Agricultural Productivity and technological gap between MENA ...

erf.org.eg

Agricultural Productivity and technological gap between MENA ...

Agricultural Productivity and technological gap between MENA region andsome European countries: A Meta Frontier ApproachMohamed Mekki Ben Jemaa *andMohamed Adel DhifLEGI-Polytechnics of TunisiaBP 743 2078 LA MARSA TUNISIATel. 00 216 71 77 46 11Fax. 00 216 71 74 88 43Preliminary vertionSeptember 2005AbstractThe paper provides a database of Total Factors Productivity growth, technical efficiency, andinput productivity for 12 MENA region’s countries and their potential competitors in term ofagricultural products into the European market. Using the metafrontier approach (Rao et al,2003), technical efficiency scores are corrected by the coefficient of technology gap sinceproduction technologies are different in the two regions. The effects of some salientdeterminants of technical efficiency are assessed in order to identify the reasons ofdiscrepancies between these two regions.Key words: Agriculture, Metafrontier, technological gap.JEL classification: C23, D24, O57, Q11* Corresponding author: Mohamed Mekki Ben Jemaa. e-mail: mekki.bendjemaa@ept.rnu.tn


IntroductionMany MENA region countries have started to liberalize agricultural sector after thesignature of the GATT agreement, and have taken part in the trade talks on agriculture heldunder the auspices of the WTO at the end of 1999. They have also engaged in a partnershipwith the European Union (EU). The agreement of association between MENA region and theEuropean Union aims at the progressive creation of a Euro-Mediterranean economic spacethat guarantees a free circulation of goods, capital and services.The need to assess the MENA region’s agricultural performances, and thereof its potentialsto compete with southern Europe countries and especially the new central and easternEuropean Union members (and future members) is straightforward. Performances are, in theframework of this work, depicted through the discrepancy in term of productivity growth,technical efficiency, technological gaps and inputs’ elasticities between MENA region and asample of the enlarged EU countries that are its potential competitors for agricultural productsinto the European market.Earlier research has identified negative productivity growth in agriculture in developingcountries. These results imply that recent growth was based mainly on an increase in resourcebase instead of an increase in the quality and efficiency of resource use and the adoption ofnew techniques. A focus on these two geographical areas will help identify characteristics ofthis evolution specific to geographical, social, or political circumstances of the MENAregion’s countries.Up to now, no studies have been dedicated exclusively to depict agricultural efficiency forMENA region. Earlier studies by Fulginiti and Perrin (1998) identified negative productivitygrowth in a set of 18 developing countries, lending support to results obtained earlier byKawagoe, Hayami, and Ruttan (1985) in a study that showed productivity regression in 22LDCs, but an increase in productivity in the 21 developed countries included. Analyses byLau and Yotopoulos (1989) also found declining agricultural productivity in the 1970s,although they used different functional forms and data. More recently, Suhuriyanto, Lusigi,and Thirtle (2001) found negative agricultural productivity growth rates in Asia from 1965 to1980 and in Africa from 1971 to 1981. They also show the rates in both regions improving insubsequent years.The analysis will examine changes in agricultural productivity in 12 MENA region’scountries (Morocco, Algeria, Tunisia, Libya, Egypt, Sudan, Jordan, Syria, Lebanon, Turkey,Iraq and Iran) and a subset of the enlarged EU (Portugal, Spain, France, Italy, Greece, Cyprus,Czech and Slovak Republics, Hungary, Poland and Romania) and Bulgaria.It is a common practice to use production frontiers to assess the level of efficiency ofproduction units or firms. Such frontiers are identified using non-parametric or parametricmethods predicated on various non-stochastic and stochastic assumptions. While technicalefficiencies of units that are measured with respect to a given frontier are comparable, this isnot normally the case among units that operate under different technologies. Such problemsarise when comparisons of units from different regions are involved.The principal objective of this paper is to derive technical efficiency scores of each countryof the studied sample using metafrontiers (Rao et al, 2003). The metafrontier concept used inthis paper is based on the concept of the metaproduction function defined by Hayami andRuttan (1971, p. 82): “the metaproduction function can be regarded as the envelope ofcommonly conceived neoclassical production functions.”The stochastic frontier analysis (SFA) is used to compare efficiencies of the agriculturalsector in regions that definitely operate under different technologies. The empirical part of the2


paper focuses on inter-region comparisons of productivity in agriculture using country-leveldata drawn from the Food and Agriculture Organization (FAO) of the United Nations.Countries are grouped into two regions using the geographical classification stated above.1. The Metafrontier approachNumber of authors such as Aigner, Lovell, and Schmidt (1977), and Meeusen and Van derBroeck (1977) developed the “stochastic composed error frontier methodology”. The mainprinciple of the model is to specify the error term as a sum of two parts, one normal and theother from a one sided normal distribution 1 .The specification of the stochastic production frontier model, allows for a non-negativerandom component in the error term to generate a measure of technical inefficiency or theratio of actual to expected maximum output, given inputs and the existing technology. Apartfrom allowing for the measure or assessment of technical inefficiency, stochastic frontiermodels also acknowledge the fact that random shocks outside the control of producers canaffect the output level. This is due to the fact that stochastic effects such as weather conditionsamong others could cause variations in maximum output. The variations in output could alsooccur as a result of firms in an industry operating at various levels of inefficiency due to poorincentives, mismanagement, inappropriate input levels or less than perfectly competitivebehavior (Kumbhakar and Lovell 2000).While technical efficiencies of units that are measured with respect to a given frontier arecomparable, this is not normally the case among units that operate under differenttechnologies. Such problems arise when comparisons of units from different regions areinvolved. The metaproduction function concept is based on the hypothesis that all producersin different groups have potential access to the same technology. However, each producermay choose to operate on a different part of it depending on circumstances such as the naturalendowments, relative prices of inputs, and the economic environment (Lau and Yotopoulos1989). Recent extensions and modification of the stochastic frontier metaproduction functionapproach is found in Battese and Rao (2001), which is reviewed below.Let consider that the whole sample is composed of K subsets (K>1) representing K regionskand firms in each region operate under a region specific technology, T (k = 1, 2, …, K).Since all the K’s technology can be considered as a subset of an “over-arching technology”*referred to as the metatechnology, which is represented by T . The metatechnology can be statedas the “absolute best” technology produced by the state of the knowledge. This leadingtechnology is accessible if neither endowments nor policy constraints are facing producers.Battese and Rao (2001), showed how technical efficiency scores for firms across regionscan be estimated using a stochastic frontier metaproduction function model, and used adecomposition result to present an analysis of regional productivity potential and efficiencylevels.If stochastic frontier models are defined for different regions within an industry, and forthe kth region, there exist sample data on N kfirms that produce one output from the variousinputs. The stochastic frontier model for this region is specified as 2k vi( k) −ui( k)i(i, β ) , 1,2,...kY = f x e i = N(1)1 or any other probability distribution providing that their support is2 Time subscript is dropped for convenience and the logarithmic linear form for the production function isassumed among the paper.+R3


kwhere Y stands for the output level, x for the vector of inputs’ levels. β is the vector ofparameters for the kth region. vi( k)’s are iid N(0, σvk ( )). uik ( )is assumed a truncation at zeroof the N ( µi( k) , σu( k)) random variable.The metafrontier production function model for units in the industry is expressed by:Y = f x i = N t = T(2)*it(it, β ), 1,2,...k; 1, 2,...where*β denotes the vector of parameters for the metafrontier function such thatx β ≥ x β(3)it* kit(3) states that the metafrontier dominates all the regional frontiers. That is themetatechnology describes the unconstrained best practice according to the state of knowledge.The observed ouput for any unit at any time period can be consequently defined by thestochastic frontier for the kth region (equation (1)) and using the metafrontier function (2),such that:kxitβ*− u eit ( k ) xitβ+ vit ( k )*xitβYit= e e(4)ewhere the first term on the right-hand side of equation (4) is the technical efficiencyrelative to the stochastic frontier for the kth region. The second term on the right-hand side ofequation (4) is the technology gap ratio (TRG) for the ith unit in the kth region at the tth timeperiod.xitβk eTRGit=*(5)xitβeThe third right-hand side of equation (4) is a pseudo stochastic metafrontier since the error*xterms of the K regions vit( k)is added to the deterministic metafrontier e β it. Dividing bothterms by the pseudo stochastic metafrontier, one can retrieve the famous equality presented byRao, Odonnell and Battese, (2003):k* kkTE ( x, y) = TE ( x, y) TGR ( x, y)(6)it it itEquation (6) implies that the technical efficiency ratio of the ith unit at the tth time periodrelative to the metafrontier (which is a common feature to all the K regions) is the product ofthe technical efficiency ratio relative to the regional frontier and the technological Gap Ratiorelative to the metafrontier. In other words, the technical efficiency scores for units that don’tproduce under the same technology can be corrected (to make them comparable) using thedistance between the regional frontier and the leading metafrontier. For example, if TE=0.9then the technical efficiency measure indicates that the observed output is 90 per cent of the4


potential output, using the same input vector. If TGR is 0.6 that is, given the input vector, thepotential output vector for region k technology is 60 per cent of that represented by themetatechnology. Then, using (6), the Technical Efficiency with respect to the metatechnologyis 0.63. Note that this implementation precludes the technical efficiency with respect to themetatechnology to be greater than the technical efficiency measure relative to the regionalfrontier.The parameters and measures associated with the metafrontier model specified above canbe estimated using the following stapes:Step 1. Obtain the maximum-likelihood estimates of parameters of the stochastic frontier forkthe kth region ( β ).Step 2. Obtain estimates for the parameters of the metafrontier function β * such that theestimated function best envelopes the deterministic components of the estimated stochasticfrontiers for the different regions. Estimates of β * can be found using either linear orquadratic programming. For the edge of this paper the linear programming solver is chosenand the optimization problem is the following:* ˆ kMin ( xitβ− xitβ)*.ˆ kst xitβ≥ xitβfor all k = 1, 2,..., KThe solution to the above problem is equivalent to:*Min xβ≥*st . xitβxitˆ kβfor all k = 1, 2,..., K(7)where x is the row vector of means of the elements of the xitvectors for all units in the dataset. This follows because the estimates of the stochastic frontiers for the different regions areassumed to be fixed for the linear programming problem.Step 3. Technical Efficiency scores relative to the metafrontier are estimated making use ofequation (6) and estimates obtained in steps 1 and 2.2. Empirical modelThe functional form chosen for the stochastic frontier function for all the K regions is theTranslog form. The choice of this functional form is driven by the amount of flexibility that itaffords in depicting changes through the data time span in estimates of the technologicalchange and factors elasticity. Indeed, since translog form is a second order approximation ofthe (unknown) production function, it provides the opportunity to estimate a second order(rather than linear) approximation of marginal effects.The translog function for the ith unit in the kth region at the tth time period is the following:5


where yitand thethe time trend andM M Mk k 1kit= β0 + ∑βm it+2∑∑βmh mt htm=1m hMk k 1 k 2k k k+ ( βt + βc confit ) t +2βtt t ++ ∑ βmt xitt + vit −uitm=1y x x xmxit’s are the logarithm of output and inputs levels respectively, t stands forkvitandkuitare specified above.(8)confitis a proxy variable of the intensity ofan armed conflict (if there is) in a country (further details on this variable are granted in thenext section). This variable is associated to the time trend in order to assess possible shift inthe stochastic frontier caused by armed conflicts. In other words, this implementationprovides the possibility to depict the effects of armed conflicts on the technological changekthrough the parameter βc.Determinants of Technical EfficiencyBattese and Coelli (1995) suggested that the technical inefficiency effects, uitin (1) could bereplaced by a linear function of explanatory variables reflecting country-specificcharacteristics. This approach gives the opportunity of accounting for resources qualitydifferences across countries and socio-political and institutional differences that might haveaffected behavior.The technical inefficiency effects are assumed to be independent and non-negativetruncations (at zero) of normal distributions with unknown variance and mean. Specifically 3 ,mjuit = δ0+ ∑ δjzit + ωit(9)whereinefficiencies; δ0andmean and finite variancej=1jzit’s are farm and time specific explanatory variables associated with technicaljit 0 j j itδjare parameters to be estimated and ω is a random variable with zero2σωdefined by the truncation of the normal distribution such thatjω ≥− ( δ +∑ δ z ). This implies that the means, µit= δ0+∑δjzitof the uitaredifferent for different units but the variance but share the same variance.Two routes are possible in investigating the determinants of technical efficiency variationamong the units in the sample. The two stage approach is the most intuitive and involves theestimation of the technical efficiency effects from both models and regressing these on a setof unit specific characteristics. This approach, though widely used, implies that theinefficiency effects which are assumed to be independently and identical distributed in theestimation of the stochastic frontier are a function of the units’ specific effects in the secondstage, thus violating the assumption that the efficiency effects are identically distributed(Battese and Coelli., 1995). The inefficiency effects would only be identically distributed ifthe coefficients of the farm specific factors are simultaneously equal to zero (Coelli et al.,1995). It is possible to overcome this problem by the use of a single stage maximumlikelihood approach (Battese and Coelli, 1995). The technical inefficiency effects arespecified to be a function of a vector of unit specific variables and a random error as in (8)and a single stage maximum likelihood estimation technique is used to estimate theparameters of the production frontier and of the technical inefficiency variablessimultaneously.mj=13 Region index subscript is dropped for convenience.6


Total Factors ProductivityThe rate of Technological Change (TC) can be obtained from the tronslog functional form.Using the definition of TC which is an autonomous change in the frontier and observablethough time, it can be derived from (8) as:k ∂yitk k k m kTCit = = βt + βtt t + ∑ βmt xit + βc confit(10)∂tmThe rate of Technological Change indicates the shift (in percentage) of the production giventhe level of inputs x m it. An estimate of TC will give an idea of the additional output produceddue to technological innovation while inputs’ level is held constant. The last term in the rightsidehand of (9) depicts the effect of armed conflicts on the technological change as statedabove.The rate of total Factors Productivity (TFP) is defined as the rate of change in output that isnot explained by the input change which is the sum of Technological Change and efficiencychange (EC):*k k TEitk kTFP & ∂it= TCit + = TCit + ECit∂t(11)where the right-hand side of (11) is the rate of change of the technical efficiency score relativeto the metafrontier. Note that this expression is empirically evaluated by the discreetapproximation of the rate of growth. Technical efficiency change, EC, is the rate at which acountry moves toward or away from the production frontier, which itself shifts through timeas measured by TC.3. DataThe present study is based on a panel data exclusively drawn from the FAOSTAT systemof statistics used for dissemination of statistics compiled at the Statistics Division of the Foodand Agriculture Organization. We used data from 1972 to 2002 of two regions: 12 MENAregion’s countries (Morocco, Algeria, Tunisia, Libya, Egypt, Sudan, Jordan, Syria, Lebanon,Turkey, Iraq and Iran) and a subset of the enlarged EU (Portugal, Spain, France, Italy, Greece,Cyprus, Czech and Slovak Republics, Hungary, Poland and Romania) and Bulgaria. Theanalysis is based on a balanced panel for the MENA region and an unbalanced panel for theEU region because Czech and Slovak Republics are integrated to the sample only in 1993, thefirst year in which their respective data were published separately.Detailed descriptions of variables are given belowOutput (Y), as the value of agricultural production in millions of 89-91 “international” dollars.We consider five important input variables.Land (A): This variable includes the arable land, land under permanent crops as well as thearea under permanent pasture, expressed in millions of hectares.Tractors (Tr): This variable includes the total number of wheeled and crawler tractors used inagriculture.Labour (L): The labour variable used is the number (in thousand) of economically activepopulation in agriculture.7


Fertilisers (F): in thousands of metric tons of nutrient units.Livestock (Lv): The livestock input variable used in the study is the sheep-equivalent of sixcategories of animals. The categories of animals considered are buffaloes, cattle, camels, pigs,sheep and goats. Data on numbers of these animals are converted into sheep equivalents usingthe following conversion factors: 8 for buffalo and cattle; and 1 for sheep, goats and pigs.As stated above, the model assumes that the efficiency effects are a function of a vector ofexplanatory variables (Battese and Coelli, 1995) that are:Soil quality (SQ): The following factors were taken as the basis for country-level rankingperformed by the Land and Water Development Division FAO (Bot, Nachtergaele andYoung, 2000):- Major soil constraints (Hydromorphy, Aluminium Toxicity, Vertic Properties, High Pfixation, Salinity, Sodicity, Shallowness, Erosion Risk)- Deserts and dryland areas- Population distribution in deserts and dryland areas- Steeplands- Land degradation severity- Human-induced land degradation due to agricultural activities- Land degradation severity and population distribution- Actual and potential available arable land.The overall rankings indicate countries with most favourable conditions (low rank numbers)or with most severe problems (high rank numbers), with respect to physical resource potentialand constraints, now and in the future.Rating for political rights and civil liberties (FREE): this variable represents an average valueof scores for each year and each country for political rights, civil liberties. Each score ismeasured on a one-to-seven scale, with one representing the highest degree of freedom. Thesescores are compiled by the Freedom House.Literacy rate (LIT): this variable is defined according to the World Bank as the Adult literacyrate is the proportion of people aged 15 and above who can, with understanding, read andwrite a short, simple statement on their everyday life.Armed conflicts (CONF): is an ordinal discreet variable indicating the existence and theintensity of an armed conflict in each country for each time point of the sample. An armedconflict- as defined by the International Peace Research Institute, Oslo (PRIO) is a “contestedincompatibility that concerns government and/or territory where the use of armed forcebetween two parties, of which at least one is the government of a state, results in at least 25battle-related deaths. The variable takes the null value if there is no armed conflict, the value1 if there is a minor conflict (More than 25 battle-related deaths per year for every year in theperiod) and the value 2 if there are More than 1000 battle-related deaths per year for everyyear in the period of the conflict. More details are available at Strand et al. (2004).Agricultural import relative to total import (RM): defined as the proportion of Agriculturalimport relative to total export.Agricultural export relative to total export (RX): defined as the proportion of Agriculturalexport relative to total export.8


Irrigated land (IR): defined as the natural logarithm of the area under irrigation in 1000 ha.The terms of trade (TT): defined as the of the export-import unit values ratio. The variable isan index with 1995 as base year 4 .The summary statistics presented in Table 1 indicate some differences in the means andstandard deviations between the two groups of countries with regards to both output andinputs. The standard deviations for all the variables are higher than the means indicating widespread around the mean of the variables. The mean values are higher for EU region thanexcept for land and labor.Furthermore, the summary statistics for efficiency variables, on the average, showconsiderable differences between the two regions. The standard deviations for all thesevariables are lower than the means indicating no significant variations intra-region.4. Estimation and resultsThe single-stage maximum-likelihood procedure of the FRONTIER 4.1 program (Coelli,1996a) was used to estimate the parameters of the stochastic frontiers and the efficiencydeterminants for each region and for pooled data. Results are presented in Table .2Two sets of specification tests were performed. In the first the null hypothesis that there areno technical efficiency effects in the models is tested using a likelihood ratio test of the onesided error. The null hypothesis is strongly rejected as the LR test statistics 372.03, 90.89 and40.41 for MENA SF, EU SF and pooled SF respectively which are all greater than the(mixed) Chi-square value of 3.84. The Stochastic Frontier specification was then held for thethree cases at the 1 percent significance level.As indicated above, Table .2 presents estimates of the parameters of the stochastic frontiersfor all the regions separately and also using pooled data. Note that if the stochastic frontiersacross regions do not differ, then it is possible to just use the pooled stochastic frontier.Therefore a LR test was performed in order to see whether regional frontiers hold 5 . Thegeneralised likelihood-ratio test statistic for the null hypothesis that the regional frontiers areidentical is LR = 629.84 which reject robustly the null hypothesis that the regional frontiersare the same. Based on this, the parameters of the metafrontier are estimated by solving theLP problem. Then, The LP problem in (7) was solved using Matlab software. Since theprogram failed to minimize the original target function, we were obliged to drop sevencovariates which are the quadratic trend, input-trend and the armed conflicts-trendinteractional variables. Even by this restriction, the metafrontier continue to lead regional*frontiers since constraints x ˆ kitβ≥ xitβare respected.The average production elasticities of inputs are presented in Table .3. Discrepancies in inputsproduction elasticities between the two regions imply that production structure and4 Note that the terms of trade was estimated for some time points for some country using available informationon the terms of trade effect. The terms of trade effect equals capacity to import less export of goods and servicesin constant prices. Note that for some missing value when no information concerning the terms of trade effectwas available, the near neighbor estimation procedure was used. Estimating missing values are performed inorder to avoid unbalanced panel data and then observation loss.5 The LR Statistic is defined by λ =−2⎡⎣ ln ( LH ( ) ( )0) −ln LH (1)⎤⎦ . Where L( H0)is the value of the likelihoodfunctions for the stochastic frontier estimated by pooling the data for all the two groups, and LH (1)is the sumof the values of the likelihood functions for the two regional stochastic production functions estimatedseparately.9


technology are different. While the greatest share of productivity is due to the land input inthe European region (0.930), land elasticity of production for the MENA region was negativewhich highlights the importance of difference in soil quality. Notable difference in inputproduction elasticities was also observed for Livestock and Labor.Explaining Technical efficiencyAnother empirical result of interest is the nature of the efficiency change, as reflected in theestimates of δ from equation (9). We can see in Table .2 that all the coefficients on theefficiency variables for the MENA region are significant and with the right sign excepting thecoefficient of the political rights and civil liberties (FREE) variable which found to be nonsignificant. It is found that literacy, agricultural export and irrigation seem to have the greatesteffects on technical inefficiency reduction. Agricultural export exposes the producers in acountry to international competitiveness which spurs efficient production process. BesidesAgricultural import is a sign of a problematic agricultural sector. An increase in the terms oftrade reduces inefficiency. This finding implies that any increase of the export unit value (orequivalently any decrease of the import unit value) enhances agricultural efficiency. Foragricultural exporter countries this phenomenon is straightforward if any amelioration in theterms of trade are driven basically by an increase in the prices of exported agriculturalcommodities. Armed conflicts seem to have a pronounced role in efficiency reduction inMENA region. Indeed, a minor conflict increases inefficiency by 4.5 percent. In the case ofwar the effect in doubled to reach 9 percent. Figure 1 plots series of technical efficiencyestimates relative to the metatechnology of Iran and Iraq. Two series are plotted for eachcountry: observed technical efficiency and potential technical efficiency if there was no armedconflicts.For the EU region, soil quality, political rights and civil liberties, Literacy, and irrigation havethe theoretical right sign and significant. Armed conflicts have not the right sign which islikely due to the countries selected in the sample.Technical efficiencyThe mean values for the regional technical Efficiencies (TE), metafrontier technicalefficiencies (TE*) and the TGR’s are given in Table .4. From the results, EU region countriesseem to be more efficient than MENA region countries.Table .4 provides average technical efficiency scores for each of the regions. For the MENAregion, the average technical efficiency score is about 0.7, indicating agricultural output byabout 70 per cent of the potential, given its regional technology. In the EU region, the meanefficiency is quite higher with respect to its own regional frontier (0.89). Turkey, Iran andSyria are the top three highly efficient countries in the MENA region. For the EU region, itwas surprising to see two Eastern Europe countries in the top Three of the regional efficiencyperformances. Indeed, Italy, Romania and Slovak Republic, seem to be the most efficientrelative to their regional frontier. Note that France is positioned in the tail of the sample whichimplies that it didn’t achieved best practice with respect to the technology observed in the EUregion.Average technical efficiency scores from the regional and metafrontier models shows nodramatically change in the counties’ rank for the MEA region except for Syria since itsaverage TGR was only 0.52 which implies that its technology is far from the overalltechnology.As expected southern European countries, Spain, Italy, France and Greece seems to be themost efficient relative to the metatechnology. Within these results, it is straightforward tostate that the technology gap ratio is the determinant component of the metafrontier technicalefficiency. Southern Europe countries (Italy, Spain, Greece and France) keep the top 4 of the10


egion as they have the highest technology gap ratios in the two regions along with Turkey.It is worth to put the stress on the low level of the metafrontier technical efficiency scores forLebanon, Iraq and Jordan that are about 0.24, 0.2 and 0.16 respectively. Note that thesecountries have low levels for both regional technical efficiency and technological gap ratioimplying that there are structural problems in their agricultural sectors.ConvergenceAs stated above, the comparison of the averages of the metafrontier technical efficiency, theregional technical efficiency and the Technological gap ratio for MENA region and EU regionshows that the EU region was definitely more efficient during the period 1972-2002. In orderto depict any cutch-up process especially in the latest years of the period, time series formetafrontier technical efficiency, regional technical efficiency and Technological gap ratio forMENA region and EU region were calculated by averaging scores of each region for everyyear. Figures .2-4 shows plots of average scores of these variables. Regional technicalefficiency series (Figur.3) shows clearly a convergence process between the two regionswhich has been accelerated during the 90’s. However, this implies only a convergence in termof regional efficiency rate since the two regions produce under different technology. Then, itwill be more informative to test convergence through the metafrontier efficiency scores whichare definitely appropriate to perform regional comparison. Even if Figure .4 shows a sightcutch-up process, three ranges of tests are performed in order to confirm whether there is ornot a convergence.Entropy index: the Entropy index is an alternative to the sigma-convergence index (betweenregionvariance) and it was preferred to the later since it provides more smoothed plots.- The Entropy Index in the case of two regions is:____ 1____ 0EIt⎛⎜TE= ln ⎜⎜⎜TE⎝____ 1t____ 0t⎞⎟⎟⎟⎟⎠where TE t and TE t are averages of metafrontier technical efficiency scores of the samplecounties at period t for regions 1 and 0 respectively. For convenience, the value at thenumerator ought to be the greater in order to get positive values of the Entropy index. So 1will stands for EU region and 0 for MENA region. As Figure .5 shows, the Entropy Index Isdecreasing along the time period implying that a cutch-up mechanism is engaged between thetwo regions.- Parametric Test: Two sample means comparison tests where performed for the five last yearof the time period and the hypothesis that the two sample means metafrontier technicalefficiency scores of are equal for 1998 to 2002 at the 1 percent significance level.- Mann-Whitney two sample rank sum test: since samples’ sizes for the two regions is small.Non parametric tests ought to be more consistent than parametric tests. Then Mann-Whitneytwo sample rank sum test is performed for the same time span than parametric test (Table .6).Results seem to conform that no significant difference between the metafrontier technicalefficiency scores of the two regions.Figure .4 plots the average of the Technological Gap Ratio of the two regions through samplespan. It can be shown in Figure .4 that before the early 80’s the two regions were operating atthe same technological level (around 65 percent). But since the mid 80’s, the gap betweenMENA region and EU region has been deepen.(12)11


Total Factors ProductivityAverage Total Factors Productivity growth for the sample time is calculated for eachcountry of the two regions (Table .7). In MENA region, only Libya, Lebanon and Egypt haverealized positive average annual growth of TFP. Six countries out of twelve had more thanone percent cut in their TFP. Note that Jordan has endured a drastic TFP deterioration duringthe sample period (about 6 percent decrease per year). This situation is likely imputed towater availability problems since the 1967 war.In order to observe the spread of the gap separating the two regions during the sample period,country-average Total Factors Productivity are plotted in Figure .6 with 1972 as a base year.5. ConclusionThe paper provided a database of Total Factors Productivity growth, technical efficiency, andinput productivity for 12 MENA region’s countries and their potential competitors in term ofagricultural products into the European market. Using the metafrontier approach (Rao et al,2003), technical efficiency scores are corrected by the coefficient of technology gap sinceproduction technologies are different in the two regions. The effects of some salientdeterminants of technical efficiency are assessed in order to identify the reasons ofdiscrepancies between these two regions. It was shown that technological gap was the mainactor playing in favor of European countries. In other words, these countries are operatingnear the possibilities’ frontier of the meta-technology. Despite this handicap, a cutch-upprocess is observed between the two regions in term of technical efficiency.TFP growth for each country in each year in the two regions was defined as the sum ofpredicted change in the production frontier in that vicinity plus predicted change in technicalefficiency for that country and year. The Battese-Coelli approach was used to estimate theefficiency effects of a set of institutional, trade, and other efficiency-changing variables, withthe production frontier specified as tranlog flexible form. We found that literacy rate irrigatedarea and agricultural exports have a considerable effect on efficiency alleviation in MENAregion. Wars and civil conflicts have considerably reduced agricultural performances ofseveral countries in the sample such as Iraq, Iran and Lebanon. The model adopted wasreshaped to account for the effect of Wars and civil conflicts on both frontier and technicalefficiency. That is, Wars and civil conflicts reduce the access to technology and induceoperators to be less efficient.In conducting, this study has characterized the current situation in MENA countries in term ofproductivity and efficiency. It is worth to examine the implementation of the Vision to Actionstrategy in MENA, develop a series of lessons learned from this work among others and thendesign strategies and policies for the development of of MENA’s “rural space,” which is thebase of the agricultural sector. The strategies and policies should grant a great deal of interestto natural resource management, rural transport, water and sanitation, telecommunication,education, health and other social services.- Rationalize water management and policies: Water is a precious and increasingly scarcecommodity in MENA region. Countries in the region should design a common policy inirrigation, water policy, water demand management, and water resource allocation. Closelyrelated to water are all the issues dealing with sustainable use of natural resources.- Improve access to social and economic infrastructure.12


- Enhance agricultural lending activities: it will be of a great interest to encourage privatesector to be a financial partner of the agricultural activity through lending and associationmechanisms.- Improve natural resource and environmental management.- Design a coherent common strategy in the MENA region concerning best practices and newtechnologies extension.13


ReferencesBattese, G. E. and D. S. P. Rao. (2001). “Productivity Potential and Technical Efficiency Levelsof firms in Different Regions Using a Stochastic Frontier Metaproduction FunctionModel.” CEPA Working Papers No. 6, School of Economics, University of NewEngland, Armidale.Battese, G.E. and D.S.P. Rao (2002), “Technology Gap, Efficiency and a StochasticMetafrontier Function”, International Journal of Business and Economics, 1 (2), 1- 7.Battese, G.E. and T.J. Coelli (1995), “A Model for Technical Inefficiency Effects in aStochastic Frontier Production Function for Panel Data”, Empirical Economics, 20, 325-332.Battese, G.E., D.S.P. Rao and C.J. O’Donnell (2004), “A Metafrontier Production Function forEstimation of Technical Efficiencies and Technology Gaps for Firms Operating UnderDifferent Technologies”, Journal of Productivity Analysis, 21, 91-103.Coelli, T. J. (1996a) “A Guide to FRONTIER Version 4.1: A Computer Program for StochasticFrontier Production and Cost Function Estimation.” working paper, Centre forEfficiency and Productivity Analysis (CEPA), Department of Econometrics, Universityof New England, Armidale, Australia, No. 7/96.Food and Agricultural Organization of the United Nations. FAOSTAT,http://apps.fao.org/page/collections?subset=agriculture.Freedom House. “Freedom in the World Country Ratings: 1972-72 to 2000-2001.”http://www.freedomhouse.org/ratings/index.htm.Fulginiti, L.. and R. Perrin(1998). “Agricultural Productivity in Developing Countries.”Agricultural Econonics 1945-51.Hayami, Y., and V. Ruttan. (1985). “Agricultural Development: An International Perspective”.Baltimore:Johns Hopkins University PressKawagoe, T., Y. Hayami, and V. Ruttan (1985). “The Intercountry Agricultural ProductionFunction and Productivity Differences among Countries.” Journal of DevelopmentEconomics (19)113-32.Lau. L.. and P. Yotopoulos. ( 1989) “The Meta-Production Function Approach and Change inWorld Agriculture.” Journal of Development Economics: 31 241-269.Strand H., Wilhelmsen L.and Gleditsch N.P. (2004), “Armed Conflict Dataset”, InternationalPeace Research Institute, Oslo (PRIO),Suhuriyanto, K.. A. Lusigi. and C. Thirtle. (2001) “Productivity Growth and Convergence inAsian and African Agricul1ure”. P. Lawrence and C. Thirtle, eds., pp. 258-273. Africaand Asia in Comparative Economic Perspective. New York: Palgrave.14


ANNEX 1: Tables and FiguresTablesTable 1. Summary Statistics of variables in the stochastic Productionfunction and efficiency covariatesMENAVariable Obs Mean Std. Dev. Min Maxy (M int $) 372 2 118 4815.394127 137 20 899Lv (number) 372 23 937 097 80117093.65 576 402 421 820 320Tr (number) 372 32 434 174956.2298 2 850 970 086A (M ha.) 372 11 674 33067.33077 308 133 898F( M kg) 372 149 132 496617.808 3 010 2 206 990l ( number) 372 1 352 3829.460895 43 14 697SQ 372 576.58 82.23 410 714free 372 5.53 1.15 2 7let 372 0.57 0.18 0.20 0.95conf 372 0.55 0.77 0 2rm 372 0.21 0.10 0.02 0.67rx 372 0.18 0.21 0 0.81ir 372 6.64 1.46 3.56 8.93term 372 106.10 24.92 31.28 210.01EUVariable Obs Mean Std. Dev. Min Maxy (M int $) 330 6 346 9716 175 35 984Lv (number) 330 31 164 806 57487973 648 003 216 592 876Tr (number) 330 184 687 525586 8 801 1 660 003A (M ha.) 330 7 544 10216 117 32 568F( M kg) 330 653 038 1519639 10 693 6 103 421l ( number) 330 1 037 1529 31 6 501SQ 330 375.26 70.19 272 540free 330 2.91 2.03 1 7let 330 0.94 0.05 0.74 1.00conf 330 0.02 0.17 0 2rm 330 0.12 0.05 0.03 0.29rx 330 0.17 0.12 0.02 0.64ir 330 6.38 1.54 3.18 8.24term 330 96.33 10.04 63.04 129.6315


Table .2 Parameters' EstimatesStochastic Frontier: Stochastic Frontier:MENAEUPooled estimation Metafrontiercoefficient Std-error coefficient Std-error coefficient Std-error coefficientintercept -16.708 5.503 21.946 2.376 -7.469 0.997 -14.708LV 2.799 1.054 -0.107 0.875 1.498 0.811 1.132TR 1.676 0.354 -7.283 0.825 0.798 0.920 -2.252A -1.051 0.594 -2.780 0.807 -2.477 0.948 2.781F 1.745 0.486 5.789 0.857 -0.026 0.091 1.584L -4.311 0.983 0.083 0.774 1.230 0.095 -0.548.5LV² -0.140 0.114 0.229 0.175 -0.182 0.400 0.336.5TR² -0.143 0.038 -0.101 0.058 0.162 0.633 0.021.5A² 0.082 0.056 -2.652 0.208 -0.434 0.631 0.108.5F² 0.023 0.034 0.395 0.073 0.068 0.470 -0.059.5L² -0.453 0.087 0.100 0.082 -0.005 0.752 -0.290lv*tr -0.019 0.038 0.430 0.105 -0.177 0.045 -0.100lv*A 0.004 0.070 0.564 0.112 0.357 0.050 -0.484lv*F -0.190 0.049 -0.779 0.118 0.098 0.200 -0.041lv*L 0.259 0.090 -0.524 0.097 -0.117 0.231 0.006tr*A 0.034 0.036 0.687 0.072 0.287 0.473 0.233tr*F -0.070 0.032 -0.168 0.050 -0.072 0.569 0.065tr*L 0.120 0.052 -0.396 0.051 -0.135 0.318 0.091A*F 0.019 0.032 0.223 0.088 -0.266 0.046 -0.016A*L -0.043 0.037 0.917 0.107 0.084 0.272 0.240F*L 0.243 0.051 0.358 0.066 0.141 0.522 -0.068t -0.226 0.028 0.121 0.024 -0.026 0.753 0.007.5t² 0.000 0.000 -0.001 0.000 0.001 0.001 -lv*t 0.020 0.002 -0.002 0.001 0.010 0.082 -tr*t 0.005 0.002 -0.001 0.001 -0.007 0.002 -a*t -0.006 0.002 -0.001 0.001 -0.010 0.022 -f*t 0.002 0.001 -0.001 0.001 0.005 0.023 -l*t -0.017 0.003 -0.003 0.001 -0.005 0.051 -T*conf -0.001 0.001 -0.002 0.001 -0.001 0.008 -Determinants of efficiencyintercept 2.193 0.16 1.207 0.34 0.023 0.623 -SQ 0.002 0.00 -0.002 0.00 0.009 0.001 -free 0.015 0.02 -0.004 0.00 0.004 0.102 -lit -1.587 0.17 -0.002 0.00 -0.056 0.511 -conf 0.045 0.02 -0.059 0.03 -0.030 0.472 -rm 0.771 0.12 -0.212 0.40 0.179 0.856 -rx -1.277 0.14 -0.172 0.28 -0.147 0.279 -ir -0.309 0.02 -0.340 0.08 0.003 0.068 -TT -0.002 0.00 0.010 0.00 0.000 0.002 -2σ 0.025 0.00 0.056 0.01 0.067 0.002 -γ a 0.968 0.01 0.788 0.06 0.102 0.003 -Log L 297.67 192.84 -17.252 -2a: σuγ =2 2σ + σuv16


Table .3 Production ElasticitiesMENA EULV 0.188 -0.023TR 0.280 0.037A -0.029 0.930F 0.028 0.060L 0.280 -0.138Sum 0.747 0.866Calculated at the sample meanTable .4Technical efficiency and Technological Gap (decreasing ranking)TE TE* TGRMENATUR 0.976 TUR 0.827 IRN 0.865IRN 0.882 IRN 0.762 TUR 0.848SYR 0.873 EGY 0.633 DZA 0.823EGY 0.859 TUN 0.550 EGY 0.745TUN 0.787 MAR 0.491 TUN 0.705MAR 0.727 SYR 0.456 MAR 0.674SUD 0.703 SUD 0.435 LYB 0.666LIB 0.668 DZA 0.383 SUD 0.618IRQ 0.662 LYB 0.362 SYR 0.523LYB 0.541 LIB 0.242 JRD 0.466DZA 0.471 IRQ 0.196 LIB 0.369JRD 0.374 JRD 0.156 IRQ 0.300Overall averages 0.710 0.408 0.633EUITA 0.959 ITA 0.884 ESP 0.923ROM 0.948 ESP 0.863 ITA 0.922SVK 0.940 GRC 0.794 FRA 0.888ESP 0.936 FRA 0.767 GRC 0.855BGR 0.936 POL 0.736 POL 0.851GRC 0.930 SVK 0.735 SVK 0.784HUN 0.921 ROM 0.546 ROM 0.577PRT 0.913 HUN 0.525 HUN 0.570CYP 0.890 PRT 0.414 PRT 0.455POL 0.868 BGR 0.393 CZE 0.439FRA 0.867 CYP 0.381 CYP 0.430CZE 0.556 CZE 0.239 BGR 0.421Overall averages 0.889 0.607 0.676Averages are for the period 1972-2002.17


Table .5 Sample means comparison test____ 1____ 0Ha: TE t > TE tt Statistic p. value Decision1998 -0.3689 0.3580 Reject Ha1999 -0.6710 0.2527 Reject Ha2000 -0.5193 0.3030 Reject Ha2001 -0.5701 0.2856 Reject Ha2002 -0.5787 0.2828 Reject HaTable .6 Mann-Whitney two sample rank sum test____ 1____ 0Ha: TE t > TE tt Statistic p. value Decision1998 0.3460 0.5420 Reject Ha1999 1.0970 0.6320 Reject Ha2000 0.8080 0.5970 Reject Ha2001 1.1550 0.6390 Reject Ha2002 1.0390 0.6250 Reject HaTable .7Annual TFP growth (in %)(Decreasing ranking)MENAEULYB 4.41 BGR 2.97LIB 1.12 SVK 2.53EGY 0.08 PRT 2.49TUN -0.11 CYP 2.47IRN -0.25 HUN 2.37TUR -0.69 GRC 1.91ALG -1.09 ITA 1.10SYR -1.27 CZE 0.74MAR -1.48 ROM -0.21IRQ -1.67 FRA -0.33SUD -2.83 ESP -0.35JRD -6.48 POL -1.0618


FiguresFigure .1Base and no war-potential Technical efficiency for Iran and Iraq10.80.60.40.2Iran-baseIraq-baseIran-no warIraq-no war019721975197819811984198719901993199619992002Figure .2 Regional technical efficiency1.000.900.800.700.600.500.400.300.200.100.001.001972197419761978198019821984198619881990199219941996199820002002Figure .3 Metafrontier technical efficiencyMENAEU0.800.600.40MENAEU0.200.001972 1975 1978 1981 1984 1987 1990 1993 1996 1999 200219


Figure .4 Technological Gap Ratio0.800.700.600.500.400.300.200.100.00MENAEU19721975197819811984198719901993199619992002Figure .5 The Entropy Index for Metafrontier technical efficiency0.400.350.300.250.200.150.100.05EntropyIndexLineartrend0.0019721975197819811984198719901993199619992002Figure .6 Evolution TFP index16014012010080604020019721975197819811984198719901993199619992002TFP MENATFP EU20