Estiamting the Profit Efficiency of Contract and Non-Contract Rice Farms in Taiwan. A Meta-Frontier and A Cross-Frontier Appraoch Applications

Size: px
Start display at page:

Download "Estiamting the Profit Efficiency of Contract and Non-Contract Rice Farms in Taiwan. A Meta-Frontier and A Cross-Frontier Appraoch Applications"

Transcription

1 Estamtng the Proft Effcency of ontract and Non-ontract Rce Farms n Tawan - A Meta-Fronter and A ross-fronter Appraoch Applcatons by hng-heng hang Research Fellow, Insttute of Economcs, Academa Snca, and Professor, Dept. of Agrcultural Economcs, Natonal Tawan Unv., Tawan. h-hung hen Professor, Dept. of Appled Economcs, Natonal hung-hsng Unv., Tawan We-hun Tseng Assstant Professor, Dept. of Appled Economcs, Natonal hung-hsng Unv., Tawan Wu-Yueh Hu* Assstant Professor, Dept. of Fnance, Provdence Unversty, Tawan orrespondng author: Wu-Yueh Hu Department of Appled Economcs, Natonal hung Hsng Unversty 250 Kuo Kuang Rd., Tachung 402, Tawan R.O. Tel: (886) ext 257 Fax: (886) Emal: wyhu@nchu.edu.tw Selected Paper prepared for presentaton at the 2015 AAEA & WAEA Jont Annual Meetng, San Francsco, alforna, July 2015

2 Estamtng the Proft Effcency of ontract and Non-ontract Rce Farms n Tawan -- A Meta-Fronter and A ross-fronter Appraoch Applcatons Abstract Trade lberalzaton and globalzaton have placed sgnfcant pressures on farmers and processors ncludng more strngent qualty control and product varetes. Small famly farms n land-locked countres are partcularly vulnerable unless proper nsttutonal arrangement s avalable to assst ther transton from subsstence farmng to market-drven producton. A meta-fronter methodology of Battese et al (2004) s adopted to estmate smultaneously the effcences and the technologcal gaps for productons under dfferent technologes relatve to the potental technology avalable as a whole as well. Another approach that s based on ummns et al (1999) s also appled to estmate the cross-fronter proft effcency. The result ndcates that an average contract farm s 20 percent more effcent than an average non-contract farm n a comparable operatng envronment. Although contract farmng has potental to mprove the proft of smallholders, t s not a suffcent condton for such mprovement. The emprcal resulst also found that the contract technology does not domnate the non-contract rce farmers technology and vce verse. However, the fact that the contract fronter domnates the non-contract fronter has been founded. Keywords: ontract farmng, Rce, Proft fronter, Meta-fronter, Swtchng regresson 1

3 Estamtng the Proft Effcency of ontract and Non-ontract Rce Farms n Tawan A Meta-Fronter and A ross-fronter Appraoch Applcatons I. Introducton ontract farmng has been proposed as an avenue for prvate sector to take over the roles prevously served by the government n the provson of nformaton, nputs or credt for small-scale farmers n the developng countres (World Bank, 2001). Increasng attenton has been gven to whether contract farmng can provde small farmers wth mproved ncome or suffcent protecton from ncurrng losses due to prce fluctuatons. In the lterature, the postve relatonshp between the use of contracts and the ncreasng productvty s found (ochrane, 1993, and Ahearn et al., 2005). A far amount of effort has been drected at assessng the relatve effcency of contract farmng over ndependent producton. Knoeber (1989) showes that the use of producton contracts n the hog and broler ndustry help the dffuson of the new technologes and lead to the mprovement of productvty. Key and McBrde (2003) provde evdence that contracts are assocated wth hgh productvty performance n hog producton n the U.S. Hu (2013) also fnds some evdence that contract farmng mght ncrease the average returns for the corn and soybean farmers n the U.S. Morrson et al. (2004) use the US farm-level data to compare the effcecny effects of contractng n major corn-producng and lvestock-producng states. Ther results show that smaller opertaons are n general less effcent than larger and more contract-ntensve enttes. Ths fndng not only suggests compettve pressures on smaller farms but also pont out the barrers for smaller growers to partcpate n contract 2

4 farmng scheme. For developng countres, Ramaswam et al. (2006) evaluate how the effcency gans of poultry contracts are shared between growers and processors n Inda. ost reducton from better technology and producton practces s counted as effcency gan for the processors, whle average return s vewed as effcency gan for the growers. However, there s lttle drect evdence regardng the potental of contract farmng to ncrease proftablty through ncreases n proft effcency. Setboonsarng et al. (2005) use a stochastc proft fronter model to examne the proft effcency of organc rce contract farmng n the north and notheastern regons of Thaland. They dentfy a sgnfcant proft margn for contract farms across all scale of operatons as compared to the non-contract farms, but effcency gan s not as sgnfcant for the largest farm szes. Ther approach assumes that all farms are operatng under a common proft fronter and cannot dstngush the effect of organc farmng from contract farmng. Thus, the large proft effcency gan may contan both the technology change from adoptng the new organc practces and the true effcency mprovement from the contract scheme. In our study by usng a very specal survey data from Tawan, besdes the technology /productvty effcency, we wll be able to compare the proft effcency between contract and non-contact farms. Meta-fronter methodology of Battese et al (2004) wll be adopted to correct the bas caused by estmatng smultaneously the comparable effcences and the technologcal gaps for productons under dfferent technologes relatve to the potental technology avalable as a whole. A swtchng regresson method wll also be used to account for the self-selecton bases. The results wll be used to test the hypothess whether contract farms are more effcent and/or proftable operatons than the non-contract ones n a comparable operatonal and techncal envronment. Polcy recommendatons on 3

5 whether contract farmng can be an effectve nsttutonal reform to ncrease proftablty for small-scale famly farms wll be drawn. The remander of ths study s organzed as follows. In the next secton, we gve some background nformaton and ntroduce the rce ndustry n Tawan. descrbes the survey desgn and a bref descrpton of the sample data. The thrd secton Secton four dscusses the meta-fronter and cross-fronter effcency usng graphc analyss n llustraton, and secton fve presents the emprcal estmaton strategy. Secton sx shows our emprcal results and the fnal secton concludes. II. Rce Porducton n Tawan Lke many of ts neghborng countres n Asa, rce producton n Tawan s largely based on small famly holdngs wth less than one hectare per farm for more than fve decades. Most rce farmers are ndependent producers who sell ther products ndvdually and have lttle barganng power wth buyers and nput supplers. However, farmers learn to gan producton and scale effcences by organzng custom farmng teams to work for those who do not own machneres (Fujk, 1999). Over the years, Tawanese government s guaranteed procurement at a support prce 20 percent above the average producton costs has ncreased rce producton and created mbalances n the supply and demand n the rce market. The government procurement scheme has also served as a major vehcle to stablze rce prces whch are two to three tmes the world level. Importaton of rce s banned to provde extra protecton for the domestc hgh-cost producers. However, after formally jonng the WTO n January 2002, Tawan began to open up ts market to mports of rce. A total of 144,000 metrc tons of 4

6 rce are mported annually under the quota system, whch s equvalent to about 8 percent of annual consumpton. The steady ncrease n mported rce has brought downward pressures on both the prce and ncome levels for the rce farmers. Moreover, the present rce marketng system has faled to provde a conducve envronment to assst ther transformaton from subsdzed farmng to market-drven producton due to the nsuffcent forward and backward lnkages after a long hstory of government protectons. On the other sde of the market, trade lberalzaton and globalzaton has also modernzed the food retal sector n Tawan, affectng consumers, producers and trade patterns. onsumers have been wllng to pay for more varetes and better qualty rce. Begnnng n 2006, the Agrcultural and Food Agency plans to open up four regonal rce tradng centers as a channel to promote the consumpton growth of hgh-qualty and safe domestc rce by enablng rce buyers, supermarkets, and food manufacturers to easly procure nspected rce whch meet the natonal safety standards. These changes have placed addtonal pressures on farmers and processors to provde more strngent qualty control and more varetes. To overcome such bottleneck, the government has launched a rce marketng contract program n 2005 to assst rce farmers and the agr-busness chan to work together as partners. The mnmum scale for each contract s 50 hectares of adjacent rce paddes wth 50 partcpants ncludng rce farmers, seedlng provders, mllers and marketng agents. Locally-adapted mproved seedlng and low-nput technologes are provded to the rce farmers and mllers under the contract program supervsed by the experts from local extenson servces and experment statons. The partcpatng farmers have to adopt the producton traceablty and book-keepng system and agree not to sell ther products to 5

7 other buyers ncludng the government. Bascally, the nsttutonal desgn s benefcal for the partcpatng rce farmers to acqure new technologes and manageral know-how and for rce mllers to meet consumers demand for qualty and safe products. Eventually, a partnershp between rce growers and rce mllers can be establshed wthout government nvolvement. III. Survey Desgn and Sample haracterstcs In order to evaluate the outcome of the 2005 rce producton-marketng contract program, a survey s conducted n the summer of 2005 after the frst (sprng) crop s harvested. Informaton of prces, value of outputs, major varables and fxed nputs are collected along wth characterstcs of the farms and farmers. The survey covers 80 contract farmers n 7 provnces producng dfferent varetes of rce throughout the major producton dstrcts. For comparson purposes, 246 non-contract farmers are also ntervewed n the same or nearby vllages wthn the same provnce or n adjacent provnces. The dstrbuton of farm sze s qute smlar wth an average of 2.09 and 2.00 hectares for the contact and non-contract farms respectvely. The survey results show that for a contract farm the average revenue s NT$145,000 per hectare, whch s about 11 percent hgher than the average revenue of the non-contract farm. The per hectare cost of producton n a contract farm s about 13 percent lower than that s n a non-contract farm. As a result, the average proft margn under contract s 50 percent hgher than wthout contract.. In the survey, both contract and non-contract farms are ntervewed. The choce of survey stes was based on the offcal record of the contract. Geographcal dspersons 6

8 are also taken nto consderaton because the rce varety s affected by the clmate and local producton envronment. For example, Tagon No. 9 and Tanon No. 71 are wdely adopted varetes n the northern and central regons, whle Tagon No. 2 and Kaoshung No. 139 are more popular n the southern and eastern regons. To enhance the sample s geographc varatons, a stratfed samplng procedure s adopted. Frst, the sample sze n each townshp s determned n proporton to the hectares under contract. Then, the sample farms are randomly drawn from the contract lst provded by the rce mllers who offer these contracts. The non-contract farms from the sample provnce or nearby provnces are selected n proporton to the plantng hectares provded by the extenson specalsts of the local Farmer Assocatons. Ten farms were ntervewed for the questonnare pretest purposes followed by a formal on-ste survey conducted n July Table 1 lsts the sample dstrbutons by provnce and townshp. The numbers of contract and non-contract farms ntervewed are 80 and 246, respectvely. Questons regardng prevous plantng experences, rce varetes, contract prces, producton costs, and demographc factors are addressed n the questonnare. Demographc factors ncluded soco-economc data, household sze and off-farm ncome. Table 2 llustrates the soco-economc characterstcs of the household heads of the sample farms. For the contract farms, most of the household heads are 50 years old, males, wth elementary-level educatons but no off-farm jobs. The non-contract farm household heads have smlar characterstcs except a much hgher off-farm job partcpaton rate. Overall speakng, the household heads of contract farms are younger and more specalzed n rce farmng than non-contract farms. Therefore, sample selecton bas may exst and should be dealt wth n our follow-up effcency comparsons. 7

9 Table 3 compares the average revenues and producton costs of contract and non-contract farms n each regon on per hectare bass. Frst, the average revenues of contract farms are hgher than those of non-contract farms n most regons. The rce yelds per hectare, however, are about the same. Therefore, the major reason of hgher revenues s due to hgher rce qualty. A floodng event n 2005 damaged the frst crops n many agrcultural provnces n the southern regon. hgher prces except those n the southern regon. So, most of the contract farms receve However, the contract farms stll outperform the non-contract farms by hgher yelds wth hgher revenues. Next, the contract farms spend more on ther seeds due to varety dfferences. However, due to strct restrcton on fertlzer and chemcal usages, these contract farms spend much less on the chemcal expendtures. Therefore, on average the total expendture of contract farms s 20 percent lower than that of non-contract farms. Thrd, both the gross and net revenues of contract farms are hgher than those of the non-contract farms for all regons. The proft margns on the gross bass range wdely from NT$6,000 n the northern regon up to NT$60,000 n the east. On the net bass, the proft margn of enterng the contract arrangement s about NT$40,000 n the central regon and NT$60,000 for those located the southern and eastern regons. However, farms n the northern regon have the smallest proft margn or no beneft at all from partcpatng contract farmng. IV. Meta-Fronter Effcency and ross-fronter Effcency Suppose the nput-output bundle for contract and non-contract rce farmers are denoted as y, x ) and y, x ). The soquant curves for contract and non-contract ( ( N N 8

10 farmers wll be (y) and N(y) as shown n Fgure 1. Therefore, the effcency measurement through the dstance functon value for non-contract and contract rce farmers are ob oh defned as follows D ( YN, X N ) 1, D ( Y, X ) 1 oa og N. Suppose there exsts a meta-fronter for these two groups and the meta-soquant wll be the enveloped curve from the soquant by contract and non-contract rce farmers. The technology gap rato for a non-contract rce farmers when the nput-output bundle s at b ad wll be. Smlarly, the technology gap rato for a contract rce farmers when the n- bd put-output bundle s at h wll be gf. Ths technology gap rato represents that the per- hf centage of rce farmers on the potental output gven the technology avalable to the ndustry as a whole. In other words, t represents the average gap between the groups wth respect to the meta-fronter. So the meta-fronter effcency wll be one mnus the value of the dstance, then tmes the technology gap rato. If the non-contract soquant s on the left-hand sde of contract rce farmers soquant for all nput combnatons, the dstance of the non-contract rce farmer from the ob contract fronter s defned as D ( yn, xn ) 1, and the dstance of the contract rce oc oh farmer from the non-contract fronter s DN ( y, x ) 1,.e., each rce farmer s us- oe ng an nput vector that domnates the other group s technology.. If the cross-fronter effcency s greater than 1, t means that a frm s nput-output bundle s nfeasble usng the other group s technology. In here, the cross-fronter eff- 9

11 cency for contract and non-contract rce farmer are defned as the followng. F ( y F N ( y, x N D ) 1 D, x N N ( y, x ) TE 1 ( y, x ) TE N ( y, x ), ( y, x ) D ( yn, xn ) TEN( yn, xn ) ) 1 1. D ( y, x ) TE ( y, x ) N N N ( If F y, x ) >0, t means contract technology domnates non-contract technology. The comparson of proft effcency s defned smlarly. N N V. Emprcal Model Ths secton descrbes the emprcal procedure we use to compare the proft effcency between the contract farmers and non-contract farmers. The proft neffcency s defned as loss of proft from not operatng on the proft fronter (Al and Flnn, 1989). It can be measured by the followng Rahman (2003) s approach n whch both techncal and allocatve neffcency are smultaneously taken nto account. Techncal neffcency s defned as the loss of profts from falng to meet the producton effcent fronter. Allocatve neffcency s the loss of profts from falng to observe or respond to the relatve prces of nputs and outputs. The stochastc proft fronter s defned as, (1) f ( p, w, Z) e where π s the vector of proft defned as gross revenue mnus varable cost; p and w are the vectors of output and nput prces, respectvely; and Z s the vector of fxed nputs. The error term (e) s assumed to be conssted of a two-sded random error (v) and a 10

12 non-negatve proft neffcency varable (u), and can be expressed as follows: e vu, (2) 2 where v s assumed to be ndependent of u, and dentcally dstrbuted,.e., v ~ N(0, ) ; u s the non-negatve random varable of neffcency whch s assumed to be ndependently dstrbuted at a truncated normal dstrbuton wth a postve mean and varance 2 u. Under ths specfcaton, proft neffcency represents the nablty of a farmer to acheve the hghest possble proft under the gven output and nput prces along wth the levels of v fxed nputs. Ths s a better measurement snce t combnes nformaton of the cost neffcency and the ncome neffcency. A rce farmer s decson on whether to sgn up a producton-marketng contract s a self-selecton problem and such a problem can be descrbed by a swtchng regresson model and a crteron functon (Lee, 1978; Huang et al., 2002). Suppose the th rce farmer has two choces, jonng or not jonng a contract, and ths decson s determned by the followng proft functons, and N, for contract and non-contract farmers respectvely: ) f ( p, w, Z e, (3a) ) N N f N ( p, w, Z e, (3b) where subscrpt and N denotes the contract farm and non-contract farm, respectvely. The farmers decson on jonng the contract depends on the proft dfferental among contract, non-contract usng and other non-proft consderatons, and can be descrbed by a crteron functon as follows: 11

13 I * X ( ) (4) N where X s a vector of non-proft varables whle the random varable represents the unobservable factors that affect the selecton of sgnng a contract; α and γ are parameters needed to be estmated. The crteron functon n equaton (4) assumes that a rce farmer may jon the contract f the proft from jonng the contract s hgher than the proft wthout the contract. Snce the farmer can only choose ether to sgn a contract or to stay ndependent, only one of the two profts ( or N ) can be observed. Therefore, the de- 註解 [WH1]: Not sure about ths. Not only the dfference of the profts wll affect the decson but also the non-proft varables X, rght? pendent varable I * 0 f the proft of jonng contract s observed, or I * 0 when the proft of non-jonng contract s found. Equatons (3) to (4) cannot be estmated drectly because the decson to sgn contract may be determned by unobserved varables (e.g., farmers characterstcs, management ablty) that may also affect performance. Therefore, the error terms n (3)~(4) wll be correlated. A standard two-stage procedure of Lee (1978) and Wlls and Rosen (1979) s adopted to allow unbased estmaton. Frst, two nverse Mlls ratos M and M N are derved from equaton (4) usng the probt model. Then, the estmated nverse Mll s ratos, ^ M and M^ N, are ncorporated nto equatons (3a) and (3b) to correct the sample selecton bas as follows: ^ f ( p, w, Z, M e for 1 ) I, (5a) ^ N N f N ( p, w, Z, M N e for 0 ) I, (5b) The dea of ths procedure s to fnd the expresson for the means of E( e 1) I and 12

14 E( e N 0), and to adjust the error terms such that they wll have zero means. In the I second stage, the coeffcents n equatons (5a) and (5b) can be estmated by a meta-fronter method based on Battese et al.(2004) to obtan proft effcency for contract and non-contract farmers under alternatve technologes as well as ther technology gap. Suppose a meta-proft-fronter functon for the th rce farmer s expressed by ^ * * * * X f ( p, w, Z, M; ) f ( X, ) e, (6) where X s the vector of explanatory varables ncludng p, w, Z, and ^ M, and * denotes the vector of parameters. Snce there are two groups of farmers (.e. contract and non-contract), the proft fronter for each group s defned by v( j ) u( j ) f X, ) e e, 1,2,..., N j, j 1, 2 (7) ( ( j) X ( j ) v( j ) u( j ) where N j s the number of samples and subscrpt j denotes the group. The alternatve stochastc proft functon n equaton (7) can be expressed n terms of the meta-fronter functon of equaton (6) by X ( j ) * U e ( j ) X V ( j ) e * * e. (8) * X e The frst term on the rght-hand sde of equaton (8) s the proft effcency relatve to the stochastc proft fronter for the j th group and can be expressed as PE e X ( j ) V ( j ) e U ( j ) (9) The technology gap rato (TGR) can be found n the second term on the rght-hand sde of equaton (8) whch s wrtten as 13

15 X ( j ) e TGR (10) * X e Fnally, the proft effcency (PE) relatve to the meta-fronter for the th frm s defned as follows: PE * e * X V ( j ) PE * TGR (11) The estmaton procedure for (9)~(11) s llustrated below: Step 1. A stochastc proft fronter model of equaton (1) s estmated assumng a truncated normal dstrbuton n whch the non-negatve random terms represent the proft neffcency as follows: 1 ( A e / A) 2 E(exp( u ) e ) exp( e A / 2), (12) 1 ( e / ) A where s v s A s , /, (1 ), e ln( ) x, and s a dstrbuton functon of a standard normal random varable. The varances 2 and 2 v can be estmated, respectvely, by the method of moments (Olson et al., 1980) as follows: 2 ˆ [ m3 ] 2 / (4 / 1) 2 / 3, and ˆ ˆ v m2 [1 ] u, where m 2 and m 3 are the second and thrd moments of the resduals. Step 2. The meta-fronter functon s a fronter that envelops all fronters of ndvdual groups such that * x x, j. The optmal ( j) * can be obtaned by mnmzng the 14

16 sum of absolute devaton (MSAD) 1 of the values on the meta-fronter from the group-specfc fronters at the observed nput levels. Battese et al. (2004) pont out that the followng lnear programmng formulaton s equvalent to the MSAD. mn s. t. * x * ln( ( x, )) ln( ( x, )) ^ ( j) (13) where x s the vector of the means of the all observatons n x. The PE * and TGR can be calculated based on the estmates of. * and the group-specfc fronter estmates ( j) VI. Estmaton Results The data n ths study were taken from the survey descrbed n Secton 2. After deletng the samples wth mssng observatons, the emprcal estmaton s based on 201 non-contract farms and 80 contract farms. Table 4 summarzes the sample statstcs for both contract and non-contract farms. The average plantng acreage of the contract farm s 2.21 hectare, whch s slghtly hgher than the average of non-contract farms 1.73 hectare. The average age of contract farms household heads s only 43 years old. Ths s much younger than the 60 years old of the non-contract household heads. 70% of the man contract farm operators are full tme farmers, whch s hgher than the 52 % of the non-contract farms. The percentage of farmers recevng hgh school or above educaton for the contract farms s also hgher than t s for the non-contract farmers. Ths suggests that contact farmers tend to have more years of educaton than the non-contract farmers. As for producton costs, Table 4 shows that chemcal and machnery costs for the 1 * The optmal can also be derved from mnmzng the sum of the squared devaton of the values on the meta-fronter from the group-specfc fronters. Ths approach wll lead to a quadratc programmng problem. 15

17 rce contract farmers are less than those for the non-contract farmers, whle there s no sgnfcant dfference n seed cost. The average proft of contract farms s 27 percent hgher than the non-contract average. The emprcal model for estmatng the decson to jon the contract versus to product ndependently s specfed as follows: I 0 ( REG), (14) 1( AGE) 2( EDU ) 3( FULL) 4( REVENUE ) 5 where I 1 f a latent proft margn from jonng the contract s postve, whle t s 0 otherwse. AGE s the man operator's age. EDU s the man operator's educaton level. FULL s a dummy varable ndcatng whether the man operator s a full-tme or part-tme farmer. REVENUE s the farm's revenue. REG s a regonal dummy varable ndcatng whether the farm s located n the east or not. The estmaton results for equaton (14) are shown n Table 5. Table 5 shows that havng a prmary occupaton on-farm rases the lkelhood of contractng. Ths s consstent wth our expectatons. Increases n years of age and schoolng lower the probablty that the farmer wll jon the contract. Older rce farm owners would less lkely to jon the contract probably because of the habt formaton effect. They are reluctant to change unless necessary. More educated farmers may also have a hgher ncome and thus a hgher reservaton wage to be nduced nto contract producton. Table 5 also suggests that farmers n the eastern regon are more lkely to accept a contract because they could earn more than farmers n other regons. The emprcal model for estmatng the mpact of contract producton on farm s proft performance takng nto account the self-selecton processes are denoted as follows: 16

18 ln( ) ln( SEED ) ln( HEM ) ln( LABOR) ln( ARE ) 0 ln( SEED * HEM ) ln( SEED * LABOR) 5 ln( HEM * LABOR) ln( HEM * ARE ) 7 ln( LABOR * ARE ) ln( FULL) ln( REG) W ln( AGE) ln( EDU ) W *ln( ARE ), (15) where subscrpt = c for contract farms and =n for con-contract farms. The estmaton results for equatons (15) are shown n Table 6 wth the selectvty bas adjusted. Frst, the farm sze (ARE) has a postve and sgnfcant mpact on the proft of contract farms, but all other nputs lke seed (SEED), chemcal (HEM) and labor (LABOR )have no sgnfcant mpact, except when they are nteracted wth the farm sze. Ths result suggests that the proft of contract farmng s hghly correlated wth the acreage devoted to the contract. Ths s expected because the contract usually requres farmers to comply wth certan nput allocaton restrctons. Thus, farmers lose control over ther manageral decsons, and the lnkage between proft and nput usage no longer exsts. The other mplcaton s that larger farms beneft more than the smaller ones once they jon the contract. The results for non-contract farms are somewhat dfferent n that nputs other than acreage are also among the major determnants of the proft. Thus, the autonomy of nput allocatons s preserved by the non-contract farmers. As for the non-nput determnants, both the geographc locaton and employment status play a sgnfcant role n the proft. Farms located n the eastern regon tend to have hgher profts than those located n other regons. However, t s only statstcally sgnfcant for those under the contract arrangement. Part-tme contract farmers earn more profts than ther full-tme peers. Thus, although full-tme farmers have larger probablty to partcpate contract farmng, they may not be better off than those part-tme contract farm- 17

19 ers. Ths result reflects an nterestng trade-off between proft gans and loss of autonomy by the contractual arrangement. In comparson, the case for non-contract farms s opposte. Full-tme non-contract farmers enjoy hgher profts than ther part-tme peers because there s no loss of manageral control when farmers produce ndependently. Because the scale of producton has a strong postve correlaton wth the lkelhood of contractng, we also add an nteracton term of sample selecton and scale (.e., W *ARE) along wth the two nverse Mll s ratos. The coeffcents of the selectvty bas adjustment (W ) are both sgnfcant but have the opposte sgns wth and 1n Ths result mples that those who choose to jon the contract are worse than c the average contract farmers n terms of proft earnngs. Those who choose to produce ndependently are better than the average ndependent farmers. The postve and sgnfcant for the nteracton term (W *ARE) confrms that larger farm sze s asso- c cated wth an ncreasng proft for contract farms. 1. Meta-Fronter Proft Effcency Estmaton The basc summary statstcs estmates of proft effcency, technology gap rato, and proft effcency relatve to the meta-fronter usng equatons (12) and (13) are presented n Table 7. The mean value of technology gap ratos for non-contract and contract rce farmers are and respectvely, whch ndcates that there s an average of 98% of the potental proft gven the technology avalable to ths rce ndustry as a whole. Please also note that all contact and non-contract rce farmers fronter are tangent to the metafronter snce the maxmum value for the technology gap rato are one. The proft effcency by mean for non-contract rce farmers s whch s small- 18

20 er than the contract rce farmers' However, contact rce farmers tended to be further from potental proft defned by the metafronter functon snce the technology gap raton for contact rce farmers s smaller than that for non-contact rce farmers. After calculatng the TGR as shown n the mddle row of Table 7, the proft effcency for each group relatve to ths meta-fronter could be estmated and are shown n Table 7. The proft effcency relatve to the meta-fronter of non-contract rce farmers s whle t s for the contract rce farmers. We could not determne whch group's technology domnates through the meta-fronter estmaton. The cross-fronter effcency wll be appled to check t. 2. ross Proft Fronter Effcency Estmaton To estmate the cross proft fronter effcency, whether the proft fronter s dentcal for rce contract and non-contract farmers needs to be tested. The ANOVA statstc s appled here and the test result shows that the null hypothess of dentcal proft fronter s rejected. Later, the estmaton approach for cross proft effcency s based on ummns et al(1999) and the estmaton results are shown n Table 8.. Both the proft effcences of the non-contract farmers relatve to the contract fronter and the proft effcences of contract farmers relatve to the non-contract fronter,.e. the cross-fronter effcences are all calculated and shown n Table 8. The non-contract proft effcency wth respect to contract fronter (.e. PE Y, X ) ) are all smaller than ( nc nc one (.e. the values of mnmum and maxmum are smaller than one) whch ndcates that the non-contract technology does not domnate the contract rce farmers technology. Smlarly, the contract proft effcency wth respect to non-contract fronter (.e. PE Y, X ) ) N( c c 19

21 are all smaller than one whch ndcates that the contract technology does not domnate the non-contract rce farmers technology. However, as we computed the cross-proft-fronter effcency for contract rce farmers (.e., F ( y, x PE ) 1 PE N ( y, x ) 1 ( y, x ) ) and t shows a negatve sgn whch means contract fronter domnates the non-contract fronter. Smlarly, the cross-proft-fronter effcency for non-contract rce farmers s postve whch means t s domnated by contract fronter. Such estmaton results mply that Tawan government may promote rce contract to farmers n order to ncrease ther proftablty when facng foregn product competton. VII. oncludng Remark ontract farmng has become an attractve polcy nstrument for many developng countres to assst small famly farms to gan access to markets, nformaton, credts, and necessary servces to manage ther rsk. On the other hand, contract farmng may have subtle mpacts on both farmers ncome and manageral control. Therefore, the success or effectveness of ths polcy nstrument depends on whether these contracts are attractve enough for the farmers by ncrease ther profts whle loss of autonomy can be mnmzed. In ths paper, we conduct an on-farm survey on more than 300 rce farmers n Tawan. The per hectare cost of producton for a contract farm s about 13 percent lower than a non-contract farm. As a result, the average proft margn under contract s more than 50 percent above those wthout contract. A swtchng regresson model s adopted to analyze farmers decson on contract partcpaton and to compare ther proft performance. The estmaton result ndcates that an average contract farms s 20 percent more effcent than an average non-contract farm 20

22 under a comparable operatng envronment. These results mply that the contract arrangement can ndeed be an effectve nsttutonal mechansm to ncrease proftablty and compettveness for small famly farms. We also fnd that the contract decson s determned not only by a proft comparson between contract and ndependent producers but also by other demographc determnants lke age, educaton level, employment status and geographcal locatons. Fnally, we fnd that contract rce farmers have more proft effcency than non-contract rce farmers gven a proft meta-fronter functon. The estmaton results mply that Tawan government may promote contracts to farmers, maybe not only n the rce ndustry but also all agrcultural products, n order to ncrease farmers' proftablty when facng foregn product competton. 21

23 References Ahearn, M, J. Yee and P. Korb (2005): Effects on Dfferng Farm Polces on Farm Structure and Dynamcs, Amercan Journal of Agrcultural Economcs, 87, 5, Al, M. and J.. Flnn (1989), Proft Effcency among Basmat Rce Producers n Pakstan Punjab. Amercan Journal of Agrcultural Economcs, 71: Battese, G..E. and T.J. oell (1988), Predcton of Frm-Level Techncal Effcences wth a Generalzed Fronter Producton Functon and Panel Data. Journal of Econometrcs, 38: Battese, G.E., D.S. Prasada Rao, and.j. O Donnell (2004), A Metafronter Producton Functon for Estmaton of Techncal Effcences and Technology Gaps for Frms Operatng Under Dfferent Technologes. Journal of Productvty Analyss, 21: ochrane, W. W. (1993). The development of Amercan agrculture: A hstorcal analyss. U of Mnnesota Press. ummns, D.J., M.A. Wess, and H. Z (1999), Organzatonal Form and Effcency: The oexstence of Stock and Mutual Property-Lablty Insures. Management Scence, 45: Delgado,.L.,.A., Narrod, and M.M., Tongco (2001), Polcy, Techncal, and Envronmental Determnants and Implcatons of the Scalng-Up of Lvestock Producton n Four Fast-Growng Developng ountres: A Synthess. Fnal Research Report of Phase II, FAO: Rome. 22

24 Fujk, H. (1999), The Structure of Rce Producton n Japan and Tawan. Economc Development and ultural hange, 47: Hu, W.-Y (2013), Effect of contract farmng on the U.S. crop farmers' average return. Agrcultural Economcs- zech, 59: Huang, M.-Y.,. Huang, and T-T. Fu (2002), ultvaton Arrangements and ost Effcency of Rce Farmng n Tawan. Journal of Productvty Analyss, 18: Key, N., and W. McBrde (2003), Producton ontracts and Productvty n the U.S. Hog Sector. Amercan Journal of Agrcultural Economcs 85: Lee, L.-F. (1978), Unonsm and Wage Rates: A Smultaneous Equaton Model wth Qualtatve and Lmted Dependent Varables. Internatonal Economc Revew, 19: Morrson,.P., R. Nehrng, and D. Banker (2004), Productvty, Economes, and Effcency n U.S. Agrculture: A Look at ontracts. Amercan Journal of Agrcultural Economcs 86: Knoeber,. (1989): A Real Game of hcken: ontracts, Tournaments, and the Producton of Brolers, Journal of Law, Economcs and Organzaton, 5, 2, Olson, J.A., P. Schmdt, and D.M. Waldman (1980), A Monte arlo Study of Estmators of Stochastc Fronter Producton Functons. Journal of Econometrcs, 13: Rahman, S. (2003), Proft Effcency among Bangladesh Rce Farmers. Food Polcy, 28: Ramaswam, B., P.S. Brthal, and P.K., Josh (2006), Effcency and Dstrbuton n ontract Farmng: The ase of Indan Poultry Growers. Workng paper, Internatonal Food Polcy Research Insttute, Washngton D. 23

25 Setboonsarng, S., P.S., Leung, and J. a (2006), ontract Farmng and Poverty Reducton: The ase of Organc ontract Farmng n Thaland. Asan Development Bank Insttute Dscusson Paper No. 49, Wlls, R. J. and S. Rosen (1979), Educaton and Self-Selecton. Journal of Poltcal Economy, 87: S07-S36. World Bank (2001), World Development Report 2002: Buldng Insttutons for Markets. Washngton D: The World Bank. 24

26 Fgure 1. Meta-Fronter Effcency and ross-fronter Effcency for ontract and Non-contract Rce Farmers 25

27 Table 1. Sample Sze and Geographcal Dstrbuton Regon Provnce Townshp ontract farm Non-contract farms North Taoyuan Shnwu 13(5.3%) Maol Yuanl 6(7.5%) 9(3.7%) Sub-total 6(7.5%) 22(8.9%) entral Tachung Tacha 19(7.7%) Wufeng 5(6.3%) 12(4.9%) hunghwa Fushng 16(6.5%) Ptou 12(15%) 15(6.1%) Erln 9(11.3%) Herme 15(6.1%) Sub-total 26(32.5%) 77(31.3%) South Yunln Tsutung 3((3.8%) 14(5.7%) hay Tabao 13(5.3%) Shnkung 13(5.3%) Mnshung 12(4.9%) Tanan Hobn 15(6.1%) Shenghua 16(6.5%) Shayn 6(7.5%) Bahe 5(6.3%) Kaoshung Dalao 12(4.9%) Pntung Wondan 13(5.3%) Sub-total 14(17.5%) 108(43.9%) East Ylang Jaosh 13(5.3%) Hualan Ful 6(7.5%) 16(6.5%) Tatung Kuanshan 11(13.8%) 7(2.8%) Tsushung 17(21.3%) 3(1.2%) Sub-total 34(42.5%) 39(15.9%) TOTAL 80(100%) 246(100%) 26

28 Table 2. Soco-Economc haracterstcs of the Sample Number of Sample Percentage n Total (%) ontract Non-contract ontract Non-contract Gender Male Female Age Below ~ ~ ~ and above Educaton None Elementary Mddle school Hgh school Vocatonal college ollege and above No. n farmng and more Type Full-tme Part-tme

29 Table 3. Revenues and Producton osts of Sample Farms by Regon North entral South East Total ontract Non-co ntract ontract Non-co ntract ontract Non-co ntract ontract Non-co ntract ontract Non-co ntract Revenue 116, , , , , , ,153 85, , ,183 Producton 7,789 7,625 7,393 7,729 9,550 7,706 6,379 5,887 7,074 7,143 Prce Total ost 97,791 91, , ,452 84, ,934 90,089 96,077 94, ,460 Drect 59,608 59,006 66,884 71,864 62, ,599 54,598 50,288 60,692 72,909 Seed 8,516 7,909 8,364 7,941 8,652 8,213 7,224 5,578 7,932 7,192 Pestcde 871 6,010 6,787 9,796 9,413 12,220 6,247 4,666 6,794 8,505 Fertlzer 12,514 13,902 11,971 9,683 9,627 11,678 9,287 8,281 10,347 9,863 Materal 0 0 2,060 3, , ,451 1,539 6,372 ustom 31,343 29,483 29,777 29,443 25,634 37,539 25,805 16,833 27,323 26,325 Hre labor 3, ,286 8,446 3,727 14,943 3,385 11,479 3,562 10,340 Energy 2,398 1,701 2,639 3,095 5,907 10,587 2,651 2,812 3,196 4,312 Indrect 38,183 32,146 40,453 37,588 21,560 35,334 35,491 45,789 33,370 39,551 Self-wage 16,696 16,696 21,376 20,987 11,800 14,734 21,484 18,848 18,892 19,298 Land rent 21,487 15,450 19,077 16,601 9,760 20,600 14,007 26,941 14,478 20,253 Gross Proft 56,884 50,726 74,813 42,785 59,036 8,047 93,556 35,447 84,596 32,274 Net Proft 18,701 18,580 34,360 5,196 37,476-27,287 58,065-10,342 51,227-7,277 Unt: Kg/Hectare; NT$/Hectare 28

30 Table 4. Sample Statstcs: Means and Standard Devatons Varables Non-ontract Farms ontract Farms Acreage (hectare) 1.73 (1.51) 2.21 (1.82) Age (9.95) (4.66) Educaton (%) 19.90% 23.75% Full Tme (%) Seed ost ($NT) 52.06% % hemcal ost ($NT) (1290.4) (96.69) (1802.0) (386.50) Fertlzer ost ($NT) (1068.8) (473.87) Labor ost ($NT) Proft ($NT 1000) (3006.5) (1208.9) (209.33) (248.01) Note: The numbers n the parenthess represent the standard devatons. Educaton s the percentage of sample wth educaton hgher than hgh school level. Full tme s the percentage of full-tme farmers n the sample. 29

31 Table 5. Probt Estmaton of the Sample Selecton Model Varables oeffcents and Standard Errors onstant ** (2.106) Age ** (0.039) Educaton ** (0.218) Full Tme 0.696* (0.440) Revenue * (0.11E-05) Regon Dummy 1.586** (0.686) McFadden R-Square LR Statstc Note: The regon dummy s defned as 1 f the farm s located n the east whle t s zero otherwse. The numbers n the parenthess are standard errors. ** sgnfcant at 5% sgnfcance level. * sgnfcant at 10% sgnfcance level. 30

32 Table 6. Estmaton Proft Functons for ontract and Non-ontract Rce Farmers Varables ontract Non-ontract onstant (40.07) * (61.14) SEED (5.44) 4.95 (7.51) HEM (3.19) 12.68** (4.53) LABOR (3.74) 14.34** (7.30) ARE 4.57** (1.82) (2.56) SEED*HEM 0.19 (0.39) (0.51) SEED*LABOR 0.56 (0.52) (0.89) HEM*LABOR (0.12) -1.43** (0.35) SEED*ARE -0.67** (0.28) 0.84** (0.33) HEM*ARE 0.17** (0.08) -0.20* (0.12) LABOR*ARE (0.05) (0.13) AGE (0.07) (0.08) EDUATION (0.02) (0.04) FULL -0.11* (0.058) 0.19** (0.08) REGION 0.16** (0.08) 0.18 (0.19) W (Inverse Mlls rato) -0.84** (0.40) 0.48* (0.26) W*ARE 0.28** (0.12) (0.07) Adjusted R-Square Note: The proft functon s a Translog functon whch means we take logarthm on both the dependent and ndependent varables except for AGE, EDUATION, FULL, and REGION varables. 31

33 Table 7. Proft Effcency for ontract and Non-contract Farms Mean Mnmum Maxmum St.Dev Non-ontract Rce Farmers Proft Effcency Technology Gap Rato Metafronter Proft Effcency ontract Rce Farmers Proft Effcency Technology Gap Rato Metafronter Proft Effcency

34 Table 8. ross Proft Effcency for ontract and Non-contract Farms Mean Mnmum Maxmum St.Dev Non-ontract Rce Farmers ontract Rce Farmers Non-ontract Rce Farmers ontract Rce Farmers Proft Effcency PE N ( Ync, X nc) Proft Effcency PE ( Yc, X c ) ross Proft Effcency PE ( Ync, X nc) ross Proft Effcency PE N ( Yc, X c )