Designing and Rating an Area YieldCrop Insurance Contract

Jerry R. Skees, J. Roy Black, and Barry J. Barnett

This article documents the design and rate-making procedures used in thedevelopment of the Group Risk Plan (GRP}-the new federal crop insurance productthat insures based on area yield. The authors of this article worked closely withpersonnel in the Federal Crop Insurance Corporation and others in developingmethodological and practical constraints needed in implementing a workable areayield contract. GRP indemnity payments are made based on percentage shonfalls inactual county yields relative to a forecasted yield. Historical county yield data areused to develop forecasted yields and premium rates.

Key words: agricultural policy, area yield, crop insurance.

This journal has published several articles onarea yield insurance. Halcrow published thefIrst such article in 1949. Miranda revisited theissue in 1991. In this article we document thedesign, rating, and implementation of the U.S.area-based yield insurance called the GroupRisk Plan (GRP). Both methodological andpractical constraints were imposed to develop aworkable area yield contract.

Much of the interest in area yield insurancehas been motivated by concerns with problemswith the traditional farm-loss crop insurance of-fered through the Federal Crop Insurance Cor-poration (FCIC). These problems are welldocumented (Congressional Commission forthe Improvement of the Federal Crop InsuranceProgram; Goodwin and Smith; Just and Calvin;Quiggin, Karagiannis, and Stanton; Smith andGoodwin; U.S. General Accounting Office1991, 1992). In some regions, the dual prob-lems of adverse selection and moral hazardhave resulted in low participation and high gov-ernment costs. Halcrow's work recognizedthese problems and recommended that insuring

Jerry Skees is professor of agricultural economics at tbe Universityof Kentucky, Roy Black is professor of agricultural economics atMichigan State University. and Barry Barnell is usistant professorof agricultural economics at Mississippi State University. Thiswork wu completed while Barnett wu a post-doctoral researchscholar in the Department of Agricultural Economics at tbe Uni-versity of Kentucky.

Funding for this research wu provided by tbe University ofKentucky Experiment Station and the U.S. Department ofAgriculture's Economic Research Stanon and tbe Risk Manage-ment Agency.

based on area yields would mitigate adverse se-lection and moral hazard problems.

Halcrow noted that, in many areas, variationsfrom forecasted farm-level yields are largely afunction of systemic risk such as the pervasivedrought that occurred in the Midwest in 1983and 1988. Area yield insurance provid~s effec-tive risk management only in areas where yieldrisks are largely systemic. There are areaswhere yield risks are not systemic. An exampleis the sub-mountainous deciduous fruit produc-ing region around Hood River, Oregon. In asub-mountainous region, freeze is the majorsource of yield risk, and the probability and ex-tent of yield loss are largely a function of local-ized topographical features such as elevation.Area yield insurance will not provide effectiverisk management in an area such as this.

Given the problems with farm-level crop in-surance-and in an attempt to provide farmerswith a viable alternative in areas where farm-level crop insurance exhibited serious short-comings-the U.S. Department of Agriculture(USDA) initiated a pilot test of area yield insur-ance for soybeans in selected U.S. counties in1993. The U.S. area yield insurance, known asthe Group Risk Plan (GRP), makes indemnitypayments based on shortfalls in county yields(the area equals the county). County yields areestimated by the National Agricultural Statisti-cal Service (NASS) of USDA.

In 1992, Skees provided the backgroundanalysis and development for the 1993 pilot testof GRP on soybeans. The president's proposed

Amer. J. Agr. Econ. 79 (May 1997): xxx-xxxCopyright 1997 American Agricultural Economics Association

2 May 1997

1994 budget strongly endorsed GRP as a re-placement for farm-based crop insurance. Inresponse, Congress did not eliminate farm-based crop insurance, but instead mandatedthat GRP be expanded "to the extent practi-cable." As a result, GRP was expanded in 1994to 1,875 county-crop programs for wheat, corn,soybeans, grain sorghum, cotton, barley, for-age, and peanuts. GRP was made available intwenty-seven states. When forage is ex-cluded, GRP is available for nearly 70% ofall U.S. acres for the seven major crops(Skees 1993).

As an example of how the GRP works,suppose that for a given year a county has aforecasted soybean yield of 40 bushels peracre. This forecast is made about six monthsbefore farmers make their insurance pur-chase decisions and plant their crops. Sincefarm yields can be greater than countyyields, and following principles of hedgingthat are developed by Miranda, farmers areallowed to scale the amount of protectionthey purchase by up to 150% of the fore-casted yield times the expected price. AsMiranda documents, such scaling is impor-tant for providing more risk protection, asfarm yields are not perfectly correlated withcounty'yields and are based on the relativevariation of farm to county yields. The scal-ing design follows the optimal hedge litera-ture. A scaling of 150% in a county with aforecasted soybean yield of 40 bushels andan established FCIC price of $6.00 allows afarmer to purchase up to $360 of protectionper acre. Suppose a farmer in the county pur-chases $360 of protection per acre and se-lects a 90% coverage leveL. This. participantwill receive an indemnity if the actual countyyield is below the critical yield of 36 bushelsper acre (40 bushels per acre x 90%). For ex-ample, if the actual county soybean yield were27 bushels per acre, or 25% below the triggeryield of 36 bushels per acre, the GRP partici-pant would receive an indemnity payment of$90 per acre ($360 x 25%). There are obvi-ous advantages to allowing a scale-up.

The United States is not the first country to

I We use the terms Uprolec:tion" and Ucoverage" as they are used

by the USDA's Federal Crop Insurance Corporation (FCIC). Read-ers who are familiar with insurance lileralure may initially findthis confusing. The FCIC uses the term "coverage" 10 mean oneminus the percenlage deductible. The FCIC uses the lerm "protec-tion" synonymously with lhe insurer's liability, or what the insur-ance lilerature oflen refen 10 as coverage.

12 ... MA Y1I7 2

Amer. J. Agr. Econ.

offer insurance based on area yields. Swedenbegan developing an area yield insuranceprogram in 1952 and implemented a programin 1961. Officials from the Canadian prov-ince of Quebec studied the Swedish programduring the mid 1970s and introduced theQuebec area yield insurance program in1977. The idea was resurrected in the UnitedStates in the 1989 principal report of theCongressional Commission for the Improve-ment of the Federal Crop Insurance Program.The Food, Agriculture, Conservation, andTrade Act of 1990 included language autho-rizing a pilot test of what was then beingcalled the "area yield program."

In 1990, Barnaby and Skees presented ar-guments for area yield insurance and de-scribed how such a program might operate.In 1991, Miranda formalized Halcrow's ear-lier insights, described the conditions underwhich area yield insurance would reduce afarmer's yield risk, and extended the litera-ture on optimal hedge ratios to show theamount of protection farmers should purchase.Miranda developed a theoretical framework forevaluating the systemic risk protection pro-vided by area yield insurance and applied thatframework to a sample of 102 western Ken-tucky soybean producers.

Using an optimal hedge format, Mirandaevaluated the effectiveness of an area yieldinsurance program with a sample of 102 soy-bean farms from a twenty-two county area inwestern Kentucky that is typical of thesouthern Corn Belt. These farms generallyhad available at least fifteen years ofdetrended yield data. Miranda concluded that"... for mQst producers, area-yield insurancewould provide better overall yield risk pro-tection than individual-yield insurance" (p.242). Miranda allowed both the scaling andthe deductible to vary beyond any politicallyacceptable limits.

In 1993, Hourigan used historical farm-level data from nearly 3,000 soybean farmsin several states to show that, during the1980s, ~ver 60% of those farms would havehad a lower coefficient of variation in grossrevenue by using GRP at the 90% coveragelevel rather than the farm-based crop insur-ance offered by the Federal Crop InsuranceProgram at the 75% coverage level. Smith,Chouinard, and Baquet evaluated the vari-ance reduction provided by an area yield in-surance program for a sample of 123 sepa-rately insured dryland wheat units in Chouteau

III'JNII7 .- DOO

Skees. Black. and Barnett

County. Montana. in 1994. They also found thatarea yield insurance provided risk protectionfor many of the farms in their sample.2

Designing an Area Yield Insurance Policy

Since indemnities are triggered only when areayields are low, an area yield policy is not liketraditional multiple-peril or named-peril (e.g.,hail) crop insurance. Rather, it is an option onan index. Just as with a put option on a futurescontract, an area yield policy has an associatedbasis risk. Farmers may experience farm-levelyield losses when area yield shortfalls are notsufficient to trigger an indemnity payment un-der an area yield policy. Lowering the chancesof such an event (i.e., lowering the basis risk) isan important objective when designing an areayield policy. The magnitude of the basis risk isaffected primarily by two elements of the con-tract design: (a) the area to be used for theyield index, and (b) the procedures for forecast-ing the central tendency in yields for the area.Two additional areas of contract design haveimportant implications for managing basis risk:(a) the indemnity payout rules, and (b) the do-main of insurance deductible and protectionchoices. All of these contract' design consider-ations are politically constrained.

Each of these contract design elements re-ceived attention during development of theGRP pilot. The reasoning for the decisions thatWeI"C made is presented below. It was decidedthat (a) counties were the only practical areafor the yield index, (b) robust double exponen-tial smoothing would be used initially to estab-lish the central tendency in yields, (c) indem-nity payments would be made based on the per-centage shortfall in area yields rather than thebushel shortfall as tYpically presented in the agri-cultural economics literature (Halcrow; Miranda;Smith, Chouinard, and Baquet), and (d) farmerswould be allowed to scale up protection levelsand purchase GRP at between 90% and 150%

. Miranda IUed actual farm yield data from the Kentucky Farm

Business Analysis program. Both Hourigan. and Smith. Chouinard.and Baquet used ten yean of actual farm yield data from federalfarm-level crop insurance reconls. Smith. o.ouinard. and Baquet con-dueled their analysis using insurance units. a lower level of agrega-tion than the farm. Units are described by ownenhip and location ofparcels on farms. Had data been available, a more complete analysiswould have considered the risk reduction provided by an area yieldinsurance policy in combination with a named-peril hail insurancepolicy. The area yield policy would protect against losses fromsystemic risks. such as drought. while the named-peril policywould protect against the largely nonsyslemic risk of hail loss.

12 su. MAV97 , ,-.

Area Yield Crop Insurance Contract 3.

of the per acre forecasted value of the crop inthat county.

Selection of the Area

Area yield insurance contracts must be basedon an index of area yields. To reduce basis risk,the area or zone boundaries for an area yieldcontract should be selected so as to group to-gether the largest possible number of farmswith similar soils and climate. In Quebec, theboundaries frequently are redrawn as new data aremade available and as farmers learn that theiryields more closely match those of a contiguouszone rather than their current zone. The data re-quirements for a system such as the one used inQuebec can be formidable. In Ontario, a less re-source-intensive index is used. For some crops(e.g., forage), the Ontario crop insurance programuses plant growth simulators to forecast yields. Itis likely that the Quebec system has less basisrisk than the Ontario system or the GRP.

In the United States, NASS county yields arethe only available historical area yield data. Anarea index based on county yields is not idealsince county boundaries do not necessarilygroup together farmers with similar patterns ofyear-to-year percentage deviations from fore-casted yields. However, since plant growthsimulators cannot capture certain causes ofcrop yield losses, county yield shortfalls shouldbe a superior measure of systemic risk. .

Two criteria were used to select county-cropprograms. First, NASS county yield data gener-ally are available from 1956 to the present. AIlcounties selected had to have at least thirty-onecontinuous years of NASS county yield dataavailable. Second, all counties selected had tohave at least 15,000 acres of the commodity(10,000 acres for peanuts) planted in the mostrecent year for which NASS data were avail-able. In short, only major production areas wereselected. Since NASS invests more resourcesinto making yield estimates for major produc-tion regions, the yield estimates from these ar-eas should be more accurate than those forfringe areas. Attention was also given to assurethat GRP counties selected would form a gener-ally contiguous block (Le., isolated countieswere eliminated).

County yield data can be calculated on eithera planted acre or a harvested acre basis. MostGRP contracts are based on planted acre yields,since these yields wiU represent the true riskassociated with abandoned acres. For most

8120187. 4:28 PM

May 19974

crops and in most areas, GRP contracts arebased on aggregate acreage and productionwithout regard for production practices. How-ever, in some cases, to improve the yield index(i.e., reduce basis risk), practice-specific GRPcontracts are offered. For example, if sufficient(as described above) practice-specific NASSdata are available, GRP contracts are offeredfor both irrigated and nonirrigated production.

Forecasting the Central Tendency in CountyYields

Any given yield outcome can be divided intotwo components: the central tendency, and thedeviation.from the central tendency. One couldargue. that the central tendency is a function ofmanagement and resource endowment (e.g.,cultural practices and the quality of land farmed),while the deviation from the central tendency isa function of nature (e.g., drought, excess mois-ture, hail, excessive temperature, frost). Thepurpose of crop insurance is to provide protec-tion against the financial consequences of yieldshortfalls due to natural hazards.

The method used to estimate central tendencyshould be robust, performing well under a widerange of circumstances and reducing the influ-ence of obvious outliers. The method must gen-erate two-year-ahead forecasts since there is atwo-year lag in the availability of NASS acre-age and production data. Estimators withknown statistical properties are preferred to es-timators whose properties are unknown, butconsideration should be given to estimators thathave withstood the test of time and have beenwidely used in analogous situations. These in-clude methods commonly discussed in the Jour-nal of Forecasting. While it may be desirable touse methods that would capture a wide array ofmarket forces, the most critical requirement isthat the method be capable of capturing the ef-fects of technology trends. Finally, the methodpreferably should not be so complex that it can-not be intuitively understood by crop insuranceagents and potential GRP purchasers.

The choice of methods for estimating centraltendency is constrained by the availability ofresources (primarily technical expertise) withinthe federal agency that administers the program.Structural models of the central tendency in yieldfor each GRP contract are not practical and thenumber of data series that would need to bemaintained for such models is cost prohibitive.

One class of candidates for estimating thecentral tendency in yields, locally weighted

12 SIr8M 1M V!I7 .

Amer. J. Agr. Econ.

scatter plot smoothing (LOWESS), was ruledout. LOWESS (Cleveland 1979, 1993; HardIe)was eliminated from consideration because ofits lack of widespread application. The firstGRP contracts estimated central tendency usinga robust double exponential smoothing proce-dure which provided a locally linear estimate oftrend (see Skees, Black, and Barnett). In 1995,central tendency was estimated for wheat GRPusing ARIMA models developed by Ker andGoodwin. Concerns were raised about the sta-bility of the AIRMA models due to the shortlength of yield data available. Further, sugges-tions in the literature and by practitioners indi.cate that specification and estimation are acase-by-case task. In 1996, the method for esti-mating central tendency changed to providemore stability in estimates. Specifically, the useof spline regression was employed to fit up totwo linear splines to the data. Robust regressionprocedures for elimination of outliers were alsoincorporated. Nonlinear optimization tech-niques were used to determine the year atwhich a change in slope was warranted.

Using a piece-wise (spline) procedure for fit-ting up to two robust linear trends provides anextrapolative forecasting method that meets therequirements presented earlier. It is a prag-matic, intuitive approach that is widely used,and it is straightforward for the administrativeagency to use in forecasting future yields with-out a need to redo the analysis every year.

Indemnity Payout Rules

For any given crop year, the area yield, y, is arandom variable. The insurer's forecast of thearea yield is given by yfcast. The insured se-lects a scale of between 0.9 and 1.5, and a yieldcoverage level (cov) of between 0.7 and 0.9(the deductible equals 1 - cov). The criticalyield, Yco is calculated as

(1) Ye = yfeast X eov.

The insured receives an indemnity, indem,whenever Y < Yco The indemnity is calculated as

(2) ind.m = -[(\~ Y}Y/cast)(.rcak)o 0]

where yfeast times scale is the insurer's liabil-ity (the insured's protection) or the maximumpossible indemnity payment. For ease of pre-

-- .-....

Skees. Black. and Barnett

sentation, indemnities are measured in units(say, bushels) per acre instead of dollars. Toconvert to dollars, simply multiply by totalacres insured and price per bushel.

Miranda, and Smith, Chouinard, and Baquetexamine the performance of several differentstylized area yield insurance contracts. All ofthese designs differ from the current GRP con-tracts in important respects. Instead of usingNASS county yields, Miranda, and Smith,Chouinard, and Baquet calculated y and yfcastbased on a weighted average of a sample of in-dividual farm yields-thus creating their ownareas with a sample of farms.

Both Miranda, and Smith, Chouinard, andBaquet utilize different indemnity payout rulesthan those used in GRP. Specifically, they cal-culate indem as

indem = max[(yc - y)scale, 0].

Miranda, and Smith, Chouinard, and Baquetmeasure the yield shortfall as the simple differ-ence between Yc and y (the shortfall is denomi-nated in units such as bushels per acre). GRPpays indemnities based on yield shortfalls mea-sured as a percentage relative to Yet as shown inequation (2). Measuring yield shortfall as a per-centage creates a disappearing deductible.When major yield shortfalls occur, indemnitypayments will be significantly higher underGRP rules than they are under the Miranda, andSmith, Chouinard, and Baquet rules. In the ex-treme, if y = 0, the percentage shortfall will be100% and the indemnity will be 100% of pro-tection regardless of the level of cov. This fea-ture makes a GRP policy with a low level ofcov more attractive for those who are con-cerned only with catastrophic yield losses.3 Thedisappearing deductible rules also increases theimportance of loading rates for catastrophicyields that have not been experienced in a shortseries of data.

, If the insunmce purchaser could optimally choose levels of

cov and/or scalll under the Miranda. and Smith. Chouinard. andBaquet indemnity payout rules. it would be possible to achieve sig-nificaatly higher indemnity payments when major yield shortfallsoccur. However. the maximum values of cov and scalll are politi-cally constrained. Federal Crop Insurance Corporation decisionmakers. whose backgrounds are in farm-level crop insurance. arenot comfortable with an area yield insurance program that allowsfor very high levels of cov or scalll. While conceptually one couldoffer an area yield insurance policy for any combination of nonnc-gative values of cov and scaill. this could lead to politically unpal-atable outcomes. Specifically. if cov > 100-.. indemnity paymentswould be made even If the actual yield. y. is greater than the fore-casted yield. ycast. If sealll is allowed to be very high. indemnitypayments may be very high.

(3)

12 SIc8M MA YII7 I

Area Yield Crop Insurance Contract 5' ~

Establishing the Domain of InsuranceDeductible and Protection Choices

The GRP is a very specific form of an areayield insurance product in which political con-straints circumscribe the domain of theinsured's choice variables. Currently, the in-sured may choose a value for cov from a lim-ited set of discrete possibilities. Specifically,cov can be set at 70%, 75%, 80%, 85%, or90%.4 The Federal Crop Insurance Reform andDepartment of Agriculture Reorganization Actof 1994 (P.L. 103-354) authorized values of covup to 95%. As of yet, a 95% cov has not beenimplemented. The insurer selects a value forscale anywhere between 90% and 150%.'

Most farm-level yields are more variable thancounty-level yields. Thus, to achieve a target in-demnity payment that will meet cash flow short-falls in low-yield years, insured farmers canchoose levels of scale up to 150%. The choice ofan optimal level of scale is analogous to an op-timal hedge ratio in the futures literature.

Rate Making for the Selected Design of AreaYield Insurance in the United States

As much as possible, an effort was made toadopt GRP rate-making procedures that wereconsistent with standard FCIC procedures. Thepure premium rate for any given level of cov issimply the average yield shortfall in percentageterms over the historical time series of avail-able data. A pure premium rate is calculated atcov = 100% since the precision of the estimatesis higher at zero deductible than for any posi-tive levels of deductibles. Ultimately, a para-metric structure of rate relativities is imposedto generate premium rate relativities. Pure pre-mium rates also are calculated for the variouslevels of cov used in the GRP (70%, 75%, 80%,85%, and 90%).

Premium Rate Relativities

Rating a county yield insurance product pre-sented some special challenges. It was deter-mined that rating the individual county without

. The Federal Crop Insurance Corporation is considering a pro-

posal to allow GRP purchasers to select a value for COy at any dis-crete level between 70~ and 9S~.

. The maximum value of scalI! is set at ISO~. The minimum

value is set at 0.6 times the maximum value of scalI! which equals90~. These values are constrained politically, unlike the work ofMiranda. and Smith. Chouinard. and Baquet. lu that work, they al-lowed the scaling to reach extremely high values.

II/!NIIII7 ""'" IIU

6 May 1997

using information from nearby counties wouldbe inappropriate. Spatial correlation amongcounty yields is important for rating. Further,

. since only thirty-five to forty years of data were

available, it was critical to load the rates for cata-strophic events that may not have been experi-enced. Providing structure to rate relationshipacross coverage levels became a key in anchoringrates. This is a common practice in rate making.The empirical rates for the various coverage lev-els (sum of all percentage shortfalls divided bythe number of years) are useful. However, usingempirical rates has some limitations. For ex-ample, within the time series of NASS yield dataavailable, some counties never may have experi-enced situations where y < Yc at the lower levelsof cov. This results in a pure premium rate of 0.0(and therefore a premium of SO.OO) at that levelof cov. bbviously, this is not realistic sincethere is always some likelihood that y will beless than Yco even at low levels of cov.

A graphical review of the many county distri-butions of the residuals about the estimate ofcentral tendency demonstrated that countyyields are typically asymmetrically distributedwith a negative skew and a thick left tail. Sta-tistical tests of skewness confirmed this obser-vation. Thus using a normal assumption foryield distribution is inappropriate.

As the first step in transformation to a nega-tively skewed, thick left-tailed distribution, anapproximation based on the Botts and Bolesprocedure was implemented. The Botts andBoles procedure calculates premium rates giveninformation on the coefficient of variation of anassumed normal yield distribution. No countycoefficient of variation was allowed to be lessthan 15%.6 The implied coefficient of variation,based on the empirical premium rate at cov =100%, was put into the Botts-Boles algorithmto develop parametric premium rates for eachlevel of cov used in the GRP. The resultingparametric premium rates were then loaded bydividing through by the cov times 0.01. The ef-fect of this procedure is to add a greater load(in percentage terms) to the lower cov levels.There are at least two reasons for loading pre-mium rates in this manner. First, lower levels ofcov will have lower premium rates. Yet, at

. While imposing a lower bound of 15"'" on the CV for yields

was somewhat arbitrary. this value gives very low premium rates.Even rates at the 90"'" COY are around 2.5"'". In a series of thirty-five years, it only require. two years with yield shortfalls of 40","below the critical yield in order to obtain a rate of about 2.5"'"(8O"'" divided by 35 = 2.29). This procedure protects against unex-perienced yield 101Ses. It Is not likely that farmers in counties withIInle or no experience of yield losses will find ORP amactive.

12 SIaMt MAYt7 .

AIII~r. J. Agr. Ecolt.

lower levels of cov, there are fewer empiricalobservations where y < Yc' Hence, there is lessempirical information on which to base an em-pirical premium rate at lower levels of cov. Sec-ond, low empirical premium rates (at any levelof cov) may simply reflect the fact that a truecatastrophic yield has not occurred during thehistorical time frame being considered. This,however, may lead to the wrong conclusionabout the possibility of such an event occur-ring. The loaded parametric premium rates areconsistent with a negatively skewed distribu-tion with a thick left tail (such as a beta). Fur-ther, rate relativities generated using this proce-dure also match the average empirically basedrate relativities.

For a given GRP contract, the "parametric"premium rate is compared to the empirical pre-mium rate at each level of cov. If the empiricalrates demonstrate a greater frequency of seriousyield shortfalls than the parametric procedure,the empirical rates are used.

Reserve Loading

It is standard insurance practice to add a re-serve load to premium rates in order to build re-serves. In principle, the concept assumes that a re-serve load will increase the chances that reservefunds can be accumulated before a major payoutoccurs. Following standard practice for the tradi-tional federal farm-level crop insurance product,reserve loading is implemented by dividing theselected premium rates (either the empirical val-ues or the parametrically adjusted rates) by 0.9.These procedures were imposed by the FCIC.

Using Contiguous Counties to Smooth Rates

Premium rates for the federal farm-level insur-ance product are smoothed cross-sectionally byutilizing information from contiguous rating ar-eas. For this reason, and following the lines ofcredibility theory, the final premium rate for eachcounty GRP contract is calculated as a weightedaverage of the premium rate for that county andthe premium rate for each contiguous GRPcounty. The weights are calculated as follows:

{ acres planted,.".,)G..S (4) w,."" = O.

100,000

subject to 0.4 S W"",ct S 0.9, where w-r.' is theweight assigned to the target county and acres

tII'XIR7 .- 8U

Sk~~$. Black. and Barn~1t Ar~a Y;~ld Crop Insuranc~ Contract

planted,."" is the average acres planted in thetarget county over the most recent ten-year pe-riod; and

(5) waIIIIi6(l} = (1 - w8IrIJn)

where W cortd,(1) is the weight assigned to the ithcontiguous QRP county. and acres plantedt:Oltlil(1)is the average acres planted over the most re-cent ten-year period for each contiguous GRPcounty. All weights sum to one. The final pre-mium rate for each county GRP contract is cal-culated as

(6) . final premium ratellJ""

= (w..".,)(rate,.",,) + I:,[(WC"",,,,(I)(rateC""dl<l)].

Information from all available contiguous GRPcounties is used to smooth rates.

Figure 1. Percentage of GRP corn insurance premium to all federal crop insurance premiumfor corn in 1995

Experience with GRP

While only two years of GRP experience isavailable since the major expansion, that expe-rience does provide some information. In thetwo crop years since the major expansion, GRPpremiums have exceeded $8 million in eachyear. The loss ratio was 1 % in 1994 and 16% in1995. While the APH loss ratio was 63% and100% of premium for 1994 and 1995, it is tooearly to make meaningful comparisons of lossratios for the GRP versus the APH. In particu-larly bad crop years, one can expect GRP lossratios to exceed APH loss ratios. The major par-ticipation has been in GRP corn policies com-prising nearly $6 million of the $8.7 million intotal premium in 1995. While $6 million is arelatively small share of the $372 million forAPH corn in 1995, there are markets whereGRP comprised a significant share of the totalpremium for corn insurance. In nearly one hun-dred counties, GRP comprised more than 5% ofthe corn insurance in 1995. As figure 1 reveals,

[ acres planted--'(I) ):t acres planted--'(I)

t?'

CJ < 1%1%-5%> 5%

III..

. May 1997

these counties are grouped together. This sug-gests that marketing efforts and knowledge bysales agents playa major role in the acceptabil-

. ity of a new product like GRP.

Another measure of acceptability is to exam-ine GRP relative to other crops insured byFCIC. Federal crop insurance was available forsixty crops in 1995. GRP corn ranked number24 when crops are arrayed by total premium.Thus, while GRP insurance has not been widelyaccepted, it does represent a niche market forthe Federal Crop Insurance program. Opportu-nities for further growth will depend on the in-dustry response in developing individual cropinsurance policies that are coupled with theGRP policy. Such wraparound policies are nowbeing developed. Companies can key thesepolicies '0 individual underwriting and reducethe incidence of moral hazard and adverse se-lection. The real advantage of a wraparoundpolicy is that it segments the functions for cropinsurance delivery ina fashion that exploits thecomparative advantage of government versusprivate companies. Government protects thesystemic risk with the GRP, and private insur-ance companies protect the individual risk withthe wraparound. Companies can use the countyyield to gauge which farms have differentyields and the extent to which these differencesare due to moral hazard and/or adverse selec-tion. Huang empirically examined such a con-tract design and concluded that it was workablefor a large percentage of farmers in the Mid-. west. Separating systemic and independent risk

in this fashion offers significant promise formixing government and insurance markets.

Further Needs in Designing GRP

Area yield crop insurance is now in place in theUnited States. The design and rating issues ad-dressed in this study should be subjected to rigor-ous review. There are opportunities to improvethe Group Risk Plan. The contract design can berefined. The use of county boundaries for the areayield needs to be evaluated. County boundariesmay be too small for some regions and too largefor other regions. If significant systemic risks arepresent, it may be practical to increase the size ofthe area (e.g., crop reporting districts). In someregions, the soils and climate may require an areasmaller than the county. In other regions, whereyield risks are independent (i.e., the systemic riskcomponent is missing), area yield crop insuranceshould not be attempted.

Amer. J. Agr. Ecofl.

Previous studies (Miranda; Smith, Chouinard,and Baquet) discuss using higher coverage lev-els or allowing for higher scaling (increasingthe liability). There are significant political andstatistical constraints to these strategies. Offer-ing coverage at or above forecasted yields com-pounds the rate-making and forecasting prob-lems discussed in this study. Offering the op-portunity to scale liability beyond the 150%level for county revenue may be workable. Fu-ture research using farm-level and county datashould investigate the potential benefits of sucha change in contract design. Another contractdesign consideration may involve incorporating"ground-level" truth in estimating the centraltendency in yields. If a radical change occurs inthe production process (e.g., sudden widespreaduse of irrigation or the elimination of irriga-tion), adjustments will be needed.

Rate-making issues that need further devel-opment include (a) the imposition of rate rela-tivities (currently the procedure uses both aparametric and nonparametric structure), (b)the current methods for catastrophic loading(i.e., it may be necessary to use additional in-formation from long series of weather data),and (c) the current methods for weighting con-tiguous counties. Each of these three issuesshould be examined rigorously. While the cur-rent procedures in these three areas have intui-tive appeal, they can be improved upon withfurther research.

Area yield insurance will not work withoutsystemic risk. The greater the systemic risk, themore workable is area yield insurance. Rede-fining GRP boundaries may improve the cur-rent area yield product in some regions.However, this activity may involve signifi-cant transaction costs and may be politicallyunlikely. Of more importance is the opportu-nity provided to private insurance companiesunder Crop Insurance Reform. These compa-nies can package farm-level insurance withthe GRP. Such policies would mitigate thebasis risk and provide a superior model forgovernment involvement in the market forcrop insurance. Farmers would receive moreprotection for systemic risk events while stillhaving farm-level protection for other ran-dom yield losses. Insurance companieswould be protected from catastrophic lossesby the GRP policy and would offer insurancefor events that are more independent.~

[Received May 1995:final revision received March 1997.J

Skees, Black, and Barnett

References

Barnaby, G.A., and J.R. Skees. "Public Policy forCatastrophic Yield Risk: An Alternative CropInsurance Program." Choices, Second Quarter1990, pp. 7-9.

Botts, R.R., and J.N. Boles. "Use of Normal-CurveTheory in Crop Insurance Ratemaking." J. FannEcon. 4O(August 1958):733-40.

Cleveland, W.S. "Robust Locally Weighted Regres-sion and Smoothing Scatter Plots." J. Amer.Statist. Assoc. 74(December 1979):829-36.Visualizing Data. Summit NJ: Hobart Press,

1993.Congressional Commission for the Improvement of

the Federal Crop Insurance Program. Recom-mendations and Findings to Improve the Fed-eral Crop Insurance Program. Washington DC,May 1993.

Goodwin, B.K. "An Empirical Analysis of the De-mand for Multiple-Peril Crop Insurance." Amer.J. Agr. Econ. 75(May 1993):425-34.

Goodwin, B.K., and V.H. Smith. The Economics ofCrop Insurance and Disaster Aid. WashingtonDC: American Enterprise Institute Press, 1995.

Halcrow, H.G. "Actuarial Structures for Crop Insur-ance." J. Fann Econ. 21(May 1949):418-43.

HardIe, W. Applied Nonparametric Regression.Cambridge: Cambridge University Press, 1990.

Hourigan, J.D. "A Farm-Level Analysis of Alterna-tive Crop Insurance Designs: Multiple-PerilVersus Area-Yield." Master's thesis, Dept. Agr.Beon., University of Kentucky, 1993.

Huang, L. "Mixing Government and Markets for anImproved Crop Insurance Program." Master'sthesis, Dept. Agr. Econ., University of Ken-tucky, 1995.

Just, R.E., and L. Calvin. "An Empirical Analysis ofU.S. Participation in Crop Insurance." Econom-ics of Agricultural Crop Insurance: Theory andEvidence. D.L. Hueth and W.H. Furtan, eds.,pp. 205-52. Boston: Kluwer Academic Publish-ers, 1993.

Ker, A., and B. Goodwin. "Rating and Yield Predict-

1? YA YD7 .

Area Yield Crop Insurance Contract 9

ing Procedures for the Group Risk Federal CropInsurance Program." Progress Report delivered toFederal Crop Insurance Corporation, May 1995.

Miranda, M.I. "Area Yield Insurance Reconsidered."Amer. J. Agr. Ecol'l. 73(May 1991):233-42.

Quiggin, I., G. Karagiannis, and 1. Stanton. "CropInsurance and Crop Production: An EmpiricalStudy of Moral Hazard and Adverse Selection."Economics of Agricultural Crop Insurance:Theory and Evidence. D.L. Hueth and W.H.Furtan, cds., pp. 2S3-72. Boston: Kluwer Aca-demic Publishers, 1994.

Skees, I.R. "Background Research and EducationalMaterial for a Pilot Test of the Group Risk Planfor Soybeans." Paper presented to the Board ofthe Federal Crop Insurance Corporation, 17September 1992.

-' "The Political Economy of a Crop InsuranceExperiment." W.I. Myers Lecture Series Paper,Dept. Agr., Resour., and Managerial Econ.,Cornell University, 14 October 1993.

Skees, I.R., I.R. Black, and B.I. Barnett. "Designingand Rating an Area Yield Crop Insurance Con-tract." Paper presented at AAEA annual meet-ing, Indianapolis IN, 6-9 August 1995.

Smith, V.H., H.H. Chouinard, and A.E. Baquet. "Al-most Ideal Area Yield Crop Insurance Con-tracts." Agr. and Resour. Econ. Rev. 23(April1 994):7S-83.

Smith, V.H., and B.K. Goodwin. "Crop Insurance,Moral Hazard, and Agricultural Chemical Use."Dept. Age. Econ. and Econ. Staff Paper No. 94-3, Montana State University, 1994.

U.S. General Accounting Office. Crop Insurance:IflDCcuTate FCIC Price Forecasts Increase Pro-gram Costs. Report to the Honorable 1. RobertKerrcy, U.S. Senate, PEMD-92-4, WashingtonDC, 1991.

-' Crop Insurance: Program Has Not FosteredSignificant Risk Sharing by Insurance Compa-nies. Report to the Chairman, Committee onAgriculture, Nutrition, and Forestry, U.S. Sen-ate, RCED-92-25, Washington DC, 1992.

.- au