Aa

HD

a

ARRA

KMLEAE

I

ceitika

cGidfliut

t(

0h

Land Use Policy 31 (2013) 182– 195

Contents lists available at SciVerse ScienceDirect

Land Use Policy

j our na l ho me p age: www.elsev ier .com/ locate / landusepol

micro-simulation evaluation of the effectiveness of an Irish grass rootsgri-environmental scheme

ugh Kelley ∗, Thomas M. van Rensburg1, Lava Yadav2

epartment of Economics, National University of Ireland, Galway, Ireland

r t i c l e i n f o

rticle history:eceived 3 November 2011eceived in revised form 13 May 2012ccepted 27 June 2012

eywords:

a b s t r a c t

Agri-environmental schemes designed to induce agricultural producers to provide environmental pub-lic goods must increasingly justify exchequer costs. This requires comparing the non-market valueof amenity externalities with government expenditures. We employ a preference calibrated micro-simulation model to analyze the economic viability of a grass roots agri-environment scheme in theWest of Ireland. We also simulate the land-use response of individual producers to variations in program

icro-simulationand-usexternalitymenityco-tourism

funding and market conditions. Our results indicate that heterogeneous preferences crucially influenceproducers’ decisions to provide environmental public goods; producers are highly sensitive to market andpolicy incentives, in particular to semi-organic product price premia, off-farm labor market conditions,and subsidy payments; and finally, even in the most unfavorable simulated market conditions, supportpayments provide a rate of return exceeding 71% and in current conditions provide returns as high as185%.

ntroduction

An important market failure is associated with managed agri-ultural landscapes. In addition to marketable outputs, agriculturalnterprises also produce un-compensated public good externalitiesn terms of landscape amenity and tourism benefits. The values ofhese externalities are implicit; that is, they are not actively tradedn the market and consequently there are no incentives via the mar-et mechanism for farmers to continue with agricultural activitiesnd management practices that produce them.

One solution to this market failure has been to introduce poli-ies such as agri-environment schemes (DAFF, 1996; Emerson andillmor, 1999; Latacz-Lohman and Hodge, 2003). However, lim-

tations of past schemes have been that they were frequentlyesigned without an analytical assessment of their economic per-ormance and they rarely included the value of public goods such asandscape amenity. Most problematically, they tended to be gener-

cally applied top down policies that did not necessarily adapt tonique regional circumstances. The implication of these limita-ions is that policy makers could not adequately assess the broader

∗ Corresponding author. Tel.: +353 091 495087, fax: +353 091 524130.E-mail addresses: hugh.kelley@nuigalway.ie (H. Kelley),

homas.vanrensburg@nuigalway.ie (T.M.v. Rensburg), lava.yadav@nuigalway.ieL. Yadav).

1 Tel: +353 091 493858; fax: +353 091 524130.2 Tel: +353 091 492501; fax: +353 091 524130.

264-8377/$ – see front matter © 2012 Elsevier Ltd. All rights reserved.ttp://dx.doi.org/10.1016/j.landusepol.2012.06.017

© 2012 Elsevier Ltd. All rights reserved.

social/cultural and economic impact of the schemes nor assess therole of local agriculture stakeholders in the provision and steward-ship of the environment. More recently, targeted initiatives suchas the EUs LIFE schemes (Cooper et al., 2009) have attempted toaddress these limitations. Although such initiatives often considerthe role of non-market public goods and are adapted to particularregional conditions, they rarely attempt to compare the exchequercosts of these programs to the value of market and non-marketgoods provided. In what follows we apply a micro-simulationportfolio-theory model including both market and non-marketpublic good values in order to provide an economic assessmentof the EU BurrenLIFE scheme used to promote farming for conser-vation in the Burren region located in the west of Ireland.

The Burren is an important and recognized landscape in Europe.Located in the counties Clare and Galway, it spans an area of approx-imately 30,000 ha, see Fig. 1. This area is principally comprised ofrural areas with scattered farm holdings that make up small farm-ing communities. Far from being a natural landscape, the Burrenis a managed landscape that has evolved around a long historyof anthropogenic influence. Neolithic farmers new to the regionwould have been presented with a landscape dominated by pine(pinus sylvestris) and hazel (corylus avellana). The farming prac-tices thereafter involved scrub clearing and grazing by cattle, whicheventually led to extensive soil loss and the exposure of the skele-

tal limestone beneath (Moles and Moles, 2002; Dunford, 2002). Asa result, the remaining soils of the upland grazing areas are gen-erally of low productivity and are best suited to extensive cattleproduction with very little arable farming occurring in the area.

H. Kelley et al. / Land Use Poli

Ddifdrat

dEgpi1mEihaG

mcCotsptsiltt

energy, or land-based inputs to generate outputs. An importantlimitation of most land portfolio allocation literature is that thefocus has almost exclusively been on market based costs and

Fig. 1. Map of the Burren study area.

espite these features, the habitat of the Burren karst limestoneisplays vegetation, with areas interspersed with perennials

ncluding hazel, shrubby thickets, woodland, and grassland. Inact, it hosts over seventy percent of Ireland’s native floral bio-iversity (Dunford, 2002). Thus, it has been termed ‘a fertileock’ (O’Rourke, 2005). This combined with the unique appear-nce of the karst landscape results in the generation of significantourism.

In recognition of its unique value, much of the Burren has beenesignated as a “Special Area of Conservation” (SAC) under the 1992U Habitats Directive. During this period, there was a shift in theoals of EU Common Agricultural Policy (CAP) from protectionistolicies to those focused on environmental protection as outlined

n the 1992 MacSharry reforms (DAFF, 1996; Emerson and Gillmor,999). The primary scheme in Ireland to achieve these environ-ental goals was introduced in 1994 and is known as the Rural

nvironmental Protection Scheme (REPS – Regulation 2078/92). Its available to all Irish farmers, is voluntary, involves a compre-ensive farm management plan, mandatory training, and includes

tiered system of payments based upon farm size (Emerson andillmor, 1999; DAFF, 2004).

Toward the end of the decade, it was being realized that the REPSeasures were not fully adequate for making significant positive

ontributions toward the conservation of the Burren (Bohnsack andarrucan, 1999). One source of inadequacy was the generic naturef the measures prescribed by REPS. These initiatives were designedo be applied to all areas in a country and did not recognize theubstantial heterogeneity of the landscape in terms of, for exam-le, productivity, topological conditions, cultural, biodiversity, andourism value, and demographic features such as population den-ity. Other sources of inadequacy included the fact that the actual

mpact of REPS upon the Burren is not clearly known due to theack of baseline information relating to the conservation status ofhe various habitats and species in the region. However, accordingo an assessment of the scheme it was reported that “the current

cy 31 (2013) 182– 195 183

management objectives and farming prescriptions of the REPS mea-sure ‘Conservation of Natural Habitats’ do not adequately reflect thespecific legal obligations for the conservation of listed habitats andspecies in the proposed candidate special areas of conservation [inthe Burren]” (Bohnsack and Carrucan, 1999). REPS measures weredescribed to be of a restrictive nature that were capable of pre-venting further damage to the habitats but lacking incentives or aproactive approach to help in the restoration process. The assess-ment conducted by Bohnsack and Carrucan (1999) led to growingawareness that the unique habitats of the Burren required a moretargeted scheme instead of generic measures prescribed by REPSfor the rest of the country. Given these limitations of REPS, andthe growing awareness that the unique habitats such as the Bur-ren required a more targeted scheme, the BurrenLIFE project wasinitiated.

The BurrenLIFE project (BLP) sets out to identify specific agro-nomic practices that are fundamental to the sustainable deliveryof biodiversity and landscape amenity that are unique to the Bur-ren region and symbolic of its heritage. The five year pilot projectinvolved intensive field studies of twenty farms (the B20) spanninga total of 3097 ha with 2485 ha of SAC and 612 ha of undesignated‘improved’ agricultural land. The object was to identify supplemen-tary measures that, instead of being generic, related specifically tothe Burren habitat. These include livestock management and outwintering3 of cattle, the use of local breeds, shrub clearing, provi-sion of water to the herd, and stonewall restoration (BurrenLIFE,2010).

The measures put together by the BLP farming methods tran-scend the protection of the Burren habitats and attempt to achievemultiple environmental as well as socio-economic goals. Forinstance, a list of individuals from the locality was drawn upto enable farmers to hire locals for meeting their recommendedlabor-intensive farming practices. Hence, the strategy was capableof generating employment and contributing to the local econ-omy. New feeding systems involved less poaching around silagefeeders, reduced feed costs and nitrate pollution, and improvedwater quality. One of the fundamental elements of the BLP schemeinvolves the revival of the traditional out-wintering of cattle. Out-wintering produces hardier and healthier cattle, which also reducesveterinary costs. A key point is that the project is targeted tothe region, involves farm practices directly linked to the conser-vation of biodiversity and maintenance of karst limestone, andinvolves researchers from a number of disciplines providing exper-tise that farm advisors or even farmers may not have (BurrenLIFE,2010).

The micro-simulation portfolio-theory approach we proposecan be useful in this context because it allows for an assessmentof profitability and risk, both of which are important features ofthe agricultural producer’s decision (Parks, 1995). The portfolioapproach is a well-established part of finance and energy eco-nomics (DeLong et al., 1990; Klibanoff et al., 1998; Awerbuch andBerger, 2003; Springer, 2003; Kelley, 2004; DeLaquil et al., 2005;Kleindorfer and Li, 2005; Kelley and Evans, 2010), and has beenapplied within the context of land use (Blank, 2001; Kelley andEvans, 2011). This approach, regardless of context, is concernedwith optimal choice of investment/production activities among aset of multiple options. The basic goal of a decision maker is tomaximize returns and minimize risk by using available financial,

3 Outwintering is a traditional farm management practice where livestock are nothoused indoors during the winter months; rather they are kept outdoors where theyare allowed to graze freely.

1 se Poli

baa(tdtse

mcsaOqPiritdc

M

T

mmaeErb

E

Eaciv(paaaspcpst

msiupamnst

84 H. Kelley et al. / Land U

enefits and has rarely accounted for non-market externalities suchs those described above. And further, the few studies that havenalyzed externalities and market failure, e.g. for deforestationPearce et al., 2002), urban sprawl (Brueckner, 2000), air pollu-ion (Henderson, 1977), and waste management (Eshet et al., 2006)id not focus on the individual farm or household. This meanshat there is generally a lack of understanding about the diver-ity of land uses that may emerge based upon variable individualndowments.

Our work therefore provides methodological and landscapeanagement contributions to the literature. Our methodological

ontribution is the construction and calibration of a micro-imulation portfolio theory model, which combines preferencesnd both market and non-market data at an individual farm level.ur management contribution is to address three key researchuestions. We first ask, are the observed actions of the BurrenLIFErogram (BLP) farms optimal and to what extent do preferencesnfluence their behavior? Second, how might these BLP farmsespond to policy changes or variations in market conditions;n particular will they continue to adopt, or will they exit fromhe BLP? Finally, how do the public good externality values pro-uced by the BLP farms compare to the program’s exchequerosts?

ethodology

heoretical structure

Below we present a static micro-simulation portfolio-theoryodel of agricultural production. In such models, a decisionaker’s objective is to maximize risk-adjusted utility derived from

particular use of productive inputs in a particular year. DeLongt al. (1990), Parks (1995), Blank (2001), Kelley (2004), Kelley andvans (2011), and many others have shown that a traditional rep-esentation of agent’s k’s expected utility in such a framework cane:

[Uk] = E[�k,j(pj, Y, F, C)] − � · �2k,�j[�k,j(p

2j , Y2, F2, C2)] + �k,j (1)

[U] represents k’s risk adjusted expected utility. �k,j representsnnual profit, which is a function of output prices p, output Y, fixedosts F, and marginal costs C. Outputs Y are derived from apply-ng input endowments. Following Parks (1995), � is a risk aversionector including a parameter for each of the j agricultural activitieszero for risk neutral, positive for risk averse), �2

k,�jis the variance of

ayoffs for k’s portfolio which is a function of output price variancend assuming no covariance among output prices for alternativectivities. ϕ is k’s preference for supplying inputs to a particularctivity. This model has been shown to be an appropriate repre-entation of decision makers when the number of options within aortfolio is small and when portfolio returns are relatively smallompared to an agent’s total wealth; both of these features areresent in the context of the Burren. The numbers of input uses aremall and the returns from a portfolio are small relative to agent’sotal wealth as represented by land value.

In traditional production optimization models, researchersight assume a representative agent. However, in this micro-

imulation application, the optimization problem is solved for eachndividual agent given his or her unique input endowments. Basedpon contextual information suggesting single activity farms, aortfolio for a Burren farmer will consist of one of nine activitiesnd one default state generating farm assist unemployment pay-

ents. Second, given data on actual activities in the Burren, these

ine activities include either dairy, dry stock beef, mixed grazing,uckler beef, or off farm labor supply. And, the first four activi-ies can either be pursued in a way consistent with the BurrenLIFE

cy 31 (2013) 182– 195

Project, thereby incurring slightly higher costs but generating anamenity and tourism spend externality and a product price pre-mia, or in a commercial manner with standard (lower) costs andproduct prices, and with no externalities. This implies a set of tenactivities, four outputs consistent with BLP, the same four outputoptions produced in a commercial manner, off farm labor supply,and unemployment.

In terms of output, the input variables for the production func-tion are restricted to include available labor hours, hectares offorageable land, herd size, and capital investment levels in euros;see Eq. (A.1). There are two principal reasons why we restrict theinputs to these four variables. First, the initial three variables areincluded as they represent the majority of non-intermediate inputsutilized (in terms of costs or opportunity costs) for our study area.This was determined by comparing the total set of inputs reportedto be used across all activities in the all Ireland National Farm Survey(NFS), to the input expenditures of the farms in our study area (theB20) and to the farms outside our study area pursuing similar pro-duction activities. Based on this comparison, these three variableswere determined to represent the majority of input usage costs.Additionally, it is useful to include the last input, even though it mayrepresent a small share of costs for the farms of interest, because itcan control for the unobserved influences of productivity on output.This allows us to control for unobserved variation that might createindependent variable endogeneity in econometric tests, see Olleyand Pakes (1996), Kelley and Jeserich (2012). The second principalreason other inputs were not included is due to data availabilityconstraints. The total set of inputs was constrained to include theinputs reported in the NFS, see Connolly et al. (2008).

Other drivers of production identified in earlier studies butexcluded from the current work due to insufficient data or irrele-vance/commonality for the B20 activities include: land slope, parcelshape, elevation, distance to road, distance to market, populationdensity, commodity and labor market characteristics, irrigationstructures, property rights, macroeconomic and trade policy condi-tions, general and sector specific technological levels, and incometaxation rates.

Amenity and tourism externalityAlthough outside a farmer’s decision problem, Eq. (2) is an

expression that is critical to our analysis. It represents the totallandscape amenity and tourism spend externalities produced byfarms operating agricultural activities consistent with BLP.

˛ = ω ·∑

k

Mk,BLP(sk,BLP) (2)

ω = ωAmenity + ωTourism and represents, respectively, the amenityand tourism spend values of a hectare of farmland operated ina manner consistent with the BLP. These values are reported inTable 1 and were obtained from Yadav et al. (2010) as described in

Section “Data”. Further,∑

k

Mk,BLP represents the sum of hectares

of land in the BLP across the k = 20 farms. As will be shown later, theamount of land in the BLP is a function of the support payments,sk,BLP . The product of the by hectare amenity and tourism value andsum of hectares under BLP usage is the estimated monetary valueof the externality produced by the B20 farms.

Rate of return on government supportIn order to address the third research question and evaluate the

rate of return (RoR) provided by government support payments, wecompare the estimated public good externality provided by all B20

H. Kelley et al. / Land Use Poli

Table 1Key experimental and other model parameters.

Parameter Value

ωAmenity (lower bound) D3061.96 (per ha){D754.85} (per ha)ωTourism D35.74 (per ha)

�BLP 1.25 (% of non-BLP)sBLP D750 (per farm per year)sREPS 1 (×100% of actual)woff D13.50 (per h)�Empl 1.00 (probability)pBLP 1.05 (% of non-BLP)

CT D0.43 (per min)L 2000 (h)sfa D11,128 (per year)ϕ D1fDParlor D50, 000 + 2000 · HDairy(MDairy)pMilk D0.34 (per liter)pBeef D713 (per head)pSuckler D442 (per head)pMixG D277 (per head)pDairy D1,300 (per head)�B,SB,MG,D 1e−08 × [1,1,1,1]�2

Milk0.003 (D2)

�2 6944 (D2)

as

R

R

TaDwttrwgse

oatsiloro

ttmptcpatea

Beef�2

SucklerB4720 (D2)

�2MixG

2723 (D2)

gents’ ˛, to the direct and indirect costs born by the exchequer toupport these programs. These are represented as,

oRDC = ˛∑k(sk,BLP) + ssunk

BLP

(3)

oRTotal = ˛∑k(sk,BLP) + ssunk

BLP + IDC(4)

he direct and indirect costs are exogenous amounts. Direct costsre known to include two payments, DC =

∑k(sk,BLP) + ssunk

BLP =(687, 000/yr).

∑k(sk,BLP) represents the direct variable costs,

hich is the sum of the participation payments paid to each ofhe 20 farms, sk,BLP = 750. ssunk

BLP reflects sunk costs spent to ini-iate the program and to identify the 20 farms. Indirect costsepresent administrative and subsequent data collection costs andere shown to be IDC = D730,000 per year, or D2.2 m for the pro-

ram duration. Importantly, we calculate the RoR for the calibratedimulation and for each of the simulation or sensitivity analysisxperiments.

The stylized model employed is designed to be representative ofur context, but must also be simplified in order to provide insightsbout our particular research questions. We therefore do not claimhat it perfectly represents all dynamics present upon the land-cape. However, although this is a stylized model, the structures based on discussions with stakeholders and on agricultural andand use research. Further, estimating preferences and the inclusionf uncompensated amenity and tourism externalities addresses aecognized limitation in the literature. This places the contributionsf our research in a clear context.

The key aspects of this theoretical environment are that: (1)here are multiple heterogeneous decision makers who differ inhe quantity of their land, labor, herd, and investment endow-

ents, and in their production preferences; (2) B20 agents makearcel level output decisions comparing actual and counterfac-ual returns; (3) counterfactual production returns for the B20 arealculated by combining the B20 input endowments data withroduction and cost function parameter estimates derived from

broader agricultural activity dataset; (4) agents are aware ofhe functions that describe the output, costs, and profits they canxpect from various activities; (5) agents activities may generaten amenity and tourism spend public good externality, but this is

cy 31 (2013) 182– 195 185

uncompensated and does not impact their decision process; (6)similar to Ahn et al. (1981) and Kelley and Evans (2011) the planninghorizon is one year.

Data

We next provide a short description of key model parametersand the three datasets we use describing Irish agricultural activ-ities and socioeconomic preferences. The first Burren 20 dataset(B20) pertains to our principal study area, the Burren, and is basedon a survey of 20 project farms. The selection criteria for thesepilot farms was based upon their willingness to participate, theextent of SAC on the farm, the level of Annex I habitats present, thegrazing levels on priority habitats, and to ensure that the farmsadequately represented the Burren in terms of geography, farmsystem, off-farm work, and gender; see BurrenLIFE (2010). Thefarms included all key agricultural activities: suckler beef (10);drystock (2); dairying (2); and mixed grazing (cattle and sheep).In terms of production inputs, the labor endowment was mea-sured as total number of own, family, and hired hours spent onfarming activities. The land endowment varied between 40 ha and448 ha with a mean of 155 ha. The average herd endowment was0.43 Livestock Units per hectare (LU/ha) (ranging between 0.19and 0.81 LU/ha). All farms were observed to have marginal soilquality, scoring in the categories 4–6, where 1 is Irish best and6 is Irish worst. In addition to this agricultural data, economicinformation describing revenues, costs, and farm investment wascollected.

The second national farm survey dataset (NFS326) is a 326 farmsubset of the National Farm Survey involving 1100 Irish farms. Thissubset includes commercial farms with the same soil classifica-tions as the B20 and those undertaking one of the four activitiesof the B20. The data for these conventional farmers were collectedthrough the National Farm Survey (Connolly et al., 2008). This datais used to estimate key parameters for the B20 simulation model.

Both datasets included a number of common variables that arerelevant to this study. These included details about, input endow-ments, outputs produced, profit margins for undertaken activities,and farm and household characteristics. Additionally, regulatorymeasures, supported by the Common Agricultural Policy (CAP) areknown to play an important role in supporting farm incomes and ininfluencing farm decisions. In both surveys data were gathered onsubsidy programs: the Single Farm Payment (SFP, 2004); REPS; andthe BLP subsidy paid for undertaking specific externality generatingagronomic practices.

The final dataset involved a socioeconomic survey and yieldsestimates of the ωAmenity and ωTourism parameters representing thenon-market benefits associated with the environmental or tourismimpact public goods provided by the Burren and the BLP man-agement practices. These estimates were based on the results ofa stated preference survey of 292 Irish national respondents incounties adjacent to the Burren region (Galway, Limerick, andClare) as well as more distant regions (Dublin, Westmeath andSligo) see Yadav et al. (2010). A key innovation of this survey was theuse of both direct and indirect willingness to pay (WTP) valuationtechniques. Together these techniques mitigated social desirabil-ity bias, a form of hypothetical bias. The latter technique involvedadditional indirect questions that asked for respondents predic-tions about other’s potential contributions (List and Gallet, 2001;Murphy et al., 2005; Harrison and Rutström, 2008). In the Yadavet al. (2010) study, the indirect question WTP was observed to be

2.5 and 3.1 times smaller than the direct question WTP. There-fore, direct response WTPs were used as the primary measures,and the indirect responses were used as a lower bound estimate ofrespondents WTP for Burren landscape amenity.

1 se Poli

ptdeflG2to

oel3t2bgvaCt(8lls

thcv1veeoooswt1aamr

T4tDtpai

atsf(

86 H. Kelley et al. / Land U

Analysis of survey results indicated a value of D80.11 per tax-ayer as the amount on average most third parties indicated thathey were willing to pay to preserve both the landscape and bio-iversity amenity aspects of the Burren. The indirect lower boundstimate was D19.75. The aggregate national benefits that resultedrom the conservation of the orchid rich grasslands and the karstimestone pavements were estimated using 2000 as the base year.4

iven approximately 1,146,672 taxpayers (Department of Finance,000) and approximately 30,000 total hectares of Burren landscapehis translated to approximately ωAmenity = 3061.96{754.85} D/haf externality.

These values are somewhat larger than those obtained fromther studies whose values are translated into D/ha amounts. Forxample, Willis and Garrod (1993) found that a similar Karstandscape in the Yorkshire Dales National Park (UK) was worth02.56 D/ha. Further, Fitzpatrick and Associates (2005) estimatedhe non-market amenity value of Irish forest walking trails to be15.90 D/ha and the total amenity value of forest in general toe 220.45 D/ha. Another study by Campbell et al. (2006) investi-ating the biodiversity amenity produced by REPS estimated itsalue at 175.61 D/ha. A more recent study proximate to our studyrea by Hynes et al. (2007) estimated the value of Roundstoneommonage and two recreational beaches in Ireland to be no lesshan 3600 D/ha. Overall these studies suggest a range of ωAmenity ⊆122.87, 215.90, 220.45, 302.56, 3, 600) D/ha with an average of92.36 D/ha; see Appendix A for details. These values suggests our

ower bound estimate of 754.85 D/ha is representative of all butocal research, and is perhaps 15% too low unless the Burren isubject to a discount.

The Yadav et al. (2010) survey also provides estimates ofhe tourism spend and associated economic impact ωTourism of aectare of BLP land. Of the respondents in this sample, 70% indi-ated that they spent money on food (an average of D26.60 perisit) 23% spent money in pubs (an average of D48.06 per visit),2% spent money on accommodation (an average of D114.68 perisit) and 7% spent money on other items such as buying gifts,tc. (an average of D42.00 per visit). Even given these stated lev-ls of expenditure across the various categories, a precise estimatef the local economic impact would require detailed informationn the nature of goods and services that are sold. Increased levelsf impact are expected if a higher share of goods and services areourced locally. According to estimates from case studies in north-estern United States by Goldman et al. (1994), typically 30–50% of

ourist expenditures are retained in the locality, with an additional0% approximately attributable to a multiplier effect resulting from

second round of spending by locals within the region. Absentdditional information about the goods and services sold, we esti-ate the overall economic impact of tourism expenditure on local

evenues to be about 50%.According to Saunders (2008), 60% of the 826,000 Domestic

ourists in the Shannon region in 2007 visited county Clare (i.e.95,600 visitors). Given our estimates, this would imply that theotal spend in county Clare by domestic tourists would amount to11,392,170 (50% of D22,784,339.95) per year. It is quite evidenthat the Burren region would be responsible for attracting a higher

roportion of this spend to the region, however, we assume thatll land in County Clare has the same value in terms of attract-ng this spend in order to develop a conservative estimate. As such,

4 Normally, the aggregation of willingness to pay estimates is done over the entiredult population. However, for a more conservative estimate we only consider theotal Irish taxpayers. This includes those who pay taxes at the higher rate (427,077),tandard rate (701,953) and the special marginal relief rate (17,642). Those exemptrom paying taxes due to income levels below the income tax exemption limits463,161) are excluded. Thus, the total number of individuals included is 1,146,672.

cy 31 (2013) 182– 195

multiplying total tourism spend by the economic impact factor anddividing by the total hectares of land in County Clare (318,784 ha)yields a value of 35.74 D/ha. This amount is of course a simple esti-mate, but given that the Burren is a principal destination of visitorsto Clare, and that we have no comparable tourism spend estimatesfor the Burren specifically, this is a reasonable lower bound esti-mate. It does mean, however, that we do not attribute all tourismspend in Clare to the Burren. In fact, we only attribute 30,000 haof Burren/318,718 ha of Clare or 9.4% of Clare tourism spend to theBurren; see Table 1.

The values we obtain for tourism impact per hectare are alsocomparable to values obtained in other studies. The general cal-culation to obtain such values is total spend times economicimpact factor divided by number of relevant hectares; with theimpact factor assumed to be approximately 50% as suggested byGoldman et al. (1994). For example, Bergin and Rathaille (1999)and Buckley and van Rensburg (2006) provide studies of all Irelandwalking tourism for 1997. For the approximately 90,000 indi-viduals that took part in walking activities (27% foreign), theirprimary expenditures were found to include travel, food, entryfees, and accommodation and were estimated to be worth 30.67D/ha. Curtis and Williams (2004) provide another study investigat-ing walking and hiking based tourism for the Irish Sports Councilin 2002. They found that the principal expenditures related to foodand drink and in some instances entry fees in addition to equip-ment, and that significant expenditures occurred in the area inwhich walks were undertaken. For the trips of this study, theysuggest an economic impact of 21.29 D/ha. Another study con-ducted by Failte Ireland (2004) found that between 2000 and 2003the economic impact of walking tourism across all of Ireland was44.21 D/ha. Another study by this organization found that walk-ing expenditures in 2004 were worth 43.52 D/ha. In terms of onlydomestic forest walking expenditures, Fitzpatrick and Associates(2005) estimated 348.86 D/ha of tourism impact and for generalforest recreational tourism 304.55 D/ha. A more recent study ofcoastal walking tourism by Barry et al. (2011) which is based uponresearch by Wilson et al. (2005) finds that coastal walking wasworth 114.87 D/ha. The values from these studies span ωTourism ⊆(21.29, 30.67, 43.52, 44.21, 114.87, 304.55, 348.86) D/ha withthe latter two numbers from the Fitzpatrick (2005) study repre-senting outliers by an order of magnitude. These values average130.66 D/ha and excluding the Fitzpatrick numbers 50.91 D/ha. Thelikely more reliable latter average lies 30% above our estimate of35.74 D/ha.

In summary, our lower bound indirect WTP estimates of theper hectare value of amenity and tourism lie within the rangesobserved in other studies, at different points in time, and for differ-ent landscapes. In fact, they are 15% or 30% lower than across studyaverages, respectively. Importantly, differences are likely due tothe fact that unique types of land (coastal, forest, standard Ireland,or Burren) have different values to the public and the economy.Also note, that given the absence of information about destinations,total spend values are normalized to the total walkable hectares ofIreland, or accessible forests, or coastal areas, or the area of countiesClare and Galway. This makes the by-hectare values we report andcalculate for other studies lower bounds, given that the land withactual tourism or amenity value is likely to be much less than thehecatares in our denominators. We next describe how we investi-gate the impact of changing the value of a key, but unobservable,parameter.

Sensitivity analysis parameters

Based upon informal discussions with stakeholders, it can be

assumed that pursuing the BLP practices involves an additional,but not strictly known, cost compared to the commercial produc-tion approach. The parameter �BLP represents this increased cost

se Poli

(ftidtp

M

itavSbor(f2osdarctB1m

C

cfilfpabS6frwmatiTsowuaaFbtuttfs

H. Kelley et al. / Land U

which is mainly due to hired labor land clearing/maintenance, andeed costs). We initially assume that the cost is 25% higher. In fact,he magnitude of this parameter is less important than the fact thatt is simply greater than one. As a result, using a utility maximizingecision strategy, a producer will always prefer to pursue a conven-ional approach due to lower cost (even if the value is 1.05) unlessreferences ϕ are included in their utility structure.

onte Carlo parametersThe fourth through eighth parameters reported at Table 1, which

ncludes �BLP above, are key policy and market condition parame-ers. These are further investigated with either model experimentsnd/or Monte Carlo analyses. First, sBLP is the direct payment pro-ided to an individual farmer for participating in the BLP each year.econd, sREP is the percent of the observed REPS payment receivedy a farm as reported in the NFS326 or B20 datasets. A value ofne indicates that an agent received 100% of the actual paymenteported, a value less (greater) than one indicates the receive lessmore) than the 2008 payment reported. Third, the average offarm industrial wage during 2008 in this area is w = D13.50 (CSO,008). Fourth, �Empl represents the probability of being able to findff farm employment. We set this value to one for the baselineimulation so that it can be reduced later to reflect the increasedifficulty of finding off farm employment during the recession-ry environment following 2008. In this period, the probability iseduced given that the majority of off farm activities were in theonstruction sector, and this sector substantially contracted duringhe recessionary period. Fifth, pBLP represents the price premiumLP producers obtain for their semi-organic products. The value of.05 indicates a 5% premium, which is obtained from internal BLParketing analysis.

onstant parametersNext in the table are the parameters that are common and held

onstant for calibration and simulation exercises. First, CT is thexed per minute commuting cost for farmers supplying off farm

abor. Without more detailed location information for each of thesearms, which is suppressed due to privacy concerns, we must sim-ly apply a common average cost for all farmers. Although therere certain remote farming areas in Ireland where this cost maye much higher, for a typical farm drawn from the National Farmurvey, the distance to the nearest urban center is between 30 and0 min. Further, B20 farms all lie in a small geographical area. There-ore, a common round trip value of 45 min can be assumed to beeasonable. We determined that the per minute commuting costas CT = D0.43/min, from the Irish Automobile Association 2009ileage cost estimates. Second, total available labor hours L are

ssumed to be 40 h per week for 50 weeks resulting in a baselineotal labor supply endowment of 2000 h. This figure is then mod-fied by the survey reported labor supply units for the BLP farms.he labor conversion factor ranges between 0.5 and 1.5, and repre-ents part time farming, employment of family members, or hiringf off farm laborers. Combining the labor unit conversion parameterith L yields Lk, which is used in Eq. (A.1). Third, sfa represents thenemployment or farm assist payment provided in the absence ofgricultural production or off farm employment. The 2009 paymentveraged D214/week for the year, Dept. of Social Protection (2009).ourth, when calibrating the preference parameter ϕ, it is adjustedy D1 per increment, thus ϕ = D1. Next in the table is the functiono determine the counterfactual dairy parlor fixed cost; this is basedpon informal discussions with B20 stakeholders. We observe that

his fixed cost is itself composed of fixed and variable costs relatedo herd size; and the herd size estimate is in turn determined by theorageable land endowment. Finally, output prices, the risk aver-ion vector, and price variances calculated over 1976–2000 follow.

cy 31 (2013) 182– 195 187

Empirical methodology

Three types of empirical methodologies are employed to addressour research questions. First, we conduct a by-farm preferencecalibration exercise exploring the extent to which agents/farmscan be parameterized to reproduce the actual B20 land uses. Thisrequires estimating the value of each counterfactual activity foreach farm. Second, simulation experiments explore how land uses(and externalities generated) respond to parameter perturbationsrepresenting policy interventions and changes in market condi-tions. Third, we analyze the results of Monte Carlo simulations,which simultaneously perturb multiple parameters across 1000simulations.

Counterfactual estimationImportantly, to calibrate preferences and determine what might

lead a B20 farmer agent to adopt or abandon BLP land uses, we mustfirst compare the utility for the activities actually undertaken (j)to estimated counterfactual utilities for pursing alternative actions(j′). Counterfactual utility can be derived from switching to an alter-native BLP activity, to the commercial version of any agriculturalactivity, supplying labor off farm, or being unemployed and accept-ing farm assist payments. For actual actions, we can calculate riskadjusted utility given that we have data on output generated andprices received, price variances, subsidies, and input costs incurred.To determine counterfactual utility we must econometrically esti-mate counterfactual output value Yk,j′ , and counterfactual totalcosts Eq. (A.4). The latter is composed of counterfactual marginalcosts Eq. (A.2) Ck,j′ and fixed-switching costs Eq. (A.3) Fk,j′ . To deter-mine these values, we regress Y, C, and F against input endowmentdata for non-BLP farms that have similar soil quality and the sameproduction activities as the B20. We use this larger NSF326 datasetto insure more than 30 farms for each by-activity sub-sample. Theseregressions provide key parameters that we can combine with theB20 farms’ endowment data allowing us to predict counterfactualutility for each B20 farm and activity.

Counterfactual output is obtained by fitting the logged Eq. (A.1)production function for each activity. This yields estimates of theby-activity total factor productivity (TFP) and MRS across inputs.In order to obtain robust TFP and MRS parameter estimates weemploy the Olley and Pakes (1996) correction method wherebyinvestment and fourth order polynomial terms are used to correctfor any unobserved variation. Despite this correction, estimatedparameters may still capture the effects of unobserved variation.However, our use of a primarily agricultural study area mitigatesthe influences of many of these factors, in particular urbanizationinfluences.

We estimate counterfactual marginal costs by relating theNFS326 variable, direct variable costs, to the input endowmentsyielding Eq. (A.2). Estimating counterfactual fixed costs is a bit orecomplex. We argue that, due to the farm already being owned,counterfactual fixed costs are switching costs. And for this area,switching costs are most directly related to the cost of acquiring anew activity specific herd. Fortunately, due to legal requirements,the herd sizes are directly related to available forageable land. Thus,counterfactual fixed costs can be determined by regressing herdsizes against the land-endowment with the NFS326 dataset. Then,the herd to land size parameter recovered in the previous step canbe combined with the land endowment of the B20 farm of interest.This yields the required herd size for a particular farm and activity.Finally, combining the known per head livestock price (see Table 1)with the estimate of the herd size required and subtracting the

value of the current herd (size of actual herd times livestock prices)yields the net fixed-switching cost, Eq. (A.3).

Combining estimates for counterfactual output Y , marginalcosts C, and fixed switching costs F , and given information about

1 se Poli

wgcFaafi

E

A

pufafvc

P

ooeaIodppucmf

S

pwattcpbfotppcidva

M

imsdMir

88 H. Kelley et al. / Land U

hether the counterfactual activity is on-farm or off-farm, andiven known subsidies paid for each activity provides the totalosts Eq. (A.4) and there expected counterfactual profit Eq. (A.5).or counterfactual costs we assume Y = Y , F = F , C = C, �off = (0|1),nd �Empl ⊆ (0, 1). Assuming ϕ = 0 and letting represent the set ofll parameters and substituting the estimated counterfactual pro-ts into Eq. (1) yields,

[Uk] = E[�k,j(pj, Y, F, C|˝)] − � · �2k,�j[�k,j(p

2j , Y2, F2, C2|˝)] (5)

gent decision makingA key aspect of our simulation model concerns agents’ decision

rocess. Each individual farmer is assumed to make an expectedtility maximizing decision. This involves comparing the knownactual returns (U) to the counterfactual estimated utilities (U) forll commercial and BLP production activities, the returns for offarm employment, and farm assist payments. That is, each indi-idual agent k selects activity j if Uk,j ≥ Uk,j′ ∀j /= j′, i.e. Eq. (1) isompared to Eq. (5) and if equal the current activity is maintained.

reference calibrationTo evaluate our first research question regarding the optimality

f the B20 decisions, we compare the actual and counterfactual pay-ffs assuming the preference ϕ for the current BLP activity at Eq. (1)quals zero. We also assume a set of baseline parameter values thatre derived from the CRO and BLP internal data sources; see Table 1.f the maximal utility is counterfactual with respective to the actualbserved choice (with ϕ = 0), this is an indication of non-optimalecision-making. We then calibrate the model by estimating thereference parameter ϕ for each farm k. This involves increasingreferences for the current BLP action by D1, i.e. ϕ = ϕ + ϕ, until thetility for the actual current activity is maximal compared to theounterfactuals. The preference values obtained provide an esti-ate of the monetary value of an individual B20 farmer’s preference

or pursuing their current BLP activity.

imulation experimentsTo address research question two regarding the sensitivity of

roduction decisions to policy incentives and market conditionse conduct numerical comparative statics experiments on policy

nd market condition parameters. Experiments 1 and 2, respec-ively, vary the BLP and REPS subsidy rates, sBLP and sREPS, reducinghe former to zero and the latter by 47%; these variations areonsistent with proposed or observed changes to farm supportayments, respectively (DAFF, 2008, 2011). Experiment 3 reducesoth the average off farm wage w and probability of securing offarm employment �Empl by 30%; these changes are consistent withbserved recessionary changes in off farm labor market condi-ions (Meredith, 2010). Experiment 4 involves reducing the outputrice premia that BLP semi-organic products receive at marketBLP, making their sale price equivalent to the prices reported forommercially produced products. This experiment is designed tonvestigate to what extent this higher profitability influences pro-ucers’ decision making. Importantly, each of these experimentsaries only one or two parameter(s) at a time and leaves the otherst their baseline values.

onte Carlo simulationsThe second method we use to address research question two

s Monte Carlo simulation analysis of the preference-calibratedodel. This involves performing 1000 simulations in which the

ix Monte Carlo parameters (see Section “Data” and the second

emarked area in Table 1) are perturbed from their baseline values.ultiple parameters may be simultaneously varied by multiply-

ng them by a random variable lying within the range 0–2; theseepresent reductions or increases in their value of up to 100%.

cy 31 (2013) 182– 195

We evaluate the results of this last set of simulations with onequalitative and two quantitative analyses. Our qualitative methodinvestigates a scatter plot of the outcome variables number of agri-cultural producers (BLP or commercial) plotted against the totalexternality produced. Qualitative groupings of results across the1000 simulations can then be observed. Our first quantitative anal-ysis calculates median values for the experimental parameters andthe outcome variables and for these groups. We also perform acrossgroup Wilcoxon hypothesis tests to determine the statistical sim-ilarity of these groups in terms of experimental parameters andoutcome variables. The second quantitative method regresses threez-score normalized dependent variables representing either num-ber of agricultural producers, the total externality produced, or anaverage of these normalized measures against z-score normalizedvalues for the key experimental parameters that were perturbed.The Monte Carlo simulations provide 1000 observations for anal-ysis. The three dependent variables provide three sets of resultsidentifying the sensitivity of simulation outcomes to key modelparameters. Normalizing the data allows coefficients to lie withinthe same scale allowing their relative influences to be easily com-pared.

Sensitivity analysisWe also perform a sensitivity analysis for the key parameter

�BLP. We focus on only this parameter as the others are expected tosimply have proportional effects and they are not directly relatedto our research questions. Its value is reduced to �BLP = 1.125 or�BLP = 1.05 while simultaneously setting sBLP = 0. This manipulationprovides evidence about whether preferences alone, in the absenceof support payments, might allow BLP viability.

Additional insights about the reliability of the results welater report can be obtained by considering how our simulationoutcomes might intuitively be impacted by changes in other param-eters. In terms of the values for our externality parameters ωAmenityand ωTourism, because they are uncompensated externalities anddo not enter the agents’ payoff structure, agents’ production deci-sions would not be impacted. Instead changes in these parameters,for example increasing our lower bound values by 15% or 30% toreflect estimates from other research, would simply increase theamenity or tourism economic impact values and the associated pro-gram rates of return proportionally. Alternatively, changes in otherparameters could impact agents’ production decision, althoughtheir impact is still likely to be proportional. We ignore parametersthat were subject to sensitivity or Monte Carlo analysis for whichwe have direct quantitative evidence, and parameters with noexpected portfolio reallocation effect (e.g. L, ϕ). If the per minutecommuting cost parameter CT were increased (decreased), equiva-lently the average distance of Burren farms to off farm work suiteswas increased (decreased), fewer (more) farmers would engage inoff farm employment. At the very least this means that more (less)farms will remain in agriculture in the face of policy or marketshocks, and potentially more (less) farms remain in the BLP provid-ing more (less) externality and program rate of return. Second, if thefarm assist unemployment payment sfa were decreased (increased)more (fewer) agents would find remaining in agricultural opti-mal compared to seeking off farm employment which potentiallyresults in unemployment; this occurs since the expected value ofoff farm employment would be lower (higher). This again implies atleast more (less) agricultural producers and potentially more (less)BLP producers, externalities, and program rates of return couldobtain. Third, increases (decreases) in the dairy parlor fixed costssimply reduces (increases) the probability that dairy production

of either form is optimal. This could imply a portfolio shift towardnon-dairy activities or off the farm. However, given that dairy activ-ities are generally the most profitable agricultural activity, a shifttoward alternative agricultural activities is likely. This suggests that

se Policy 31 (2013) 182– 195 189

ttabdtoneaooItao

(pItrv

R

mstmptwtr

P

odpsisatbBBoio

wtBba

tmmo

0

1

2

3

4

5

0

1000

3000

5000

7000

9000

11000

13000

15000

17000

19000

21000

23000

25000

Preference Estimate (φ) Catago ry

Fre

qu

en

cy

H. Kelley et al. / Land U

he number of agricultural producers will be unchanged, althoughhe externality and rate of return might fall. Finally, changes ingricultural output prices, risk aversion, and price variances cane relative or across the board. If there is an increase in one price orecrease in one activity’s risk aversion or price variance this implieshe associated agricultural activity is more likely relative to thethers. This suggests most likely no change (or increases) in theumber of agricultural producers and potentially no change in thexternality and rate of return. Increases in prices or decreases in riskversion or price variances across the board reduce the likelihoodf off farm activity and either produces no change in the numberf agricultural producers and externalities, or possibly increases.n addition to direct effects, there may also be interaction effects;herefore, a more complete Monte Carlo analysis would be mostppropriate. However, given the interest in limiting the scope ofur research questions, we reserve this for future work.

Overall, in terms of the reliability of our results, we may over-under-) estimate the public good externality and government sup-ort rate of return if sfa and fDParlor are higher (lower) than expected.

n addition, we may over- (under-) estimate the number of agricul-ural producers (and also potentially the externality and rate ofeturn) is prices are higher (lower), and if risk aversion and priceariances are lower (higher) than expected.

esults

We first discuss the results of model calibration, which esti-ated the B20 farmers’ production preferences ϕ. We next conduct

imulation experiments and a 1000 run Monte Carlo analysis usinghe preference-calibrated model. These exercises allow us to deter-

ine the sensitivity of land uses and externalities generated to keyolicy and market condition parameters. With predictions abouthe externalities generated, and the observed government support,e can calculate the exchequer’s rates of return. Finally, a sensi-

ivity analyses for a key unobservable parameter investigates theobustness of our results.

reference calibration results

Assuming all simulated BLP farmers’ preferences for theirbserved action are zero (ϕ = 0), and observing their predicted pro-uction decision given estimated counterfactual production values,rovides an indication of how many farmers are operating in atrictly maximizing way. In this case, five farms continue to partic-pate in the BLP, three farms pursue off farm wage earning activity,even farms pursue the conventional version of their current BLPctivity, and two suckler beef, two beef, and one dairy farm converto commercial mixed grazing. This suggests that preference cali-ration is crucial for providing an accurate account of 15 out of 20urrenLIFE farmers’ activities. Or equivalently, participating in theLP is not strictly utility maximizing for three quarters of farms, ifne excludes stakeholders preferences. Crucially, this implies thatn the absence of preferences, the BLP payment was insufficient tovercome the increased cost of the BLP method.

We next calibrate the preference parameter for all BLP farms,hich results in correct predictions of all farms actions. We observe

hat the majority of estimated preference values for pursuing theLP method lie between D0 and D10,000, with the majority valuedetween D0 and D3000; see Fig. 2. Four farms display outlier riskdjusted preferences.

Intuitively, a value for this preference parameter that is greater

han zero indicates that the BLP activity is not strictly utility maxi-

izing for a farm, and that to continue producing as they are theyust be deriving preference value equivalent to the magnitude

f this parameter to offset suboptimal utility. Alternatively, this

Fig. 2. Histogram of estimated preferences for observed practices of Burren 20.

parameter (to the extent it is positive) could be thought of as theestimate of a farm’s opportunity cost of pursuing a particular BLPactivity.

Simulation experiment results

We next report the results of numerical comparative staticsexperiments for five key policy and market structure parameters.For all experiments only one, or in one case two, parameters are per-turbed; all others are set to the baseline calibrated model values.We focus on the impacts on agricultural activity decisions, deci-sions to maintain/abandon the BLP, and decisions to pursue off farmemployment. Additionally, given information about which farmsmaintain the BLP, we can estimate the amenity and tourism exter-nality produced, and the rate of return for government support.To determine the monetary value of the environmental amenityand tourism spend externalities, we need to multiply the hectaresof usable land the Burren farmers keep in the BLP by the perhectare willingness to pay for the Burren landscape and poten-tially by the 50% tourism economic impact factor. Importantly, theamenity/tourism value of the Burren is not directly part of theowner’s payoff structure, so a positive value represents an uncom-pensated externality. Further, to determine the rate of return forvarious experimental outcomes, we assume that only the variablecost part of the government payments, sBLP = D750, are reducedwhen a farm abandons the BLP. Alternatively, the initial sunk costof program start up for these farms, ssunk

k,BLP, and administrative costs,

IDC, cannot be avoided; see Eqs. (3) and (4).When calibrating the preference parameters we are able to

reproduce the actual actions of all B20 farms; i.e. all farms arepredicted to pursue the observed BLP activity. In this baselinecase with the maximal externality, a value of ˛Amenity = D8,820,100{D2,174,468} per year is produced compared to direct programpayments of D687,605 per year. These numbers indicate that theseprograms are producing rates of return of 1283% {316%} per Euroof subsidy investment for average {lower bound} amenity WTPestimates. Including the multiplied economic impact income gen-erated from tourism income per hectare increases the values to

= D8,923,030 {D2,277,417} or rates of return of 1298% {331%}.Including annualized indirect costs of D730,000, these total rates of

return become 629% {161%}.

Our first experiment involves setting the BLP subsidy payment,sBLS, to zero; all other parameters are kept at their baseline val-ues. In this case, all 20 farms abandon BLP practices, implying that

190 H. Kelley et al. / Land Use Poli

Fp

tAcg

wpfat{dccw

ipfco

ffpas˛oo

c

ated with low to moderate numbers of agricultural producers, but

TM

ig. 3. Scatter plot of Monte Carlo simulations perturbing key policy and structuralarameters reported at top of Table 1.

he externality produced is predicted to eventually fall to zero.lthough all farms abandon the BLP, 12 switch to the commer-ial version of their BLP activity, five switch to commercial mixedrazing, and three switch to off farm (un)employment.

Our second experiment reduces the REPS payment by 47% asas suggested to be the case for the 2010–2011 period, all otherarameters are set to the baseline values. In this case, 18 of 20arms retain their BLP practices, however two abandon agricultureltogether and switch to off farm (un)employment. This suggestshat the total externality value will eventually fall to = D8,599,316D2,194,796}. To calculate the rate of return, we can remove theirect sBLP payments from program expenditure, but initial sunkosts ssunk

k,BLPhave already been incurred in program start up. In this

ase, the rates of return with direct costs are 1251% {319%} andith both direct and indirect costs they are 607% {155%}.

Our third experiment reduces the off farm wage and probabil-ty of finding off farm employment by 30%; i.e. w = D9.45/h androboff = 0.7. This was suggested to be representative of current anduture labor market conditions for Ireland (Meredith, 2010). In thisase, all farms retain their BLP activity and the externality and ratef return is equivalent to baseline.

Our fourth experiment reduces the price premium receivedor BLP semi-organic products to the same price as receivedor the commercial versions, i.e. pBLP = 1. In this case, 12 of 20roducers switch to the commercial form of their current BLPctivity, the other eight retain their current BLP activity. Thisuggests that the total externality value will eventually fall to

= D3,956,739{D1,009,875} generating direct cost rates of returnf 575% {147%} and for cumulative direct and indirect costs rates

f return of 279% {71%}.

The results of all these experiments and the calibrated baselinease are graphically displayed in Fig. 3 as open circles. In this figure,

able 2edian parameter values for Monte Carlo simulation outcomes in groups 1–4.

Group Nobs Number Ag. Total extern. (D) sBLP (D) sREPS

Group 1 282 20 8,923,000 962.15 1.01

Group 2 481 20 447,620 584.03 0.97

Group 3 86 14 6,837,000 892.71 0.79

Group 4 132 16 1,700,700 536.37 0.62

cy 31 (2013) 182– 195

the horizontal axis represents the total externality values producedand the vertical axis represents the number of farms remaining inthe agricultural sector, either in a commercial form or consistentwith the BLP. These experimental outcomes can be used to cate-gorize the outcomes of the Monte Carlo simulations we describenext.

Monte Carlo results

We next discuss the results of Monte Carlo simulations perturb-ing our calibrated model.

Qualitative resultsFig. 3 also provides a visual representation of the results of these

1000 simulations, which randomly perturbed the six key experi-mental parameters (the six listed in the second section of Table 1).On the figure, dots represent the aggregate (across the 20 farms)Monte Carlo outcomes. Markers lying on the vertical axis indicatesimulations with at least one commercial agricultural producerwho does not create an externality value. Because of there beingonly 20 farms with fixed amounts of land, there are limited amountsof discrete simulation or experimental outcomes. This implies eachdot potentially represents multiple outcomes. The most obviousfeature of this figure is that there is a set of eight farms that areable to maintain at least commercial agricultural activity in the faceof all parameter perturbations, i.e. no dots lie below eight on thevertical axis.

Upon closer inspection, one also sees that the simulationoutcomes are grouped in four areas, which are approximatelydemarked with the large ovals. Within these groups, we observecommon parameter characteristics. Group 1 provides the maxi-mal externality, group 2 represents farms transitioning from BLPto commercial agricultural activities, group 3 represents farmstransitioning from BLP directly to off farm activities, and group 4represents a mixture of the previous cases with no systematic pat-tern. More specifically, group 1 (top right) includes the calibratedsimulation and the experiment 3 outcomes near the same loca-tion, and the experiment 2 outcome. These represent the secondmost frequent outcomes and involve 18 or more farms maintain-ing the BLP and thereby providing maximal externality value, i.e.more than D8 m. Group 2 is the most frequent outcome and includescases located along the top part of the figure in which there arealways more than 18 agricultural producers of either commercialor BLP type; this group includes the experiment 4 outcome. As oneconsiders simulation outcomes moving to the left in this area, theexternality value is falling as farms transition from BLP agriculturalpractices to commercial agricultural activities. Group 3 is the leastfrequent outcome and includes cases along an upward sloping linespanning the lower middle to the top right in the figure. Outcomesmoving toward the bottom left in the figure represent BLP farmstransitioning directly out of BLP agricultural production to off farmactivities or unemployment. Finally, group 4 are outcomes associ-

producing highly variable amounts of externality values. The exper-iment 1 outcome is roughly in this area. There are varied transitionsof farms in this area, moving among all activities.

(%) woff (D/h) Probability employ. Price prem.(D) �BLP (%)

8.81 0.79 1.49 0.729.40 0.83 0.60 1.51

21.03 1.00 1.36 0.7221.53 1.00 0.79 1.69

H. Kelley et al. / Land Use Policy 31 (2013) 182– 195 191

Table 3P-values for Wilcoxon by-group test of the significance of parameter differences.

Groups compared Number Ag. Total extern. (D) sBLP (D) sREPS (%) woff (D/h) Probability employ. Price prem. (D) �BLP (%)

Group 1 vs. group 2 0.69 0.00 0.00 0.79 0.10 0.64 0.00 0.00Group 1 vs. group 3 0.00 0.00 0.19 0.02 0.00 0.00 0.39 0.96Group 1 vs. group 4 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

0.020.000.50

Q

cpWpt

atpabc

aggpc4l

phige

gcsbft

oaeadrvpbicw

TR

Group 2 vs. group 3 0.00 0.00 0.00

Group 2 vs. group 4 0.00 0.11 0.64

Group 3 vs. group 4 0.00 0.00 0.01

uantitative resultsOur first quantitative analysis of these Monte Carlo simulations

ompares summary statistics for the parameters associated witharticular groups. The sample sizes, median parameter values, andilcoxon p-values indicating the statistical distinguishability of

arameters across groups are provided in Tables 2 and 3, respec-ively.

Tables 2 and 3 together indicate that in terms of outcome vari-bles, group 1 includes significantly more agricultural producershan all but group 2; and includes higher externality values com-ared to all other groups. For parameters, group 1 has higher sBLSnd BLP price premia and lower �BLP and off farm wage than allut group 3, and higher sREPS and lower probability of employmentompared to all but group 2.

Group 2 involves significantly more agricultural producers thanll but group 1, and lower externality values compared to all butroup 4. For parameters, group 2 involves lower sBLS than all butroup 4, higher sREPS and lower probability of employment com-ared to all but group 1, and marginally higher off farm wageompared to group 1 but lower wage compared to groups 3 and, higher �BLP than groups 1 and 3 but lower than group 4, and

ower price premium compared to all other groups.Group 3 involves significantly fewer agricultural producers com-

ared to all groups; and lower externality values than group 1 butigher externalities than groups 2 and 4. For parameters, group 3nvolves higher sBLS and price premium and lower �BLP for all butroup 1, lower sREPS and higher wage than all but group 4, and highermployment probability compared to all.

Group 4 involves significantly fewer agricultural producers thanroups 1 and 2, but more than group 3; and lower externality valuesompared to all but group 2. For parameters, group 4 involves lowerBLS for all but group 2, lower sREPS for all but group 3 higher proba-ility of employment and �BLP compared to all groups, a higher offarm wage compared to all but group 3, lower product premiumhan groups 1 and 3 but higher premium than group 2.

Our second quantitative analysis pools all 1000 simulationutcomes and regresses three z-score normalized dependent vari-bles, the number of agricultural producers, the total amount ofxternality produced, and an average of these two normalized vari-bles, with normalized values for the six experimental parametersescribed in Tables 2 and 3. The first two dependent variables areepresented as the two axes in Fig. 3, and a higher value for the thirdariable (the average of the first two) would be associated withoints in the top right of the figure. All variables are normalized

y calculating z-scores across the simulation runs. A benefit of this

s that the estimated parameter coefficients allow us to effectivelyompare their relative influences since all independent variables lieithin the same scale. Further, given that parameters are randomly

able 4esults correlating Monte Carlo outcomes and model parameters; t-crit. = 1.96.

Dependent variable (abs. t-stat) Constant sBLP (D) sREPS (%)

numAg 0.00 (0.00) −0.04 (1.65) 0.28 (12.80)

Externality 0.00 (0.00) 0.20 (8.71) 0.05 (2.33)

numAg × extern. 0.00 (0.00) 0.11 (7.61) 0.17 (10.60)

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.02

0.99 0.03 0.00 0.00

perturbed around the baseline values in the process of Monte Carlosimulation, violations of the normality requirement for employinga z-score can be avoided. Table 4 provides the results of these OLSregressions. Unless stated, significance is defined at less than the 5%level and the variable influences are described in decreasing orderof magnitude.

The first dependent variable, the number of agricultural pro-ducers, is only significantly positively influenced by the sREPSparameter, and is significantly negatively influenced by woff, theprobability of off farm employment, and the BLP price premium.Insignificant parameters include the constant term, sBLP, and �BLP.The largest overall effect is negative and follows from the off farmwage. The results are mostly intuitive, higher wages pull produc-ers off their farms and reduce the number of active farms, a similareffect follows from the probability of employment. However, thenegative effect of the price premium is a bit unusual and mayreflect the fact that groups 2 and 4, with the lowest price pre-mium, together represent nearly two thirds of all outcomes andinvolve many agricultural producers. Together these facts suggesta significant negative association. The insignificance of the �BLP andsBLP payments most likely results from the fact that the dependentvariable, number of agricultural producers, pools both the BLP andcommercial farms. And, in such a sample, the sREPS payment, whichis larger and applies to both groups of farmers, is likely to have adominating influence.

The second dependent variable, total amenity and tourismexternality, is most and positively influenced by the price premium,sBLP, and sREPS, and is significantly negatively influenced by the�BLP cost parameter, woff, and the probability of off farm employ-ment. The results are intuitive, higher wages and the probabilityof off farm employment pull producers off their farms and reducethe number of externality generating producers, while higher BLPproduction costs push producers into non-externality generatingcommercial activity, or off the farm altogether. Alternatively, higherBLP premia and farm subsidy payments of both forms help supportagricultural production in general, and BLP externality generatingactivities in particular.

The final dependent variable is a composite of these earliertwo. Because the variables are normalized and may be positiveor negative, an average is the most appropriate interaction term.The analysis indicates that this composite variable is significantlypositively influenced by the price premium, sREPS, and sBLP, and issignificantly negatively influenced by woff, the probability of offfarm employment, and the �BLP cost parameter. Only the con-

stant term has an insignificant effect. The largest overall effect isnegative and follows from the off farm employment wage. Theresults are again intuitive, higher wages and probability of off farmemployment pull producers off their farms and reduce the number

woff (D/h) Probability employ. Price prem. (D) �BLP (%)

−0.60 (27.56) −0.30 (13.61) −0.15 (7.09) −0.02 (1.28)−0.13 (5.83) −0.06 (2.52) 0.54 (23.53) −0.28 (12.18)−0.36 (23.44) −0.17 (11.31) 0.20 (12.57) −0.15 (9.93)

1 se Poli

octfd

S

oiopncBitpdt1pwo

ffttwwpvpd

fitm{wptc

C

lmvolmofdmftpaa

92 H. Kelley et al. / Land U

f externality generating producers, while higher BLP productionosts push producers away from externality generation. Alterna-ively, higher BLP price premia and farm subsidy payments of bothorms help support BLP activities yielding more agricultural pro-ucers and potentially more BLP externality.

ensitivity analysis results

The final simulations are akin to a sensitivity analysis on one ofur key assumptions, which cannot be verified with field data. Wenvestigate the extent to which the magnitude of the additional costf producing in a manner consistent with the BLP, �BLP, impacts ourredictions about land uses and the amount of public good exter-ality produced. Specifically, we consider if farmers, given theiralibrated preferences, might purse activities consistent with theLP in the absence of the BLP payment if additional costs of produc-

ng with this method were not so high. Intuitively, this tells us abouthe extent to which BLP practices can be maintained with farmers’references, but without direct payments and despite higher pro-uction costs. We remove the sBLP payment of D750 and reduce �BLP

o 1.125 (12.5% more than commercial) and then 1.05 (5%) from the.25 (25%) higher cost in the baseline calibration case. Calibratedreferences and all other model parameters are maintained. Earlierhen �BLP = 1.25 and sBLP = 0 all BLP farms switched to commercial

r off farm activities.When we set �BLP = 1.125 and sBLP = 0 we observe that 11 of BLP

arms maintain the BLP approach. Three producers switch to offarm (un)employment, one switches to the commercial form ofheir BLP activity, and five switch to commercial mixed grazing. Theotal externality produced in this case is D4,486,142 {D1,144,994}hich yields a direct cost rate of return of 652% {166%} andhen including indirect costs 317% {81%}. Thus, without the BLPayments and when the BLP is only 12.5% more costly than a con-entional approach, nearly half of the farms are predicted to ceaseroducing the public good externality, although the remaining pro-ucers provide a significant but fragmented externality.

When we set �BLP = 1.05 and sBLP = 0 we observe that, 12arms maintain the BLP approach. The same production patterns observed except that the producer that previously switched tohe commercial form of their BLP activity now maintains the BLP

ethod. The total externalities produced in this case are D4,966,291D1,267,542} yielding direct cost rates of return of 738% {195%} andhen including indirect costs 350% {89%}. Thus, without the BLPayments and when the BLP is only 5% more costly than a conven-ional approach, again nearly half of the farms are still predicted toease producing the public good externality.

onclusions

We set out to achieve two primary objectives with our micro-evel modeling evaluation of the BurrenLIFE program. First, our

ethodological objective was to construct and calibrate an indi-idual household model and to integrate environmentally basedutputs of production into its theoretical structure. Second, ourand management objective was to perform numerical experi-

ents and Monte Carlo analysis with the calibrated model inrder to address three research questions. These included identi-ying first, how optimal are stakeholders’ agricultural productionecisions and to what extent do their preferences account for non-aximizing and heterogeneous decisions? Second, how might BLP

arms production decisions respond to policy or market condi-

ion changes? And finally, how do the public good externalitiesroduced by the BLP farms compare to the program’s costs? Inddressing this last question, we went to great lengths to identify

conservative value for the public goods. This involved the use of

cy 31 (2013) 182– 195

lower bound estimates for amenity and tourism spend and maxi-mum values for program costs. And, when comparing our estimatesto the extant literature, our lower bound estimates were from 15%to 30% lower than the averages across previous research.

Our results indicate that preference calibration is critical forreproducing the farmers decisions, as 15 of the 20 farms are shownto not be pursuing the utility maximizing activity, and farmers’preferences, although lying mostly in the range D0 to D3000, werequite heterogeneous. Once calibrated, the model correctly predictsall farmers participate in the BLP and the externality and tourismspend public good values produced total over D9 m {D2.3 m} peryear compared to annual direct and indirect program costs ofD1.4 m. The annualized rates of return of this program range froman upper bound of 1313% if using direct question WTP elicitationmethods and excluding indirect sunk costs, to the most conser-vative lower bound estimate of 71% when using the indirect WTPvaluation method, assuming all unfavorable parameter perturba-tions (Experiment 4), and including indirect costs. The conclusionwe reach regarding research question 1 is that farmers do notappear to be acting in a maximizing way, their unique preferencesare related to this behavior, and the farmers are quite heteroge-neous. Further, in terms of research question 3, even in the mostunfavorable policy and market environments and using our lowerbound estimates, funds expended to support this scheme gener-ated at least 1.71D of public good value for every 1D spent; and inmost cases much more.

We may take a more meta-analytical approach by compar-ing our externality value estimates to other studies that haveattempted to estimate the monetary value of amenity and biodi-versity and the economic of tourism spend. As stated in Section“Data”, our lower bound indirect WTP ωAmenity estimate was foundto be 15% smaller than the average value reported in similar stud-ies. Further, our lower bound ωTourism estimate was found to be30% smaller than the average across various other studies of walk-ing in forests, coastal lands, or across general areas across all Irelandbetween 1997 and 2007. This first suggests we should focus on ourlower bound estimates of rates of return, as the indirect elicitationmethod is more representative of earlier work. And, if we adjustour values up to the across study averages, the value of amenityand tourism spend for our calibrated model become ˛′ = D2.6 mil-lion yielding a direct and indirect cost rate of return of 185%. Forthe most unfavorable simulation case described above ˛′ = D1.2 myielding a rate of return of 82%. Thus, given others and our research,government exchequers can be confident of a rate of return of71–185%. And, if the direct WTP elicitation method is relevantto this unique Burren landscape, these values could be substan-tially larger. Alternatively, as described in the Section “Empiricalmethodology” sensitivity analysis discussion, the reliability of ourestimates of the public good externality and expenditure ratesof return may be over- or under-stated if our assumptions aboutother model parameters are inaccurate. Specifically, externalitiesand rate of return may be overstated (understated) if the values ofsfa (unemployment assistance) and fDParlor (dairy parlor fixed costs)are higher (lower), if agricultural output prices are higher (lower),or if risk aversion and/or price variability were lower (higher) thanreported or expected.

These results have important management and methodologicalimplications. In terms of management of the Burren and researchquestion two, manipulations predicted to provide the largest singlepolicy effects (i.e. maintaining the number of agricultural producersor the amount of public good externality) will be policy sensitiveto the opportunity costs associated with off farm labor market

conditions. This is especially true for maintaining the number ofagricultural producers. Additionally, the price premia for BLP prod-ucts and as well as mitigating the higher costs associated withBLP style activities are critical for maintaining the public good

se Poli

etomFugetgtafdfmhttsrimct

FtmtpatabTvpoiifpelafmdi

oplopOaaoaOBpfm

H. Kelley et al. / Land U

xternality. The sREPS subsidy is predicted to be the most effec-ive policy for maintaining the numbers of agricultural producersf either commercial or BLP form, while sBLP is more effective foraintaining the environmental public good of particular interest.

urther, results suggest that achieving the joint goal of contin-ed agricultural production and high public good externality, theeneral agricultural subsidy (sREPS), as opposed to the target agri-nvironmental subsidy (sBLP), may be more effective. However,his is most likely related to the much larger magnitude of theeneral subsidy and its availability to non-BLP farms and the facthat fixed costs for switching among similar commercial and BLPctivities are assumed to be zero. Again, however, lower costsor BLP producers and less favorable off farm labor market con-itions will have the largest impacts. In terms of unfavorable offarm labor market conditions, one should in fact interpret this to

ean that when off farm labor market conditions are favorable,igher degrees of government support for either general agricul-ural production or agri-environmental activities will be necessaryo overcome the opportunity cost of remaining in agriculture. Thisuggests subsidies might be beneficially indexed to off farm wageates. Absent such indexing, the four simulation outcome groupsdentified suggests that in unfavorable (favorable) off farm labor

arket conditions maintaining high sREPS subsidy and �BLP higherosts but reducing sBLP subsidy results in transitions from the BLPo commercial agricultural activity (off farm employment).

There are a number of methodological benefits of our approach.irst, the inclusion of non-market values such as environmen-al/cultural amenity and tourism economic impact provides a much

ore complete understanding of the links between the agricul-ural sector and the general population. With valuation of theseublic goods and comparison to program costs, both citizensnd governments can better identify the most beneficial schemeso support. Secondly, our household-level modeling approachllows much more precise predictions to be made regarding theehavioral response of stakeholders to policy and market shocks.his allows more targeted, efficient, and economical policy inter-entions to be made. Third, our unique method of productionreference calibration allows quantification of the monetary valuef non-maximizing decision-making. Given such estimates, one candentify the marginal sensitivity of decision makers to changesn their environment, and identify situations where preferencesor traditional methods may temporarily substitute for direct sup-ort. Finally, employing a portfolio-allocation approach comparingxpected returns for multiple activities is critical. And, the use of aarger dataset for farms with similar soil qualities and productivectivities provides more robust estimates of the value of counter-actual production. Overall, this combination of elements provides

aximal precision and robustness in predicting the response ofecision makers and the externalities they produce given a chang-

ng socioeconomic and institutional environment.The methodological limitations of our approach are centered

n a few primary issues. First, estimating the non-pecuniaryreference value of individual land owners employing particu-

ar production techniques is difficult. This is because a numberf household specific preferences must be aggregated into onereference measure that can be represented in monetary terms.ur approach of aggregating these items is appropriate given thebsence of more specific information regarding an appropriate dis-ggregation of household preferences. Next, our counterfactualutput value predictions rely solely upon cross sectional variationmong the NFS farms given the absence of B20 time series data.ur analysis could be strengthened once time series data for the

20 becomes available. Finally, there may be more relevant inde-endent variables useful for predicting production value and costsor counterfactual activities. Although some of this predictor infor-

ation may be available for the NFS, we are limited by what is

cy 31 (2013) 182– 195 193

reported for our farms of interest, the B20. As an attempt to con-trol for this unobserved variation, constant terms and investmentinstruments are included in all econometric exercises. Interest-ingly, in most cases these terms appeared to have an insignificantor small impact on outcomes, suggesting a limited effect of unob-served variation. Finally, it should be noted that the externalityvalues associated with the BurrenLIFE farming system representextreme lower bounds. Values are limited to the visual impactsdespite the fact that many other ecosystem services such as betterwater quality and health of livestock are also produced. Further-more, the survey used to identify WTP for the externality waslimited to Irish Nationals, was conducted during the height of the2008 recession, and the aggregated value is only normalized tothe population of adult Irish taxpayers. Hence, the true external-ity value is likely larger given the significant numbers of underageand international visitors (e.g. 27% more) to the area.

Appendix A. Theoretical structure

Output. To chose their production activity, agents consider therelation among outputs produced and inputs available assuming aCobb–Douglas production function. Below Aj is total factor produc-tivity (TFP) for activity j. Labor is defined as labor used in excessof the farmers own L = 2000 h endowment, L = max(L − 2000, 10).In log form (lower case) using 2000 h (full-time) or less (part-time)results in a labor input of 1, while employing other family mem-bers or hired labor results in l > 1. M is land available for activity j,H is the activity specific herd, and IN is activity independent farminvestment. Farm k’s output value is then,

Yk,j = Aj · Lˇl,jk · Mˇm,j

k,j · Hˇh,jk,j · INˇin,j

k (A.1)

In Eq. (A.1) the marginal rates of substitution (MRS) sum to one,∑vˇv,j = 1 and v ⊂ (l, m, h, in). Further, Yk,j can be off farm labor

supply, or either agricultural output produced in a commercialor BLP manner. Note that labor and investment are the actualendowments associated with current production. Alternatively, theland and herd inputs are either the current levels or the esti-mated levels needed to perform a counterfactual activity. For offfarm, Aoff = 1, ˇl,off = 1, ˇm,off = 0, ˇh,off = 0, ˇin,off = 0, makingoff farm ‘output’ simply the value of available working hours. Log-ging Eq. (A.1) allows the values for TFP and the MRS to be estimatedfor the four agricultural activities using the NFS326 cross sectionalsample of output values and input costs. Combining these with theB20 endowments yields counterfactual output Y .

Costs. Next we assume agents are aware of the costs associatedwith agricultural activities. These are either actually observed orcounterfactual and estimated marginal and fixed-switching costs,C and F. For counterfactual costs, we estimate two expressions withthe NFS326 dataset for each agricultural activity. For marginal costswe estimate, Cj = ıl,j · lj + ım,j · mj + ıh,j · hj + ε. With the ıs fromthis and k’s input endowments, k’s counterfactual marginal costsfor activity j′ are

Ck,j′ = ıl,j′ · lk + ım,j′ · mk,j′ + ıh,j′ · hk,j′ (A.2)

To obtain fixed-switching costs, we assume fixed costs only occur ifswitching among alternative agricultural activities, but not amongsimilar BLP or commercial activities; i.e. switching from BLP dairyto commercial diary only involves different marginal costs. For dif-ferent activities, we first relate the activity specific herd and landendowments by estimating hj = �j · mj + ε. With the �j ’s ∀j andfarm k’s land endowment we define the counterfactual herd sizefor activity j′ to be h ′ = � ′ · m ′ . With Table 1 per head prices

k,j j k,j

for herds relevant to j (pj), the net fixed-switching cost for farm kcurrently pursuing activity j and considering j′ is,

Fj′ = pj′ · hk,j′ (mk,j′ ) − pj · hk,j (A.3)

1 se Poli

Fotbc

T

Ifststith

pCotwsc(Aw

E

Ftiv(Ffefa(

fBtataCi((aDbt1

R

A

A

94 H. Kelley et al. / Land U

or off farm activity we assume a per minute commuting costf CT = D0.43/min. Next, define � to be the average commutingime per year, � = 45 min × 5 day × 50 week = 11,250, Irish Automo-ile Association. Inserting fixed and marginal costs yields actual orounterfactual total costs.

Ck,j = �k,off · �Empl · � · CT + (1 − �k,off ) · (Fk,j + Ck,j · �k,BLP · Yk,j)

(A.4)

n Eq. (A.4) �k,off = 1 if an agent attempts to supply labor off thearm, else zero and �Empl is the common and fixed probability ofuccessfully finding employment. Multiplying these terms withhe others provides the expected commuting costs if an agentuccessfully seeks out off farm employment. Finally, an assump-ion supported by contextual information is that BLP productionnvolves higher costs; represented by �k,BLP . �k,BLP = 1 if one opera-es their farm in a commercial manner, while �k,BLP > 1 indicatesigher costs when producing consistent with BLP.

Profits. Output prices pj, off farm labor supply wages w, subsidyayments sj, and farm assist payments sfa are obtained from theentral Statistics Office Ireland (CSO) and Teagasc. Also, evidencef random walk prices from Deaton and Laroque (1992) allows uso define expected output prices and wages as pt+1,j = pt,j + ε and

t+1 = wt + ε. Finally, BLP marketing research has shown that theemi-organic outputs produced by the program tend to garner aommon price premium at the market, represented by pBLP. pBLP = 1>1) occurs if a farm is producing in a commercial (BLP) manner.gent k’s actual or expected counterfactual profits are then Eq. (A.5)hich can then be substitution into Eq. (1).

(�k,j) = (1 − �k,off ) · pj · pBLP · Yk,j − TCk,j + sk,j

+ �off · �Empl · w · L + (1 − �Empl) · sk,fa (A.5)

inally, in terms of the cross study by hectare values of amenity andourism economic impact, the values were based upon the follow-ng information. For the value of the amenity externality the totalaluation amounts and relevant hectares were: Willis and Garrod1993), £41 million (D53.3 m) in 1993 with 680 mi2 (176,165 ha);itzpatrick and Associates (2005), D95 m for forest trails and D97 mor forest generally in 2005 with 440,000 of public accessible for-st (Coillte Ireland, 2005); Campbell et al. (2006), D737.6 m in 2005rom D643.22 per person and 1.15 m taxpayers for potentially allgricultural land (64% of all Irelands 7 m ha) of 4.2 m ha; Hynes et al.2007), D1.8 m for 500 ha of commonage, beach and visible areas.

For tourism economic impact, assuming a 50% economic impactactor, the total expenditure amounts and relevant hectares were:ergin and Rathaille (1999) and Buckley and van Rensburg (2006)otal expenditure of D146 m. Given the absence of informationbout their specific destinations, a simple calculation can usehe 2.38 m ha of walkable Irish land (given 2% urban and 64%gricultural, this leaves 34% of the total 7 m ha of Ireland);urtis and Williams (2004), from an estimated tourism spend-

ng D83.2 m and with 2.38 m ha of walkable land; Failte Ireland2004) from a total across 2000–2003 expenditure of D690.8 m2004, D170 m) and again 2.38 m ha of walkable land; Fitzpatricknd Associates (2005) forest walking (general forest) generated307 m (D268 m) in 2005 and with 440,000 ha of public accessi-le forest; Barry et al. (2011) coastal walking tourism expenditureotaled D340 m with 1,480,000 ha of coastal land (Marine Institute,996: 7400 km × 200 m wide).

eferences

hn, C., Singh, I., Squire, L., 1981. A model of an agricultural household in a multicropeconomy: the case of Korea. Review of Economics and Statistics 63 (4), 522–525.

werbuch, S., Berger, M., 2003. Energy Diversity and Security in the EU: Apply-ing Portfolio Theory to EU Electricity Planning and Policy-making. InternationalEnergy Agency, Paris, February, Report No. EET/2003/03.

cy 31 (2013) 182– 195

Barry, L., Rensburg, T., Hynes, S., 2011. Improving the recreational value of Ireland’scoastal resources: a contingent behavioral application. Marine Policy 35 (6),764–771.

Bergin, J., Rathaille, M.O., 1999. Recreation in the Irish Uplands. Report for Moun-taineering Council of Ireland, Republic of Ireland, Dublin.

Blank, S., 2001. Producers get squeezed up the farming food chain: a theory of cropportfolio composition and land use. Review of Agricultural Economics 23 (2),404–422.

Bohnsack, U., Carrucan, P., 1999. An assessment of farming prescriptions under theREPS in the uplands of the Burren karstic region, Co. Clare. The Heritage Council.

Brueckner, J., 2000. Urban sprawl: diagnosis and remedies. International RegionalScience Review 23 (2), 160–171.

Buckley, C., van Rensburg, T.M., 2006. An analysis of the financial returns associatedwith commonage in Ireland. In: Contributed Paper Presented at the AgriculturalResearch Forum, Tullamore, Republic of Ireland, March 15–16.

BurrenLIFE, 2010. Farming for conservation in the Burren. Technical FinalReport covering the project activities from 01.09.2004 to 31.01.2010.LIFE04NAT/IE/000125.

Campbell, D., Hutchinson, W., Scarpa, R., 2006. Using discrete choice experimentsto derive individual-specific WTP estimates for landscape improvements underagri-environmental schemes. FEEM Working Paper 26.

Central Statistics Office Ireland, 2008. 2006 Census of Population. CSO, Dublin,Ireland.

Connolly, L., Kinsella, A., Quinlan, G., Moran, B., 2008. National Farm Survey. Teagasc,Rural Economy Research Centre, Athenry.

Cooper, T., Hart, K., Baldock, D., 2009. The provision of public goods through agri-culture in the European Union. Report Prepared for DG Agriculture and RuralDevelopment, Contract No 30-CE-0233091/00-28, Institute for European Envi-ronmental Policy, London.

Curtis, J.A., Williams, J., 2004. A national survey of recreational walking in Ireland,Economic and Social Research Institute. Report for Irish Sports Council, Dublin,Republic of Ireland.

DAFF, 1996. Rural Environment Protection Scheme, Report on Agri-EnvironmentalSpecifications. Department of Agriculture, Food and Forestry.

DAFF, 2004. Terms and Conditions of the Rural Environment Protection Scheme.Department of Agriculture, Food and Forestry.

DAFF, 2008. Annual Report 2007: Rural Economy and Environment. Department ofAgriculture, Food and Forestry.

DAFF, March 2011. Brief for Minister for Agriculture, Fisheries and Food. Departmentof Agriculture, Food and Forestry.

Deaton, A., Laroque, G., 1992. On the behavior of commodity prices. Review of Eco-nomics and Statistics 59, 1–24.

DeLaquil, P., Awerbuch S., Stroup K., 2005. A portfolio-risk analysis of electricitysupply options in the commonwealth of Virginia. Chesapeake Climate ActionNetwork Report.

DeLong, J.B., Shleifer, A., Summers, L., Waldmann, R.J., 1990. Noise traders and riskin financial markets. Journal of Political Economy 98 (4), 703–738.

Department of Finance, 2000. Budget 2000. Summary of Budget Measures, Annex C.Distribution of Tax Units in the Income Tax System.

Department of Social Protection, 2009. Annual SWS Statistical Information Report.Dunford, B., 2002. Farming and the Burren. Teagasc Publications, Ireland, pp. 32–51.Emerson, H.J., Gillmor, D.A., 1999. The rural environment protection scheme of the

Republic of Ireland. Land Use Policy 16 (4), 235–245.Eshet, T., Ayalon, O., Shechter, M., 2006. Valuation of externalities of selected waste

management alternatives. Resources Conservation and Recycling 46, 335–364.Failte Ireland, 2004. Hiking/Walking 2003. Dublin.Fitzpatrick and Associates, 2005. The Economic Value of Trials and Forest Recreation

in the Republic of Ireland. Coillte Publications, Wicklow.Goldman, G., Nakazawa, A., Tayor, D. 1994. Impact of Visitor Expenditures on Local

Revenues. University of Alaska, Fairbanks, College of Rural Alaska. WREP-00145.Harrison, G.W., Rutström, E.E., 2008. Experimental evidence on the existence of

hypothetical bias in value elicitation methods. In: Plott, C.R., Smith, V.L. (Eds.),Handbook of Experimental Economics Results, vol. 1. North-Holland, New York.

Henderson, J., 1977. Externalities in a spatial context. Journal of Public Economy 7,89–110.

Hynes, S., Buckley, C., van Rensburg, T.M., 2007. Recreational pursuits on marginalfarmland: a discrete-choice model of Irish farm commonage recreation. TheEconomics and Social Review 38 (1), 63–84.

Kelley, H., 2004. Asset pricing with behavioral traders: a test of the country funddiscount. Journal of Behavioral Finance 5 (4).

Kelley, H., Evans, T., 2010. Measuring the impact of behavioral traders in the marketfor closed-end country funds from 2002 to 2009. In: Bruce, B. (Ed.), Handbookof Behavioral Finance. Edward Elgar, Cheltenham.

Kelley, H., Evans, T., 2011. The relative influences of land-owner and landscape het-erogeneity in an agent-based model of land-use. Ecological Economics 70 (6),1075–1087.

Kelley, H., Jeserich, N., 2012. The impact of unobserved productivity on agriculturalproduction functions estimates. Working Paper, National University of Ireland,Galway.

Kleindorfer, P., Li, L., 2005. Multi-period VaR-constrained portfolio optimizationwith applications to the electric power sector. Energy Journal 26 (1), 1–26.

Klibanoff, P., Lamont, O., Wizman, T., 1998. Investor reaction to salient news inclosed-end country funds. Journal of Finance 53, 673–700.

Latacz-Lohman, U., Hodge, I., 2003. European agri-environmental policy for the21st century. Australian Journal of Agricultural and Resource Economics 47 (1),23–139.

se Poli

L

M

M

M

O

O

P

H. Kelley et al. / Land U

ist, J.A., Gallet, C., 2001. What experimental protocol influence disparities betweenactual and hypothetical stated values? Environmental & Resource Economics20, 241–254.

eredith, D., 2010. Rural areas in transition. In: Winchell, D.G., Ramsey, D., Koster, R.,Robinson, G.M. (Eds.), Geographical Perspectives on Sustainable Rural Change. ,pp. 220–242.

oles, N., Moles, R., 2002. Influence of geology, glacial processes and land use onsoil composition and quaternary landscape in the Burren National Park, Ireland.Catena 47, 291–321.

urphy, J.J., Allen, P.G., Stevens, T.H., Weatherhead, D., 2005. A meta analysis ofhypothetical bias in stated preference valuation. Environmental & ResourceEconomics 30, 313–325.

lley, S., Pakes, A., 1996. The dynamics of productivity in the telecommunications

equipment industry. Econometrica 64 (6), 1263–1297.

’Rourke, E., 2005. Socio-natural interaction and landscape dynamics in the Burren,Ireland. Landscape and Urban Planning 70, 69–83.

arks, P., 1995. Explaining ‘irrational’ land-use: risk aversion and marginal agricul-tural land. Journal of Environmental Economics and Management 28, 34–47.

cy 31 (2013) 182– 195 195

Pearce, D.W., Putz, F., Vanclay, J.K., 2002. Is sustainable forestry economically pos-sible? In: Pearce, D., Pearce, C., Palmer, C. (Eds.), Valuing the Environment inDeveloping Countries: Case Studies. Edward Elgar, Cheltenham, pp. 447–500.

Saunders, J., 2008. Tourism in the Burren: Summary of Literature and Policy Docu-ments for Burren Connect.

Springer, U., 2003. Can the risks of Kyoto mechanisms be reduced through portfoliodiversification. Environmental and Resource Economics 25 (4), 501–513.

The Single Payment Scheme, 2004. An Explanatory Guide. Department of Agricultureand Food.

Willis, K.G., Garrod, G.D., 1993. Valuing landscape: a contingent valuation approach.Journal of Environmental Management 37 (1), 1–22.

Wilson, M., Costanza, R., Boumans, R., Liu, S., 2005. Integrated assessment and valu-ation of ecosystem goods and services provided by coastal systems. In: Wilson,

J. (Ed.), The Intertidel Ecosystems: The Value of Ireland’s Shores. The Royal IrishAcademy, Dublin, pp. 1–24.

Yadav, L., van Rensburg, T., Kelley, H., 2010. A Comparison Between the ConventionalStated Preference Valuation Technique with a Prediction Approach. WorkingPaper, NUIG, Ireland.