A Hedonic Pricing Model for Large Parcels on the Urban Fringe in Chester County, Pennsylvania

Transcription

1 A Hedonic Pricing Model for Large Parcels on the Urban Fringe in Chester County, Pennsylvania DERALD J. HAY, ESQUIRE Fox Rothschild, LLP, 747 Constitution Drive; Exton, PA 19341; Abstract This article examines the use of the hedonic model in developing a method of examining parcel larger than five acres in Chester County, Pennsylvania. Many variations on methods for estimating the value of out-of-sample properties have been developed through the years. This article examines the attributes of parcels on the urban fringe which have both urbanizing and rural influences which play an important role in a parcel s market value. In Chester County, many of the agricultural areas are being converted to residential or commercial development. The anticipated development can increase the value of a large parcel in ways that make the more traditional methods of estimating the value of farmland inaccurate. Likewise, the value of a property for development is in its size, location, and potential uses rather than the attributes of a the existing structures on the property; thus, the traditional methods for estimating residential real estate parcels can be inaccurate. This study attempts use the hedonic method to develop a model to explain the values of large parcels in a market with a wide variation of values and influences that can affect the market value of a parcel. Key Words: Hedonic models, urban fringe, property prices. Introduction The market value of real property is only known upon the sale of that particular parcel. Therefore, any attempt to determine the market value of a property before or even shortly after a sale would be to estimate. The process of estimating the value of a property is price discovery. (Geltner et al. 2003). There are three primary methods used to develop price discovery models. The three methods are hedonic pricing, repeat sales, and assessed value. (Meese & Wallace 1997). Although all three methods are distinctly different, each uses some form of regression analysis. (Birch & Sunderman 2003). Hedonic pricing was first introduced in (Birch & Sunderman 2003). The basic theory of hedonic pricing is that the market value of a property can be expressed as a function of the value of a property s attributes in the aggregate. The attributes in the hedonic method usually include square footage, number of bedrooms and bathrooms, neighborhood characteristics and location, age of house, and many others. The equation for hedonic pricing is: (1) 1

2 where P it is the transaction price of a property; S it and L it are the property s attributes; b is the coefficient for the respective characteristic; Q is a variable depicting the time period of the sale; and c is the time period s coefficient. (Clapp & Giaccotto 1992). There are a number of advantages of using the hedonic pricing method as compared to other methods. One advantage is the ability to standardize pricing to allow a researcher to create a constant-quality house price index. (Haurin & Hendershott 1991). This is useful because it allows for the side-by-side comparison of out-of-sample properties. For instance, hedonic pricing is helpful when considering that an appraiser would probably reject a sale of a 4,000 square foot house as a useful comparison when valuing a 1,500 square foot house, even if the sales were quite close in space and time. Hedonic pricing captures most transactions because it only requires the sales price and attributes of the property; thus hedonic pricing can act as a fairly reliable gauge of market price even with a limited sample size. (Schwann 1998). While there are advantages to hedonic pricing, there are also difficulties associated with using the model. The most critical disadvantage is the sheer amount of data that must be collected to develop a reliable model. (Clapp & Giaccotto 1992). The level of detail required on each transaction can be prohibitive. There is also ignorance of both the functional form of the relation and of the appropriate set of house characteristics to include in the analysis. (Meese & Wallace 1997). Each variable also contains its own source of error, thus increasing the overall error within the analysis. (Clapp & Giaccotto 1992). Finally, there remains the heroic assumption that all omitted variables are orthogonal to those included. (Meese & Wallace 1997). Meaning that the research assumes that all of the included variables explain the influences that the excluded variables had on the transaction. On the other hand, including excessive variables can clutter the resulting equation without adding much predictive power and lead to correlation between the regressors. (Hushak 1975). Therefore, a balance must be struck between inclusion and exclusion of variables. Recognizing the disadvantages that were present with hedonic pricing, another method was introduced that reduced the amount of information that needed to be collected about each transaction. The repeat sales method was first introduced in (Baily et al. 1963). As the name would suggest, the repeat sales method compares prices of a property resale with an earlier sale, where both sales happened in the same period. There can be no significant changes to the attributes of the property during the period between the sales since that would not result in a direct comparison of the same product. (Meese & Wallace 1997). There are two underlying assumptions that must be satisfied before the repeat sales method can be applied. The first assumption is that the sample of homes is representative of all the homes sold during that sample period. (Meese & Wallace 1997). The second assumption is that the price of attributes remains constant over time. The basic equation for the repeat sales method uses a ratio of a subsequent sale price of a property to an initial sale price of that same property: (2) 2

3 where R itt' is the ratio of the subsequent sale price in time t' to the initial sale price in time t; B is the sale price of a house; and U represents the residuals, in log form, which are assumed to sum to zero. (Baily et al. 1967). Then, to estimate the market price of a property, the B can be treated as a regression problem: (3) Despite reducing the level of standard error in estimating real estate prices when there is a large variation of the quality and types of attribute, the repeat sales method also has disadvantages. (Baily et al. 1967). For instance, it is difficult to remove and adjust for properties that have had renovations or changed characteristics between sales. Additionally, the sample selection process is biased because there are very few properties that are resold in the same time period. (Clapp & Giaccotto 1992). The bias presented in the sample of houses in the repeat sales method is problematic for two reasons. The first reason is that this method can only be applied on a large-scale; otherwise, the sample would be too small to gather reliable results. The second reason is the tendency for the sample to be unrepresentative of the overall population of houses because properties that are sold many times within a short period of time are usually a starter home, a lemon, or a fixer-upper. The latest method to be introduced to estimate prices for real estate is the assessed value method. (Clapp & Giaccotto 1992). In the assessed value method, the market value of a property is estimated by using linear regression based on its assessed value. While it is recognized that an assessed value is not the true value of a parcel, the assessment is a professional s estimate of the most likely transaction price using knowledge of buyer and seller behavior. (Geltner 1991). The assessed value is modeled as: where P it is the price of the i th property at time t; c is the level at which the assessor is required by law to assess the property; and Q is the corresponding time period of the actual sale, the purpose of which is to capture temporal influences on sale prices; and A i0 is the modeled assessment below: where A is the assessed value a time i; V is the true market value of the property; and z is a random error term with mean zero. Under the assessed value method, all of the variables of the hedonic method are theoretically compiled into one statistic the assessment. Assessment values are contained in a database compiled by the state and are available to the public. The data will represent every transaction because every property is assessed for property tax purposes. (4) (5) 3

4 There are some disadvantages to using the assessed value method. First, there is likely to be a bias in the assessment ratio in underestimating market values. (Clapp & Giaccotto 1992). This is because property owners whose properties are assessed too high will appeal the assessment in order to reduce the tax levied on the owner based on the assessment, but the opposite is not true. Another disadvantage is that properties sold at the beginning of a period may be assessed higher or lower than properties sold at the end of a period due to real or perceived market fluctuations. (Ihlanfeldt 2004). It is an assumption of the model that the assessment methods are uniform throughout the entire assessment time period. The Urban Fringe For large parcels, the contribution a structure can add to the value of the property can vary significantly. On the urban fringe, this variety is magnified due to the constant changes of use of a property. For instance, if the property is used primarily for agricultural purposes and only has a small residence with a barn on a hundred-acre parcel, the value of the structure compared to the value of the land is much less than a residence on a quarter-acre lot. Therefore, when creating a model for large parcels, it is important to examine factors beyond the structures on the properties. Recognizing the fact that other factors are the driving force behind the value of large parcels, hedonic models were created to determine the value of a rural property primarily used for agriculture. These models focused on the size and soil characteristics of the farm. The size of the farm is important because it relates to the total surface area that can be used for crop production. Soil fertility is an important characteristic of agricultural land because the productivity of the soils translates into the total crop yield that the property is capable of sustaining. Both of these factors directly influence the predicted rents that can be derived from the property. The total rents include all anticipated future profits from the use of the property, including the prospective value of developing the parcel. Therefore, using soil characteristics, as a primary independent variable, in a hedonic model is problematic when there are other external forces influencing the value of a property. Few empirical analyses, however, have concentrated on land values in urban fringe areas, where agricultural and urban forces interact and complicate the estimation and understanding of values. (Chicoine 1981). Therefore, other empirical data need to be included in a hedonic price model that appropriately accounts for external factors beyond the soil characteristics when there is a possibility that a purchaser may seek to develop the parcel because of the possibility of emigration from urban centers to the rural areas along the fringe of these urban centers. The rural-urban fringe is the zone of transition in land use, social and demographic characteristics lying between (a) the continuously built up urban and suburban areas of the central city, and (b) the rural hinterland. (Pryor 1968). Pryor also noted that the land use in the urban fringe is distinctively intermingled and transitional between residential, agriculture, and commercial uses. Land values in the urban fringe increase with the anticipation of urbanization. Many of the urban fringe characteristics noted by Pryor accurately describe Chester County, Pennsylvania. Chester County is located in relatively close proximity to 4

5 Philadelphia, Pennsylvania and Wilmington, Delaware. It is also close to King of Prussia, a corporate center in Montgomery County. However, to the west of Chester County is Lancaster County and to the north is Berks County. Both of these counties major economic activity is agriculture. That would make Chester County the transition zone between the city and rural cultures. While much of the county has been subdivided into residential developments there are still many areas in the county that are devoted to agriculture in some form or another. Many of the agricultural properties closer to population centers are popular for use as horse farms while agricultural properties in the western part of the county are usually used for diary or crop production. Many times, farms are surrounded by residential or commercial developments. The surrounding land matrix has an influence on the value of the farm. However, in the same respects, the presence of a farm will have an influence on the values of the residential developments surrounding it. As shown in many studies, agriculture can have an amenity or disamenity impact on the values of nearby homes. (Ready & Abdalla 2005). In the studies that have focused on rural land prices on the urban fringe, several different independent variables have been used in hedonic models. The two most frequently used predictors are the size of the parcel and the parcel s radial distance to a large urbanized area. This is not surprising because size clearly has an influence on the eventual price because of the larger the parcel, the more surface area to farm or more lots that can be developed. (Colwell & Sirmans 1978). As for the distance to urban centers, it has been widely noted that parcels have higher values closer to labor centers. (Shi et al. 1997). Another important characteristic that has been used by many of the models is the distance to the nearest highway. The distance to a highway controls the access to a parcel. If a parcel is not accessible, then it will be less likely to be sold at a premium price. The timing of the sale was used by several studies. Studies that did not use a variable to account for the temporal difference from sales used sales data over an insignificantly short time period. Table 1: Variables Used in the Literature Estimating Parcel Value at the Urban Fringe Models Based on Characteristics Including Urbanization (Shi et al. 1997). Variable Used in Model Other Type of Variable Population Density Time Size of Parcel Land Values Zoning District Distance to Highway Distance To City Buyer/Seller Type Article and Author Chicoine 1981 X X X X X X Clonts 1970 X X X X X Colyer 1978 X X X Dunford et al X X X X X X Folland & Hough 1991 X X X X Hushak 1975 X X X X Hushak & Sadr 1979 X X X X Shonkwiler & X X X X 5

6 Reynolds 1986 Shi et al X X Frequency The Data For this study, five acres is the minimum threshold sample property. In Chester County, there were 11,826 parcels greater than 5 acres. Between January 1, 2001 and December 31, 2006 there were 2,804 parcels greater than 5 acres which were sold. Many transactions involve the transfer of property from one party to another for a nominal value; these transactions are not reflective of the market. Usually these transactions involve a transfer between family members, a transfer from an estate to a beneficiary, a transfer of a property at a tax sale where significant tax liability attaches and runs with the property, or a transfer of a person to the business operated by the previous owner. Since these transfers do not involve a transaction where the selling price reflects the market price, these transactions were excluded from the data set. The hedonic model expresses the market price of a parcel as a function of the values of its characteristics in the aggregate. Accordingly, data on the attributes of properties in the market are needed to create the hedonic model. The following categorical and continuous variables were included in the model: Categorical Variables: Zoning Characteristics Clean & Green School District Presence of Structures Continuous Variables: Acres per Parcel Distance to City Centers Distance to Nearest Highway Sales Period Previous studies have used the current owner of a parcel as a predictor of the price that was paid to purchase the property. However, the hedonic method assumes an efficient market. For a market to be considered efficient, buyer and seller characteristics should be irrelevant. An efficient market assumes equal bargaining power between the prospective sellers and purchasers. The market price of a property, in the hedonic method, is expressed as a function of its attributes. Including buyer or seller characteristics imputes on the property a characteristic that is not tied to the property itself. Dunford et al. and Chicoine both used a binary indicator variable to indicate whether the buyer was an individual person or non-person. The authors theorized that this variable can be important for several reasons. Individuals are more likely to purchase a property for personal use than for purely investment purposes. If an individual purchased a property for personal use, it is possible the property will be used as a domiciliary. It is well known that the (present) value of rural land is dependent upon anticipated future net benefits appropriately discounted. (Dunford et al. 1985). Therefore, individual landowners will anticipate fewer net benefits from the property than businesses, which will generate yearly income from the use of the property. It was found that partnerships and corporations paid more than individuals for similar properties. 6

7 Dunford hypothesized that individuals likely paid less because businesses are more capable of more intensive land uses that yield higher rents. Additionally, he noted that businesses have more access to a broader financial base. In Chester County, each municipality has zoned every parcel in such a way that permits some uses of the property but prohibits other uses. The zoning of a property can limit the permissible uses that the landowner may engage in, such as limiting uses to only agriculture, residential, or commercial activities. These limitations have been shown to influence the value of a property. (Shi et al. 1997). In Pennsylvania, the authority to zone lays with the municipality and each municipality is has its own method of zoning. In order to avoid having an excessive number of categories, five major zoning categories were defined: residential, rural, agriculture, commercial, and performance based zoning. Many properties are enrolled in Pennsylvania s Clean and Green Program. 72 P.S et seq. The Clean and Green program is a method of reducing the property taxes of a landowner by assessing a property s use value as an agricultural production unit rather than the maximum market value, the property s developmental value. The program provides a reduced real estate tax assessment for a property so long as the property is not developed. If the property is developed, penalties are assessed to the property owner. This may affect the market value of the property. Information about a property s enrollment in Clean and Green is readily available public information. An indicator variable was used to note if a property was enrolled in the program. There are several major highways in Chester County. Several authors found that distance to the nearest highway had an influence on the market value of large parcels. The authors explained that the influence was because accessibility is important aspect of a parcel s value. (Hushak 1975). The major highways in Chester County were defined as any federal interstate highway or first class state highway with two or more travel lanes of traffic. The distance to each highway was measured as a radial distance from the center of the parcel. The radial distance to a highway does not present a completely accurate portrayal of the accessibility of the parcel since roads are rarely accessible via the shortest distance possible. Access to highways is controlled by the location of onramps and the path that the tributary roads follow to connect to the on-ramps. Even with the closer access points, tributary roads rarely follow direct paths to these access points. Despite these weaknesses, other studies have successfully used the radial distance measurement. There are fourteen school districts in Chester County. School districts affect a parcel s market value because many home buyers prefer properties in a school district with a good reputation. In Chester County, there is a noticeable disparity between the quality of the school districts. This difference may be reflected in the ultimate market value of a home. The disparity may be the result of funding issues, with the wealthier school districts able to afford more expensive facilities to aid teaching. Additionally, property tax rates vary between school districts. Higher taxes may result in lower sale prices. The location of parcels plays an extremely important role in determining the demand for a parcel, which will influence the market value of a parcel. In order to capture the variation in market price due to the location of a parcel, the radial distance from a parcel to the two major metropolitan centers and two large towns was measured. The major metropolitan city near Chester County is Philadelphia. King of Prussia is also 7

8 a commercial center near Chester County which may provide an influence on the property value as it is a major labor center which commuters travel to on a daily basis. The radial distance from the center of the parcel and the geographic center of each city was calculated. Some parcels within the sample include improvements to the property such as houses, buildings, barns, or sheds. Information regarding the type, size, and quality of these improvements was not included in the county data set. However, in assessing a property, the tax assessment office breaks the total assessment into two values- the value of the land and the value of the structures. In order to capture the additional value that a structure contributes to the eventual sale price, the value of structures will be accounted for by using a dummy variable to indicate whether or not there was a structural assessment for the property or not. While this dummy variable lacks the capacity to capture the nuances of the difference between structures, it also avoids adding bias into the equation that would result from using the assessment information for structures. Empirical Results Of all the zoning districts, only two coefficients were significant in the model. The commercial/industrial zone and the R-4 high density zone were the only zoning variables with a positive coefficient. It was anticipated that the commercial zone would have the highest coefficient because there are broader uses that are permissible in the commercial and industrial zones. These parcels can be used to generate income, thus increasing any anticipated rents that can be generated from the parcel. Additionally, the high density residential zones positive coefficient can be explained by the increased rents that could be anticipated through the more intense development that is permitted in the zone. The remaining zoning districts have a negative coefficient. This means that the properties in the rural zone, the intercept, have a higher value than the properties in residential zones. This result was not anticipated. It was thought that parcels in residential zones would have higher market values because of the possibility of subdividing large parcels and developing residential neighborhoods. One reason for this result is a wide variability of the permissible uses among the residential zones. The most obvious difference between the residential zoning districts is the permissible housing densities. The low-density residential zones, like the R-1 and R-2 zones may actually have similar permissible uses as the rural zone. However, the higher density zones, like the R-4 zone, permit more intense development. Table 2: Hedonic Model Predictor Coef SE Coef T P Constant ln Acres per Parcel Sales Period Structures Present Clean & Green Low Price School District Medium Price School District Intercept High Price School District Distance to Philadelphia (km) Distance to King of Prussia (km) Distance to Highway (km)

9 Rural Zone Intercept Agricultural Zone Commercial/Industrial Zone Conservation Zone Performance Bases Zoning Residential Agricultural R-1 Residential Zone R-2 Residential Zone R-3 Residential Zone R-4 Residential Zone S = R-Sq = 33.0% R-Sq(adj) = 32.0% Interestingly, a few key variables did not result with a significant coefficient. For instance, the distance to the highway yielded a p-value of A reason for this result is that the model may have classified too many roads as highways. If only a few key highways were selected, a more significant result may have occurred. Another reason why the result may not have been significant is because of the dual influence that a being near a major highway has to the value of a house. While being close to a highway increases accessibility for the parcel; nearby highways increase traffic, noise, and detracts from views around residential parcels. However, being located on a highway may be a benefit for commercial parcels because it increases the visibility of the business. Increased visibility equates to an opportunity for more potential customers to notice the businesses along the highway. In order to capture this possibility, an interaction term between the distance to the highways and in a commercial or industrial zoning district was included in the analysis. The interaction term for parcels came out negative and significant. This result suggests that there is indeed an interaction going on between the specific land use of the parcel and the distance to a highway. The negative coefficient means that the value of a parcel is reduced as the distance between the parcel and a highway increases. The literature did not suggest a method to evaluate the dual impact that nearby highways have on residential properties the benefit of faster access and the impairment of the nuisances. One possibility would have been to measure the distance only to major artery routes, like route 202 and the Route 30 bypass, as a measure of accessibility because are the two most traveled routes in Chester County. (Chicoine 1981). As a measure of the negative influence being near a highway, a dummy variable could be include for properties that are within a specified distance of a highway. Chicoine did something similar to this in measuring the negative influence of being near a railroad for properties in his study. The coefficient associated with properties enrolled in a preferential assessment program was also unanticipated. The clean and green variable had a positive coefficient. It was anticipated that this encumbrance would be a burden on a property and reduce the eventual market value of the property. However, the positive coefficient suggests that when a property is enrolled in preferential assessment program, the value of the property increases. One reason for a positive coefficient is that enrollment in the Clean and Green program does not act as a disincentive for a landowner to develop his parcel. If the program does not create a disincentive for development, the price that a purchaser pays for the parcel would include the developmental value of the parcel. In all likelihood, the parcels that are enrolled in Clean and Green are surrounded by other properties that have 9

10 considerable open space. Surrounding open space has been considered an amenity to a parcel s market value. (Ready & Abdalla 2005). This would further explain the reason for a positive coefficient. The size of a parcel has a negative coefficient, conforming to the theory that as parcel size increases, the price paid per acre decreases. The negative coefficient was expected because the data were standardized on a per acre basis. If the analysis commenced on the sale price (as opposed to price per acre) versus the number of acres of the parcels, the coefficient for the size of the parcel would have been positive. The coefficient would have been positive because as the parcel size increases, the total price paid for the parcel increases. However, that price increases at a diminishing rate as the parcel size increases because each additional acre adds less value to the property than the previous acre. This is the economic theory of the rate of diminishing returns. The distance to Philadelphia has a negative coefficient, which indicates that as the distance between a parcel and the center of Philadelphia increases, the value of the property is negatively influenced. This conforms to the other studies that showed that as the distance to the labor centers increased, the relative value of parcels decreased. A negative coefficient is expected because as the distance between a parcel and a labor center increases, the property owner will have to commute a farther distance to get to that labor center for work or to purchase services not available closer to the parcel. There are also fewer businesses located as the radial distance from a labor center increases. With fewer businesses, there are fewer job opportunities and fewer services that the landowner can use. Travel distances generally increase with fewer businesses because grocery stores, other retail centers, churches, and restaurants are dispersed more sparsely. The increased travel distance translates to reducing the appeal a property has to purchasers, thus decreasing the market value of the parcel. Interestingly, the coefficient associated with the distance to King of Prussia was slightly positive. This is unexpected because, along with Philadelphia, King of Prussia is a labor market that Chester County services. More than 11% of the county s residents travel there for work. Additionally, King of Prussia has attractions for residents such as the King of Prussia Mall and other retail centers. It is not clear why King of Prussia would have a positive coefficient. One reason that might explain why King of Prussia had a slightly positive coefficient is because of the influence that the distance to King of Prussia is inversely related to other labor markets. For instance, West Chester, the county seat, is located southwest of King of Prussia. Malvern, the location of the Great Valley Corporate Center is located roughly halfway between King of Prussia and West Chester. Also a significant portion of the population from Chester County, about 9%, works in Delaware County or Wilmington, which are both south of King of Prussia. Because those locations were not included in the analysis, it is possible that the measurement to King of Prussia is not only explaining the distance to King of Prussia, but also the distance from another location. One method that could have been used to isolate the relationships labor centers have to one another collectively on a parcel is to calculate the mean distance to labor centers. This statistic can be calculated by weighing the distance to each labor center by the number of jobs that the location supplies that the sample population and then computing the mean weighted distance to each location. However, using this statistic will not explain the influence each labor center has in the hedonic equation. 10

11 Summary and Conclusion Deducing trends in the market price of a unique product has an inherent amount of variability. No sale of real estate is identical to another sale, even when the same parcel is sold multiple times. Each parcel and transaction has unique features. Each feature contributes to the value of the parcel in a different fashion in each transaction. Parcels are sold for very low prices in some transactions where there is a relationship between the buyer and seller. There is no model that can capture and explain the transactions where a relationship between the parties influences that transaction. Outlier analysis can help to ferret out those transactions where a personal relationship is likely to have influenced the transaction. However, transactions happen along a spectrum and at some point the bright line between a nominal sale and an arms-length transaction gets blurred. The previous studies have not offered an explanation of how anomalous transactions were excluded from the sample. Further analysis and explanation would be helpful to develop a mechanism that can help create a repeatable method of removing anomalous transactions without introducing too much bias into the sample by indiscriminately removing transactions that seem out of place given the other data. There are also other influences on the transaction that cannot be captured by any model. For instance, when a property is sold with tenants leasing a building, those leases add value to the parcel. However, the value of those leases is unknown without the disclosure of the terms of the leases. Other times, a landowner may be the last holdout parcel of a larger developmental plan that involves adjacent parcels. Being the last piece of a developer s multi-parcel plan escalates the value of the parcel, but that hold out is not a characteristic bound to the land. Rather being a holdout parcel is a characteristic imputed upon parcel by a unique situation. A model that examines the physical features of the parcel in a transaction cannot capture this type of situation and explain the unusually high price paid for that type of a parcel when other similar parcels are selling for much less than that last holdout parcel. While models can still help to indicate general trends for market values of real estate in a labor market on a macro-scale, bringing a macro-scale equation into microscale estimation can be problematic. The models have a high standard error and large prediction intervals. The estimate of an individual property s value from a model may not be helpful because of the large prediction interval. The correlations from the models in this study were lower than the correlations provided in the literature. Most studies reported in the literature had correlations around 80% while the highest correlation from this study was less than 40%. In general, one reason why the results were not as correlated as previous research could be that many nominal sales were retained in the analysis at the risk of removing valid observations. Another reason for the lower correlation could be that there were variables that were not included in the model developed here even though they would significantly explain the variability in the sale price of a parcel. One noteworthy area of concern was the distance to the nearest highway. This variable lacked significant predictive power in the hedonic model while other researchers have found this variable useful. The lack of significance is likely the result of using too many highways to measure the distance to the parcel and an inability of the measurement 11

12 to actually account for access to the highway. Although this paper followed a similar approach in designating highways as the other researchers, it is possible that the unique nature of Chester County as a hub of many travel corridors should have required a more restrained and moderate definition of a highway. References Bailey, Martin, Richard Muth, and Hugh Nourse. (1963). A Regression Method for Real Estate Price Index Construction, Journal of the American Statistical Association 58, Birch, John, and Mark Sunderman. (2003) Estimating Price Paths for Residential Real Estate, The Journal of Real Estate Research 25(3), Case, Bradford., John Clapp, Robin Dubin, et al. (2004). Modeling Spatial and Temporal House Price Patterns: A Comparison of Four Models, Journal of Real Estate Finance and Economics 29(2), Chicoine, David. (1981). Farmland Values at the Urban Fringe: An Analysis of Sales Prices. Land Economics 57(4), Clonts, Jr., Howard. (1970). Influence of Urbanization of Land Values at the Urban Periphery, Land Economics 46(4), Colyer, Dale. (1978). Socio-Economic Determinants of Rural Land Values at Greenbrier County, West Virginia. Journal of the Northeastern Agricultural Economics Council 7, Colwell, Peter and C. Sirmans. (1978). Area, Time, and the Value of Urban Land. Land Economics 54(4), 514. Dunford, Richard, Carole Marti, and Ronald Mittelhammer. (1985). A Case Study of Rural Land Prices at the Urban Fringe Including Subjective Buyer Expectations, Land Economics 61(1), Folland, Sherman and Robbin Hough. (1991). Nuclear Power Plants and the Value of Agricultural Land. Land Economics 67(1), Geltner, David. (1991). Smoothing in Appraisal Based Returns, Journal of Real Estate Finance and Economics 4(3), Geltner, David, Bryan MacGregor and Gregory Schwann. (2003). Appraisal Smoothing and Price Discovery in Real Estate Markets, Urban Studies 40(5-6), Hushak, Leroy. (1975). The Urban Demand for Urban-Rural Fringe Land, Land Economics 51(2), Hushak, Leroy, and Kazem Sadr. (1979). A Spatial Model of Land Market Behavior. American Journal of Agricultural Economics 61(4), Haurin, Donald and Patric Hendershott. (1991). House Price Indexes: Issues and Results. AREUEA Journal 19(3), Ihlanfeldt, Keith R. (2004). The Use of an Econometric Model for Estimating Aggregate Levels of Property Tax Assessment Within Local Jurisdictions. National Tax Journal 57(1),

13 Meese, Richard and Nancy E. Wallace. (1997). The Construction of Residential House Price Indices: A Comparison of Repeat-Sales, Hedonic Regression, and Hybrid Approaches. Journal of Real Estate Finance and Economics 14(1), Pryor, Rachel. (1968). Defining the Rural-Urban Fringe, Social Forces 47(2), Ready, Richard, and Charles Abdalla. (2005). The Amenity and Disamenity Impacts of Agriculture: Estimates from a Hedonic Pricing Model, American J. of Agricultural Economics 87(2), Schwann, Gregory. (1998). A Real Estate Price Index for Thin Market, Journal of Real Estate Finance & Economics 16(3), Shi, Yue Jin, Timothy Phillips, and Dale Colyer. (1997). Agricultural Land Values under Urbanizing Influences, Land Economics 73(1), Shonkwiler, J. S. and J. Reynolds. (1986). A Note on the Use of Hedonic Price Models in the Analysis of Land Prices at the Urban Fringe, Land Economics 62(1),