A PANEL DATA ANALYSIS OF THE DETERMINANTS OF FARMLAND PRICE: AN APPLICATION TO THE EFFECTS OF THE 1992 CAP REFORM IN BELGIUM

Transcription

1 A PANEL DATA ANALYSIS OF THE DETERMINANTS OF FARMLAND PRICE: AN APPLICATION TO THE EFFECTS OF THE 1992 CAP REFORM IN BELGIUM Romain Duvivier, 1 Frédéric Gaspar and Bruno Henry de Frahan Universié caholique de Louvain Paper prepared for he XIh EAAE Congress (European Associaion of Agriculural Economiss) The Fuure of Rural Europe in he Global Agri-Food Sysem Copenhagen, Denmark, Augus 23-27, 2005 Copyrigh 2005 by R. Duvivier, F. Gaspar, and B. Henry de Frahan. All righs reserved. Readers may make verbaim copies of his documen for non-commercial purposes by any means, provided ha his copyrigh noice appears on all such copies. 1 Corresponding auhor : Romain Duvivier, Unié d économie rurale, Place Croix du Sud 2-be 15, B-1348 Louvain-la-Neuve, Belgium ; duvivier@ecru.ucl.ac.be

2 Absrac This sudy idenifies he effecs of he 1992 and subsequen CAP reforms on arable farmland price in Belgium. We firs propose a brief lieraure review of sudies idenifying he deerminans of farmland price. Aferwards, we use a panel daa se o esimae a capializaion model of farmland price. We firs show ha he compensaory paymens inroduce by he 1992 CAP reform exer a posiive effec on he arable farmland price in Belgium. We also idenify a srucural break in he land price equaion afer he adopion of he new suppor insrumens as well as a regional break beween Wallonia and Flanders. Keywords: farmland price, land marke, Common Agriculural Policy, capializaion of agriculural suppor Jel classificaion: Q10, Q15, Q18 1. Inroducion The main obecive of mos agriculural suppor policies consiss in rising farmers income. One of he maor criics opposed o such policy is he weak efficiency of he income ransfer. An imporan par of he income ransferred o he agriculural secor leads o larger consumpion of variable inpus and creaes a ren ha capializes ino asse value. As a fixed facor, land is seen as very sensiive o his capializaion process. The rise of farmland value can be seen as favorable for an individual farm operaor. Indeed, he land value plays a crucial role as collaeral o obain credis and deermines he purchasing power of he farmer afer reiremen. However, ha he benefis of he suppor are capialized in farmland value rises some equiy consideraions. Because wo hirds of Belgian farmland are culivaed by enans, no land owners, an imporan par of he suppor is capured by non-operaors. As inflaed land values also imply inflaed land rens, his ype of policy migh even make enan farmers worse off. Moreover, if a farm operaor purchases farmland wih he expeced fuure suppors already capialized in is price, he does no fully benefi from he suppor. Those benefis are used o finance a more expensive invesmen (Barnard e al., 2001). Finally, he inflaed farmland price due o direc paymens makes more difficul for a young farmer o sar his own business. In his paper, we are ineresed in analyzing he effecs of he 1992 and subsequen Common Agriculural Policy (CAP) reforms on he price of arable farmland in Belgium. The adopion of he 1992 reform corresponds o a srong inflexion in he CAP suppor insrumens. Relying on price suppor since is foundaion, he CAP progressively swiches o direc paymens decoupled from producion. To undersand he reasons of his significan change, we have o remember ha he 1992 CAP reform also corresponds o a paricular momen of he rade negoiaion: he ongoing Uruguay Round. For he firs ime agriculure was on he global rade agenda. Many specialiss poined ou ha he elemens of he 1992 reform permied o achieve an agreemen a he Uruguay Round. Oher announced obecives of he reform were o promoe rural developmen and conain environmenal problems. The adopion of he 1992 and subsequen reforms leads o a subsanial cu of he price suppor bu compensaed wih direc paymens. For cereal, oilseed and proein crops, he amoun of direc paymens depend upon he culivaed area. Depending on he exen of he subsidised area, he direc paymens can be condiional o a se aside provision. For beef producion, he pre-exising premiums per head of cale are increased. However, hese premiums are condiional o a limi densiy of cale per hecare of fodder crops. There is also a ceiling on he number of cale heads eligible for premium. We see ha he reform induces a sronger link beween he eligibiliy o he benefis of he agriculural suppor and he use of land. Moreover, i is relaively well esablished ha he effec of agriculural suppor on he price of land is sronger in case of direc paymens decoupled from producion. As a 2

3 resul, we can suspec ha he CAP 1992 reform has caused a subsanial change in he deerminans of farmland price. Figure 1 shows he evoluion of he arable farmland price in real erm during he period in Belgium. We can see ha afer a coninuous decrease up o 1994, he evoluion of farmland price ends o sabilize and hen increase in he years following he 1992 CAP reform. During he period, he real price of arable land increases a a yearly average of abou 3.8% in Belgium. Showing he evoluion of he Belgian agriculural area, Figure 2 gives us anoher insigh of he effec of he 1992 CAP reform. We can observe a significan change in he evoluion of he series during he period succeeding he reform. Alhough ha movemen does no correspond o farmland acquisiion, i allows us o conclude ha he main message of he reform was well undersood by farmers and landowners. They anicipae ha forhcoming agriculural suppor will be linked o he farmland use. Consequenly, hey are more rigorous in reporing heir agriculural acreage Figure 1 : Arable farmland price evoluion ( ; in consan euro by hecare) Source : Belgian Naional Insiue of Saisics (NIS) Figure 2 : Evoluion of he Belgian Agriculural Area ( ) 1440 Agriculural Area (in 1000 hecares) Source : Belgian Agriculural Census, Belgian Naional Insiue of Saisics (NIS) The remainder of his paper is organized as follow. The nex secion presens a brief lieraure review of sudies idenifying he deerminans of farmland price. The following secion discusses a heoreical model of farmland price formaion. The fourh secion presens he empirical model and he daa used in he empirical analysis. The fifh secion presens and inerpres he resuls of he empirical analysis. 3

4 We show ha he compensaory paymens exer a posiive effec on arable farmland prices. We also show ha he compensaory paymens have a disinc effec each year and ha he exen o which he agriculural suppor is capialized ino farmland value is region specific. The final secion of he paper offers a summary and concluding remarks. 2. Lieraure Review Mos research aemping o idenify and quanify he deerminans of farmland price rely on he capializaion approach. This approach assumes ha he price of farmland equals he presen value of all fuure expeced cash flow aached o he use of land for producive purposes. In his conex, an increasing farmland price should be explained by an increasing land ren. In USA, agriculural economiss observed ha he evoluions of he real land value and agriculural income go in he same direcion from 1910 o Those rends convince agriculural economics o rely on he capializaion approach. However, a divergen evoluion of land value and agriculural income is observed since Because of he observed decreasing agriculural income and increasing land price he classical model was reeced. Several alernaives hypohesis were proposed o explain he evoluion of farmland price. Feldsein (1980) proposes a firs alernaive explanaion. He poins ou ha he increasing farmland price observed in he sevenies ook place during a period characerized by a srong inflaion. Feldsein (1980) proposes a porfolio choice model wih wo asses: a classical financial asse and land. He shows ha an anicipaed inflaion could lead o a decrease in he acualizaion rae applied o land and explain an increase in farmland price. Oher auhors explain ha as a real asse wih fixed supply, land ends o hold is real value during inflaionary periods. Consequenly, here is an inflaionary hedging moive o buy land during an inflaionary period (Casle e al., 1982). Mos sudies aemping o evaluae he effec of inflaion on farmland prices found an insignifican effec. Alson (1986) shows ha he inflaion has a significan bu negaive effec on farmland price. On he basis of his resul, Feldsein s hypohesis is reeced. For Melichar (1979), he diverging rends of agriculural income and farmland price canno conduc o reec he basic capializaion model. Firs, we need o consider ha he ne farm income usually used as a measure of he reurn o land is an inappropriae indicaor. The ne farm income corresponds o he reurn o land bu also he reurn o farm labor and managemen. The land ren is he ne farm income lef afer subracing he reurn o farm labor and managemen. Second, he share of each ype of remuneraion in he ne farm income canno be considered as consan hrough ime. During he second half of he weny cenury, he capial-labor subsiuion phenomenon leads o an increase in he share of capial and, consequenly, land in he ne farm income and a decrease in he share of farm labor. Consequenly, i is possible o reconcile boh a decrease in he ne farm income and an increase in he land ren explaining an increase in land value. Following Melichar s argumens, Alson (1986) shows ha beween 1962 and 1982 he periods characerized by an increasing land value are also characerized by an increasing land ren. Moreover, he rae of growh of he land ren and he rae of growh of land value are never saisically differen. A number of sudies have aemped o esimae he exen o which suppor policies increase farmland price. Mos of hose sudies invesigae he relaionship beween he benefis of agriculural suppors and he value of land using he presen value approach. In he land marke, he paricipans accep o pay a higher price as long as hey expec o yield a larger sream of cash flow. The subsidies aimed o suppor he agriculural secor increase he cash flows aached o he propery of farmland. Consequenly, he agriculural suppor also increases he land value. In wha follow, we presen he mos significan sudies aemping o provide a quaniaive esimae of he exen o which governmen paymens are capialized in farmland value. Goodwin e al. (1992) esimae an empirical model relaing land value o expeced level of producer suppor, expeced yield and expeced producer prices ne of governmen suppor in six whea producing regions beween 1979 and The governmen suppor is proxy by he producer subsidy equivalen. This sudy shows ha a one percen increase in he producer subsidy equivalen increases land value by 0.38%. Barnard e al. (1997) regress he cropland value on governmen subsidies and oher characerisics like agriculural produciviy, non-agriculural influence and sae-specific insiuional environmens. 4

5 In heir sudy, governmen paymens are measured by he couny-level averages of he annual amoun of direc paymens received par acre. The auhors carry he same analysis for weny U.S. regions. They show ha he elasiciy of cropland value o he governmen subsidies ranges from 0 o 0.69 according o he region. This sudy also poins ou ha he sensibiliy of farmland value o governmen suppor is spaially variable. Two elemens can explain his spaial variabiliy; 1) wheher or no he dominan crops in a given region are eligible o he suppor, and 2) he level of agronomic flexibiliy of a given region ha deermines he abiliy o adus oupu in response o changing governmen policy. Goodwin e al. (2005) use farm level daa o esimae he capializaion rae of governmen paymens ino farmland value. They firs noe ha he formaion of land value is based upon expecaions abou he long-run sream of reurns aached o land. Assuming ha he realizaion of a paricular source of reurn correcly reflecs wha is expeced in he long run could lead o an error in variable problem. Error in variable problem resuls in inconsisen esimaors. To represen he expeced paymen, he auhors use a four-year average value of he realized paymens a he couny level. They firs consider he aggregaion of all suppor programs ino a single caegory. They show ha using he acual realized paymen of each farm as proxy of he expeced ren gives a coefficien of Wih he couny average, hey obain a coefficien of They conclude ha he use of he farm observed paymens may resul in an aenuaion bias ha forces he implied capializaion raes oward zero. Because he four-year couny average of he paymens is more represenaive of he long run benefis, a larger coefficien is obained. They also show ha he rae of capializaion of one dollar of paymen is program specific. Breaking ou he overall measure of governmen paymens ino heir individual componens, hey find a specific rae of capializaion for each ype of suppor program. Like Barnard e al. (1997), Goodwin e al. (2003) show ha he exen o which suppor policies affec land value is spaially variable. They also poin ou ha he implied effec of a suppor insrumen on land value differs from year o year. Clark e al. (1993a) poin ou a condiion for he presen value model o hold. The ime series of ren and land values mus be ineger of he same order for he series of he ren o explain he movemen in farmland value. Campbell e al. (1987) rigorously demonsrae ha for he presen value model o hold i is necessary ha ime series of land prices and expeced rens are coinegraed. On he basis of his condiion, Clark e al. (1993b) show ha farm incomes alone canno explain land values. Because he ime series of land value is found o conain one uni roo while he series of farm income is saionary, he presen value model is reeced. However, a ime series aggregaing farm income and governmen paymens is found o have a single uni roo. Moreover, hey find some empirical evidence suggesing a coinegraion relaionship beween land value and income including subsidies. They conclude ha farm income alone canno explain land value. However, considering ogeher farm incomes and subsidies, he presen value formulaion canno be reeced. This could imply ha subsidies are capialized ino farmland value. Mos sudies ineresed in he explanaion of farmland prices also ake ino accoun he demand of land for alernaive uses (residenial, recreaional, ec.). Generally, he variable inended o capure he influence of non-agriculural demand appears o be significan. Goodwin e al. (2005) poin ou ha each addiional inhabian per square mile in a given counry adds $1.85 in value o an acre of farmland. A 1% increase in he populaion growh rae adds $64 per acre o he farmland value. Planinga e al. (2002) show ha he price of land reflecs curren as well as poenial uses of land. If a poenial buyer expecs a conversion of farmland for a more profiable purpose in a forhcoming year, he curren land value also reflecs a conversion value associaed wih he fuure poenial use of land. The auhors show ha he imporance of ha conversion value in farmland value depends upon wo main facors: he proximiy of large urban area and he availabiliy of farmland. Wih French daa, Cavailhès e al. (2003) show a negaive relaionship beween he price of farmland and he disance from a cenral business disric (CBD). The price of farmland is he highes he closes o he CBD. When he disance o CBD increases, he price of land declines reflecing a decrease in he conversion value. A abou 40 km from he CBD, he price of farmland becomes independen of he disance. The auhors conclude ha when farmland is so far from a CBD, he value of ha farmland in agriculural producion is he sole deerminan of farmland price. 5

6 3. The Theoreical Model : A Presen-Value Model for Farmland Prices Like for he value of any asses, he classical economic heory suggess ha he value of a piece of agriculural land should be equal o he presen value of all fuure expeced cash flows. The cash flows associaed wih he use of land for agriculural producion are called he rens (R), i.e., he income lef afer subracing hem from revenue of all oher producive facors (labour, capial and variable inpus). If he price of land is below his fundamenal value, buying land allows exra profis. In his case, he compeiion among poenial buyers of land is such ha he price increases. We follow Goodwin e al. (2002, 2005) and propose an equaion explaining he formaion of arable land price of he following form: P = 0 R * + e r d (1) The acual value of land (P ) is equal o he discouned expeced value of all fuure expeced rens from * agriculural producion ( ). The erm r represens he discoun rae, e.g., he rae used by land R + marke paricipans o discoun fuure paymens. Ineresed in esimaing he exen o which he CAP compensaory paymens are capialized ino farmland value, we break he expeced ren from agriculural producion ino wo componens: a firs componen arising from he sale of he agriculural producion in he marke (MS) and a second componen linked o he agriculural suppor (AS): P = 0 E e a ( MS + ) e d + A ( AS + ) e d 0 Boh funcions E (.) and A (.) represen an expecaion formaion process for he expeced revenue arising from he marke and from he agriculural suppor. For example, he erm A (AS + ) is he level of agriculural suppor expeced in period for he period +. The erms e and a are respecively he discoun raes apply o he revenue from he sale o he marke and o he agriculural suppor. In any presen value model, he weigh of any income source depends upon he opporuniy cos of capial and a facor reflecing he expeced variabiliy of ha paricular income source. Indeed, if he agens are assumed o be risk averse, he anicipaed variabiliy of a fuure cash flow may affec he deerminaion of he discoun rae of ha paricular cash flow. If a given source of paymen is seen as a more variable and/or more emporary, ha paricular paymen akes a smaller weigh in he presen value. Consequenly, we define each discoun rae as follows: (2) e = o + ve a = o + va Where he erm o is he opporuniy cos of he capial engaged in agriculure in and he erms ve and va respecively reflec he expeced variabiliy of he revenue arising from he sale o he marke and from he agriculural suppor. In equaion (2), he erm: A ( AS + ) e a represens he presen value in of he agriculural suppor expeced for he period +. We use a disinc discoun rae for each income source o reflec ha he expeced variabiliy is paymen specific. If a given income sream is seen as more variable, i is discouned a a larger rae. We also consrain he erms E (MR + ) and A (AS + ) o be consan independenly of he period + a which he paymens is expeced : E (MR + ) = E (MR) A (AS + ) = A (AS) 6

7 However, o remain as general as possible, we allow he paricipans o he land marke o expec increasing (or decreasing) cash flow for each source of paymen. For example, if he agriculural suppor is expeced o grow a a consan rae ga each year, he presen value in period of a cash flow expeced for he period + is be given by: A ( o ve ) ga ( o ve + ga ) ( AS ) e e = A ( AS ) e A similar expression can be obained if land marke paricipans anicipae increasing (or decreasing) revenue from he sale in he marke a a consan rae ge. Wih he las modificaions, equaion 2 becomes: P = 0 E ( o ve + ge ) ( o va + ga ) ( MS) e d + A ( AS) e d 0 where ge and ga are respecively he rae of growh expeced in for he sources of paymen E (MR) and A (AS). Recognising ha each inegral corresponds o he presen value of a perpeual consan flow, equaion (3) simplifies as follows: (3) P = o ge = α E ( MS ) ve E β ( MS ) + o A ( AS ) + 1 va ga A ( AS ) (4) where α and β respecively represen he rae of capialisaion of he expeced fuure revenues arising from he sale o he marke and he rae of capialisaion of he expeced agriculural suppor. Boh erms α and β represen he reciprocal of he corresponding discoun facor. 4. The Economeric Model and Variable Definiion The previous secions sugges a number of facors ha may influence he price of arable farmland. This secion defines he empirical model and he daa used o esimae he deerminans of arable farmland price. The empirical analysis proceeds from equaion (5) ha is esimaed using a balanced panel of 42 Belgian disrics, each of which has daa for he enire period from 1980 o PAL + β MA + β MS + u i i = βremri + βccpi + βd PDi 1 + βl PRLi 1 + β ppigi 1 m i s i i (5) The PAL i, variable is he average sale price of arable land (per hecare) observed in disric i during year. The variables EMR and CP are respecively used as indicaors of he expeced land ren from he sales o he marke and he land ren from he agriculural suppor. Due o limiaion in available daa, we choose o approach he realized land ren by he sum of he agriculural family income and land coss by hecare. Those daa are available in he yearly publicaions of he Belgian Cenre for Agriculural Economics. Because he compensaory paymens defined hereafer are also a componen of he agriculural family income, we subrac he variable CP of he realized land ren o obain he land ren arising from sales o he marke. Of course, ha variable is he curren realisaion of ha paricular ren and, consequenly, i is a wrong indicaor of he expeced land ren from sales o he marke. To approach he expeced marke ren, we follow Goodwin e al. (2005) and consruc a four year average of ha land ren realized during he curren and pas hree years. The variable CP is used o approach he level of suppor for cereals, oilseed and proein crops. Because i is impossible o obain he realized paymen per hecare, we choose o approach ha paricular ype of suppor by he following formula: 7

8 CP i = ϖ icpi wihϖ i = area i area wih = cereals, rapeseed, flax and fallow. The CP i and area i variable are respecively he amoun of suppor by hecare of crop and he area of ha crop in disric i during year. Boh variables PD i -1 and PRL i -1 are inroduced o conrol for he expeced fuure capial gains inheren in farmland in areas facing non-agriculural pressure. The variable PD is he populaion densiy per square kilomere. The variable PRL is he growh rae of he price of residenial land. We choose o express he price of residenial land in growh rae o break a poenial mulicollineariy beween PD and PRL since he correlaion beween PD and PRL is close o 0.7. The variables PAL, PD and PRL are direcly available in he yearly publicaions of he Belgian Naional Insiue for Saisics (NIS). The variable PIG corresponds o he pig densiy by hecare of agriculural area. The series is calculaed from he Belgian Agriculural Census (NIS). The pig densiy is inroduced o conrol for a possible capializaion of manure spreading ren in he land marke (Le Goffe e al., 2005). Wih all variables presened above farmland price is mainly explained by demand facors: agriculural profiabiliy, governmen suppor, compeing uses for land, ec. However, demand facors are he sole deerminans of farmland price if and only if farmland supply is perfecly inelasic. We, however, hypohesize ha he area of farmland enering ino he marke in a paricular place a a paricular dae also plays a role in he explanaion of farmland value. In oher words, we wan o conrol for supply facors explaining farmland price. The variables MS and MA are used o conrol for his supply. The variable MS represens he size of he arable farmland marke. I is approached by he fracion of arable farmland exchanged in a paricular disric a a paricular year. An increase in MS is expeced o induce a decrease in he farmland price. The variable MA is he average area of farms in he disric. This variable MA reflecs he pressure of agriculural aciviies on farmland marke. The variable MA is expeced o yield a negaive coefficien. The variables PIG, MA and MS are calculaed from he Belgian Agriculural Census of NIS. To correc for he effec of common inflaionary pressure, all moneary variable (PAL, EMR, and CP and PRL) are adused o 2000 euros using he Gross Domesic Produc deflaor published in he Belgosa daabase. For he esimaion procedure, all variables are ransformed ino naural logarihm o obain a direc inerpreaion in erms of elasiciy. 5. Esimaion of he Land Price Equaion 5.1. Specificaion Tess and Model selecion Our model of farmland formaion presened in equaion (5) can be esimaed in several ways. The appropriae esimaion mehod depends upon he srucure of he error erms, u i, and he correlaion beween he observed deerminans of farmland prices and he differen componens of he error erm. In wha follows, we wan o deermine he mos appropriae esimaion echnique. We sar by assuming a ime consan coefficien for boh he variable EMR and CP. The second column of Table 1 gives he OLS esimaion of he farmland price equaion. This model assumes ha all OLS assumpions be saisfied and ha he parameers of he equaion be consan for all ime periods and Belgian disrics. However, i migh be possible ha boh disricspecific and ime-specific unobservable facors affec on farmland price. In case of such ime and/or disric heerogeneiy in he equaion explaining he formaion of farmland price, OLS esimaors are biased. For ha reason, he possibiliy of a cross-secional or a ime specific heerogeneiy is considered in he four las columns. In he disric fixed effec (DFE) model, a dummy variable specific for each disric accommodaes for disric heerogeneiy. The fifh column of Table 1 presens resuls from he disric random effec (DRE) model where he unobserved disric effec is included in he error erm. This error erm is decomposed ino wo componens: a cross-secional specific error (v i ) and an individual error (ε i ). By i 8

9 inroducing he individual effecs in he error erm, we assume ha he inercep is a random oucome variable. The DRE and DFE models differ by one criical assumpion: he DRE model assumes no correlaion beween he cross-secional effecs and he observed regressors. This assumpion migh be criical for he presen sudy. Some unobserved parameers specific o a disric ha explain he formaion of land price migh also have an impac on some independen variables. For example, he agro-climaic condiions of a paricular disric migh influence boh farmland price and marke ren. In presence of such correlaion, he DRE esimaor is inconsisen while he DFE esimaor remains consisen. In he absence of such correlaion boh esimaors are, however, consisen, bu he DRE esimaor is more efficien. If we are convinced ha he disric heerogeneiy needs o be aken ino accoun, we need o decide wheher he DFE or DRE model is he more appropriae (Sevesre, 2002 ; Balagi, 2001). In a similar way, he possibiliy of a ime-specific heerogeneiy is considered in fourh and sixh columns of Table 1. The fourh column presens he resuls from he ime fixed effec (TFE) model, which accommodaes for ime heerogeneiy by inroducing a dummy variable specific o each ime series. The sixh column presens he resuls from he ime random effecs (TRE) model in which he error erm accommodaes for ime unobserved heerogeneiy. Table 1. Esimaion resuls of he farmland price equaion (5) using ordinary leas squares (OLS), ime-fixed effec (TFE), disric-fixed effec (DFE), ime-random effec (TRE) and disric-random effec (DRE) esimaors (Numbers in parenheses are p-value). VARIABLE OLS DFE TFE DRE TRE Expeced Land Ren from Marke Sales (EMR ) (0.464) Compensaory Paymen (CP ) (0.055) Populaion Densiy -1 (PD -1 ) (0.854) Price of residenial Land Growh -1 ( PRL -1 ) (0.619) (0.938) (0.689) (0.874) (0.916) Pig Densiy -1 (PIG -1 ) Mean Area (MA ) Marke Size (MS ) (0.527) (0.045) To selec he appropriae specificaion, wo quesions have o be addressed. Firs, is here evidence of a need o conrol for cross-secional heerogeneiy and ime-effecs? If no, we can rely on he OLS esimaor. Second, if such evidence exiss, can we reasonably assume ha he random disric and/or ime effec is he Bes Linear Unbiased Esimaor (BLUE)? To answer he firs quesion, we firs es he oin significance of all disric effecs in he DFE model. We perform his oin significance es by a radiional F-es. The value of he saisics is Compared wih a F(41, 875) we reec he null hypohesis a more han one percen significance level. We now es he need o conrol for ime effecs by esing he oin significance of all ime effecs in he TFE model. The value of he saisics is Compared wih a F(21, 895), we reec he null hypohesis. The wo F-ess indicae he need o accommodae wih boh unobservable disric heerogeneiy and unobservable ime effecs. We now have o decide on wheher a fixed or random effec specificaion is he more reliable one. For he unobservable ime effecs, we es he assumpion ha he ime specific effecs are uncorrelaed wih he independen variables. Under his assumpion, we choose he TRE model which is consisen and more efficien han he TFE model. Under he alernaive, we choose he TFE model, which remains consisen. The former assumpion is no direcly esable. Bu, Hausman (1978) suggess an indirec es of he null hypohesis. Under he null hypohesis, boh esimaors are consisen and so converge o he same value. If he difference beween 9

10 he wo esimaors can be considered as sysemaic, we reec he null hypohesis and prefer o focus on he TFE model sill consisen under he alernaive. The resuls of he esimaions repored in Table 1 indicaes a large difference in some of he coefficiens obained from he TFE and TRE models, a sign of failure of he null of no correlaion. The value of of he Hausman saisics confirms his inuiion. Under he null hypohesis, his saisics is disribued as a χ² wih K degrees of freedom equal o he number of regressors. Consequenly, he Hausman es reecs he null ha he difference in coefficiens is no sysemaic and we rely on he resuls on he TFE model. For he disric heerogeneiy, he Hausman saisics akes a value of Once again, i is a sign of failure of he no correlaion assumpion and we prefer o rely on he resuls of he DFE model. 5.2 Correcion for Heeroskedasiciy and Serial Correlaion The specificaion ess indicae he need o conrol for boh disric unobservable heerogeneiy and unobservable ime effecs. In boh cases, he fixed effec specificaion seems o be more reliable han he random effec specificaion. For ha reason, we presen he resuls of a wo-way fixed effec model. Moreover, we wan o say as general as possible and allow for he compensaory paymen o have a disinc effec for each year beween 1993 and Consequenly, we esimae he following equaion: PAL + β MA + β MS + u i i = αi + λ + βremri + βccpi + βd PDi 1 + βl PRL i 1 + β ppigi 1 where he regression disurbances, u i,, are assumed o be homoskedasic and uncorrelaed. If hose assumpions are no saisfied, he wo-way fixed effec model is sill consisen bu no more efficien. Before o presen he esimaion resuls, we es for a possible correlaion in he residuals. To es he null hypohesis of no firs-order serial correlaion in he residuals, we use he Balagi-Wu locally bes invarian es saisics. The saisics is disribued as an N(0,1). To es Ho: ρ = 0 agains Ha: ρ>0, we refer o he lower ail of he normal disribuion. If he alernaive is ρ<0, we compare he realisaion of he saisics wih he upper ail criical value of he disribuion (Balagi e al., 1999). The realizaion of he saisics is Consequenly, we reec he null hypohesis of no serial correlaion in he residuals. Anoher es proposed by Bhargava e al. (1982) is he generalizaion of he Durbin-Wason for he fixed effecs model. The realisaion of he saisics is As in he classical Durbin-Wason case, he null hypohesis of no correlaion in he residuals is reeced if he realisaion of he saisics is below he lower bound criical value, and is no reeced if he saisics is larger han he upper bound criical value. The criical value are repored in Bhargava e al. (1982). As he realisaion of he saisics is below 1.5, we srongly reec he null of no serial correlaion in he residuals. The reecion of he null hypohesis of no serial correlaion indicaes he need o correc he sandard errors for serial correlaion. Table 2 repors he resuls of he esimaion of he wo-way fixed effec model using robus sandard error. We assume ha he disurbance follows an AR(1) process wih a specific coefficien for each cross-secion. To correc for possible heeroskedasiciy, we assume ha each panel has is own variance. The esimaion resuls of equaion (6) are presened in Table 2. Esimaes of he disric and ime dummy variables (generally significan) are no presened here due o space limiaion. Table 2 shows ha he compensaory paymens are consisenly significan wih he correc sign from 1996 o Boh variables aimed o represen he supply side are also found o be significan wih he expeced sign. However, he variable EMR plays only a very weak effec on he price of arable farmland. An increase in EMR of 1% only induces a increase in farmland price of %. Moreover, EMR is only significan a 6% significan level. Boh variables represening he non-agriculural demand (PD -1 and PRL -1 ) are found o be insignifican. Tha PD -1 and PRL -1 are found insignifican and ha he coefficien of EMR is low and weakly significan do no mean ha hose variables are no explicaive of land price variaion. To undersand he source of hese counerinuiive resuls, he hird column of Table 2 presens he esimaion resuls of he following equaion: m i s i i (6) 10

11 PAL + β MA + β MS + u i = λ + βremri + βccpi + βd PDi 1 + βl PRLi 1 + β ppigi 1 m i s i i (7) Table 2. Esimaion resuls of he farmland price equaions (6), (7) and (8). Prais-Winson regression wih panel correced sandard errors and AR[1] disurbances (Numbers in parenheses are p-value). Variable Equaion 6 Equaion 7 Equaion 8 wih τ=1993 Equaion 8 wih τ=1994 Expeced Land Ren from Marke Sales (EMR ) (0.064) Expeced Land Ren from Marke Sales (EMR ) (<τ) Expeced Land Ren from Marke Sales (EMR ) ( τ) Compensaory Paymen 1993 (CP 1993 ) (0.280) (0.309) (0.313) (0.270) Compensaory Paymen 1994 (CP 1994 ) (0.214) (0.264) (0.251) (0.268) Compensaory Paymen 1995 (CP 1995 ) (0.123) (0.052) (0.043) (0.047) Compensaory Paymen 1996 (CP 1996 ) (0.011) (0.003) (0.002) (0.002) Compensaory Paymen 1997 (CP 1997 ) (0.010) (0.004) (0.003) (0.003) Compensaory Paymen 1998 (CP 1998 ) (0.002) Compensaory Paymen 1999, (CP 1999 ) (0.014) (0.001) Compensaory Paymen 2000 (CP 2000 ) Compensaory Paymen 2001 (CP 2001 ) Populaion Densiy -1 (PD -1 ) (0.697) Price of residenial Land Growh -1 ( PRL -1 ) (0.849) (0.781) (0.710) (0.739) Pig Densiy -1 (PIG -1 ) (0.840) Mean Area (MA ) (0.001) Marke Size (MS ) Equaion (7) only differs from equaion (6) by removing he disric dummies. Esimaion resuls of equaion (7) show a larger elasiciy for he expeced land ren from marke sales and indicae ha he populaion densiy has an expeced and significan effec. Tha he elasiciy of EMR is of smaller size in he esimaion resuls of equaion (6) indicaes ha he cross secional difference across disrics plays a very imporan role in explaining farmland value. In equaion (6), he disric dummies absorbed mos of he cross-secional effec of he expeced land ren from marke sales. 11

12 5.3 Tess of he Null Hypohesis of Non-Saionariy The classical panel-daa esimaion mehod presened upon relies on he assumpion ha he series are saionary. The presence of non-saionariy series could invalidae he resuls presened above. When boh dependen and independen variables are rend-dominaed, we are very likely o find significan coefficien and high R², even in he absence of causaliy beween hose rending variables. This phenomenon known as he spurious correlaion problem migh explain he very aracive resuls presened above. We, herefore, examine he order of inegraion of our series using he panel uni roo es proposed by Levin e al. (2002) and anoher similar es proposed by Im e al. (2003). In boh ess he null hypohesis assumes a uni roo in he daa generaing process. In he Levin e al. es, i is assumed ha he coefficien of he auoregressive process is common o each crosssecion. In conras, he Im e al. es allows a specific coefficien for each cross-secion. More deailed informaion on he consrucion of hose wo ess can be founded in Balagi (2001, chaper 12). For each es, we incorporae an inercep. I is lef o decide wheher or no i is appropriae o inroduce a ime rend. The inspecion of he daa indicaes o inroduce such a rend only for he EMR series. Of course, he absence of a significan rend in he series represening land prices mainly reflecs ha we have wihdrawn he effec of inflaion by working on real erm daa. Table 3 presens he resuls of he Levin e al. and he Im e al. ess of he null hypohesis of uni roo process for each variable. Boh ess reec he null of non saionariy for he series represening he price of arable land, he expeced land ren from marke sales and he price of compensaory paymens. Tha hose series are found o be ineger of he same order is o be linked wih he resuls of he sudy of Clark e al. (1993). We canno reec he assumpion ha he movemens in he expeced land ren from marke sales and he compensaory paymens explain he movemens in land price. For he series of PD, PIG and MA, he Im e al. es does no allow o safely reec he null of non saionariy. Table 3. Panel uni roo ess Levin, Lin and Chu Im, Pesaran and Shin Variable Tes saisic p-value Tes saisic p-value Price of Arable Land (PAL) Expeced land Ren from he marke sales (EMR) Compensaory Paymen (CP) Populaion Densiy (PD) Price of residenial Land Growh ( PRL) Pig Densiy (PIG) Mean Area (MA) Marke Size (MS) Tha hree series are non-saionary could invalidae he resuls presened upon. However, by forming he linear combinaion of he PD, PIG and MA series using he esimaed coefficien repored in Table 2 (hird column), we find ha he linear combinaion is saionary. The Levin e al. saisics akes a value of We can reec he null of non saionariy of he linear combinaion a more han 1% significan level. This means ha he lef and righ hand sides of equaion (7) are balanced. Hence, he resuls of equaion (7) presened in Table 2 canno be invalidaed by spurious correlaion. Unforunaely, i is no possible o show ha boh pars of equaion (6) are balanced by a similar argumen. Tha boh sides of equaion (6) are unbalanced makes any inference drawn from he resuls of equaion (6) quesionable. 12

13 5.4 Srucural Break Tes In wha follows, we wan o es he assumpion ha he 1992 CAP reform has caused a srucural break in he land price equaion. We expec o find ha he elasiciy of he variable EMR is smaller afer he 1992 reform. The assumpion of a decrease in he rae of capializaion of he expeced land ren from marke sales relies on wo argumens. Firs, because of he 1992 reform and he concomian price cu, he CAP insrumens are less able o sabilise agriculural incomes. Consequenly, he expeced land ren from marke sales could be expeced o be more volaile and, hence, have a larger capializaion facor aached o i. Second, he adopion of he Agreemen on Agriculure a he Uruguay Round ogeher wih he discussions concerning he fuure of he CAP insrumens indicae subsequen price cus and, hence, subsequen decrease in farm incomes. On he basis of equaion (4), we can conclude ha boh a larger anicipaed variabiliy of he expeced land ren from marke sales (increase in ve ) and an expeced decreasing land ren from marke sales (ge <0 for >1992) should induce a decrease in he rae of capializaion of he expeced land ren from marke sales. To es he exisence of a srucural break in a given year τ, we esimae an exended version of equaion (7): PAL + β MA + β MS + u i i = λ + βremri + βccpi + βd PDi 1 + βl PRL i 1 + β ppigi 1 where β r = β r if < τ and β r = β r + θ if τ. (8) Table 2 (fourh and fifh columns) presens he resuls of he esimaion of wo versions of equaion (8). The fifh column of able 2 shows he esimaion resuls of equaion (8) inroducing he possibiliy of a srucural break a he firs year of he inroducion of he reform, i.e.,1993. The elasiciy of EMR is quaniaively smaller in he period succeeding he break. We es he null of equaliy of he EMR elasiciy before and afer The saisics of he es akes a value of Compared wih a χ² wih one degree of freedom, we canno reec he null hypohesis of elasiciy equaliy. The las column of Table 2 shows he esimaion resuls of equaion (8) considering he possibiliy of a srucural break in Tesing again he assumpion of elasiciy equaliy, we find a saisics of Consequenly, we can reec he hypohesis of elasiciy equaliy. 5.5 Regional Effec of he CAP Compensaory Paymens Because he share of arable land planed in crops eligible o direc suppor (cereals, rapeseeds, flax and fallow) is greaer in he Belgian region of Wallonia (50%) han in he region of Flanders (29%), we check wheher or no he elasiciy of he CAP compensaory paymens have a regional-specific effec. We allow for all variables of equaion (8) o have a separae effec in each region. Esimaion resuls repored in Table 4 show ha he elasiciy of farmland price o compensaory paymens is sysemaically sronger in Wallonia han in Flanders. On he basis of ha resul, we canno reec ha he regional specialisaion of agriculure has an impac on he capialisaion rae of direc agriculural suppor. Resuls in Table 4 also show an increase in he rae of capialisaion of he compensaory paymens in boh regions. Those rends can reflec he expecaion of he land marke paricipans. An increase in he rae of capialisaion of he compensaory paymens could be explained by he belief ha his ype of direc paymens will coninue in he forhcoming years. Resuls in Table 4 also show ha he rae of capializaion of he expeced land ren from marke sales remains seady in Flanders. In conras, we observe a significan change in he rae of capializaion of ha expeced land ren in Wallonia. Before 1994, a 1% increase in he expeced land ren from marke sales leads o a 0.15 % increase in farmland price. Since 1994, he movemens in he expeced land ren from marke sales badly explain farmland price movemens. Once again his resul mus be linked o he agriculural specialisaion in each par of Belgium. Due o is greaer agriculural specialisaion owards farm aciviies suppored by inervenionis prices, an imporan par of he expeced land ren from marke sales arises from he price suppor in Wallonia. Consequenly, he 1992 and forhcoming announced inervenion price cus have a sronger impac on he rae of capialisaion of he expeced land ren from marke sales in Wallonia han in Flanders. m i s i i 13

14 Table 4. Esimaion of equaion (8) wih τ=1994 and allowing a regional specific elasiciy for each variable. Prais-Winson regression wih panel correced sandard errors and AR[1] disurbances (Numbers in parenheses are p-value). Variable Wallonia Flanders Expeced Ren from Marke Sales (EMR ) (<1993) (0.012) Expeced Ren from Marke Sales (EMR ) ( 1994) (0.158) Compensaory Paymen 1993 (CP 1993 ) (0.135) Compensaory Paymen 1994 (CP 1994 ) (0.094) Compensaory Paymen 1995 (CP 1995 ) (0.048) Compensaory Paymen 1996 (CP 1996 ) (0.004) Compensaory Paymen 1997 (CP 1997 ) (0.003) Compensaory Paymen 1998 (CP 1998 ) Compensaory Paymen 1999, (CP 1999 ) (0.002) Compensaory Paymen 2000 (CP 2000 ) Compensaory Paymen 2001 (CP 2001 ) Populaion Densiy -1 (PD -1 ) Price of residenial Land Growh -1 ( PRL -1 ) (0.992) Pig Densiy -1 (PIG -1 ) Mean Area (MA ) (0.004) Marke Size (MS ) (0.063) (0.003) (0.003) (0.123) (0.480) (0.259) (0.026) (0.021) (0.001) (0.016) (0.493) (0.019) 6. Summary and conclusion Few exising sudies provide informaion on he exen o which CAP suppor insrumens are capialized ino farmland sale price. In his sudy, we esimae he effec of he CAP 1992 reform on he price of arable land in Belgium. Indeed, afer a long period of declining prices, he real price of arable farmland increased a a yearly average of 3.8% during he period. The CAP 1992 and subsequen reform also correspond o an imporan inflexion of he CAP suppor insrumens. Saring from an inervenion sysem largely based on he price suppor, he CAP progressively shifs o direc paymens parially decoupled from producion. However, economic heory suggess ha farmland prices also reflec direc agriculural suppor. Using a panel of 42 Belgian disrics from 1980 o 2001, we firs observed ha he exchange price of arable farmland is affeced by he compensaory paymens. Depending on he year and region considered, he elasiciy of arable farmland price o compensaory paymens ranges from 0.12 o Those parameers are in he same range of hose found by oher sudies like, for example, Goodwin e al. (1992) and Barnard e al. (1997). This resul also indicaes ha, by creaing a ren ha capializes 14

15 ino land value, he new CAP insrumens also benefi o landowners. Because abou wo hirds of he Belgian agriculural land are rened by farmers, non-operaors capure an imporan share of he agriculural direc suppor. Our resuls also indicae ha a emporal variabiliy exiss in he elasiciy of arable farmland value o compensaory paymens. We show ha he sensiiviy of arable farmland values o compensaory paymens increases during he period. All resuls obained from he ime fixed effec model show ha, besides he ime dummies and oher conrol variables, he expeced land ren from marke sales exer an imporan effec on arable farmland price. Moreover, because resuls repored in Table 2 preven o reec he presence of a srucural break in he period subsequen o he CAP 1992 reform, we can conclude ha he inervenion price cu induced by he 1992 CAP reform conducs o a decrease in he rae of capialisaion of he expeced land ren from marke sales. Before he reform, an 1% increase in he expeced land ren from marke sales induces a farmland price increase of 0.24%. In he years following he reform, he farmland price elasiciy declines o a value of 0.18%. Resuls from Table 4 also show ha he sensiiviy of arable farmland price o boh expeced land ren from marke sales and compensaory paymens is region specific. The elasiciy of farmland price o compensaory paymens is much sronger in Wallonia han in Flanders. Second, he decrease in he capializaion rae of he expeced land ren from marke sales is mosly aribuable o a decrease in he capialisaion rae in Wallonia. The comparison of he resuls of he wo-way and one-way ime effec models drawn from Table 2 provides us anoher imporan informaion concerning he deerminans of farmland prices. The oneway ime model gives a sronger and more significan elasiciy of farmland prices o expeced land rens from marke sales han he wo-way fixed effec model. To undersand his, we need o consider ha our measured expeced land ren from marke sales has wo componens. The firs componen is a sricly srucural and ime-invarian componen. The second one is a ransiory and ime varying componen. In he wo-way fixed effec model, he disric dummies absorb he srucural componen of he expeced land ren from marke sales. In conras, in he one-way fixed effec model, he elasiciy of farmland price o he expeced land ren from marke sales oally reflecs he srucural componen. As a resul, we can conclude ha he srucural deerminans of he expeced land ren from marke sales are he sole deerminans of farmland prices. Farmland price is insensiive o shorrun expeced ren flucuaions. This resul is coheren wih he heory saing ha long-run expecaions of rens are he rue deerminans of curren land prices. Since our sample covers only 20 years, he long run expecaions are absorbed in he disric fixed effecs. References Balagi, B.H. (2001). Economerics Analysis of Panel Daa. U.K.: John Wiley & Sons. Balagi, B.H. and Wu, P.X. (1999). Unequally spaced panel daa regression wih AR(1) disurbances. Economeric Theory 15: Barnard, C.H., Nehring, R., Ryan, J. and Collender, R. (2001). Higher cropland value from farm program paymens: Who gains? Agriculural Oulook 286: Barnard, C.H., Whiaker, G., Wesenbarger, D., and Ahearn, M. (1997). Evidence of capializaion of direc governmen paymens ino U.S. cropland values. American Journal of Agriculural Economics 79: Bhargava, A., Franzini, L. and Narendranahan, W. (1983). Serial correlaion and he fixed effecs model. Review of Economic Sudies 49: Campbell, J.Y. and Shiller, R.J. (1987). Coinegraion and ess of presen value models. Journal of Poliical Economy 64:

16 Casle, J.S. and Hoch, I. (1982). Farm real esae price componens, American Journal of Agriculural Economics, 64: Cavailhès, J. and Wavresky, P. (2003). Urban influences on periurban farmland prices. European Review of Agriculural Economics 30: Clark, J.S., Fulon, M.E. and Sco, J.T. (1993a). The inconsisency of land values, land rens and capializaion formulas. American Journal of Agriculural Economics 75: Clark, J.S., Klein K.K. and Thompson, S.J. (1993b). Are subsidies capialized ino land values? Some ime series evidence from Saskachevan. Canadian Journal of Agriculural Economics 41: Feldsein, M. (1980). Inflaion, porofolio choice and he prices of land and corporae sock. American Journal of Agrivculural Economics, 62: Goodwin, B.K. and Oralo-Magne F. (1992) The capializaion of whea subsidies ino agriculural land values. Canadian Journal of Agriculural Economics 40: Goodwin, B.K., Mishra, A.K. and Oralo-Magné, F.N. (2003). Wha s wrong wih our models of agriculural land values? American Journal of Agriculural Economics 85: Goodwin,, B., Mishra, A.K. and Oralo-Magne, F.N. (2005) Landowners'Riches: The Disribuion of Agriculural Subsidies. Paper presened a he CESIfo Applied Economics Conference, Munich, Germany, March Hausman, J.A. (1978). Specificaion ess in economerics. Economerica 46: Im, K.S., Pesaran, M.H. and Shin, Y. (2003). Tesing for uni roos in heerogenous panels. Journal of Economerics 115: Le Goffe, P., Saliné, J. (2005). Pricing manure spreading righs : measure from land marke. Paper presened a he 79 h Annual Conference of he Agriculural Economic Sociey. Levin, A., Lin, C.F. and Chu, C.S. (2002). Uni roo ess in panel daa: Asympoic and finie sample properies. Journal of Economerics 108: Planinga, A. J., Lubowski, R.N. and Savins, R.N. (2002). The effecs of poenial land developmen on agriculural land prices. Journal of Urban Economics 52: Sevesre, P.(2002) Economérie des Données en Panel. France : Dunod. 16