An Analysis on Taiwan Broiler Farm Prices under Different Chicken Import Deregulation Policies

Transcription

1 An Analysis on Taiwan Broiler Farm Prices under Differen Chicken Deregulaion Policies MENG-LONG SHIH Deparmen of Social Sudies Educaion Naional Taiung Universiy Taiwan SHOUHUA LIN Deparmen of Applied Foreign Languages Naional Kaohsiung Universiy of Applied Sciences Taiwan Biing-Wen Huang *, Wei-Yu Hu Deparmen of Applied Economics Naional Chung-Hsing Universiy Taiwan * Corresponding auhor * bwhuang@dragon.nchu.edu.w CHI-I HSU Deparmen of Informaion Managemen Kainan Universiy Taiwan imchsu@mail.knu.edu.w Absrac: - According o he changes of chicken impor policy, i could be divided ino hree periods: conrol impor, quoa impor and free impor. This sudy adops GARCH models and weekly daa o analyze he broiler farm prices during hese differen periods. Based on he empirical es resuls, i indicaes ha here is larger long-run persisence effec of shocks in he free impor period. The reacions of broiler farm prices o he lag broiler farm prices, survey number of chick afer six weeks, pig prices, colorful broiler farm prices, chick prices and feedsuff prices are differen during hese hree periods. Keywords: - regulaion periods, Volailiy, Price response, GARCH model ISSN: Issue 9, Volume 5, Sepember 8

2 . Inroducion The broiler indusry is one of he imporan animal husbandry programs in Taiwan wih he chicken impor policies undergoing hree differen conrol sages during Taiwan s applicaion for enry ino he WTO. As a WTO official member; Taiwan has o comply wih relevan norms o fulfill is commimen of membership negoiaions and herefore, opens is domesic primary-producs marke o he globe. The governmen enforced sric conrol on chicken impor before is enry ino he WTO in. Afer enering he WTO, Taiwan gradually opened is impor of foreign chickens. The quoa impor period was from o 4, during which he impor quoa was 4% o 8% of he domesic chicken consumpion. Wihin he quoa, he imporers were levied wih lower ariffs, and vice versa. During ha period, he impor of chickens increased from 9, ons in o 45, ons in 4, which was an increase of more han double. Beginning 5, Taiwan opened is marke compleely, hus he broiler indusry faced severe challenges. According o he saisics of he Council of Agriculure, he oupu value of he broiler indusry in 6 reached NT$.6 billion, and i became he hird larges indusry in he animal husbandry. Alhough he indusry developmen was in a growing rend, he slaugher number increased from 6 million in 996 o 7 million in 4, hen decreased o 8 million in 6 due o he open marke policy. The govenmen s policy was o avoid he impac of impored chicken on domesic chicken prices. According o changes o chicken impor conrols, he ime series for farm prices of broilers can be divided ino hree periods: conrol impors, quoa impors, and free impors. All social circles pay aenion o he effecs of such conrol changes affecing he farm prices of broilers. Before opening chicken impors, he changing rend of domesic farm prices of broilers could be analyzed based on he quaniy of conrolled impors and he siuaion of domesic broiler breeding. During he period of conrolled impors, he changes o farm prices of broilers wen from NT$4.7/kg in 999 o NT$34.5/kg in, wih an average price, over hree years, approximae NT$36.36/kg (USD.35/kg). However, afer he opening of chicken quoa impors in, he quaniy of domesic chicken had gradually increased, exering heavy pressure on he domesic farm price of broilers, for example: from NT$36.53/kg in o NT$36.8/kg in 4, and he average price over hree years was approximaely NT$35.4/kg, if compared be wih average price of he conrolled impor period, he price decreased approximaely NT$ average per kilogram. Afer opening chicken impors in 5, i caused volailiy o he domesic farm price of broilers; domesic poulry-breeders did no adop he measure of reducing producion o reac o he relaively lower producion price because of free chicken impors increase in quaniy, which would have offse he effecs of domesic poulry-breeders reducions exered on price. I can be seen ha i is more difficul o analyze he farm prices informaion abou broilers afer opening he marke o chicken impors; moreover, volailiy and reacion o indusry relevan facors may be differen from former sages. Based on hese facs, under differen chicken impor conrol measures, he invesigaion of he difference beween volailiy and reacion of broilers farm price becomes an imporan research issue. Previous researches ofen adop he ime series mehod o analyze price and o learn he price change (e.g., [][][3][4][5]). However, various domesic documens pu emphasis on he effecs of opening chicken impors o he broiler indusry and pay lile aenion o he volailiy and reacion of farm prices of broilers during Taiwan s enry ino WTO [6]. There is a lack of research in comparing price changes of differen chicken impor sages; moreover, relevan demonsraion daa ofen adops monhly daa; whereas i is appropriae o use weekly daa in order o obain more accurae informaion of broiler prices, according o he characerisics of shor broiler producion cycles (approximaely six weeks). This hesis analyzes hese issues. Based on above menioned, his research adops weekly daa, dividing he changes of he farm price of broilers ino hree periods: conrol impors, quoa impors, and free impors o consruc a demonsraion model, which compares he differences beween price reacion and he volailiy of farm prices of broilers during he hree periods, and analyzes and explains he changes in he informaion of farm price of broilers afer imporing chicken freely. This can be regarded by relevan unis as reference of price informaion analysis. Therefore, demonsraion analysis, farm price of broilers daa will be used as he saionary uni roo es o consruc a model of farm prices of ISSN: Issue 9, Volume 5, Sepember 8

3 broilers. Generalized Auo-Regressive Condiional Heeroskedasiciy; (GARCH) based on es resuls of residual error heerogeneiy, which compares and analyzes price volailiy and reacion of he hree periods, hen uses differen predicive indices o compare he predicive abiliy of he demonsraion model of he differen periods, and proposes relevan suggesions on research resuls, providing poulry-breeders wih reference on producion sraegy and agriculural adminisraive unis o guide he indusry.. Research Mehod. Uni Roo Tes Because he farm price of broilers is ime series daa, i is necessary o deermine wheher i belongs o saionary before he analysis. Thus, uni roo ess should be conduced, or else, spurious regression may be resuled. In Dickey and Fuller s [7] uni roo es, ime series only concerned he firs-order auocorrelaion; such resuls may cause he residual iems o have significan auocorrelaion problem. In response o his problem, Dickey and Fuller [8] proposed ADF (augmened Dickey-Fuller es) mehod, which assumed ha he ime series is p-order auocorrelaion, and i adds differenial sub-iem of explained variable (Δy ) lag phases in he regression equaion, uses OLS o esimae he regression equaion, and selecs he model wih he opimum lag number of periods, in order o solve he auocorrelaion problem of residual iems, and ensure he residual iems o conform o he whie-noise process. The es models can be divided ino hree ypes: () A model wih non-inercep iems and ime rend iem Δy = ρy p + βiδy i + () ()A model including inercep iems bu no ime rend iem Δy = α + ρy p + βiδy i + (3) A model including inercep iems and ime rend iem () Δy = α + ρy p + α + β iδy i + (3) where, Δy i = y i y i, Δy is he lag iem, p is he opimum lag number of periods. The null hypohesis and alernaive hypohesis are: H : ρ = H : ρ < If es saisics are greaer han DF criical value, i does no rejec he null hypohesis, indicaing ha y has uni roo phenomenon, which is a non-saionary ime series.. Residual auoregraion and normaliy ess In he analysis of he ime-series model, he residual iems may appear auoregraion, and require furher consideraion on wheher i conforms o he assumpion of saionary disribuion. In oher words, he residual iems mus conform o he non auocorrelaion and normal disribuion rules, or else, he parameers esimaed by he model may be invalid. Therefore, i is assumed ha a residual iem in he regression equaion is no whie noise or linear, he residual squares iems would be inerruped. Common diagnosic saisics are Q es and JB es, of which he Q es checks wheher he residual iems in he diagnosic model have auoregraion, whereas he JB es checks wheher he residual iems conform o normal disribuion. The auoregraion Ljung-Box Q es equaion is wrien as: p ri Q( p) = T ( T + ) ~ χ ( p) ( T i) (4) where, T is he sample number, p is he r auoregraion exponen number, i is ih-order auoregraion coefficien, Q( p) is he χ disribuion close o freedom p. The null hypohesis and alernaive hypohesis are wrien as: H : Q( p) < χ ( p) H : Q( P) > χ ( p) If auoregraion Q( p) saisics is less han he χ disribuion saisics, i canno rejec he null hypohesis. Thus, he model residual iems do ISSN: Issue 9, Volume 5, Sepember 8

4 no have auoregraion problem. In normaliy ess, when calculaing JB saisics, he coefficien of he skewness (R) of he residual iem and coefficien of kurosis (S) should be calculaed firs. If he here are n parameers o be esimaed in he model, and he oal sample number of he residual iems is T, hen he JB es is: T n JB = R + ( S 3) 6 4 (5) The null hypohesis and alernaive hypohesis are, respecively: : H JB < χ ( ) H : JB > ( ) χ The disribuion of JB saisics belong o χ disribuion of freedom, hus, when JB saisics are less han he χ disribuion es saisics, i canno rejec he null hypohesis, herefore, he model s residual iem is a normal disribuion..3 Residual Heerogeneiy Tes In he ime series model, residual iems may have heerogeneiy. The residual heerogeneiy means ha, he variance of ime series residual iems is no fixed consans, and changes wih ime, hus, his research used he LaGrange Muliplier (LM) es proposed by Engle [9] o es he significance of he regression coefficiens of he residual squares of differen periods, hen deermine if he variance beween he wo ime periods conain he heerogeneiy phenomenon, ha is, if i has ARCH (auoregressive condiional heeroskedasiciy) effec. The mehod provides regression o he curren residual squares, inercep iems, and lag,, LL, q iems, and obains he coefficien of deerminaion R. Tes saisics was obained by muliplying he sample number o he coefficien of deerminaion ( T R ), which was closes o he χ disribuion of freedom q. This model is as follows: q = β + β s i + v (6) Where, is he sample number, is he residual squares series, obained from esimaed equaions. The null hypohesis and alernaive hypohesis are: H : TR < χ ( q) H : TR > χ ( q) If TR saisics is less han he χ disribuion es saisics, i canno rejec he null hypohesis, hereby, indicaing ha his model does no have ARCH effec..4 GARCH Mode Bollerslev [] expanded he ARCH model o propose he GARCH model, which allows he condiional variance of error iem o cover more pas informaion. The ypical GARCH (p,q) model is shown as following:; y = xb + (7) q p = + α + α i i β i h h i (8) where α >, α i > β i > o Eqs.(7) and (8) are average and variance equaions, respecively. The necessary and sufficien condiion of he condiional variance α i + β i < of he GARCH (p,q) is. In α i he variance equaion, here are ARCH ( ) and GARCH ( α i + βi ) effecs. ARCH effecs show he shor run persisence of shocks and GARCH effecs show long run persisence of shocks. These wo effecs can be used o measure he volailiy of farm price of broilers. To choose he lag period of broiler farm price, he Schwarz Bayesian Informaion Crierion(SBIC) is adoped. The decision-making process is firsly used o esimae he model wih differen lag periods. Then he error sum of squares (SSE) is used o calculae SBIC values. The model wih lag periods has minimum SBIC values. Then which lag periods will be chose as model s opimum lag number of periods. The SBIC crierion is defined as following: SBIC = T ln( SSE ) + mln(t ) (9) where T is he sample size, SSE is model s error sum of squares, and m is oal number of esimaed parameers. In addiion, for he model diagnosis he esimaed model mus es wheher is error iem q p ISSN: Issue 9, Volume 5, Sepember 8

5 exiss auocorrelaion or no, examine i is belong o normal disribuion, and verify condiion variance does no have heerogeneiy. Those could apply Ljung-Box s Q saisics, Jarque-Bern s JB saisics, and Lagrange Muliplier s (LM) saisics o es. If all saisics would no refuse null hypohesis, i would show he esimaed model is suiable..5 Crieria of model predicive abiliy In order o deermine if he esimaed model has good analogy abiliy, beside having R and all esimaed parameer no equal o zero, wheher daa generaed from he whole model conforms o he facs should also be considered. Therefore, i is necessary o weigh he accuracy beween he model s predicive value and acual values. This research adoped roo mean square error (RMSE), meaning absolue error (MAE), and mean absolue percenage error (MAPE) o evaluae he model predicabiliy. The measuremen definiion is as follows: Roo mean square error RMSE = N N = Mean absolue error MAE = N Y Y N = Y Y Mean absolue percenage MAPE = N N = Y Y Y () () () 3.Empirical Analysis 3.. Empirical Model Specificaion Variables affecing he farm price of broilers can be divided ino inpu coss (e.g. chick price, feed price), oupu coss (e.g. broiler producion quaniy), and relevan produc prices (e.g. colored chicken price, pig price), and impor quaniy, in addiion o prior periods price. Regarding farm price of broilers, poulry-breeders ofen consider prior period locaion of producion price as reference a sale ime when hey se sale price. Regarding inpu coss, chick and fodder prices are main broiler breeding coss, when chick and fodder prices change he breed cos of broilers will change and exers is effecs on farm price of broilers. Regarding oupu, when he supply of broilers is high, he price will be reduced in order o increase sales; whereas when he supply is low and falls shor of demand, price will be increased. Regarding relevan producs, he main succedaneum are colored chickens and pigs, when he price of colored chickens or pigs increases, i will decrease he demand of colored chickens or pigs o buy broilers, which will increase he farm price of broilers. Regarding impor quaniy, he impor quaniy of chickens affec he oal supply of domesic chickens; if he quaniy of impor chickens increases, i will bring pressure o decrease he farm price of broilers. According o above effecs, he variables of Taiwan farm price of broilers, and resuls of uni roo ess, his research will se he demonsraion model as following: Δ m price = b + b Δprice + λ dsq + λimq + λ3 pkp + λ4 + λ scp + λ6cfp + 5 (3) q p = + α α i i + cop h βih i (4) (3) Expresses he reacion equaion of he farm price of broilers, among which endogenous Δprice variable price ( price ) meaning NTD/kg afer underaking order differences, meaning he variables of he farm price of broilers. Secondly, exogenous variable mean Δprice is he lag period of he farm price of broiler variables, dsq is a survey number afer six weeks of domesic chick (dsq), imq is impor quaniy of domesic chicken (imq), pkp is pig price, cop is colored chicken locaion of producion price, scp is chick price (NTD/a chick), cfp is chicken fodder price. (4) shows, condiional variances equaions, and curren condiion variance is affeced by prior period. residual squares and prior condiional variances. 3.. Sample Descripive Saisics The empirical daa of his research adoped weekly daa for analysis. The research period was from January 999 o July 7, and daa were colleced from saisical daa of he Taiwan Poulry Associaion and he Naional Animal Indusry Foundaion. The impor quaniy of chickens only has monhly daa, hus, he daa ISSN: Issue 9, Volume 5, Sepember 8

6 were revised o weekly daa based on he number of weeks in ha monh. The weekly ime series daa were divided ino hree periods. The firs period was he conrolled impor period, which was compared o he quoa impor of he second period, and adoped he daa of he same lengh (hree years), namely from January 999 o December, oaling 56 pieces of daa. The second period was he quoa impor period, from January o December 4 wih a oal of 56 pieces of daa. The hird period was he unresriced impor period, from January 5 o July 7, wih a oal of 35 pieces of daa.. The descripive saisics of he main variables are as shown in Table. In erms of he farm price of broilers, he average price during he unresriced impor period was he highes (NT$37./kg). As for oher variables, he average pig price was NT$43.9/kg during he conrolled impor periods, which was relaively lower han he average price during quoa and unresriced impor period. The average farm prices of colored chicken and chick were NT$5.763/kg and NT$3.4/chick, respecively, which were relaively higher han he average price of oher periods. As for chicken impor quaniy, he impor quaniy increased grealy during he unresriced impor period. The survey number of domesic chick afer six weeks (which is he possible supply of chicken during he curren period) was 3.78 million during he quoa impor period, which was more han he average quaniy of oher periods. As for fodder price, he average price was NT$.5/kg during unresriced impor period, which was higher han he average price during oher periods. Table Sample Descripive Saisics Conrol (//999-3//) Variable price dsq (million pcs/week) imq (million on/week) pkp cop scp (NTD/pc) cfp Average SD Quoa (//-3//4) Variable price dsq (million pcs/week) imq (million on/week) pkp cop scp (NTD/pc) cfp Average SD Free (//5-3/7/7) Variable price dsq (million pcs/week) imq (million on/week) pkp cop scp (NTD/pc) cfp Average SD Noe: price is farm price of broilers, dsq is chick survey number of afer six weeks, imq is chicken impor quaniy, pkp is pig price, cop is colored pig place of producion price, scp is chick price, cfp and is chicken fodder price 3.3 Tes of farm prices characerisics of broiler This research firs adoped ADF o es wheher source daa is saionary. If variances are found o be non-saionary, he source daa were reaed wih firs-order differenial, and furher esed. The opimum lag number of periods is deermined based on AIC values. The daa on farm price of broiler can be divided ino hree sages: conrolled impor, quoa impor, and unresriced impor. The es model can be divided ino hree ypes: inerceped iems and rend iems, inerceped iems and non-rend iems, and non-inerceped iems and rend iems. As shown in Table, he uni roo es on source daa indicaes ha under he significance levels of %, 5%, and %, he source daa resuls canno rejec he null hypohesis. In oher words, he source daa of all hree sages have uni roo phenomenon, indicaing ha hey are non-saionary daa. Therefore, he source daa were differeniaed, and esed according o hree models of ADF es. The resuls showed ha i is under significance level of %, all rejecing he null hypohesis. In ISSN: Issue 9, Volume 5, Sepember 8

7 oher words, he firs-order differenial of he hree periods has no uni roo phenomenon, indicaing ha hey are saionary daa. Therefore, fuure sudies will conduc empirical analysis on he farm price of broilers in he hree periods by firs-order differenial daa. Table Tes of farm price of broilers Source daa s Conrolled Quoa Unresrice d impor Non-inercep iem and rend iem -.3 (3) -.38 (7) -.87 () Inercep iem and non-rend iem -.95 (7) -.6 (7) () Inercep iem and rend iem (6) (7) -.49 () Non-inercep iem and rend iem *** (3) *** (6) -5.8*** () Daa afer differenial Inercep iem and non-end iem -3.85*** (3) -5.7*** (6) -5.6*** () Noe:. The number in (.) indicaes he lag number of periods.. * means significance level of %, * * means significance level of 5%, * * * means significance level of %. Inercep iem and rend iem *** (3) -5.87*** (6) -5.8** () 3.4 Resuls of residual error auocorrelaion and saionary es In erms of auocorrelaion and saionary es of residual iems, as shown in Table 3, in Ljung-Box s Q saisics, he residual iem lag 6 periods and periods of he farm price of broilers during he conrolled impor, quoa impor, and unresriced impor periods were seleced, and he LB (6) saisics were 9.77, 6.35, and , respecively. LB () saisics were 4.64, 9.543, and 64.9, respecively. The resuls suggesed ha he residual iems of he ime series daa of he farm price of broilers have auocorrelaion during he hree periods. As for saionary ess, JB saisics were 4.496, 35.38, and 56.36, respecively, during he hree periods, indicaing ha he residual iems of he ime series daa of he farm prices during he hree periods did no conform o normal disribuion. Table 3 Self-correlaion and normaliy es of farm price of broilers Q es &JB es LB(6) LB() JB s Conrol 9.77 *** 4.64 *** *** Quoa 6.35 *** *** *** Unresriced impor *** 64.9 *** *** Noe: LB indicaes Ljung-Box s Q saisics; JB indicaes Jarque-Bera saisics.resuls of residual error heerogeneiy es In residual iem heerogeneiy es, as shown in Table 4, under significance levels of %, 5%, and %, LM es values were.67, 3.36, and 4.63, respecively. Thus, he residual iem variance of he farm price of broilers during he hree periods had heerogeneiy. As shown by he residual error heerogeneiy es, all hree periods are suiable o be esed by he GARCH model. Table 4 Heerogeneiy Tes of Farm price of broilers Heerogeneiy Tes s LM Tes Conrol.67*** Quoa 3.36*** Unresriced impor 4.63*** ISSN: Issue 9, Volume 5, Sepember 8

8 3.5. Empirical Resuls The esimaion model coefficien is calculaed by using E-View5. economeric sofware, which chooses he opimum model based on he principle of SBIC minimum value, resuls are shown in Table. The suiable models for he hree periods are ARIMA(,,)-GARCH(,), ARIMA(,,)-GARCH(,) and ARIMA(,,)-GARCH(,). The coefficien value of he condiional variance equaion all α mee he resricive requiremens, ha is >, α i > β >, i α and i + β i <, all mee he saionary condiions. Regarding he goodness of fi abou hose models, according o LB, JB, and LM saisics of he hree periods, which show ha error erms are no exi auocorrelaion and has a normal disribuion, and is variances does no have heerogeneiy characerisics. Thus, esimaed resuls of he hree periods are appropriae The volailiy of broiler farm prices Price volailiy is no fixed, and he volailiy of each period has Auocorrelaion. The analysis of he GARCH model on price volailiy is mainly based on variance equaions, which include boh ARCH and GARCH effecs. As for ARCH effec, during he hree periods, he curren residual iem condiional variances have significan relaionship wih he previous residual squares, is values are.68 ( α ),.35 ( α + α ) and.56 ( α ), respecively, indicaing ha he volailiy of previous farm price of broilers has duraive effec over a shor ime on he curren volailiy of farm price, among which he volailiy of quoa impor period is larger han ha of he conrolled and unresriced impor periods. I indicaes ha, here are higher duraive effecs o shor volailiy periods of farm price of broilers during he conrolled impor periods. As for he GARCH effec, previous residual squares and previous residual iem condiion variances have significan effecs on he curren residual iem condiional variance during he hree periods. The values were.988,.954, and.973, respecively, indicaing ha he volailiy of previous farm price of broilers has duraive effec over he long erm on he curren volailiy of farm price, among which he volailiy of he unresriced impor period is larger han ha of he quoa impor and conrolled impor periods. I shows ha he previous volailiy of farm price of broilers has longer duraive effecs on he curren farm price during he unresriced impor period. If he previous farm price is more volaile, he curren farm price would be volaile as well, indicaing ha he impac on he unresriced impor of chicken afer Taiwan s enry ino WTO has led o longer impac on he farm price of broilers. In oher words, poulry-breeders have o face higher risks in producion under he volailiy of farm prices The reacion analysis of broiler farm prices The expeced resuls of he farm price reacion o main exogenous variable changes during he hree periods are shown in Table 5. The changes of farm price of broilers lagging one period have significan effec on he change amouns of he curren farm price during he hree periods. The values were.936,.9 and.77, respecively. The unresriced impor period has he sronges effec; when he change amoun of lag from one period of farm price increases NT$, he change amoun of he curren farm price will be increased by NT$.936. As for broiler producion quaniy, he chick survey number afer six weeks showed significan negaive effecs on he change amouns of he curren farm price of broilers, indicaing ha when he supply of he broilers of curren period may increase, i would move he change amouns of he curren farm price o a negaive rend, hus he chicken prices would decrease. The esimaed coefficien during he unresriced impor period is relaively small, indicaing ha he supply of broilers has a small effec on he curren price during ha period. In erms of chicken impor, he chicken impor quaniies of he hree periods and he previous period have no significan effecs on he change amouns of he curren farm price of broilers. There are wo possibiliies for his: one is ha he impored chicken were released o he marke gradually, hus he shor-erm effec was no significan; anoher is ha he scale of domesic poulry-breeders was adjused appropriaely, so ha he impor of chicken had no significan effec on he domesic farm price of broilers. In erms of subsiuion produc prices, pig prices have significan posiive effec on he change amouns of he curren farm price of broilers during he unresriced and conrolled impor periods. The esimaed coefficiens were.8 and., respecively, among which he effec during he unresriced impor period was ISSN: Issue 9, Volume 5, Sepember 8

9 Table 5 Esimaed Resuls of Demonsraion Model during Conrol, Quoa and Free s Conrol Quoa Free Model ARIMA(,,)- GARCH(,) ARIMA(,,)- GARCH(,) ARIMA(,,)- GARCH(,) Price Reacion Equaion Inercep Iem 5.64 (5.887) Farm price of broilers(lag one period) Chick Survey Number afer six weeks Chick Survey Number afer six weeks(lag one period ) Chicken Quaniy (.49) *** (.69) ** (.8) *** (.3) (.389) *** (.) *** (.) *** (.99) * (.3) (.99) *** (.7) *** (.37) *** (.3) *** (.9) s Conrol Quoa Free Model ARIMA(,,)- GARCH(,) ARIMA(,,)- GARCH(,) ARIMA(,,)- GARCH(,) Chicken Quaniy (lag one period) -.59 (.3) -.9 (.) -. (.9) Pig Price Colored Chicken Place of producion price. (.4) ***. (.9). (.3) Chick Price (lag six periods).64 (.5).5 (.7) **.3 (.3) Chicken Fodder price (lag four periods) (.5) Chicken Fodder price (lag five periods).59 (.55) (.4) * (.4) * (.4) * (.4) * s Model Condiion Variance Equaion α α α β α i + β Conrol ARIMA(,,)- GARCH(,). (.7).68 (.69) *** (.63) ***.973 Quoa Free ARIMA(,,)- GARCH(,) ARIMA(,,)- GARCH(,).64 (.) **. (.33) **.93 (.59) ***.649 (.6) ** (.4) *.56 (.73) *** (.55) **.988 s Conrol Quoa Free Model ARIMA(,,)- GARCH(,) ARIMA(,,)- GARCH(,) ARIMA(,,)- GARCH(,) SBIC Value Log likelihood LB(6) LB() JB LMS saisics Noe:, * means noable level %, * * means noable level 5%, * * * means noable level %., Number in (.) is progressive sandard deviaion ISSN: Issue 9, Volume 5, Sepember 8

10 he sronges. Thus, when pig prices increase, he change amoun of he curren farm price of broilers would increase as well. As for farm price of colored chicken, here has posiive effec on he change amoun of he curren farm price of broilers during he quoa and unresriced impor periods. The esimaed coefficiens were.5 and., respecively, indicaing ha here was small effec of he change amoun of he curren farm price during he unresriced impor period. As for inpu coss, since he growing period of chicks requires five o six weeks before hey can be sold, he chick prices lag for six periods may affec he curren farm price of broilers. The chick price has posiive effec on he change amoun of he curren farm price of broilers during he unresriced impor period. The esimaed coefficien was.9, indicaing ha when he price of six lag periods increases, he change amoun of he curren farm price of broilers would also increase. In erms of fodder price, chicken fodder is used during he mos period of he breeding course, hus, he lag of four o five weeks of fodder price may have effec on he curren farm price of broilers. Fodder price has significan posiive effec on he change amoun of he curren farm price of broilers during he unresriced impor period. The esimaed coefficien was.6, indicaing ha when he fodder price lagging 4 periods increases, he change amoun of he curren farm price of broilers would also increase Model predicive abiliy and performance The comparaive analysis of he predicive abiliies of he models during he hree periods was conduced by roo mean square error (RMSE), meaning absolue error (MAE), and mean absolue percenage error (MAPE). The resuls are shown in Table 6. The predicive abiliy indices of RMSE, MAE, and MAPE during he unresriced impor period were.368,.88, and.757, respecively, which were less han hose of oher periods. Since he smaller he index value, he beer he predicive abiliy, hus, ARIMA (,,)+GARCH(,) se in he unresriced impor period was a beer model o predic he fuure farm price of broilers. This research used four ses of weekly daa from Augus 7 for ou-of-sample predicion, in order o compare he predicive value esimaed from he model and he acual values, and furher analyze he predicive performances of farm prices of broilers models during he unresriced impor period. This mehod is based on progressive renewal predicive mehods, and subsiues he observaional value of he T period and T- period ino model o find T+ predicive values, in order o deduce he resuls. The resuls are shown in Table 7. The four ses of weekly daa from Augus 7 generaed he predicive values of farm price of NT$38.6/kg, NT$38.9/kg, NT$4.6/kg and NT$4.8/kg, respecively. If compared wih acual values, he predicive error of each week was NT$./kg, NT$.4/kg, NT$.6/kg and NT$.4/kg, respecively. The error value was wihin NT$.3/kg, and fairly close o he acual values. Therefore, he empirical model of farm price for he unresriced impor period is feasible o analyze he changes of domesic farm prices of broilers in he fuure. Empirical model ARIMA (,,) -GARCH (,) ARIMA (,,) -GARCH (,) ARIMA (,,) -GARCH (,) Table 6 Predicive abiliy indices of he empirical model s Conrol impor period RMSE MAE MAPE Quoa impor period RMSE MAE MAPE Unresriced impor period RMSE MAE MAPE ISSN: Issue 9, Volume 5, Sepember 8

11 Table 7 Ou-of-sample predicive performance of farm price of broilers during he unresriced impor period Farm price Unresriced Acual value Predicive value Predic -ive error impor 8/7 (s week) /7 (nd week) /7 (3rd week) /7 (4h week) Noe: he predicive value is convered from he change of prices. 4. Conclusions The price volailiy analysis on he farm price of broilers showed significan shor-erm duraive effec (ARCH effec), of which he volailiy during he quoa impor period was larger han ha of oher wo periods, and ha of he unresriced impor period was he lowes. The long-erm duraive effec (GARCH effec) was he highes in he unresriced impor period, indicaing ha if he farm price of broilers in he previous period was subjeced o impac, here was long-erm duraive effec on he curren farm price of broilers. Especially afer opening unresriced impor of chicken, he volailiy of he curren farm price of broilers is affeced by ha of he previous period. A he same ime, here is long duraive volailiy of he farm price, indicaing ha he change of impor policy would lead o increasing risks faced by poulry-breeders. The reacion analysis of farm price of broilers showed ha, he effec of he previous farm price on he curren farm price is greaer during he unresriced impor period. The effec of he possible supply quaniy on he curren farm price is smaller during he unresriced impor period. I indicaes ha afer Taiwan s enry ino he WTO, broiler breeders may be more careful on he planning of he producion quaniy. There is no grea increase or decrease as in he conrolled impor periods, hus, he effec on he curren farm price is moderae. Alhough he quaniy of impored chicken has no significan effec on he farm price of broilers, here is sill pressure he farm price. In oher words, imporers may sock impored chickens and sell according o he marke demand, which will affec he marke of aggregae supply and price. As for he effec of he farm prices of pig and colored chicken, he effec of pig price on he curren farm price of broilers during unresriced impor period is relaively larger han ha of oher wo periods. However, he effec of he farm price of colored chicken on he farm price of broilers is relaively smaller han ha of oher wo periods. In erms of he inpu cos variables, chick price and chicken fodder price have significan posiive effecs on he farm price of broilers only during he unresriced impor period. In oher words, when he chick price and chicken fodder price increase, he predicive breeding cos increases, which will increase he farm price of broilers. Based on his, he change of chicken impor policy would affec he farm prices of broilers, among which here is significan reacion o he chick and chicken fodder prices during he unresriced impor period. Because he raw maerials of fodder in Taiwan are mos impored, he volailiy of inernaional crop prices would affec he domesic fodder price; i no only affecs he broiler breeding coss, bu also increases he chick prices due o he rise of chick breeding cos, and indirecly affecs he broiler breeding cos. Because he farm price of broilers in Taiwan lacks a price warning price sysem, he GARCH model consruced by his research can be used o predic price under unresriced impor of chicken. I is an urgen issue for agriculural bureaus o esablish a price warning price sysems for domesic farm price of broilers, in order be updaed wih he price flucuaion and relaed informaion, provide reference daa o breeders on decision-making relaed o producion and agriculural bureaus for indusrial guidance, and promoe he susainable developmen of broiler indusry in Taiwan. References: []. M. T. Hol and S. V. Aradhyula, Price Risk in Supply Equaion: An Applicaion of GARCH Time-Series Models o he U.S. Broiler Marke, Souhern Economic Journal, Vol.57, No., 99, pp ISSN: Issue 9, Volume 5, Sepember 8

12 []. M. T. Hol and S. V. Aradhyula, Endogenous Risk in Raional- Expecaions Commodiy models: A Mulivariae Generalized ARCH-M Approach, Journal of Empirical Finance, Vol.5, 998, pp [3]. J. Elza, S. Pivac and J. Arneric, Hisorical and Prognosic Risk Measuring Across Socks and Markes, WSEAS TRANSACTIONS on BUSINESS and ECONOMICS, Vol.8, Issue4, 7, pp [4]. Y. H. Liang, Evoluionary Neural Nework Modeling for Taiwan Sock Price Index Predicion: a Comparison of SARIMA Model, WSEAS TRANSACTIONS on MATHEMATICS, Vol.4, Issue 3, 6, pp [5]. Y. H. Liang, Combining Seasonal Time Series ARIMA Mehod and Neural Neworks wih Geneic Algorihms for Sock Price Index Forecasing, WSEAS TRANSACTIONS on MATHEMATICS, Vol.6, Issue6, 7, pp [6]. Y. H. Chen and K. D. Lin, Effecs on Domesic Broiler Indusry afer Enry ino WTO, Collecion of Essays on Agriculural Finance, Issue43,, pp [7]. D. A. Dickey and W. Fuller, Disribuion of he Esimaors for Auoregressive Time Series wih a Uni Roo, J. Amer Sa Asso, Vol.74, 979, pp [8]. D. A. Dickey and W. Fuller, Likelihood Raio Tes for Auoregressive Time Series wih a Uni Roo, Economerica, Vol.49, 98, pp [9]. R. F. Engle, Auoregressive Condiional Heeroskedasiciy wih Esimaes of he Variance of Unied Kingdom Inflaion, Economerica, Vol.5, 98, pp []. T. Bollerslev, Generalized Auoregressive Condiional Heeroskedasiciy, Journal of Economerics, Vol.3, 986, pp ISSN: Issue 9, Volume 5, Sepember 8