Adjusting the value of houses located on a golf course.
Despite the increasing abundance of golf course properties, appraisers have little to guide them in determining the value adjustment for a house that abuts a golf course. A recent study examined the impact of golf-course location on the value of single-family residences.(1) This study, which hypothesized that a golf course will have a positive effect on property value, focused only on houses that abut a golf course. The distinction made between houses on, rather than just some measurable distance from, a golf course was important for several reasons. First, generally the only residents with a view of a golf course are those who have houses adjacent to it. They benefit even if they are not golfers because their view is both natural and unobstructed, as opposed to the restricted suburban property view of a solid fence or a neighbor's backyard. Second, these residents may also associate their golf course view with the desirable feature of low population density. Third, residents of golf course houses may feel a greater amount of privacy because of the open space and the attractiveness of a golf course landscape.
Appraisers can adopt the following procedure to make a value adjustment for such a property and then derive an estimate of the relative magnitude of this adjustment for a sample of golf course properties in a suburb of a metropolitan city.
The frequent use of multiple regression by appraisers suggests that it is an appropriate technique for "unbundling" from the sale price the implicit value attributable to a house's physical characteristics, the market conditions at the time of its sale, and any location-specific factors. Accordingly, an appropriate model to apply to the analysis of the data may be stated as follows:
[Price.sub.i] = [[Beta].sub.0] + [Sigma][[Beta].sub.j][X.sub.ij] + [[Beta].sub.k][T.sub.i] + [[Beta].sub.m][L.sub.im] + [[Beta].sub.g][GOLF.sub.i]
[Price.sub.i] = Selling price of house i as reported at the closing
[Beta] = Estimated regression coefficients
[X.sub.ij] = Set of j physical characteristics of house i
[T.sub.i] = Tune trend variable representing the month house i sold
[L.sub.im] = Matrix of dummy variables indicating the location of the house i in relation to the country club m
[GOLF.sub.i] = Dummy variable indicating whether house i abuts a golf course
The regression coefficients or [Beta]s from the above model represent the impact of each component or factor on the selling price value of a house. The marginal effect on the selling price of a house abutting a golf course is represented by the regression coefficient [[Beta].sub.g]. If the coefficient [[Beta].sub.g] is significantly greater than zero and is statistically significant, then golf course location is held to influence prices positively in this housing market. The magnitude of this coefficient is important because it indicates to the appraiser how much to add to the appraised value of a subject property because it abuts a golf course.
For this study, data on sales prices, property characteristics, date of sale, and golf course location were collected from the local multiple listing service (MLS). Houses in this sample were located on one of three privately owned and operated golf courses. The data represents single-family residential sales in Rancho Bernardo, California, a relatively compact residential suburb of San Diego located approximately 32 miles away from San Diego's central business district (CBD).
To guard against mistakenly attributing a selling price adjustment to a house's golf course location when, in fact, the underlying forces really affecting the adjustment might be other location-specific variables, such as closeness to neighborhood schools, distance from the noise caused by high-speed/high-volume traffic, and proximity to local shopping centers, two control procedures were incorporated into our design. The first control procedure, which was aimed at neutralizing the effect of other location-specific factors was to include in the sample one house not on a golf course but located within the same country club as a match for the house that abuts a golf course. In fact, because this control procedure often resulted in matching houses located across or down the street from each other, implicit housing characteristics, such as quality of construction, reputation of builder, and luxury features were often equalized too. Because there may be explicit differences in the matched houses (e.g., a golf course house may have been larger or resided on a more spacious lot than its matched-pair counterpart), independent variables describing these houses are included in the pricing model.
The second control procedure intended to negate the effect of other location-specific factors on the selling price of a house on the golf course and its matched pair is to code each house according to its location in relation to one of four golf courses in the Rancho Bernardo area. This control procedure, which was implemented by locating all houses in a matrix of dummy variables, should cause the sales prices of houses situated in an exclusive area to be adjusted upward for their location on top of their separate adjustment for golf course location.
Our total sample consists of 314 sales transactions, covering the period February 1990-July 1993. Summary statistics for the total sample, for the 157 houses on a golf course, and for the 157 "control" houses not on a golf course are given in Tables 1, 2, and 3.
The average sales prices for the overall sample and subsamples of golf course homes and homes not on a golf course are $246,362, [TABULAR DATA FOR TABLE 1 OMITTED] [TABULAR DATA FOR TABLE 2 OMITTED] [TABULAR DATA FOR TABLE 3 OMITTED] $250,511, and $242,212, respectively. The age in years and size in square footage of the houses in the overall sample, and for subsamples of the golf course houses and the houses not on golf courses are 7.86 years (1,899 square feet), 7.81 years (1,879 square feet), and 7.91 years (1,919 square feet), respectively. The average property of the total sample and the two subsamples contains 2.4 bathrooms, 3.2 bedrooms, 1.6 stories, and 2.1 parking spaces. The typical lot size of the overall sample, houses on the golf course, and houses not on a golf course is 9,428, 9,507, and 9,349 square feet, respectively. Finally, 87% of the houses on a golf course have central air conditioning, while only 73% of the houses not on a golf course have central air conditioning.
The Rancho Bernardo area has four golf courses. The number of houses in each of these golf course locations is given in Table 4.
Table 5 gives the results of estimating two price equations on the aforementioned data. Both equations employ a dummy variable to distinguish the effect of golf course location. [TABULAR DATA FOR TABLE 4 OMITTED] The difference between Model A and Model B is expressed in the dependent variable. In Model B the dependent variable of the sale price of Model A is expressed in its natural logarithmic form so that a percentage value adjustment for golf course location can be obtained.
TABLE 5 Regression Results for the Dependent Variables of Selling Price (Model A) and the Natural Log of Selling Price (Model B) Variable Model A Model B Constant 31,106 11.5934 (2.115) (221.868) Age in years -711 -0.0032 (-1.420) (-1.827) Number of bathrooms -6,944 -0.0050 (-0.896) (-0.180) Number of bedrooms -3,716 -0.0005 (-1.233) (-0.046) Central air conditioning 10,364 0.0480 (2.741) (3.577) Square footage of the house 106 0.0004 (15.805) (16.160) Size of the lot 3.67 0.00001 (7.823) (7.475) Number of stories -17,645 -0.0823 (-3.378) (-4.436) Number of parking spaces 18,139 0.0446 (3.646) (2.525) Time trend -735 -0.0034 (-6.779) (-8.691) Rancho Bernardo location -4,873 0.01798 (-0.049) (0.505) Bernardo Heights location 12,534 0.6412 (3.281) (4.725) Carmel Mountain location 17,232 0.8105 (3.043) (4.029) Golf course location 12,914 0.0467 (4.523) (4.408) Adjusted [R.sup.2] 0.85 0.86 F-value 134.40 147.93 T-statistics in parentheses.
Given recent studies that have used regression analysis and a knowledge of the local housing market, the results are as expected. About 85% of the variation in the selling prices of houses can be explained by the variation in the independent variables. To determine if we obtained a relatively uniform fit across prices, we divided our sample into five subsets arrayed on price. The following tables present the results of calculating a mean absolute percentage error (MAPE) between the actual and predicted values for each house. These results indicate that problems of scale at the upper and lower price brackets do not exist.
The regression coefficients for living area, lot size, and number of parking spaces are all positive and statistically significant. This indicates that additional units of these variables, all other things being equal, have the effect of increasing the sales price of a house. The same could be said about the positive incremental effect on sales price if a house has central air conditioning.
Mean absolute percentage errors across price ranges Absolute Number of Prediction Errors Price Range Houses Mean Variance $160,000-$195,500 62 0.0890 0.0046 $196,000-$219,000 62 0.0836 0.0048 $220,000-$245,000 63 0.0779 0.0045 $245,000-$283,000 63 0.0748 0.0042 $285,000-$490,000 64 0.0738 0.0031 Total sample 314 0.0798 0.0044 Mean absolute percentage errors across size of house Absolute Size Number of Prediction Errors (square feet) Houses Mean Variance 1,034-1,600 52 0.0601 0.0023 1,601-1,800 106 0.0756 0.0033 1,801-2,000 61 0.0808 0.0063 2,001-2,300 43 0.0980 0.0056 2,301-3,358 52 0.0913 0.0049 Total sample 314 0.0798 0.0044 Mean absolute percentage errors across date of sale Absolute Date of Sale Number of Prediction Errors (months) Houses Mean Variance 0-8 58 0.0819 0.0036 9-17 51 0.0843 0.0058 18-27 64 0.0676 0.0027 28-35 51 0.0858 0.0073 36-42 90 0.0752 0.0035 Total sample 314 0.0798 0.0044
[TABULAR DATA FOR TABLE 6 OMITTED]
The regression coefficients associated with the number of stories and age of a house are negative and statistically significant. They reflect the buyer's perception of the relative value of a multiple-story house in Southern California and the perception that an older house is less desirable than a newer house. Additionally, the negative and significant coefficient of the time trend variable, T, indicates an overall decrease in housing prices during the time period from which the sample sales transactions were drawn. The negative coefficient of this variable is consistent with the general decline in residential values in Southern California during the 1990s.
Lastly, the regression coefficients for the number of bedrooms and bathrooms are not statistically significant. This may, in part, be due to the fact that independent variable values, such as square footage, number of bedrooms, and number of bathrooms, almost always move together (e.g., the larger the square footage, the more bedrooms; and the more bedrooms, the more bathrooms). The study's results are quite robust with respect to the functional form, especially over the golf course location variable. In addition, Table 6, which presents a correlation matrix of the independent variables, substantiates that this data set is not unusual for a study that uses the hedonic pricing equation.
Adjusting for Golf Course Location
Focusing on Model A in Table 5, the GOLF coefficient is positive and significant (i.e., T-statistic = 4.523). This indicates that for this sample, a house abutting a golf course sold for $12,914 more, on average, than a comparable house not on a golf course.
Studying the GOLF coefficient of semilogarithmic Model B perhaps gives a more relevant measure of the magnitude of the adjustment that should be made when appraising a subject property that abuts a golf course. The coefficient of 0.047 for the GOLF variable suggests an upward adjustment of 4.81% to the selling price of a house located on a golf course.(2)
Consider, for example, a subject property located on a golf course within the study's geographic area. A comparable property similar to the subject in every way except that it is located across the street and not on a golf course sold for $250,000. The results indicate that the subject property's appraised value should be adjusted upward by approximately $12,000 ($250,000 x 4.81%), to $262,000 to reflect its golf course location.
The effect of location-dependent factors is important in appraising real estate. For the sample of sales transactions studied, the effect on the selling price of homes that abut a golf course was found to be both positive and statistically significant. Specifically, our data supports an upward adjustment to the selling price of a house because it abuts a golf course.
Some appraisers who work outside of California or similar housing markets may have difficulty taking into account the fact that the large confidence interval obtained (at 90%, it ranges from $7,341 to $16,325) applies to houses in this sample, the average selling price of which was approximately $246,000. Perhaps these appraisers could better relate to mean and confidence interval estimates generated from the log normal regression model, which gives a percentage adjustment for houses located on the golf course instead of a fixed dollar amount. Thus, for a $100,000 house on the golf course, the log normal regression predicts a value increase of $4,810 (4.81% x $100,000) because of its location, with a 90% confidence interval ranging from $3,010 to $6,611.
Although the actual magnitude of the adjustment may be unique to a housing market, this study has outlined a multiple regression, matched-pair-based procedure that can be used by an appraiser to derive the local impact on prices because of a house's golf course location. Once this adjustment is derived for a geographic area, an appraiser can factor it into the sales comparison valuation approach to adjust the value of a subject property abutting a golf course to comparable properties not on a golf course.
1. A Quang Do and Gary Grudnitski, "Golf Courses and Residential House Prices. An Empirical Examination," The Journal of Real Estate Finance and Economics (June 1994) 100-109.
2. The interpretation of a dummy variable coefficient when the dependent variable is in log-form is described in Peter E. Kennedy, "Estimation with Correctly Interpreted Dummy Variables in Semilogarithmic Equations," American Economic Review (1991): 801, and can be calculated as: Percentage change = EXP(coefficient) - 1, or 4.81% = EXP(0.047) - 1.
Gary Grudnitski, PhD, is professor of accountancy at San Diego State University. He received a PhD from the University of Massachusetts, Amherst, and has written books as well as numerous articles for academic and professional journals.
A. Quang Do, PhD, is professor of finance at San Diego State University. He received a PhD, an MS, and an MBA from Louisiana State University, Baton Rouge. He conducts research on real estate and teaches courses in real property valuation and has had articles published in numerous real estate journals.