.4. . . . . . . . . . _ . Linton-K J . ., . . . ...WOH. n . 2333:... .6. .vl. . . ofi furl Ilhrvuiiadis . I". .h an“: .1 . .. 3.111%... . . . ..V C]: X... : . .... . . . . I...“ - ..v r .. up . . ow {HINDI-7P ‘1 lll'po . . . . . v 31 0': . . . , ‘ tloolo ‘- :.. . 06.. A . . la -. AJu"... I "l . . . . .nkbqwmnu , 1.9%..- . I.ur“vU-rwbrl.rdcnvLP }{¥.¢H Sud“ .. : . . 1C.Du..r,n.rv§:t vfifgfihnflfinumw #53 071‘ .... 01"“. i ... I. ‘ .\ ...... .3- Likeli £114 at ..(VL) -..... $109.46.}; ...!) . p‘hxl r-bvl D . I’M-1% z I $5.9»... ...... :nunxw L. .5. ., purgiflut 1.0{1 ..yltJ,.r\...p!.fl .. r-rlv till! I..r). llvb . 4.1!!) 3" #5., H.833! ......Lhaphcrztsrfléé. 9’ ’Amgbriuu'. I ‘.‘ o IMfiu§g‘ .15! y I“ us 05.?" ... n ”49.. HY... .. .... «up: . .2”.an In .r. .1...- . . ..nfla...§§¥f.£ _.t. ‘ a o u u ’5'. .. { . ....mE ... r‘ . urkl .LI(, ..>1:.- cl. 0...... y.-. y, . .33.. 1.0 l' l I. ..I.. ...! .1 I 0 \ 1w; m! 2’“! - . I'M'W» i .4 I": “In. .LIL 1.41;.» It! . . .7...| Fyll ..., .oi‘ldpn 0 . . 7 . :rv . MHZ? . \ . .. ... y... . v . . . ‘ ..t...f . . . . .. . . . . . ..-.3... .. .. . ... . . 1 I . . . ., . . Dlhl ... . . . . . . . . .. .. .. .. . {I‘ . I. u . l . ... . .I- A n at. .. q. .r . I.‘l. . T . . . ... Ill... . i. ..-. .. . . I .. .. 53.9..) . . ......vc.i.cl.... .4 .. . .. v.02 . . . . il.r. I}... ....lll. ti...) :0 .I.: . . I v . I. .. ..\rh‘hNfiv (Suva; .rt . ... : .Arllt‘ v. :lbro... r..- A: . . 1...}... ., tiri...‘ 14. .tll...l;. . ... . .. . ...—Gm... .Y: §4l!l.vb.|¥. (0111‘. ..l! O I v . I A 1|. v §\.si . if. ..»$Lu.v'l|l$vp .iuéslpz r< it). are. 7:... av I.Q46V.\v.l.$\o ... ...-w. : ..(.15.|2.IA.- >..I’I.:.~ 6. £1.511) I ....» I195 ZII...\t3I»IIu.. . Co. I 5......9‘: .rf...) >hp|Huprl£f . 4"...) Dunc. I ... z ’3.) ..\‘t .. 4. . \ 7V3; }¢.J .\ ... . _ .12.. .y a) -acesDttr...\ar\}4.v . . ... to 3 \I {Ivvlié .. .. .é . $p¢.! I ...»;thh)! If! 3-.. |\ *1 a I‘. .. . ...... u(.fvltv..ts.§.ldz '3. .k '. I ... I Hi. .2-.. . ..\\v .... ..r4( 3“) ;. .1.’ .. vX .. .. 15!..pl: 11.1.- \ofi .. I “‘ -.AVDQ [\IQQV-‘O‘ - .1 .anu aox...!:...1lqy ‘ -..)...Y Iv..-‘ It 1. E V tr lllll 4 . .1») I)! .‘.|.rlul y- .41: I- \3: . . 1.--..- .lfi».--.olo_)9 ,.. . .r. I. . 0.. ~ fivtli lg. . . . .. ll ‘kL-A~I. I-c, Il“ u l...~ .. 70 I: .v -- LII ¢ ,1: 15 v ‘. ’uu... . .... .. . u ,. . . .. u . . x4: . y)!\ vo.a.l!‘l~ .. .. ., . . .... . . f . .. . . .. 32:1 ’7. 45.“.-. ‘1. . . . . ,. . _ : .. . . . .1 cu)?’.:\.( -5111 y. .. . . . . : .y. . -v. :- A\“.ll| .-3'“ . , I . . v on . in: < u . . it, I’D". 1.0 5;! ... .... \o 0!, ... . , ...: 1|. 1: .1! o l..‘ .v ‘V A (I .vvvtr.§vl.q‘ a-u’vbl . . I. ....o .w .... . Hurst» . tutti). . v... v . 3 U 3 ..- .l. . I110. . 3.3:... (It; . 90.). Zliit.‘ cit“, I . .yvvtr #1.!1. 3% ‘I u)¢0, ‘s‘ PIC.'2atp,'r$.:l»J}.|nika\cr'f¢z hr Iv!!'.‘l\l. ll:..!b¢r.(.7. II :1.» ..\. ... ....v...;$10:...,$.l{lnlvtt\l..€ I; Illa), VI 2... - . . ....v | . .. .1 3‘9}: .. I: bauhxiiihi 'At‘o‘llk 1‘ ‘.I . ' .“ ’i ..I... . \ ‘O-i cl \AK 0 l .r. A.. III .111 ~L '1 ... r... ai.r UH) 9. ‘...fl.rx... . . ‘8! > I ’I' .. C. 14 I'll - I, run. I ah? . . . ...u! VHF; a u... #1.... ...4).1. (If 6 f. , u l . . . | . . . . Ft) , . . u I . r . K. . 9 iii?! ‘1 3M0 I > ~ I . n.‘ . . W! {rig-I3!- ‘t‘ilr? lilting. Ir c v ‘4 . ’ y .l .II‘AI: .[tit‘\.l\ b\.‘(.,.l‘ b . a . .eflfift '5‘“ ll‘*;£-‘I(l .ltth .-\.. b) .47.].I3 . . Es\‘... . . . . . '1‘ ....ttE-4." ...-‘11.? .1 pl .9 . . IA.‘ . ...K. . . . . . . . . . . . iflr.l.t..5§n.}:vvaf ‘33:}... A) . I, a.\ .¥...$ :- ii \1. . . . , . , . . .. . . r fittiit’Iln... 1.214% fxuuno.‘ .4 5., 3 .y A. i 17 . . .. I. .. . ...Vo"‘~(l...llut() wig-......{QIIIOr ,a .“nvll . - \L. . (.... C...v| £0... . . . . .. ...,.»y.7£-?nib|lt¢.v5u 1!:.A1..?Vu.ra..\ll.!{ In ,. . . .3 r . . V .. . n)....al:r¢-abrlvu ....PilbllvioH 53‘. 1b.! ‘6.\Mn[3 {7“‘3‘ Hu., ‘ | 1 \4 n . . ‘ ....I..\: .. ‘. . uV .., . v A . . V- ..\ l .1‘0 iOll.-yIAl - 3 . 3. I ........ x .-.l,.. b' I.‘ I‘l‘r~ . 1 .. b... ...I..I ‘ I l I V‘l“.v‘\u . ~ ... all .1 .u / . n {l t .- , I l . .... All .1 I . . .'ul .. 0. ‘ vI'I.v»\., .‘ , I |¢n .11.. L . . u .I x... c . I}. i. . u ... , U cl -...)‘llkfi‘l V \‘ Irv ¢ '0 ID. x 0 .- “\“‘I‘l\“l .( r . I 2... I 1133...... I.'l I ‘. ' ." m, u MI. . .. A ~ . v. I .) .1 ..l v - . l ... .u . "2‘ v, .« In: .lu . . . .. ... I Cb .. v. A . wt ll l v . . v I . y I'3Pu1 ... .3. o. . v I 0|}. I \ 1 In I. «Es€l ‘Il. .\..‘.‘Iv.l .... ....~|.,..~u. o v» 0 o-!|¢»v.£:(... . . ..3» 6tx.... . y o. . . I . . ‘( II‘ I . . . .. . "1.. . , Iv .I‘ . .Q‘t‘}... . ... )).- -1.-- S:\..I'i~ta)\6.. .wl ‘ .1 tv «I..I}'.Alx'lntux.. |.. 6‘ I ‘.‘.‘[i ‘1‘; .r ’r I . I 0 I -I 1‘.‘ tl‘ i 1 0| I I -u ‘11. [1-3.3: .. I t (l ..\vi‘vl’..\lv .r 1 l v .1 ... ,-tl:.0\.. .\5$t\|.. V r , -1’l. 1:-..) ‘3. .. . ,I v . . u n 4... -\t . .1 ..( .v...v!01v. ... .. . I.61I ‘Fx. :. ..«IVI. | .. ‘ql. . -. , ...I 1.1 .Q“: . . . .121: u .. . t ‘ ...»l -Iitv't. . . . . {1‘31}... . . . . . . . . . .. .. v . . ...: L... n . ~. .‘ .I . y A Yv ll bl) . . . ...c‘C... . u it}. . . . G3,. . 4 i‘lflvul . .. II... v ..If . It? (II.!:I. . . {nitriVytl‘t/prr): . p ..\ i3‘l-5'Il111fliét It. -.(4éél’ I. I uC'HPu¢ru.Df€(rv ..Npu; nuvrr-c... . .‘l’. I. 3. ..r1. Inr'tlvtto o'lflaci.‘ . .2». y... .. on... 9 {0:15 .ri‘.‘ {tail '3 ... 5.14 2.1.51 . .3 - .\ 3-55:... .15.! ‘3. lirlal‘v \r.“rrnl‘l .0...o.r.;: ylui‘t .. ).V$..tl)n»!lll.l . ;..LI.(‘Nv..E. 1‘1 . . (....l‘y‘tttrl ..llilttfv l.\ll'l|€r. {n’g . , 4. . $.st}.1y; .0 ~31... I-.. ob,v.(.suf.u\.lalinco’ I 31‘... ..r «Eil‘m 0:7,.1'. ‘g’g‘nilv .¢n..‘s..lil.c la... id‘s}: 72.31921?!» 110' if} vI‘h’Or. 8.115))!!- n V . \n A 1‘ . A ‘ v I v 1.5%: I .., . . . 4 I‘Il-cv-V: h‘l‘ul T- I. .n I _ .51-..“ . I, . . .. . . to. Iv Etfiy .- . .5: I- . . curs.(.\‘,v14.)1.rw:.0k,l. . Y3! I, . .th. ~l)» ...“ I‘ll}; \u‘ . . .( 5kg» . \él...’ . X; ill-x! , ...-1‘3 . .70 .1 (’33). as. a. ..O. r . lo.) ... . f . .10 . I. 0 l. ..- o it: .. ...IIFS t. ‘1 .4. culls)? 1(1p5‘4telfl....: ..\‘Pl’. Li): ,3; It“. . .05. 53.4 rllvlll G)“: .. \. stillinu“. UV“H0~».C:£ ‘filtN? .IL... )1: .. . ~ : .. .5! \, iaflSciharli.‘.t{flw(l.LlIl.‘3 .23... A21 (Isl? . 3C. _ ‘Ulypfilflhllfl.’ r In”? {it'll-r i}! l . 9»... ‘91.. n1. Tsar. .t‘LEiQt‘l \ is .r; .4! 6 ......) It? if‘os. 51. us} .. '7 v (0113‘ n .3 N. imMMJ :‘ ... .§ 1.5!, 6 El- U025}! ..ttglltlltl. 1‘10“?! .0 2’ (flat!!!) Vlvtlzv lo I flit. . ll I14? .190. .1. Pin... z: 111‘ .3“ ... 30 Price. -313... «.359... ....vvdohot 5.5.”). PM” 0.. l..§||.li.$_.v..o:r . . .5. o : Iln.‘ .1 . . l :49! I. VIII . :triaiicglo' rt €30.11. 3- . . .....If4.o: ti.tl.l.vviclo\l_v.. r fo‘ '1’. . . II. (13" v\ will f. L... .3. I . (.- . {1,14 ... 511.: .... ..r I: . . it i “q: :3 4: . $5.... 332.. t. ..tlx‘slu: V..14.18.I.n-..lt.v. . . $413. ..\evitt‘llleFQH‘wcil. <1. ‘55.. u c . .1711... . .s,.. . . .. . b Y 55......Kv.......hu.....c...!. . . . . . ... « h' 1r... . _ , I. r~...fl?.-n1. ,,.. u . . .. . s... o..- .--- 1 s. .e . ..Yt 1....afitquigkrwfli?gflgrin”? . . . ...... J . ....v..........\\1 . )n I! .1 .1 ...... .11. \(‘l l. 1: -.. . . . u. n ,l r'. k .- I .c .0. 1... .. f.....d.1u4.uu.t¢0$a a _ . . . - 1 ..‘O. ‘l‘u'l| .AVKI. o y; . . v ..e.. , r- o 1 . .-..-. wfl-....2v.n..1n:..r¢.-1 .1. w. . . 1 II p . ti. | .. . - L’ {i ‘wems gm; 551 lb NllVERS TY LIBRARIES I\1|ljluljv9|s3.wy§;2||;ml'H 1| L LIBRARY .‘ Michigan State University .-.-r— .— - .’ This is to certify that the dissertation entitled Panel Data Analysis on Farm-Level Efficiency Input Demand and Output Supply of Rice Farming In West Java, Indonesia - presented by Erwidodo has been accepted towards fulfillment L of the requirements for l Ph.D. degreein Agricultural Economics amass. AMMajor professor Date l2 April l990 MS U i: an Affirmative Action/Equal Opportunity Institution 0-12771 PLACE IN RETURN BOX to remove thIs checkout from your record. TO AVOID FINES return on or before date duo. DATE DUE . DATE DUE DATE DUE 1n) Mama IJL‘ W =bb \ .‘N 0: I995 \ {aflt ~ qty-2.193 ' 6' or; .‘ 14 . ...!) MSU Is An Affirmative Action/Equal Opportunity Institution PANEL DATA ANALYSIS ON FARM-LEVEL EFFICIENCY, INPUT DEMAND AND OUTPUT SUPPLY OF RICE FARMING IN NEST JAVA, INDONESIA By ERHIDODO A DISSERTATION Submitted to Michigan State University in partial fulfilment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Agricultural Economics 1990 Id“ LG Ul Q ABSTRACT PANEL DATA ANALYSIS ON FARM-LEVEL EFFICIENCY, INPUT DEMAND AND OUTPUT SUPPLY OF RICE FARMING IN NEST JAVA, INDONESIA By enutoooo This study has three main objectives: (i) to generate and evaluate parameters of rice production system, especially those related to efficiency, returns to scale, the demand for inputs and the supply of rice, (ii) to identify factor affecting farmers’ decisions in the adoption of new rice varieties, and (iii) to evaluate current agricultural policies, particularly price support and input subsidy policies. This study uses stochastic production and stochastic profit frontier models. Two functional forms (i.e. Cobb-Douglas and translog) are used in the analysis. The models are estimated using panel data from rice farming regions in West Java, Indonesia. Both single equation and system equation estimation methods are used in the analysis. In addition, a multinomial logit model is used to assess farmers’ behavior in adopting new rice varieties. This study confirms that rice production in Nest Java is in the stage of constant returns to scale. Rice farmers in West Java are found to be allocatively inefficiency and are, on the average, about seven percent technically inefficient and fourteen percent economically (profit) inefficient. Using average figures of profit per hectare and annual harvested area in Ilest Java, the total profit loss amounts to about seventy eight billions or eighty million US dollars annually. The benefits of promoting increased efficiency in rice farms in Indonesia appear to be considerable. The results also show that individual level of technical and profit inefficiency does not have any association with individual farm size, meaning that large farms may or* may not be technically or economically more efficient than small farmers. Provision of reasonable incentives for rice farmers is necessary if the adoption of new rice varieties is an important government’s priority. If increasing rice production and promoting rural employment are the government’s primary concern, this study verifies the argument that price support policy is more effective and less costly than fertilizer subsidy policy. Considering government’s budget constraints, the implementation of improved price support policy, coupled with a gradual reduction in input subsidies, may be one alternative to maintain the long-run rice self-sufficiency goal. Technological and institutional changes including efforts in improving marketing efficiency are necessary to simultaneously realize both higher producer and lower consumer prices. ACKNONLEDCEMENT Without the guidance, assistance, and support of so many people, this dissertation would never have been completed. In particular, I would like to thank Professor Eric Crawford, my'major professor, Professor Roy Black, the Chairman of ‘thesis committee, and other' members of 'the thesis committee, Professors Peter Schmidt and Rick Bernsten, who have given me the guidance, assistance and constructive participation which are important through out my doctoral program and the dissertation work. The guidance and assistance provided by other guidance comittee members, Professors Steve Harsh and Jack Meyer, are also sincerely appreciated. My doctoral program would not be possible without the support from the Government of Indonesia through the Badan Litbang Pertanian. In this regard, I would like to thank, Dr. Syarifuddin Baharsjah and Dr. Faisal Kasryno, both former’directors of the Pusat Penelitian Agro Ekonomi (PAE), and Dr. Effendi Pasandaran, who now the new director of PAE. To all my colleagues at PAE, who provided assistance while I was away from home, I also greatly appreciate them. I gratefully acknowledge Dr Robert Herd at the Rockefeller Foundation who provided me with a grant to cover the last three months living allowances, and in this regard the help of Professor Bernsten is highly appreciated. My appreciation is also extended to a number of persons at the Hinrock International for all the assistance they have given me. All friends and colleagues in the»department of Agricultural Economics, who have made my time at MSU enjoyable, deserve special recognition. The iv help of Meg Brown, who spent some time correcting my dissertation draft, is appreciated. I also want to thank to all my Indonesian friends at MSU for their friendship and cooperation. I would like to thank my parents, sisters and brothers for their ever- ending encouragement and moral support. The person that deserves the most recognition is my wife, Sri. Her sacrifices, patience, full understanding and encouragement are greatly appreciated. To my daughter, Andini, who was too little to understand what his dad was doing, but frequently missed his dad’s attention, I would like to thank her and hope she forgives me. TABLE OF CONTENTS LIST OF TABLES ........................................................ ix LIST OF APPENDICES .................................................... xi 1. INTRODUCTION ..................................................... l 1.1. Background and Problem Setting .............................. 1 1.2. Research objectives ......................................... 3 1.3. Organization of the Thesis .................................. 6 11. REVIEW OF INDONESIAN AGRICULTURE AND THE DATA .................... 8 2.1. Review of Indonesian Agriculture ........................... 8 2.1.1. Role of the Agricultural Sector ..................... 8 2.1.2. The Importance of Rice ............................. 11 2.1.3. Past and Current Agricultural Policies ............. 15 2.2. Data Set .................................................. 21 2.2.1. The survey ......................................... 21 2.2.2. The Survey Area and Sampling Procedure ............. 22 2.2.3. The Panel Nature of the Data ....................... 23 2. L 2. Sample Village Characteristics ..................... 24 III. LITERATURE REVIEW AND ANALYTICAL FRAMEHORK ...................... 31 3.1. Panel Data ................................................ 31 3.1.1. Fixed Effect: Dummy Variable Model ................. 34 3.1.2. Random Effect: Error Component Model ............... 37 3.1.3. Fixed or Random Effects ............................ 42 3.2. Production Function and Efficiency ........................ 45 3.2.1. Deterministic Non- -parametric Frontier .............. 46 3.2.2. Deterministic Parametric Frontier .................. 47 3. 2.3. Deterministic Statistical Frontier ................. 48 3. 2. 4. Stochastic Statistical Frontier .................... 48 3.2.5. Frontier Systems ................................... 51 3.3. Dual Approach: Profit, Demand and Supply Function ......... 54 3.3.1. The Case of Cobb-Douglas Profit Function ........... 55 3.3.2. The Case of Translog Profit Function ............... 60 3.4. Stochastic Profit Frontier and Profit Efficiency .......... 63 3.5. Panel Data and Stochastic Frontiers ....................... 67 3.6.. Adoption of New Rice Varieties: A multinomial Logit Model ............................................... 73 vi IV. VI. 3.7. Policy Evaluation: Rice Price Support and Fertilizer Subsidy ........................................ 75 MODEL SELECTION AND ESTIMATION METHODS .......................... 81 4.1. Production Function ....................................... 81 4.1.1. Model Selection and Functional Forms ............... 82 4.1.2. Estimation Methods ................................. 85 4.2. Profit Function ............................................ 86 4.2.1. Model Selection and Functional Forms ............... 87 4.2.2. Estimation Methods ................................. 88 4.3. System Equations: Profit function and Factor Share equations ................................................. 89 4.3.1. Model Selection and Functional Forms ............... 89 4.3.2. Estimation Methods ................................. 91 4.4. Multinomial Logit Model ................................... 92 4.4.1. Model Selection .................................... 92 4.4.2. Estimation Methods ................................. 94 4.5. Evaluation of Price Support and Fertilizer Subsidy ........ 94 4.6. Data Transformation ....................................... 98 EMPIRICAL RESULTS AND DISCUSSION ............................... 102 5.1. Results of Descriptive Analysis .......................... 102 5.2. Production Function and efficiency ....................... 109 5.2.1. Cobb-Douglas Production Function .................. 109 5.2.1. Translog Production Function ...................... 121 5.3. Profit Function and Profit Inefficiency .................. 126 5.3.1. Cobb-Douglas Profit Function ...................... 127 5.3.2. Translog Profit Function .......................... 136 5.4. Profit Maximizing Output Supply and Variable . Input Demand Functions ................................... 139 5.5. Adoption of New Rice Varieties ............................ 150 5.6. Evaluation of Price Support and Fertilizer Subsidy ........ 156 SUMMARY AND RECOMMENDATION ..................................... 165 6.1. Empirical Findings and Conclusions ....................... 167 6.1.1. Results of Descriptive Analysis ................... 167 6.1.2. Model Selection and Estimation Methods ............ 168 6.1.3. Production Function ............................... 170 vii 6.1.4. Profit Function ................................... 172 6.1.5. Output Supply and Derived Input Demand ............ 173 6.1.6. Multinomial Logit: Adoption of HYV ................ 174 6.1.7. Evaluation of Price Support and Fertilizer Subsidy ........................................... 176 6.2. Policy Implication ........................................ 178 6.3. Recommendation for Further Research ....................... 180 BIBLIOGRAPHY ................................................... 182 APPENDICES ..................................................... 187 viii LIST OF TABLES RAGE Sectoral Share (%) to Gross Domestic Product (GDP) and Employment ............................................. 9 Number of Farm Households According to Farm Size and Status ............................................ 11 Area Harvested (000 Ha), Production (000 Ton) and Yield (00 Kg/Ha) of Rice in Indonesia ....................... 14 Per Capita Consumption and Food Balance Sheet for Rice (in Million Tons) ...................................... 16 Number of Respondents in Each Village ........................... 24 Sample Village Characteristics .................................. 25 Mean Values of Input and Output Per Hectare for Rice Farming ................................................... 104 Deflated Prices of Inputs, Rough Rice and Profit ..................................................... 107 Estimated Parameters of the Cobb-Douglas Production Function ............................................ 112 Frequency Distribution of Farmers Based on the Level of Technical Inefficiency from the Cobb-Douglas Production Frontier ........................... 119 Allocative Inefficiency Measures ('d” ratio) in Rice Production ............................................. 121 Estimated Parameters of the Translog Production Function ............................................ 123 Frequency Distribution of Farmers Based on the Level of Technical Inefficiency from the Translog Production Frontier ............................... 126 Estimated Parameters of the Cobb-Douglas Normalized Profit Function ................................................ 128 Direct and Indirect Production Elasticities of the Cobb-Douglas Profit Function ............................ 132 ix .10. .11. .12. .13. .14. .15. .16. .17. .18. .19. .20. .21. .22. The Own- and Cross-Price Elasticities and the Elasticities with Respect to Land of the Output Supply and Variable Input Demand ........................ 134 Frequency Distribution of Farmers Based on the Level of Profit Inefficiency from the Cobb-Douglas Normalized Profit Frontier ..................................... 135 Estimated Parameters of the Translog Normalized Profit Function ................................................ 137 Frequency Distribution of Farmers Based on the Level of Profit Inefficiency from the Translog Normalized Profit Frontier ..................................... 140 Estimated Parameters of the Cobb-Douglas Normalized Profit Function Using Seemingly Unrelated Regression Method ......................................................... 143 Estimated Parameters of the Translog Normalized Profit Function Using Seemingly Unrelated Regression Method ......................................................... 144 The Own- and Cross-Price Elasticities and the Elasticity with Respect to Land of Rough Rice Supply and Variable Input Demand ...................................... 147 Direct and Indirect Elasticities of the Cobb-Douglas Production Function ............................................ 150 Estimated Coefficients of the Multinomial Logit Model .................................................... 152 Mean Probabilities of Varietal Choise .......................... 154 Elasticities Derived from the Multinomial Logit Model .................................................... 155 The Impacts of Rice Price Support and Fertilizer Subsidy Policies .................................... 158 Government’s Additional Cost for an Additional Unit of Rough Rice and Rural Employment ............................. 159 LIST OF APPENDICES AEEENQIX 28.6.5 2.1. The BIMAS Package per Hectare ................................... 187 2.2. Fertilizer Price Under BIMAS and INMAS Program .................. 188 2.3. Trends in Imported and Actual Rice Prices in Jakarta (USS per Ton) .................................................. 189 2.4. Price Structure for Urea and Triple Superphosphate (TSP) in 1982 .................................................. 190 2.5. Government Budget Cost of Fertilizer Subsidy (1981/1982) ....... 191 3.1 Derivation of Response Elasticities of the Multinomial Logit Model .................................................... 192 5.1 Input and Output Per Hectare of High Yielding Rice Variety (HYV) .................................................. 194 5.2. Input and Output Per Hectare of Traditional Rice Variety (TV) .............................................. 195 5.3. Input and Output Per Hectare of Mixed Varities (MV) ............ 196 5.4. Abbreviation of the Variables .................................. 197 5.5. Individual Level of Technical Inefficiency Estimated from Cobb- -Douglas Production Frontier .......................... 200 5.6. Individual Level of Technical Inefficiency Estimated from Translog Production Frontier .............................. 204 5.7 Individual Level of Profit Inefficiency Estimated from Cobb-Douglas Profit Frontier .............................. 208 5.8. Individual Level of Profit Inefficiency Estimated from Translog Profit Frontier .................................. 212 5.9. Evaluation of Rice Price Support and Fertilizer Subsidy ........................................................ 216 5.10. BULOG Domestic Rice Procurement (Tons) in Indonesia ............ 221 xi I. INTRODUCTION 1.1. Research Background and Problems Setting. During the first half of the 19603, rice production in Indonesia remained static while population grew rapidly, resulting in a significant food shortages and eventually'making Indonesia as the world’s largest rice importing country. Faced with this situation, the government has implemented numerous policy instruments and programs. The rehabilitation and expansion of the irrigation systems, continued investment in agricultural research capacity, the implementation of nation-wide rice intensification programs, the introduction of high yielding rice varieties, and continued provision of highly subsidized inputs (fertilizers and pesticides) were, among others, the» major factors signifying the increase of rice production in Indonesia. All these efforts have resulted in significant benefits during the late of 19605 up to the late of 19705 when the rate of growth of rice was as high as etght percent annually. Since 1985, after struggling more than two decades, Indonesia has become self-sufficient in rice. Much of the early success, however, did not persist and the rate of growth of rice has been declining. During the period of 1982-1985, for instance, the annual growth rate of rice production declined to five percent, and continuously declining to only 1.3 percent during the period of 1985-1987. Given this declining growth rate of the rice production on the one hand, and a rapid growth of total rice demand (stemming from a steady population growth rate of 2.3 percent and increased private income) on the other hand, critics recently have questioned the ability of the 1 2 Indonesian government to maintain rice self-sufficiency in the long run. Some scientists consistently assert that the strategy for reaching rice self-sufficiency in Indonesia is very costly and can be achieved only at the expense of efficiency. Thes strategy, as applied in most Asian countries, has been characterized by high levels of subsidies on capital infrastructure, production inputs, as well as subsidies for governmental marketing systems, resulting in heavy annual budgetary expenditures by the government. Considering the government’s recent budget difficulties, particularly due to the significant reduction of oil revenues, new policies should be directed to find the least costly alternatives. The most recent policy direction in Indonesia is a general movement toward phasing out all type of subsidies. Besides the macro issues and problems mentioned above, attention should also be directed to the micro level to assess common problems faced by farmers, since farmers are the key factor for successful implementation of any agricultural policy. For example, the optimal application of new technologies and cultivation practices depend, among other things, on the farmer’s understanding of these new technologies as well as their capability to adopt them. Better understanding will lead farmers to optimally allocate production inputs, which in turns leads them to achieve maximum profits. Identification of factors affecting the adoption of new rice varieties seems urgently needed, since to date the traditional varieties are still widely grown, despite the high yielding varieties have long been introduced. Measurement of farm level efficiencies along with identification of factors affecting them are also important for policy makers to improve efforts in increasing rice production. This study aims to examine the above issues, particularly those related 3 to the following policy' questions. Do farmers, with their' existing constraints and knowledge, behave rationally, i.e. utilize the available resources to produce in economically (technically and allocatively) efficient ways? If most farmers behave inefficiently, how inefficient are they? What factors affect individual farm efficiency. Does the rice yield increasing technology increase farms profit and rural employment? If it indeed increases profitability why are traditional varieties still widely used? Nhat factors determine the adoption of high yielding varieties? 1s a tendency for land consolidation in Java theoretically justifiable? Are existing programs, price support and subsidies among others, economically justifiable, especially in terms of increasing productivity, farmer’s income and promoting employment in the rural areas? Which one is more desirable, price support or fertilizer subsidy' policy, in ‘terms of increasing rice production and rural employment? 1.2. Research Objectives Using a stochastic frontier model and panel data, this research has the following main objectives: (a) to generate and evaluate parameters of rice production system, especially those related to farm level efficiency, return to scale, demand for inputs and supply of rice. (2) to identify factors affecting farmers’ decisions in the adoption of new rice varieties. (3) to evaluate current agricultural policies, particularly price support and input subsidy policies. 4 Efficiency measures of rice farming are important for the government to improve its efforts to increase rice production in less costly ways. Better information about efficiency is also needed to assess policies in agrarian reform, in general, and in particular on planning and reorganizing rice production system in terms of resource allocation and agricultural extension policies. If farmers, for example, are found to be inefficient, and the level of efficiency is not the same among (groups of) farmers, the government can use this information to set up group-specific extension policies to improve efficiency. A better measure of efficiency provides decision makers information about the potential benefits from promoting improved efficiency. Another illustration of the potential value of information on farm level efficiency is related to credit policy. Since the late 19705, for example, the government has subsidized tractor adoption by providing credit at low interest rates to farmers wishing to purchase tractors. In fact, only large farmers benefit from this policy due to their ability to provide collateral to obtain the loan. If the economic efficiency level of large farmers is the same as that of small farmers, providing fair tractor-rental arrangements is probably a more desirable policy, since participation is accessible to both groups of farmers. Knowledge of returns to scale is very important in assessing the potential economic impact of alternative land reform policy. If, for instance, increasing returns to scale exist, it can be used as a basis for supporting any effort toward land consolidation, or at least not excessively opposing land consolidation, which has spontaneously been occurring in Java. New rice varieties, commonly referred to as high yielding varieties 5 (HYV), have long been promoted by the government to accelerate the growth of rice production. In some places, however, the traditional varieties which are usually low yielding varieties, are still used by farmers. Identifying factors which affect the adoption process is urgently needed by ‘the policy' makers to formulate policy alternatives to encourage farmers’ adoptionof new varieties. The Indonesian government’s price policy to increase rice production consists of (i) subsidizing inputs, and (ii) establishing a floor price for unmilled and «filled rice. Some scientists, such as Timmer (1975), argue that using fertilizer subsidies does not provide a strong incentive. for farmers to increase production. Some argue that price supports would be more effective than input subsidies in increasing rice production and promoting employment. The estimates of own- and cross-elasticities of factor demand and output supply, which will be obtained in this study, can be used to determine whether to put more emphasis in the implementation of input subsidy or price support policy. The above issues and questions are not new and some of them have already been evaluated in some previous studies. This study, however, attempts to evaluate the above questions using a different analytical framework. For instance, instead of using non-frontier models in a cross- sectional framework. in assessing efficiency (e.g., Sugianto (1982), Kasryno (1986)), this study uses stochastic frontier models in a panel data framework. Unlike Hutabarat (1985) and Gunawan (1988) who applied probit and logit model, respectively, this study applies a multinomial logit model in assessing determinants of ‘the adoption of new rice varieties. 6 Farm-level panel data used in the analysis are obtained from the Center for Agra-Economic Research, Ministry of Agriculture, Indonesia. Panel data, which are time-series on a set of cross-sectional data, offer several major advantages over conventional cross-sectional or time-series data (described later). To date several studies have used these data (Hutabarat, 1985; Kasryno, 1985; Gunawan, 1988), but these studies did not use the data in the framework of panel data analysis. 1.3. Organization of the Thesis Chapter 2 presents a brief review of Indonesian agriculture along with the description of the survey and the data set used in this study. In reviewing Indonesian agriculture attention is given to presenting the role of the agricultural sector, the importance of rice, and the past and current agricultural policies. The description of the data set starts by presenting the general description of the study area, the nature of the survey, followed by the description of village characteristics. Chapter 3, is a literature review of the analytical framework used in this study taking into account the nature of the data, which mainly consists of reviewing panel data analysis, frontier models, dual approach of the production theory, a multinomial logit model, and a literature of price policy evaluation. Chapter 4 describes the model selection, formulation as well as the parameter estimation procedures. The proposed policy evaluation procedure and data transformation for obtaining the variables included in the model are also discussed in this chapter. Discussion of empirical results is presented in chapter 5, beginning with the results of the descriptive analysis, and followed by production function and production inefficiency, 7 profit function and profit inefficiency, output supply and input demand function and the results of the multinomial logit model. Finally, chapter 6 presents the conclusions of the study consisting of the summary of the empirical findings, policy implications and recommendations for further research. II. REVIEH OF INDONESIAN AGRICULTURE AND THE DATA SET This chapter consists of two sections. Section one briefly reviews Indonesian agriculture, beginning with the role of agricultural sector in the Indonesian economy, the importance of rice and the past and current agricultural policies. Section two describes the data set used in this study, consisting of the description of the survey, the survey area and the sampling procedure and a brief description of characteristic of the sample villages. 2.1. Review of Indonesian Agriculture 2.1.1. Role of the Agricultural Sector The relative importance of a sector in the economy can be assessed from its contribution to gross domestic product (GDP) and to employment. Through 1985, the agricultural sector has played a dominant role in the Indonesian economy in these two respects. In 1985, the agricultural sector accounted for 23.6% of the GDP, followed by mining and quarrying, public and other services, trade and commerce, and manufacturing industry, which accounted for 16.2%, 15.8%, 15.4%, and 13.2%, respectively. At the same time, the share of the agricultural sector of the total employment amounted to 54.7%, followed by the trade and commerce (14.9%), public and other services (13.4%), and the manufacturing industry (9.3%). The breakdown of sectoral contribution to GDP and employment is presented in table 2.1. 9 Table 2.1. Sectoral Share (%) to the Gross Domestic Product (GDP) and Employment Sectors GDP: _§mplgymgnt_ 1971 1985 1971 1985 Agriculture {419 'm23.6 66.4 _54.7 Farm Food Crops 26.2 14.7 na na Farm Non-Food Crops 5.3 3.2 na na Estate Crops 2.9 0.6 na na Livestock Products 3.4 2.4 na na Forestry 3.9 0.1 na na Fishing 3.2 1.6 na na Mining and Quarrying 8.0 16.2 0.2 0.7 Manufacturing Industry 8.4 13.2 6.8 9.3 Electricity, gas and water 0.5 0.8 0.1 0.1 Construction 3.5 5.3 1.7 1.7 Trade and Commerce 16.1 15.4 10.8 14.9 Transport/Communications 4.4 6.5 2.4 3 1 Banking & Finance 1.2 2.9 0.3 0.4 Public 5 other services 13.0 15.8 11.2 13.4 Source: CBS. Statistical Year Book of Indonesia * Based on current price na - not available From table 2.1, it is obvious that the relative importance of the agricultural sector in the Indonesian economy, even though still dominant, 10 is gradually declining. Thjgwghjft Suggests a structural change of the economy, the shift from agriculture to others. In 1971, for instance, the agricultural sector accounted for about 45%.of the GDP and declined to 24% in 1985. Similarly, its share in employment has declined from about 66% in 1971 to about 54% in 1985. On the other hand, the contribution of mining and quarrying, and manufacturing industry to GDP have increased significantly during this period. Table 2.1 also shows that the trade and comerce and the manufacturing industry are becoming relatively more important in generating employment. » Within the agricultural sector, food crops contributes the highest share to GDP, followed by non-food crops and livestock products (see table 2.1). The farm food crops can be classified into two distinct groups: (1) rice as the country’s primary food, and (ii) five other crops (traditionally classified as secondary food crops) consisting of maize, cassava, sweet potatoes, peanuts and soybeans. In terms of area grown (harvested) in the country, rice roughly accounts for more than 60% of the total food crap harvested area. The farm subsector is dominated by small-scale and traditional farm households. Table 2.2 shows that 45.7% of the farm households in 1973 operated farm sizes which averaged less than 0.5 hectare. This percentage increases to 63.3% in 1980. Number of landless farmers has consistently increased at a higher rate than owned-operated farmers. The landless farmers amounted to 3.2 percent of the total farm households in 1973, and increased to 14.9 percent of the total in 1980. 11 Table 2.2. Number of Farm Households According to Farm Size and Status Land - - Status Number Percent Number Percent W M511 WM Owned operated 4,907,495 34.2 7,914,305 45.4 Owned and rented 1,356,843 9.4 1,018,048 5.8 Rented 296,420 2.1 2,095,300 12.0 Mgrg than 0.5 ha; .ZIQIZLZQQ 55;} QISEQIQQZ §§&2 Owned operated 5,839,027 40.6 4,935,162 28.3 Owned and rented 1,813,831 12.6 999,254 5.7 Rented 159,926 1.1 506,491 2.9 Total 14,373,542 100.0 17,468,560 100.0 Source: adopted from Hutabarat, B. 1985 2.1.2. The Importance of Rice Rice is the most important food commodity in the Indonesian economy. It is the major domestically produced staple foodstuff. The production, processing and distribution of rice is one of the largest primary sources of income and employment in the Indonesian economy. Rice production accounts for 72 percent of food crop sector GDP, for 11 percent of national GDP and 14 percent of non-oil GDP (World Bank (1987) in Tabor (1988)). Rice is the main source of calories and protein in the Indonesian diet. It accounts for 60 percent of the total calories and more than 50 percent of the total protein in the diet. Moreover, rice accounts for a major 12 share of total consumer expenditures. Private consumption expenditures in Indonesia are dominated by outlays on necessities, such as food, housing, clothing and education. Based on 1984 National Household Expenditure Survey (SUSENAS), as reported in Tabor (1988), food expenditures alone accounted for 55 percent of the total private consumption expenditures. Rice expenditure made up 45 percent of the total food expenditures. For the poorest population quartile, rice expenditures account for 58 percent of the total household expenditures. , Not only does rice expenditure use up a major part of the household budget, but rice is commonly used as the primary'wage good of the economy. The direct consequence is that it is one of the cost push inflation factors. Beyond that, rice appears to have a psychological role in determining anticipated inflation and is considered a price leader by policy makers. In rural areas of Java, production of rice is the main source of employment and incomes. According to the Ministry of Agriculture (1988), as reported by Tabor (1988), rice production is estimated to provide 9.4 million full-time person-years of employment each year. In 1986, income from rice production and distribution, valued in 1985 wholesale prices, was about 11,000 billion rupiah, or about ten billion U.S. dollars (the exchange rate in this period was roughly Rp 1100/USS). Rice in Indonesia is considered a politically sensitive and strategic comodity. A national stockpile of rice is maintained and a National Logistics Agency (BULOG) is designated to handle this task as the sole importer and exporter of rice. The government provides large subsidies from its recurrent and development budget for irrigation investment and for the production and distribution of chemical fertilizers. 13 In Indonesia, rice is produced both in irrigated land (wetland) and in dryland. However, wetland rice production accounts for more than 90% of the total rice production. The yield of the wetland rice is much higher than that of dryland rice. This is not surprising since not only more government efforts are concentrated in the wetland rice production, but the dryland rice cultivation is dominated by subsistence farmers using traditional varieties with very simple cultivation and management practices. There is a distinct geographic distribution of rice production system in Indonesia. In.1982, for example, more than 62% of the total rice was produced in Java. Wetland rice is mostly cultivated in Java, which in 1983 accounted for more than 64%»of total wetland rice production. 0n the other hand, about 70% of total dryland rice was produced off Java. The total rice production, due to various government efforts and policies discussed later, has been increasing in the past few years with the annual growth rate 8.5% in the period of 1979-1982 and 5.1% in the period of 1982-1985 as presented in Table 2.3. This increase of the total rice production is mostly attributable to the significant increase in yield, for both wetland and dryland rice production. The annual growth rate of rice yield in the period of 1979-1982, for instance, was 7.8% for wetland rice and 5.6%ifor dryland rice. For the period 1982-1985, however, the growth rate of the rice yield declined significantly to only 4.8% for wetland and 2.6 for dryland rice. 14 Table 2.3. Area Harvested (000 Ha), Production (000 M.Ton) and Yield (00 Kg/Ha) of Wetland and Dryland Paddy in Indonesia (1979-1985). 1979 1982 1985 9:9!1n_r§13* 1979-82 1982-85 W Area Harvested 7,675 7,873 8,756 0.9 3.6 Production 24,732 31,776 37,027 8.7 5.2 Yield 32.22 40.36 42.29 7.8 4.8 W Area Harvested 1,128 1,116 1,147 -0.4 0.9 Production 1,551 1,808 2,005 5.2 3.5 Yield 13.74 16.20 17.49 5.6 2.6 M Area Harvested 8,803 8,988 9,902 0.7 3.3 Production 26,283 33,584 39,033 8.5 5.1 Yield ' 29.85 37.36 39.77 7.8 2.1 Source: CBS, Statistical Year Books of Indonesia * computed from Vt . Vo (1+r)t \ Although rice production has been increasing during the past two decades, it has not been sufficient to meet the total consumption of the country. Consequently, the country has been forced to import rice. During the period of 1968-1983, Indonesia was well known as the world’s largest rice importing country. The significant growth in the total rice consumption is attributed to a rapid population growth and a steady + a... . //< r' 15 improvement of the Indonesian people’s living standard. Table 2.4 shows a steady increase in per capita rice consumption. Since 1985 Indonesia has been a rice self-sufficient country. Despite this success, however, some potential problems still face the Indonesian government in the near future. A decreasing growth rate of the total rice production and steady growth rate of the population and the people’s standard of living have recently led some Indonesian scientists to question the ability of the Indonesian government to maintain the rice self-sufficiency goal in the long run. Their concern comes at a time when budget constraints have forced the government to reconsider some of agricultural programs which were responsible for achieving the rice self- sufficiency goal. The next section presents a brief overview of the past and current agricultural policies and programs. The recent policy issues which will be evaluated in this study will also be briefly outlined. 2.1.3. Past and Current Agricultural Policies Until the late of 19705, the primary objective of agriculture policy was to increase food, particularly rice, production in order to meet the accelerating growth of food consumption. Thus the policy target was very clear: to produce as much food as possible, regardless of any post- problems. Provincial targets for food production, by type of crop, were set annually, and all government efforts were focused on that target. A substantial portion of the government budget was devoted to obtain the target. No significant financial problems were facing the government, since as an OPEC member, the Indonesian government has been benefited from the monopolistic nature of the international oil market. Problems related 16 to rice quality and inefficiency, among others, did not concern policy makers at that time. Table 2.4. Per Capita Consumption and Food Balance Sheet for Rice (in Millions Tons) Year Product- Change Waste Import Export Availa- Per Cap. ion in Seed ble for cons. Stock Feed Cons. (kg/year) 1970 13.05 0.22 1.87 1.18 0 12.13 104 1971 13.66 0.04 2.01 0.57 0 12.19 102 1972 13.24 -0.31 1.92 0.73 0 12.36 101 1973 14.45 0.32 2.06 1.71 0 13.78 110 1974 15.21 0.48 2.15 1.07 0 13.64 106 1975 15.19 0.09 2.15 0.67 0 13.63 104 1976 15.75 -0.14 2.21 1.29 0 14.96 112 1977 15.87 -0.05 2.23 1.96 0 15.65 114 1978 17.35 0.61 2.44 1.84 0 16.14 115 1979 17.84 -0.42 2.51 1.93 0 17.69 123 1980 19.97 1.27 2.82 2.03 0.01 17.89 122 1981 22.07 0.33 3.13 0.53 0 19.13 128 1982 22.78 0.11 3.20 0.30 0 19.77 129 1983 23.90 -1.08 3.38 1.15 0 22.76 145 1984 25.75 1.35 3.62 0.38 0 21.15 132 1985 26.47 - 3.72 0.03 0.27 22.52 137 Source: adopted from Tabor et al. (1988) -.. 1.3....- 1- .5’ -.-\ . "“‘"‘-\-‘.. 17 Several policies and programs deserve to be mentioned here. During the first half of the 19605, rice production remained static while population grew very fast, resulting in a significant food shortages. Faced with this situation, the government has introduced nation-wide rice intensification programs which consist of the 'mass guidance“ program (BIMAS) started in the wet season 1964/1965, and the "mass-intensification“ (INMAS), started in 1967. The BIMAS program embodies three basic principles (Horld Bank, 1978; Nestel, 1985): (i) the ideology of modern rice farming (PANCA-USAHA or five endeavors) which consists of: proper soil preparation, proper irrigation, improved seeds, proper fertilizer application and proper use of pesticides, (ii) credits to purchase a package of improved inputs (see appendix 2.1), and (iii) intensive guidance (extension) for participating farmers. The first and the third components are part of the INMAS program. The introduction of the high yielding varieties (HYV) of rice and the continuous supply*of'highly subsidized inputs (fertilizers and pesticides) coupled with the rehabilitation, upgrading and expansion of irrigation systems were, among others, the major factors stimulating the increase of rice production in Indonesia. Rehabilitation and establishment of new transportation and other facilities were also undertaken as a set of extensive development strategies, especially in relation to dry-land development both on Java and off Java. All these efforts resulted in obvious benefits from the late 19605 to the late 19705 when the rate of growth of rice was as high as 8% annually. Much of the previous success, however, did not persist and the rate of growth of rice has been declining. During the period of 1982-1985, for instance, the growth rate of rice production was 5% per year, declining to only 1.3% during the period of 1985-1987. Given this declining growth 18 \ rate of the rice production and a steady population growth rate of 2.3%, Lit will not be too surprising if Indonesia, in the near future, returns :to its former role as the world’s largest rice importing country. In addition to the rice self-sufficiency objective, there are three other objectives of the agricultural development in Indonesia (Nestel, 1985): (i) improvement of farm incomes and rural employment in the interest of achieving better income distribution within the society, (ii) provide urban consumers with rice at a “reasonable” and relatively stable price, and (iii) control the budget subsidies to producers and consumers which have been given in pursuit of the other objectives. These objectives are often in conflict with each other, in the sense that attaining more i Lgof one requires some sacrifice of another. A principle instrument in the prursuit of these objectives is pricing policy. i Rice was the first food commodity for which the government seriously intervened in the market. The basic philosophy of the rice price policy, 1 as summarized by Mears (1981), was: (i) support for floor prices high enough to stimulate domestic production, (ii) ceiling price protection assuring a reasonable price for consumers, (iii) sufficient range between 1 these two prices to provide traders and millers reasonable profit after I holding rice between crop seasons, (iv) appropriate price relationships within Indonesia and internationally. In addition, inter-regional price spreads were intended to be sufficient to enable traders to cover costs of rice movement from surplus to deficit areas, and domestic prices were to be insulated from world prices to avoid large fluctuation in the domestic price. The floor price of rice is set high enough to stimulate domestic Production, improve farm income and promote rural employment. The floor 19 price is determined on the basis of an incremental benefit-cost ratio that results from the participation in the BIMAS program. It is set annually such that the magnitude of the B/C ratio, as well as the rice-fertilizer ratio, is sufficient to induce farmers to join the intensification program. The ceiling price, on the other hand, is set low enough to provide a price subsidy to consumers and to contain the rate of inflation, but it is reasonable enough in order to provide incentives for trader. The prices of fertilizers and the floor prices of rough rice and rice are presented in appendix 2.2. Since the basic philosophy of the government pricing policy was first implemented in the early 19705, its application has evolved in response to changing circumstances and pressures. In particular, substantial economic and budget subsidies, especially for fertilizer, have been introduced, which to some extent have resulted in departures from the original principles of the rice price policy. Currently two types of *mw'fi“ mane—mm ‘M subsidies are used; namely, budget subsidies which involve government cash “I... am o-n-w .- "ill-'5 ...—r AM“ a 9319;0th from.develeposnt.-.by§seh andhemmmisfiubsidies which. involve «0110112951995 ..be].ow_.the .anortunitx-.CPst as reflected by long-run world prices (Nestel, 1985). During the period 1970-1982, Indonesia generally maintained a domestic price for rice below the import parity price (see appendix 2.3). Only in 1976, 1977, and 1982, when the world price was well below its long-run trend level, was the domestic rice price above the import price (Nestel, 1985). The domestic prices of fertilizers are also kept well below their import parity prices. In 1982, for example, the economic prices (i.e., price in the absence of any subsidies) of urea and triple superphosphate (TSP) at the farm-gate were estimated to be Rp ISO/kg and Rp 171/kg, 20 respectively, compared to the official price of Rp 70/Kg (see appendix 2.4). Moreover, there is also an economic subsidy involved in the domestic production of urea, where the suppliers of natural gas .for urea manufacture receive a price lower than the opportunity cost of that gas. Although economic subsidies may involve efficiency costs for the economy, they do not necessarily involve cash outlays from the government budget. Specifically, differences between domestic and import parity prices will give rise to a budget subsidy only when rice or fertilizer is imported. Thus, imports of these commodities during 19705, for example, were a substantial burden on the budget. Since Indonesia is now a competitive urea producer, and its urea production will continue to grow rapidly, no urea imports are needed in the coming years. The economic subsidy implicit in the low price of natural gas does not have a direct budget impact, rather it simply involves foregone revenues for the gas producer (the gas producer is a government owned enterprise).. The government budget for fertilizer subsidies consists of two main components: .Lily'the costs for selling urea and TSP fertilizers at prices considerably below the full production cost, (it); the import costs for some imported fertilizers, including TSP. The 1981/1982 government budget for these subsidies was estimated to be USS 370 million (appendix 2.5), which is equivalent to 30% of the total agricultural development budget. The other type of subsidy arises from the selling price of rice which does not reflect its full costs of storage and other marketing costs. In 1982, the price subsidy for rice consumers was estimated about Rp 30/kg. Given the government’s recent difficulties in raising the development budget in the one hand, and the developmental needs of other sectors in the economy in the other hand, attention is now focused on controlling 21 subsidies such as fertilizer and rice price subsidies, which have a substantial impact on the government budget. A recent policy issue among Indonesian social scientists and policy makers is a general movement toward a phasing out all types of subsidies. The question effects of such policy changes on the rice production and consumption in general, on farm income, and on the real cost to the government, need to be carefully evaluated. As stated in the previous chapter, this study attempts to evaluate a small part of the above policy question. 2.2. Data Set 2.2.1. The survey The data set used in this study was collected by the Agro Economic Survey, as part of the Rural Dynamic Study in the rice production area of the Cimanuk River Basin, West Java, and obtained from the Center for Agro Economic Research, Ministry of Agriculture, Indonesia. Originally, the project was to be conducted over several consecutive years in order to observe individual farm response to economic and social stimulus. Farm household activities related to production, labor utilization, and consumption were to be continually monitored. Started in 1977, the project was terminated due to lack of funds, after having recorded two full crop years. The project was fortunately reinstated in 1983 in the same location and with the same farmers to cover the wet season of 1982/1983 and the dry season of 1983. In 1977, the survey was conducted twice, that is at the beginning and the end of the year. The first survey gathered information on farming practices in the wet season of 1975/1976 and the dry-season of 1976. The second survey covered farm household activities in the wet season of 22 1976/1977. A similar survey' was undertaken in 1978 to cover farm management activities in the dry season of 1977. The resurvey of 1983 to the same areas and same farmers was conducted with a different emphasis on labor utilization, asset holding, and land tenure arrangements. 2.2.2. The Survey Area and Sampling Procedure The survey area, which is the rice production area in the Cimanuk river basin, is characterized by irrigated rice farms and an almost uniform agroclimate. It covers six desa (villages) located in five kabupaten (the administrative unit between district and province level), namely: desa Wargabinagun in kabupaten Cirebon, desa Lanjan in kabupaten Indramayu, desa Gunung Wangi and Malausma in kabupaten Majalengka, desa Sukaambit in kabupaten Sumedang and desa Ciwangi in kabupaten Garut. The six sample villages were drawn using a multi-stage stratified random sampling, based on the following four criteria: (i) percentage of irrigated sawah (paddy field), (ii) latitude stratum, (iii) accessibility to transportation, and (iv) proximity to township (Hutabarat, 1985). In addition, the desa selection was also based on the consideration to cover all five kabupaten. From each desa, 60 farmers were randomly drawn from four farm size strata, 15 farmers each, namely (1) below 0.25 ha, (ii) 0.25 to 0.50 ha, (ii) 0.5 ha to 1 ha, and (iv) above 1 ha. The attempt was to get a better representation of the farm households in the survey area. In practice, the number of farmers recorded in each season varied between and within villages. 23 2.2.3. The Panel Nature of the Data The data set generated by the survey is commonly referred to as a panel data set, since the individual farmer was observed over time. This data set will be used in its advantage manner, that is in the framework of panel data analysis discussed in the next chapter. To date, several studies have used these data. However, these studies, Sugianto (1982), Hutabarat (1986) and Gunawan (1987) among others, analyzed these data using separate cross-sectional analysis or by simply pooling the data The analysis of this study uses the so called balance design, where individuals are observed for the same lengths of time. Using the individual identification number to check and match individual respondents, only 171 respondents were found to have been continuously recorded for six seasons (Table 2.5). Some respondents, for various reasons, were replaced by the new ones in the next survey. Others were not recorded in a particular planting season since they were absent. All these respondents were excluded from the analysis. Some respondents were also excluded from the analysis because of incomplete information associated with them. It is possible, of course, to analyze panel data where individuals are observed for different lengths of time, that is t-1,2....T1, where i is individual’s subscript, which commonly referred to as imbalance design. Due to its potential computational difficulties, however, this study does not use this approach. In addition, the balance design is much simpler computationally, and the number of individuals is large enough to obtain reasonable degrees of freedom. 24 Table 2.5. Number of respondents in each Sample Village Desa Kabupaten Number of (Village) (Regency) Observation Wargabinangun Cirebon 19 Lanjan Indramayu 24 Gunung Hangi Majalengka 37 Malausma Majalengka 33 Sukaambit Sumedang 22 Ciwangi Garut 36 Total Observations 171 2.2.4. Sample Village Characteristics This section attempts to provide a brief description of the sample village characteristics. This is important since it is hypothesized that the variation of the level of production and profit is associated with the region where the individual farmer lives. Regions can be regarded as individual non-specific time invariant variables in the framework of panel data analysis. It should be noted that the description presented here is based on the information gathered in the first survey in 1976, and may differ from the present situation. Table 2.6 presents village characteristics in terms of available land types. As noted earlier, sample villages were selected such that they best represented irrigated rice farming in West Java. Three villages 25 (Wargabinangun, Lanjan and Gunungwagi) can be considered as lowland areas while the other three (Malausma, Sukaambit and Ciwangi) can be considered as highland areas. The following is the brief’description of each village. Table 2.6. Land Use Characteristic of the Sample Village Desa % Irrigated % Dryland Total Area land & Others (Ha) Wargabinangun 95 5 302 Lanjan 80 20 125 Gunung Wangi 56 44 338 Malausma 16 84 909 Sukaambit 30 70 578 Ciwangi 17 83 1726 11311931211110.4110 This village is relatively remote and difficult to reach since no public transportation connects this village to other villages. The village roads are unpaved and were constructed by local villagers. During the wet (rainy) season these roads are almost impassable. Bicycle is the main form of transportation for villagers. The village can be classified as a lowland area with an average altitude of 10-15 meters. As indicated in Table 2.2, this village is dominated by irrigated rice farms. Rice farming in this village, however, is not without problems. Poor drainage and water control are problems 26 regularly faced by farmers. During the wet season flooding is almost always a problem, while during the dry season most paddy fields lack irrigation water. There is no permanent local market available in the village. Farmers sell their products to small traders, often referred to as collectors, who periodically come to the village, particularly during harvest season. Host farmers cultivate TV, since HYVs are known not to grow optimally under local conditions. In addition, farmers prefer the taste of TV rice. In the past rice was only planted once a year, but since 1968 farmers now plant twice a year. Lanjan Lanjan is located in the northern part of Hest Java along the major Kerawang-Jakarta highway and thus is much more accessible than Hargabinangun. Unlike Wargabinangun, year-round public transportation is available. The village is relatively small, 125 ha total area, of which 80% is irrigated sawah and 20% is dryland. Most of the sawah can be cultivated with paddy twice a year: the remainder, due to water control problems, is cultivated only once a year. Rice is grown primarily on small farms with 60% of farms being less than one hectare. Interestingly, only 27% of the land is owned by villagers, the rest is owned by people outside the village. The majority of villagers are landless farm laborers. Farmers purchase fertilizer and other'production inputs from either the village cooperative unit (KUD/BUUD) as part of BIMAS credit or from a local retailer. The retailer’s price of fertilizer was lower than the cooperative’s price. 27 It should be noted that Lanjan is characterized by seasonal migration. During the harvest season, when employment opportunities in the village decline, some villagers temporarily leave to seek harvesting jobs in other villages and even other kabupaten. They move from one village to another until the harvest season ends. During the dry season, when there are no longer field activities available, they migrate to nearby cities such as Kerawang, Bekasi or even Jakarta to get temporary jobs, particularly as construction workers. GununsJiansl Gunung Wangi can be classified as a highland area, with the average altitude 875 meters. It is relatively remote, about 14 km away from the nearest kecamatan (district) town. Although available, transportation facilities are relatively poor, especially during the rainy season. This village covers 338 hectares, of which 189 hectares are sawah and 144 hectares are dryland. Most of the sawah can be cultivated with paddy twice a year. Traditional varieties of rice are commonly grown. Sawah in the village is irrigated by a local irrigation facility and irrigation water from the mountain and is available year round. Fertilizer is obtained as part of the BIMAS package, and since farmers grow TV rice, they receive package-B (150 kg/ha urea, and 50 kg/ha TSP). Farmers feel this amount of fertilizer'is insufficient, and purchase additional amounts of fertilizer from local private retailers. The majority of villagers are farmers some of whom are landless farm laborers. There is no evidence of out-migration from the village, but there is strong evidence of in-migration. Gunung Wangi has no permanent local market, so farmers rely on local or out-of-village traders who come 28 periodically to collect agricultural produce. 11mm Malausma is located about 12 km away from the district capital. It is a mountainous area with the altitude ranging from 600 to 1100 meters. This village has paved all-weather village roads. Public transportation is available, making this village highly accessible. The village covers 909 hectares, consisting of 150 hectares of irrigated sawah and the remaining area of dryland. Unlike other villages covered in the survey, most villagers are involved in small-scale home industry as well as working part time on their farms. Only about 34% of the villagers are involved purely in farming. Irrigated sawah is cultivated twice a year with both TV and HYV rice, while the dryland is planted to rice once a year which is then followed by secondary crops. Most of the rice is produced for own-consumption and the rest is sold to the rice hullers owners or to local traders. Fertilizer and other production inputs are obtained from the village cooperative unit, as part of the BIMAS credit or the INMAS program. These inputs can also be purchased from the local retailers at slightly higher prices. Sukaambit Sukaambit is about 10 km away from the capital of kabupaten Sumedang. Public transportation connecting the village to other villages and to the capital of the kabupaten is available. Thus there is no transportation problem. Moreover, although no village market is available, there are many small retailers, selling all basic needs and agricultural inputs. With 29 an average altitude of 375 meters, it is not clear whether the village should be categorized as a lowland or a highland area. Sukaambit covers 578 hectares, of which 70% is dryland and 30% is sawah. Rice, both TV and HYV, is cultivated twice a year. Some farmers follow a cropping pattern of rice-rice-secondary crops, but most (72%) follow a rice-rice pattern. Only 10%.of sawah is cultivated with rice once a year. Vegetables are also grown in this village, particularly in the dryland area. Most of the rice produced is for home consumption, with only a small part sold to the local traders or rice huller owners in the village. Since no local market is available, farmers usually sell their farm products to the»market in kecamatan. Fertilizer and other inputs are obtained from the village unit as part of BIMAS credit or purchased from the local retailers or retailers in kecamatan. 911131191 Ciwangi is located about 3 km away from the district capital. It can be classified as a highland area with the average altitude of 700 meters. Village roads connecting the village to district capital can be used by trucks or other small vehicles year round. Thus the village is highly accessible. This village covers 1726 hectares consisting of 83% dryland and 17% irrigated sawah. Most of the sawah is planted with rice twice a year. Water for the local irrigation facilities is obtained from a small river in the village and from a nearby mountain, but is not always available during the dry season. HYV rice has been widely adopted by farmers in this village. 30 Farmers under the BIMAS program obtain fertilizers and other inputs from ‘the 'village cooperative unit, while non-BIMAS farmers purchase production inputs from local or kecamatan retailers. Farm products are sold either to local traders or at the district capital market. The majority of villagers (80%) are involved in farm activities. Temporary or seasonal migration is common, particularly for those who work as traders. III. LITERATURE REVIEW AND ANALYTICAL FRAMEWORK This chapter reviews the literature related to the analytical framework used in this study. It consists of seven sections, and is organized as follows. The first section provides a brief review of panel data literature, followed by section two which reviews literature related to production function and efficiency'concept. Section three presents a short review of the dual theory of the production function, that is profit function approach, along with the derivation of output supply and input demand. Section four presents a derivation of stochastic profit frontiers, followed by section five which reviews stochastic frontier models in relation to panel data. A literature review of the multinomial logit model, which will be used to identify factors affecting the adoption of new rice varieties, is described in section six. Finally, section seven presents a brief review on the framework of agricultural and food policy evaluation in general, particularly related to price support and input subsidy policies. 3.1. Panel Data A panel data set, that is a cross section of individuals observed over time, provides additional observations and new sources of variation in the exogenous variables; in particular, variation between individuals and variation within individuals over time become distinguishable. In principle, this leads to more efficient estimates of the common parameters. More importantly, panel data also make possible consistent estimation when unobservable factors, specific to time or to individual, 31 32 are thought to be correlated with other explanatory variables. When panel data are used, one of the ultimate goals is to use all available information to make inferences on the parameters of interest. A simple model commonly postulated is that dependent variable (Y) is a linear function of the regressors (X). To run a least-squares (LS) regression with all observations (pooled), one needs to assume that the regression parameters take values common to all cross-sectional units for all time periods. If this assumption is not valid, the pooled LS estimates may lead to false inferences. ‘ One of the most simple models to take account of heterogeneity across individuals and/or through time is to use the variable-intercept models. The basic assumption of such a model is that, conditional on the observed explanatory variables, the effects of all omitted variables are driven by the following three types of variables (Hsiao, 1986): (i) individual time- invariant variables that are the same for a given cross-sectional unit through time but vary across cross-sectional unit, such as individual firm management ability, sex, and socioeconomic variables, (ii) period individual-invariant variables that are the same for all cross-sectional units at a given point in time but that vary through time, such as prices, interest rates, and widespread optimism and pessimism, and (iii) the individual time-varying variables that vary across-sectional units at a given point in time and also exhibit variation through time, for examples: firm profits, sales, capital stocks. The variable-intercept models can provide a fairly useful specification for fitting regression models using panel data. Let consider the following model (Hsiao, 1980): 33 YIt - 30 + 2k ak int + eit (3.1) eit - “MI + ZPt + vIt where i-1,2...N refers to a cross-sectional unit or individual, t-1,2...T refers to a given time period, M1 and Pt are individual- and time-effects variables while v": represents the effects of all remaining omitted variables. Unfortunately, there usually are no observations on "i and Pt' It is, therefore, impossible to estimate 11 and 2 directly. A natural alternative would then be to consider the effectsof the product, "i - uMi and zt - th, which then leads to a variable-intercept model. There are two basic approaches in treating the effects, that is to consider them as fixed or random. The fixed-effects model is viewed as one in which the investigator’makes inferences conditional on the effects that are in the sample. The random-effects model is viewed as one in which the investigator makes unconditional or marginal inferences with respect to the population of all effects. It is the investigator’s choice to decide whether to make inference with respect to the population characteristics or only'with respect to the effects that are in the sample, and it depends on the context of the data, the manner in which they were collected, and the environment from which they came. The following presentation is on the variable-intercept model, as this study will be using. More specifically, it is the model with constant slope coefficients and an intercept that varies over individuals but is constant over time. A varying intercept term is assumed to capture differences in behavior over individuals (individual effects). The presentation in this section drawn heavily from Judge et al.(1982), Judge et al.(1985), and Hsiao (1986). 34 Equation (3.1) can be rewritten as Ylt ' 301 + 2k 3k int + Vlt (3.2) where (i) i and t are defined as in equation (3.1), (ii) Y” is an observation on the dependent variable for the ith individual and tth time period, (iii) int is an observation on the kth explanatory variable for the ith individual and tth time period and assumed non-stochastic, (iv) Vit is the random error for ith individual and tth time period and is assumed to have zero mean and constant variance and independently distributed over time and individuals, (v) ak, k-1,2...K, are the slope coefficients which are assumed to be constant over time and individuals, and finally (vi) aoi’ i-1,2...N are the intercept terms that are assumed to be different for each individual but constant over time. 3.1.1. Fixed Effect: Dummy Variable Model In this section we consider the model in equation (3.2) with the assumption that a0, are fixed parameters and v“ are independent and identically distributed with E[Vit]'0 and E[v1t2]-ov2. This model is often known as a dummy variable model since it is possible to rewrite it as YIt ' 23 301 Djt + 2k bk int + vlt (3.3) where the Djt are dummy variables and take values zero or one, that is Djt equals to 1 if j is i and equals to zero if j is not 1. Using a Kronecker product notation, the complete set of NT observation can be written compactly as 35 as Y-[INeJT Xs][ao]+v (3.4) where Y'-(Yl’,...YN’), Xs’-(Xsl',..xsn’), v’-(vl',..vn’), ao-(aol,. ..aoN), as’-(al,...ak),«e is a symbol of Kronecker product, JT is a (TxI) vector of ones and IN 8 JT is a the (NTxN) matrix of dummy variables. Theoretically, given the above assumptions of v and X5, there is no estimation difficulties since the LS estimator can directly be applied and it is best linear unbiased. However, there could be numerical problems, if there are many individuals (N large) since the inversion of matrix (N+K) may become unreliable. Under these circumstances it is advisable to estimate a0 and as using partitioned inverse as as - (xg'uN e 0T)xs)' xs'uN e 0111! (3.5a) in - xi. -X.-.' a. (3.51,, where 01 - Ii- (Jr’il/T x - m 2. vi. 31d. ' VT 2t xkit X1. - (lli." ........ XKi.’) Matrix ”T is idempotent. When it is used to transform the observations on the ith individual, it has the effect of expressing each variable in terms of its deviation from the mean. The LS estimator of the slope coefficient as in the dummy variable model is obtained by simply applying 36 LS to the transformed model, that is (Yit ' 11.) ' 3k ak (xkit ' in.) + (Vit ' 1i.) (3-5’ This estimator is often referred to as least-squares dummy variable (LSDV) estimator or covariance estimator. It is also called the ”within“ estimator since it utilizes variation within individuals. It is worthwhile to note the estimation of avz. Both the LS with dummy variable and the “within" procedures lead to the same residual vector. The unbiased estimator of av2 is computed by dividing the residuals sum of squares (SSE) by (NT-N-K), which is obtained by simultaneously estimating a0 and as. However, if we use the “within“ transformation technique the variance estimator is biased, since it is a division of SSE by (NT-K). Under these circumstances it is advisable to correct this estimator by multiplying it by (NT-K)/(NT-N-K) (Judge et al., 1982). Statistical tests can be performed to test the null hypothesis that individuals have the same intercepts against the alternative hypothesis that they are not the same. One can compute relevant F statistic using restricted and unrestricted SSE as (SSER ' SSEUR)/(N'l) F - (3.7) SSEUR/(NT-N-K) where R and UR stand for restricted and unrestricted, (N-l) is the number of restrictions and (NT-N-K) is number of degrees of freedom in the 37 unrestricted model. Under the null hypothesis this statistic has an F distribution with (N-l) and (NT-N-K). In applied work judgments are frequently made whether some dumy variables should be excluded by examining the t values associated with individual coefficients. Such a practice, however, is not recommended (Judge et al., 1982). 3.1.2. Random Effect: The Error Component Model We again consider the same model as in equation (3.2). Instead of assuming that an, are fixed coefficients, we now assume that they are independent random variables with a mean a0 and variance one. The model can be rewritten as YIt - 80 + 2k 3k int + “I + vIt (3.8) when written in matrix notation this becomes YI ' X1 3 + "10T + V1 (3.9) where Y1, JT and Vi were previously defined. xi'(JT X51) is a Tx(K+1) matrix of'observations of the explanatory variables including the constant term for the ith individual and a’-(ao, 81,...ak). The term (u‘JT + v1) can be regarded as a composite disturbance vector that has mean zero and covariance matrix Y ‘ E[(U1JT + V1)(U1JT + VI)] (3.10) 2 I 2 ' a“ JTJT + 0v IT p 2 2 2 2 l I 0“ ”V on ...... 0U 2 2 2 2 ”u "u +"v "u 2 2 2 2 a“ 0“ ...... a“ +0v The structure of this covariance matrix is such that, for a given individual, the correlation between any'two disturbances in different time periods is the same. Thus in contrast to a first-order autoregressive model, the correlation is constant and does not decline as the disturbance become farther apart in time. Another feature of this covariance matrix is that it does not depend on i, and is identical for all individuals. The complete set of NxT observations can be presented more compactly as Y - X a + u o JT + v (3.11) with the covariance matrix 0 - IN eiV , which is block diagonal with each block.given by equation (3.10). The block.diagonal property arises because the disturbance vectors corresponding to different individuals are assumed to be uncorrelated. It is clear that the covariance matrix for the error component model is not of the scalar-identity type, and consequently although the LS applied directly to (3.11) is unbiased, there is a question whether it still remains best. This because the covariance matrix of the estimates will likely be biased, either understating or overstating the true 2 2 variances and covariances. If au are known, the generalized least and 0V squares estimator (GLS) is relevant in this instance and it is the best linear unbiased estimator. The GLS estimator is 39 a - (x' 0'1 x1'1 x' 0'1 v (3.12) If we partition this estimator as a'- (30 25’), it is possible to show that as' is a matrix weighted average of the “within“ estimator and another estimator which is known as the "between" estimator (Judge et.al, 1985; Hsiao, 1986). The latter is obtained by running the LS regression of the individual means, that is by averaging equation (3.2) over time. The weights are the inverses of the covariance matrices of the respective estimators. Thus the GLS estimator can be viewed as an efficient combination of these two estimators. The easiest way to obtain the GLS estimator is by transforming the data 1 and c is any by a transformation matrix P, where P is such that P’P - cO' scalar, and then running the LS regression of the transformed data using a standard LS computer package. Fuller and Battese (1973) suggest the transformation matrix P-IN x Pi’ where Pi - IT - (1 - w) JTJT/T (3 13) where w - ov/o and o2 - To“z + ovz. Pi’Pi - av2 01'] and hence P’P - av2 0“. In practice, the transformed observations are simply obtained by (i) calculating the individual means for each variable, and then (ii) expressing the observations in terms of deviations from a fraction 8, where 8 - 1 - w, of their individual means. For the constant term this means replacing a column of ones with a column containing the constant (1 - 8). The GLS estimator is obtained by applying LS regression on the 4O transformed data, that is ('it ' 911.) ' (1 ' Olin t 2k (xkit ' oflunk + (v1t - Ox, ) (3.14) It is obvious that if w 9 1 or 0 e 0 the GLS estimator (3k) converges to OLS estimator, while if w e 0 or 0 e 1 it converges to covariance or within estimator (equation 3.6). In essence, w measures the weight given to the between-individual variation. In the covariance procedure, this source of’variation is completely ignored. The OLS procedure, on the other hand, takes this source of variation completely by adding it to the within-individual variation. Thus, we can view OLS and covariance procedures as somewhat all-or-nothing ways of utilizing the between- individual variation (Hsiao, 1986). 2 are unknown we can not use the above GLS estimator. 2 If av and all However, it is possible to replace them with their estimates. When this is done the resulting estimator (3*) is often called the estimated generalized least squares (EGLS) 2* - (x' 01"1 X)'1 x' 0*'1 v (3.15) 2 u documented in literature (Judge et.al, 1985). One method for obtaining A large number of alternative estimators for o and 0‘,2 have been unbiased estimators of these variances is to use the residuals from the 2 v and the residuals from the ”between" 'within' estimator to estimate a estimator to estimate 02. 41 The EGLS can briefly be summarized in the following steps: (1) Calculate the dummy variable estimator or the ”within" estimator. (2) Use the residuals in step 1 to estimate ovz - SSE/(NT-N-K) (3.16) (3) Calculate the "between“ estimator, that is LS regression using the observations on the individual means. (4) Use the residuals from step 3 to estimate OZ/T - SSE/(N-K-l) (3.17) It is worthwhile to note that the between estimator can be performed by using NT rather than N observations, that is by repeating each of the N individual means T times. This practice will not change the estimator, but it will make SSE T times larger. In this case we estimate 02, not oz/T. (5) Calculate the estimator for the o 2 as U ouz - ( 02 - av2 )/T (3.18) The disadvantage of this estimator is the possibility of a negative 2 u . (6) Calculate the weight as 8 - 1 - w, where w - ov/o 0 (7) Transform the original observations using 8 and the individual means (8) Run a LS regression on the transformed observation as in equation (3.14) 42 The best linear unbiased predictor of the random components "1 can be obtained by using the following formula (Lee and Griffiths, 1979) 0 - ( To 2 /02 )( x - a - s a x ) (3 19) i u i. o k k ki. ° This prediction is sometimes of interest since it gives information on the future behavior of individual. 2 To test the presence of individual random effects, that is to test the 2 null hypothesis that ou - 0, one can use the Lagrange Multiplier test suggested by Breusch and Pagan (1980) as follows NT [ 2, (2t e”)Z LM - - 1 )2 (3.19) 2(T-1) 21 2t eitz Under the null hypothesis this statistic is asymptotically distributed as Chiz(1), where e is the vector of LS residuals obtained by regressing Y on X (equation 2.8). 3.1.3. Fixed or Random Effects Whether to treat the individual effects as fixed or random is not an easy question to answer. It can make a surprising amount of difference in the estimates of the parameters in the cases where T is small and N is large (this is the case of the panel data of this study). There are two strong opposing arguments. Advocates of the random effect model will argue that if the effects of omitted variables can be appropriately summarized by a random variable 43 and the individual (or time) effects represent the ignorance of the investigator, it does not seem reasonable to treat one source of ignorance (“1) as fixed and the other source of ignorance (“it) as random. It appears that one way to unify the fixed-effects and random-effects models is to assume from the outset that the effects are random. Mundl ak (1978) criticized the random effects formulation on the grounds that it neglects the correlation that may exist between the individual effects ("1) and the explanatory variables (xit)' There are reasons to believe that in many circumstances these two are correlated. For instance, the output of each firm (Yit) may be affected by unobservable managerial ability (“1)' Firms with more efficient management tend to produce more and use more inputs (Xi), while less efficient firms produce less and use fewer inputs. Ignoring this correlation leads to biased estimates of the parameters. One way to decide whether to use a fixed effects or random effects model is to test the null hypothesis that there is no correlation between the individual effects and the included explanatory variables against the alternative hypothesis that such correlation exists. For this purpose, we can use either Hausman test (1978) or an asymptotically equivalent test suggested by Mundlak (1978). If the null hypothesis holds we use the random-effects model, otherwise we use the fixed-effect model. The Hausman test is basically as follows. Under the null hypothesis, E[X’uJ-O, the GLS estimator should not be significantly different from the covariance estimator. Provided no other classical assumptions are violated, a statistically significant difference of these two estimators indicates that the alternative hypothesis holds, that is E[X’u] is different from zero. Hausman takes the difference of these two estimators, 44 say 9, and its variance, var(g), as a basis for the relevant test statistic. Var(g) is simply the difference between the variance of these two estimators, since the covariance of these two is zero. The test statistic is 0'1 mu) 1‘1 0 (3.20) where the null hypothesis is distributed as Chi2(K) and K is the number of explanatory variables (excluding the constant term). The Mundlak test is an F test based on the residuals from transformed equation (3.14) and the residuals from its modification as Y* - X*a + Xb + v* (3.21) where X is a matrix of the individual means of included explanatory variables (Kmenta, 1986). The test of the null hypothesis is therefore a test of significance of b, which is a standard F test. 3.1.4. Seemingly Unrelated Regression (SUR) With Error Component The concept of an error components model can be generalized beyond a single equation. Given a set of panel data, for example, one might wish to estimate a system of consumer demand equations or to estimate a system of equations consisting of profit (cost) function and factor share equations. Single equation error component procedures outlined in the previous section can not be used to obtain efficient parameter estimates of a system of equations unless error correlations between equations are assumed to be zero. 45 The derivation of the GLS estimator of SUR with an error component can be found in Avery (1977) and in Baltagi (1980). The problem with this estimator, particularly for applied users, is that no ready computer packages are available at this time. Moreover, the transformation matrix to allow the use of standard LS computer packages is apparently very complicated and is not yet in the literature. The only way, at this time, is to write the necessary computer program. 3.2. Production Function and Efficiency A production function describes technical relationships that transform inputs into outputs. It also shows the maximum possible output (frontier) attainable from a given combination of inputs. There cannot be any point above the production frontier. The distance a firm lies below its production frontier measures the level of inefficiency. A production process can be inefficient in two ways. It is technically inefficient if it fails to produce maximum output from a given input bundle. It is price or allocatively inefficient if the marginal revenue product of an input is not equal to the marginal cost of that input, resulting in utilization of inputs in the wrong proportions for given input and output prices. A combination of these two is usually referred to as total productive or economic inefficiency. There is, however, a discrepancy between the above definition and the one that is statistically estimated. The latter, which is usually referred to as a 'non-frontier' or an ”average“ production function and estimated by ordinary least square (OLS), allows some firms to be above the 'fitted' function. One can use the ”average" function to estimate technical inefficiency under a certain restrictive assumption. The result, however, 46 cannot be called a pure measure of technical inefficiency since it also includes random variability. The frontier production function is designed to bridge the gap by introducing the error term to represent an (in)efficiency measure. Forsund, Lovell and Schmidt (1980), in their survey on frontier production functions, distinguished between best practice frontiers and absolute frontiers. The former is estimated by those methods which fit a frontier without assuming the form of the distribution of the one-sided error. Consequently, the former yields 100% efficient firms. The latter does not yield 100% efficient firms since it is defined with respect to all the firms which could conceivably exist and embody current technology. 3.2.1. Deterministic Non-Parametric Frontier The concept of frontiers and efficiency measurement in production theory was initiated by Farrel (1957). His measure of technical and allocative inefficiency is defined in terms of a non-parametric and deterministic framework. In this approach output is assumed to be bounded by a deterministic production function. With the assumption of constant returns to scale, the production technology can be fully characterized by a unit isoquant. Since the efficient unit isoquant is unobservable, it has to be estimated from the observed input-output ratios of a sample firms by linear'(mathematical) programming techniques subject to the restriction that no observation on the input-output ratio can lie below the isoquant. The principal advantage of the approach is that no functional form is imposed on the data. The disadvantage is that the assumption of constant returns to scale (CRS) is restrictive, and its extension to non-CR5 is cumbersome. The other disadvantage is that the frontier is constructed 47 from a supporting subset of observations from the sample, and is therefore susceptible to outliers and measurement errors. 3.2.2. Deterministic Parametric Frontier In view of the difficulties in the non-parametric approach, Farrel proposed computing a parametric convex hull of the observed input-output ratios. Aigner and Chu (1968) were the first to follow Farrel's suggestion. They estimated a CD function: ln y - a0 + 2i a, ln xi + u, u s 0 (3.22) by programing methods, i.e., linear programing and quadratic programming. Once the parameters are estimated, technical inefficiency for each firm can be computed directly from the residuals. The principal advantage of the parametric approach compared to the non-parametric approach is the ability to accomnodate non-CRS. The problem is the specification of functional form which often imposes a limitation on the number of observations that can be technically efficient. In the homogenous CD case, for example, when linear programming algorithm is used, there will in general be only as many technically efficient observations as there are parameters to be estimated. Like the non- parametric approach, this is very sensitive to outliers. One possibility to improve the estimates, suggested by Aigner and Chu and implemented by Timmer (1971), is to discard some extreme (most efficient) observations until the resulting estimates stabilize. If the rate of change of the estimates with respect to succeeding deletions of observations diminishes rapidly, this suggestion will be useful. Finally, a common problem of all 48 the non-statistical models is that the estimators do not possess any statistical properties so that inferential results cannot be obtained. 3.2.3. Deterministic Statistical Frontiers In the Aigner-Chu model, once someldistributional assumptions are made on the one-sided error u, it becomes a statistical frontier model. The usual assumptions are (i) u is independently and identically distributed with finite mean and variance, (ii) u is uncorrelated with xi’s. Under these assumptions OLS gives consistent estimators of all parameters except an. A consistent estimator of a0 can be obtained once some distributional assumption is imposed on u. This is sometimes referred to as corrected OLS (COLS) originally proposed by Richmond (1974). The difficulty with the COLS approach is that we can get residuals, which are our estimates of inefficiency, with the wrong sign. One way of resolving the problem is to correct the estimate of a0 in such a way that no residual is positive and one is zero. This amounts to assuming that one firm is 100 percent efficient and then measuring the efficiency of other firms relative to this one. Another difficulty with the COLS methods is that the correction to a is dependent of the distribution of 11. For 0 example, the game and exponential distributions yield systematically different corrections for ac, and systematically different estimates of technical inefficiency, except for the special case when variance of u is one (Forsund et al., 1980). 3.2.4. Stochastic Statistical Frontiers. In the deterministic model the variation in firm performance relative> to the frontier is attributed to inefficiency, thereby ignoring the 49 possibility of variation due to factors not under the control of any firm such as weather variation, machine breakdown, luck, which is usually referred to as statistical noise. Combining these two together, as in deterministic frontiers, and labelling it as inefficiency is not appropriate. The statistical noise needs to be separated from the controllable factors that are designated as inefficiency. This is the essential idea behind the stochastic production frontier model developed independently by Aigner, Lovell and Schmidt (1977) and Meeusen and van den Broeck (1977). Their model, which can also be referred to as a composed error model, is stochastic in a sense that it captures exogenous shocks beyond the control of firms. The model is as follow: Y1 ' f(X1;B) + 81 (3.233) where '1 is the maximum amount of output obtainable from X1, a vector of non-stochastic productive inputs of the ith firm, and B is a vector of unknown parameters to be estimated. In addition e1 ' VI ' “1 (3.23b) where 'i’ the error component representing random noise, is assumed to be distributed normally with zero mean and variance of‘ovz, while "i’ the non-negative error component representing technical inefficiency, is assumed to be distributed either with a "half normal" density or with an exponential density, both with mode at u-O. 50 The major weakness of the model is its difficulty in estimating individual inefficiency although average inefficiency for the population and its variance are available as E[u] - u - au (2/11)°'5 (3.24.) m) - of (1 - 2m (3.240) Jondrow, Lovell, Materov and Schmidt (1982) show that by assuming v, is normal and “i is positive half-normal, v1 and “i are independent, and that inefficiency is independent of the regressors, then an estimate of the individual inefficiency for each firm can be obtained, although not consistently. The estimate “i is based on the conditional mean of “1 given e1, that is: E(u1/e1) - 0* [f(e1 s/a*)/(l-F(ei s/o*)) ] - ei /o* (3.25) where 0* - (ou2 ovz)(ou2 + of)"1 2 2 s - ou / ov The symbols f and F represent the standard normal density (pdf) and cumulative distribution function (cdf), respectively. By replacing e1 with its estimate (éi) and 0*, s by their estimates, one can estimate “1' Using this method B391 and Huang (1983), as cited in Seale (1985), estimate individual efficiency for 193 farms in Tennessee. More recently, Battese and Coelli (1988) presented a generalization of this method for a given 51 panel data of the Australian dairy farms, which will be discussed later. 3.2.5. Frontier Systems The earlier applications of the frontier model have been used to estimate production frontiers primarily to measure technical inefficiency. In addition, one may be interested in measuring allocative inefficiency, that is, whether a firm is producing off its least cost expansion path. The stochastic production frontier model has been extended by Schmidt and Lovell (1979) to allow measurement of average technical inefficiency for the population and allocative inefficiency for individual firm. The system of equations to be estimated are the production function as in Aigner et al.(1977) and factor share (proportion) equations derived from the first-order conditions for cost minimization. The behavioral assumption they used is cost minimization given aldesired level of output, i.e., treating output as exogenous and inputs as endogenous. Let us briefly review the model of Schmidt and Lovell (1979) by first rewriting the Cobb-Douglas production function of Aigner et al. (1977) as In y . a + 2 a1 ln xi + v - u (3.26) 0 where the condition u z 0 permits production to occur beneath the stochastic production frontier. They assume that the first order conditions for cost minimization are not satisfied, by writing ln(x1 /xn ) - ln(ai wn /an w1 ) + e1 . (3.27) where e, is symmetrically distributed, multivariate normal with zero mean. 52 This condition permits production to occur off the least cost expansion path. The combination of technical and allocative inefficiency yields a stochastic cost frontier of the form ln(w’x) - bo + I/r lny + 21 b1 lnwi - 1/r(v-u) + E (3.28) where bisai/r and r-Ziai. Actual cost exceeds the stochastic cost frontier for two reasons (1) due to technical inefficiency by an amount (l/r)u 2 0, and (ii) due to allocative inefficiency by an amount E z 0, where E is well-specified function of the 81. The parameters of the system equation (3.26) and (3.27) were estimated from cross-sectional data for a sample of US steam-electric plants by using MLE under three different assumptions. Firstly, firms were assumed to be allocatively efficient, but technically inefficient. Secondly, firms were assumed to be both allocatively and technically inefficient but without any systematic tendency to over or under utilize any input. Thirdly, firms may be both technically and allocatively inefficient and may systematically tend to over or under utilize any input relative to any other input. In this model they were able to compute the mean technical inefficiency over the sample [E(u)], the individual allocative inefficiency (81), the cost of technical inefficiency [(1/r)E(u)], and the cost of allocative inefficiency (E). Schmidt and Lovell (1980) extended this model in two directions. In the first place, they replaced the assumption that ei has mean zero by the assumption that it has mean M, to permit a test of the hypothesis that allocative inefficiency is symmetric rather than random. In the second place, they relaxed the assumption that technical and allocative 53 inefficiency are independent by permitting correlation between "i and e1. This permits a test of the hypothesis that firms that are relatively technically efficient are also relatively allocatively efficient. It should be pointed out, however, that the model works with a fairly restrictive functional form (Cobb-Douglas), and it requires data on both input prices and input quantities. A system consisting of a deterministic translog cost frontier, a type of flexible functional form, and the associated factor share equations has been estimated by Greene (1980). The major weakness of the model is the difficulty in providing an explicit solution for the associated production function, or vise versa, which makes it impossible to see the relationship between inefficiency in one function relative to the other function. It is well-known that either the production, cost or profit function uniquely define the technology. Which one is to be used depends on the behavioral assumption of the firms and/or data availability. The behavioral assumption underlying direct estimation of the cost function is cost minimization with output exogenous, which is usually related to the case of the regulated firm. This approach requires data on output quantity and input prices but not input quantities. The direct estimation of the profit function, which requires data on input and output prices, assumes that firms are profit maximizers. Cost and profit frontiers, like production frontiers, can be either deterministic or stochastic. The cost frontier yields information on the extra cost, while the profit frontier provides information on the forgone profit due to technical and allocative inefficiency. Separation of these two type of inefficiency is not possible without additional assumptions. 54 3.3. Dual Approach: Profit, Demand and Supply Function One objective of is this study to derive the supply and input demand functions of the rice farms. Product supply and factor demand equations consistent with a firm’s optimizing behavior can be obtained by two different but equivalent approaches. One approach, referred to as the primal approach, is to explicitly solve an optimization problem. Another approach, called the dual approach, allows one to obtain product supply and factor demand equations by partial differentiation of the objective function, either profit maximization or cost minimization. Applications 'of the dual approach begin with specification of a functional form for the profit or cost function, as opposed to the applications of the primal approach which begin with a production function specification. Although duality theory does not offer any particular profound insights into production economic theory, it offers a.more convenient way to obtain supply and demand equations than the primal approach. The dual approach is also useful in generating a more flexible functional specification for a consistent set of supply and demand equations for econometric estimation. Two principal advantages of dual approach are»given by Diewert (1974). First, it enables us to derive systems of demand equations which are consistent with maximizing or minimizing behavior of an economic agent, simply by differentiating a profit (cost) function, as opposed to explicitly solving a constrained maximization or minimization problem, which is sometimes difficult and explicitly unsolvable in the case of more flexible functional forms. Most qualitative results of production theory follow from properties of the dual functions, without restrictive assumptions on the divisibility of commodities and convexity and smoothness of production possibilities. The second principal advantage is 55 that it enables us to easily derive the “comparative static” theorems originally deduced from maximizing behavior. Another obvious advantage is that profit, supply and the derived demand functions are explicitly written as functions of variables that are normally considered to be determined independently of a firm’s behavior. Therefore, by estimating these functions directly, the problem of simultaneous equation bias can be avoided. Yotopoulos and Lau (1973; 1979) and Lau and Yotopoulos (1974) have explored the advantages of the dual approach by developing the so-called normalized restricted profit function using a Cobb-Douglas functional form, and implementing the model several empirical studies (Lau, Lin and Yotopoulos, 1979; Leo, Myers, and Chou, 1979). Sidhu (1974) used the Lau and Yotopoulos model for assessing relative economic efficiency in wheat production in the Indian Punjab. This model was also used by Sugianto (1982) and Kasryno (1985) to evaluate the relative economic efficiency of the irrigated rice farms in West Java, Indonesia. Other empirical studies using the Lau-Yotopoulos model can be found in Somel (1979), Kuroda (1979), and Tamin (1979). The duality between the profit and the production function, following Yotopoulos and Lau (1979), can be briefly outlined below. 3.3.1. The Case of Cobb-Douglas Production Function Consider a Cobb-Douglas type of production function of the ith firm as ak bh where Xik’s are the quantities of the variable inputs (k-1,...K) and the Zih’s are the quantities of the fixed inputs (h-1,...H), 3k is production 56 elasticity of the kth variable input, bh is production elasticity of the hth fixed input, while aoi represent the level of technical efficiency of the 1th firm. It is assumed that the production function is in the stage of decreasing returns in variable inputs, in order to guarantee that maximum profit exists, that is r - 3k 3k < 1. Let Wk denotes the price of k”I variable input, and P denotes the price of output. The normalized restricted profit function, that is profit over variable cost normalized by price of output, corresponding to the above production function is '1 ' Y1 ' 2k elk Xik (3.30) where ck - Wk/P is the normalized price of the kth variable input. Hoch (1962) and Yotopoulos and Lou (1979) incorporated the possibility of an allocatively inefficient firm (one that is unable to perfectly maximize profit) by introducing a non-unit factor of proportionality (dik) in the first order condition for profit maximization. The profit maximizing condition in this circumstance can be stated as Xik - ak Y1 /clk dlk (3.31) where dik - 1 for perfect profit maximization, dik is less than or greater than one when profit maximization is not perfect. The output supply function (3.32) can be derived by substituting equation (3.31) into equation (3.30) as . Y1 . 3011/11") ( nk dik‘ak/(l‘r) H nk akak/il‘r) ) (3.32)~ ( nk cik'ak/(l'r) ) ( uh zjhbh/(l-r) ) 57 We can then derive the actual normalized profit function (3.33) for a 1“ firm simply by substituting back equation (3.31) and (3.32) into profit equation (3.30), as follows: 1, - nail/(1") (1- 2k ak/dik ) ( nk dik'3k/(l'rl ) (3.33) ( nk akak/(l‘r) ) ( nk cik‘ik/(l'r) ) ( uh zihbh/(l-r) ) Simplifying equation (3.33) and then taking natural logarithms we get ln '1 ' In 001 + 2k Ck In cIk + 2h fih In Zlh (3.34) where: ln ooi - (1-r)‘l ln a0i + ln(1 - 2k ak/dik) - 31 ak(1-r)‘l ln d,k + 2k .iku-r)‘l ln ak “k - -ak(1-r)'l 3h ' bh(1-r)'1 so that the actual normalized restricted profit function of two or more farms differ only by a multiplicative constant aoi' The 1th farm is said to be relatively more efficient than the jth farm if 11 is greater than '1' In the present Cobb-Douglas case, this implies “oi>°oj’ which is also sufficient for the ith farm to be globally (for all normalized prices and fixed inputs) relatively more efficient than jth farm. 1th Similarly by direct computation the demand function for variable input is given by 58 X.” . Ail/(hr) (3] / d” C”) ( 11k dik'PJ/(l’r) ) ( nu akak/(l-r) ) ( Bk cik'ak/(l-r) ) ( 11h 211W ‘1") ) (3.35) where k, l - 1,2 ..... K. From equations (3.33) and (3.35), we can derive the ratio of lth variable cost relative to profit, hereafter referred to as factor share function, as so that the ratio of expenditure on the lth input to the actual normalized profit function is a constant. Moreover, this constant depends only on the vector of price ,efficiency parameter dik and the elasticities of production of the variable inputs. Equation (3.36) can be rewritten as -c1] X11 /‘i - -(aj /dil )/(l - 2k ak /dik ) - 011* (3.37) Equations (3.34) and (3.37) are the system equations from which Lau and Yotopoulos estimate the parameters. Several hypotheses related to relative efficiency and returns to scale can be constructed from these equations. First, two farms are equally price (in)efficient (di'dj) if and only if “ik* - “jk*' Second, a farm is price efficient (di-l) if and only if a1k* - ok*. Third, two farms are of equal relative economic efficiency if and only if lno°1*-lnaoj. Fourth, two farms are equal in technical efficiency and price efficiency if and only if lnoo1-lnaoj and “ik* - * ajk . 59 The next steps are to estimate the actual derived demand functions for variable inputs and the output supply function. Note that the actual profit function equation (3.34) can be rewritten as III“ ' 17300.. 'i‘ 2k Ck ank ' (2k Ck) I"? '4’ III flh anh (3.38) Substituting equation (3.38) into equation (3.37) and then solving for lnXiI, that is demand for lth input, we get lnXiI - [ln (‘°1l*) + 1““011 + (“I - 1)an1] + 2k.] “k 1""ik + (1 - r*) lnP1 + 2h 5h lnlih (3.39) where r* - 2k ck, and -ai] is the coefficient of the lth factor share function of the ith farm as in equation (3.37). By similar substitution process we can derive the output supply function with the final form as follows lnYi - [ln(1 - 2k “k/dik) + lnaoi] + 2k “k 1""ik - r* lnP1 + 2h 5h lnlih (3.40) The own- and cross-price elasticities of output supply and variable input demand can be obtained from equation (3.39) and (3.40) respectively. In addition, the elasticities of'output supply and variable input demand with respect to the quantity of fixed input can also be obtained from these two equations. 60 3.3.2. The Case of Translog Profit Function The development of the translog profit function as a more flexible functional form permits applications of duality theory for more disaggregated analysis of the production structure than is possible with the Cobb-Douglas functional form (Christensen, Jorgensen and Lau, 1976; Sidhu and Baanante, 1981). The main advantage of this functional form is that it allows the exogenous variables to produce different impacts across input demand functions. In the case of the Cobb-Douglas function, due to the well-known property of constant unitary elasticity of substitution among all input pairs, the impact across variable input demand functions of a given change in any of the exogenous variables is symmetric. Another disadvantage of the Cobb-Douglas function is that it fixes a priori the own-price elasticity of input demand to values greater than one, which is very unrealistic. The generalization of the normalized restricted translog profit function for a single output is given by : ln a1 - ao‘ + 2k “k lncik + 1/2 2k 21 Tkl lncik lnciI + 1/2 2h 29 Ihg lnlih 1"zig (3.41) where 'kl - ’lk (symmetry) for all k, l, and the function is homogenous of degree one in prices of all variable inputs and output. Differen- tiating the above profit function with respect to the normalized price of variable inputs gives a system of equations of variable input-profit ratios (factor share) or a system of variable input demand equations as 61 Sik ' ' (cik xik) / ‘1 ' Ck + 2] 7k] Inci] + 2h 8k" anih (3.42) Ideally, profits and variable inputs are determined simultaneously from equation (3.41) and (3.42). Several statistical tests may be done. The first statistical test is for the validity of the symmetry and parametric constraints across profit and SJ equations, or we may simply impose a symmetry restriction. The second statistical test is performed to test for the Cobb-Douglas hypothesis. For the profit function to be Cobb- Douglas, coefficients of all second order terms of the translog profit function should be zero. If the hypothesis is rejected, the translog representation is more suitable than Cobb-Douglas representation, or vice versa. The demand equation which is derived from equation (3.42) can be written as Xk ' I/"k [' dint/aank ] (3.43) or in the logarithmic form becomes lnxk - lna - ank + ln[- alnz/aank ] (3.44) where a is a symbol of partial derivative. Given equation (3.44) we can derive (i) the own-price elasticity of input demand for kt" variable input (Ekk), (ii) the cross-price elasticity kth of demand for variable input with respect to the price of lth input (Ekl)’ (iii) the elasticity of demand for the kth variable input with 62 respect to output price (Eky)’ and (iv) the elasticity of demand for the kth variable input with respect to fixed input Zh (Ekzh)' The derivation of these elasticities can be found in Sidhu and Baanante (1981). The following is the final result: Ekk - '51. - 1 - rkk /§,k (3.45) E11 ' ‘51 ‘ 'kl /5k (3°45) sky - 2k 5,, + 1 + 2, (7k, /§k) (3.47) Ekzh ‘ 3k 51.11 Wk t 5h ' ‘kh /5k (“3) where 5k is a simple average of 5k and 5k is the simple average price of the kth variable input. The supply function can be derived by substituting equation (3.43) into equation (3.30) and solving for Y as Y - a [ 1 - 2k alnx/aank] (3.49) or in logarithmic form becomes lnY - lnx + ln[l - 2k alnx/aank] (3.50) The own-price elasticity of supply (Eyy), the elasticities of supply with respect to price of kth variable input (Eyk)’ and the elasticity of output supply with respect to fixed input 2h (E are respectively yzh) 63 computed from: 3.4. Stochastic Profit Frontier and Profit Efficiency The restricted profit function of Yotopoulos and Lea (1979) implies that the underlying production function is an I'average" function which is not consistent with the neoclassical notion of a firm-specific production function. Furthermore, the Yotopoulos and Lau’s profit function approach can be used to analyze productivity differences between group of firms only in relative terms, assuming equal efficiencies within each group. Individual specific efficiency, however, cannot be quantified. Kalirajan (1985), following the frontier production model of Aigner, Lovell and Schmidt (1977), developed a stochastic restricted profit frontier and implemented the model on micro-level data for Indian farm production. Thelderivation of the stochastic profit frontier is straightforward and follows directly the steps of the derivation of the non-frontier profit function outlined above. The only difference is that the error terms of the production function as formulated by Aigner et al. (1977) are taken into account in the derivation process to keep track the primal-dual error relationship. Once again, for convenience and for theoretical purpose, we use a Cobb-Douglas type of function since it is not possible to assess 64 dual-primal relationship by using more flexible functional, such as translog. Following Kalirajan (1985), consider the Cobb-Douglas production function as formulated by Aigner, Lovell and Schmidt (1977) v, - ac uk x“,ak nh z“,bh e‘“l+V‘ (3.54) where Y1, X‘k and 11h have been defined in equation (3.29), while the error components u and v are defined in equation (3.24). Recall that v1 captures random variation in input due to factors beyond the control of the ith firm, such as weather and luck, and is assumed to be distributed as N(0,ov2), while "i is non-negative disturbance capturing randomness under the control of the ith firm, that is technical inefficiency. Let assume that firm is technically inefficient but allocatively efficient. Recall the normalized restricted profit equation (3.30). For simplicity let us omit the i subscript. The profit maximizing condition, assuming firm is maximizing anticipated (expected) profit as in Hoch (1962), is Xk ' (ak/Ck )Y (3.55) Substituting (3.55) into equation (3.54) yields: v - aol/(I‘I’ nk akaK/(l‘rl nk ck'35/(1'V) (3.55) uh zhbh/(l-r) e(u+v)/(l-r) where r - 2k ak < 1. Substituting equation (3.55) and (3.56) into equation 65 (3.30) yields the normalized restricted stochastic profit frontier as: I - eel/(1") (l-r) nk akak/(l'r) nk ck-ak/(l-r) uh Zhbh/(l-r) e(u+v)/(1-r) (3.57) Note that the profit equation (3.57) is bounded by the stochastic profit frontier as: a - aol/(l'r) (l-r) Bk akak/(l-r) Bk ck-ak/(I-r) (3.58) nh zhbh/(l'r) ell/(l’r) This represents the maximum possible profit for a given normalized price and amount of fixed input. The profit frontier is stochastic, as is the production frontier, because of the randomness of the v which reflects any shocks beyond the control of the firm. The term u/(l-r) represents the percent by which actual profit is less than the profit frontier. In other words it measures the profit forgone of producing below the production frontier due to technical inefficiency. Now we allow for the possibility that the firm may also be allocatively inefficient. Allocative inefficiency is modelled by permitting the profit maximizing (cost minimizing) condition to fail to hold exactly. Let us assume further that a firm makes systematic errors in seeking to equate the marginal value with marginal factor cost for any particular input. These assumptions yield the first-order profit maximizing conditions as follows: xk - (ak Y / ck) e“" (3.59) 66 where w - (wl, ..... wK) has a multivariate normal distribution with mean u and variance-covariance matrix 2. Substituting equation (3.59) into the production frontier equation and simplifying, gives: Y - aol/(l'r) nk akak/(l-r) nk ck-ak/(l-r) (3.60) 11h zhbh/(l-r) e(u+v)/(l—r) e-Zk (ak wk)/(1-r) Substituting equation (3.59) and (3.60) into equation (3.30), gives ' - aol/(l'r) (l _ 2k ak e'wk ) nk akaK/(l'r) nk ck-ak/(l-r) “h zhbh/(l-r) e(n+vl/(1-r) e-Zk (ak wk)/(1-r) (3.51) This can be simplified as: f ' 00 UK Ckak Hh thh eu*+v*+s* (3.62) where: “o _ a01/(1-r) (1 _ 2k ak e-wk ) nk akak/(l-r) “k - - ak / (1-r) 3h ' bh / (1") u* - u/(I-r) v* - v/(l-r) - 2k (ak wk )/(1-r) 5* Taking logarithms on both sides of equation (3.62), yields the following stochastic profit frontier: 67 Int - lnao + 2k “k lnck + 2h 5h lnzh + u* + 5* + v* (3.63) Kalirajan (1985) estimated equation (3.63) by using the method of maximum likelihood, assuming the density functions for u*, v*, and 5* as specified above. I If we are not interested in separate estimates of u* and 5*, we can simply combine them together and define this combined error component as an economic or profit inefficiency measure. The model can then be viewed as a generalization of stochastic production function previously described. This approach was used by Ali and Flinn (1989) for measuring farm-level profit efficiency among the Basmati rice producers (Pakistan Punjab) using cross-sectional data. 3.5. Panel Data and Stochastic Frontiers The estimation of individual technical inefficiency from a set of panel data was done by Hoch (1955; 1962). He, however, used an "average" (non- frontier) function rather than frontier function. Hoch assumed that firms maximize anticipated profit, and then estimated the production parameters using covariance analysis. There are great potential advantages for modifying existing frontier models to allow the use of panel data. Schmidt and Sickles (1984) pointed out three difficulties in applying stochastic production frontier models reviewed so far’ using cross section data. First, one can estimate technical inefficiency of each firm but not consistently. Second, separation of inefficiency measures from statistical noise depends on specific assumptions about the distribution of technical inefficiency. 68 Third, the assumption that inefficiency is independent of regressors is not valid if a firm knows its level of technical inefficiency. These difficulties will analogously be found in using cross-section stochastic profit frontiers. With the availability of panel data these problems can be avoided. First, if there are T observations on each firm, then the technical inefficiency of'a particular firm can be estimated consistently as T tends to infinity. Second, any distribution of technical inefficiency need not be assumed if these are treated as firm-speCific effects. Third, no assumption is needed regarding the independence of technical inefficiency and the regressors. This section is drawn heavily from Schmidt and Sickles (1984) article, with some modification in notation. Consider a production function as: ’it ' ao + xit a + Vit ' “i (3.64) Here, i - 1,2...N indexes firms and t - 1,2...T indexes time period. The value yit is output of the ith farm at time t, whereas xit is a vector of K inputs. The Vit are assumed uncorrelated with regressors and distributed iid N(0,ov2). The u, represent technical inefficiency and "i z 0 for all 2 u independent of the Vit' A particular distribution for "i may or may not i. It is also assumed that "i is iid with mean U and variance 0 and is be assumed. Furthermore, the “i may or'may not be assumed to be correlated with regressors. For T-I (a single cross section) the model is the stochastic frontier of Aigner, Lovell and Schmidt (1977). For T > 1, it is a straightforward generalization of that model, and it fits the usual framework in the 69 panel-data literature with individual effects but no time effects, as previously'discussed in section 2.1. The only difference from the standard panel data model is that individual effects are one-sided. The equation (3.64) can be rewritten in two ways. First, let E[u1] - g > 0, and define ‘ a°* - ao - n and u1* . "i - u so that u1* are iid with mean 0. Equation (3.64) then can be rewritten as Yit - a * + Xit’a + Vit - "1* (3.65) 0 with the error terms ”it and u1* have zero mean. Most of the results of panel data literature can be applied directly, except those that hinge on normality. Secondly, define - e- ,e a - ao g - ao u1 oi and then rewrite the model into Yit - aoi + Xit’a + Vit (3.66) This is exactly a variable intercept model as equation (3.2) 70 DWIII_IAI1Ahl9_£511!§19£.IE1391.£II2§11 As mentioned in section 3.1.1, this estimator treats the "i as fixed, that is, it estimates a separate intercept for every individual firm as in (3.66). This can be done by suppressing the constant term and adding a dummy variable for each of the N farms or, equivalently, by keeping the constant term and adding (N-I) dummies. Another equivalent procedure is to apply the within transformation, that is, to apply OLS after transforming the data in terms of deviations from the farm means. The advantage of the within estimator is that its consistency does not hinge on uncorrelatedness of the regressors and the effects. It also does not depend on the distribution of the effects, since in treating them as fixed it simply proceeds conditionally from whatever their realizations are. The estimates of a is consistent as either N or T tends to infinity. Consistency of the individual estimated intercept requires T e o. A considerable disadvantage of the within estimator is that it is impossible to include in the specification the time invariant regressors even though they vary across farms. In this case the estimated individual effects will include the effects of all variables that are fixed within the sample at the farm level, possibly including some that are not in any sense a representation of inefficiency (Schmidt and Sickles, 1984). In the case of the frontier function, if N is large, we can use the fact that "i z 0 to appropriately normalize the individual effects (“i) and the overall constant (a0). If' N estimated intercepts are 501, a 2""°30N’ simply define O 30 - max (501) (3.673) 01 - so - 301 (3.67b) 71 This definition amounts to counting the most efficient firm in the sample as 100%.efficient. The estimates 3° and 01 are consistent as N and T go to infinity. W With 0V2 and ouz known, the GLS estimator of a0* and a of equation (3.65) is consistent as either N or T approaches infinity. It is more efficient than the within estimator in the case of N e o and T fixed, but 2 2 v a not known, their consistent estimates need to be estimated. Consistent 2 this difference in efficiencies disappears as T + a. When a and au re estimation of ou requires N 4 n. Thus the strongest case for GLS is when N is large and T is small. If the opposite is true the GLS is useless, unless on2 were known a priori. Given estimates of a, we can recover estimates of the individual firm intercepts (501) from the residuals, that is, the mean (over time) of the residuals of each individual firms. so, - 1/1 2t 5” (3.68) These estimates are consistent as T -> 0, provided that estimates of a are consistent. Note that 301 can be decomposed into 30 and "i’ for which consistency requires N - o and consistency of the 301. Another way to estimate the individual effects (inefficiency) is by using Battese and Coelli (1986) method, which is actually a generalization of the method suggested by Jondrow et.al (1982) as described in equation (3.25). The Battese and Coelli method is presented in a slightly different notation 72 as follows: [1- For - ,/a*) 1 01 - 1 - exp(-m1 + o*/2) (3.69) 1 1 - F1-n,/a*) 1 where 0* - on2 ovz (0‘,2 + Touzf1 ”i - -(ou2 9i)(°u2 + oVZ/T)’l 91"‘ai"n Note that u and 801 have been described in equation (3.24a) and (3.68), respectively, while F is a symbol for standard normal cumulative distribution function (cdf). The important advantage of GLS estimator relative to within estimator in the present context is not efficiency, but rather the ability to include the time invariant regressors. In cases where time-invariant regressors are relevant, this is important so that their effects do not contaminate measured efficiency. The problem with Fixed Effect (FE) model is that if there are any time invariant variables that are excluded, the firm dummies will reflect this influence. This would make inefficiency comparisons difficult unless the excluded time-invariant variables affect all firms equally. Since this is not always the case, inefficiency measures relative to the best firm (which has to be assumed 100%iefficient in FE model) might give misleading results. Thus, even though FE models have the advantage of allowing correlation among inefficiency and the regressors and no distributional assumption on inefficiency is required, the results should be interpreted 73 carefully. Khumbakar (1986), in his study of the 0.5. Class 1 railroads, found that the estimates of inefficiency in the FE models are much bigger than those of random effect (RE) models. 0n the other hand, all the production function parameters in the FE model are much lower than those in the RE model. The choice between these two has nothing to do with the frontier model as such. The only problem with the FE framework in the context of a production or cost (profit) frontier is that the firm-specific effects pick up the effect of variables that differ across firms but are invariant over time. These effects are not in any sense a representation of inefficiency. This might be one of the reasons why estimated inefficiencies in the FE models are much greater than in the RE models. Given the potential differences in results, both in magnitude of inefficiency and efficiency ranking of the firms, one may think to conduct a specification test, to examine whether the random effect models is correct, i.e., the Hausman (1978) or the Mundlak (1978) test, as in equation (3.20) and (3.21) respectively. Since inputs might be correlated with technical inefficiency, there is a strong justification for performing this test when dealing with the estimation of the production function by single equation methods. But for a profit (cost) function, inefficiencies can be assumed with greater confidence to be uncorrelated with input prices. 3.6. Adoption of New Technology: A Multinomial Logit Model. Another objective of this study is to determine factors affecting farmer decisions in choosing rice variety. This objective is motivated by the fact that there were many farmers in the sample still growing TV, 74 despite the fact that HYVs have long been introduced by the government. Applications of probability models to evaluate varietal adoption in the West Java rice farming have been done by Hutabarat (1986) using a probit model and Gunawan (1988) using a logit model. This study instead will use a multinomial logit model. An example of an emirical study using a multinomial logit model is found in Schmidt and Strauss (1975), which analyzed patterns of employment in the US labor market using race, sex, educational attainment and labor market experience as explanatory variables. The derivation of the multinomial logit model can be found in, among others, Judge at al. (1985) and Fomby et al. (1984). The multinomial logit model is based on the assumption that the error terms are independently and identically distributed with Wibull density functions. Let there be N possible choices, with probabilities P1, P1,...PN. The multinomial logit model, omitting individual (observation) subscript, can then be written as where j-2,....N, X is a IxK vector of explanatory variables, and ”j is a le vector of unknown parameters. The N-1 equation in (3.70) plus the requirement that the probabilities for each individual sum to one, determine the probabilities uniquely. The probability of each choice is explicitly as follows: 75 P1 . (3.71) 1 + SJ exp(XflJ) exptxfii ) p1 . (3.72) 1 + 23 exp(XBj) These probabilities are comonly evaluated at the mean value of the explanatory variables. Given equation (3.71) and (3.72), one can derive response elasticities for any change in explanatory variables (appendix 3.1). This measure represents a percentage change in probability of particular event as a result of a one percent change in the value of any explanatory variable. This elasticity is evaluated at the mean value of explanatory variables. 3.7. Policy Evaluation: Rice Price Support and Fertilizer Subsidy In general the evaluation of policy alternatives can be done by assessing their impact on the welfare of the producers and consumers and the government budget, using the world prices as a standard of reference to judge the opportunity costs of particular policies. Timer (1986) argued that understanding the logic of the border prices (also referred to as parity prices) is the essential first step in addressing any further policy concerns about the implementation and impact of changes in domestic prices for food and agricultural commodities. Evaluating domestic price policy relative to border prices starts with the supply and demand functions of the domestic market for a commodity. 76 The supply function reflects the initial stage of technology and productivity of the commodity being analyzed, including the cost of inputs and alternative commodities that farmers might grow. The demand function represents the tastes of consumers, the level and distribution of incomes, and the prices of alternative commodities. These supply and demand curves reflect only part of the reaction to a change in the price of the commodity in question, which commonly is referred to as ”partial equilibrium” models. When the coumiodity is important to the entire economy, a staple food such as rice, the partial equilibrium models provide only an initial glimpse of the full adjustments likely to occur when the price changes (Timmer, 1986; Timmer, 1987). More complicated analytical techniques are necessary to understand the full effect of the price policy. The simple supply and demand model may be used to eXplain basic price policy issues as well as to provide a base from which to identify the necessity for more complicated analysis. A price change for the main staple food, or significant price changes simultaneously for multiple commodities, are likely to cause multimarket spillover effects. Timmer (1986:1987) measures the ”importance" of the conmodity in one of the following four ways. First, the commodity is the chief wage good in the society and forms a significant share (20-50%) of the average consumer’s budget. Second, the commodity is a major source of farm income. Changing the price alters farm incomes and hence farm expenditures on goods and services. Price changes also cause farmers to change input use and alter their cropping patterns and therefore affect national agricultural production. Third, the commodity is important in the country’s international trade as an export or an import. Either way, changes in 77 domestic prices are likely to alter trade volumes and hence the foreign exchange balance. Finally, the commodity is important to the government budget, either as a generator’of revenue or as a significant drain because of large subsidies. The standard partial equilibrium analysis using border prices as the measure of domestic opportunity cost can be used to assess the impact of price changes by measuring gains and losses in producer and consumer welfare, transfer to and from the government budget, changes in foreign exchange earnings and expenditures, and efficiency losses (deadweight losses). The partial equilibrium analysis assumes that there are no adjustments in the economy other than those being analyzed. ’The most difficult situation of agricultural price policy analysis occurs when the commodity is important for all four reasons. This is a common situation in developing countries in which a primary food staple is grown domestically and imported or exported. Rice in Indonesia easily meets all these criteria. Timmer (1987) pointed out that analysis of such comdity extends beyond the boundaries of partial equilibrium models even if the political dimensions remain outside the formal analysis itself. All markets are linked together by substitution possibilities in production and consumption. Timmer (1986:1987) argued that one approach to determine whether general equilibrium consequences are significant is to work from the simple to the complex, starting with a single-market analysis. When this first step is carried out carefully and thoughtfully, it provides clear directions about where to look for the most immediate impact on other'markets. A policy intervention that lowers the rice price, for instance, will lead to a reduction in rice production that, in turn, will be accompanied by diminished used of inputs including labor and by 78 some increased effort on other crops or activities. Immediately, at least three other markets, input market such as fertilizer, alternative output markets and rural labor market, are identified as having potentially significant links to the rice market which would adjust to a change in rice price. Another approach is comonly referred to as a sector-wide model. Because price policy changes can lead to a complex combination of changes in incomes of producers and to shifts in use of regional or national resources, such as land, water and labor, a sector-wide model is required to trace the most important effects. Agricultural sector models have been used since the mid-19605 to check the consistency of agricultural development plans and to examine resource requirements of alternative cropping patterns (Timmer, 1987). Unlike single-market or multi-market equilibrium models, sector models contain detailed farm models as the basis for supply response and alternative crops in the face of changed prices. ’ World market prices, as a measure of domestic opportunity cost, should be used with caution due to the instability of world commodity markets. A domestic floor price announced in the beginning of planting season, for example, may well reflect the opportunity cost of imports at the time of the announcement but be too low or too high by the time of harvest. Instability in international prices is therefore a serious issue for the design of domestic price policy in the short run, whereas long-run structural changes in the world food system affect investment decisions about agricultural research, rural infrastructure, and sectoral priorities. According to Timmer et al. (1983), agricultural and food price policies 79 are commonly judged by their effects on the following four food policy objectives: (1) promoting economic efficiency and hence faster growth of income, (ii) distributing income more equally, (iii) guaranteeing adequate nutritional status for all people, and (iv) providing security of food supplies. Empirical analysis of a policy requires measurement of the size as well as the direction of its impact. In addition, the weights given to the objectives and the constraints, including international repercussions, determine the actual feasibility and efficacy of a price policy. In evaluating output price support and input subsidy policies in the Philippine rice economy, Barker and Hayami (1976) used a simple demand- supply model combined with a benefit-cost analysis. They started the analysis by evaluating the programs of rice price support and fertilizer subsidy for achieving rice self-sufficiency as the main goal. Benefits and costs associated with these alternative programs and their income distribution effects are estimated and compared with the case of no price support or fertilizer subsidy. They concluded that a subsidy applied to modern input, such as fertilizer, that are being used below optimum level can be more beneficial than supporting the price of rice. Unfortunately, the Barker and Hayami approach requires large amounts of data. Another empirical study in agricultural policy evaluation was done by Hayami, Bennagen and Barker (1977). Using rice economy in the Philippines as a case and benefit-cost analysis as the main tool of analysis, they evaluated and compared the irrigation investment and price incentive policies. They concluded that, in the long run, despite its large initial capital cost, irrigation investment imposes less financial burden on the government than the manipulation of product and input prices. In terms of the social benefit-cost ratio, the irrigation 80 development is clearly more efficient than rice price support, but it becomes inferior to fertilizer subsidy if high discount rate is applied to a large-scale and high-cost irrigation project. In this study, we attempt to evaluate rice price support and fertilizer subsidy policies in much simpler way. The main reason for not using benefit-cost analysis, as in Barker and Hayami (1976), is lack of data. We also do not intend to assess these policy options with those above four objectives specified by Timmer et al. (1983). We instead will focus on the impact of these policies on increasing total rice production as well as promoting rural employment. The evaluation is based on the estimated cross- and own-price elasticities of the rice supply and variable input demand, given the following information: (i) total quantity of subsidized fertilizer for rice production, (ii) total quantity~ of excess rice supply in the market purchased by the government, and (iii) total rice production, (iv) total labor used in the rice production, (v) the price of rough rice, and (vi) the price of fertilizer. The main idea of the proposed evaluation procedure, which will be described in the next section, is based on a comparison between rice price support and fertilizer subsidy policies on the the government’ costs in generating an additional kilogram of rough rice and promoting an additional unit of employment in the rural areas. The more desirable policy is the one which is less costly. Note, however, that this procedure only evaluates the impact of the policy on the government budget, and ignores theiwelfare impact on consumers and producers of rice. IV. MODEL SELECTION AND ESTIMATION The purpose of this chapter is to present the models, estimation methods and the»data transformation. This chapter consists of six sections and is organized as follows. Section one and two present the model selection and estimation methods of the production and profit function approaches, respectively. Section three describes the model selection and estimation method of the system of equations consisting of the profit function and variable factor share equations. The specification of the multinomial logit model for varietal choices, along with the estimation method, is outlined in section four; Section five describes the procedure for evaluating the rice price support and fertilizer subsidy policies. Finally, section six describes the data transformation with regard to the variables used in the analysis. 4.1. Production function Direct estimation of the production functionlgives consistent parameter estimates only when input can be treated as exogenous. The Hoch’s (1962) argument of anticipated profit maximization and the Zellner et al. (1966) argument of expected profit maximization can be used to treat inputs as exogenous. The main purpose of the direct estimation of the production function in this study, in addition to the estimation of production elasticities, is to measure individual technical inefficiency. One can use a non- frontier or an 'average' function to estimate technical inefficiency but only under’certain restrictive assumptions. The result of the inefficiency 81 82 measure, however, cannot be called a pure measure of technical inefficiency since it also includes random variability. The frontier production function is designed to bridge the gap by introducing the error term to represent an (in)efficiency measure. 4.1.1. Model Selection and Functional Forms The per farm (household) production function, using the Cobb-Douglas type and the more flexible transcendental logarithmic (translog) functional forms, will be estimated in this study. One of the objectives is to check whether or not the individual level of technical inefficiency is insensitive to the functional form used. 999W The total output per farm, measured in kilograms of rough rice, is the dependent variable, while the total quantity of seed, fertilizer, labor and farm size of the corresponding farm household, are the independent variables. By assuming that farmer is maximizing anticipated profit, all of the production inputs can be treated as exogenous variables, and therefore, the distinction of whether a particular input is variable or fixed is irrelevant. In logarithmic form, the per farm Cobb-Douglas production function to be estimated is specified as follows: ]"Yit - ln a0 + 2k ak lnxkit + aGDPit + a7DVlit + aBszit + agDSS + aloDSIZE + allDRl + alzDRZ + a130R3 + a14DR4 + alsDRS + Vit - "i (4.1) 83 where: i - 1,2, ..... 171 subscript for individual observations t - 1,2, ..... 6 subscript for time k - 1,2, ..... 5 subscript for production inputs v - the error component represents random noise, and is assumed to be distributed normally with zero mean and variance of ovz. u - the non-negative error component representing technical inefficiency. lnY - anGOUT: total production in the form of rough rice in kilogram. lnXl - anGS: the amount of seed (kg) lnX2 - anGN: the amount of urea (kg) lnX3 - anGP: the amount of TSP (kg) lnX4 - lnLAB: the amount of labor (hours) lnX5 - lnHA: area planted with rice (ha) DP: dummy variable of pesticide use, equals 1 if farmer uses pesticides and equals 0 otherwise DVI: dummy HYV variety, equals 1 if HYV, zero otherwise 0V2: dummy of Mixed Varieties (MV), equals 1 if mixed varieties are used, zero otherwise. Note: traditional variety (TV) is the control 055: dummy variable of season, equals 1 if wet season, zero otherwise DSIZE: dummy variable of farm size, equals 1 if farm size greater than 0.5 ha, zero otherwise 0R1: dummy village, equals 1 if desa Lanjan kabupaten Indramayu, zero otherwise. 0R2: dummy village, equals 1 if desa Gunung Wangi kabupaten Majalengka, zero otherwise 0R3: dummy village, equals 1 if desa Malausma kabupaten Majalengka, zero 84 otherwise 0R4: dummy village, equals 1 if desa Sukaambit kabupaten Sumedang, zero otherwise 0R5: dumy village, equals 1 if desa Ciwangi kabupaten Garut, zero otherwise. Note: Wargabinangun (kabupaten Cirebon) is the control village W The translog is a flexible functional form based on the second order Taylor series approximation of any function. The Cobb-Douglas functional form is a very special type of the translog functional form in the case that all the coefficients of the second order terms are zero. Note that the translog specification only applies to the production inputs, not to the categorical (dummy) variables. The specification of the per farm translog production function used in this analysis is as follows: ]"Yit - ln a0 + 2k ak lnxkit + 1/2 2k 2] ak] ]"xkit*]"xlit aGDP1t + a7DVIit.+ a8szit + agDSS + aloDSIZE allDRl + alzDRZ + 313DR3 + a14DR4 + alsDRS + Vit - "i (4.2) The variable definition is the same as that of Cobb-Douglas described above. The symmetry restriction is a 11.119111 imposed implying that the coefficient of (1/2) of the second order terms will occur only if k - l. 85 4.1.2. Estimation Methods Two estimators, i.e.,. the'within and EGLS estimators, are used in this study. In addition to these two estimators, the simple OLS estimator will also be presented for comparison purposes. As indicated in the previous chapter, it is possible and even more efficient to use MLE, since it can fully utilize the information of the distributional assumption of the error component, especially the one-sided error component representing inefficiency'measure. However, due to the complicated nature of the latter (which sometimes offsets its potential gain of efficiency) there was no attempt to use this estimator in this study. The within estimator is simply obtained by running OLS on the transformed data as described in equation (3.6) in chapter 3. To obtain the EGLS estimator, the steps described in chapter 3 section 3.1.2 are used. One way to decide whether to use fixed effects (FE) or random effects (RE) model is to test the null hypothesis that there is no correlation between the individual effects and the included explanatory variables against the alternative hypothesis that such correlation exists. For this purpose, we can use either Hausman test (1978) or an asymptotically equivalent test suggested by Mundlak (1978) as discussed in chapter 3. If the null hypothesis holds we use the random-effects model: otherwise we use the fixed-effect model. The individual level of technical inefficiency is measured using Battese and Coelli (1986) method as described in equation 3.69 in chapter 3. The methods presented in section 3.5, specifically equation (3.67) for FE model and equation (3.68) for RE model, will also be employed for comparison purposes. 86 4.2. Profit Function As pointed out in the previous chapter, the dual approach offers an easier way to obtain output supply and input demand equations compared to the primal approach. The dual approach is also useful in generating a more flexible functional specification for a consistent set of supply and demand equations for econometric estimation. Unlike the case of production function presented previously, a distinction between fixed and variable inputs is relevant in this case, since it has an implication of whether or not the price of a particular input needs to be taken into account in the production decision process. In this analysis, we make the following specification. To produce rice, each farmer must first decide how many hectares to plant. The farmer can then determine the amount of seed, fertilizer and labor to use for his land. Thus, the planning horizon for the farmer is a short-run period covering only one season of rice production. This planning and production process is assumed to be repeated for each planting season. In this case we can therefore treat hectareage of land (farm size) as fixed input, while seed, fertilizers and labor as variable inputs. Farmers are assumed to maximize restricted profit, that is profit over variable costs, subject to a production function; thus, variable factor prices, quantities of fixed factors, and price of output are considered as exogenous, but not output. This is consistent with the Zellner, Kmenta and Dreze (1966) approach of maximizing expected profit, assuming that prices and quantities of fixed factors of production are exogenous variables (Kalirajan, 1985). . 87 4.2.1. Model Selection and Functional Forms The Cobb-Douglas and translog functional forms, as in the production function approach, will be used in ‘this analysis, to determine if individual level of profit inefficiency is insensitive to the functional form. In this analysis, the translog specification only applies to the prices and fixed inputs, not to the categorical variables. W In its notation, the specification of the Cobb-Douglas restricted profit function is the same as the production function. The difference is in the independent variables involved in the model. In the logarithmic form, the Cobb-Douglas restricted profit function to be estimated is ln‘it - In 30 + 2k ak lnCHt + a4 an + 350Pit + ‘GDVIit + a7DV21t + aBDSS + agDSIZE + aloDRI + allDR2 + alZDR3 + a13DR4 + al4DR5 + Vit - “i (4.3) where: k - 1,2,3 is subscript for variable input prices. v - the error component represents random noise, and is assumed to be distributed normally with zero mean and variance of ovz. u - the non-negative error component representing profit inefficiency. lna - ln LPRO: restricted profit, that is profit over variable cost, normalized by per kg price of rough rice lnCl - lnLPS: per kg price of seed normalized by per kg price of rice lnCz - lnLPF: per kg price of fertilizer normalized by per kg price of rice 88 lnC3 - lnLWG: per hour labor wage normalized by per kg price of rice an - LnHA: farm size, as a fixed factor, in hectare. The definition of the remaining variables is the same as before. W The specification of the translog profit function used in this analysis is as follows: ]"'it - ln a0 + 2k ak lanit + 1/2 2k 21 ak’] ]"ckit*]"clit + a4 an + 1/2 a4’4 an*an + 2k ak’4 lan1t*an + “Sopit + aBDVIit + a7DV2it + aaoss + agDSIZE + aloDRl + allDRZ + alzDR3 + a13DR4 + + a14DR5 + Vit - "i (4.4) Again, the symmetry restriction will be imposed a priori, that is ak,l - a1,k, for all k and l. This implies that the coefficient of (1/2) will occur only if I - k. The definition of variables is the same as previously described. 4.2.2. Estimation Methods As in the case of production function approach, two estimators, i.e., the within and EGLS estimators, are used in this study. In addition to these two estimators, the simple OLS estimator will also be presented for comparison purposes. The test to decide whether to use the within (fixed effects model) or the EGLS (random effects model) has been described in section 4.1.2. 89 The individual level of profit inefficiency is measured using the Battese and Coelli (1986) method as described in equation 3.25 in chapter 3. The methods presented in section 3.5, specifically equation (2.67) for FE model and equation (2.68) for RE model, will also be employed for comparison purposes. 4.3. System of Equations: Profit Function and Factor Share Equations A system of equations consisting of a profit function and variable factor share equations can be regarded as a typical seemingly unrelated regression (SUR). As mentioned in chapter 3 section 3.3, in the non- frontier framework, some parameters of the profit function can be estimated either from the profit function itself or from the factor share functions. Theoretically, if“ profit. maximizing conditions hold, the parameter estimates from the profit function have to be the same as the corresponding estimates from the factor share function. Thus, imposing an equality restriction will improve the efficiency of the estimates. 4.3.1. Model Selection and Functional Forms Cobb-Douglas and translog profit functions, and corresponding factor share equations, are employed in the analysis. The specification and notation of the profit function of each functional form is the same as described in the previous section. coo-IAa‘I ’. "o l '1 '1‘. 1 a 11 ‘. 1.. 01 The Cobb-Douglas profit function and its corresponding variable factor share equation are specified as follows: 90 1"‘it - ln a0 + 2k ak lanit + a4 lnlit + a5DPit + aGDVlit + a7DV21t + aBDSS + agDSIZE + aloDRI + allDRZ + alzDR3 + a13DR4 + 314DR5 + ”it (4.5) sikt ' ' cikt xikt /'it ' aikt t 'ikt (for k - 1, 2, 3) where 51k is the kth variable factor share for the ith farmer, that is the ratio of the normalized kth input value to the normalized restricted profit. The term "ikt represents random noise, which is assumed to be distributed normally' with. a constant variance. Definition of’ other variables in the model is the same as that previously described. Izans19s_Er9I1t_Eunst1on_and_Eastor_§bars_£9uations A system of equations consisting of the translog profit function and variable factor share equations are: 1"‘it - In 80 + 2k ak Inc“t + 1/2 2k 2] ak,1 lan1t*lnC11t + a4 lnlit + 1/2 34,4 an1t*anit + 2k ak’4 Inckit*‘"zit + aSDPit + aGDVIit + a7DV21t + a80$S + agDSIZE + aloDRl + a110R2 + alzDR3 + 313DR4 + al4DR5 + ”it (4.6) sikt ' ‘ (cikt xiktl / 1‘it - ak + 2] ak’1 Inc”t + ak’4 lnlit + 'ikt (for k - 1, 2, 3) The definition of the variables is the same as before. 91 4.3.2. Estimation Method The concept of FE and RE models can be generalized beyond a single equation described above. Given a set of panel data, we can estimate a system of equations consisting of profit function and factor share equations. Single equation error component procedures outlined in the previous section can not be used to obtain efficient parameter estimates of a system of equations unless error correlations between equations are assumed to be zero. The SUR type of model with error components can be estimated using the procedure outlined in the articles of Avery (1977) and in Baltagi (1980). The problem with this estimator, however, is that no ready computer packages are available at this time. Moreover, the transformation matrix required to use standard LS computer package is apparently very complicated and is not yet in the literature. Given the above difficulties, the system of equations (4.5) and (4.6) will not be estimated using. either the fixed effects or the error component framework. Instead they will be estimated using the Zellner’s (1962) SUR method, applied on the pooled cross-section and time series data. This amounts to estimating the»models in the non-frontier framework, and therefore, the individual specific effects (inefficiencies) will not be relevant in this case. A more detailed discussion of this approach, will be presented in the next chapter'dealing with the empirical findings. This approach is primarily intended to derive the profit maximizing (i.e., profit maximization as a maintained hypothesis) input demands and output supply, and related elasticity parameters. As mentioned in the previous chapter, some parameters of the profit function can be estimated 92 either from the profit function itself or from the factor share functions. Theoretically, if the profit maximizing condition holds, the parameter estimates from the profit function have to be the same as the corresponding estimates from the factor share function. Thus, imposing equality restriction (i.e., profit maximization is maintained) will improve the efficiency of the estimates. Note, however, that profit maximization can alternatively be tested following the method of Yotopoulos and Lau (1979). 4.4. Multinomial Logit Model 4.4.1. Model Selection Framers face three choices when selecting rice varieties: (i) to grow TV, (ii) to grow HYV, or (iii) to grow both TV and HYV, hereafter referred to as mixed varieties (MV). The multinomial logit model is used to evaluate economic factors affecting the probability of an individual farmer choosing one of those three alternatives. Five economic variables are hypothesized to affect farmers’ decisions regarding varietal choice, i.e., prices of TV, HYV and fertilizer, labor wage, and farm size. The model to be estimated is specified as follows (omitting individual subscripts): where: P1 : probability of farmer growing TV. PJ : where j - 2, 3, represent probability of farmer growing HYV and MV, respectively. 93 k - I, 2...5, subscript for explanatory variables X1 - PTV : price of TV (Rp/kg) X2 - PHYV : price of HYV (Rp/kg) X3 - PF : price of fertilizer (Rp/kg) X4 - WAGE : labor wage (Rp/hour) X5 - HA : land in hectare. Comparison for choice '3' and '2' can simply be made as "'(P3/Pz) " (33o ‘ 32°) + 2k (53" ‘ 52k)xk ”-3) Having estimated the coefficient of 825 and B35 and the corresponding covariance we can then estimate the (£30 - £20) and (53k - 82k). Fertilizer price, labor wage and farm size used in this model are similar to those used in the profit function model presented before. Prices of TV and HYV need some further explanation. The survey data provide either the TV or HYV price for ‘the individual respondent, depending on the variety used. For example, from the respondent who grew TV, only the TV price was obtained. Thus, a value of the HYV price has to be assigned to this respondent. For this purpose, the average HYV price for each season is used. More troublesome is the case of farmers who grew both TV and HYV, since only one price was recorded without indicating whether it was a TV price or HYV price or an average of these two. In this case, the average prices of HYV and TV for each season are used. Note that variable prices and labor wage used in this analysis are measured in nominal terms. The reason for using nominal rather than real (deflated) prices, is that farmers are commonly influenced by nominal 94 price levels in their decision making process. 4.4.2. Estimation Method The ML method will be used to estimate the model on the pooled data (N-1026). The GLS estimation, based on the availability of repeated observations, is not possible since many farmers did not switch from one choice to another, even though some did so. In addition, the time periods (T) are too short. Of course, it is also possible to estimate a seasonal (pure cross-sectional) multinomial logit model using ML method. A primary reason to pool the data is to obtain larger variation on the price (wage) variables. 4.5. Evaluation of Price Support and Fertilizer Subsidy Policies In this study, we attempt to evaluate rice price support and fertilizer subsidy policies, focusing on their impact on increasing total rice production as well as promoting rural employment. The evaluation is based on the estimated cross- and own-price elasticities of the rice supply and variable input demand, obtained from the analysis outlined above. In addition, we need the following information: (1) total quantity of subsidized fertilizer for rice production, (ii) total quantity of excess rice supply in the market purchased by the government, and (iii) total rice production, (iv) total labor used in the rice production, (v) the price of rough rice, and (vi) the price of fertilizer. The main idea of the proposed evaluation procedure is to compare the rice price support and fertilizer subsidy policies in terms of their costs to the government in producing an additional kilogram of rough rice and in promoting an additional unit of employment in the rural area. The 95 desirable policy is the one which is less costly. The procedure involves the following calculation: (1) We a. Government's additional cost: GPR . 0.01 * PR * OER (4.9) b. Effects on: b1. Rice Production: AQRP(1) ' ERR * 0.01 * QRP (4.10) b2. Rural Employment: AQL(1) - ELR * 0.01 * QLRP (4.11) (2) ' ' I a '0 ’ ‘ I‘ ‘. ' e a 1 a. Government’s additional cost: GPF - 0.01 * PF * QFSR (4.12) b. Effects on: b1. Rice Production: AQRP(2) ' ERF * 0.01 * QRP (4.13) b2. Rural Employment: AQL(2) ' ELF * 0.01 * QLRP (4.14) The definition of the variables is as follows: QRP - total rice production (ton/year) OER - total excess supply of rice purchased by the government (ton/year) 96 QFSR. - total quantity of subsidized fertilizer used in rice production (ton/year) QLRP - total labor used in rice production (mandays/year) GPR - governments’ additional costs due to specified price support policy (Rp/year) GPF - government's additional costs due to specified fertilizer subsidy policy (Rp/year) AQRP(1) - additional quantity of rice production due to specified price support policy (ton/year) A°RP(2) - additional rice production due to specified fertilizer subsidy policy (ton/year) A°L(1) - additional rural employment due to specified price support policy (mandays/year) A°L(2) - additional rural employment due to specified fertilizer subsidy policy (mandays/year) PR - price of rice (Rp/ton) PF - price of fertilizer (Rp/ton) ERR - estimated own-price elasticity of rice supply ERF - estimated cross-price elasticity of rice supply with respect to fertilizer price ELR - Estimated cross-price elasticity of labor demand with respect to price of rice ELF - estimated cross-price elasticity of labor demand with respect to price of fertilizer In the equation (4.9), based on empirical evidence discussed in the next chapter, we assume that no additional excess supply of rice must be 97 purchased by the government as a result of a one percent increase in the price of rice. Stated differently, the additional excess supply of rice which is purchased by the government is small, and can be ignored. The additional government cost is therefore equal to the product of the increase in the unit price and the total rice previously purchased, as in equation (4.9). A similar assumption in the case of fertilizer, equation (4.12), is also made. Next we need to calculate the government’s costs per additional unit of rice (Rp/kg) as well as the costs for every additional unit of employment opportunity (Rp/manday) generated by each policy, as follows: (1) W223: (b) Cost per unit of rice produced: c11(1) ' GPR“0119(1) (b) Cost per unit of employment generated: (2) -a - -- . '- ° . I- -. . I - ,il (a) Cost per unit of rice produced: CR(2) ' GPF/AQRP(2) (b) Cost per unit employment generate: CL(2) ' GPF/AQL(2) The final step of the proposed procedure is comparing CR(1) against CR(2) and CL(1) against CL(2). The smaller the value, the more desirable is the policy in question. 98 The critical points of this procedure will be on estimating the OER as well as the QFSR' The better the estimates, the more reliable is the conclusion made. This procedure only evaluates the impacts of the policy alternatives on the government budget, and ignores the welfare impact on consumers and producers of rice. Even though the procedure described here is very rough, it provides a simple measure for identifying theimore cost- effective policy alternative for attaining the specified policy goals. 4.6. Data Transformation Landing Farmers in the survey area often do not have contiguous plots of land to cultivate, but rather have several parcels scattered throughout the village area. Land or farm size, in this analysis, is defined as the total area that farmers cultivate with rice. The survey did not record input- output data for each parcel; instead it recorded the total inputs used and total output of the whole hectareage. Some farmers may plant other crops in some of their parcels, but these parcels, including the crops, are excluded from the analysis. Note that the terms of “land" and “farm size' are used interchangeably in this study. WW Two types of fertilizers are comonly used by farmers, namely N (Nitrogen) in the form of urea (48% N) and P (Phosphorus) in the form of TSP (Triple Super Phosphate, 46% P). The amount of fertilizer in this analysis, which is measured in kilograms (kg), represents the bulk of urea and TSP, respectively. These fertilizers are part of the BIMAS credit 99 package, but non-BIMAS farmers as well can easily find these fertilizers at the local market. As mentioned in chapter 1, the price of fertilizer is set by the government as part of input subsidization policies. The price of these two fertilizers is officially the same. Prices on local markets, however, are sometimes lower than the BIMAS price and the price of urea and TSP may differ from each other. In this analysis the price of fertilizer is the weighted average of the prices of urea and TSP paid by farmers, and is obtained by dividing the total costs of fertilizers (urea and TSP) by the total quantity used. MW Farmers may use seed they produce themselves or they may purchase seed from the village cooperative unit. The latter (certified seed) is usually available in the village unit, and has a higher germination rate than self-produced seeds. Therefore, farmers who use self-produced seed usually use more seed per unit of land. The price of seed per kilogram recorded in this survey is based on information obtained directly from the farmer. There is no information regarding the type of seed used by the farmer. This, however, can be inferred from the price and the quantity recorded in the survey. Low price and large quantity per unit of land may indicate use of self-produced seed. WE! The measurement of labor units and wages is more complex than for other inputs. There are two kinds of laborers in the rice production reported 100 in the survey: (i) human labor, and (ii) animal (bullock) labor. Although the use of tractors is not uncommon, there was no tractor use reported in the survey. The two categories of labor are subdivided into family labor and hired labor. Human labor consists of male and female, the wage and quality of which this study assumes is equal. It is true that male labor may be superior in one type of job while female is superior for another. Male labor usually is applied to land preparation, while female labor is applied to planting and crop management. The wage for hired labor, except for harvesting, is usually paid in a combination of cash and food. The actual wage per unit of time varies among farmers due to the valuation of food given to the labor. Wages also vary according to sex and the type of work performed. The male wage for plowing, for instance, is usually higher than the male wage for weeding. In some villages, the male wage is higher than the female wage for the same type of work performed. Harvesting involves cutting paddy in the field, carrying it to the owner’s home, and separating the rough rice from the straw. Harvest workers are commonly paid in terms of rough rice, as a portion of the total amount of rough rice they have collected. This portion, which is called 'bawon', varies among farmers. In general it is determined on the basis of local custom, and the distance of the paddy field to the owner’s home. It is necessary to note that the number of workers involved in harvesting is large but not precisely known. In this study, animal labor is converted into human labor using the animal and human labor wage ratio as the weight. Thus it is the total money wages of animal labor divided by the average wage of human labor 101 paid by the farmer. Similarly, the total amount of labor in harvesting is simply the total money wage divided by the average human labor wage paid by the farmer. The total money wage of harvesting is approximated by multiplying the amount of rough rice paid to harvest workers by the price of rough rice. The amount of labor, in hours, is the total labors used in the production process including the labor used in harvesting. W235: Output is measured in kilograms of rough rice. It is the total of rough rice before deducting harvest wage. The price of rough rice (Rp/kg) is used in this analysis, and it is the price farmers gave in the survey response. The variation of the output price is usually attributable to variety, moisture content and other criteria related to rice quality. TV are usually more expensive than HYV. WW Revenue is defined as quantity of output multiplied by the price of output. As mentioned before, restricted profit is defined as profit over variable costs, which in a sense is a return attributable to the fixed costs. Labor, seed and fertilizer are treated as variable inputs, while land is the fixed input. Total variable cost is computed as the sum of seed cost, fertilizer cost and labor cost. V. EMPIRICAL RESULTS AND DISCUSSION This chapter consists of six sections and is organized as follows. The first section presents the results of a descriptive analysis of the data. A discussion of production function parameters and technical and allocative inefficiency is covered in section two, followed by section three which presents results of the profit function approach. The discussion on output supply and derived input demand is covered in section four. It should be pointed out that the results presented in sections two and three are based on the estimation of the single equation, i.e., direct estimation of production and profit function, while the results in section four are based on an equation system estimation procedure. Section five describes the results of the estimated multinomial logit model, emphasizing on the discussion of the factors affecting the adoption of the high yielding rice varieties. Finally, section six attempts to evaluate the rice price support and fertilizer subsidy policies using the proposed evaluation procedure described in previous chapter. 5.1. Results of Descriptive Analysis Some simple statistics on the data are presented before proceeding to more complicated analysis. Statistics such as mean values, provide basic information which assists in understanding the more advanced analysis. However, caution should be used in interpreting such statistics since they could easily lead to wrong conclusions. The results of analysis of variance for seasonal and varietal comparisons are also presented. 102 103 The mean values of the quantities of inputs and output, prices, costs and profit are presented in Table 5.1. The standard deviation of the mean values of each variable are also computed but not presented here. The figures presented in Table 5.1 were obtained by first calculating the per hectare measure of the individual farm and then averaging this measure over individuals. The per hectare measure of the inputs and output will tend to be overestimated if the size of cultivated land is very small. The result shows that the average size cultivated land during the dry season is smaller than average size during the wet season. Avoiding drought risk, lack of irrigation water due to inappropriate irrigation systems, and growing relatively less risky crops (secondary crops and vegetables) could be the reasons for the seasonal difference. In some villages, as described in section 2.2.4, farmers cultivate part of their land with secondary crops and vegetables in the dry season. The result of the analysis of variance indicates that the seasonal difference in farm size is in fact significant at the 0.01 level. However, there is little tendency of decreasing farm size over time. The average rice yield is greater for the wet season than for the dry season. This is because farmers use greater amounts of fertilizer in the wet season than in the dry season. There is no information to explain the low yield of the wet season 1976/1977. The dry season of 1977 was relatively long, causing significant yield reductions. The result of the analysis of variance confirms that seasonal yield difference is in fact significant at 0.10 significant level, with the computed F-statistic - 2.80. 104 Table 5.1. Mean Values of Input and Output Per Hectare for Rice Farm Input and - t. D-Drvl Output H75/76 076 H76/77 D77 H82/83 083 Land (Ha) 0.5054 0.3798 0.4522 0.3583 0.4812 0.4219 Output (Kg/Ha) 3216.6 3030.9 2515.7 2358.6 4196.8 4003.0 Seed (Kg/Ha) 39.5 45.5 40.6 38.3 42.3 38.2 Urea (kg/Ha) 194.4 186.8 219.9 N198.5 267.9 251.8 TSP (kg/Ha) 59.0 53.0 63.4 58.2 119.8 111.9 Labor (hrs/ha) 1429.4 1311.4 1193.9 1259.0 1421.3 1390.7 Hired preharvest 623.4 547.7 484.5 542.6 562.3 520.9 Hired harvest 508.5 456.9 368.5 358.1 447.4 , 473.7 Family 297.5 306.7 340.9 348.3 414.7 396.1 Price (Rp/Kg): Rough Rice 62.57 66.94 65.29 74.58 127.21 149.19 Seed 68.75 72.04 74.98 94.49 176.49 185.68 Urea 79.81 79.72 70.58 70.92 87.50 89.65 TSP 78.14 79.56 70.49 71.28 89.08 90.85 Hage 49.50 48.40 55.83 51.99 139.08 137.74 Revenue (Rp/Ha) 201255 202892 164238 175899 533846 597198 Var.Cost (Rp/Kg) 103857 95499 103136 102439 279079 271033 Profit (Rp/Kg) 97398 107393 61102 73460 254767 326166 The amount of seed used is relatively stable over time. There is no consistent relationship between the amount of seed used and season. For 105 example, the amount of seed used per hectare for the wet season of 1975/76 is smaller than for the dry season of 1976, while the opposite is true for the rest of the period. The result of the analysis of variance, where the computed F-value is 0.92, supports the conclusion that there is no seasonal difference in the quantity of seed per hectare. The amount of fertilizer used per hectare tends to increase over time. The mean of the fertilizer use per hectare, except for the last two seasons of the observation, is lower compared to the recommended rate of the BIMAS program (see appendix 2.1). The ' reconlnended urea and TSP application, under package C of the BIMAS program, are 250 kg and 75 kg per hectare, respectively. The result also shows that farmers use more fertilizer in the wet season than in the dry season. This may be because soil moisture content in the wet season is usually more appropriate for higher levels of fertilizer application than in the dry season. The computed F-value for urea is 3.78 (a-0.05), while the computed F-value for TSP is 2.99 (a-0.08). This result indicates that the seasonal difference in fertilizer application is more significant for urea than for TSP. There are no apparent seasonal differences in the means of quantity of labor use per hectare. The result of the analysis of variance, however, indicates that the seasonal difference is in fact significant at the 10% significance level. About 70-80%iof labor used per hectare is hired labor. This is surprising since the majority of the farmers in this study have farms less than 0.65 hectare, and were expected to be more dependent on family labor. The fact that family labor is only a small portion of the total labor use justifies the use of the current labor wage as opportunity cost of labor to calculate the imputed labor cost of family labor. As discussed in chapter 1, the government sets the price of rough rice 106 and the price of fertilizer. The average price of rough rice actually received by the farmers is very close to the floor price. Similarly, farmers buy fertilizer at prices close to the government-established price. The government policy to control fertilizer prices has thus been very effective, primarily because fertilizer is part of the BIMAS credit program for rice farmers. The nominal labor’wage, which the government has no direct control, has increased faster than the prices of rough rice and of fertilizer. The figures of nominal prices, costs and profit show a significant positive trend, particularly in the period of 1977-1982. This is due to a high inflation rate within this period. It should be pointed out that the Indonesian government devalued the rupiah twice during this period, in November 1978 from Rp 415/USS to Rp 615/055, and in March 1983 from Rp 615/USS to Rp 970/055. Table 5.2 shows no significant positive trend in real (deflated) prices. It is not surprising that the real prices of fertilizer and rough rice have decreased over time, as a result of government price controls. As discussed in chapter 1, the government sets the price of rough rice and the price of fertilizer. The average price of rough rice actually received by the farmers is very close to the floor price. Similarly, farmers buy fertilizer at prices close to the government-established price. The government policy to control fertilizer prices has thus been very effective, primarily because fertilizer is part of the BIMAS credit program for rice farmers. Table 5.2 also shows that the wage rate has increased faster than the prices of rough rice and of fertilizer. 107 Table 5.2. Deflated Prices of Inputs, Rough Rice and Profit - D-DrYI H75/76 D76 H76/77 D77 H82/83 D83 RCPI* 225 283 292 301 654 710 Deflator 1.00 1.25 1.30 1.34 2.90 3.15 Rough Rice 62.57 55.78 50.22 55.22 43.86 47.36 Seed 68.75 57.63 57.68 70.51 60.86 58.95 Fertilizer** 78.98 63.71 54.26 53.06 30.44 28.65 Labor _49.50 38.80 42.95 38.80 47.49 43.73 Profit 97398 85915 47002 54821 87851 103545 * Source: CBS, 1983. ** Simple average of the urea and TSP figures in Table 5.1 RCPI: Price indexes of 9 essential goods in rural area. The government operates a parallel system of rice procurement and distribution, partly to provide civil servants, army officers and other budgetary group members with a monthly rice ration and partly to hold as a buffer stock for stabilizing rice price in domestic market (Tabor, 1988). BULOG (National Logistic Agency), acting through it’s network of 00LOGs (Logistic Depots) or through a procurement Task Force, purchases rough rice or rice, mainly from private mills and partly from government sponsored farmers’ cooperatives. Rice is purchased at a floor price set by the government prior to the planting season. Thus, the government purchases rice not only when the market price of rice is below the floor price, but it purchases rice in order to maintain rice stocks for monthly rice rations, in addition to maintaining the buffer stocks for rice price 108 stabilization. Hhen price rises above a ceiling price level, the government begins to distribute rice. Harket operations continue in an effort to maintain prices below the target ceiling price. The floor price is set uniformly for the entire nation, but the target ceiling price is set differently so as to allow for normal movement of rice from surplus to deficit regions. Appendices 5.1, 5.2 and 5.3 present the mean values of the per hectare quantities of inputs and output, costs and profit of the HYV, TV and MV (mixed-variety) of rice, respectively. Detailed discussion and comparison of these three groups of farmers will not be made in this section. However, some interesting findings need to be mentioned here. The average land of the HYV farmers is larger than the average land of the TV farmers. This indicates that large farmers are more responsive to new technology than small farmers. Large farmers are relatively wealthier and usually have more access to credit than small farmers. 0n the average, the HYV farmers use more urea than TV farmers. This is understandable since HVVs tend to be more responsive to fertilizer application (as well as to water availability). Moreover, the HYV farmers are mostly large farmers, who can afford fertilizer. As a result of a significantly higher urea applications, the yield of the HYV farmers is on the average much higher than the yield of TV farmers. The average quantities of seed, TSP and labor per hectare are not significantly different between these two groups. The results of ‘the analysis of’ variance support all these conclusions. 109 5.2. Production Function and Production Inefficiency In addition to obtaining production elasticities, this approach is intended to evaluate the level of technical inefficiency of individual rice-producing farms. Two functional forms of stochastic production frontiers, i.e., Cobb-Douglas and translog, are estimated from panel data (H-171 and T-6). The reason for' estimating the translog production function, in addition to obtaining non-constant production elasticities, is to examine whether the individual level of technical inefficiency is invariant to the functional form. It is assumed that the elasticities of production are constant between firms and over time. This assumption may become increasingly difficult to defend as the time span of the observations increases. However, for the four-year period of the sample, the assumption seems fairly reasonable. One purpose of this study was to compare farmers who planted HYV with those who planted TV rice. However, the survey reported that some farmers switched from HYV to TV or to mixed varieties (HYV and TV), during the six periods of the survey. Separate analysis of TV and HYV is therefore not possible in the panel data analysis framework. The separation of these two varieties is possible using a dummy variable which varies over time (dummy intercept, dummy slope or both). This dummy variable can be treated as other time varying variables in the panel data framework. Pesticide use will also be treated like the dummy for rice variety. 5.2.1. Cobb-Douglas Production Function Three estimators were used, namely the ordinary least squares (OLS), the dummy variable (within) and the EGLS estimator. Recall that only the 110 last two estimators yield a frontier function, while the OLS, which is intended for comparison purposes, gives the usual non-frontier function. The OLS estimator is obtained simply by applying the ordinary least squares method to the pooled data. The dummy variable estimator, hereafter referred to as within estimator, is obtained by applying 0L5 to the transformed data, that is after transforming the data in terms of deviations from individual means. The EGLS estimator, which can be viewed as a weighted average of the within and the between estimator, is calculated by first transforming the data in terms of deviations from a fraction of the individual means and then running OLS on the transformed data. These three estimators are described in chapter 3, section 3.1.1 and 3.1.2. Three statistical tests were performed. The first test is related to the fixed effect (FE) specification to test the null hypothesis that individuals have the same intercept against the alternative hypothesis that their intercepts are not the same. The computed F-statistic (equation 3.7) equals 1.4818 and the critical F0.05(l70, 845) equals 1.2214. Thus, the null hypothesis is rejected at the 0.05 significance level. The second test is the LH test (equation 3.19), related to random effect (RE) specification to test the presence of individual random effects by testing the null hypothesis that on2 equals zero. The computed LH statistic equals 9.5864 and the critical Chi2(1)o.05 equals 3.8415. Since the LH statistic is larger than its critical value, we can therefore reject the null hypothesis. This implies that individual random effects exist. The third test was performed on the null hypothesis that there is no correlation between the individual effects and the included variables 111 against the alternative hypothesis that such correlation exists. The results determine whether to use RE or FE specification. The Mundlak test (equation 3.21) was used in this case. The computed F-statistic is 1.5378 while the critical F0.05(8, 1009) is 1.9384. Thus the RE model rather than the FE model is justified statistically. Note that in performing the Hundlak test, only the time-varying variables which are statistically significant were included. In order to examine the difference between OLS, within and EGLS estimators, the results are presented in Table 5.3. The individual intercepts of the within estimator, which can be estimated directly by using equation (3.5b), are not presented here. The results show that the parameter estimates of the EGLS lie between the corresponding parameter estimates of the other two estimators. The EGLS estimator gives the best fit of the production function compared to the others. The computed coefficient of multiple determination (R2) of the EGLS estimator equals 0.9967, meaning that almost one hundred percent of the variation of the dependent variable is associated with the variation of the specified independent variables in the model. The R2 of’ OLS and the *within estimators are 0.8843 and 0.7497, respectively. The sign of the coefficients estimated by OLS and EGLS is the same and the magnitudes are very close to each other. This is not too surprising, since the (1 - w) is very small, and therefore running LS regression on the transformed data (EGLS estimator) yields similar results to running LS regression on the original data (OLS estimator). Note that 'w' measures the weight given to the between-individual variation. In the covariance (within) estimator, this variation is completely ignored (w-O), while in the OLS estimator this variation is completely incorporated (w-l). 112 Table 5.3. Estimated Parameters of the Cobb-Douglas Production Function Independent E Variables OLS Hithin GLS Constant 5.0868*** - 5.0690*** (0.1916) (0.1938) LKGS 0.1339*** 0.1176*** 0.1304*** (0.0271) (0.0271) (0.0271) LKGN 0.1175*** 0.0878*** 0.1110*** (0.0175) (0.0193) (0.0179) LKGP 0.0735*** 0.0912*** 0.0778*** (0.0114) (0.0116) (0.0115) LLAB 0.2159*** 0.2378*** 0.2211*** (0.0288) (0.0296) (0.0290) LHA 0.4759*** 0.4323*** 0.4676*** (0.0318) (0.0333) (0.0321) DP 0.0066 0.0325 0.0127 (0.0284) (0.0293) (0.0286) 0V1 0.1743*** 0.1768*** 0.1756*** (0.0385) (0.0377) (0.0383) 0V2 0.1389*** 0.1792*** 0.1477*** (0.0541) (0.0531) (0.0539) DSIZE 0.0198 0.0881** 0.0349 (0.0359) (0.0400) (0.0368) 055 0.0496** 0.0555*** 0.0503** (0.0218) (0.0196) (0.0211) 0R1 -0.0505 - -0.0519 (0.0435) (0.0499) 0R2 -0.0403 - -0.0465 (0.0546) (0.0591) 0R3 -0.0640 - -0.0736 (0.0575) (0.0621) 113 Table 5.3 (continued) 024 0.0240 - 0.0118 (0.0527) (0.0585) 0R5 0.0801 - 0.0734 (0.0557) (0.0502) av2 - 0.1059 0.1059 52 - - 0.1525 ouz - - 0.0075 wbov/o - - 0.8372 8 - 1 - w - - 0.1628 E[u] - - 0.071 F-Statistic 514.70 304.34 19208.10 R2 0.8843 0.7497 0.9957 N (individuals) 171 171 171 T (seasons) 6 6 6 Variable definition can be seen in chapter 4 or appendix 5.4 Figures in parentheses are standard deviations *** statistically significant at a-0.01 ** statistically significant at o-0.05 * statistically significant at a-0.10 The results also show little difference in magnitude between the EGLS estimator and corresponding estimate of the within estimator, and no difference in the sign of the estimate. This is consistent with the result of the model specification test previously described. If the null hypothesis that E[X’u]-0 holds, and in fact it is not rejected by the test, the EGLS estimator should not be very different from the within estimator. The significant difference between these (two estimators 114 indicates that the alternative hypothesis holds. This is the basic idea underlying the Hausman test, or equivalently the Hundlak test. The EGLS estimators for the production elasticities with respect to seed, urea fertilizer and land are) greater than the corresponding elasticities of the within estimator. For production elasticities with respect to phosphate fertilizer and labor, the reverse is true. The returns to scale coefficients of these three estimators are 1.0167, 0.9667 and 1.0082, respectively, for OLS, within, and EGLS. A statistical test performed, to test the null hypothesis of'constant returns to scale, could not reject the null hypothesis at the 0.10 level. Thus, these three estimates confirm that the rice production technology is in the stage of constant returns to scale. The following discussion will focus only on the EGLS estimates. However, some comparisons to the within estimates might be made. Before proceeding to production elasticities, let us first interpret the coefficient estimates of the dunlny variables. The dumy variable for pesticide use is not significant, indicating the use of pesticide does not have any effect on the level of production. It was reported that during survey periods no significant crop damage due to insect attack or plant diseases occurred in the study area. The dummy variables of HYV and NV are significant at the 0.01 level. Thus, farmers with HYV and NV produce more output than TV farmers. This is consistent with a priori expectation. The season dummy has a positive sign and is significant at the 0.05 significance level. This indicates that the level of production is greater in the wet season than in the dry season. This is understandable because the soil moisture content in the wet season is usually more optimal for plant growth than in the dry season. In addition, the optimality of soil 115 moisture content induces farmers to use more fertilizers, which in turn increases rice yield. Thus, the lack of water during the dry season is possibly the key reason for the seasonal yield difference. As mentioned in chapter 2, most of the rice fields lack water during dry seasons. The difference in the level of production is represented directly by the coefficient for the season dummy. The regression coefficient for the farm size dummy is not significant, indicating that there is no significant difference in productivity between small and large farmers. The region dummies representing individual non- specific time invariant variables such as climate and soil quality, are not significantly different from zero. This result indicates that there is no signifant difference, statistically, in the level of per household production between regions. The nation-wide rice intensification program (BIHAS/IHHAS), which has already been implemented intensively since the early 1960’s, particularly in Hest Java, could be the main reason. The interpretation of'a Cobb-Douglas production function is very simple and straightforward, since the regression coefficients directly represent the production elasticities of the corresponding independent variables (inputs). Table 5.3 shows that all input coefficients have correct signs and are significantly different from zero at the 0.01 level. The seed coefficient of 0.1300, indicates that a one percent increase in quantity of seed, other things being fixed (ggtgris airings), will result in 0.13 percent increase in the level of rice production. The production elasticities with respect to urea and TSP fertilizer equal 0.1110 and 0.0778, respectively, and both are significant at the 0.001 level. The result shows that rice is more responsive to N fertilizer than P fertilizer. The slow decomposition of phosphorus relative to 116 nitrogen in the soil could be the reason for this difference. Thus the paddy fields may have sufficient phosphorus, but insufficient nitrogen, resulting in a relatively smaller yield response from additional phosphorus compared to yield response from additional nitrogen. The fact that the production elasticities of these two fertilizers are different could be used to support the argument for differentiating the prices of these two fertilizers. This is particularly important if the government aims to gradually reduce fertilizer subsidies. Up to now, however, the prices of these two fertilizers are kept the same by the government. The production elasticity with respect to labor is 0.22. This means that a one percent increase in labor hours will increase the production level by 0.22 percent. Similarly, the interpretation of the production elasticity with respect to land is that a one percent increase in area cultivated per household will result in 0.47 percent of increase in the level of production, indicating the condition of diminishing marginal returns to land. As mentioned before, the results also show that rice production in'west Java is in the stage of constant returns to scale. The test of the hypothesis of constant returns to scale could not be rejected at the 0.10 significance level. This means that a one percent increase of all inputs will result in a one percent increase in the level of production. If this is true, natural land consolidation which has occurred in some places would not have any effect in increasing total rice production, although some advantages may occur particularly those related to post harvest activities such as product processing and marketing. One of the weaknesses of the Cobb-Douglas type of functional form is that it has a constant elasticity property at any level of output and 117 input use. This difficult to defend both theoretically and empirically. Another serious weakness of this functional form is the well known property of constant unitary elasticity of input substitution which will be discussed later in relation to the estimation of cross-price elasticities of the input demand function. Let us now turn to technical inefficiency measures. Individual level of technical inefficiency is estimated using three different methods. The first method obtains individual technical inefficiency by differencing the individual intercepts from the intercept of the most efficient farm, as in equation (3.67). The second method is based on the residual of the EGLS estimator as presented in equation (3.68). The third method is the one suggested by Battese and Coelli (1986) as presented in equation (3.69). Note that these methods, particularly for the first two, give consistent estimates of individual effects (“i) only if both N and T are large. Since in our case T is relatively small, the consistency of the estimates of “i is questionable. The first two methods do not use any distributional assumptions regarding “i’ while the third method uses a half-normal distribution with mode at "i - 0 The first two methods give much larger estimates than the third one. In the case of the first method, the estimate of "i may always be larger than the others since it includes the effects of time invariant variables which could not be included in the model. Given the obvious disadvantage of the first two methods, their estimation results are not discussed. However, it is important to note that although the magnitudes of these three estimators are quite different, the individual ranks based on the level of efficiency are quite similar. The estimates of the individual level of technical inefficiency and 118 their ranks, based on the third method (equation 3.69), are presented in appendix 5.5. The range of technical inefficiency using the third method is 3.4% - 12%, with the mean of 6.5%. The mean is 7% if the calculation is based on equation (3.24a). This figure tells us that rice farms in Hest Java are, on the average, 6.5% technically inefficient or 93.5% technically efficient. One could also interpret that the fitted individual production function is 3.4% - 12%, or 6.5% on the average, below the frontier production function. Given the intensive nature of wetland rice farming in Hest Java, this relatively small figure of technical inefficiency is reasonable. The figure, presumably, would be much greater in the case of dryland rice farming since the government has given relatively less attention to intensifying dryland farming. Policy implications of the efficiency measure are debatable. Some advocates of frontier analysis claim that a firm can move from the interior of the production function surface to the frontier without any cost to the firm. They assert that better use of the existing technology in terms of cultivation and crop-management practices will definitely increase yield. They do not, however, specifically address the question of how this can be achieved. On the other hand, nonadvocates would argue that free correction is very unrealistic, since movement to frontier requires adjustments of factors of production including management skills which could be regarded as a fixed factor. Improving management skills is of course not without cost. Table 5.4 shows that no significant difference in the level technical inefficiency between small farms and large farms; small farms may or may not be more technically efficient than large farms 0r vise versa. This is consistent with the regression results which yield a coefficient on the 119 farm size dummy variable not significantly'different from zero. Therefore, it seems reasonable to interpret the mean of technical inefficiency level (6.5%), as a per-hectare output loss due to technical inefficiency. Alternatively, we can simply reestimate a per-hectare production frontier and find both the mean and the individual level of technical inefficiency. Using rice yield (kg/ha) figures as presented in Table 5.1, and the annual harvested area of rice farms in Hest Java (1.74 million hectares), we could roughly 'estimate the annual output loss, due to technical inefficiency. Using rice yield for the 1983 dry season (4003 kg/ha), the per-hectare quantity of rice loss was 260 kg, and the total quantity of rice loss in Hest Java rice farms would be 0.45 million ton annually. Thus, better use of existing technology of rice production provides an opportunity to somewhat increase rice yield and total rice production in Hest Java. Table 5.4. Frequency Distribution of Farmers Based on The Level ,of Technical Inefficiency from the Cobb-Douglas Production Frontier Range of Technical % farms % from Inefficiency s 0.5 Ha > 0.5 Ha total farms 5 5% 13.4 13.6 13.5 5% < u s 10% 83.5 84.1 83.6 10% < u s 15% 3.1 2.3 2.9 Total % 100 100 100 Total farms 127 44 171 120 Let us now evaluate allocative efficiency of the input use. As mentioned in chapter 3, a farm is allocatively efficient if the input use maximizes profit, that is if the value of marginal product (VHP) of particular input equals its marginal factor cost (HFC). This condition implies that at the point of profit maximization, the ratio (d) of VHP to HFC for each input is equal to one. This also means that the last dollar spent on each input must return exactly one dollar, and most if not all previous units will have given back more than a dollar (Debertin, 1986). The accumulation of the excess dollars in returns over costs represents the profits or net revenues accruing to the farm. A simple evaluation for allocative inefficiency can be conducted by calculating the "(1" ratio for individual farm based on the estimated production function and the price levels reported in the survey. There is no intention to interpret the level of allocative (in)efficiencies for an individual farm. Seasonal average and grand average values of the 'd' ratio for each input are presented in Table (5.5). On the (grand) average the 'd' ratio for seed, urea, TSP, and labor are 10.34, 3.09, 14.60, and 0.89, respectively, indicating underutilization of seed and fertilizers and overutilization of labor. This, however, does not necessarily mean that farmers do not attempt to maximize profits; rather it may mean that farmers, for various reasons, were not able to maximize profit. 121 Table 5.5. Allocative Inefficiency Measures ('d' ratio) in Rice Production. Season Seed Urea Phosphate Labor H-75/76 12.04 3.74 16.74 1.06 D-76 11.45 3.79 19.38 0.93 H-76/77 8.58 2.46 17.00 0.78 0-77 7.39 2.97 16.11 0.74 H-82/83 10.67 2.58 6.03 0.85 D-83 11.92 3.04 12.37 1.03 Average 10.34 3.09 14.60 0.89 5.2.2. Translog Production Function As in the case of Cobb-Douglas, the translog production function is estimated using three different estimators, namely OLS, the within and EGLS. However, none of the statistical tests outlined in section 5.2.1. is repeated here. The regression result is presented in Table 5.6. An F test was conducted to test the null hypothesis that all coefficients of the second order terms of the inputs are equal to zero, in order to test whether a Cobb-Douglas functional form is the correct one. The computed F(15,995) statistic is 6.3868, while the critical F0.05(15,995) is 1.6664. The null hypothesis, therefore, is rejected. Further examination, however, indicates that this model suffers from serious multicollinearity problems, since numerous regression coefficients were not significant at 0.01-0.10 significance level while the computed 122 F-value of the model was very high. Although multicollinearity does not cause bias of 'the estimates, it does enlarge the variance of ,the estimates, resulting in a greater tendency to accept the null hypothesis. To more formally measure the degree of multicollinearity, one can compare the coefficient of multiple determination (R2) of the model and the R2 of each auxiliary regression (i.e., regression between each independent variable and the remaining independent variables). If R2 of the main regression is greater than R2 of the auxiliary regression the presence of multicollinearity can be ignored. Because multicollinearity is directly dependent upon the sample of observations, little can be done to resolve it unless more information is available. The information may be additional sample observations or non- sample information (i.e., imposing restrictions on the parameter space). The second choice is sometimes preferable, though not without risk, since the first choice is usually very costly and/or may not be feasible. Since non-sample information is uncertain, its use may or may not improve estimator performance in repeated sampling, although its use does ensure that the variances of the parameter estimates are reduced. Using incorrect restrictions incurs the risk of biasedness of the parameter estimates. On the other hand, if nonsample information exists about which we are relatively certain, then it can be used in conjunction with the sample data to obtain more precise estimators (Fomby et al., 1984). More discussion about multicollinearity problem will be given later in relation to the estimation of output supply and input demand elasticities. 123 Table 5.6. Estimated Parameters of the Translog Production Function Independent E Variables OLS Hithin GLS Constant 2.8946* - 2.9631* (1.8055) (1.7896) LKGS 1.1509*** 1.0506*** 1.1297*** (0.3699) (0.3636) (0.3682) LKGH 0.1165 0.3251 0.1763 (0.1628) (0.2587) (0.2617) LKGP -0.3551** -0.447l*** -0.3742** (0.1639) (0.1596) (0.1621) LLAB 0.6856 0.5188 0.6407 (0.5801) (0.5656) (0.5759) 0.5*LKGSKGS -0.0259 -0.0241 -0.0265 (0.0246) (0.0244) (0.0245) LKGSKGN -0.0088 -0.0069 -0.0074 (0.0376) (0.0365) (0.0372) LKGSKGP 0.0536** 0.0555** 0.0549** (0.0236) (0.0230) (0.0234) LKGSLAB -0.1587*** -0.1502*** -0.1578*** (0.0614) (0.0606) (0.0612) 0.5*LKGNKGN 0.0029 -0.0027 0.0017 (0.0140) (0.0137) (0.0139) LKGHKGP 0.0325* 0.0315* 0.0324* (0.0191) (0.0187) (0.0189) LKGNLAB -0.0210 -0.0518 -0.0307 (0.0372) (0.0365) (0.0370) 0.5*LKGPKGP 0.0206*** 0.0219*** 0.0208*** (0.0078) (0.0079) (0.0078) LKGPLAB -0.0047 0.0101 -0.0016 (0.0243) (0.0232) (0.0239) 0.5*LLABLAB 0.0086 0.0297 0.0154 (0.0500) (0.0491) (0.0497) Table 5.6 (Continued) 124 LHA 0.1824 0.2412 0.1893 (0.4473) (0.4315) (0.4429) 0.5*LHALHA 0.0257 0.0370 0.0293 (0.0339) (0.0327) (0.0335) LHAKGS 0.1172** 0.0982* 0.1119** (0.0555) (0.0543) (0.0552) LHAKGN -0.0595 -0.0301 -0.0523 (0.0451) (0.0452) (0.0451) LHAKGP -0.0877*** -0.1082*** -0.0925*** (0.0271) (0.0259) (0.0270) LHALAB 0.1018 0.0903 0.1008 (0.0709) (0.0597) (0.0705) DP 0.0103 0.0372 0.0143 (0.0277) (0.0285) (0.0282) 0V1 0.1719*** 0.1755*** 0.1741*** (0.0377) (0.0371) (0.0375) 0V2 0.1022* 0.1379*** 0.1100** (0.533) (0.0527) (0.0531) 055 0.0455** 0.0489*** 0.0478** (0.0212) (0.0191) (0.0204) 081 -0.0441 - -0.0454 (0.0428) (0.0503) 082 0.0240 - 0.0175 (0.0541 (0.0589) 0R3 -0.0017 - -0.0171 (0.0559) (0.0514) 084 0.0585 - 0.0523 (0.0518) (0.0580) 085 0.1375*** - 0.1255*** (0.0547) (0.0591) av2 - 0.1004 0.1004 c2 - - 0.1528 125 Table 5.6 (Continued) a"z - - 0.0087 w-av/o - - 0.8106 8 - 1 - w - - 0.1894 E[u] - - 0.075 F-Statistic 287.82 138.97 10751.28 82 0 8934 0.7590 0.9958 N (individuals) 171 171 171 T (seasons) 6 6 6 Variable definition can be seen in chapter 4 and appendix 5.4 Figures in parentheses are standard deviations *** statistically significant at a-0.01 ** statistically significant at a-0.05 * statistically significant at o-0.10 The level of individual technical inefficiency is very close to that obtained from the Cobb-Douglas production frontier. The range of the individual level of technical efficiency is slightly wider than the range obtained in Cobb-Douglas case, that is 3.4% - 13.4%, with the mean 7.0%. The mean is 7.4% if equation (3.24a) is used. The rank of the individual technical inefficiency is not affected at all; it is exactly the same as the rank obtained from the Cobb-Douglas production frontier. Thus, the measure of individual technical inefficiency is invariant to the functional form of the production. The frequency distribution of farms based on technical inefficiency level is described in Table 5.7. 126 Table 5.7. Frequency Distribution of Farmers based on The Level of Technical Inefficiency from the Translog Production Frontier Range of Technical % farms % from Inefficiency s 0.5 Ha > 0.5 Ha total farms 5 5% 10.2 11.3 10.5 5%. < u s 10% 84.3 79.6 83.0 10% < u s 15% 5.5 9.1 6.5 Total % 100 100 100 Total farms 127 44 171 5.3. Profit Function and Profit inefficiency The direct estimation of profit function also uses three different estimators as in the case of the production function. The three statistical tests outlined in section 5.2.1, were repeated here. Statistical tests for both the FE specification (standard F test) and the RE specification (LH test) confirm the existence of individual specific effects representing individual levels of inefficiencies. The third test, the Mundlak test, justifies the use of RE specification, since the test could not reject the null hypothesis that no correlation exists between the individual effects and the included exogenous variables. This test, therefore, insures that the EGLS estimator is more efficient than the within estimator. 127 5.3.1. Cobb-Douglas Profit Frontier The parameter estimates of the Cobb-Douglas normalized profit function are presented in Table 5.8. The computed F-value of OLS, within and EGLS is 162.0, 117.2, and 4053.6, respectively. The EGLS estimator provides a best fit compared to other estimators, with R2 - 0.9825, compared to 0.6755 for OLS and 0.4458 for the within estimator. As in the case of production frontier, the parameter estimates of these three estimators are close enough to each other both in sign and magnitude. Again, this indicates that the random effect specification is valid. In the situation where the assumption of random effect is correct, the EGLS is more efficient than the within estimator. The following discussion will focus only on the EGLS estimator, although some comparisons to the others will also be made. Except for labor wage, the coefficient estimates of the variable input prices have negative signs as expected, and are statistically significant at the 0.01 level for fertilizer price and at the 0.10 level for seed price. The regression coefficient of the normalized labor wage has a positive sign, which is unexpected, and is significant at the 0.10 significance level. The profit elasticity with respect to the normalized seed price is - 0.1331, meaning that a one percent increase in normalized (real) seed price will reduce the normalized profit by 0.13 percent. This relatively small elasticity is reasonable since seed input, in terms of its value, accounts for a very small percentage of the total value of output. 128 Table 5.8. Estimated Parameters of the Cobb-Douglas Normalized Profit Frontier Independent E Variables OLS Hithin GLS Constant 6.9280*** - 6.9315*** (0.1093) (0.1184) LPS -0.1408* -0.1173 -0.1331* (0.0845) (0.0818) (0.0758) LPF -0.3670 *** -0.3394*** -0.3596*** (0.0785) (0.0673) (0.0759) LHG 0.1621 0.2817** 0.1937* (0.1049) (0.1891) (0.1021) LHA 0.9905 *** 0.9194*** 0.9794*** (0.0250) (0.0359) (0.0272) DP 0.1241** 0.1343** 0.1279** (0.0543) (0.0563) (0.0550) 0V1 0.2931*** 0.2133*** O.2734*** (0.0786) (0.0778) (0.0785) 0V2 0.1793* 0.2657*** 0.2019** (0.1094) (0.1071) (0.1088) DSS 0.0218 0.0555 0.0222 (0.0477) (0.0696) (0.0569) 0R1 -0.0179 - -0.0161 (0.0863) (0.1024) DR2 0.2555** - 0.2475** (0.1090) (0.1200) 0R3 0.2109** - 0.1940** (0.1058) (0.1182) 0R4 0.1553 - 0.1373 (0.1017) (0.1165) 0R5 0.4336*** - 0.4212*** (0.1046) (0.1163) 129 Table 5.8 (Continued) ovz - 0.4278 0.4278 02 - - 0.5572 5“2 - - 0.0382 w-ov/o - - 0.8068 8 - 1 - w - - 0.1932 E[u] - - 0.155 F-Statistic 152.0 117.2 4053.5 82 0.5755 0.4458 0.9825 N (individuals) 171 171 171 T (seasons) ’ 6 6 6 Variable definition can be seen in chapter 4 or in appendix 5.4 Figures in parentheses are standard deviations *** statistically significant at o-0.01 ** statistically significant at a-0.05 * statistically significant at a-O.10 The profit elasticity with respect to the normalized fertilizer price is -0.3596, which indicates that a one percent increase in normalized price of fertilizer results in 0.36 percent reduction in normalized profit. Again the relatively low elasticity is due to the fact that the total value of fertilizer is only a minor percentage of the total profit or total value of output. The sign of the labor wage coefficient is unexpected, and is rather difficult to interpret. However, the magnitude of the coefficient is very small, that is 0.1937 and significant at the 0.10 level. This can be interpreted that the variation of the normalized profit is weakly associated with the variation of the normalized labor wage. The 130 explanation of this finding could be as follows. Labor is a dominant production input in Indonesian rice farming in general, and particularly in densely populated areas such as Hest Java. No substitutes are available for this input. In addition, economic considerations are not the only driving factor in hiring labor, and there may be many other factors which are more relevant in the rural situation. For example, it is very common for farm households to hire and to be hired by other households. Since money' wage is only part of the total wage, (in some places like Hargabinangun it is only small part), this labor exchange situation occurs regardless the level of wage rate. This may explain why farmers are not so responsive to wage levels in their farm activities. Gunawan (1988), using the same data and cost function model, also found an unexpected (negative) sign for the labor wage coefficient. The profit elasticity with respect to land is positive and statistically significant at the 0.01 significance level. The magnitude of 0.9794 of this coefficient tells us that one percent increase in farm size will result in 0.9794 percent increase in profit. In our case, this coefficient directly represents the returns to scale coefficient of the underlying production function. The within estimator of this coefficient is smaller, that is 0.9194, and is significant at the 0.01 significance level. The test related to scale coefficient will be discussed later. The coefficient estimates of the dummy variables for pesticide and rice variety are all significant with positive signs, indicating that the use of pesticide and HYV increases the profit. The question is then why large numbers of farmers are not using HYV. It is very likely that economic factors are not the only ones farmers use to make decisions regarding the use of HYV. In Gunung Hangi (kabupaten Majalengka), for example, efforts 131 to increase adoption of HYV have been unsuccessful, and almost all respondents in this village grew TV. Preference for growing TV could also be explained by the fact that most of the rice produced is for own- consumption, and since villagers find TVs taste better than HYV they prefer growing TV. Unlike in the production function, the dummy variable for season is not significantly different from zero, indicating that season is not an explanatory variable for the profit variation. Profit is a function of price and total product. Price in the dry season sometimes is higher than in the wet season due to inelastic nature of rice supply function. The relatively higher price in the dry season may offset the reduction of total product, resulting in a non-significant difference between the wet and dry seasons profits. Dummy variables for regions all have positive signs, although only three of them are significantly different from zero, indicating that, compared to Hargabinangun (as a control) the restricted profit earned by farmers in the other villages is higher. In light of individual village’s accessibility factors described in chapter 3, ‘this finding is reasonable, since these three villages have better product marketing channels due to better transportation facilities. The higher the price received, the higher the profit earned. Before proceeding to the discussion of profit inefficiency, let us consider several advantages of Cobb-Douglas functional form, despite its weaknesses discussed later. One of its advantages, in addition to its simplicity, is that the dual-primal relationship is tractable, i.e., one can derive the production parameters from the profit (cost) function, or vice versa. Recalling equation (2.34), note that coefficient of the kth 132 kth production variable input price (0k) equals ak/(l-r), where ak is the parameter and r - 2k ak. After some algreba we find that ak - '“k /(1 - r*), where r* - 2k ck. Similarly for the fixed input (land), we get that bh - 5h /(1 - r*), where b and p are the production and profit elasticity with respect to farm size, respectively. Using these equations and the results in Table 5.7, for the EGLS estimator, we can obtain the indirect estimates of production elasticities, as presented in Table 5.9 below. The result shows that, except for seed variable, the corresponding direct and indirect estimates are very different from each other. The indirect estimate of production elasticity with respect to labor is negative, indicating the production function is not well-behaved. This is because the profit function, from which indirect estimates are derived, is also not well-behaved, as marked by the positive sign of the labor. wage. Table 5.9. Direct and Indirect Production Elasticities of the Cobb-Douglas Production Function Input Direct Estimatea Indirect Estimateb Seed 0.1304 0.1025 Fertilizer 0.0944c 0.2758 Labor 0.2211 -0.1491 Farm Size 0.4676 0.7540 a) This is the EGLS estimator from Table (5.3) b) Computed from the EGLS estimator in Table (5.7) . c) A simple average of LKGN and LKGP coefficients 133 Another advantage of Cobb-Douglas functional form is that the test of constant returns to scale for underlying production function can easily be performed. He know that 3h - bh /(1-r). If there is only one fixed input, as in our case, then this following relation holds: if p - 1, then b + r - 1 where b + r is the returns to scale coefficient of the production function. Thus, the hypothesis of constant returns to scale can simply be tested by setting up the hypothesis whether or not 8 equals one. Following this procedure, we found that the null hypothesis of constant returns to scale for the OLS and EGLS estimators could not be rejected at the 0.10 level. Thus, the underlying production function is in the stage of constant returns to scale. This conclusion is consistent with the result of the direct production function estimation presented before. In contrast, the within estimator confirms that production function is in the stage of decreasing returns to scale, since the null hypothesis was rejected at the 0.10 level. The rate of return to the fixed input in the Cobb-Douglas case is also readily available, equal to the partial derivative of the normalized profit function with respect to corresponding fixed input. In our case, the rate of return to land is equal to fll/Z, where 8 is the coefficient of farm size, while n/Z is the average normalized profit per hectare. Since constant returns to scale is confirmed, the rate of return to land therefore is exactly the profit per hectare. This result can be generalized for more than one fixed input. 134 The computation of output supply and input demand elasticities from the Cobb-Douglas profit function is also simple, compared to the translog profit function. The elasticities of output supply and input demand presented in Table 5.10 are estimated using equation (3.39) and (3.40) and the EGLS estimates in Table 5.8. More discussion about output supply and input demand elasticities will be covered in section 5.4. Table 5.10. The Own- and Cross-price Elasticities and the Elasticities Hith Respect to Land of Rough Rice Supply and Variable Input Demand Primf Farm Rice Seed Fertilizer labor size Rice Supply 0.2290 -0.1331 -0.3596 -0.1937 0.9794 Demand for: Seed 1.2290 -1.1331 ~0.3596 -0.1937 0.9794 Fertilizer 1.2290 -0.1331 -1.3596 -0.1937 0.9794 Labor 1.2290 -0.1331 -0.3596 -0.8063 0.9794 Let us now examine the level of profit inefficiency. The individual level of profit inefficiency and its individual rank are presented in appendix 5.7. The frequency distribution of farms based on technical inefficiency level is shown in Table 5.11. The computation, as in the case of technical inefficiency, follows the equation (3.69) of the Battese and Coelli (1986). The level of profit inefficiency ranges from 6.9%.to 28.9% with the mean 13.8%. If the equation (3.24a) used, the mean is 18.4%8 This 135 indicates that on the average rice farmers are 13.8% profit inefficient. This percentage can also be interpreted as a percentage of profit loss. Table 5.11 also shows, as in the case of production function, that the level of profit inefficiency does not have any association with the farm size, meaning that small farms may or may not be more efficient than large farms. Therefore, as in the case of production frontier discussed above, it seems reasonable to interpret the mean of profit inefficiency level (13.8%), as a per-hectare profit loss due to inefficiency. Alternatively, we can reestimate a per-hectare profit frontier to get the same measure of profit inefficiency. Table 5.11. Frequency Distribution of Farmers Based on the Level of Profit Inefficiency from the Cobb-Douglas Normalized Profit Function Level of Profit % from farms with % from Inefficiency s 0.5 Ha > 0.5 Ha total farms 5 5% 0.0 0.0 0.0 5%.< u S 10% 11.8 6.8 10.5 10% < u s 15% 58.3 63.6 59.6 15% < u s 20% 24.4 20.5 23.4 20% < u s 25% 3.9 6.8 4.7 25% < u s 30% 1.6 2.3 1.8 > 30% 0.0 0.0 0.0 Total % 100 100 100 Total Farms 127 44 171 136 Given a figure of profit per hectare as presented in Table 5.1, and the annual harvested area of rice farms in Nest Java (1.74 million hectare), we can approximate the annual profit loss due to inefficiency (both technical and allocative inefficiencies). Using perehectare profit figure in the dry season 1983 (Rp 326,000/ha), the per-hectare profit loss amounts to about Rp 45,000, and the total profit loss in rice farms in Hest Java amounts to about 78 billion Rupiahs annually (USS 81 million, using exchange rate Rp970/USS in 1983). Thus, the benefits of promoting increased efficiency in rice farm in Indonesia, particularly'in Hest Java, appear to be extremely attractive. 5.3.2. Translog Profit Frontier As in the case of translog production frontier, the estimated translog profit frontier also suffers from multicollinearity problem. While most coefficient estimates have correct signs, many are not statistically significant (Table 5.12). There is no attempt to interpret the coefficient estimates individually, since they do not provide valuable information. Apart from the fact that some of them are not significant, the profit elasticities with respect to variable input prices are not constant in the case of translog profit function. The more meaningful interpretation would be associated with the measure of the elasticities of output supply and input demand, which is discussed later. 137 Table 5.12. Estimated parameters of the Translog Profit Function Independent E Variables OLS Hithin GLS Constant 6.6701*** - 6.6905*** (0.1208) (0.1298) LPS -0.1913 -0.1802 -0.1931 (0.1807) (0.1744) (0.1785) LPF -0.6695*** -0.6372*** -0.6564*** (0.0785) (0.1498) (0.1590) LHG -0.2972 -0.0797 -0.2171 (0.2228) (0.2033) (0.2154) 0.5*LPSLPS 0.1906 0.2635 0.2201 (0.1849) (0.1782) (0.1827) LPSLPF -0.4571* -0.5434** -0.4841** (0.2496) (0.2332) (0.2437) LPSLHG 0.2767 0.0363 0.1950 (0.3698) (0.3456) (0.3617) 0.5*LPFLPF 1.0926*** 1.1027*** 1.0984*** (0.2350) (0.2108) (0.2262) LPFLHG -2.1349*** -2.0544*** -2.1114*** (0.4321) (0.3990) (0.4207) 0.5*LHGLHG 0.1349 0.1188 0.1300 (0.3155) (0.2941) (0.3081) LHA 1.0305*** 1.0189*** 1.0381*** (0.0602) (0.0742) (0.0631) 0.5*LHALHA 0.0398** 0.0529*** 0.0448*** (0.0157) (0.0189) (0.0166) LHALPS 0.0426 0.0600 0.0454 (0.0744) (0.0745) (0.0745) LHALPF -0.2828*** -0.2975*** -0.2879*** (0.0744) (0.0696) (0.0720) LHALHG -0.2078** -0.1310 -0.1778* (0.1061) (0.0981) (0.1033) 138 Table 5.12 (Continued) DP 0 0555 0.0548 0.0559 (0.0542) (0.0571) (0.0553) DV1 0.2869*** 0.2072*** 0.2517*** (0.0774) (0.0755) (0.0772) ovz 0.1775* 0.2509*** 0.2034** (0.1075) (0.1052) (0.1059) oss 0.0005 0.0045 0.0022 (0.0454) (0.0595) (0.0578) 081 -0.0702 - -0.0725 (0.0852) (0.1052) 082 0.3523*** - 0.3534*** (0.1088) (0.1218) 0R3 0.2895*** - 0.2593** (0.1059) (0.1205) 084 0.3455*** - 0.327s*** (0.1058) (0.1228) 0R5 0.5323*** - 0.5191*** (0.1074) (0.1210) av2 - 0.3924 0.3924 02 - - 0.5582 2 au - - 0.0450 w-ov/o - - 0.7664 8 - 1 - w - - 0.2335 E[u] - - 0.171 F-Statistic 101.5 58.5 2341.8 82 0.5997 0.4458 0.9825 ADJ-R2 0.5928 0.4420 0.9821 N (individuals) 171 171 171 T (seasons) 6 6 6 139 The coefficient of the dummy variables can be interpreted individually as before. The coefficient of the dummy varieties have the same signs and relatively the same magnitude compared to the Cobb-Douglas functional form. Similarly the coefficients of the dummy regions do not change in signs, the only difference being the coefficient of 0R4 which previously was not significant now is highly significant. The individual level of profit inefficiency, presented in appendix 5.8, is very close to the level obtained from Cobb-Douglas functional form. This is not surprising, and indeed it is expected. The level of profit inefficiency should be invariant from the functional form. The frequency distribution of farms based on the profit inefficiency level is described in Table 5.13. 5.4. Profit Maximizing Output Supply and Variable Input Demand Function In general, the results of the direct estimation of the profit frontier outlined so far are not satisfactory. In the case of the Cobb—Douglas profit frontier, for example, the wage coefficient has the wrong sign which violates the behavioral assumption of profit maximization. The severe multicollinearity problem which apparently occurs in the estimation of translog profit frontier, makes the coefficient estimates meaningless since many of them, although having the correct signs, are not significantly different from zero. Thus, despite the fact that we were able toimeasure and interpret the individual level of profit inefficiency, we face a serious problem making meaningful interpretations of the estimated coefficients. The output supply and variable input demand functions derived from the above profit frontier'will also be meaningless. 140 Table 5.13. Frequency Distribution of Farmers Based on the Level of Profit Inefficiency from the Translog Normalized Profit Frontier Level of Profit % from Earns with % From Inefficiency s 0.5 Na > 0.5 Ha Total Farms 5 5% 0.0 0.0 0.0 5% < u s 10% 7.1 4.5 6.4 10% < u s 15%) 53.5 54.6 53.8 15% < u s 20% 28.4 25.0 27.5 20% < u s 25% 8.7 9.1 8.8 25% < u s 30% 1.6 6.8 2.9 > 30% 0.8 0.0 0.6 Total % 100 100 100 Total Farms 127 44 171 The coefficient estimates presented so far are obtained from a single equation estimation procedure, that is by direct estimation of the profit function with no restrictions imposed on these estimates. Notice that, as discussed in chapter 3 section 3.4, some of the profit frontier parameters (i.e the parameters of variable factors of production), can alternatively be estimated from a set of factor demand (factor share) equations. Under the hypothesis of profit maximizing and price taking behavior on the part of farmers, the parameters from the factor share equations must be equal to the corresponding parameters in the profit function. Yotopoulos and Lau (1979) explore this equality property as a 141 basis for testing the hypothesis of profit maximization, and simultaneously estimating the profit function and factor share functions. It may also be desirable»to maintain the hypothesis of profit maximization as part of model specification. The output supply and input demand functions derived from this model should be referred to as profit maximizing output supply and input demand functions. Given the facts that (i) EGLS estimators of the slope parameters, in the above single equation case, are very close to the OLS estimators of corresponding slope parameters, (ii) the complexity in estimating the SUR type for a random effect model, (iii) the unavailability of a statistical package to handle such model, and (iv) limited time available, it is desirable to approximate the slope parameters by simply running joint generalized least squares to the system equation of the pooled data. In this case we turn back to the non-frontier type of functions, rather than frontier function previously estimated. Since the slope parameters are more relevant in deriving output supply and input demand functions, it seems reasonable to use the average function as an approximation of the frontier function. System equations of Cobb—Douglas profit function (equation 4.5) and system equations of translog profit function (equation 4.6), are estimated using the Zellner’s SUR estimation method. In this analysis the "exact“ profit maximizing hypotheses is maintained, by imposing equality restrictions of the parameters in profit function with the corresponding parameters in factor share equations. In the case of the translog profit function, this hypothesis also implies a symmetry restriction. However, as mentioned before, there is a tradeoff in making the decision to impose restrictions. The main advantage for maintaining 142 certain hypotheses by imposing restrictions is that the efficiency of the estimator of the parameters can be increased, whether or not the restrictions are correct. The main disadvantage is that the parameter estimate is biased if the restrictions are in fact incorrect. Consequently, we are faced with a decision whether to use unbiased estimator that has larger variance or the biased restricted estimators that has smaller' variance. In our' case, the restriction of’ profit maximization is considered to be reasonably'correct, since this behavioral assumption is intuitively and theoretically acceptable. The restricted parameter estimates of the Cobb-Douglas profit function are presented in Table 5.14. All the input price coefficients are significant at the one percent significance level and have correct signs. The result indicates that a one percent increase in prices of seed and fertilizer and in labor wage will reduce profit by 0.031, 0.1932 and 0.6007 percent, respectively. The coefficient estimates of‘ dummy variables are very close in magnitude to the corresponding coefficient estimates in the case of single equation estimators previously presented (Table 5.8), and except for the dummy variable of region one (0R1) they all have the same sign. Thus, all conclusions related to the dummy variables are the same as those previously made. For example, the total profit of farms with HYV and MV are significantly greater than the profit of the TV rice farms. Table 5.15 describes the restricted parameter estimates of the translog profit function. Unlike the Cobb-Douglas function, the regression coefficients of the translog function, except those related to the dummy variables, are not readily interpreted. The profit elasticities with respect to variable input prices and quantity of fixed inputs are not 143 Table 5.14. Estimated Parameters of the Cobb-Douglas Normalized Profit Function Using Seemingly Unrelated Regression Estimation Method Independent Profit fit Variables Function Seed Fertilizer Labor Intercept 7.0441*** -0.0318*** -0.1932*** -0.6007*** (0.0694) (0.0028) (0.0113) (0.0373) LPS -0.0318*** - - - (0.0028) LPF -0.1932*** - - - (0.0113) LNG -0.6007*** - - - (0.0373) LHA 0.9547*** - - - (0.0168) DP 0.1599*** - - - (0.0363) 0V1 0.3236*** - - - (0.0507) 0V2 0.2753*** - - - (0.0732) 0R1 -0.1382** - - - (0.0583) 0R2 -0.0457 - - - (0.0695) 0R3 0.0839 - - - (0.0691) 0R4 0.1051* - - - (0.0678) 0R5 0.2751*** - - - (0.0687) Variable Definition can be seen in chapter 4 or in appendix 5.4 *, **, and *** are difined as in the previous Tables. 144 Table 5.15. Estimated Parameters of the Translog Normalized Profit Function Using Seemingly Unrelated Regression Method Independent Profit Variables Function Seed Fertilizer Labor Intercept 6.9888*** -0.0360*** -0.3148*** -1.0465*** (0.0721) (0.0053) (0.0206) (0.0698) LPS -0.0360*** -0.0591*** -0.0208*** -0.0338*** (0.0053) (0.0074) (0.0072) (0.0103) LPF -0.3148*** -0.0208*** -0.2082*** -0.2930*** (0.0206) (0.0072) (0.0271) (0.0364) LHG -1.0465*** -0.0338*** -0.2930*** -1.5378*** (0.0698) (0.0103) (0.0364) (0.1035) 0.5*LPSLPS -0.059l*** - - - (0.0074) LPSLPF -0.0208*** - - - (0.0072) LPSLHG -0.0333*** - - - (0.0103) 0.5*LPFLPF -0.2082*** - - - (0.0271) LPFLHG -0.2930*** - - - (0.0364) 0.5*LHGLHG -1.5378*** - - - (0.1035) LHA 1.0274*** 0.0024 0.0120 0.0395 (0.0368) (0.0028) (0.0114) (0.0378) 0.5*LHALHA 0.0253** - - - (0.0103) LHALPS 0.0024 - - - (0.0028) LHALPF -0.0120 - - - (0.0114) 145 Table 5.15 (Continued) LHALHG 0.0395 - - - (0.0378) DP 0.1283*** - - - (0.0347) 0V1 0.3511*** - - - (0.0485) 0V2 0.2872*** - - - (0.0598) 081 -0.1072* , - - - (0.0570) 082 0.1321** - - - (0.0575) 083 0.1540** - - - (0.0553) 084 0.2093*** - - - (0.0554) 085 0.3594*** - - - (0.0559) Variable Definition can be seen in chapter 4 or in appendix 5.4 Figures in parentheses are standard deviations *** statistically significant at o-0.0l ** statistically significant at a-0.05 * statistically significant at o-O.10 constant. A common practice is to estimate these elasticities for average levels of variable input prices and fixed input quantities. The magnitudes of the coefficient estimates of the price variables are very different from the corresponding estimates in the case of single equation estimators previously described (Table 5.12), even though some have the same signs. Similarities, however, hold for the coefficient estimates of the first order term of farm size and the coefficient 146 estimates of the dummy variables. He now come to the estimation and evaluation of the profit maximizing input demand and output supply elasticities. The own- and cross-price elasticities as well as the elasticities with respect to farm size of the output supply and input demand are estimated using equations (3.39) and (3.40) for the Cobb-Douglas profit function, and equations (3.45 - 3.48) and equation (3.51 - 3.53) for the translog profit function, described in chapter 3. The means of factor shares to restricted profit for seed, fertilizer and labor are 0.063, 0.386 and 1.125, respectively. The mean (over time) of 'the deflated prices in Table 5.2 are used in the computation. The results are presented in Table 5.16. The translog functional form is superior to the Cobb-Douglas form in the measures of input demand and output supply elasticities. The former gives more reasonable elasticity figures than the latter. In the Cobb- Douglas case, fori example, the impact across variable input demand functions for a given change in any of the exogenous variables is synlnetry. This is due to the well-known property of constant unitary elasticity of substitution among all input pairs for Cobb-Douglas function. This is of course very unreasonable, unless only two production inputs are involved. The impact of a similar change in the case of translog function, on the other hand, varies across input demand equations, which is very consistent with a DLiQIJ. theoretical expectations. In the Cobb-Douglas case, moreover, the own-price elasticities for the well-behaved profit function, are definitely greater than one in absolute values, which are of course difficult to defend theoretically and empirically. 147 Table 5.16. The Own- and Cross-price Elasticities and the Elasticities with Respect to Land of the Rough-Rice Supply and Variable Input Demands. Pricg_gf: Farm Rice Seed Fertilizer Labor Size W Rice Supply 0.8257 -0.0318 -0.1932 -0.6007 0.9545 Demand for: Seed 1.8257 -1.0318 -0.1932 -0.6007 0.9545 Fertilizer 1.8257 -0.0318 -l.1932 -0.6007 0.9545 Labor 1.8257 ‘-0.0318 -0.1932 -1.6007 0.9545 mm Rice Supply 0.6026 -0.0189 -0.1832 -0.4006 1.2128 Demand for: Seed 0.7692 -0.1265 -0.0558 ~0.5885 1.1957 Fertilizer 1.2217 -0.0091 -0.8467 -0.3659 1.2027 Labor 0.9166 -0.0330 -0.1256 -0.7581 1.1987 In general, the elasticity estimates from Cobb-Douglas profit function are much greater than corresponding estimates from the translog profit function, except for the elasticity with respect to fixed input (farm size). Negative cross-price elasticities from these two functional forms are consistent with theoretical expectation, which indicates a complementary nature of the production inputs. Positive own-price I48 elasticities are consistent with the behavioral assumption of profit maximization. The following discussion covers only the estimates from translog profit function. The own-price elasticity of rice supply is 0.6026, which is inelastic. This figure indicates that a one percent increase in price of rough rice, getgris airings, will increase rice supply by 0.6026 percent. The cross-price elasticities of rice supply with respect to prices of seed, fertilizer and labor’ wage are -0.0189, -0.1832 and -0.4006, respectively. Thus the own—price elasticity of output supply is more elastic than cross-price elasticities of output supply with respect to variable input prices. This always be the case for the well-behaved profit function, where the own-price elasticity of the output supply has to be equal to the summation of the cross-price elasticities (in absolute terms) of the corresponding output supply with respect to variable input prices. The significant difference (in absolute terms) between the own-price elasticity of rice supply (0.6026) and the cross-price elasticity of rice supply with respect to fertilizer (-0.1832) roughly indicates that if increasing total rice production is the government's primary concern, the rice price support policy will likely be more effective than fertilizer subsidy policy. This because a one percent increase in the price of rice will increase the rice supply by 0.6026 percent, while a one percent reduction (subsidy) in the fertilizer price will increase the rice supply by only 0.11832 percent. Note, however, that the former involves movement along the supply curve while the latter involves a shift of the supply curve. The own-price elasticities of'demand for seed, fertilizer and labor are -0.1265, -0.8467 and -0.7581, respectively. The cross-price elasticity of 149 demand for labor with respect to output price is much greater in absolute terms than corresponding elasticity'with respect to fertilizer price. This implies that price support policy will likely be more effective in promoting agricultural employment than the fertilizer price subsidy. This because a one percent increase in rice price will increase labor demand by 0.9166 percent, while a one percent reduction in fertilizer price will increase labor demand by only 0.1256 percent. The own-price elasticities of demand for seed, fertilizer and labor are -0.1265, -0.8467 and -0.7581, respectively. A one percent decrease in the prices of seed, fertilizer and labor wage, other things being constant, will increase the demands for seed, fertilizer and labor by 0.1265, 0.8467 and 0.7581 percent, respectively. Table 5.16 shows that the cross-price elasticities of labor'demand with respect to the prices of rice, seed, and fertilizer are 0.9166, -0.0330, and -0.1256, respectively. In absolute terms, the cross-price elasticity of labor demand with respect to rice price is much greater than corresponding elasticity with respect to fertilizer price. Again, with regard to the policy options in question, this roughly indicates that price support policy will likely be more effective in promoting agricultural employment than the fertilizer price subsidy. This because a one percent increase in rice price will increase labor demand by 0.9166 percent, while a one percent reduction in fertilizer price will increase labor demand by only 0.1256 percent. The elasticity of output supply with respect to land is 1.2128. Lau, Lin and Yotopoulos (1979) pointed out that this figure reflects the mutatis,mutand1§ effect of a change in the quantity of land, allowing the farm to adjust its output and variable inputs optimally. Hence, this 150 elasticity is not comparable to the production elasticity of land, which reflects the ggtgris parihus effects of a change in the size of land, holding the quantities of variable inputs constant. The mutatis,mutandis effect is greater than the 9.9.1.911: 1111111115 effect, and in fact was confirmed in this analysis. Table 5.17 presents the indirect estimates of the Cobb-Douglas production elasticities using the results in Table 5.14. The table shows that, except for the seed variable, the corresponding direct and indirect estimates are relatively close to each other. The indirect estimates in this table appear more reasonable compared to the corresponding indirect estimates presented in Table 5.9. Table 5.17. Direct and Indirect Elasticities of Cobb-Douglas Production Function Input Direct Estimatea Indirect Estimate 588d 0.1304 0.0174 Fertilizer 0.0944b 0.1058 Labor 0.2211 0.3290 Farm Size 0.4676 0.5229 a) This is the EGLS estimator from table (5.3) b) A simple average of LKGN and LKGP coefficients 5.5. Adoption of New Rice Varieties Using a multinomial logit model, this study attempted to identify the determinants of the farmers’ decisions in choosing rice varieties. This 151 objective is motivated by the fact that there were many farmers in the sample still growing TV, despite the fact that HYVs have long been promoted by the government. Hith regard to rice varieties, farmers face three alternatives: to grow TVs, HYVs, or both. The maximum likelihood estimation method was used to estimate the model on the pooled data (1026 observations). The coefficient estimates and the standard deviations of the model are presented in Table 5.18. The negative coefficient of TV price indicates that any increase in TV price will reduce the probability of farmers growing HYV relative to TV. However, the statistically non significant of this coefficient, which is unexpected, indicates that TV price is not an important determinant in farmers’ decisions on adopting HYV. The coefficient of HYV price is positive and is statistically significant at the 0.01 significance level, indicating that any increase in HYV price will likely increase the probability of farmers growing HYV relative to TV. This finding is consistent with theoretical expectation of producer behavior. A policy implication of this result is that adoption of' new ‘technology (HYV) can be accelerated by providing sufficient incentives for farmers, for instance by increasing the price of HYV. The price of fertilizer has a negative effect on the probability of farmers choosing HYV relative to TV, and is statistically significant at the 0.05 level. This indicates that an increase in fertilizer price will discourage farmers to grow HYV. This is consistent with the fact that HYV rice is more responsive to fertilizer application, as discussed in previous sections; an increase in fertilizer price will therefore reduce fertilizer application and in turn discourages farmers from growing HYV. 152 Table 5.18. Estimated Coefficients of the Multinomial Logit Model Independent Variables ln(P2/Pl) ln(P3/Pl) ln(P3/P2) Constant 0.5964 -5.8011 -6.3975 (0.8913) (1.8158) (2.1425) PTV -0.0018 0.0056 0.0074 (0.0061) (0.0120) (0.0181) PHYV 0.0376*** 0.0041 -0.0335 (0.0100) (0.0193) (0.0292) PF -0.0366** 0.0220 0.0586 (0.0136) (0.0270) (0.0405) RAGE -0.0094* 0.00003 0.0094 (0.0051) (0.0094) (0.0145) HA 1.0265*** 1.1712*** 0.1447 (0.1728) (0.2317) (0.3558) Variable definition can be found in chapter 4 and in appendix 5.4 Figures in parenthesis are standard deviations. *** statistically significant at o-0.01 ** statistically significant at o-0.05 * statistically significant at a-O.10 Labor wage has a negative effect on farmers’ decision to grow HYV relative to TV and is statistically significant at the 0.1 level; thus an increase in labor wage will discourage farmers from using HYV relative to TV. This could be explained by the fact that (even though the mean of the total labor use indicates no difference between HYV and TV), some crop management activities of the HYVs, especially weeding, are very labor intensive. It is well known that physical characteristics of HYV (which is much shorter in the height) is inferior compared to TV in terms of its ability to compete with weeds. 153 Farm size positively affects the probability of using HYV relative to TV, and is statistically significant at, the 0.01 significance level, indicating that larger farmers, other things being equal, are more likely to grow HYV. This is consistent with the notion that large farmers are commonly more responsive to new technologies. Sayogyo and Collier (1973), as cited in Gunawan (1988), using simple frequency distribution, found that the percentage of HYV adopters are much higher in the large farmer group compared to the small farmer group. The same conclusion was also made by Gunawan (1988). Coefficient regressions for the probability of farmers growing MV relative to TV are not statistically significant, with the exception of the intercept and land coefficient. Growing MV is a compromise choice, and is a typical strategy to minimize risks, before totally adopting HYV. Again, large farmers are more likely to grow MV relative to TV, other things being equal. Using equation (3.71) and (3.72) we can calculate the probability of each choice evaluated at the sample means for each explanatory variable, as presented in Table 5.19. Ignoring the non-significant coefficients, the probability of farmers with TV, HYV and MV are 0.5287, 0.4697 and 0.0016, respectively. These figures are significantly different from the varietal frequency distribution of the sample which are 0.6657, 0.2865 and 0.0478, respectively. The coefficient estimates presented in Table 5.18 do not represent a percentage change in probability for a percentage change in explanatory variables, thus they are not typical elasticity measures (as in the case of linear probability model). Using the procedure described in appendix (3.1), we can derive the response elasticities of the probability of 154 particular event with respect to changes in the explanatory variables, as presented in Table 5.20. Note that the calculation is carried out by setting the non-significant coefficients of Table 5.18 to zero. Table 5.19. Mean Probabilities of Varietal Choice Varietal Probabilities derived frgm Choice sample distribution Multinomial logit TV 0.6657 0.5287 HYV 0.2865 0.4697 NV 0.0478 0.0016 The interpretation of the elasticity figures is straight forward, and it is much simpler than the interpretation previously made based on figures on Table 5.18. The conclusion made, however, is consistent. A one percent increase in HYV price (ggtggig paribgg) will, on the one hand, reduce the probabilities for growing TV and MV by 5.19 percent and 1.45 percent, respectively, and on the other hand will increase the probability for growing HYV by 1.64 percent. A one percent increase in the price of fertilizer will increase the probabilities for choosing TV and MV by 4.87 percent and 1.36 percent, respectively, but reduce the probability for choosing HYV by 1.54 percent. Similarly, a one percent increase in labor wage will result in, respectively, 1.27 percent and 0.36 percent increase in the probabilities for growing TV and MV, but it will result in 0.40 percent reduction in the probability for growing HYV. Finally, a one percent increase in farm size 155 will, on the one hand, increase the probability of farmer choosing HYV and NV by 0.23 percent and 0.30 percent, and will on the other hand reduce the probability of farmer choosing TV by 0.75 percent. Table 5.20. Elasticities Derived from The Multinomial Logit Hodel. Independent Variables P1 P2 P3 PTV 0.0000 0.0000 0.0000 PHYV -5.1958 1.6395 -1.4523 PF 4.8686 -l.5363 1.3608 HAGE 1.2704 -0.4009 0.3551 HA -0.7474 0.2341 0.2966 Assessing these results for their policy implications, one can conclude that providing sufficient incentives is necessary if the adoption of HYV is an important priority. These incentives can be in 'the forms of reasonable input and output prices. This finding supports the notion of the induced innovation hypothesis that economic variables, including prices of inputs and output among others, are the main factors affecting the behavior of farmers in the adoption of new technologies. Thus, improved price support policy which allows the price of rice to gradually approach the real market price would appear to be a reasonable incentive to encourage farmers to adopt new HYVs. Of course there are many other factors, particularly non-economic factors, which may significantly influence farmers’ decisions in choosing 156 variety. It was reported that the significant rice damage due to brown planthgpng: (BPH) attack in 1979/80 induced farmers in Nest Java and other areas to adopt new HYV which was promoted to be more resistant to BPH biotype 1 and 2. The development and diffusion of rice varieties in Indonesia is well-documented in the article of Bernsten et al. (1982). To date, it is not uncommon that farmers are obligated to adopt new BPH resistant varieties as part of pest (insect) control strategy promoted by the government. 5.6. Policy Evaluation: Rice Price Support and Fertilizer Subsidy As described in chapter 2, increasing rice production and promoting employment in rural areas are two primary goals in agricultural development in Indonesia. To attain these goals, the Indonesian government subsidizes fertilizers and other production inputs but controls the price of rough rice. The domestic fertilizer prices are set well below their import parity prices (appendix 2.5). Moreover, the government maintains nominal fertilizer prices constant for several periods of time (appendix 2.1). As a result, the real (deflated) prices of fertilizers have decreased over time. Since 1971 the government has implemented floor and ceiling prices of rice. The floor price is set high enough to stimulate domestic production, while the ceiling price is set low enough to provide affordable rice to consumers and to contain the rate of inflation. The floor price is determined on the basis of an incremental benefit-cost ratio which results from farmer’s participation in the BIMAS program. Since the rate of increase in the nominal price of rice is usually below the general rate of inflation, the real (deflated) price of rice has been deteriorating. 157 Since pricing policies for both fertilizer and rice involve government’s budget or subsidies, the question is then which subsidy should be reduced in light of the recent government budgetary constraints. Note that, as mentioned in chapter 2, the current government’s policy concern is to phase out all kinds of subsidies. The question can also be raised differently as choosing the policy which can best achieve the specified policy goals at least cost to the government. The estimates of own- and cross-price elasticities presented above can be used to roughly evaluate the impact of the price support and fertilizer subsidy policies with respect to the above two policy goals. Recall, for example, that the cross-price elasticities of rice supply with respect to variable input indicate the percentage change of the rice supply due to a one percent change in any input prices. These figures, however, do not provide any information about which one of these two policies is more desirable, neither in terms producer and consumer welfare nor government budget. Table 5.21 illustrates the likely' performance of ‘these two policies in terms of increasing rice production and rural employment. This table shows that the price support policy should be more effective than fertilizer subsidy policy in attaining the objectives of increasing rice production and rural employment. 'The estimates from both Cobb-Douglas and Translog functional forms consistently confirm this conclusion. Table 5.21. The Impact of Rice Price Support and Fertilizer Subsidy Policies 158 Policy t on Scenario Seed Fertilizer Labor Rice Use Use Use Production WW 1% subsidy in 0.1932 1.1932 0.1932 0.1932 fertilizer price 1% support (increase) 1.8257 1.8257 1.8257 0.8257 in rice price W90 1% subsidy in 0.0558 0.8467 0.1256 0.1832 fertilizer price 1% support (increase) 0.7692 1.2217 0.9166 0.6026 in rice price Let us now evaluate these two policy alternatives by using the proposed procedure described in chapter 4. The procedure is based on the idea of comparing the two policies in terms of the additional government costs in generating an additional unit of rice and employment opportunity. The smaller the costs, the more desirable is the policy. The results of the calculation are presented in Table 5.22, while the detail is presented in appendix 5.9. The elasticity estimates from Cobb-Douglas and translog functional forms are used in the evaluation, and as expected, they both give consistent conclusions. 159 Table 5.22. Government’s Additional Cost for an Additional Unit of Rough Rice and Rural Employment Policy | Options Cobb-Douglas Translog WM Cost/Unit Rice (Rp/Kg) 12.26 16.80 Cost/Unit Employ.(Rp/Manday) 356.84 321.45 W Cost/Unit Rice (Rp/Kg) 27.50 29.03 Cost/Unit Employ.(Rp/Manday) 801.12 1232.30 Table 5.22 shows that if ‘the specified price support policy is implemented, the government’s additional cost for an additional unit (kg) of rice will be Rp 12.26 if estimated elasticities from Cobb-Douglas function are used, and Rp 16.80 if the translog estimates are used. These figures are much smaller compared to the corresponding figures derived from the implementation of the fertilizer subsidy policy, which are Rp 27.5 and Rp 29.03, respectively. Thus, as previously indicated, the price support policy is less costly than the fertilizer subsidy policy for as increasing rice production. Similarly the price support policy provides much cheaper’way to promote rural employment compared to the fertilizer subsidy policy. This because the government’s additional cost to generate one manday of employment is only Rp 356.84 if the Cobb-Douglas estimates are used and Rp 321.45 if the translog estimates are used, compared to the costs derived from the 160 implementation of the fertilizer subsidy policy, which are Rp 801.12 and Rp 1232.30, respectively. Moreover, price support policy may also have positive direct consequences for supporting diversified-food-consumption program in Indonesia to reduce people’s dependency on rice. Setting the price of rice too low as is currently done may (i) induce further dependencies on rice since it is affordable for all income levels, (ii) accelerate total rice demand due to population pressure, and (iii) turn Indonesia back to the largest rice importing country as. Thus, the implementation of improved price support policy, while at the same time gradually reducing input subsidies, could be one of the strategies to maintain rice self- sufficiency goal in the future. The conclusion made here, however, contradicts the conclusion made in Barker and Hayami (1976). Using rice in the Philippines as case and a benefit-cost analysis as the main tool of analysis, they attempted to evaluate rice support and input subsidy as policy alternatives to achieve food self-sufficiency in developing countries. The benefits and costs associated with these alternative programs and their income distribution effects are estimated and compared with the case of no programs. They concluded that in terms of the social benefit/cost ratio, fertilizer subsidy is more efficient than rice price support. Which one of these two policies is more desirable is debatable and the conclusion made is highly dependent on the country’s situation. Since the results of this study support price support policy, it would be worthwhile to consider the opposing arguments in order to see what likely problems created by this policy; problems particularly related to consumer and producer welfare, which are not incorporated in the evaluation procedure 161 of this study. To date, there are two opposing arguments regarding the price support policy. Economists generally accept that farmers are responsive to price incentives and that production will tend to increase when rewards are greater. Yet even the most ardent advocates of higher prices for agricultural producers would admit that many other things are necessary to call forth the extra supply. Among these is the removal of physical, social, and administrative barriers to increased supply (Streeten, 1987). Institutional arrangements which ensure that the benefits of the higher prices accrue to the farmers and not to monopolistic private middlemen or to inefficient public marketing authorities must be in place. New technology must always be available so that incentives of higher prices can significantly accelerate the growth of rice production. Proponents differ on whether raising prices without these other measures is better than nothing or whether raising prices, by itself, is futile or even counterproductive. The critics conlnonly point to cases where the introduction of higher supply prices without an appropriate scale-neutral technology or without the appropriate institutions has accelerated the transfer of land from small to large farmers and violated equity and poverty alleviation objectives (Streeten, 1987). If the productivity per hectare on the small farms was greater than on large farms, total output fell. Increased agricultural prices will in the long run lead to lower food prices than would otherwise have prevailed; the improved short-term incentives to invest, innovate, and adopt technical change will bring about a downward shift of the whole supply curve. This is perfectly possible and appears to have happened, but, according to Streeten (1987), 162 the following two qualifications are necessary. First, this very much depends on the initial land distribution and tenure system, on access to agricultural inputs, on institutions, on infrastructure, and on information. In societies with unequally distributed power and wealth, an increase in the food price may make rich richer by leading to transfers of land without leading to downward shift of the supply curve. A second qualification relates to the period of transition, in terms of the impact of food price increases on the impoverished food consumers. A rise in price of rice, for example, raises the real incomes of food producers and lowers, in the short run, the real income of the food consumers, since in the short run the supply of rice does not rise. In developing countries, such as Indonesia, there are many very poor people in both groups. In the medium and long run, the detrimental impact on the impoverished food consumers can be mitigated or offset by changing technology, increasing supply of food, increasing employment and perhaps reducing rural-urban migration which in turn increases urban income level. The very poor food consumer, however, may starve to death before the blessings of the medium and the long run materialize. Streeten identified three groups vulnerable to raising food prices: (i) landless laborers and deficit farmers, (ii) small surplus farmers, and (iii) the urban poor. Streeten, furthermore, pointed out three different alternatives to protect the above vulnerable groups. One way to protect these groups in the short run is to increase the price gradually. Very large price increases might even have perverse effects on supply with producers choosing to increase leisure time rather than increasing production. An alternative way is to make subsidies as selective and targeted as possible by concentrating them either on vulnerable groups or on basic food staples 163 consumed by the poor. A third way is to provide income subsidies to the poor. The importance of remunerative food prices as supply incentives is not questioned, although price policies are but one of'many policy instruments available to governments for expanding food production. In similar fashion, Pinstrup-Andersen (1987) argues that policies that attempt to strengthen incentives to expand food production through higher food prices may result in reduced real incomes and severe hardships for the poor, at least in the short run. Moreover, it is likely that the long-run effects are of little or no interest to the poor who are adversely affected in the short run, and uncertain future gains may be insufficient to compensate for immediate losses of the poor. The impact of food price increases on those poor who derive their income from food production would be expected to be positive provided the price increase is reflected in higher farmgate prices. Higher prices would add to revenues obtained from marketable surpluses, and labor demand for rice production would be expected to increase. However, total demand for rural labor need not increase if the food price increases cause the substitution of less labor-intensive for'more labor intensive commodities. Recent research indicates that food price increases may be much less favorable for the rural poor than is often expected. Many of the rural poor do not derive a large share of their income from either wage labor in food production or from the sale of food, and a large proportion are net purchasers of food. A study in Thailand conducted by Trairatvorakul (1984), as cited by Pinstrup-Andersen (1987), shows that the rural poor would not benefit greatly from increased domestic rice prices. Even though many of the rural poor are rice producers, their marketable surplus is I64 often small and large proportion are net buyers of rice. Trairatvorakul concludes that increasing rice prices would primarily benefit larger farmers and would create severe hardships among rural as well as urban poor. Increasing output prices without technological change in production, improved input markets, and better rural infrastructure may have little impact on total supply, although the commodity mix may change. Furthermore, because of market imperfections, only a relatively small part of the price increase in the final market may be transmitted to the producers. Technological change that reduces unit cost plays a particularly important role because it facilitates expanded food supply at equal or lower prices and employment creation in rural areas. Institutional changes may be needed to assure that food price increases are transmitted to the producers. In particular, improving marketing efficiency is necessary to simultaneously realize both higher producer and lower consumer prices. VI. SUMMARY AND RECOMMENDATION The role of the agricultural sector in the Indonesian economy, although declining, is still dominant, not only in terms of its contribution to national income but particularly in terms of its substantial contribution to labor absorption. In 1985, the agricultural sector accounted for about 24 percent of the GDP, and 54 percent of the total employment. Hithin the sector, the farm food subsector has the greatest contribution. It is not surprising if a high development priority has been given to the agricultural sector, especially the farm food subsector. In the past, the primary objective of agriculture policy'was to achieve production targets to meet an ever-increasing demand for rice resulting from population pressures and increased households’ income, and to promote employment opportunity in rural areas. Thus the policy target was very clear, that was to produce as much food as possible, regardless of what will be the problems after producing them. A great deal of effort and a large portion of the government budget were devoted to achieve the goals. Introduction of high yielding varieties of rice along with the rehabilitation, upgrading and expansion of irrigation system and provision of continuous supplies of subsidized inputs (fertilizers and pesticides) were, among others, the major contributors to the increase of rice production in (Indonesia. Investment in transportation and other infrastructure was also undertaken as part of the extensive development strategies. All these efforts resulted in obvious benefits during the late of 19605 up to the late of 1970s when the rate of growth of rice was as high as 8% annually. 165 166 The elevated production, however, did not persist over the long run and the rate of growth of the rice production has been declining in recent years. During the period of 1982-1985, for instance, the growth rate of rice production was 5% per year, and it declined to only 1.3% during period of 1985-1987. Given a steady annual population growth rate of 2.3%, and the continuing decline of rice production, it will not be too surprising if Indonesia returns to its former role as the world biggest rice importing country in the near future. The recent policy issue is a general movement towards a phasing out all subsidies. To date, subsidies are used as a major instrument in food price policies. Some productive inputs, such as fertilizers and pesticides, are highly subsidized in order to offset the production disinsentive effects of the rice subsidies to consumers. Recently, critics have questioned the ability of the Indonesian government to maintain the goal of long-run rice self-sufficiency, given the rapid growth of total rice demand in the one hand, and relatively stagnated growth of total rice production on the other hand. Some scientists consistently assert that Indonesia’s success in achieving the goal is at the expense of efficiency. This study aims to (i) generate and evaluate parameters of rice production system, especially those related to farm level efficiency, returns to scale, rice supply'and demand for inputs, (ii) identify factors affecting farmers’ decisions in the adoption of new rice varieties, and (ii) evaluate current agricultural policies, particularly price support and input subsidy policies. 167 6.1. Empirical Finding and Conclusion 6.1.1. Results of Descriptive Analysis 1. The average size of land cultivated with rice in the dry season is smaller than the average size in the wet season. This seasonal difference is statistically significant at 0.01 significance. level. Risk averse behavior (e.e. avoiding drought risks) and the lack of irrigation water due to inappropriate irrigation systems, could be the»main reasons for the seasonal difference in hectareage of rice. 2. The average yield per hectare is greater for the wet season than for the dry season. The result of analysis of variance verifies that seasonal difference in yield is statistically significant at 0.10 significance level. 3. The amount of seed used is relatively stable over time. The average figure does not show any consistent relationship between the amount of seed and season. The result of the analysis of variance, where the computed F-value is 0.92, supports the conclusion that there is no seasonal differences in the quantity of seed used. 4. The amount of fertilizer used per hectare tends to increase over time. On the average, farmers use more fertilizer in the wet season than in the dry season. The result of analysis of variance confirms the seasonal difference in fertilizer use, both for urea and phosphate, with computed F-statistic 3.78 (a-0.05) and 2.99 (a-0.08), respectively. This result indicates that seasonal differences in fertilizer application is more significant in urea than in TSP. 5. Seasonal difference in labor use is significant at 0.10 significance level. Hired labor is dominant, that is about 70-80% of the quantity of labor use per hectare is hired labor. 168 6. The average land of the HYV farmers is larger than the average land of the TV farmers, confirming the contention that large farmers are more responsive to new technology than small farmers. On the average, the HYV farmers use more urea than TV farmers, and as a result the HYV yield is significantly higher than the TV yield. The average quantities of seed, TSP and labor per hectare are not significantly different between these two groups. The results of the analysis of variance support all these conclusions. 6.1.2. Model Selection and Estimation methods 1. In a single equation estimation framework, the Cobb-Douglas functional form gives statistically more reliable parameter estimates than translog functional form. The possibility of high degree multicollinearity exists in the translog case, marked by large number of statistically non- significant parameter estimates while the computed F-statistic of the model is large. 2. Both statistical tests for fixed effect (FE) specification and random effect (RE) specification confirm the existence of individual specific effects representing the individual level of inefficiencies. 3. As far as the Cobb-Douglas functional form is concerned, the Mundlak test (standard F test) justifies the use of RE specification, since the test could not reject the null hypothesis of no correlation existing between the individual effects and the included exogenous variables. This test, therefore, ensures that the EGLS estimator is more efficient than the within estimator. The EGLS estimator in fact gives the best fit (i.e. the highest coefficient of multiple determination - R2) for both the production and the profit frontiers. 169 4. The results show little difference in magnitude between the OLS, the within and the EGLS estimators. The sign of each parameter estimate of the EGLS estimator is exactly the same as the sign of the corresponding estimate of the within and the OLS estimators. In fact, the magnitude of the EGLS estimator lies in-between the magnitude of the other two estimators. This confirms that the RE model is correct. 5. The own- and cross-price elasticities of output supply and input demand were derived from a system of equations of the profit and variable input share functions using seemingly unrelated regression estimation framework. In this case we turn back to the usual non-frontier type of function. Profit maximization is part of the maintained hypothesis in the parameter estimation. 6. Compared to the translog profit function, the Cobb-Douglas profit function gives, in general, greater (in absolute terms) elasticity estimates of the output supply and derived input demand function. In the Cobb-Douglas case, due to the well-known property of unitary elasticity of substitution among all input pairs, the impact across variable input demand functions for a given change in any of the exogenous variables is symmetry. On the other hand, the impact of a similar change in the case of translog varies across input demand equations, which is very consistent with 3 mm theoretical expectation. In the Cobb-Douglas case, moreover, the own-price elasticities of input demand for the well-behaved profit function, are definitely greater than one in absolute terms. Thus due to these weaknesses, the Cobb-Douglas functional form is considered to be inferior despite its many advantages including its simplicity. 7. Despite the differences in magnitude of the elasticity estimates, the results of ‘these two functional forms consistently support the 170 argument that price support policy is more effective in increasing production and promoting employment than fertilizer price subsidy. 6.1.3. Production Function 1. The dummy variable for season has a positive sign and is statistically significant at 0.05 significance level, indicating that the level of production for wet season is greater than that for the dry season. The lack of water availability during the dry season due to inappropriate irrigation facilities is the key reason for this seasonal yield difference, as reported in the survey. 2. The dummy variable for farm size (i.e. small or large farm category), is not significantly different from zero, indicating that there is no significant difference in the productivity level between small and large farms. In other words, small farms may or may not be more productive than large farms. 3. The dummy variables for varieties, HYV and NV, are statistically significant at 0.01 significant level, with positive signs. This strongly indicates that farms with HYV and NV produce more output than TV farms, which is consistent with a pziggi expectations. 4. The dummy variable for pesticide use is not significant, indicating no difference in production with or’without pesticide use. It was reported that during survey periods no significant crop damage due to insect attack or plant diseases occurred in the study area. 5. There is no statistically significant difference in the level of production between regions. This is not surprising given the fact that BIMAS/INMAS program have been intensively implemented in nearly all of the West Java province, particularly in the regions covered in the survey 171 which are major rice production areas. 6. The production elasticities with respect to seed, labor, land, urea and phosphate fertilizer are 0.1304, 0.2211, 0.4676, 0.1110 and 0.0778, respectively, and are statistically significant at 0.01 significance level. Thus, a one percent increase in the amount of each of these inputs, ggtgris.par1bg§, will increase the level of production by that percentage amount, respectively. The production elasticity with respect to urea fertilizer is greater than the elasticity with respect to phosphate fertilizer. This finding could be used to support the argument for differentiating the prices of these two fertilizers, if the government’s primary concern is to gradually reduce fertilizer subsidies. Up to now, however, the prices of these two are kept the same by the government. 7. Rice production in Nest Java is found to be in the stage of constant returns to scale, indicating that a one percent increase of all inputs will result in a corresponding one percent increase in the level of production. 8. The range of individual technical inefficiency is 3.4% - 12% with the mean 6.5%. These figures simply tell us that the rice farms in Nest Java are, on the average, 6.5% technically inefficient or 93.5% techni- cally efficient. Using rice yield for the 1983 dry season (4003 kg/ha) and the figure of total annual harvested area in Nest Java (1.74 million hectare), the estimate of yield loss was 260 kg per hectare, and the total quantity of production loss would be about 0.45 million tons annually. 9. The individual level of technical inefficiency is invariant to the choice of functional form. Both the Cobb-douglas and the translog give approximately the same mean and range of this measure, and the individual rank based on the level of inefficiency is exactly the same. 172 10. This study confirms that farmers in the study area are not able to optimally allocate the production inputs. There is a tendency that farmers underutilize both seed and fertilizers, but overutilize labor. 6.1.4. Profit Function 1. The dunlny variables for varieties, HYV and MV, of the profit function all have positive signs and are statistically significant, indicating that HYV and MV farmers are making more profits than the TV farmers. Interestingly, the majority of farmers in the study area were still using TV rice. Preference for growing TV could be justified given that (i) consumers prefer the taste of TV over HYV, and (ii) most of the rice produced is for own-consumption. 2. Unlike the production function case, the dummy variable for season is not significantly different from zero. The relatively higher rice price in the dry season (due to inelastic nature of rice supply function) may offset the reduction of total output, resulting in a nonsignificant difference between profit in the dry season and the wet season. 3. Dummy variables for regions all have positive signs, although only three of them are statistically significant, indicating that, compared to Nargabinangun (as a control), the profit earned by farmers in other regions is significantly higher. This finding is reasonable, since the other villages have relatively better product marketing channels due to better transportation facilities. The higher the price received, the higher the profit earned. 4. The individual profit inefficiency ranges from 6.9% to 28.9% with the mean 13.8%, indicating that on the average rice farms are 13.8%.profit inefficient or 86.2% profit efficient. Thus, on the average, 13.8% of 173 profits are foregone due to inefficiency. The results also show that individual level of profit inefficiency'does not have any association with individual farm size, meaning that large farms may or may not be more profit efficient than small farms. Using a per-hectare profit figure in the dry season 1983 (Rp 326,000/ha) and the total harvested areas in Nest Java per year (1.74 million hectare), a roughly estimate of per-hectare profit loss amounts to about Rp 45,000, while the total profit loss in Nest Java rice farms amounts to about Rp 78 billion annually, or about USS 81 million at the 1983 exchange rate (Rp970/USS). Thus, the benefits of promoting increased efficiency in rice farms in Indonesia appear to be very attractive. 6.1.5. Output Supply and Derived Input Demand 1. In general, the elasticity estimates of output supply and input demand from the Cobb-Douglas Profit function are much greater than corresponding estimates from the translog profit function. The negative signs of cross-price elasticities from these two functional forms are consistent with theoretical expectation, indicating a complementary nature of production inputs. Similarly, positive signs of own-price elasticity of supply and negative sign of own-price elasticities of input demand are consistent with a priori theoretical expectations of profit maximizing behavior. ' 2. The supply function of rough rice is found to be inelastic, as expected, with the own-price elasticity of 0.6026. This figure indicates that a one percent increase in the price of rough rice results in 0.6026 percent increase in rice supply. The cross-price elasticities of rice supply with respect to prices of seed, fertilizer and labor wage are - 174 0.0189, -0.1832 and -0.4006, respectively. Thus, the own-price elasticity is more elastic than cross-price elasticities of output supply with respect to input prices. 3. The own-price elasticities of demand for seed, fertilizer and labor are -0.1265, -0.8467 and -0.7581, respectively. The cross-price elasticities of fertilizer demand with respect to seed price, labor wage and rice price are -0.0091, -0.3659 and 1.2217, respectively. The cross- price elasticities of labor demand with respect to prices of seed, fertilizer and rough rice are, -0.0330, -0.1256, and 0.9166. Finally, the cross-price elasticities of seed demand with respect to fertilizer price, labor wage and rice price are -0.0558, -0.5885 and 0.7692, respectively. 4. The elasticity of output supply with respect to land (farm size) is 1.2128, indicating that a one percent increase in the size of land, ggugufig,mmt§ngi§, will result in 1.2128 percent increase in the quantity of output. 6.1.6. Multinomial Logit: Adoption of HYV I. The price of HYV positively affects the probability of farmers growing HYV relative to TV, and was found statistically significant at the 0.01 significance level. The higher the price of HYV, other things being equal, the more likely farmer to grow HYV. This is consistent with theoretical expectation. Even though statistically insignificant, the price of TV negatively affects the probability of farmers growing HYV relative to TV, meaning that the higher the price of TV the less likely farmer to grow HYV. ' 2. The price of fertilizer has a negative effect on the probability of farmers choosing HYV relative to TV, and is statistically significant at 175 0.05 significance level, indicating that an increase in fertilizer price, other things constant, will discourage farmers from growing HYV. This is consistent with the fact that HYV rice is more responsive to fertilizer application. An increase in fertilizer price will therefore reduce fertilizer application and this in turn discourages farmers from growing HYV. 3. Similarly, labor wages negatively affect farmer decisions on grow HYV relative to TV, and is statistically significant at the 0.1 level, indicating that an increase in labor wages will reduce the probability of farmers using HYV relative to TV. 4. Farm size positively affects the probability of using HYV relative to TV, and is statistically significant at 0.01 significance level, indicating that larger farmers, other things being equal, are more likely to grow HYV. This is consistent with the notion that larger farmers are usually more responsive to new technologies. . 5. All coefficient regressions of the probability of farmers growing NV relative to TV are not statistically significant, with the exception of the intercept and land coefficient. Large farmers are more likely to grow mixed varieties (NV) relative to TV, other things being equal. Growing MV is a likely compromise choice, is a typical strategy to minimize risks, before totally adopting HYV. 6. Ignoring the non-significant coefficients, the derived probability of farmers growing TV, HYV and NV are 0.5287, 0.4497 and 0.0016, respectively. These figures are different from the frequency distribution derived from the sample'which are 0.6657, 0.2865 and 0.0478, respectively. 7. A one percent increase in HYV price (tgtgrit paritgs) will, on the one hand, reduce the probabilities for growing TV and NV by 5.19 percent 176 and 1.45 percent, respectively, and on the other hand will increase the probability for growing HYV by 1.64 percent. 8. A one percent increase in the price of fertilizer will increase the probabilities for choosing TV and NV by 4.87 percent and 1.36 percent, respectively, but reduce the probability for choosing HYV by 1.54 percent. 9. Similarly, a one percent increase in labor wages will result in, respectively, 1.27 percent and 0.36 percent increase in the probabilities for growing TV and NV, but it will result in 0.40 percent reduction in the probability for growing HYV. 10. A one percent increase in farm size will, on the one hand, increase the probability of farmers growing HYV and NV by 0.23 percent and 0.30 percent, respectively, and will on the other hand reduce the probability of farmer growing TV by 0.75 percent. 11. Obviously there are many other factors, particularly non-economic factors, which may significantly influence farmers’ decisions in choosing rice varieties. Brown plantngppgr (BPH) and other insect attacks were apparently important factors in inducing farmers in Nest Java and other areas to adopt new HYVs which are more resistant to BPH biotype 1 and 2. 6.1.7. Evaluation of Rice Price Support and Fertilizer Subsidy 1. The estimates of own- and cross-price elasticities can be used to roughly evaluate the impact of the price support and fertilizer subsidy policies with respect to the above two policy goals. These estimates, however, do not provide any information about which one of these two policies is more desirable, neither in terms producer and consumer'welfare nor government budget. 2. The cross-price elasticity of demand for labor with respect to 177 output price is much greater in absolute terms than corresponding elasticity with respect to fertilizer price, implying that price support policy will likely be more effective in promoting agricultural employment than the fertilizer price subsidy. This because a one percent increase in rice price will increase labor demand by 0.9166 percent, while a one percent reduction in fertilizer price will increase labor demand by only 0.1256 percent. 3. In absolute terms, the cross-price elasticity of labor demand with respect to rice price is much greater than corresponding elasticity with respect to fertilizer price. Again, with regard to the policy options in question, this roughly indicates that price support policy will likely be more effective in promoting agricultural employment than the fertilizer price subsidy. This because a one percent increase in rice price will increase labor demand by 0.9166 percent, while a one percent reduction in fertilizer price will increase labor demand by only 0.1256 percent. 4. The results of price and subsidy policy evaluation indicate that rice price support is less costly than fertilizer subsidy policy in increasing total rice production and rural employment. If rice price support policy is implemented, the government’s additional cost for an additional kilogram of rice will be Rp 12.26 if the Cobb-Douglas elasticity estimates are used, and Rp 16.80 if the translog estimates are used. These figures are much smaller compared to the corresponding figures derived from the implementation of the fertilizer subsidy policy, which are Rp 27.5 and Rp 29.03, respectively. 5. Similarly the price support policy provides much cheaper way to promote rural employment compared to the fertilizer subsidy policy. This because the government’s additional cost to generate one manday of 178 employment is only Rp 356.84 if the Cobb-Douglas estimates are used and Rp 321.45 if the translog estimates are used, compared to the costs derived from the implementation of the fertilizer subsidy policy, which are Rp 801.12 and Rp 1232.30, respectively. 6.2. Policy Implication This study confirms that a yield increasing technology is sill the best choice in the future, for the following reasons: (i) relatively small land holdings and land intensive nature of Indonesian rice farming in general, and Nest Java in particular, (ii) objective of increasing total rice production to meet accelerating growth for rice demand, and (iii) objective of increasing farmer’s income. However, given the fact that varietal quality and taste are also important part of farmers’ considerations in choosing varieties, efforts have to be directed to finding new HYVs with better taste and quality. Other criteria such as resistance levels to insect attacks and diseases, must also be considered. Improving irrigation systems to control and guarantee optimal irrigation water availability throughout the year, appears essential and is confirmed by the results of this study. In the very near future, given government’s budget limitations, attention should be given to the improvement of the existing irrigation systems, particularly those categorized as local or simple irrigation systems. Farmers’ skills, motivation and willingness are the key factors for making any policy succeed. Continuing efforts in improving extension services in order to motivate and improve farmer’s management practices are important, particularly given the fact that the potential benefits of promoting increased efficiency both in terms of total rice production and 179 total income are very high. To guarantee improvement in farmer’s income, in addition to the improvement in yield through HYV and improved irrigation, the adoption of new rice technology will also require an improvement of other services (e.g., a.more efficient supply of inputs and farm credit) and reasonable economic incentives for farmers. This study verifies that, despite the input production incentives they receive, farmer’s real income is in fact deteriorating over time. Assuming that the Cobb-Douglas production function used in this study properly represents the true response curve of the existing technology, the results confirm that the rice production is in the stage of constant returns to scale. If this is true, natural land consolidation which has gradually come into existence in some places, would not have any effect in increasing total rice production, despite some advantages related to post-harvest activities. Considering (i) direct consequences of land consolidation, e.g., increasing number of landless, and (ii) HYV is in fact not labor intensive technology, as verified in this study, attention must be given to the potential social problems which might occur in rural areas due to unemployment. Nith regard to the government budget constraints and the objectives of increasing rice production and rural employment, this study verifies that price support policy is more desirable than fertilizer subsidy policy. The price support is less costly than fertilizer subsidy policy for attaining those objectives. Moreover, this policy will encourage farmers to adopt HYV and therefore to increase total rice production. Price support policy may also have positive direct consequences for supporting diversified- food-consumption program in Indonesia. Setting the price of rice too low as in the case now, may have further consequences as follows: (i) induces 180 consumers to further dependent on rice since it is affordable for almost all income levels, (ii) accelerates the total rice demand due to population pressure, and (iii) eventually forces the country back to its former role as the world's largest rice importing country. Thus, the implementation of improved price support policy, coupled with a gradual reduction in input subsidies, may be one policy alternative to maintain the long-run rice self-sufficiency goal. The short-run detrimental impact of the price support policy on poor rice consumers must be carefully considered. In the medium and long run, this detrimental impact may be offset by changing technology, increasing supply of rice and increasing employment opportunity in rural areas. The poor rice consumers, including landless laborers and deficit farmers, may starve to death before the blessings of the medium and the long run materialize. One way to protect these groups in the short run is to increase the price of rice gradually. Increasing the price of rice without technological change in the rice production, improved input markets, and better rural infrastructure may have little impact on the total rice supply. Furthermore, because of market imperfections, only a relatively small part of the price increases may be transmitted to the rice producers. Institutional changes are needed to assure that price increases are transmitted to the producers. In particular, improving marketing efficiency is necessary to simultaneously realize both higher producer and lower consumer prices. 5.3. Recommendation for Further Research This study was not able to identify factors affecting individual level of technical as well as profit inefficiency, due to lack. of' data 181 availability; Socioeconomic factors such as education, farming experience, tenancy, off-farm employment, involvement in any intensification programs, may affect individual level of inefficiencies. Knowing these factors, we can simply regress these factors as independent variables against the estimated individual level of technical (profit) inefficiency as the dependent variable. The panel data set used in this study covers wetland and dryland farming, both rice and secondary crops. It would be interesting to assess the inefficiency level of the dryland rice and secondary crop farming practices, using a similar approach used in this study. This approach can also be extended to handle a multiproduct production as well as multiproduct profit function. The multiproduct framework may best represent the real farming activities. The impact of price support and input subsidy policies on consumer and producer surplus is not analyzed in this study. Future research is needed to assess the welfare impact of the policy alternatives on both producers and consumers, in addition to the government cost. BIBLIOGRAPHY BIBLIOGRAPHY Aigner, D.J. and Chu, S.F. 1968. On Estimating the Industry Production Function. American Economic Review 58: 826 - 839. Aigner, D.J. C.A.K. Lovell, and P. Schmidt. 1977. Formulation and Estimation of Stochastic Frontier Production Function Models. Journal of Econometrics 6: 21 - 37. Ali, N. and J.C. Flinn. 1989. Profit Efficiency Among Basmati Rice Producers in Pakistan Punjab. American Journal of Agricultural Economics. Vol.2, No.2: 303 - 310. Agro-Economic Survey (Survey'Agro Ekonomi). 1976. Deskripsi Keadaan Sosial Ekonomi Desa Contoh: Nargabinangun, Lanjan, Gunungwangi, Malausma, Sukaambit, dan Ciwangi. SAE, Indonesia. Avery, R. B. 1977. Error Components and Seemingly Unrelated Regressions. Econometrica, 45: 199 - 209. Baltagi, B.H. 1980. In Seemingly Unrelated Regressions with Error Components. Econometrica, 48: 1547 - 1551. Barker, R. and Y. Hayami. 1976. Price Support versus Input Subsidy for Food Self-Sufficiency in Developing Countries. Am. J. Agr. Econ., 58: 617 - 628. Battese, G.E. and T.J. Coelli. 1988. Prediction of Firm-Level Technical Efficiencies Nith A Generalized Frontier Production Function and Panel Data. Journal of Econometrics, 38: 387 - 399. Bernsten, R.H, B.H. Siwi, and H.H. Beachell. 1982. The development and Diffusion of Rice Varieties in Indonesia. IRRI Research Paper Series no. 71, Los Banos, Laguna, Philippines. Breusch, T.S. and A.R. Pagan. 1980. The Langrange Multiplier Test and Its Applications to Model Specification in Econometrics. Review of Economic Studies, 47: 239 - 253. Central Bureau Statistics. 1976; 1980; 1985; 1986; Statistical Year Book of Indonesia. Jakarta, Indonesia. Central Bureau Statistics. 1983. Monthly Statistical Bulletin: Indicator Ekonomi, December 1983. Jakarta, Indonesia. Christensen, L.R, D.N. Jorgenson, and L.J. Lau. 1973. Transcendental Logarithmic Production Frontiers. Review Economics and Statistics. 55: 28 - 45. Deptan/Deperin/APPI. 1984. Study Pupuk National 11 (National Fertilizer Study II): Annexes I - IV. Kerjasama antara Departeman Pertanian, Perindustrian dan Assosiasi Pengusaha Pupuk Indonesia (APPI). 182 183 Diewert, N.E. 1974. Application of Duality Theory. In M. Intrilligator and D. A. Kendrick, eds. Frontiers of Quantitative Economics. Vol.II. Amsterdam, North Holland. Farrel, M.J. 1957. The measurement of productive efficiency. Journal of the Royal Statistical Society, 70: 253 - 281. Fuller, N. A. and G.E. Battese. 1973. Transformations for Estimation of Linear Models with Nested Error Structure. Journal of the American Statistical Association, 68: 626 - 632. Gittinger, J.P., J. Leslie, and C. Hoisington. 1987. Food Policy: Integrating Supply, Distribution, and Consumption. EDI Series in Economic Development. The John Hopkin Univ. Press, Baltimore, Maryland, USA. 4 Greene, N. H. 1980. Maximum Likelihood Estimation of Econometric Frontier Functions. Journal of Econometrics, 13 : 27 - 56. Gunawan, M. 1988. Adoption and Bias of'New.Agricultural Innovation in Jawa Barat, Indonesia. Unpublished Ph.D thesis, Univ. of Minnesota, St.Paul, Minnesota. Hausman, J.A. 1978. Specification test in Econometrics. Econometrica, 46: 1251 -1272. 'Hayami, Y., R. Barker and E. Bennagen. 1976. Price Incentive versus Irrigation Investment to .Achieve Food Self-Sufficiency in the Philippines. Am. J. Agr. Econ., 59: 717 - 721. Hoch, I. 1962. Estimation of Production Function Parameters Combining Time-Series and Cross-Section Data. Econometrica, Vol.30, no.1: 34 - 53. Hsiao, Cheng. 1986. Analysis of Panel Data. Cambridge Univ.Press. Hutabarat, B. 1985. An Assessment of Farm-Level Input Demands and Production Under Risk on Rice Farms in Cimanuk River Basin, Jawa Barat, Indonesia. Unpublished Ph.D thesis, Iowa State Univ. Ames, Iowa. , IFPRI/CAER. 1986. An Assessment of Food Demand/Supply Prospects and Related Strategies for Indonesia. Preliminary Draft of Final Report. International Food Policy Research Institute, Nashington, D.C., and ll’hg Center for Agro Economic Research, Ministry of Agriculture, n ones a. Jondrow, J, C.A.l(. Lovell, I.S. Materov, and P.Schmidt. 1982. On The Estimation of Technical Inefficiency in the Stochastic Frontier Production Function Model. Journal of Econometrics, 19: 233 - 238. 184 Judge, G.E, R.C. Hill, N.E. Griffiths, H. Lutkepohl, and T-C Lee. 1982. Introduction to the Theory and Practice of Econometrics. John Niley and Sons, N.Y, USA. Judge, G.E, N.E. Griffiths, R.C. Hill, H. Lutkfiaohl, and T-C Lee. 1985. The Theory and Practice of Econometrics. 2 edition. John Niley and Sons, N.Y, USA. Kalirajan, K. 1985. On Measuring Absolute Technical and Allocative Efficiencies. The Indian Journal of Statistics, vol.47, series B, Pt.3: 385 - 400. Kasryno, F. 1986. Supply of Rice and Demand for Fertilizer for Rice Farming in Indonesia. Jurnal Agroekonomi (JAE), vol.5, no.2: 27 - 42. Khumbakar, S.C. 1986. Estimation of Inefficiency in Production: Methodological and Empirical Issues. Unpublished Ph.D thesis, Univ. of Southern California, Los Angeles, California. Khumbakar, S.C. 1987. Production Frontiers and Panel Data: An Application to U.S. Class 1 Railroads. Journal of Business and Economic Statistics, vol.5, no.2: 249 - 255. Kmenta, J. 1986. Elements of Econometrics. 2nd Edition. Macmillan Publishing Company, New York, USA. Kuroda, Y. 1979. Production Behavior of the Japanese Farm Household in the Mid-1960s. In Yotopoulos, P.A. and L.J. Lau; Resource Use in Agriculture: Application of the Profit Function to Selected Countries. Food Research Institute Studies, vol 17, no.1, 1979. Lau, L.J. and PuA.Yotopoulos. 1974. Profit, Supply, and Factor Demand Functions. Amer. J. Agr. Econ., Feb. 1974, 54: 11 - 18. Lau, L.J., N.L. Lin and P.A. Yotopoulos. 1979. Efficiency and Technological Change in'Taiwan’s Agriculture. In Yotopoulos, P.A and L.J. Lau; Resource Use in Agriculture: Application of the Profit Function to Selected Countries. Food Research Institute Studies, vol 17, no.1, 1979. Lau, L.J., R.H. Myers, and E.C. Chou. 1979. Economic Efficiency of Agricultural Production in Northeast China, Circa 1940. In Yotopoulos, P.A and L.J. Lau; Resource Use in Agriculture: Application .of the Profit Function to Selected Countries. Food Research Institute Studies, vol 17, no.1, 1979. Maddala, 0.5. 1971. The Use of Variance Components Models in Pooling Cross-Section and Time-Series Data. Econometrica, vol.39, no.2: 341 - 358. 185 Meeusen, N. and J van den Broeck. 1977. Efficiency Estimation From Cobb- Douglas Production Functions Nith Composed Error. International Economic Review, vol.18, no.2: 435-444. Mundlak, Y. 1978. On the Pooling of Time Series and Cross Section Data. Econometrica, 46: 69 - 85. Nestel, B. 1985. Indonesia and the CGIAR Centers: A Study of Their Collaboration in Agricultural Research. Consultative Group on International Agricultural Research (CGIAR) Study Paper No.10. Pinstrup-Andersen, Per. 1987. Food Prices and the Poor in Developing Counties. In Gittinger, J.P., J. Leslie and C. Hoisington. Eds; Food Policy: Integrating Supply, Distribution, and Consumption. EDI Series in Economic Development. The (John Hopkins Univ. Press, Baltimore, USA. Pitt, H.H. and L.F. Lee. 1981. The Measurement and Sources of Technical Inefficiency in The Indonesian Neaving Industry. Journal of Development Economics, 9: 43 - 64. Richmond, J. 1974. Estimating The Efficiency of Production. International Economic Review, vol.15, no.2: 515 - 521. Schmidt, P. and C.A. Knox Lovell. 1979. Estimating Technical and Allocative Inefficiency Relative to Stochastic Production and Cost Frontiers. Journal of Econometrics, 9: 343 - 366. Schmidt, P. and C. A. Knox Lovell. 1980. Estimating Stochastic Production and Cost Frontiers Nhen Technical and Allocative Inefficiency are Correlated. Journal of Econometrics,13: 83 - 100. Schmidt, P. and R.C. Sickles. 1984. Production Frontiers and Panel Data. Journal of Business and Economic Statistics, vol.2, no.4: 367 - 374. Seale, J.L. 1985. Fixed Effect Cobb-Douglas Production Functions For Floor Tile Firms, Fayoum and Kalyubiya, Egypt, 1981-1983. Unpublished Dissertation, Michigan State University. Sidhu, S.S. 1974. Relative Efficiency in Nheat Production in the Indian Punjab. Amer. Econ.Review, v0l.64: 742 - 751. .Sidhu, S.S. and C.A.Baanante.1981. Estimating Farm-Level Input Demand and Nheat Supply in Indian Punjab Using A Translog Profit Function. Amer.J. Agr.Econ., vol. 63, no.2: 237 - 246. Somel, K. 1979. Technological Change in Dryland Nheat Production in Turkey. In Yotopoulos, P.A. and L.J. Lau; Resource Use in Agriculture: Application of the Profit Function to Selected Countries. Food Research Institute Studies, vol 17, no.1, 1979. 186 Sugianto, T. 1982. The Relative Economic Efficiency of Irrigated Rice Farms, Nest Java, Indonesia. Unpublished Ph.D thesis, Univ. of Illinois, Urbana- Champaign, Illinois. Tabor, S.R. et.al. 1988. Supply and Demand for Food Crops in Indonesia. Directorate General of Food Crops, Ministry of Agriculture, Indonesia. Tabor, S.R. 1988. Price and Quality of Rice in Java: An Investigation into the Demand for Closely Related Goods. Directorate General of Food Crops, Ministry of Agriculture, Indonesia. Tamin, M. 1979. Microeconomic Analysis of Production Behavior of Malayan Farms: Lessons from Muda. In Yotopoulos, P.A and L.J. Lau; Resource Use in Agriculture: Application of the Profit Function to Selected Countries. Food Research Institute Studies, vol 17, no.1, 1979. Timmer, C.P. 1971. Using A Probabilistic Frontier Production Function to Measure Technical Efficiency. Journal of Political Economy, vol.79, no.4: 776 - 794. Timmer, C.P. 1975. The Political Economy of Rice in Asia: Indonesia Food Research Institute Studies, 14: 197 - 231. Timmer, C.P., N.P. Falcon, and S.R. Pearson. 1983. Food Policy Analysis. A world Bank Publication. The John Hopkins Univ. Press, Baltimore, Maryland, USA. Timmer, P. 1986. Getting Prices Right: The Scope and Limits of Agricultural Price Policy. Cornell Univ. Press, Ithaca, N.Y. Timmer, C.P. 1987. Price Policy and the Political Economy of Markets. In Gittinger, J.P., J. Leslie and C. Hoisington. Eds; Food Policy: Integrating Supply, Distribution, and Consumption. EDI Series in Economic Development. The John Hopkins Univ. Press, Baltimore, USA. Yotopoulos, P.A. and L. J. Lau. 1973. A Test for Relative Economic Efficiency: Some Further Results. Amer. Econ. Rev., vol.63, no.1: 214 - 223. ' Yotopoulos, P.A. and L. J. Lau. 1979. Resource Use in Agriculture: Applications of the Profit Function to Selected Countries. Food Research Institute Studies, vol.17, no.1. Zellner, A.1962. An Efficient Method for Estimating Seemingly Unrelated Regressions and Test for*Aggregation Bias. J. Amer. Stat. Ass., June 1962, 57: 348 - 368. Zellner, A, J. Kmenta and J. Dreze. 1966. Specification and Estimation of Cobb-Douglas Production Function Models. Econometrica, vol .34, no.4: 784 - 795. APPENDICES 187 Appendix 2.1. The BIMAS Package per Hectare, 1980 Items _TathgkaggleE— -—Tagg%kag%§$03— T3E2131933753— (Rp) (Rp) (Rp) Urea (kg) 7 200 14000 100 7000 250 17500 TSP (kg) 50 3500 35 2450 75 5250 KCL/KZO (kg) 50 3500 50 3500 50 3500 Insecticide (It) 2 2460 2 2460 2 2460 Rodenticide (gr) 100 400 100 400 100 400 Seed - 5000 - - - 5000 Spraying Cost - 2000 - 2000 - 2000 Additional Cost - 10000 - 10000 - 10000 Total 40860 27810 46110 Source: Adopted from Sugianto (1982) 188 Appendix 2.2. Fertilizer Price Under BIMAS and INMAS Program fertilizer Rough Rice Rice Floor Rough Rice Year Urea TSP Floor Price Price -Urea price (Rp/k9) Rp/k9) (Rp/k9) (Rp/k9) Ratio 1974 40.00 40.00 41.80 68.50 1.05 1975 60.00 60.00 58.50 97.00 0.98 1976 80.00 80.00 68.50 108.00 0.86 1977 70.00 70.00 71.00 110.00 1.01 1978 70.00 70.00 75.00 119.50 1.07 1979 70 00 70 00 85.00 140.00 1.21 1980 70.00 70.00 105.00 175.00 1.50 1981 70.00 70.00 120.00 195.00 1.71 1982 90.00 90.00 135.00 214.00 1.93 1983 90.00 90.00 145.00 238.00 1.61 1984 90.00 90.00 165.00 270.00 1.83 1985 100.00 100.00 175.00 285.00 1.75 Source: IFPRI/CAER, 1986. 189 Appendix 2.3. Trends in Imported and Actual Rice Prices in Jakarta (USS per ton). imported Bite Act. Jakarta Year FOB Bangkok Cost to Retail Price (25% broken) Jakarta 1970 125.3 148.6 112.4 1971 93.9 115.5 109.3 1972 103.6 127.5 119.0 1973 116.3 175.8 205.2 1974 493.2 558.7 242.2 1975 311.8 380.5 262.7 1976 222.3 263.3 209.6 1977 237.4 287.3 319.6 1978 327.9 382.2 318.8 1979 308.3 362.0 272.5 1980 403.9 466.4 319.0 1981 416.4 470.1 325.0 1982 271.6 320.9 348.0 Source: Norld Bank 1982 (adopted from Nestel, 1985) Exchange Rate at Rp 660 - U55 1 190 Appendix 2.4. Price Structure For Urea and Triple Superphosphate (TSP) in 1982 Fertilizer Price USS/Ton Rp/Kg Urea: FOB Europe 185 Ex-Factory Price (Palembang) 198 Handling/distrib. to Retail. +40 Transport to Farm 4 4 Farm-gate Price 242 160 Financial Farm-gate Price 106 70 TSP: FOB Florida 185 Ocean Freight 8 Insurance +60 Handling/distrib. to Retail. +35 Transport to Farm + 4 Farm-gate Price 259 171 Financial Farm-gate Price 106 70 Source: Nestel (1985) Exchange Rate at Rp 660 - USS 1 191 Appendix 2.5. Budget Cost of Fertilizer Subsidy (1981/1982) Quantity ________§gh§idy (000 Ton) Rp.Billion USS Million Domestically Produced: Urea 1,758 76,289 115.6 TSP 487 93,445 141.6 Ammonium Sulphate 120 8,594 13.0 Subtotal 23195 1253328 21932 Imported: Urea 200 27,633 41.9 TSP 150 20,724 31.4 Ammonium Sulphate 100 10,296 15.6 Potassium Chloride 50 6,824 10.3 Subtotal 500 55.411 2252 Total 2,865 243,805 369.4 Source: Ministry of Finance (Adopted from Nestel, 1985) Exchange Rate at Rp 660 . USS 1 192 Appendix 3.1. Derivation of Response Elasticities of the Multinomial Logit Model. Probability of each event (choice) is Pl ' 1 + 21 exp(XBJ) exp(Xfi, ) 1 + SJ exp(XBJ) In our case subscript 1 represents an event of farmer growing TV, while i and j - 2, 3, represent HYV and MV, respectively. Let exp(XBj) - Zj and I + 23 exp(Xfij) - Y. Taking a partial derivative of P with respect to kth explanatory variables, we get pjk 21 Y ‘ (fljk 21 + flik ZI) ZJ aPJ/axk Y2 flJk 23 Y ' ka ZJ ZJ ' flik 21 ll Y2 fijk 23 Y - fijk 21 23 - flik Zj ZI v2 v2 v2 (”1k Pi ‘ 53k P1 P3 ' 5ik Pi Pi) 193 Appendix 3.1. (Continued) 8 z p z aPI/axk _ - 2k 2 _ 3k 3 Y2 Y2 ' ‘ 52k (Pg/P1) ‘ 33k (P3/P1) where a is a symbol of partial derivative. The response elasticities (Eik) can be calculated from Elk - (dPl/dxk)(Xk/Pl) EZk - (dPZ/dxk)(Xk/P2) 53k - (dP3/ka)(Xk/P3) evaluated at the mean value of P and Xk. 194 Appendix 5.1. Input and Output Per Hectare of High Yielding Rice Variety (HYV) Input and Sea;gn_ifl:ugt, D-dryl Output N75/76 D76 N76/77 D77 N82/83 083 Land (Ha) 0.6696 0.6213 0.7573 0.5007 0.6431 0.5160 Output (Kg/Ha) 3619.4 2867.7 2370.6 2536.4 4821.9 4235.3 Seed (Kg/Ha) 42.8 45.6 43.5 37.9 46.0 36.8 Urea (kg/Ha) 225.2 181.0 240.2 '212.0 257.7 244.4 TSP (kg/Ha) 47.8 57.4 58.1 56.3 120.5 108.2 Labor (hrs/ha) 1538.2 1315.7 1321.2 1221.8 1424.9 1356.0 Hired preharvst 524.7 529.7 607.7 533.8 507.6 474.4 Hired harvest* 767.4 605.5 472.5 486.4 581.3 578.7 Family 246.1 180.5 240.9 201.7 335.9 302.9 Price (Rp/Kg) Rough Rice 58.28 59.32 61.30 66.86 111.00 136.56 Seed 81.41 77.16 (76.29 104.02 217.16 212.80 Urea 77.31 77.03 70.91 70.65 84.86 88.15 TSP 77.31 77.04 70.67 71.09 86.12 89.23 Nage 45.43 46.25 46.37 51.48 141.21 136.72 Revenue (Rp/Ha) 210940 170112 145318 169584 535231 578372 Var.Cost (Rp/Ha) 94470 86734 85721 85820 243445 224422 Profit (Rp/Ha) 116470 83378 59597 83764 291786 353950 N (Observations) 32 37 43 44 65 74 Variety (TV) 195 Appendix 5.2. Input and Output Per Hectare of Traditional Rice Input and Sea;gn_ifl=wet. Dydryl Output N75/76 D76 N76/77 D77 N82/83 083 Land (Ha) 0.3793 0.3063 0.3508 0.3025 0.3796 0.3265 Output (Kg/Ha) 3220.3 3236.0 2633.9 2255.2 3461.5 3677.0 Seed (Kg/Ha) 35.4 38.6 38.5 39.0 38.9 39.7 Urea (kg/Ha) 195.5 192.2 205.1 195.3 262.7 248.5 TSP (kg/Ha) 76.5 68.1 61.0 57.9 111.4 110.7 Labor (hrs/ha) 1364.6 1409.4 1202.1 1289.0 1377.3 1547.4 Hired preharvst 610.3 610.2 396.1 555.8 597.1 639.4 Hired harvest* 396.6 418.4 292.3 295.3 297.8 336.0 Family 357.7 380.7 413.7 437.9 482.3 572.0 Price (Rp/Kg) Rough Rice 63.70 69.21 66.56 77.97 138.87 136.67 Seed 65.05 70.66 74.54 89.46 150.91 155.40 Urea 77.86 77.62 70.80 71.02 89.12 89.13 TSP 78.32 76.88 70.44 71.39 91.00 92.22 Nage 50.60 49.18 58.93 52.20 136.91 130.19 Revenue (Rp/Ha) 205133 223964 175312 175838 480698 502536 Var.Cost (Rp/Ha) 92565 92584 92528 88778 227986 239983 Profit (Rp/Ha) 112568 131099 82784 87060 252712 262553 N (Observations) 131 129 125 119 91 87 196 Appendix 5.3. Input and Output Per Hectare of Mixed Varieties (NV) Input and Sea;gn_ifl=ugt. D-dry) Output N75/76 D76 N76/77 D77 N82/83 083 Land (Ha) 1.9139 0.5365 0.3140 0.4148 0.4050 0.3832 Output (Kg/Ha) 2640.9 3007.9 3371.9 2330.0 4076.4 4104.4 Seed (Kg/Ha) 48.5 45.7 56.1 33.2 36.1 40.2 Urea (kg/Ha) 147.7 163.6 233.6 152.4 254.7 252.1 TSP (kg/Ha) 27.5 47.5 86.8 32.7 161.3 157.1 Labor (hrs/ha) 1347.6 1588.2 1252.3 1283.7 1647.5 1665.7 Hired preharvst 664.2 652.7 385.6 466.1 739.4 553.8 Hired harvest* 509.2 650.7 436.6 488.1 336.0 473.2 Family 174.3 284.9 387.8 329.5 572.0 638.8 Price (Rp/Kg) Rough Rice 61.25 63.73 63.50 66.67 126.67 141.08 Seed 78.75 69.25 75.00 105.50 155.40 170.40 Urea 79.00 74.50 70.00 70.78 89.13 91.00 TSP 79.50 76.50 70.00 70.78 90.27 91.00 Nage 47.76 46.69 48.30 51.57 143.06 132.66 Revenue (Rp/Ha) 161755 191695 214116 155341 516358 579046 Var.Cost (Rp/Ha) 82035 93140 87122 82804 278536 265059 Profit (Rp/Ha) 79720 98555 126994 72537 237822 313987 N (Observations) 8 5 3 9 15 10 197 Appendix 5.4. Abbreviation of the Variables .. , ,1 - .._ ., 1,. 4 . , ..f- LKGS: natural logarithm of kilograms of seed used. LKGN: Natural Log. of kilograms of urea used. LKGP: Natural Log. of kilograms of TSP used LLAB: Natural log. of hours of laborers used. LHA : Natural Log. of hectares of cultivated land. LKGSKGS - LKGS*LKGS LKGNKGN - LKGN*LKGN LKGPKGP - LKGP*LKGP LLABLAB - LLAB*LLAB LKGSKGN - LKGS*LKGN LKGSKGP - LKGS*LKGP LKGSLAB - LKGS*LLAB LKGNKGP - LKGN*LKGP LKGNLAB - LKGN*LLAB LKGPLAB - LKGP*LLAB LHALHA - LHA*LHA LHAKGS - LHA*LKGS LHAKGN - LHA*LKGN LHAKGP - LHA*LKGP LHALAB - LHA*LLAB LPS: natural Logarithm of per kilogram seed price normalized by price of rice. LPF: natural logarithm of per kilogram fertilizer price normalized by per kilogram price of rice. 198 Appendix 5.4. (Continued) LNG: natural logarithm of per hour labor'wage normalized by price of rice. LPSLPS - LPS*LPS LPFLPF - LPF*LPF LNGLHG - LNG*LNG LPSLPF - LPS*LPF LPSLNG - LPS*LNG LPFLNG - LPF*LNG LHALPS - LHA*LPS LHALPF - LHA*LPF LHALNG - LHA*LNG DP: dummy variable of pesticide use, equals 1 if farmer uses pesticides and equals 0 otherwise. 0V1: dummy HYV variety, equals 1 if HYV, zero otherwise 0V2 : dummy of Mixed Varieties (MV), equals 1 if NV, zero otherwise. Ngtg; traditional variety (TV) is the control. 055: dummy variable of season, equals 1 if wet season, zero otherwise. DSIZE: dummy variable of farm size, equals 1 if farm size greater than 0.5 ha, zero otherwise. DRI: duimny village, equals 1 if desa Lanjan kabupaten Indramayu, zero otherwise. 0R2: dummy village, equals 1 if desa Gunung Nangi kabupaten Majalengka, zero otherwise. 0R3: dummy village, equals 1 if desa Malausma kabupaten Majalengka, zero otherwise. 199 Appendix 5.4. (Continued) 0R4: dummy village, equals 1 if desa Sukaambit kabupaten Sumedang, zero otherwise. 0R5 : duiIIny village, equals 1 if desa Ciwangi kabupaten Garut, zero otherwise. flgtg; Nargabinangun kabupaten Cirebon is the control village. Wm Pl: probability of farmer grows traditional rice variety (TV). P2: probability of farmer grows high yielding rice variety (HYV). P3: probability of farmer grows mixed variety (MV), both TV and HYV. Ln(P2/Pl): natural logarithm of (PZ/P1)' Ln(P3/Pl): natural logarithm of (P3/Pl). Ln(P3/P2): natural logarithm of (P3/P2). PTV: price of TV (Rp/Kg). PHYV: price of HYV (Rp/Kg). PF: price of fertilizer (Rp/Kg). NAGE: labor wage (Rp/hour). HA: hectares of area cultivated with rice. 200 Appendix 5.5. Individual Level of Technical inefficiency Estimated from Cobb—Douglas Production Frontier OBS ‘ID HA TE TIE 1 501041 0.43900 0.965581 0.0344191 2 608215 0.38067 0.965450 0.0345496 3 606133 0.12167 0.963772 0.0362275 4 608207 0.58933 0.962232 0.0377680 5 101056 0.33600 0.961426 0.0385738 6 302192 0.16617 0.957572 0.0424276 7 302195 0.22933 0.956930 0.0430696 8 401032 0.08633 0.956897 0.0431025 9 301075 0.55167 0.956171 0.0438295 10 504167 0.13800 0.954806 0.0451938 11 204096 2.48833 0.954338 0.0456625 12 606145 0.37650 0.954329 0.0456711 13 101068 0.42033 0.953795 0.0462048 14 302189 0.55500 0.953326 0.0466741 15 607188 0.89317 0.953202 0.0467975 16 101057 0.18983 0.952472 0.0475282 17 401125 0.46883 0.951562 0.0484383 18 609234 0.49300 0.951515 0.0484854 19 101069 0.46300 0.951474 0.0485256 20 607164 0.54067 0.950991 0.0490087 21 606151 0.25483 0.950862 0.0491377 22 605116 0.26167 0.950615 0.0493852 23 401109 0.21867 0.950374 0.0496262 24 401002 0.24533 0.949882 0.0501179 25 607167 0.98983 0.949467 0.0505330 26 302197 0.49700 0.949163 0.0508366 27 302116 0.36100 0.948565 0.0514351 28 401075 0.16633 0.948453 0.0515467 29 401138 0.16767 0.948427 0.0515728 30 301023 0.15300 0.948242 0.0517579 31 205153 1.45233 0.947771 0.0522287 32 302169 0.48900 0.947522 0.0524776 33 201003 0.48800 0.947268 0.0527317 34 201002 0.52567 0.946468 0.0535319 35 302134 0.49450 0.946295 0.0537049 36 502112 0.18800 0.946243 0.0537566 37 204116 1.19033 0.946199 0.0538007 38 501020 0.41933 0.946085 0.0539154 39 607168 1.88700 0.946083 0.0539173 40 402162 0.18817 0.945522 0.0544779 41 607170 0.37100 0.945391 0.0546087 42 401122 0.56200 0.944937 0.0550629 43 203079 0.80850 0.944438 0.0555621 44 401092 0.14817 0.944196 0.0558038 45 302161 0.31400 0.944137 0.0558625 46 402203 0.15350 0.944001 0.0559985 47 101067 0.23567 0.943627 0.0563726 48 101026 0.15583 0.943118 0.0568818 49 302205 0.55350 0.942850 0.0571505 50 607195 0.13183 0.942830 0.0571702 51 101035 0.32117 0.942420 0.0575795 52 207209 0.36933 0.942350 0.0576500 53 302163 0.41767 0.942279 - 0.0577214 54 202039 2.48600 0.942160 0.0578395 55 504204 0.43583 0.942083 0.0579166 Appendix 5.5 (Continued) OBSID 56 504161 57 504162 58 - 206169 59 402171 60 302194 61 302146 62 302137 63 302151 64 401095 65 102119 66 202066 67 201001 68 102157 69 206158 70 603052 71 402201 72 302144 73 402167 74 502080 75 503135 76 603070 77 502062 78 401069 79 503143 80 402155 81 609241 82 402150 83 504168 84 402168 85 401041 86 301070 87 504201 88 402179 89 606147 90 604074 91 401077 92 301038 93 501045 94 302209 95 401124 96 401034 97 301004 98 609227 99 302142 100 502057 101 101073 102 301067 103 401043 104 102113 105 102111 106 609242 107 201009 108 204124 109 209250 110 102220 201 HA 0.40183 0.11033 0.30833 0.21700 0.42667 0.40417 0.26200 0.35033 0.33133 0.41150 0.74467 0.65083 0.71283 0.51200 0.21400 0.09067 0.64267 0.22617 0.16667 0.27433 0.16650 0.18900 0.08433 0.07000 0.40133 0.14450 0.15817 0.17633 0.12367 0.14850 0.14333 0.38200 0.09283 0.29200 0.35967 0.12117 0.95633 0.22150 0.67633 0.10617 0.31667 0.23967 0.32817 0.15917 0.23567 0.11417 0.64550 0.28583 0.73783 0.60083 0.06317 0.27233 0.21333 0.17783 2.85650 TE 0.941709 0.941700 0.941579 0.941095 0.941082 0.941020 0.940991 0.940852 0.940680 0.940596 0.940314 0.940290 0.940219 0.940037 0.940018 0.939721 0.939697 0.939608 0.939258 0.939228 0.939147 0.938967 0.938917 0.938422 0.938164 0.937971 0.937959 0.937623 0.936693 0.936455 0.936450 0.936097 0.936079 0.935972 0.935804 0.935557 0.935456 0.935381 0.935380 0.934833 0.934703 0.934555 0.934407 0.934108 0.933890 0.933778 0.933558 0.933526 0.932380 0.932220 0.932208 0.931696 0.931656 0.931635 0.931201 TIE 0.0582906 0.0583001 0.0584212 0.0589046 0.0589180 0.0589802 0.0590086 0.0591485 0.0593204 0.0594039 0.0596862 0.0597103 0.0597807 0.0599633 0.0599817 0.0602792 0.0603029 0.0603924 0.0607419 0.0607719 0.0608535 0.0610330 0.0610827 0.0615783 0.0618356 0.0620291 0.0620411 0.0623773 0.0633066 0.0635449 0.0635501 0.0639025 0.0639212 0.0640283 0.0641956 0.0644431- 0.0645440 0.0646193 0.0646203 0.0651667 0.0652972 0.0654450 0.0655935 0.0658917 0.0661095 0.0662219 0.0664424 0.0664735 0.0676202 0.0677801 0.0677919 0.0683039 0.0683438 0.0683646 0.0687995 Appendix 5.5. (Continued) OBSID 111 302131 112 301105 113 601010 114 208225 115 609245 116 603043 117 203080 118 205151 119 301110 120 402176 121 302182 122 202061 123 301055 124 605108 125 504197 126 401036 127 302120 128 502081 129 205132 130 503136 131 609244 132 602034 133 601005 134 609231 135 101001 136 209232 137 302199 138 401058 139 209241 140 101017 141 301058 142 402208 143 605109 144 302147 145 501034 146 608205 147 205136 148 401049 149 302143 150 204114 151 402169 152 302153 153 101094 154 102194 155 603068 156 501008 157 601016 158 504169 159 101089 160 401063 161 102126 162 603065 163 401006 164 302207 165 501001 202 HA 0.36833 0.18633 0.38867 0.90200 0.37650 0.15933 0.25383 0.60733 0.15733 0.18433 0.31650 2.08900 0.46500 0.07233 0.17350 0.08567 0.06333 0.20933 0.58050 0.16700 0.18850 0.11683 0.05433 0.17733 2.52400 0.10833 0.92267 0.15967 0.68867 0.72300 0.21667 0.39217 0.35533 0.08433 0.15583 0.65983 0.11983 0.52400 0.61900 0.24000 0.16067 0.92017 0.46417 0.35700 0.50317 0.46250 1.10250 0.17033 0.29000 0.17917 0.76700 0.71200 0.39383 0.55433 0.37333 TE 0.931190 0.931188 0.931021 0.930998 0.930912 0.930761 0.930562 0.930370 0.930123 0.930064 0.929702 0.929681 0.929674 0.929179 0.928175 0.928097 0.928068 0.927409 0.926778 0.925246 0.925022 0.924062 0.923591 0.923468 0.923444 0.923203 0.923195 0.922682 0.922072 0.921714 0.921315 0.920935 0.919889 0.919719 ‘0.919S40 0.918774 0.918177 0.917970 0.917885 0.917564 0.916763 0.916563 0.916240 0.915884 0.913612 0.913259 0.912696 0.911583 0.911079 0.910556 0.908945 0.906839 0.905463 0.903213 0.901846 TIE 0.0688103 0.0688125 0.0689787 0.0690022 0.0690878 0.0692394 0.0694383 0.0696304 0.0698772 0.0699363 0.0702982 0.0703193 0.0703256 0.0708210 0 0.0718248 0.0719032 0.0719324 0.0725915 0.0732221 0.0747541 0.0749777 0.0759377 0.0764089 0.0765317 0.0765557 0.0767972 0.0768055 0.0773178 0.0779277 0.0782861 0.0786854 0.0790652 0.0801111 0.0802808 0.0804599 0.0812264 0.0818229 0.0820298 0.0821155 0.0824361 0.0832369 0.0834368 0.0837602 0.0841157 0.0863877 0.0867411 0.0873039 0.0884173 0.0889215 0.0894440 0.0910548 0.0931606 0.0945367 0.0967868 0.0981537 Appendix 5.5. (Continued) 203 OBS ID HA 166 167 168 169 170 171 Notes : OBS : observation 301010 401037 301084 603062 603067 603053 0.102333 0.190333 0.611500 0.212000 0.350000 0.343167 ID: respondent's identification number HA: average (over time) farm size TE: technical efficiency TIE: teclmical inefficiency TE 0.901552 0.899745 0.899374 0.893669 0.891260 0.880566 TIE 0.098448 0.100255 0.100626 0.106331 0.108740 0.119434 204 Appendix 5.6. Individual Level of Technical Inefficiency Estimated From‘Translog Production Frontier OBSID HA TE TIE 1 501041 0.43900 0.965619 0.0343810 2 608215 0.38067 0.965478 0.0345218 3 606133 0.12167 0.963660 0.0363401 4 608207 0.58933 0.961980 0.0380204 5 101056 0.33600 0.961097 0.0389034 6 302192 0.16617 0.956836 0.0431642 7 302195 0.22933 0.956120 0.0438797 8 401032 0.08633 0.956084 0.0439165 9 301075 0.55167 0.955271 0.0447288 10 504167 0.13800 0.953741 0.0462588 11 204096 2.48833 0.953214 0.0467859 12 606145 0.37650 0.953204 0.0467957 13 101068 0.42033 0.952603 0.0473969 14 302189 0.55500 0.952073 0.0479266 15 607188 0.89317 0.951934 0.0480659 16 101057 0.18983 0.951108 0.0488924 17 401125 0.46883 0.950076 0.0499243 18 609234 0.49300 0.950022 0.0499779 19 101069 0.46300 0.949976 0.0500236 20 607164 0.54067 0.949427 0.0505726 21 606151 0.25483 0.949281 0.0507194 22 605116 0.26167 0.948999 0.0510010 23 401109 0.21867 0.948724 0.0512756 24 401002 0.24533 0.948164 0.0518363 25 607167 0.98983 0.947690 0.0523102 26 302197 0.49700 0.947343 0.0526571 27 302116 0.36100 0.946658 0.0533420 28 401075 0.16633 0.946530 0.0534698 29 401138 0.16767 0.946500 0.0534998 30 301023 0.15300 0.946288 0.0537118 31 205153 1.45233 0.945748 0.0542518 32 302169 0.48900 0.945462 0.0545376 33 201003 0.48800 0.945170 0.0548295 34 201002 0.52567 0.944250 0.0557501 35 302134 0.49450 0.944051 0.0559494 36 502112 0.18800 0.943991 0.0560089 37 204116 1.19033 0.943940 0.0560597 38 501020 0.41933 0.943808 0.0561920 39 607168 1.88700 0.943806 0.0561942 40 402162 0.18817 0.943159 0.0568409 41 607170 0.37100 0.943008 0.0569919 42 401122 0.56200 0.942483 0.0575167 43 203079 0.80850 0.941906 0.0580940 44 401092 0.14817 0.941626 0.0583739 45 302161 0.31400 0.941558 0.0584419 46 402203 0.15350 0.941401 0.0585994 47 101067 0.23567 0.940967 0.0590330 48 101026 0.15563 0.940376 0.0596239 49 302205 0.55350 0.940064 0.0599358 50 607195 0.13183 0.940041 0.0599588 51 101035 0.32117 0.939566 0.0604345 52 207209 0.36933 0.939484 0.0605164 53 302163 0.41767 0.939401 0.0605995 54 202039 2.48600 0.939263 0.0607369 55 504204 0.43533 0.939173 0.0608265 Appendix 5.6. (Continued) 205 OBSIDI-IA 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 504161 504162 206169 402171 302194 302146 302137 302151 401095 102119 202066 201001 102157 206158 603052 402201 302144 402167 502080 503135 603070 502062 401069 503143 402155 609241 402150 504168 402168 401041 301070 504201 402179 606147 604074 401077 301038 501045 302209 401124 401034 301004 609227 302142 502057 101073 301067 401043 102113 102111 609242 201009 204124 209250 102220 0.40183 0.11033 0.30833 0.21700 0.42667 0.40417 0.26200 0.35033 0.33133 0.41150 0.74467 0.65083 0.71283 0.51200 0.21400 0.09067 0.64267 0.22617 0.16667 0.27433 0.16650 0.18900 0.08433 0.07000 0.40133 0.14450 0.15817 0.17633 0.12367 0.14850 0.14333 0.38200 0.09283 0.29200 0.35967 0.12117 0.95633 0.22150 0.67633 0.10617 0.31667 0.23967 0.32817 0.15917 0.23567 0.11417 0.64550 0.28583 0.73783 0.60083 0.06317 0.27233 0.21333 0.17783 2.85650 TE 0.938738 0.938727 0.938586 0.938022 0.938007 0.937934 0.937901 0.937738 0.937537 0.937440 0.937110 0.937082 0.937000 0.936787 0.936765 0.936417 0.936390 0.936285 0.935876 0.935841 0.935745 0.935535 0.935477 0.934896 0.934594 0.934367 0.934353 0.933959 0.932867 0.932586 0.932580 0.932166 0.932144 0.932017 0.931821 0.931529 0.931410 0.931321 0.931320 0.930676 0.930522 0.930347- 0.930172 0.929820 0.929563 0.929430 0.929169 0.929132 0.927776 0.927586 0.927572 0.926966 0.926918 0.926894 0.926378 TIE 0.0612620 0.0612730 0.0614141 0.0619775 0.0619931 0.0620657 0.0620987 0.0622619 0.0624626 0.0625600 0.0628896 0.0629178 0.0630000 0.0632134 0.0632349 0.0635826 0.0636104 0.0637151 0.0641240 0.0641592 0.0642546 0.0644649 0.0645230 0.0651038 0.0654055 0.0656325 0.0656465 0.0660412 0.0671332 0.0674135 0.0674197 0.0678344 0.0678564 0.0679825 0.0681795 0.0684711 0.0685900 0.0686788 0.0686800 0.0693242 0.0694783 0.0696527 0.0698279 0.0701801 0.0704374 0.0705703 0.0708309 0.0708677 0.0722244 0.0724138 0.0724278 0.0730344 0.0730817 0.0731064 0.0736221 Appendix 5.6. (Continued) .206 OBS ID HA TE 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 I45 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 302131 301105 601010 208225 609245 603043 203080 205151 301110 402176 302182 202061 301055 605108 504197 401036 302120 502081 205132 503136 609244 602034 601005 609231 101001 209232 302199 401058 209241 101017 301058 402208 605109 302147 501034 608205 205136 401049 302143 204114 402169 302153 101094 102194 603068 501008 601016 504169 101089 401063 102126 603065 401006 302207 501001 0.36833 0.18633 0.38867 0.90200 0.37650 0.15933 0.25383 0.60733 0.15733 0.18433 0.31650 2.08900 0.46500 0.07233. 0.17350 0.08567 0.06333 0.20933 0.58050 0.16700 0.18850 0.11683 0.05433 0.17733 2.52400 0.10833 0.92267 0.15967 0.68867 0.72300 0.21667 0.39217 0.35533 0.08433 0.15583 0.65983 0.11983 0.52400 0.61900 0.24000 0.16067 0.92017 0.46417 0.35700 0.50317 0.46250 1.10250 0.17033 0.29000 0.17917 0.76700 0.71200 0.39383 0.55433 0.37333 0.926365 0.926362 0.926165 0.926137 0.926036 0.925856 0.925620 0.925392 0.925099 0.925028 0.924598 0.924573 0.924566 0.923977 0.922782 0.922689 0.922654 0.921869 0.921117 0.919289 0.919022 0.917874 0.917311 0.917164 0.917135 0.916846 0.916836 0.916223 0.915493 0.915063 0.914585 0.914130 0.912875 0.912672 0.912457 0.911537 0.910821 0.910573 0.910470 0.910085 0.909122 0.908882 0.908493 0.908066 0.905334 0 . 904 908 0.904231 0.902891 0.902284 0.901655 0.899715 0.897179 0.895522 0.892812 0.891166 TIE 0.073635 0.073638 0.073835 0.073863 0.073964 0.074144 0.074380 0.074608 0.074901 0.074972 0.075402 0.075427 0.075434 0.076023 0.077218 0.077311 0.077346 0.078131 0.078883 0.080711 0.080978 0.082126 0.082689 0.082836 0.082865 0.083154 0.083164 0.083777 0.084507 0.084937 0.085415 0.085870 0.087125 0.087328 0.087543 0.088463 0.089179 0.089427 0.089530 0.089915 0.090878 0.091118 0.091507 0.091934 0.094666 0.095092 0.095769 0.097109 0.097716 0.098345 0.100285 0.102821 0.104478 0.107188 0.108834 Appendix 5.5. (continued) 207 OBSIDI-IA 166 167 168 169 170 171 Notes: Variable definition as in appendix 5.5. 301010 401037 301084 603062 603067 603053 0.102333 0.190333 0.611500 0.212000 0.350000 0.343167 TE 0.890812 0.888637 0.888191 0.881334 0.878444 0.865664 TIE 0.109188 0.111363 0.111809 0.118666 0.121556 0.134336 208 Appendix 5.7. Individual Level of Profit Inefficiency Estimated fromiCbbb-Douglas Profit Frontier mSID HA PE PIE 1 501041 0.43900 0.930522 0.069478 2 608215 0.38067 0.930165 0.069835 3 608207 0.58933 0.925434 0.074566 4 606133 0.12167 0.925035 0.074965 5 101056 0.33600 0.914863 0.085137 6 504167 0.13800 0.909389 0.090611 7 606145 0.37650 0.909346 0.090654 8 605116 0.26167 0.905866 0.094134 9 401109 0.21867 0.905827 0.094173 10 302195 0.22933 0.905262 0.094738 11 607164 0.54067 0.903816 0.096184 12 606151 0.25483 0.903656 0.096344 13 204096 2.48833 0.902685 0.097315 14 401125 0.46883 0.902267 0.097733 15 101068 0.42033 0.901974 0.098026 16 504204 0.43583 0.901798 0.098202 17 504168 0.17633 0.901602 0.098398 18 607170 0.37100 0.900488 0.099512 19 401032 0.08633 0.899904 0.100096 20 301075 0.55167 0.897939 0.102061 21 302189 0.55500 0.896692 0.103308 22 609234 0.49300 0.896558 0.103442 23 302192 0.16617 0.895092 0.104908 24 607188 0.89317 0.894791 0.105209 25 607168 1.88700 0.894440 0.105560 26 401095 0.33133 0.894414 0.105586 27 607167 0.98983 0.893996 0.106004 28 501020 0.41933 0.893730 0.106270 29 401002 0.24533 0.893542 0.106458 30 201002 0.52567 0.892572 0.107428 31 201003 0.48800 0.892442 0.107558 32 201001 0.65083 0.892167 0.107833 33 401075 0.16633 0.892058 0.107942 34 101026 0.15583 0.891980 0.108020 35 302151 0.35033 0.891334 0.108666 36 302197 0.49700 0.890950 0.109050 37 504162 0.11033 0.890909 0.109091 38 302144 0.64267 0.890215 0.109785 39 202066 0.74467 0.889810 0.110190 40 607195 0.13183 0.889323 0.110677 41 102119 0.41150 0.889134 0.110866 42 302169 0.48900 0.888701 0.111299 43 102220 2.85650 0.888352 0.111648 44 302116 0.36100 0.887692 0.112308 45 204116 1.19033 0.887678 0.112322 46 101057 0.18983 0.887677 0.112323 47 503143 0.07000 0.886307 0.113693 48 401043 0.28583 0.886164 0.113836 49 301058 0.21667 0.885950 0.114050 50 402162 0.18817 0.885215 0.114785 51 401122 0.56200 0.884704 0.115296 52 504161 0.40183 0.884692 0.115308 53 101069 0.46300 0.884372 30.115628 54 401069 0.08433 0.884183 0.115817 55 302134 0.49450 0.884023 0.115977 Appendix 5. 7. (Continued) 209 OBSID 56 302194 57 209232 58 205153 59 606147 60 402179 61 401092 62 203079 63 302163 64 503135 65 402201 66 301038 67 504201 68 302137 69 402203 70 102113 71 208225 72 302146 73 302161 74 207209 75 609242 76 302142 77 401124 78 401138 79 401077 80 302199 81 401034 82 301023 83 202061 84 402167 85 609241 86 101067 87 302209 88 301067 89 502080 90 601010 91 402176 92 402171 93 202039 94 402155 95 206158 96 301004 97 609245 98 604074 99 401041 100 501045 101 205151 102 302205 103 608205 104 204124 105 504197 106 101035 107 102157 108 609231 109 101094 110 101073 HA 0.42667 0.10833 1.45233 0.29200 0.09283 0.14817 0.80850 0.41767 0.27433 0.09067 0.95633 0.38200 0.26200 0.15350 0.73783 0.90200 0.40417 0.31400 0.36933 0.06317 0.15917 0.10617 0.16767 0.12117 0.92267 0.31667 0.15300 2.08900 0.22617 0.14450 0.23567 0.67633 0.64550 0.16667 0.38867 0.18433 0.21700 2.48600 0.40133 0.51200 0.23967 0.37650 0.35967 0.14850 0.22150 0.60733 0.55350 0.65983 0.21333 0.17350 0.32117 0.71283 0.17733 0.46417 0.11417 PE 0.883347 0.882712 0.882663 0.882375 0.881468 0.881172 0.880792 0.880780 0.880706 0.880547 0.880525 0.880005 0.879705 0.879275 0.878830 0.878772 0.878443 0.877798 0.876965 0.876394 0.875209 0.874334 0.874043 0.873322 0.872338 0.872017 0.870838 0.870275 0.870177 0.870094 0.868848 0.868127 0.868034 0.867687 0.867334 0.867250 0.867223 0.866331 0.866015 0.865964 0.865961 0.865659 0.865300 0.865217 0.865141 0.864151 0.863792 0.860377 0.859565 0.858515 0.858109 0.857865 0.857826 0.857787 0.857532 PIE 0.116653 0.117288 0.117337 0.117625 0.118532 0.118828 0.119208 0.119220 0.119294 0.119453 0.119475 0.119995 0.120295 0.120725 0.121170 0.121228 0.121557 0.122202 0.123035 0.123606 0.124791 0.125666 0.125957 0.126678 0.127662 0.127983 0.129162 0.129725 0.129823 0.129906 0.131152 0.131873 0.131966 0.132313 0.132666 0.132750 0.132777 0.133669 0.133985 0.134036 0.134039 0.134341 0.134700 0.134783 0.134859 0.135849 0.136208 0.139623 0.140435 0.141485 0.141891 0.142135 0.142174 0.142213 0.142468 Appendix 5.7. (Continmd) 210 OBSIDHAPE 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 I45 146 I47 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 I64 165 302131 402150 301070 302182 209250 605109 401058 502112 601005 205132 609244 603052 102126 503136 609227 301105 203080 301110 603043 101089 302143 301055 401036 602034 206169 502062 504169 102111 603070 101017 102194 592057 302147 209241 501008 205136 402208 302153 301084 204114 401049 605108 401063 201009 603053 401006 502081 601016 402169 401037 301010 501034 302120 402168 101001 0.36833 0.15817 0.14333 0.31650 0.17783 0.35533 0.15967 0.18800 0.05433 0.58050 0.18850 0.21400 0.76700 0.16700 0.32817 0.18633 0.25383 0.15733 0.15933 0.29000 0.61900 0.46500 0.08567 0.11683 0.30833 0.18900 0.17033 0.60083 0.16650 0.72300 0.35700 0.23567 0.08433 0.68867' 0.46250 0.11983 0.39217 0.92017 0.61150 0.24000 0.52400 0.07233 0.17917 0.27233 0.34317 0.39383 0.20933 1.10250 0.16067 0.19033 0.10233 0.15583 0.06333 0.12367 2.52400 0.857507 0.856403 0.856108 0.855870 0.855182 0.855075 0.854247 0.853608 0.851629 0.851247 0.849845 0.849066 0.847935 0.847578 0.847269 0.845104 0.842738 0.842202 0.841087 0.840512 0.840171 0.839616 0.839027 0.832751 0.832463 0.830935 0.830279 0.829947 0.829388 0.829100 0.828266 0.826818 0.826384 0.825109 0.823593 0.822588 0.820492 0.820399 0.820187 0.818403 0.818318 0.817736 0.817155 0.815700 0.814173 0.811903 0.809714 0.807777 0.802963 0.800271 0.797227 0.796413 0.796208 0.795409 0.792792 PIE 0.142493 0.143597 0.143892 0.144130 0.144818 0.144925 0.145753 0.146392 0.148371 0.148753 0.150155 0.150934 0.152065 0.152422 0.152731 0.154896 0.157262 0.157798 0.158913 0.159488 0.159829 0.160384 0.160973 0.167249 0.167537 0.169065 0.169721 0.170053 0.170612 0.170900 0.171734 0.173182 0.173616 0.174891 0.176407 0.177412 0.179508 0.179601 0.179813 0.181597 0.181682 0.182264 0.182845 0.184300 0.185827 0.188097 0.190286 0.192223 0.197037 0.199729 0.202773 0.203587 0.203792 0.204591 0.207208 Appendix 5.7. (Continued) 211 OBS ID . HA 166 302207 0.554333 167 603065 0.712000 168 501001 0.373333 169 603068 0.503167 170 603062 0.212000 171 603067 0.350000 Notes : 08s : observation ID: respondent's identification umber HA: average (over tine) farm size PE profit efficiency PIE: profit inefficiency PE 0.788484 0.768216 0.763551 0.746916 0.745791 0.711396 PIE 0.211516 0.231784 0.236449 0.253084 0.254209 0.288604 2312? Appendix 5.8. Individual Level of Profit Inefficiency IE151:1318113883. 1818:401‘1xriaiissliori’ Itrradii;t: Iirxaritziiair (IEES III) 1111 1?13 III}: 1 508215 0.38057 0.930592 0.069408 2 501041 0.43900 0.927429 0.072571 3 505133 0.12157 0.923098 0.075902 4 508207 0.58933 0.920333 0.079557 5 101055 0.33500 0.915599 0.083401 5 504157 0.13800 0.905507 0.094493 7 505145 0.37550 0.905310 0.094590 8 401125 0.46883 0.905135 0.094155 9 507154 0.54057 0.901904 0.098095 10 504168 0.17533 0.901454 0.098546 11 505115 0.25157 0.900878 0.099122 12 401109 0.21857 0.899479 0.100521 13 302195 0.22933 0.899415 0.100584 14 509234 0.49300 0.899252 0.100748 15 504204 0.43583 0.898705 0.101295 15 505151 0.25483 0.895884 0.103115 17 401032 0.08633 0.895992 0.104008 18 101068 0.42033 0.895839 0.104151 19 401002 0.24533 0.895528 0.104472 20 302192 0.15517 0.895411 0.104589 21 507188 0.89317 0.892238 0.107752 22 301075 0.55157 0.891512 0.108488 23 402152 0.18817 0.889718 0.110282 24 507170 0.37100 0.889227 0.110773 25 401095 0.33133 0.889201 0.110799 25 201001 0.65083 0.889179 0.110821 27 302189 0.55500 0.889138 0.110862 28 507157 0.98983 0.889069 0.110931 29 501020 0.41933 0.887485 0.112515 30 401075 0.15533 0.886599 0.113401 31 201002 0.52557 0.886555 0.113445 32 204095 2.48833 0.886388 0.113512 33 204115 1.19033 0.886316 0.113554 34 201003 0.48800 0.886050 0.113950 35 504152 0.11033 0.885754 0.114245 35 401122 0.55200 0.885485 0.114515 37 101059 0.45300 0.885147 0.114853 38 302159 0.48900 0.884407 0.115593 39 302197 0 49700 0.882498 0.117502 40 505147 0 29200 0.882491 0.117509 41 507158‘ 1.88700 0.879745 0.120255 42 101025 0.15583 0.879310 0.120590 43 101057 0.18983 0.879110 0.120590 44 ,302144 0.54257 0.879067 0.120933 45 302151 0.35033 0.878899 0.121101 45 301058 0.21557 0.878590 0.121410 47 209232 0.10833 0.878581 0.121419 48 302137 0.25200 0.878528 0.121472 49 302134 0.49450 0.878236 0.121754 50 302115 0.35100 0.878175 0.121125 51 302151 0 31400 0.877684 0.122315 52 102119 0.41150 0.877654 0.122345 53 401092 0.14817 0.877293 0.122707 54 302153 0.41757 0.875158 0.123532 55 102220 2.85650 0.876150 0.123550 Appendix 5.8. (Continued) 213 OBSIDHA 56 57 58 59 60 61 62 63 66 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 86 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 503135 607195 301023 205153 504161 208225 302196 207209 401043 205151 402203 102113 302166 302142 203079 202066 101067 503143 401077 504201 402155 101035 502080 402167 401034 601041 401138 302131 402179 302209 402201 402171 401069 202039 609231 301067 609245 401124 301070 501045 601010 609242 604074 301038 206158 402176 301105 608205 204124 302205 101094 502112 609241 202061 605109 0.27433 0.13183 0.15300 1.65233 0.40183 0.90200 0.42667 0.36933 0.28583 0.60733 0.15350 0.73783 0.40417 0.15917 0.80850 0.74467 0.23567 0.07000 0.12117 0.38200 0.40133 0.32117 0.16667 0.22617 0.31667 0.14850 0.16767 0.36833 0.09283 0.67633 0.09067 0.21700 0.08433 2.48600 0.17733 0.64550 0.37650 0.10617 0.14333 0.22150 0.38867 0.06317 0.35967 0.95633 0.51200 0.18433 0.18633 0.65983 0.21333 0.55350 0.46417 0.18800 0.16650 2.08900 0.35533 PE 0.875717 0.872934 0.872700 0.872555 0.872231 0.871745 0.871068 0.870099 0.870076 0.869253 0.869098 0.868632 0.867953 0.867852 0.867673 0.866988 0.866451 0.865759 0.864755 0.863499 0.863221 0.863176 0.863031 0.862756 0.862292 0.861602 0.860860 0.860538 0.860464 0.860180 0.859615 0.858791 0.858650 0.857767 0.857745 0.857551 0.856318 0.855853 0.855852 0.854260 0.854110 0.853929 0.853769 0.853211 0.852751 0.850276 0.850244 0.850206 0.849108 0.868916 0.848809 0.848204 0.847970 0.847751 0.846367 PIE 0.126283 0.127066 0.127300 0.127445 0.127769 0.128255 0.128932 0.129901 0.129924 0.130747 0.130902 0.131368 0.132047 0.132148 0.132527 0.133012 0.133549 0.136241 0.135245 0.136501 0.136779 0.136824 0.136969 0.137264 0.137708 0.138398 0.139140 0.139662 0.139536 0.139820 0.140385 0.161209 0.141350 0.142233 0.142255 0.142449 0.163682 0.144147 0.144148 0.145740 0.145890 0.146071 0.146231 0.146789 0.147249 0.169726 0.149756 0.149794 0.150892 0.151084 0.151191 0.151796 0.152030 0.152249 0.153633 Appendix 5.8. (Continued) OBS 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 167 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 ID 504197 602150 203080 301004 609244 302182 102157 301110 302199 206169 503136 101089 609227 209250 603052 101073 205132 102126 602034 601005 603063 401058 302143 401036 101017 504169 301055 204114 603070 502062 102194 502057 401049 302147 401063 302153 501008 605108 201309 102111 402169 205136 402208 209241 401006 601016 502081 301010 603053 301084 402168 401037 501034 302207 302120 214 HA 0.17550 0.15817 0.25585 0.25957 0.18850 0.51550 0.71285 0.15755 0.92257 0.50855 0.16700 0.29000 0.52817 0.17785 0.21400 0.11417 0.56050 0.75700 0.11585 0.05455 0.15955 0.15957 0.51900 0.05557 0.72500 0.17055 0.45500 0.24000 0.15550 0.16900 0.55700 0.25557 0.52400 0.05455 0.17917 0.92017 0.45250 0.07255 0.27255 0.50035 0.15057 0.11985 0.59217 0.53557 0.59585 1.10250 0.20955 0.10255 0.54517 0.51150 0.12557 0.19055 0.15505 0.55455 0.05555 PE 0.845914 0.845787 0.845162 0.866815 0.844719 0.844054 0.844000 0.842951 0.861552 0.840690 0.860126 0.839526 0.837451 0.836253 0.836051 0.836973 0.832090 0.830830 0.825856 0.824945 0.824370 0.822232 0.821609 0.820907 0.818028 0.817339 0.816312 0.814310 0.814123 0.816063 0.814021 0.813685 0.813106 0.812661 0.811137 0.806862 0.804205 0.802340 0.802319 0.801816 0.798555 0.793269 0.794265 0.794180 0.793915 0.792065 0.790267 0.790209 0.787270 0.778221 0.778118 0.776636 0.775066 0.768929 0.704088 PIE 0.154086 0.154213 0.154838 0.155185 0.155281 0.155946 0.156000 0.157049 0.158448 0.159310 0.159874 0.160476 0.162549 0.163767 0.163949 0.165027 0.167910 0.169170 0.174144 0.175055 0.175630 0.177768 0.178391 0.179093 0.181972 0.182661 0.183688 0.185690 0.185877 0.185957 0.185979 0.186515 0.186896 0.187339 0.188863 0.195138 0.195795 0.197660 0.197681 0.198184 0.201445 0.201751 0.205735 0.205820 0.206085 0.207955 0.209753 0.209791 0.212730 0.221779 0.221882 0.223364 0.224954 0.231071 0.235912 Appendix 5.8. (Continued) OBS ID 155 101001 157 503055 158 501001 159 503052 170 503058 171 503057 Note: 215 HA 2.52400 0.71200 0.37333 0.21200 0.50317 0.35000 PE 0.749106 0.762298 0.736825 0.725962 0.720022 0.691229 Variable definition as in appendix 5.7. PIE 0.250894 0.257702 0.263175 0.274038 0.279978 0.308771 216 Appendix 5.9. Evaluation of Rice Price Support and Fertilizer Subsidy A. Data and Parameters Total Rice Production in Rest Java Total rice production in Best Java in 1982 amounted to 7,431,497 tons rough rice (CBS, 1984). This is the total wetland rice production. Total Labor in The Rice Production in Hest Java The total labor absorbed in rice production, in Hest Java, is approximated by' multiplying the per hectar labor use by the total harvested area. The per hectar labor use is approximated by using the figures of table 5.1, that is about 150 mandays/hectare. The total annual harvested area of rice in Nest Java in 1982 was 1,702,504 hectare (CBS, 1984). Using these figures, the labor absorption in West Java’s rice production is approximately 255,375,600 mandays/year. Total Subsidized Fertilizer for Rice Production Total subsidized fertilizer (Urea+TSP) for rice production in West Java (1982) amounted to 439,180 tons. This figure is the product of total BIMAS fertilizer in West Java (1982) and the ratio of fertilizer use for rice relative to other crops. The total BIMAS fertilizer in West Java in 1982 amounted to 585,576 tons (Deptan/Deperin/APPI, 1984). Note that BIMAS fertilizer was mainly intended to be used for rice production. It is not uncommon, however, that farmers use some of this fertilizer for other crops. To avoid an overestimation a ratio of 0.75 is used; estimated by taking into account per hectare fertilizer use and total harvested area of rice relative to other crops. 217 Appendix 5.9 (Continued) Domestic Rice Procurement in West Java Excess supply of rice (which is assumed to be purchased by the government) due to the implementation of rice floor price is approximated by the annual domestic rice procurement during the period of 1979-1986 (Tabor, 1988). On the average (over time) the domestic procurement of rough rice in Indonesia amounted to approximately 7.5%.of the total rough- rice production. Using this ratio and the total rough-rice production in West Java in 1982, we get the estimate of the Nest Java’s domestic procurement in 1982 at 557,360 tons. Consulting to appendix (5.10), the domestic procurement increases only when the percentage increase of the rice price more than 10%. This can be interpreted that the excess supply of rice (which has to be purchased by the government) will increase only when the percentage increase in the floor price of rice high enough, which could probably need to be above the general inflation rate. This empirical evidence is used to formulate the proposed evaluation procedure, particularly reflected in equation 4.13 and 4.17 of chapter 4. Price of Rough Rice The price of rough rice used in this analysis is Rp 135 per kilogram. This was the floor price of rough rice in 1982. Price of Fertilizer The fertilizer price used in the analysis is the 1982’s price which was Rp 90 per kilogram. 218 Appendix 5.9 (Continued) Own- and Cross-Price Elasticity Parameters Translog are 0.8257 and 0.6026, respectively. The cross-price elasticity of rice supply with respect to fertilizer price is 0.1932 (Cobb-Douglas) and 0.1832 (translog). The cross-price elasticities of labor demand with respect to the prices of rice and fertilizer estimated from Cobb-Douglas (translog) function are 0.8257 (0.9166) and 0.1932 (0.1256), respectively The own-price elasticity of rice supply estimated from Cobb-Douglas and 8. Calculation (1) Price Support Policy: 1% Increase in Rough-Rice Price W21; Government’s additional cost: GPR - 1.35 * 557,360 * 1000 - Rp 752,436,000 Additional quantity of rice produced: AQRP(1) - 0.008257*7,431,497 - 61,362 tons Additional employment generated: AQL(1) - 0.008257 * 225,375,600 - 2,108,636 mandays Cost per unit of rice produced: CR(1) - 752,436,000/61,362,000 - Rp 12.26 /kg Cost per unit of employment generated: CL(1) - 752,436,000/2,108,636 - Rp 356.84 /manday. E1a51is1tx_E5t1mated_8a5ed.en.1raneleg_fiunstien; Government’s additional cost: 898 - 1.35 8 557,350 * 1000 - Rp 752,435,000 219 Appendix 5.9 (Continued) (2) Additional quantity of rice produced: AQRP(1) - 0.006026 * 7,431,497 - 44,782 tons Additional employment generated: AQL(1) - 0.009166 * 225,375,600 - 2,340,773 mandays Cost per unit of rice produced: CR(1) - 752,436,000/44,782,000 - Rp 16.80 /kg Cost per unit of employment generated: CL(1) - 752,436,000/2,340,773 - Rp 321.45 /manday. Fertilizer Subsidy Policy: 1% Reduction in Fertilizer Price W90; Government's additionai costs: GPF - 0.90 * 439,180 * 1000 - Rp 395,262,000 Additional quantity of rice produced: AQRP(2) - 0.001932 * 7,431,497 - 14,358 tons Additional empioyment generated: AQ](2) - 0.001932 * 255,375,600 - 493,385 mandays Cost per unit of rough rice produced: CR(2) - 395,262,000/14,358,000 - Rp 27.5 /kg Cost per unit of employment generated: CL(2) - 395,262,000/493,385 - Rp 801.12 /manday WWW Government’s additional costs: GPF - 0.90 * 439,180 * 1000 - Rp 395,262,000 220 Appendix 5.9 (Continued) Additional quantity of rice produced: AQRP(2) - 0.001832 * 7,431,497 - 13,615 tons Additional employment generated: A01(2) - 0.001256 * 255,375,600 - 320,752 mandays Cost per unit of rough rice produced: CR(2) - 395,262,000/13,615,000 - Rp 29.03 /kg Cost per unit of employment generated: CL(2) - 395,262,000/320,752 - Rp 1232.30 /manday 221 Appendix 5.10. Domestic Rice Procurement (Tons) in Indonesia. Year Total Rough Domestic Ratio x floor Change in Rice Product. Procur. Price Change Dom. Proc. (1) (2) (3-1/2) (4) (5) 1979 24,669,444 331,066 1.34 - - 1980 27,626,886 2,439,206 8.83 24.0 2,108,140 1981 30,640,867 3,098,871 10.11 14.0 659,665 1982 31,683,652 3,145,635 9.93 12.5 46,764 1983 33,148,256 1,489,808 4.49 7.4 -1,655,827 1984 35,725,992 3,853,488 10.79 14.0 2,093,680 1985 36,870,300 3,123,817 8.79 6.0 -729,671 1986 36,378,453 2,363,033 6.32 0.0 ~760,784 (l) and (2) adopted from Tabor (1988): rough rice equivalent (4) computed from appendix 2.1, using: (P1 - Po)/Po * 100% (anually) (5) computed anually from (2) RIES 11111111111111u