|
|
|
 |
Fritz M. Roka and Raymond B. Palmquist Fritz M. Roka is at the Food & Resource Economics Department, University of Florida. Raymond B. Palmquist is at the Department of Economics, North Carolina State University. Hedonic techniques have attracted the interest of economists as a means of measuring values of non-market goods. By studying the market transactions of differentiated products such as automobiles and houses, implied values and corresponding demand schedules can be estimated for underlying characteristics such as automobile safety features, two-car garages, and air quality of residential neighborhoods. The subject of this paper is farmland and estimating the values of characteristics such as soil productivity and erodibility. In 1974, Rosen and Freeman independently developed theoretical models for differentiated consumer products that now serve as the basis for empirical estimates of marginal prices of product characteristics. Procedures to estimate demand schedules of underlying characteristics were outlined in Palmquist (1984). Palmquist (1989) further extended Rosen's theoretical model to consider land as a differentiated factor of production. Several hedonic analyses have focused on farmland values and sought to estimate the values of important land variables. Most of these studies fall in one of two categories. One group of studies estimates the values of soil characteristics such as topsoil depth, pH. erosiveness, and drainage (Ervin and Mills, Gardner and Barrows, Miranowski and Hammes, Palmquist and Danielson). Another collection of work focuses on the effects of urbanization on farmland values (Chicoine; Pardew, Shane, and Yanagida; Shonkwiler and Reynolds). The empirical studies mentioned above have relied on market transactions. For instance, Nagy assessed the impact of surrounding housing developments on the rural land parcels by collecting sales data and descriptions of six at-tributes from over 1,400 Iowa land sales. While such data sets are rich with specific information, the high costs to assemble these data sets limit the geographic boundaries of potential study areas. Public policies concerning the withdrawal of commodity support programs and limiting movement of non-point source pollutants will impact millions of acres throughout the country. The cost of conducting a national hedonic study to estimate the impact of such policies on farmland values with market transaction data could be prohibitive. Alternative land value data sets need to be investigated. One such data set comes from the June Agricultural Survey (JAS). The JAS had been the principle survey instrument to collect the annual production intentions of U.S. crop and livestock producers. Enumerated during June by personal interview, the JAS is a stratified random sample of individual land tracts throughout the country. In 1994, a new section was added to the survey questionnaire. In addition to reporting crop acreage and livestock inventories, farmers and landowners are asked about their knowledge of area land values. Land values are solicited for pasture, woodland, irrigated cropland, and non-irrigated cropland. The important feature of this data set is that the land values reported in the JAS reflect respondents' opinions and not actual market transactions. The purpose of this paper is to examine the use of land value information from the JAS in hedonic analyses of farmland values. The first objective is to estimate a hedonic price function and evaluate the quality of parameter estimates. A second objective is to discuss the feasibility of utilizing the JAS land value to estimate demand schedules for selected soil attributes. The five-state Corn Belt region, including Illinois, Indiana, Iowa, Missouri, and Ohio, serves as a case study. Data Previous studies of farmland values are useful in identifying potential explanatory variables. Soil characteristics that indicate high levels of inherent productivity are expected to positively affect land value (Ervin and Mills, Miranowski and Hammes). In contrast, soil characteristics such as erodibility, wetlands, or excessive water permeability suggest agricultural limitations and therefore exert a negative influence on land values. Investment in technology that negates these limitations has a positive effect on land values (Palmquist and Danielson; Xu, Mittelhammer, and Barkley). Surrounding residential or industrial development competition with agriculture for land is expected to inflate the value of the farmland (Chicoine, Nagy). Several authors have included "size of land tract" as an explanatory variable (Elad, Clifton, and Epperson; Nagy; Palmquist and Danielson; Xu, Mittelhammer, and Barkley). The expected sign on this variable is ambiguous and depends on the particular land market being analyzed. If the potential exists for conversion of farmland into residential or industrial uses, the expense of subdividing may cause smaller tracts to be more valuable per acre. However, in an agricultural land market, value per acre may increase with size of tract as an overall parcel approaches a commercially viable size. Data provided by the JAS include land value estimates, agricultural land use patterns, and descriptive characteristics of the farm enterprise and operators. Respondents are asked to estimate the per acre land values by agricultural land use within the specified tract boundary. Land value estimates are collected for pasture, woodland, irrigated cropland and non-irrigated cropland. The JAS also elicits information about the size of the farming operation, the predominate farm enterprise, and characteristics describing the respondents. The JAS allows for fourteen farm classifications. For this paper, farm enterprise categories are consolidated into four groups-cash grain, livestock, farms enrolled in the Conservation Reserve Program (CRP), and farms that grew specialty crops such as vegetables, tree crops, and ornamentals. Respondents are characterized as sole proprietors, partners, or hired managers. Missing from the JAS are data describing specific soil attributes. While the JAS collects data on planting intentions, expected yields are not recorded. Further, no information is available concerning general topography, presence of wetlands, or other factors that would suggest limitations on agricultural uses. Some of these data are available from the National Resource Inventory (NRI). The latest NRI was administrated by the Natural Resources Conservation Service (the old Soil Conservation Service) in 1992. Field personnel visit sample sites and record general topographical features and soil characteristics. The type of data collected during an NRI survey would be ideal for inclusion in a hedonic price analysis of farmland. Unfortunately, the sampling frame of the 1992 NRI did not correspond to the sampling frame of the JAS. Consequently, very few NRI sample points overlap with JAS sample points. For NRI soil information to be merged with the JAS land values, NRI point data have to be aggregated. The lowest level of aggregation is an NRI polygon that is the intersection of county, major land resource areas, and eight-digit hydrologic unit boundaries. NRI variables are calculated as acreage-weighted averages for the NRI polygon. Each observation in the JAS is placed in an NRI polygon using its respective location identifier. Variables from the NRI data set considered in this study include the percentages of highly erodible land and prime farmland within the NRI polygon. In addition to the JAS and NRI, other data sources provide county level information that may be candidate variables in a hedonic analysis. The 1990 Census of Population and Housing provides estimates of county population densities. Population density serves as a measure of urban pressures. Per acre land values are expected to increase with population density. From the Census of Agriculture, a county com yield variable is constructed by averaging the county per acre yields reported in the 1982, 1987, and 1992 censuses. Corn yields reflect soil productivity, which is hypothesized to be directly correlated with land value. Using crop yields as a measure of soil productivity becomes problematic when comparing soils across regions, particularly when the predominate crops change. For instance, little inference can be made concerning soil productivity when comparing yields from a citrus grove in south Florida to yields from a cornfield in central Iowa. To preserve crop yield as a measure of soil productivity, the study area in this paper is restricted to the USDA Crop Production Region 3 where corn is the predominate crop. Estimating a Hedonic Price Function The JAS land value data set is relatively new. Started in 1994, farmland value data have been collected for three years. A primary objective of this paper is to evaluate the feasibility of using this data in hedonic price analysis. Table I lists the variables considered in the analysis and presents some of their descriptive statistics by survey year. The analysis in this paper focuses on reported values of non-irrigated cropland in the USDA Corn Belt Production Region. The predominate respondent is a sole proprietor of a cash grain operation. Average cropped acreage was 91 acres. On average, over 70% of the land is described as prime farmland and another 25% is considered highly erodible. None of the survey respondents within the Corn Belt region irrigate cropland. Further, respondents from Missouri did not report land values and therefore are excluded from the analysis. Over 6,200 people per survey year from Illinois, Iowa, Indiana, and Ohio responded to land value questions. Reported in table 2 are regression results of four models estimating the 1995 hedonic prices of the variables listed in table 1.The variables are grouped by data sources and added sequentially to the regression model. Since the hedonic price schedule is an equilibrium between suppliers and demands of product attributes, the functional form of a hedonic pricing model must be determined empirically. A limited Box-Cox analysis considered linear, semilog, and double log specifications. Superior function forms have lower residual sums of squares when the dependent variable has been adjusted by multiplying by the inverse of its geometric mean. The residual sum of squares from a semilog specification with the adjusted dependent variable was slightly less than the residual sum of squares from the double-log form and substantially less than the residual sum of squares from the adjusted linear model. Parameter estimates from the semilog functional form are reported in subsequent table 3. Model (1) in table 2 includes only variables from the JAS that are hypothesized to have an effect on the reported land value. Cash grain farms and sole proprietors are included in the intercept term. Overall, this model has little explanatory power. Size of operation (CROPLD) is positive and significant. Significant negative coefficients for alternative enterprises, such as enrollment in the Conservation Reserve Program (SIC3), may reflect land or soil limitations that prevent cash grain farming and consequently, lower expected returns. Models (2) and (3) incorporate additional in-formation from outside the JAS. The proportion of prime farmland within the NRI polygon (PRIMEPC) is included as an explanatory variable in model (2). Explanatory power of the model improves, and the coefficient of PRIMEPC is highly significant and positive. The higher the percentage of prime farmland within the polygon, the greater the probability that the specific land value estimate represents a parcel of prime farmland. The proportion of land that is highly erodible (TOTHELPQ is another NRI variable that was considered in this analysis. Collinearity between PRIMEPC and TOTHELPC was revealed by regression diagnostics (Belsley, Kuh, and Welsch), preventing the joint inclusion of both variables. Model (3) adds county data on population density (PDEN) and corn yield (CTY-YLD). Both variables are significant; they have the expected positive influence on land value and improve explanatory power of the model. In 1994 and 1995, JAS respondents were asked if they had bought or sold land during the previous year. In model (3), the SOLD variable represents 1.4% (88 individuals) who answered affirmatively to recent market experience. The estimated coefficient is not significantly different from zero, suggesting that recent market experience did not influence stated 1995 land values. A similar model was specified for 1994 land values and again, the SOLD variable had no influence in explaining 1994 land values. Model (4) drops variables that categorize farm enterprise, respondent tenure, and market experience. The null hypothesis that alternative enterprises (SIC2, SIC3, and SIC4), partnerships (OPER2), hired managers (OPER.8), and market experience (SOLD) are not jointly significant is not rejected. Nearly all the explanatory power comes from the remaining variables-cropland acreage, percentage of prime farmland, county average corn yields, and county population densities. Implications for Demand Analysis Estimating demand equations for product characteristics such as soil quality or urban population densities requires an accurate specification of the hedonic price function. The fact that the JAS land values are based on opinion rather than market observation brings into question the "accuracy" of the JAS land value data. The usefulness of the JAS data set in hedonic applications depends directly on how well a person's opinion corresponds to current market conditions. The fact that the estimated coefficient on an indicator variable (SOLD) of market experience is not significant, provides some evidence that the land value opinions given in the JAS match market observations. However, two important caveats need to be noted. First, the conclusion that opinions match market reality only holds for those years and regions where such a comparative check can be made. The 1996 version of the JAS dropped all questions regarding recent market transactions. The absence of a check against market transactions weakens any opinion about whether 1996, or subsequent year, land value responses match market conditions in the Corn Belt. The second caveat relates to the fact that the JAS contains imputed data for missing land value observations. In earlier work, Roka and Palmquist found a significant and negative effect from the SOLD indicator variable. This earlier work was based on the raw 1994 observations. With additional survey years, analysts at the Economic Research Service were able to identify apparent outliers and established rules for imputing land values for both the outliers and missing data points. These techniques led to a substantial increase in the number of observations. More work needs to be done to judge the statistical implications of these data adjustments. Another indicator of a model's performance or reasonableness is the consistency of parameter estimates. If parameter estimates switched signs or changed magnitudes appreciably, one could question the consistency with which land value opinions were offered. Table 3 presents some evidence suggesting that implicit values for farm size (CROPLD), soil quality (PRI-MEPC and CTY_YLD), and population density (PDEN) are stable for the land values collected through the JAS. Model (4) from table 2 is replicated with corresponding parameter estimates using 1994 and 1996 land value data. Accurate estimates of marginal prices of land characteristics depend on the accuracy of the model's specification. The best model presented in this paper explained only one-third of the overall land value variability, suggesting that important variables explaining the reported values have been omitted. In its current version, the JAS offers a limited selection of variables that could be used to estimate implied values of underlying land characteristics. Incorporating information from other data bases with different geographic scales (i.e., NRI) is not entirely satisfactory. Market sales for differentiated products such as farmland depend on site-specific conditions. The lack of "tract"-specific information that connects soil quality and neighborhood attributes to reported values diminishes the confidence of estimated hedonic prices and ability to estimate demand equations of farmland attributes. The significance of crop yield as an indicator of soil productivity suggests the potential benefit of including supple-mental JAS questions about expected crop yield. Further, information on soils within the JAS tract sites would improve the potential quality of hedonic price estimation. Recent interest and developments in GIS technology increase the possibility of merging production, environmental, and socioeconomic data to the same geographic scale with the JAS land values. The principal advantage of the JAS data set is that land values are collected in a uniform manner from across the country. Hence, the data set affords the potential opportunity to conduct analysis on the impacts of national policies. The spatial scope of the data set would allow separate land markets to be defined and thereby facilitate demand estimation of underlying attributes. References Belsley, D.A., E. Kuh, and R.E. Welsch. Regression Diagnostics. New York: John Wiley & Sons,1980. Chicoine, D.L. "Farmland Values at the Urban Fringe: An Analysis of Sale Prices." Land Econ. 57(August 1981):353-62. Elad, R.L., I.D. Clifton, and J.E. Epperson. "Hedonic Estimation Applied to the Farmland Market in Georgia." J. Agr. and Appl. Econ. 26(December 1994):351-66. Ervin, D.E., and J.W. Mills. "Agricultural Land Markets and Soil Erosion: Policy Relevance and Conceptual Issues." Amer. J. Agr. Econ. 67(December 1985):938-42. Freeman, A.M. "On Estimating Air Pollution Control Benefits from Land Value Studies." J. Environ. Econ. and Manage. l(May 1974):7483. Gardner, K., and R. Barrows. "The Impact of Soil Investments on Land Prices. " Amer. J. Agr. Econ. 67(December 1985):943-47. Miranowski, J.A., and B.D. Hammes. "Implicit Prices of Soil Characteristics for Farmland in Iowa." Amer. J. Agr. Econ. 66(December 1984):745-49. Nagy, J. "The Residential Impact on Rural Farm-land Price." Unpublished, USDA/ERS/RTD Cooperative Agreement Report, 1995. Palmquist, R.B. "Estimating the Demand for the Characteristics of Housing." Rev. Econ. and Statist. 66(August 1984):394-404. Palmquist, R.B. "Land as a Differentiated Factor of Production: A Hedonic Model and its Implications for Welfare Measurement." Land Econ. 65(February 1989):23-28. Palmquist, R.B., and L.E. Danielson. "A Hedonic Study of the Effects of Erosion Control and Drainage on Farmland Values." Amer. J. Agr. Econ. 71(1989):943-46. Pardew, J.B., R.L. Shane, and J.F Yanagida. "Structural Hedonic Prices of Land Parcels in Transition from Agriculture in a Western Community." W.J. Agr. Econ. ll(July 1986):5157. Roka, EM., and R.B. Palmquist. "Hedonic Farmland Prices and the 1994 June Agricultural Survey." USDA/ERS Cooperative Agreement 43-3AEM-4-80105 Final Report, 1995. Rosen, S. "Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition." J. Polit. Econ. 82(January 1974):34-55. Shonkwiler, J.S., and J.E. Reynolds. "A Note of the Use of Hedonic Price Models in the Analysis of Land Prices at the Urban Fringe." Land Econ. 62 (February 1986):58-63. U.S. Department of Commerce. 1990 Census of Population and Housing. Washington DC: Economics and Statistics Administration, Bureau of Census, October 1993. U.S. Department of Commerce. 1992 (and 1982 and 1987) Census of Agriculture. Washington DC: Economics and Statistics Administration, Bureau of Census, October 1993. Xu, F, R.C. Mittelhammer, and PW. Barkley. "Measuring the Contributions of Site Characteristics to the Value of Agricultural Land." Land Econ. 69(November 1993):356-369. |
|
|
 |
|
|
|
||
){kind=link}
){kind=link}
){kind=link}