Evaluating the risk of housing investment.
This article evaluates the patterns of housing risk and returns in Greensboro, NC, and Houston, TX. It shows the price risk facing the average homeowner is high and varies substantially among cities. The return on an individual housing transaction is positively associated with national returns, but the association is not strong, which indicates a high level of nonsystematic risk in housing transactions. The probability of a loss on sale in the average housing transaction is found to be large and inversely related to national housing returns and to local employment growth. It also is strongly influenced by location and other housing features.
Households in the United States have long tended to concentrate their wealth in housing. In 2001, 68% of U.S. households were homeowners, according to the Federal Reserve's Survey of Consumer Finances. (1) The net worth of the median homeowner household was $171,700. The value of the median household's primary residence was $123,000, and its housing equity was $53,000, or about 31% of its total net wealth.
The risk confronting households holding large amounts of undiversified real estate has been recognized by Case, Shiller, and Weiss, and by Caplin et a1. (2) Case, Shiller, and Weiss propose the establishment of cash-settled real estate futures and options markets to allow better hedging and diversification. Caplin et al. urge the creation of partnership markets for housing equity. In December 2004, the Chicago Mercantile Exchange announced it was proceeding to offer futures contracts based on the cost of housing in different major cities in the United States. (3)
This article focuses on assessing the risk of housing investment. Although most homeowners consider housing to be a consumption good, it is clear that housing also represents an investment. Although imputed rent is consumed when a homeowner occupies a house, this benefit is excluded from the risk and return analysis in this study.
When judged on the basis of movements in aggregate price indexes, housing appears to be a relatively low-risk investment. For example, the Office of Federal Housing Enterprise Oversight (OFHEO) produces a series of aggregate price indexes for existing housing that are often used to assess current housing trends. Drawing on annual data from the national OFHEO House Price Index for 1975-2003 yields the series of capital gain returns shown in Table 1 for one-, three-, five-, seven-, and ten-year holding periods.
Table 1 shows that between 1975 and 2003 there were 28 one-year holding periods (n). The average one-year capital gain return to housing investment (ignoring transaction costs) was 5.7% with a standard deviation of 3.2% and a coefficient of variation (CV) of 0.52. (4) The risks and returns for housing investment decline with longer holding periods, as shown in Table 1. (5)
When compared to common stocks over the same periods, housing investment appears to have substantially lower risk. Comparing Tables 1 and 2, housing investment has lower risks and returns for almost all corresponding holding periods shown, the exceptions being the CVs for housing with 7- and 10-year holding periods, which are higher than the equivalent CVs for stocks.
The risks and returns shown in Tables 1 and 2 are associated with national portfolios of houses and stocks. For stocks, this is appropriate because stock investors can purchase shares in funds that hold such portfolios; however, for housing investors, purchase of shares in a national housing portfolio is not possible. Most housing investors are able to purchase a share only in the house they occupy. This means that housing investors have no way of diversifying away the nonsystematic risk of housing investment.
To evaluate the patterns of housing risk and returns, this study examines samples of repeat-sale, single-family homes in Greensboro, NC (1975-2003), and in Houston, TX (1989-2004). The first section of this article reviews the literature on housing risks and returns, and the second section examines the pattern of housing risks and returns. The third section estimates a series of housing return models to explore the determinants of returns. The fourth section presents a probit model to examine the factors that influence the probability of loss, and the final section provides a summary and evaluation of the results.
A number of past studies have examined the risks and returns of housing market investment, including papers by Coyne, Goulet, and Picconi; Alberts and Kerr; Hendershott and Hu; Peiser and Smith; Webb and Rubens; Miller and Sklarz; Case and Shiller; Ermer, Cassidy, and Sullivan; Harris; and Jud and Winkler. (6) Most of these studies have assessed the risks and returns of housing investment using aggregate indexes of market prices.
Case and Shiller relate the price risk of an individual house to the movements in a metropolitan statistical area (MSA) price index. (7) They show that a relatively low percentage of the variation in the price of an individual house can be accounted for by the variation in the MSA-wide price index.
Goetzmann investigates the risk and return in an investment portfolio using the weighted repeat sales to produce a quality-adjusted housing price index. (8) He finds that investing in an individual house has about twice the risk of a well-diversified portfolio (investing in a regional or inter-regional portfolio).
Englund, Hwang, and Quigley study the investment implications of housing choices using housing prices in Stockholm. (9) Their quarterly data consists of one-family houses in eight regions of Sweden from January 1981 to August 1995. The analysis consists of optimal portfolio analysis of real estate stocks, a general stock portfolio, T-bills, bonds, and houses. The results of their study suggest that return losses get larger with increasing portfolio shares in housing by an average of several percent. An efficient portfolio would include no housing for short holding periods, but for longer periods, low-risk portfolios would include between 15% and 50% housing.
Using data for five cities, Eichholtz, Koedijk, and de Roon use a mean-variance framework to examine residential property holdings. (10) They find that residential real estate provides significant diversification benefits, and that most U.S. investors have the optimal diversification benefits by allocating about 30% of their investment portfolio to residential real estate. However, neither stocks nor bonds are found to be a good hedge for housing.
Flavin and Yamashita, using housing price data from the Panel Study of Income Dynamics (PSID), estimate the risk and return to investments in residential real estate. (11) They confirm the absence of a positive correlation between housing and financial asset returns. Also, the inclusion of housing as an asset in the household wealth portfolio is found to dramatically improve the unconstrained mean-variance performance of the household portfolio. The PSID housing price data is based on annual household estimates of the value of their housing, not actual transactions. For the 1968-1992 period, Flavin and Yamashita estimate the average capital gain return to housing from the PSID data at 6.59% with a standard deviation of 14.24%.
Patterns of Housing Risk and Return
The total risk of housing investment theoretically can be decomposed into market risk (systematic risk) and house-specific risk (nonsystematic risk) as shown in the following equation:
Total housing risk = market risk (1) + house-specific risk
Market risk is that portion of total risk that is related to factors that systematically affect the returns for most homes, such as inflation, recession, and high interest rates. Market risk is measured in Table 1. House-specific risk is that portion of total risk that is related to factors that are specific to a particular house and its location, such as its structural characteristics, neighborhood environment, public services, infrastructure, local governmental regulations, and taxes. To assess the risk of single-family housing investment, this article draws on two samples of repeat home sales. The first sample is Greensboro, North Carolina (Guilford County). (12) This sample consists of 18,942 repeat sales of single-family homes that were bought and sold in Greensboro from 1975 through 2003. The data was taken from the master appraisal file maintained by the Guilford County Tax Assessor's Office. The geographic distribution of repeat sales in the county is illustrated in Figure A-1 of the Appendix.
The second sample is Houston, Texas (Harris County), from 1989 through 2004. This data was obtained from the Harris County Appraisal District. The sample contains 92,485 repeat sales. Unlike the Greensboro sample, the Houston sample is not geocoded and contains only a limited number of property attributes. The Appendix presents Table A1, which contains descriptive statistics for all variables used in the analysis for both samples.
Capital gain returns for discrete holding periods were calculated for each of the properties, as follows: (13)
[r.sub.it] = [([p.sub.i,t + n]/[p.sub.i,t]).sup.1/n] - 1 (2)
[r.sub.it] = annualized return for the ith property purchased in year t;
[p.sub.it] = sale price of the ith property in year t;
n = the number of years between the year of purchase t and the year of sale t + n.
The holding period return for an individual property ([r.sub.it]), was calculated by dividing the sales price ([p.sub.i,t + n]) by the purchase price ([p.sub.i,t]), taking the nth root, and subtracting one. The result is the annualized compounded return on the property held over n years.
The distribution of capital gain returns is shown in Figure I for Greensboro and in Figure 2 for Houston. They suggest that the total risk of investment in a single-family home in Greensboro or Houston is higher than the market risk shown in Table 1. In Greensboro, while the average return is 5.59%, the standard deviation of returns is 13.13% and the coefficient of variation is 2.35. In Houston, the average return is 4.8%, the standard deviation 5.36o and coefficient of variation is 1.12. In both cities, the standard deviation of returns is greater than the average return. However, the coefficient of variation indicates that the risk per each percentage return is more than double in Greensboro compared to Houston. In contrast, in Table 1, where risk and return are tabulated from the aggregated OFHEO national price index, the standard deviation is substantially less than the average return, resulting in very low risk as indicated by the small coefficient of variation.
Also shown in Figures 1 and 2 are the Jarque-Bera statistics that provide a test for whether the observed distributions of housing returns are normally distributed. The Jarque-Bera statistic is distributed as [chi square] with 2 degrees of freedom. The associated Probability is the probability that the Jarque-Bera statistic exceeds (in absolute value) the observed value under the null hypothesis. The small probabilities shown in Figures 1 and 2 suggest rejection of the hypothesis of normal distribution. To further examine the risk of housing investment, the risks and returns of investment for one-, three-, five-, seven-, and ten-year holding periods were tabulated, corresponding to the holding periods shown in Table 1. The results are shown in Table 3.
[FIGURES 1-2 OMITTED]
Table 3 shows that the average one-year capital gain return to housing investment in Greensboro (ignoring transaction costs) was 6.46% with a standard deviation of 25.52% and a coefficient of variation (CV) of 5.92. In Houston, the average one-year capital gain return was 6.62%, with a standard deviation of 9.80% and a CV of 1.48. The Houston sample suggests substantially less risk than Greensboro. (14) For comparison, Flavin and Yamashita using the PSID data reported an average return of 6.59% and a standard deviation of 14.24%. (15)
As in Table 1, the risk of housing investment declines with longer holding periods. However, the risk of housing investment in Greensboro (measured by the standard deviation of returns or the coefficient of variation) is substantially higher than the level of market risk shown in Table 1 for every holding period. In Houston, risks by holding period also are higher, but much lower than Greensboro.
Figure 3 plots the risk of the OFHEO index measured by the standard deviation of returns (market risk) against the standard deviation of returns in the Greensboro and Houston samples (total risk). For longer holding periods the differential between total and market risk declines, but total risk is still substantially higher than market risk even with a ten-year holding period.
Models of Housing Returns
To further explore the relationship between total risk and market risk, a series of single-factor return models for specific holding periods (n) were estimated. The models are defined as follows:
[r.sub.it] = [alpha] + [beta][RN.sub.1] + [epsilon.sub.i.t] (3)
[RN.sub.t] = annualized return for the average of all properties nationally that were purchased in year t and sold after n years (16)
[[epsilon].sub.i.t] = a random error term
The estimated return models are shown in Table 4; t-values appear in parenthesis below the estimated beta coefficients, and F-values are shown in parenthesis below the [R.sup.2] statistics. The models are estimated using the White adjustment of heteroskedasticity. (17) The estimated return models for five-, seven-, and ten-year holding periods are statistically significant at the 0.01 level and above.
Looking first at Greensboro, the models for one-and three-year holding periods are not statistically significant. For the five-, seven-, and ten-year holding period models, the estimated beta coefficients are positive, but less than one. For these holding periods, the beta coefficients indicate that housing returns in Greensboro tend to move in the same direction as returns nationally, but less so, on average, than in other markets nationwide. The size of the estimated beta coefficients tends to rise with longer holding periods. However, in every case, the betas are less than one, indicating that the level of market risk in Greensboro is below that for housing markets nationally.
In the Houston sample, all of the models are statistically significant and all of the beta coefficients are positive. The magnitude of the estimated betas rises with the length of the holding period except for the ten-year holding period, which drops. The estimated betas suggest that housing investment in Houston has more market risk than in Greensboro except for the ten-year period, but for all holding periods except the seven-year period, market risk in Houston is below the national average.
The [R.sup.2] statistics for all of the estimated equations in both Greensboro and Houston are all low, indicating that market risk accounts for a very small fraction of the total risk of housing investment. The [R.sup.2] statistics for Greensboro are highest with the ten-year holding-period model; however, even in this case, it is only 0.0615. Overall, the [R.sup.2] statistics are substantially higher in the Houston sample.
To try to increase the explanatory power of Equation (2), a multifactor return model was estimated using the Greensboro sample. A similar multifactor model could not be estimated using the Houston sample because data for the additional variables were not available. In addition to [RN.sub.t], the new variables included the following: (18)
[HP.sub.i,t] = the holding period for the ith property, or n
[CEMP.sub.t] = annualized growth in employment in the Greensboro/Winston-Salem/ High Point MSA from year t when the home was purchased until it was sold n years later
[HVAL.sub.i,t] = average value of homes in the neighborhood (zip code)
[AGE.sub.i,t] = the age of the ith property when purchased
[SQFT.sub.i,t] = the square footage of the ith property in 1,000s
[ATYP.sub.i,t] = a measure of the atypicality of the ith property (19)
[GSO.sub.i,t] = a dummy variable equal to one if the ith property was located in Greensboro
[HIPT.sub.i,t] = a dummy variable equal to one if the ith property was located in High Point
The results for the multifactor return model are shown in Table 5. The model is estimated using the White adjustment of heteroskedasticity. All of the estimated coefficients are statistically significant at the 0.01 level or above.
The [R.sup.2] is 0.081, indicating that the model with the additional variables still explains only 8.1% of the variation in housing returns and that the unsystematic risk of housing investment is quite large.
The estimated coefficient on [RN.sub.t] is positive, but less than one, indicating that Greensboro housing investment is positively associated with returns nationally. However, housing returns in Greensboro are less sensitive to housing return movements nationwide, when compared to housing returns in other housing markets including Houston. The growth in local employment is a positive and significant determinant of housing returns. The coefficient on holding period is negative, showing that returns fall with longer holding periods. This is the same pattern shown in Table 1.
The coefficients on age, square footage, and atypicality are interesting. Returns appear to rise with age and size of the home, but returns are lower for homes that are more atypical. These results suggest the importance of structural characteristics, layout, and design as determinants of returns.
The coefficient on neighborhood housing values is negative, showing that returns tend to be lower in neighborhoods where housing values are highest. Similarly, the coefficients for location in Greensboro and High Point are negative, indicating lower returns in these cities. The statistical significance of the neighborhood and city variables confirms the importance of location as a strong determinant of housing returns.
Assessing the Probability of Loss
Another way to measure the risk of housing investment is the percent of repeat sales that resulted in a loss, where a loss is defined as: [p.sub.t + n] - [p.sub.t] < 0. For the entire Greensboro sample of 18,942 observations, 10.9% of the transactions resulted in a loss. In Houston, 9.1% of transactions resulted in a loss. For most homeowners, this is an important statistic because they are likely to assess the risk of an investment by the probability of loss, rather than the variation of returns.
The foregoing assessment of the probability of loss ignores transaction costs. If it is assumed that transaction costs are approximately 7% for an average residential repeat sale, the loss (LOSS7) can be defined as: 0.93 * [p.sub.t+n] - [p.sub.t] < 0. (20) Using this estimate of transaction costs, 21.0% of the repeat sales in Greensboro and 23.4% in Houston resulted in losses.
To explore the determinants of loss on sale, a probit model is estimated using the Greensboro data in which the dependent variable was defined as [LOSS.sub.i,t], a dummy variable equal to 1 if the property sold for less than its purchase price, and 0 otherwise. The independent variables included all those in Table 5 discussed above.
The estimated probit model is shown in Table 6. The model is estimated using the Huber/White procedure for robust standard errors. The probability of loss is positively and significantly related to the following variables: (1) the atypicality of the property, and (2) location in the cities of Greensboro and High Point. The probability of loss is negatively and significantly associated with the following variables: (1) local employment growth, (2) holding period, (3) neighborhood property values, (4) age, and (5) size of the structure. Interestingly, the loss probability is not significantly related to the growth in housing prices nationally.
For housing investors, the results indicate that investments in atypical houses carry substantially more risk. Housing investment risk is less in cities where employment is rapidly expanding. Risk also is lessened with longer holding periods of larger, older homes in neighborhoods with higher overall property values.
The interpretation of the coefficient values in Table 5 is complicated by the fact that estimated coefficients from a probit model cannot be interpreted as the marginal effect on the dependent variable. The marginal effect of an independent variable (x) on the conditional probability of loss (L) is given by:
[delta]E(L | x, [beta])/[delta][x.sub.j] = f(-x'[beta])[[beta].sub.j] (4)
where f(x) = dF(x)/dx is the density function associated with F. The marginal effects of each independent variable on the probability of loss are shown in the third column of Table 6. This approach is not appropriate for dummy variables. The marginal effects of the dummy variables are analyzed by comparing the probabilities that result when the variables take on their two different values with those that occur with the other variables held at their sample means. (21) Atypicality of the property has the largest positive marginal effect on the probability of loss, while size and holding period have the largest negative marginal effects. A one-year increase in holding period reduces the probability of loss by 2.9%.
Table 7 presents an alternative probit model where the loss variable is calculated to consider transactions costs as discussed above. Here loss is defined as: [LOSS7.sub.i,t] = 1, if 0.93*[p.sub.t + n] - [p.sub.t] < 0, and 0 otherwise.
As in Table 6, the model in Table 7 is estimated using the Huber/White procedure for robust standard errors. The probability of loss is positively and significantly related to the following variables: (1) the atypicality of the property, and (2) location in High Point. The probability of loss is negatively and significantly associated with the following variables: (1) the growth rate in housing values nationally, (2) local employment growth, (3) holding period, (4) neighborhood property values, (5) age, and (6) size of the structure. Again, the atypicality of the property has the largest positive marginal effect on the probability of loss, while holding period has the largest negative marginal effect. A one-year increase in holding period in this case reduces the probability of loss by 5.0%.
For housing investors the results are clear: investments in atypical houses carry substantially more risk. Housing investment risk is less in cities where employment is rapidly expanding and also when housing returns are increasing nationwide. In comparison to the probit model with no transaction costs, risk is lessened to an even greater extent for longer holding periods as well as for older homes. But, while risk is still less for large houses and for homes in neighborhoods with higher overall property values, these variables have less effect than in the probit model with no transactions costs.
Summary and Evaluation
The foregoing analysis of housing risk in Greensboro and Houston shows that the price risk facing the average homeowner is high and may vary widely among cities across the country. Further research needs to focus more closely on the extent and determinants of the variation in risk levels among cities.
The analysis shows that the return on an individual housing transaction is positively associated with national returns, but the association is not strong, which indicates a high level of nonsystematic risk in housing transactions. Factors such as local employment growth and characteristics of the house's location and other features are very important in explaining individual housing returns.
The probability of loss on sale in the average housing transaction is large. The probability is inversely related to holding period, housing returns nationally, and local employment growth. It also is strongly influenced by location and other features.
The analysis presented here is drawn from just two metropolitan areas. Whether the pattern of results reported is found to apply generally in other area across the county is a topic that merits additional inquiry.
(1.) Ana M. Aizcorbe et al., "Recent Changes in U.S. Family Finances: Evidence from the 1998 and 2001 Survey of Consumer Finances," Federal Reserve Bulletin (January 2003): 1-32.
(2.) Karl E. Case Jr., Robert J. Shiller, and Allan N. Weiss, "Index-Based Futures and Options Markets in Real Estate," Journal of Portfolio Management 19, no. 2 (Winter 1993): 83-92; Andrew Caplin et al., "Household Asset Portfolios and the Reform of the Housing Finance Market," TIAA-CREF Research Dialogues (February 1999).
(3.) Chicago Mercantile Exchange, "CME and MARCO Securities Research to Explore Development of Futures Contracts Based on Housing Prices," M2PressWIRE, December 6, 2004.
(4.) Standard deviation is a measure of variability. It is defined as:
[sigma] = [[[n.summation over (t=1] [(r.sub.t - [bar.r]].sup.2]]].sup.05],
where [r.sub.t] = is the holding period return at time t and [bar.r] is the average return for all similar holding periods. The average return is calculated as:
[bar.r] = [n.summation over (t=1)[r.sub.t]/n
The coefficient of variation (CV) is risk per unit of return, or [sigma]/[bar.r]. In the financial literature, the standard deviation and the coefficient of variation are used as measures of investment risk, and the average is the typical value of a set of values, sometimes called the mean.
(5) The national OFHEO House Price Index is tabulated quarterly from 1975.1 through 2003. Table I shows assumed holding periods only for 10 years and less.
(6.) Thomas J. Coyne, Waldemar M. Goulet, and Mario J. Picconi, "Residential Real Estate Versus Financial Assets," Journal of Portfolio Management 7, no. 1 (Fall 1980): 20-24; William W. Alberts and Halbert S. Kerr, "The Rate of Return From Investing in Single-Family Housing," Land Economics 57, no. 2 (May 1981): 230-242; Patric H. Hendershott and Sheng Cheng Hu, "Inflation and Extraordinary Returns on Owner-Occupied Housing: Some Implications for Capital Allocation and Growth," Journal of Macroeconomics 3, no. 2 (Spring 1981): 177-203; Richard B. Peiser and Lawrence B. Smith, "Homeownership Returns, Tenure Choice and Inflation," AREUEA Journal 13, no. 4 (Winter 1985): 343- 60; James R. Webb and Jack H. Rubens, "Tax Rates and Implicit Rates of Return on Owner-Occupied Single-Family Housing," Journal of Real Estate Research 2, no. 1 (Winter 1987):11-28; Norman G. Miller and Michael A. Sklarz, "A Comment on 'Tax Rates and Implicit Rates of Return on Owner-Occupied Single-Family Housing,'" Journal of Real Estate Research 4, no. 1 (Spring 1989): 81-84; Karl E. Case and Robert J. Shiller, "Forecasting Prices and Excess Returns in the Housing Market," AREUEA Journal 18, no. 3 (Fall 1990): 253-273; Charles M. Ermer, Steven M. Cassidy, and Michael J. Sullivan, "Modeling Returns to Owner-Occupied Single-Family Residences," Journal of Economics and Finance 18, no. 2 (Summer 1994): 205-217; Jack Harris, "Investment Performance of Owner-Occupied Housing" (paper presented at the annual meeting of the American Real Estate Society, Naples, FL, April 2002); G. Donald Jud and Daniel T. Winkler, "Returns to Single-Family Owner-Occupied Housing," Journal of Real Estate Practice and Education 8 (forthcoming).
(7.) Karl E. Case and Robert J. Shiller, "Prices of Single-Family Homes Since 1970: New Indexes for Four Cities," New England Economic Review 22, no. 2 (September/October 1987): 45-56.
(8.) William N. Goetzmann, "The Single Family Home in the Investment Portfolio," Journal of Real Estate Finance and Economics 6, no. 3 (May 1993): 201-222.
(9.) Peter Englund, Min Hwang, and John M. Quigley, "Hedging Housing Risk," Journal of Real Estate Finance and Economics 24, no. 1/2 (January-March 2002): 167-200.
(10.) Piet M. A. Eichholtz, Kees G. Koedijk, Frans A. de Roon, "The Portfolio Implications of Home Ownership" (manuscript, Maastricht University and University of Amsterdam, 2002).
(11.) Marjorie Flavin and Takashi Yamashita, "Owner-Occupied Housing and the Composition of the Household Portfolio," American Economic Review 92, no. 1 (March 2002): 345-362.
(12.) Guilford County is located in the Piedmont area of North Carolina, in the northern center of the state. It lies astride 1-40, which connects the state east and west from Wilmington to Knoxville and 1-85, which extends south from Washington through the county and on to Atlanta. The county's two main cities are Greensboro and High Point. The population of Guilford County in 2000 was 421,048, making it the third-largest county in the state, behind Mecklenburg (Charlotte) and Wake (Raleigh). From 2000 through 2003, county population grew 3.0%, lagging North Carolina's overall increase of 4.4%. The countywide unemployment rate was 5.2% in July 2004, compared to the statewide rate of 5.0%. In the second quarter of 2004, there were 219,444 persons employed in the county (from the household survey). From 1990 through 2003, employment in Guilford County rose 1.0% annually, trailing the 1.3% increase in employment recorded for North Carolina as a whole. During this period, the county sustained large employment losses in the manufacturing sector. Manufacturing employment, which accounted for 15.4% of employment in 2003, fell 1.8% annually from 1990 through 2003, recording large losses in the apparel, textile, and furniture sectors.
(13.) In the calculation of the annualized returns, an effort was made to exclude those properties where there was substantial change in the attributes of the property from time t to t + n; however, it cannot be certain that all such transactions have been excluded. Nonetheless, the number of such properties is likely to be small, and therefore, have a minimal impact on the findings.
(14.) The Greensboro sample has more risk than the Houston sample even when restricted to 1989-2003, almost the same time span as the Houston sample. Here the average return in Greensboro was 5.07%, with a standard deviation of 13.45% and a CV of 2.65. There were 7,256 repeat sales in the Greensboro sample during this period.
(15.) Flavin and Yamashita.
(16.) The inclusion of [RN.sub.t] on the right hand side of Equation (3) captures the effects of inflation and interest rate trends as they affect the overall return to housing nationally.
(17.) Halbert White, "A Heteroskedasticity-Consistent Covariance Matrix and a Direct Test of Heteroskedasticity," Econometrica 48, no. 4 (May 1980): 817-838.
(18.) The means and standard deviations of all variables used in the analysis are shown in Appendix Table A-1. Appendix Table A-2 provides variable definitions. Neither a time-trend variable nor a variable for year of second sales was included in the expanded model because each one is collinear with [RN.sub.t].
(19.) To measure the atypicality ([ATYP.sub.i,t]) of a particular property an index is developed which takes the absolute value of the deviation of the property's actual sale price from its predicted price on final sale, The index is defined as follows:
[ATYP.sub.t,t] = [n.summation over (i=1)] [absolute value of][p.sub.i][a.sub.i] - [??][a.sub.i]
The hedonic price equation used to construct the index along with sample statistics are shown in Appendix Table A-3 and Table A-4. Table A-3 shows the estimated regression equation and Table A-4 shows the sample statistics.
(20.) Transaction costs include mortgage origination fees, brokerage charges, and other closing costs involved in the purchase and sale of residential property. The largest of these costs is sales commissions; a recent article in The Wall Street Journal Online estimated that the average sales commission is 5.1% of the selling price. Kelly A. Spors, "What You Need to Know About Commission Rates," The Wall Street Journal Online, September 20, 2004, http://www.realestatejournal.com/buysell/agentsandbrokers/20040920-spors.html. While the estimate of 7% in the present study is difficult to verify empirically, it represents the authors' best judgment of the burden borne by the average homeowner.
(21.) William H. Greene, Econometric Analysis (New York: Macmillan Publishing Co,, 1990), 704-705.
G. Donald Jud, PhD, is professor emeritus of finance in the Bryan School of Business and Economics at the University of North Carolina at Greensboro were he taught courses in finance and real estate. He received his PhD from the University of Iowa, and MBA and BA degrees from the University of Texas. He is author of over 80 academic articles and three books; his research has appeared in numerous academic and professional journals. Jud serves on the editorial boards of the Journal of Real Estate Finance and Economics, Journal of Real Estate Research, and Journal of Real Estate Literature and is a member of The Appraisal Journal's academic review panel. He is a past editor of the Journal of Real Estate Research, a past president of the American Real Estate Society (ARES), and former ARES Director of Publications. He is a research fellow of the Homer Hoyt Advanced Studies Institute, where he is an emeritus member of the Weimer School faculty and the board of directors of the Institute. Contact: Bryan School of Business and Economics, University of North Carolina at Greensboro, Greensboro, NC, 27412-5001; T 336-334-3091; E-mail: firstname.lastname@example.org
Stephen E. Roulac, PhD, JD, is founder of The Roulac Group, an international firm advising senior management and investors on complex real estate decisions. He received his PhD from Stanford University and his JD from the University of California, Berkeley. Roulac is the author of 12 books and many articles dealing with various aspects of real estate; his research has appeared in numerous academic and professional journals. Roulac's professional and scholarly work has been recognized by the American Real Estate Society with the prestigious ARES James A. Graaskamp Award (1997). He is a research fellow of the Homer Hoyt Advanced Studies Institute. He has been a visiting professor/lecturer at the University of Ulster, University of California at Davis, Cleveland State University, Dartmouth College, Texas A&M University, and the University of Chicago. Roulac is a member of The Appraisal Journal's Academic Review Panel, and he serves on the editorial boards of the Journal of Housing Research, International Real Estate Review, and Journal of Real Estate Practice and Education. He is a past president of the American Real Estate Society and continues to serve as its director of strategy. Contact: The Roulac Group, Inc., San Rafael, CA; T 415-451-4300; E-mail: email@example.com
Daniel T. Winkler, PhD, is a professor of finance and department head in the Bryan School of Business and Economics at the University of North Carolina at Greensboro (UNCG). He holds a PhD from the University of South Carolina and an MBA from the University of Central Florida. Since arriving at UNCG in 1986, he has taught courses in corporate finance, investments, and real estate. Winkler has published over 40 academic articles in areas of real estate, finance, and insurance; his real estate research has been published in numerous academic and professional journals. Winkler is a co-editor of the Journal of Real Estate Practice and Education. Contact: Bryan School of Business and Economics, University of North Carolina at Greensboro, Greensboro, NC 27412-5001; T 336-334-4524; E-mail: firstname.lastname@example.org
Table A-1 Sample Statistics Greensboro, NC (n = 18,942: 1975-2003) [r.sub.i,t] [LOSS.i,t] [LOSS7.i,t] [RN.sub.t] Mean 5.59 0.11 0.21 5.82 Median 4.20 0.00 0.00 5.13 Std. Dev. 13.13 0.31 0.41 3.07 [CEMP.sub.t] [HP.sub.i,t] [HVAL.sub.i,t] [AGE.i,t] Mean 2.91 6.19 114.68 31.75 Median 2.83 5.00 107.10 30.00 Std. Dev. 2.31 4.58 29.07 19.98 [SQFT.i,t] [ATYP.i,t] [GSO.i,t] [HIPT.i,t] Mean 1.64 0.22 0.77 0.21 Median 1.46 0.15 1.00 0.00 Std. Dev. 0.71 0.25 0.42 0.41 Houston, TX (n = 92,485: 1989-2004) [r.sub.i,t] [LOSS.i,t] [LOSS7.i,t] [RN.sub.t] Mean 4.80 0.09 0.23 4.92 Median 4.10 0.00 0.00 4.83 Std. Dev. 5.36 0.29 0.42 1.99 [CEMP.sub.t] [HP.sub.i,t] [HVAL.sub.i,t] [AGE.i,t] Mean n.a. n.a. n.a. n.a. Median n.a. n.a. n.a. n.a. Std. Dev. n.a. n.a. n.a. n.a. [SQFT.i,t] [ATYP.i,t] [GSO.i,t] [HIPT.i,t] Mean n.a. n.a. n.a. n.a. Median n.a. n.a. n.a. n.a. Std. Dev. n.a. n.a. n.a. n.a. Table A-2 Variable Definitions Variable Name Definition GSO Location in Greensboro HIPT Location in High Point HVAL Ave. neighborhood home value (zip code) Brick Brick veneer construction Central Air Conditioning Central air conditioning # of Bedrooms Number of bedrooms # of Bathrooms Number of bathrooms # of Fireplaces Number of fireplaces Age Age of structure Square Feet Square feet Golf Course Location on a golf course Front Feet Front feet Depth of Lot Depth of lot Quality Quality index used by tax appraiser Table A-3 Hedonic Price Equation Used to Calculate the Atypicality Index (Dependent Variable = Log(Price)) Variable Coefficient t-value Intercept 8.4818 337.84 1976 0.0804 4.30 1977 0.1424 8.08 1978 0.2518 14.52 1979 0.2550 14.79 1980 0.3506 19.78 1981 0.3403 18.57 1982 0.3837 20.81 1983 0.5285 30.63 1984 0.5949 35.34 1985 0.6933 41.66 1986 0.8186 48.55 1987 0.8556 49.64 1988 0.8891 51.58 1989 0.8941 51.67 1990 0.9297 53.49 1991 0.9524 54.12 1992 0.9807 57.15 1993 1.0167 59.89 1994 1.0452 62.16 1995 1.0944 64.18 1996 1.1426 66.91 1997 1.2463 73.38 1998 1.2594 74.99 1999 1.2911 77.23 2000 1.3135 77.70 2001 1.3672 81.71 2002 1.3933 86.96 2003 1.3909 84.20 GSO 0.1121 9.31 HIPT 0.0156 1.26 HVAL 0.0032 43.20 Brick 0.0648 16.72 Central Air Conditioning 0.1325 27.20 # of Bedrooms 0.0259 6.88 # of Bathrooms 0.0380 8.57 # of Fireplaces 0.0655 33.89 Age -0.0036 -33.72 Square Feet 0.0002 41.09 Golf Course 0.1225 2.93 Front Feet 0.0004 5.53 Depth of Lot -0.0001 -2.17 Quality 0.1852 46.48 [R.sup.2] 0.6642 N 51,215 Note: The hedonic index equation is estimated on the basis of a sample of 51,215 residential properties sold in Greensboro, NC, between 1975 and 2003. Data from the master appraisal file of the Guilford County Tax Assessor's Office. Table A-4 Sample Statistics of Properties Used to Estimate Hedonic Equation in Table A-3 Variable Mean Standard Deviation 1976 0.021 0.144 1977 0.028 0.166 1978 0.031 0.174 1979 0.032 0.176 1980 0.028 0.164 1981 0.024 0.152 1982 0.023 0.150 1983 0.032 0.177 1984 0.037 0.190 1985 0.040 0.197 1986 0.037 0.189 1987 0.033 0.178 1988 0.033 0.178 1989 0.032 0.177 1990 0.031 0.175 1991 0.029 0.169 1992 0.034 0.182 1993 0.036 0.187 1994 0.039 0.193 1995 0.036 0.186 1996 0.036 0.186 1997 0.037 0.189 1998 0.040 0.197 1999 0.042 0.201 2000 0.040 0.195 2001 0.042 0.201 2002 0.060 0.237 2003 0.047 0.212 GS0 0.749 0.433 HIPT 0.223 0.416 HVAL 111.085 29.621 Brick 0.420 0.494 Central Air Conditioning 0.757 0.429 # of Bedmomn 3.049 0.697 # of Bathrooms 1.647 0.690 # of Fireplaces 2.456 1.253 Age 28.746 21.103 Square Feet 1637.040 738.351 Got Courue 0.002 0.045 Front Feet 82.262 27.157 Depth cf Lot 163.620 43.777 Quality 3.195 0.635 Log(Price) 11.096 0.719 N 51,215 Figure 3 Total Risk Versus Market Risk Greensboro, NC Houston, TX National 1 25.3% 9.8% 3.2% 3 14.2% 5.8% 2.9% 5 8.7% 4.1% 2.3% 7 6.8% 3.4% 1.8% 10 4.9% 2.4% 1.5% Note: Table made from bar graph. Table 1 Holding Period Risks and Returns for Housing, 1975-2003 (Capital Gains Only) Holding Periods (Years) 1 3 5 7 10 Ave. Return 5.7% 5.6% 5.3% 5.0% 4.8% Std. Dev. 3.2% 2.9% 2.3% 1.8% 1.5% Cv 0.56 0.52 0.44 0.36 0.31 n 28 26 24 22 19 Computed using annual data for the national OFHEO House Price Index. Table 2 Holding Period Risks and Returns for Stocks, 1975-2003 (Capital Gains Only) Holding Periods (Years) 1 3 5 7 10 Ave. Return 9.8% 10.0% 11.1% 11.4% 11.5% Std. Dev. 13.6% 9.1% 6.0% 3.8% 2.2% Cv 1.39 0.91 0.54 0.33 0.19 n 28 26 24 22 19 Computed using annual data for the S&P 500 Index. Table 3 Holding Period Risks and Returns for Housing in Greensboro, NC (1975-2003) and Houston, TX (1989-2004) (Capital Gains Only) Holding Periods (Years) 1 3 5 7 10 Greensboro, NC Ave. Return 6.46% 6.30% 4.85% 4.94% 4.53% Std. Dev. 25.32% 14.15% 8.71% 6.78% 4.92% CV 3.92 2.25 1.80 1.37 1.09 No. of Repeat Sales 2,106 2,233 1,672 1,253 738 Houston, TX Ave. Return 6.62% 5.23% 4.57% 4.07% 3.92% Std. Dev. 9.80% 5.78% 4.06% 3.38% 2.40% CV 1.48 1.11 0.89 0.83 0.61 No. of Repeat Sales 10,778 26,039 19,916 13,629 7,095 Table 4 Determinants of Housing Returns Greensboro, NC Holding Period (n) Constant [beta] [R.sup.2] N 1 6.9462 -0.0553 0.0001 2,106 (-0.5789) (0.3060) 3 6.6918 -0.0639 0.0002 2,233 (-0.5774) (0.3752) 5 3.1603 0.3226 0.0061 1,672 (3.0573) (10.2495) 7 1.2793 0.7433 0.0348 1,253 (6.6067) (45.0684) 10 0.5262 0.8268 0.0613 738 (6.8298) (48.1033) Houston TX Holding Period (n) Constant [beta] [R.sup.2] N 1 4.0047 0.4660 0.0137 10,778 (12.2436) (149.94) 3 1.5899 0.6731 0.0650 26,039 (42.6013) (1,812.30) 5 0.4587 0.8316 0.1728 19,916 (64.4651) (4,162.22) 7 -0.3629 1.0101 0.2356 13,629 (64.7500) (4,202.53) 10 1.2782 0.6370 0.0708 7,095 (23.2482) (541.64) Table 5 Determinants of Housing Returns, Greensboro, NC Variable Coefficient t-value Constant 9.188 9.10 [RN.sub.t] 0.191 4.32 [CEMP.sub.t] 0.211 3.56 [HP.sub.i,t] -0.185 -9.65 [HVAL.sub.i,t] -0.018 -4.76 [AGE.sub.i,t] 0.059 9.96 [SQFT.sub.i,t] 1.266 8.21 [ATYP.sub.i,t] -14.171 -17.11 [GSO.sub.i,t] -2.749 -3.50 [HIPT.sub.i,t] -4.088 -5.03 [R.sup.2] 0.081 N 18,942 Table 6 Probit Model of the Probability of Loss (Dependent Variable = [LOSS.sub.it]) Variable Coefficient Marginal Effect z-Statistic Constant -0.910 n.a. -6.51 [RN.sub.t] -0.004 -0.002 -0.96 [CEMP.sub.t] -0.050 -0.020 -7.20 [HP.sub.i,t] -0.072 -0.029 -18.17 [HVAL.sub.i,t] -0.004 -0.002 -6.96 [AGE.sub.i,t] -0.005 -0.002 -6.65 [SQFT.sub.i,t] -0.117 -0.047 -4.49 [ATYP.sub.i,t] 1.880 0.748 26.75 [GSO.sub.i,t] 0.391 0.047 3.27 [HIPT.sub.i,t] 0.679 0.105 5.58 N 18,942 Log likelihood -5,139.09 Restr. log likelihood -6,529.87 LR statistic (9 d.f.) 2,780.57 Table 7 Probit Model of the Probability of Loss (Dependent Variable = [LOSS7.sub.it]) Variable Coefficient Marginal Effects z-Statistic Constant 0.244 n.a. 2.42 [RN.sub.t] -0.039 -0.015 -9.81 [CEMP.sub.t] -0.047 -0.018 -8.78 [HP.sub.i,t] -0.126 -0.050 -32.66 [HVAL.sub.i,t] -0.002 -0.001 -4.45 [AGE.sub.i,t] -0.007 -0.003 -10.82 [SQFT.sub.i,t] -0.068 -0.027 -3.72 [ATYP.sub.i,t] 1.480 0.583 25.72 [GSO.sub.i,t] 0.105 0.025 1.31 [HIPT.sub.i,t] 0.291 0.073 3.51 N 18,942 Log likelihood -8,134.90 Restr. log likelihood -9,726.32 LR statistic (9 d.f.) 3,182.85
|Printer friendly Cite/link Email Feedback|
|Author:||Jud, G. Donald; Roulac, Stephen E.; Winkler, Daniel T.|
|Date:||Sep 22, 2005|
|Previous Article:||Evaluating environmental stigma with multiple regression analysis.|
|Next Article:||Expropriation claims for increased development costs and loss of profit.|