetary Fund (IMF) Tunisian Dinars/ US$ Interest rate R Banque Centrale de Tunisie. Percentage Money Supply M International Monetary Fund (IMF) Million dinars Consumer Price Index P International Monetary Fund (IMF) Index (1990 = 100) Gross Domestic Product GDP International Monetary Fund (IMF) 1990 Million dinars Rate of Commercial Openness RCO International Monetary Fund (IMF) Percentage Farm output prices PP Institut National de la Statistique. Tunisia. Index (1990 = 100) Farm input prices IP Institut national de la Statistique. Tunisia. Index (1990 = 100) Agricultural exports AX FAO. 1990 Million dinars Agricultural Output AP Institut national de la Statistique. Tunisie. 1990 Million dinars