Measuring Residential Real Estate Risk and Return

IMPORTANCE OF RESIDENTIAL REAL ESTATE FOR HOUSEHOLD PORTFOLIOS AND THE ECONOMY

Lustig and Van Nieuwerburgh [2005] reported that real estate is one of the largest asset classes and was often the major household asset. It is generally considered that the value of residential real estate is twice the principal value of outstanding mortgages. Residential mortgages were at about $10 trillion in 2009, therefore suggesting a $20 trillion residential real estate market (Federal Reserve [2010]). In addition, commercial real estate was worth about $6.5 trillion in 2009 and was financed by an estimated $3.1 trillion in debt (Patalon [2009]).

Real estate constitutes a nontrivial fraction of an individual’s asset portfolio, mainly through the purchase of the primary residence. Nearly 70% of U.S. households own their home. Consequently, residential real estate wealth accounts for 64.6% of the household’s nonfinancial assets (out of which 48% are attributed to the primary residence) and 40% of the household’s total net worth (Federal Reserve [2009]).² For the median-income family in particular, the primary residence alone constitutes almost 72% of the household’s nonfinancial assets and 47.4% of the household’s total net worth (Federal Reserve [2009]).

The relative importance of housing in the total asset portfolio depends on household income and age. Specifically, the weight of housing decreases with income and is particularly low for the top 10% of the income distribution, for which the primary residence constitutes only 19.8% of the household’s net worth. Similarly, the ratio of housing to net worth declines over the lifecycle for all income categories, as income and savings increase, and the housing constraint generally creates a lifecycle pattern in the portfolio ratio of stocks and bonds.

In addition to serving a salient role in the household’s asset portfolio, real estate wealth has an important link to consumption and gross domestic product (GDP) growth. Specifically, fluctuations in housing wealth affect household consumption more than fluctuations in financial assets. Case, Quigley, and Shiller [2005] showed that the elasticity of consumption with respect to housing wealth ranges from 0.05 to 0.09. Therefore, for housing wealth of $100, a one-dollar increase in house prices results in a consumption increase of 5 to 9 cents in a given year. In contrast, a number of studies find that the elasticity of consumption with respect to financial assets is significantly smaller than it is with respect to housing wealth. For example, Bostic, Gabriel, and Painter [2009] found that from 1989–2001, elasticity of consumption with respect to housing wealth was 0.06, compared with 0.02 for financial wealth. Similarly, Benjamin, Chinloy, and Jud [2004] examined the 1952–2001 period and reported that a one-dollar increase in real estate wealth increases consumption by 8 cents while the same increase in financial wealth increases consumption by only 2 cents.

High elasticity of consumption with respect to housing wealth is particularly critical for GDP growth since nearly 70% of GDP growth is attributed to consumption. Bostic, Gabriel, and Painter [2009] showed that a 10% decline in housing wealth from 2005 levels results in a 1% decline in real GDP growth. Compared with an approximate average annual real GDP growth of 4%, such a decline in GDP is nontrivial.

The evidence about the role of real estate in individual portfolios and the overall economy discussed above highlight the considerable importance of this asset class. Combined, the size of the real estate market, the weight of real estate in an individual’s portfolio, and the relation of real estate to consumption and GDP all point toward economy-wide risks associated with potential declines in the real estate market. Unfortunately, these risks were largely overlooked until the collapse of real estate prices around the world that triggered the financial crisis of 2008.

Along with the overall economy, the real estate market follows business cycles. Since the 1970s, the United States has experienced four boom–bust cycles in the real estate housing market with peaks in 1972, 1978, 1984, and 2006 (Case [2008]). Interestingly, however, both individuals and lenders may not have explicitly acknowledged this cyclicality and have acted, especially in recent years, with the expectation that home prices rarely or never fall. Indeed, the national quarterly repeat sales index of the Federal Housing Finance Agency (FHFA, formerly, Office of Federal Housing Enterprise Oversight [OFHEO]) rose 532.4% between 1975 and 2007. Also, even though the S&P/Case–Shiller index shows falling home prices in various quarters, the overall trend of this index between 1987 and 2005 was upward sloping. Similarly, rating agencies used the data from the previous several decades of continuously rising home prices to back test their default models. Further, the models assumed that as long as a portfolio was regionally diversified, the effect of prices falling only regionally should be smoothed.³ However, when prices synchronized nationally after 2006, the models failed to fit the data (Case [2008]).

Several other important factors distinguished the most recent real estate cycle of 2006. Specifically, the troughs of the earlier three cycles were accompanied by inflationary pressure, which necessitated contracting monetary policy. With tighter availability of credit, demand for housing declined while unsold inventories increased, necessitating a fall in new housing starts. Eventually, household formation absorbed the excess inventories, thus allowing for output to rebound.

In contrast, the last cycle was characterized by a lack of inflationary pressure. The slowdown of 2001 called for a fast response to stimulate the economy. In 2001 alone, the Federal Reserve cut interest rates 11 times, from 6.5% to 1.75%. Importantly, real estate, unlike other asset classes, is considerably more sensitive to changes in monetary policy and credit availability (Case [2008]).

Credit was widely available in the early 2000s because the United States experienced a tremendous inflow of cheap capital from developing countries.⁴ Between 2000 and 2008, the United States received over $5.7 trillion from abroad to invest, which constituted 40% of 2007 GDP. Large amounts of capital and historically low levels of interest rates resulted in a credit boom. Households and the financial sector absorbed most of the capital (see Exhibit 1).⁵ Financial sector debt grew from 20% to 120% of GDP in the last three decades. Household debt grew from 50% to 100% of GDP in the last 20 years. Such extraordinary expansion of the financial sector and household debt was mainly used to finance investments in real estate equity and debt markets, because stock and bond markets provided relatively unattractive returns. The size of the U.S. residential mortgage market increased from $4.8 billion to $10.5 billion between 2000 and 2007 (Federal Reserve [2010]).

• Download figure
• Open in new tab
• Download powerpoint

Exhibit 1

Private Sector Debt Outstanding (in billions)

Source: Federal Reserve Board, Bureau of Economic Analysis.

One of the reasons for the fast expansion of the residential mortgage market was the increased popularity of adjustable rate, interest only, negative amortization, and other exotic mortgages, which offered very low monthly payments for the first few years of the loan. Low initial monthly payments accompanied with small, if any, down payment requirements and the ability to state income without verification allowed more individuals to qualify for mortgages. Many borrowers assumed that interest rates on such mortgages would remain at these historically low levels. As payments on such loans increased, many borrowers became delinquent, as their ability to pay at closing did not amount to the ability to pay over the long run.

Shiller [2005] argued that upward pressure on house prices shaped the public’s opinion that house prices can only go up and that if one did not buy a house now, he or she would never be able to become a homeowner. This change in public sentiment amplified the real estate bubble. In fact, individuals, who share predictable cognitive errors, such as overconfidence and excessive optimism, have caused many bubbles.

The real estate market started to cool off in 2008, affecting the broader economy. There has been extensive debate seeking to identify factors that caused the crisis. Homebuyers took out mortgages larger than they could afford. Mortgage originators lowered underwriting standards and loaned more money than they should have, because they earned a fee for originating mortgages and no longer bore the risk of mortgage default. Investment banks did not bear the risk of mortgage default either and could not securitize mortgages fast enough to satisfy the demand for highly rated securities that offered high yields. Rating agencies, paid by investment banks, failed to properly estimate the riskiness of the mortgage pools that they rated. Buyers of mortgage-backed securities blindly relied on the ratings and assumed they were buying low-risk assets. The government contributed to the crisis by mandating that banks and government-sponsored enterprises (GSEs) lower lending standards so that more low-income individuals could afford homeownership.⁶ Moreover, regulators overlooked the broken chain of incentives and excessive risk-taking by all parties involved in securitization.

Clearly, no single party is solely responsible for the crisis. Yet had individuals not purchased houses they could not afford, simply hoping for price appreciation, it is likely that the vicious circle would not have arisen and no financial crisis would have occurred in 2008. Unfortunately, many individuals are not financially literate, and even those with business degrees rarely learn basic concepts about real estate. Even the interpretation of real estate returns often appears to be clouded by numerous misconceptions, leading one to believe that house prices rarely or never fall and that returns are superior and almost guaranteed. We discuss such misconceptions in the next section, after which we describe proper ways of calculating returns on real estate.

COMMON MISCONCEPTIONS ABOUT REAL ESTATE RETURNS

Behavioral finance suggests that some individual errors are predictable and are shared by a large group of market participants. Such established common errors in the financial markets include anchoring, representativeness, availability bias, and overconfidence. Anchoring referrers to the reliance on past information and not fully incorporating new information. Stock market underreaction to many events is often explained by this cognitive error. Representativeness is a form of bias in which beliefs heavily rely on stereotypes. Availability bias denotes overreliance on easily available information. Overconfidence bias often translates to investors overestimating returns and underestimating risks. As a result of these cognitive errors, average opinion and asset valuation can be systematically biased (De Bondt et al. [2008]). Systematic cognitive valuation errors are especially severe for real estate because of the limit of arbitrage due to high transaction and information costs as well as the length of the construction cycle.

Shiller [2005, p. 20] argued that one of the reasons so many individuals believed that there had been a continuing upward trend in home prices was because they compared simple return on infrequent real estate transactions.⁷ As an example, assume one sells a property in 2005 for ten times as much as what it had been purchased for in 1945. Simple return from such a transaction is 900%. In addition, the claim of selling for ten times more appears impressive. However, this transaction yields a real annualized return of less than 1%. In contrast, stock market performance is usually discussed in terms of annualized percentage returns. Further, because stock prices are often subject to stock splits, which usually keep prices around the $30-a-share level, it rarely happens that a stock price has increased ten times over a certain period. Nonetheless, it may have exhibited outstanding performance, as evidenced by historical percentage returns. This misunderstanding of return on real estate likely generated representativeness bias and contributed to investors’ overconfidence about real estate in 2008.

Another likely investor error pertaining to the analysis of the rate of return on residential real estate is ignoring the various transaction costs, such as closing costs, real estate taxes, house maintenance, and improvements costs. Prior literature has established investor ignorance of transaction costs for other types of investment. For example, Barber, Odean, and Zhen [2005] showed evidence that investors neglected a fund’s operating expense ratio when purchasing a mutual fund. Operating expenses average about 1% for mutual funds but are much higher for residential real estate, especially in high-tax areas.⁸ Ignoring such expenses would lead to substantial overestimation of the rate of return on real estate, which would contribute to unrealistic assumptions about real estate returns. In fact, for our earlier example of selling a property in 2005 for 10 times as much as what it had been purchased for in 1945, a conservative 1% annual maintenance fee implies a zero real annual percentage rate of return.

An average investor may also conclude that house prices appreciate continuously because he or she may compare prices of smaller, older houses to those of newer, bigger houses, with more amenities. Clearly, a bigger and better built house should have a higher price than a smaller, outdated house. Therefore, if one compares how much his or her house costs relative to what his or her parents paid for a house 40 years ago, one might incorrectly assume that the entire difference in price is due to price appreciation rather than size and various structural differences.

In contrast, an ounce of gold or General Electric Company (GE) common stock remains homogenous through time. Therefore, when one learns that the price of an ounce of gold or the price of GE stock increased by 3%, the interpretation of such information is straightforward. However, learning that a house on the other side of town, or even a house across the street, sold for $300,000 provides only limited information about the value of one’s home. Information about the current state of the real estate market, even along with the data for specific sales transactions, is valuable only in conjunction with information about the differences with one’s home and information about how much those differences are worth.

In order to make inferences regarding house price appreciation, one needs to compare the price of the same house at points t and t + 1. This is not a trivial exercise because real estate is one of the most heterogeneous commodities and is traded infrequently. Reputable publications, such as the Wall Street Journal, the Associated Press, and Forbes devote substantial attention to changes in median house prices.⁹ The median price is the value for which 50% of sales are more expensive and 50% are less expensive. Comparison of median house prices of sold houses can lead to distorted inferences—if characteristics of sold houses change from quarter t to quarter t + 1, then one is evaluating prices without a common basis for comparison. For example, on May 12, 2009, the Wall Street Journal published an article entitled “Median Home Price Falls to $169,000, NAR Reports.” The article discussed that the median U.S. single-family house price in the first quarter fell 14 percentage points from a year earlier. It further added, “sales were up sharply in some of the areas hardest hit by the housing bust, largely because bargain hunters were out in force.” It is plausible that the decline in the median price was mainly driven by an increased number of sales of cheaper houses in the depressed areas, while the prices of comparable houses may have increased.

PROPER WAYS OF COMPUTING RESIDENTIAL REAL ESTATE RETURNS

In this section, we discuss several ways of calculating returns on real estate. Hedonic regression and repeat sales methods allow for constructing house price indices that measure changes in price (capital gain) as a percentage of the market price of the property. We also discuss measures of the rate of return on real estate that account for imputed rental income, the cost of ownership, and tax considerations.

Hedonic Pricing Model

Hedonic regression is a common way of constructing a house price index that accounts for interperiod variation in the quality of housing sold. The hedonic regression method defines an average house and tracks the change in the value of this average house over time (Court [1939], Lancaster [1966], and Rosen [1974]). The hedonic pricing model originates with Court [1939] and has been applied to many durable assets. It assumes that each attribute (q) of the durable asset has its own implicit price (v). In the case of real estate, attributes usually include property and location characteristics. Rosen [1974] derived conditions under which the price of a durable asset (P) is the sum of the implicit market prices times the quantity of attributes. Prior literature generally assumes that the relationship is not linear and includes the log of price as a dependent variable when estimating a hedonic pricing model with ordinary least squares (OLS) as follows:

1

where i indices N individual property sales in the cross-section, each with K hedonic characteristics. The disturbance term e_i, typically assumed to be independently and identically distributed (IID), is the result of negotiations between buyers and sellers who exhibit idiosyncratic preferences.

Sirmans et al. [2006] analyzed 125 hedonic price regressions and found that the physical characteristics of a property were among the most popular hedonic characteristics used, namely, square footage, lot size, age, number of bedrooms and bathrooms, garage, swimming pool, fireplace, and air conditioning. Interestingly, despite the difference in samples, these 125 regressions suggest consistency in the hedonic prices of many of these hedonic characteristics. For example, the elasticity of the hedonic price of interior square footage is generally between 0.0004 and 0.0007.

The hedonic pricing model allows for constructing house price indices across both space (space-to-space indices) and time. For example, if one is interested in comparing house prices in New York, St. Louis, and Denver, one first estimates a hedonic pricing model separately for each city at a given point in time. This approach provides estimates of implicit market prices (vector v), which are then multiplied by the vector of typical characteristics (q). The exponent of the sum of products of v and q is the constant quality price for each city. To construct a house price index, one city is chosen as a base city, with an index of 100, relative to which all other cities are compared.

Malpezzi, Chun, and Green [1998] estimated place-to-place hedonic price indices for 272 owner-occupied U.S. Metropolitan Statistical Areas using the 1990 census. They include 15 hedonic characteristics, such as number of bedrooms, age of structure, length of tenure, indicators for single-family detached, type of sewer, and water. They find that Honolulu has the most expensive owner-occupied housing, while Joplin, Missouri has the lowest.

The hedonic pricing model can easily be used to estimate a house price index for a given housing market over time. To do so, one needs to add dummy variables T_j to Equation (1), which indicates that the sale took place in time period j (where j can represent any time interval, such as a month, quarter, or year):

2

There are N+1 time periods in Regression (2). The intercept term v ₀ represents the normalized period, which can be set to 100; the values of the other period’s indices are derived using b_j relative to v ₀.

Hedonic time price index models assume constant hedonic coefficients for the same location across time (i.e., vector v is constant across time). A potential caveat with hedonic price indices is that the vector of hedonic characteristics q can omit important variables and is thus subject to omitted variable bias. Other potential challenges with hedonic pricing models include the choice of the appropriate functional form and costly data collection.

Repeat Sales Method

The repeat sales method provides an alternative way of constructing a cross-time price index.¹⁰ It tracks changes in houses that sold more than once (Bailey, Muth, and Nourse [1963]). The main advantage of the repeat sales method is that it keeps the quality of the observed house constant by focusing on the sales of the same property at two or more points in time. For example, if a property sells twice at time periods 1 and 2, the log of price for these properties at these two time periods can be represented as follows:

3 4

If the vector of hedonic characteristics q does not change from time period 1 to time period 2, then the change in the log of prices is:

5

where ? is the new error term. Equation (5) is then estimated for each pair of repeat sales and results in the following regression (without observation indices) of the change in the log of price for each pair (Coulson [2010]):

6

where T_j takes the value of -1 (+1) for observations that had their first (second) observation in a given time period. Usually, the first time period is normalized, and the rest of the coefficients are represented relative to it. It is common to set b ₁ = 0. Case and Shiller [1989] argued that house prices follow a random walk because of shocks to house prices in each time period. While these shocks, on average, will not affect the price over time, they will affect the variance, leading to heteroskedasticity in Equation (6). The authors propose to estimate Equation (6) using a weighted or generalized least squares method (WLS or GLS, respectively).

Note that the repeat sales price index shows only percentage change in price. It does not provide information on house price levels or the value of a particular characteristic. Unlike hedonic price indices, the repeat sales index does not suffer from an omitted variable bias, because it examines the change in price of the same property at different points in time. Moreover, it does not require specification of functional form of relation between attributes. Nevertheless, properties that sell frequently might not be representative of the entire population of properties. Clapp and Giaccotto [1992b] showed that properties that sell frequently tended to be lemons and starter homes.¹¹ Another potential caveat with the repeat sales methodology is that it is sometimes unclear whether the change in price is partly attributed to property renovation, and, therefore, whether vector q has changed. Furthermore, the repeat sales index assumes that implicit market prices of hedonic attributes (vector v) remain unchanged, which may not necessarily hold true. Perhaps the biggest drawback of this methodology is that it utilizes only a small percentage of the sample of house sale data because of the requirement that the sale takes place more than once.

While hedonic price models have been largely used in academic studies and by appraisers, widely reported U.S. FHFA and S&P/Case-Shiller indices use the repeat sales methodology. FHFA publishes a quarterly house price index that uses the repeat sales method for refinancing in 363 metropolises. The limitation of this index is that it uses data for only those properties whose mortgages have been sold to Fannie Mae and Freddie Mac (conforming mortgages). Therefore, it omits the sales of more expensive houses (the jumbo mortgage market), the sales of houses to less creditworthy consumers, as well as those that do not provide sufficient documentation to obtain conforming loans. The S&P/Case-Shiller index is released monthly and tracks prices in 20 metropolitan statistical areas.

Using the repeat sales method, Shiller [2005] showed that the inflation-adjusted appreciation in house prices over 116 years starting in 1890 and ending in 2006, when the price index witnessed the highest level, was only 0.61% (i.e., real prices increased 103% over 116 years). The increase in real house prices over 1980–2009 was only 36%, or 0.26% per year. The run-up during the period of the most recent housing boom generated an increase in house prices of 7.07% per year between 1997 and 2006 and 8.22% per year between 2000 and 2006.

RISK AND RETURN RELATION FOR RESIDENTIAL REAL ESTATE

A cornerstone of the traditional asset-pricing theory is that higher risk should be rewarded with higher return. This relationship also holds true for real estate. Specifically, Cannon, Miller, and Pandher [2006] analyzed cross-sectional differences in house prices at zip code level in 155 Metropolitan Statistical Areas and found that housing returns increased by 2.48% annually, for a 10% rise in volatility.

Strikingly, however, and unlike in the case of stocks, real estate prices also reward idiosyncratic risk.¹⁴ ,Cannon, Miller, and Pandher [2006] found that a 10% increase in idiosyncratic volatility raised returns by 1.88% annually. This result perhaps should not be surprising considering that investment in real estate is undiversified.

An analysis of time series of real estate home prices suggests that the market for single-family homes is less efficient than the financial market due to transaction costs, carrying costs, and tax considerations. Case and Shiller [1989] analyzed sales price data for 1970–1986 for Atlanta, Chicago, Dallas, and San Francisco/Oakland. They found that year-to-year changes in prices tended to be followed by changes in the same direction in the subsequent year and that house prices did not incorporate information about real interest rates.

As for other asset classes, standard deviation is also employed as a measure of risk for real estate. Examining the 1952–2005 period, Hasanov and Dacy [2009] compared the rate of return and standard deviation of real estate with the rates of return and standard deviations of seven other asset classes. The authors reported that real estate offered a 10.9% return when computed based on equity value or a 7.3% return when computed based on market value and a 5.2% or 3.4% standard deviation, respectively. In turn, T-bills return 5.1%; corporate bonds, 7.1%; and common stocks (S&P 500), 12.7%, with 2.9%, 9.8%, and 17.1% standard deviations, respectively. Therefore, according to these calculations, real estate offers the best risk–reward ratio. Indeed, it equals 2.09 for real estate (or 2.18 when return is computed based on market value), 1.75 for T-bills, 0.72 for corporate bonds, and 0.74 for common stocks.

The comparison above, however, should be viewed cautiously because the real estate return calculations do not include transaction and various other costs, thereby introducing an upward bias to the computed return measure. Moreover, standard deviation is not a perfect risk measure, especially for real estate. Total risk on investment in real estate should be higher than the standard deviation because of high liquidity risk and because idiosyncratic risk due to location cannot be removed through diversification. These warnings render real estate to be a less attractive asset than might be first interpreted from the above statistics.

CONCLUSION

One of the lessons of the financial crisis of 2008 is that real estate is an important asset class. With about $20 trillion in market value as of 2009, residential real estate is one of the largest asset classes and constitutes a large portion of an average household’s portfolio. Moreover, real estate has a greater wealth effect on consumption and, in turn, on GDP growth, than other financial assets. As a result, the real estate market imposes economy-wide risks, which were displayed as the events of 2008 unraveled.

The concept of real estate return appears to be clouded by various misconceptions. Homeowners often compute simple rather than real annualized return on infrequent real estate transactions and ignore transaction costs and maintenance fees. Compared with financial markets or the stock market in particular, the real estate market is highly heterogeneous and illiquid. Due to these peculiarities, the information in news articles about the state of the real estate market or even specific real estate statistics is applicable to one’s home only to the extent that its features are representative of the reported averages and thus is often subject to misinterpretation. The lack of knowledge regarding real estate basics contributed to the real estate bubble and its dramatic collapse, which caused the financial crisis of 2008.

The hedonic pricing model and repeat sales method are two major ways of proper computation of real estate returns as they account for interperiod variation in the quality of housing sold. While academics and appraisers have used mostly hedonic pricing models, the widely known U.S. FHFA and S&P/Case-Shiller indices are based on repeat sales methodology. Although hedonic pricing and repeat sales methods account only for capital gain in computations, more comprehensive measures that account for rental income and tax advantages are also available. Overall, about 80% of real estate return is attributed to capital gain. Real estate appears to have a high risk–return ratio when risk is measured as standard deviation.

This article provides background knowledge for the concept of risk and return in real estate that is essential given the size of the residential real estate market and its ability to pose economy-wide risk. Investors should carefully consider this information when making asset allocation decisions. Investors should use appropriate measures of risk and return on real estate, adjusted for transaction costs. All transaction costs should be considered: upfront (closing fees), annual (insurance, taxes, maintenance), and back-end (commissions and taxes on gains). Respectively, better understanding of real estate risk and return will help investors’ ability to forecast market conditions and make better investment choices.