Identification of climatic effect on crop yield of Marathwada region by using multiple linear regression & stochastics frontier approach

Transcription

1 016; (5): ISSN Prnt: ISSN Onlne: Impact Factor: 5. IJAR 016; (5): Receved: Accepted: Satyvan Yashwant Research Scholar, Department of Statstcs, Dr. B.A.M.U Aurangabad, Maharashtra, Inda. SL Sananse Professor, Department of Statstcs, Dr. B.A.M.U Aurangabad, Maharashtra, Inda. Identfcaton of clmatc effect on crop yeld of Marathwada regon by usng multple lnear regresson & stochastcs fronter approach Satyvan Yashwant, SL Sananse Abstract The am of ths study s estmate clmatc effect on soybean crop yeld and measures ts crop producton effcency by usng multple lnear regresson and stochastc fronter model. In ths study we have use 0 years (1980 to010) data of soybean and ten clmactc factor n Marathwada regon. Ths regon s stuated between N and N lattude and E and E longtudes n Maharashtra state In ths regon consst of eght dstrcts such as, Aurangabad, Beed, Hngol, Jalna, Latur, Nanded, Osmanabad and Parbhan. The economy of ths regon depends upon agrculture sector. The last ten year ths regon s less ranfall affected regon n Maharashtra. The result show that ranfall, area, dry spell max temp are postvely effect on soybean crop producton and mn humdty, mnmum temp, total could are negatve effect on crop producton of soybean. The ftted multple lnear regresson model of parameter estmate by usng least square regresson method. result show that value of multple R-square of Later and Hngol dstrct are 0.91 and 0.90 respectvely ths ndcate that ftted model can explan more varablty of repressors varable. The Cobb-Douglas stochastc fronter model s appled to Latur and Hngol of soybean crop producton for measure producton effcency. The result show that effcency of Latur and Hngol dstrct are 0.86 and 0.84 respectvely Ths result ndcates that 86%and 84% of total composed error varance of the producton functon s explaned by the varance of the techncal neffcency term. Ths result s mportance to ncorporate techncal neffcency n the producton functon.maxmum lke hooded method s used for estmaton of parameter of stochastc fronter model. The value of gamma s used to testng the hypothess to check stochastc nose and techncal neffcency. Ths result s helpful to framer and regonal and local planer n Marathwada regon. Keywords: Multple Regresson model, Stochastc Fronter model, Maxmum lkelhood Estmator, lkelhood Rato test Correspondence Satyvan Yashwant Research Scholar, Department of Statstcs, Dr. B.A.M.U Aurangabad, Maharashtra, Inda. Introducton Marathwada of Maharashtra comprsng eght dstrcts (Aurangabad, Jalna, Parbhan, Nanded, Latur, Beed, Omarabad and Hngol) s tradtonally a drought-prone regon. The stuaton has worsened wth ranfall defcency for three consecutve years. Ranfall defct for 014 was 45 per cent; ths was followed by a defct of more than 50 per cent n many dstrcts tll the end of August 015. The regon has struggled wth frequent droughts for the past 50 years. Even a moderate ranfall defcency s mpactng agrcultural output sgnfcantly due to ncreased water demand. Soybean s ganng popularty on account of ts unque characterstcs and adaptablty to vared agro-clmatc condtons. It has unmatched composton of 40 per cent proten and 0 per cent ol and nutrtonal superorty on account of contanng essental amno acds, unsaturated fatty acds, carbohydrates, vtamns and mnerals. n study we use two method for study the soybean crop yeld n Marathwada regon such as multple lnear regresson and stochastc fronter analyss. Regresson analyss s a statstcal technque that utlzes the relaton between two or more quanttatve varables on observatonal database so that outcome varable can be predcted from the others. One of the purposes of a regresson model s to fnd out to what extent the outcome (dependent varable) can be predcted by the ndependent varables. Ranfall forecast methods are employed n weather forecastng at regonal and natonal levels. Fundamentally, there are two approaches to predct ranfall. They are Emprcal method and dynamcal methods. The emprcal approach s based on analyss of hstorcal data of the ranfall and ts relatonshp to varety of atmospherc and oceanc varables over dfferent parts of the world n ths study usng ~ 841 ~

2 regresson analyss we have check the clmatc effect on soybean crop producton n Marathwada regon. The stochastc fronter analyss s an analytcal approach that utlzes econometrc (parametrc) technques whose models of producton recognze techncal neffcency and the fact that random shocks beyond producers control may affect the product. Dfferently from non-parametrc approaches that assume determnstc fronters, SFA allows for devatons from the fronter, whose error can be decomposed for adequate dstncton between techncal effcency and random stock (ranfall, dry spells, max temp, etc) Stochastc fronter models allow to analyse techncal neffcency n the framework of producton functons. Producton unts (frms, regons, countres, etc.) are assumed to produce accordng to a common technology, and reach the fronter when they produce the maxmum possble output for a gven set of nputs. Ineffcences can be due to structural problems or market mperfectons and other factors whch cause countres to produce below ther maxmum attanable output Over tme, producton unts can become less neffcent and catch up to the fronter.1 It s also possble that the fronter shfts, ndcatng techncal progress. In addton, producton unts can move along the fronter by changng nput quanttes. Fnally, there can be some combnatons of these three effects. The stochastc fronter method allows decomposng growth nto changes n nput use, changes n technology and changes n effcency, thus extendng the wdely used growth accountng method. When dealng wth productvty, two man problems arse: ts defnton and ts measurement. Tradtonally, emprcal research on productvty has suffered from a number of shortcomngs. Most emprcal studes have employed the so called Solow resdual (Solow 1956) [15]. The use of ths measure s problematc: Abramovtz (1956) [] refers to the dfference between the growth rates of output and the weghted sum of nput growth rates as a measure of our gnorance about the causes of economc growth. In ths study we have use soybean crop yeld s output varable and other ten are nput varable. Usng ths technque have check the man effcency of model for cheekng utlzaton of maxmum level of output of soybean by usng nput n Marathwada regon. Related Work Many researchers have study of crop producton by usng multple lnear regresson and stochastc fronter such as, Díaz and Sánchez (004) [9] nvestgated the temporary employment and techncal effcency n Span s productvty growth that occurred between md-1995 to the end of 000. Agner and Chu (1968) [1], who proposed the use of specfc econometrc models consstent wth the fronter--the bestpractce notons of Farrell (1957) [10]. Contemporary researchers famlar wth econometrc modellng would prefer the use of stochastc fronter analyss (SFA) n ther effcency studes (Agner et al. 1977; Meeusen and Broeck 1977) [1, 9]. Dolado et al, (001) []. The use of SFA software developed by Battese and Coell (1995) [4] has been wdely appled, not only n manufacturng and agrculture sectors, but also n fsheres studes that wsh to compare the performance of frms (see Roy 00, Basr et al. 006) [4]. They should be hghly accredted for makng ths robust computer program freely avalable to those who wsh to use t for teachng and research. Koop et al. (1999, 000a, b) [11], and Koop (001) [16] adopt a Bayesan approach to estmate stochastc producton fronters. Whle there are certanly advantages of the Bayesan estmaton method, the choce of Maxmum Lkelhood estmaton n large sample s justfed. Km and Schmdt (000) [1]. examne a large number of classcal and Bayesan procedures to estmate the level of techncal effcency usng dfferent panel data sets. They fnd that Maxmum Lkelhood estmaton based on the exponental dstrbuton gves smlar results to the Bayesan model n whch the pror dstrbuton for effcency s exponental and there s an unnformatve pror for the exponental parameter. Kumbhakar and L othgren (1998) [18] assume n ther Monte Carlo study that the true values of the underlyng parameters are unknown and must be replaced by ther ML estmates. They found that the result s true for all value of neffcency and for sample szes less than 00. Afrat (197) and Rchmond (1974) [] explctly assume that the dsturbances follow a one-sded dstrbuton, such as exponental or half normal. See Temple (1999) and the ntroducton for a more detaled dscusson. Addton, t allows to nclude explanatory varables n both the producton uncton and the effcency term. These are related work. Data and Methodology Study area and data collecton Marathwada regon consst of eght dstrcts such as, Beed, Hngol, Jalna, Latur, Nanded, Osmanabad and Parbhan. Ths In ths study we have used 0 years (1980 to010 ) of tme seres data n ranfall, mnmum &maxmum temperature, mnmum &maxmum humdty, wnd speed, wnd drecton & total could cover(octa) of eght metrologcal staton n Marathwada regon. In ths regon consst of eght dstrcts such as, Beed, Hngol, Jalna, Latur, Nanded, Osmanabad and Parbhan. Ths regon s stuated between N and N lattude and E and E longtudes. The data of two dstrcts of 0(1980 to010) years were obtaned by IMD & Maharashtra agrculture department. In ths study we have use the crop yeld of soybean s dependent (out varable) and other ten are ndependent (nput varable) such as, ranfall, dry spell, max. temp, mn temp, wnd speed, wnd drecton, total could cover, area of crop growng (hector), max. Humdty and mn.humdty 1. Regresson analyss Regresson analyss s a statstcal process for estmatng the relatonshps among varables. It ncludes many technques for modellng and analysng several varables, when the focus s on the relatonshp between a dependent varable and one or more ndependent varables. More specfcally, regresson analyss helps one understand how the typcal value of the dependent varable changes when any one of the ndependent varables s vared, whle the other ndependent varables are held fxed. In ths study use have used seven varables such as Ranfall, Max temp, Mn temp, Total could cover, Max humdty, Mn humdty, Wnd speed, Wnd drecton yeld and Area, Yeld s dependent varable. In that study more than two varable present n data we have use multple regresson A. Multple Regresson analyss Multple regressons ft a model to predct a dependent (Y) varable from two or more ndependent (X) varables. Multple lnear regresson models are often used as ~ 84 ~

3 approxmatng functons. That s, true functonal relatonshp between y and x1,x, x, s unknown, but over certan ranges of the repressor varables the lnear regresson model s an adequate approxmaton to the true unknown functon. In the present study ranfall was treated as dependent varable and maxmum & mnmum temperature, mnmum &maxmum humdty, wnd speed, wnd drecton & total could cover (octa) as ndependent varable. The form of the multple lnear regresson equaton ftted to the weekly average weather parameters s gven below. Y 0 X 1 1 X X X 4 4 X 5 5 X 6 6 X 7 7 X X 910X 10 Y = Crop yeld X1=ranfall (mm) X= Area (hectare) X= Maxmum temperature ( C) X4= Total cloud cover (octa) X5= Maxmum Relatve humdty (%) X6= Wnd speed (kmph) X7= Wnd drecton (deg) X8= Dry spell X9 = Mnmum temperature ( C) X10= Mnmum Relatve humdty (%) 0 = ntercept = regresson coeffcent of th ndependent varables (=1,, 7 ϵ = error term.. The Producton Fronter: Theoretcal Framework The standard defnton of a producton functon s that t gves the maxmum possble output for a gven set of nputs; the producton functon therefore defnes a boundary or a fronter. All the producton unts on the fronter wll be fully effcent. Effcency can be of two knds: techncal and allocate. Techncal effcency s defned ether as producng the maxmum level of output gven nputs or as usng the mnmum level of nputs gven output Allocate effcency occurs when the margnal rate of substtuton between any of the nputs equals the correspondng nput prce rato. If ths equalty s not satsfed, t means that the country s not usng ts nputs n the optmal productons. An ntal justfcaton for computng effcency can be found n that ts measure facltates comparsons across economc unts. Secondly, and perhaps more mportantly, when dvergence n effcency s found some further research needs to be undertaken to understand whch factors led to t. Fnally, dfferences n effcency show that there s scope for mplementng polces addressed to reduce them and to mprove effcency. A producton fronter model can be wrtten as Y f ( x ; ) TE Where, Y s the output of producer ( = 1, N); x s a vector of M nputs used by producer s f ( x; ) the producton fronter and β s a vector of technology parameters to be estmated. Let TE be the techncal effcency of producer, () TE Y f ( x; ) In the case, TE 1, () Y acheves ts maxmum feasble TE, t measures techncal output of f ( x; ).f 1 neffcency n the sense that observed output s below the maxmum feasble output. The producton fronter f ( x; ) s determnstc. That means that the entre shortfall of observed output Y from maxmum feasble output f ( x ; ) s attrbuted to techncal neffcency. Such a specfcaton gnores the producer specfc random shocks that are not under the control of the produce. We have to specfy the stochastc producton fronter Y f ( x; ) exp( v ) TE where, f ( x; )exp( v ) TE s the stochastc fronter, whch conssts of a determnstc part f ( x; ) common to all producers and a producer-specfc part whch exp( v ) captures the effect of the random shocks to each producer TE can be computed for Stochastc Fronter producton of th producer. Y TE f ( x; )exp( v ) Techncal effcency can be estmated usng ether the determnstc producton fronter model gven by equatons () and (), or the stochastc fronter model gven by equatons (4) and (5) snce the stochastc fronter model ncludes the effect of random shocks on the producton process, ths model s preferred to the determnstc fronter. Stochastc Fronter Productons Functon The econometrc approach to estmate fronter models uses a parametrc representaton of technology along wth a twopart composed error term. Under the assumpton that s of ( ; ) f x s of Cobb-Douglas type, the Stochastc fronter model n equaton can be wrtten as. lny X Where, s an error term wth? v u In ths study usng crop yeld of soybean s output varable and other ten are nput varable. The Cobb-Douglas type of stochastc fronter model n equaton can be wrtten as follows. ln (Y)= β 0+ β 1 ln(ranfall) + β ln (Area) + β l ln (Max temp) +β 4 ln (Total Could cover)+ β 5 ln (Max. Humdty)+ β 6 ln (Wnd speed)+ β 7 ln (Wnd drecton)+ β 8 l ln g (dry spell)+ β 9 ln (Mn.Temp)+ β 10 ln (Mn. Humdty))+ ϵ The economc logc behnd ths specfcaton s that the producton process s subject to two economcally dstngushable random dsturbances. Statstcal nose represented by v and techncal neffcency represented by (4) (5) (6) ~ 84 ~

4 u There are some assumptons necessary on the characterstcs of these components. The errors v assumed to have a symmetrc dstrbuton, n partcular, they are ndependently and dentcally dstrbuted as N (0, v ).The component u s assumed to be dstrbuted ndependently of v and to satsfy u 0 (e.g. t follows a one-sded normal dstrbuton N (0, v ).The non-negatvty of the techncal neffcency term reflects the fact that f u 0 the country wll not produce at the maxmum attanable level. Any devaton below the fronter s the result of factors partly under the producton unt s control, but the fronter tself can randomly vary across frms or over tme for the same producton unt. Ths last consderaton allows the asserton that the fronter s stochastc, wth a random dsturbance v beng postve or negatve dependng on favourable or unfavourable external events. It s mportant to note that gven the non-negatvty assumpton on the effcency term, ts dstrbuton s non-normal and therefore the total error term s asymmetrc and non-normal. Ths mples that the least squares estmator s neffcent. Assumng that v and u are dstrbuted ndependently of X estmaton of (6) by OLS provdes consstent estmators of all parameters but the ntercept, snce E( ) E( u ) Moreover, OLS does not provde an estmate of producerspecfc techncal effcency. However, t can be used to perform a smple test based on the skewness of emprcal dstrbuton of the estmated resduals. Schmdt and Ln (1984) propose the test statstc. 1/ m b m / Where, m and (7) m are the second and the thrd moments of the emprcal dstrbuton of the resduals. Snce v s symmetrcally dstrbuted, m s smply the thrd moment of the dstrbuton of u.the case m 0 mples that OLS resduals are negatvely skewed, and that there s evdence of techncal neffcency. In fact, f u 0 then v unegatvely skewed. The postve skewness n the OLS resduals,.e. m 0, suggests that the model s ms-specfed. Coell (1995) [5] proposed an alternatve test statstc. 1/ m b (6 m / N) 1/ Where, N s equal to the number of observatons. Under the null hypothess of zero skewness n the OLS resduals, (8) ~ 844 ~ m 0, the thrd moment of OLS resduals s asymptotcally dstrbuted as a normal random varable 1/ wth mean zero and varance (6 m / N ).Ths mples that the test statstc (10) s asymptotcally dstrbuted as a standard normal random varable N (0,1). Coell (1995) [5] presents Monte Carlo experments where these tests have the correct sze and good power. The asymmetry of the dstrbuton of the error term s a central feature of the model. The degree of asymmetry can be represented by the followng parameter. u v (9) The larger s, the more pronounced the asymmetry wll be. On the other hand, f s equal to zero, then the symmetrc error component domnates the one-sde error component n the determnaton of. Therefore, the complete error term s explaned by the random dsturbance v, whch follows a normal dstrbuton. Therefore has a normal dstrbuton. To test the hypothess that 0, we can compute a Wald statstc or lkelhood rato Test both based on the maxmum lkelhood estmator of Coell (1995) [5] tests as equvalent hypothess 0 aganst the alternatve 0, where u u v (10) A value of zero for the parameter ndcates that the devatons from the fronter are entrely due to nose, whle a value of one would ndcate that all devatons are due to techncal neffcency. The Wald statstc s calculated as W Y (11) Where,Y s maxmum lkelhood estmate of and s ts estmated standard error. Under H 0 s true, the test statstc s asymptotcally 0 dstrbuted as a standard normal random varable. However, gven that cannot be negatve, the test s performed as a one-sded test. The lkelhood test statstc s LR [ log( L ) log( L )] 0 1 (1) Where, log( L 0) s the log-lkelhood valued under the null log( L ) s the log-lkelhood value under hypothess and 1 the alternatve. Ths test statstc s asymptotcally dstrbuted as ch-square random varable wth degrees of freedom equal to the number of restrctons. Coell (1995) [5] notes that under the null hypothess 0, the statstc les on the lmt of the parameter space snce γ cannot be less than zero. He therefore concludes that the lkelhood rato statstc wll have an asymptotc dstrbuton equal to a mxture of ch-square dstrbuton.

5 4. Estmaton of parameter of Cobb-Douglas stochastc fronter model There are two man methods to estmate the parameter of stochastc fronter models s modfed ordnary least Squares method and other conssts of maxmum lkelhood method. A. Modfed Ordnary Least Squares In ths method all the assumptons of the classcal regresson model apply, wth the excepton of the zero mean of the dsturbances. The OLS estmator wll be a best lnear unbased and consstent estmate of the vector β. Problems arse for the ntercept term α. ts OLS estmate s not consstent. B. Maxmum Lkelhood Estmator method In ths method consstent estmates of all the parameters of the fronter functon can be obtaned smply usng a modfcaton of the OLS estmator. However the dstrbuton of the composed error term s asymmetrc (because of the asymmetrc dstrbuton of the neffcency term). A maxmum lkelhood estmator that takes nto consderaton ths nformaton should therefore gve more effcent estmates, at least asymptotcally. Ths has been nvestgated by Greene (1980a, b) [0] who argues that the Gamma dstrbuton s one of the dstrbutons whch provdes a maxmum lkelhood estmator wth all of the usual desrable propertes and whch s charactersed by a hgh degree of flexblty. Ths dstrbuton should therefore be used to model the neffcency error term. However, t has been notced that the flexblty of the Gamma dstrbuton can make the shapes of statstcal nose and neffcency hardly dstngushable IV. Result and Dscussons In ths study present methodology s appled to fve selected metrologcal staton of Marathwada regon. In ths regon Latur dstrct shows the hghest soybean crop producton dstrct of Marathwada average producton of soybean of Latur s 01 tonnes.n that skews and kurtoss value s 1.08 and 0.66 respectvely ths ndcate that ncreasng the crop producton of soybean. Area per hector of soybean of Latur s hgher than other dstrct. Table (1) show the descrptve statstcs of selected varable of dstrct of Marathwada regon. Table 1: Descrptve statstcs selected varable of metrologcal staton (dstrct) Staton Varable Mean StDev Mnmum Maxmum Skewnes Kurtoss Yeld Ranfall Area Latur Max Temp Total could cover Max. Humdty Wnd. Speed Wnd. Drecton Yeld Ranfall Area Parbhan Max Temp Total could cover Max. Humdty Wnd. Speed Wnd. Drecton Yeld Ranfall Area Nanded Max Temp Total could cover Max. Humdty Wnd. Speed Wnd. Drecton Yeld Ranfall Area Osmanabad Max Temp Total could cover Max. Humdty Wnd. Speed Wnd. Drecton Yeld Ranfall Area Hngol Max Temp Total could cover Max. Humdty Wnd. Speed Wnd. Drecton ~ 845 ~

6 Multple Lnear Regresson model (MLR) In multple regressons a common goal s to determne whch ndependent varables contrbute sgnfcantly to explanng the varablty n the dependent varable. A goal n determnng the best model s to mnmze the mean square error (MSE), whch would ntern maxmze the multple correlaton value (R).In ths study we have use crop yeld of soybean s dependent varable and other ten are ndependent varable for fttng multple lnear regresson model of two dstrcts such as Latur and Hngol. The Latur dstrct s hghest soybean crop producton n Marathwada regon Table : show the parameter estmates of ftted Multple Regresson model for measurng the clmatc effect on soybean crop of Latur dstrct (staton) Coeffcent Estmate Std. Error t-value Pr(> Intercept Ranfall Area Max.Temp Toat.could Max. Humdty Wnd Speed Wnd Drecton Dry Spell Mn.Temp Mn.Humdty Table. Show that multple regresson equaton of soybean crop of Latur dstrct. n ths method we have use crop yeld of soybean s dependent varable and other ten are ndependent varable. The ranfall, Area, max. Temp, total could cover, wnd speed, wnd drecton s postve effect on crop producton of soybean and Max & mn humdty are negatve effect on crop producton of soybean. In multple regresson model, multple R-squared s 0.905, whch mples that 90.05% of the total varatons can be explaned by the repressor varables and Adjusted R-squared s , whch mples that % varaton can be explaned by the repressor after adjustment varables after adjustments and from the overall test. The level of sgnfcance (α=5%) used for ths study. Multple regresson model s good ft to soybean crop producton of Latur dstrct on bass value of multple R-square and adjusted R-square. Table : show the parameter estmates of ftted Multple Regresson model for measurng the clmatc effect on soybean crop of Hngol dstrct (staton) Coeffcent Estmate Std. t- Error value Pr(> Intercept Ranfall Area Max.Temp Toat.could Max. Humdty Wnd Speed Wnd Drecton Dry Spell Mn.Temp Mn.Humdty Table. Show that multple regresson equaton of soybean crop of Latur dstrct. n ths method we have use crop yeld of soybean s dependent varable and other ten are ndependent varable. The ranfall, Area, max.temp, total could cover, wnd speed, wnd drecton s postve effect on crop producton of soybean and Max& mn humdty are negatve effect on crop producton of soybean. In multple regresson model, multple R-squared s 0.905, whch mples that 90.05% of the total varatons can be explaned by the repressor varables and Adjusted R-squared s , whch mples that % varaton can be explaned by the repressor varables after adjustments and from the overall test. The level of sgnfcance (α=5%) used for ths study. Stochastc Fronter model The Cobb-Douglas stochastc fronter model s used for measure the producton effcency of soybean crop yeld producton. In ths method we have use crop yeld s output varable and other ten such as, ranfall, dry spell, max. Temp, mn temp, wnd speed, wnd drecton, total could cover, area of crop growng (hector), max. Humdty and mn. humdty Table 4: show the parameter estmates of ftted Cobb-Douglas Stochastc Fronter model. for measurng effcency of clmatc effect on soybean crop of Latur dstrct (staton) Coeffcent Estmate Std. Z- Error value Pr(> Intercept <.e-16 Ln(Ranfall) e-11 Ln(Area) e-1 Ln(Max.Temp) Ln(Toat.could) Ln(Max. Humdty) Ln(Wnd Speed) Ln(Wnd Drecton) e-11 Ln.(Dry Spell) Ln(Mn.Temp) Ln (Mn.Humdty ) SgmaSq gamma The Cobb-Douglas stochastc fronter model assumng a logarthmc transcendental (translog) technology, the parameters estmates of the producton fronter and the techncal neffcency component are presented n Table 4. The statstcally sgnfcant parameters at the level of 5% are essentally related to soybean crop producton, as well as the measures of regonal techncal neffcency expressed by dummy varables. The Lkelhood Rato statstc presents sgnfcant value at 1% level, ndcatng effects of techncal neffcency n the model. The results show that of techncal neffcency. Presents approxmated value of 0,90. Ths result ndcates that 90% of total composed error varance of the producton functon s explaned by the varance of the techncal neffcency term. Ths reveals the mportance to ncorporate techncal neffcency n the producton functon. The estmated coeffcents of all the parameters of producton functon of ranfall, area, max. temp, total could cover are postve there s postve effect on soybean crop producton and other s negatve effect on soybean crop producton. ~ 846 ~

7 The value Coell s test H0: γ= 0, whch gves the value of gamma s1 and t s p-value for testng the hypothess s , ndcates hghly nsgnfcant and all of the devatons arses due to stochastc nose and there s no techncal neffcency. The test statstcs value of lkelhood rato test s Pr(> 0=0.458 whch mples to accept the null hypothess that there s no producton neffcency of soybean producton due to clmates change n Marathwada regon. Table 5: Show the parameter estmates of ftted Cobb-Douglas Stochastc Fronter model. For measurng effcency of clmatc effect on soybean crop of Hngol dstrct (staton) Coeffcent Estmate Std. Error Z-value Pr(> Intercept <.e-16 Ln(Ranfall) Ln(Area) Ln(Max.Temp) Ln(Toat.could) Ln(Max. Humdty) Ln(Wnd Speed) Ln(Wnd Drecton) Ln.(Dry Spell) Ln(Mn.Temp) Ln(Mn.Humdty ) SgmaSq gamma The Cobb-Douglas stochastc fronter model assumng a logarthmc transcendental (translog) technology, the parameters estmates of the producton fronter and the techncal neffcency component are presented n Table 5. The statstcally sgnfcant parameters at the level of 5% are essentally related to soybean crop producton, as well as the measures of regonal techncal neffcency expressed by dummy varables. The Lkelhood Rato statstc presents sgnfcant value at 1% level, ndcatng effects of techncal neffcency n the model. The results show that of techncal neffcency. Presents approxmated value of 0,95. Ths result ndcates that 9.5% of total composed error varance of the producton functon s explaned by the varance of the techncal neffcency term. Ths reveals the mportance to ncorporate techncal neffcency n the producton functon. The estmated coeffcents of all the parameters of producton functon of ranfall, area, max. humdty, total could cover are postve there s postve effect on soybean crop producton and other s negatve effect on soybean crop producton. The value Coell s test H0: γ= 0, whch gves the value of gamma s0.976 and t s p-value for testng the hypothess s 0.084, ndcates hghly nsgnfcant and all of the devatons arses due to stochastc nose and there s no techncal neffcency. The test statstcs value of lkelhood rato test s Pr(> 0=0.0118whch mples to accept the null hypothess that there s no producton neffcency of soybean producton due to clmates change n Marathwada regon. Concluson The am of ths study s to estmate clmatc effect on soybean crop yeld and measures ts crop producton effcency by usng multple lnear regressons and stochastc fronter model.in multple regresson model we have use crop yeld s dependent varable and other ten are ndependent varable n Latur and Hngol dstrct respectve. The result show that ranfall, area, dry spell max temp are postvely effect on soybean crop producton and mn humdty, mnmum temp, total could are negatve effect on crop producton of soybean. The ftted multple lnear regresson model of parameter estmate by usng least square regresson method. result show that value of multple R-square of Later and Hngol dstrct are 0.91 and 0.90 respectvely ths ndcate that ftted model can explan more varablty of repressors varable. The Cobb-Douglas stochastc fronter model s appled to Latur and Hngol of soybean crop producton for measure producton effcency. The result show that effcency of Latur and Hngol dstrct are 0.86 and 0.84 respectvely Ths result ndcates that 86%and 84% of total composed error varance of the producton functon s explaned by the varance of the techncal neffcency term. The estmated coeffcents of all the parameters of producton functon of ranfall, area, max. temp, total could cover are postve there s postve effect on soybean crop producton and other s negatve effect on soybean crop producton. The value Coell s test H0: γ= 0, whch gves the value of gamma testng the hypothess to check stochastc nose and there s no techncal neffcency. The statstcs value of lkelhood rato test s Pr(> 0= whch mples to accept the null hypothess that there s no producton neffcency of soybean producton due to clmates change n Marathwada regon. Ths study s help farmers n makng decson concernng wth ther crop & regonal and local plan they depend on ranfall of Marathwada regon References 1. Agner DJ, Lovell CAK, Schmdt P. Formulaton and Estmaton of Stochastc Fronter Producton Functon Models, Journal of Econometrcs Abramovtz M. Resource and output trends n the unted states snce 1870, Amercan Economc Revew. 1956; 46:5.. Afrat SN. Effcency estmaton of producton functons, Internatonal Economc Revew 1, Oct, 197, Battese GE, Coell TJ. A Stochastc Fronter producton Functon Incorporatng a Model for Techncal Ineffcency Effects, Workng Paper n Econometrcs and Appled Statstcs, No.69, Department of Econometrcs, Unversty of New England, Armndale Coell T. Estmators and hypothess tests for a stochastc fronter functon: A Monte carol analyss, Journal of Productvty Analyss. 1995; 6: Coell TJ. Battese GE. Identfcaton of Factors whch Influence the techncal neffcency of Indan Farmers. Australan Journal of Agrcultural Economcs. 1996; 40(): Coell VJ. Gude to Fronter Verson 4.1: A Computer Programmed for Stochastc Fronter Producton and Cost Functon Estmaton, Department of Economcs, Unversty of New England, Armanda, Australa Constantn PD, Martn LM, Rvera EBBR. Cobb- Douglas, Translog Stochastc Producton Functon and Data Envelopment... Journal of Operatons and Supply Chan. 9. Díaz-Mayans MA, Sánchez R. Temporary employment and techncal effcency n Span, Internatonal Journal of Manpower. 004; 5(): ~ 847 ~

8 10. Farrell MJ. The measurement of productve effcency, Journal of the Royal Statstcal Socety (A, general) 10, pt. 1957; : Ngwenya, S, Battese GE, Flemng EM. The Relatonshp between Farm Sze and Techncal Ineffcency of Producton of Wheat Farmers n Eastern Orange Free State, South Afrca, Agrekon (South Afrca), FAO, Schmdt P, Lovell, CAK. Estmatng Techncal and Allocatve Ineffcency Relatve to Stochastc Producton and Cost Fronters. Journal of Econometrcs North-Holland 1979; 9: Paulo Dutra Constantn, Dogenes Leva Martn, Edward Bernard Bastaan de Rvera Y Rvera, Cobb- Douglas. Translog Stochastc Producton Functon and Data Envelopment Analyss n Total Factor Productvty n Brazlan Agrbusness, the flagshp research journal of nternatonal conference of the producton and operatons management socety, 009, (). 14. Racsko P, Szedl L, Semenov L. A seral approach to local stochastc weather models. Ecologcal Modellng. 1991; 57: Robert M. Solow A Contrbuton to the Theory of Economc Growth The Quarterly Journal of Economcs. 1956; 70(1): Koop G, Osewalsk J, Steel M. The component of output growth: A stochastc fronter analyss, Oxford Bulletn of Economcs and Statstcs 1999; 61: Km Y. Schmdt P. A revew and emprcal comparson of Bayesan and classcal approaches to nference on effcency levels n stochastc fronter models wth panel data, Journal of Productvty Analyss 000; 14: Kumbhakar S, Lovell C. Stochastc Fronter Analyss, Cambrdge Unversty Press, Cambrdge, Satyvan Yashwant, Sananse SL. Comparng Neural Network and Multple Regressons Models to Estmate Monthly Ranfall Data, Internatonal Journal of Scence and Research (IJSR). 015, 4(1). 0. Wllam H. GREENE maxmum lkelhood estmaton of econometrc fronter functons Journal of Econometrcs. 1980; ~ 848 ~