34
JRER | Vol. 36 | No. 2 2014 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market Sentiment Authors Changha Jin, Go ¨kc ¸e Soydemir, and Alan Tidwell Abstract We explore the pricing patterns of the residential real estate market in the United States in the context of the recent housing bubble and subsequent deflation. We examine 10 consolidated metropolitan statistical areas and calculate excess residential market return per risk. Then, using an error correction model, we regress excess residential market return per risk on fundamental market risk factors from a range of demand- and supply-side variables together with a non-fundamental-based sentiment variable. Our long-run findings reveal that non- fundamental-based (irrational) consumer sentiment is a significant exogenous variable in the pricing pattern of U.S. residential real estate. Although there have been justifiable discussions pertaining to the impact of consumer sentiment particularly non-fundamental (irrational) sentiment on residential real estate pricing, an empirical study examining the relationship between the residential real estate market and consumer (market) sentiments is absent in the literature. The testing procedures in most previous housing studies are grounded in ‘‘rational’’ theoretical models, which use fundamental economic variables contingent upon the supply and demand side of the housing market when explaining the residential housing market. 1 This approach to explaining market behavior assumes that investors and homebuyers are fully rational and will only make decisions that reflect knowledge of all available fundamental information. Given the increased importance of real estate on consumer consumption and the business cycle post 1980s financial deregulation (Miles, 2009), investigations into the rationality of the pricing patterns of residential real estate are warranted. Case and Shiller (2003) contend that a real estate bubble can be characterized as a state in which housing prices experience rapid appreciation due to excessive public expectations of future prices. As the authors suggest, patterns in fundamentals such as income growth and declining interest rates may account for some but not all of observed price increases in the housing markets as price

The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

  • Upload
    others

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

Page 1: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4

T h e U . S . H o u s i n g M a r k e t a n d t h e P r i c i n g

o f R i s k : F u n d a m e n t a l A n a l y s i s a n d

M a r k e t S e n t i m e n t

A u t h o r s Changha Jin , Gok ce Soydemir, and Alan

Tidwell

A b s t r a c t We explore the pricing patterns of the residential real estatemarket in the United States in the context of the recent housingbubble and subsequent deflation. We examine 10 consolidatedmetropolitan statistical areas and calculate excess residentialmarket return per risk. Then, using an error correction model,we regress excess residential market return per risk onfundamental market risk factors from a range of demand- andsupply-side variables together with a non-fundamental-basedsentiment variable. Our long-run findings reveal that non-fundamental-based (irrational) consumer sentiment is asignificant exogenous variable in the pricing pattern of U.S.residential real estate.

Although there have been justifiable discussions pertaining to the impact ofconsumer sentiment particularly non-fundamental (irrational) sentiment onresidential real estate pricing, an empirical study examining the relationshipbetween the residential real estate market and consumer (market) sentiments isabsent in the literature. The testing procedures in most previous housing studiesare grounded in ‘‘rational’’ theoretical models, which use fundamental economicvariables contingent upon the supply and demand side of the housing market whenexplaining the residential housing market.1 This approach to explaining marketbehavior assumes that investors and homebuyers are fully rational and will onlymake decisions that reflect knowledge of all available fundamental information.Given the increased importance of real estate on consumer consumption and thebusiness cycle post 1980s financial deregulation (Miles, 2009), investigations intothe rationality of the pricing patterns of residential real estate are warranted.

Case and Shiller (2003) contend that a real estate bubble can be characterized asa state in which housing prices experience rapid appreciation due to excessivepublic expectations of future prices. As the authors suggest, patterns infundamentals such as income growth and declining interest rates may account forsome but not all of observed price increases in the housing markets as price

Page 2: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

1 8 8 u J i n , S o y d e m i r , a n d T i d w e l l

increases are likely to be a result of fundamental variables and irrationalexpectations. Wheaton and Nechvey (2008) empirically observe that the increasein housing prices from 1998 to 2005 could not be explained by demandfundamentals, and that MSAs characterized with high levels of speculativepurchasing and subprime mortgage activity had greater forecast errors. Nneji,Brooks, and Ward (2013) find evidence that the housing market in the UnitedStates recently experienced a rational bubble, where buyers nonlinearlyextrapolated historic growth patterns in house prices rather than relying onfundamentals. In this study, we consider consumer sentiment, particularly theportion of sentiment unexplained by fundamental variables, as a possibleexplanatory variable in explaining residential real estate pricing patterns.

Shiller (2007) argues that fundamentals do not explain the recent housing boom.Rather, a psychological theory better explains the phenomenon, since it candescribe a so-called feedback mechanism or social epidemic, formulatingperception on real estate as an investment vehicle. Likewise, behavioral studies(e.g., Diaz, 1990; De Bondt, 1998; Hansz and Diaz, 2001) reveal that participantsin real estate are not fully assumed to be rational, so that they tend to showsystematic biases. In addition, market sentiment is conceived of as expectationsor judgments that are not fully justified by available information on marketfundamentals, and thus beliefs on future market conditions can be misguided,resulting in noise traders who misprice investments against rational expectations(Shiller, 1989; De Long, Shleifer, Summers, and Waldmann, 1990; Baker andWurgler, 2007). We assert that the behavior driven by market sentimentparticularly irrational market sentiment does not have a negligible impact, butrather considerable economic impact on residential markets.

Theoretically this phenomenon can be explained by a behavioral concept referredto as overreaction. In this study, the concept of overreaction posits thathomebuyers respond disproportionately to new information. This causes housingprices to fluctuate more than fundamentals might suggest as homebuyers overreactto fundamental market information; in turn, at any given point the price of a housewill not fully reflect the property’s fundamental value. Despite the relevance ofmarket sentiment to real estate consumers and investors, there is only limitedliterature examining the relationship between market sentiment and the real estatemarket.

Case and Shiller (2003) attribute the excess return exceeding the predicted valuesto an increase in sentiment. This is verified through a survey of home buyersconducted in 2003 on subjects in four metropolitan areas. They find thathomebuyers who recently purchased a home expect home prices to increasebetween 7% to 11% annually, indicating that people generally viewed housinginvestment as an on-going escalator, meaning that this market state will remainas it was in the early 2000s (i.e., status quo bias). This mentality is consistentwith prior periods (e.g., 1980s) of exuberant market sentiment, which eventuallyresulted in price declines. The linkage between this suspected belief and marketconditions can be explained from a behavioral concept known as a status quo: the

Page 3: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

T h e U . S . H o u s i n g M a r k e t a n d t h e P r i c i n g o f R i s k u 1 8 9

J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4

state in which naıve expectations on future events are similar to the way theypresently are (Kahneman, Knetsch, and Thaler, 1991). If mispricing is driven bya deviation from current market fundamentals, then the deviation may be from anaıve expectation that current price increases will remain the same in followingyears.

The research expectation for this study is that market sentiment unexplained byfundamental variables significantly impacts subsequent pricing patterns of the U.S.residential real estate market. This contention suggests that the changes in priceare not sufficiently explained by fundamental variables, but rather are influencedby non-fundamental (irrational) sentiment. We test to see if the market sentimentunexplained by fundamental variables is a significant factor, and suppose that pricechanges are not completely dependent on fundamentals. Support for our researchexpectation is based on the behavioral concepts of overreaction and status quo.

Correspondingly, we make the following contributions to the literature. First, ourstudy is contingent upon the consideration of a behavioral factor, e.g., non-fundamental based (possibly irrational) consumer sentiment, as an importantexogenous variable in analyzing the pricing patterns of residential real estate. Wehypothesize that consumer (market) sentiment unexplained by fundamentalvariables is a significant explanatory factor in future housing prices. If marketmispricing is driven by a deviation in housing prices from what currentfundamental economic and finance conditions would indicate, then the deviationmay be a result of naıve consumer expectations that the contemporaneous pricechanges will reoccur in following years. This expectation or sentiment could leadto artificial (non-fundamental based) pricing patterns. Consistent with Baker andWurgler (2006), we decompose consumer sentiment by regressing consumersentiment on a set of fundamental variables. We then use the residual as thedecomposed and orthoganalized irrational sentiment factor in error correction andOLS regression models examining residential real estate pricing patterns.

This study provides empirical evidence of the impact of non-fundamentalconsumer sentiment on residential real estate markets and also providesinformation on the dynamic relationship between non-fundamental-basedconsumer sentiment and fundamental economic, financial, and real estatevariables. We identify both long-run and short-run models from a set of potentialexplanatory variables. We hypothesize the relationship of the models as a functionthat assumes the deviation from the short-run model will converge to a long-runmodel. Also, both the long- and short-run models provide dynamic models of thepricing of the residential housing markets in recent years. Our findings aregenerally consistent with the conjecture that non-fundamentally-based consumersentiment explains a significant amount of variation in residential housing risk-adjusted and unadjusted returns in the long-run, and that this behavior is dynamicin the short-run.

This paper proceeds as follows. We first describe the literature and then providea description of the data. We next outline the methodology used to test ourhypothesis. We then present empirical results and close with concluding remarks.

Page 4: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

1 9 0 u J i n , S o y d e m i r , a n d T i d w e l l

u L i t e r a t u r e R e v i e w

Recent research on the role of individual and institutional sentiment in the pricingpattern of stocks generally finds significant co-movements between stock marketreturns and sentiment. Sentiment measures include direct measures of investorsentiment (i.e., Conference Board Consumer Confidence Index, the Investors’Intelligence Survey Index, and the University of Michigan Consumer SentimentIndex) and indirect measures (i.e., closed-end fund discounts, mutual fund flow-based measures, trading activity-based measures, derivative variables, and IPOrelated variables).

Closed-end fund discounts were one of the first measures modeled as proxies ofinvestor sentiment. Lee, Shleifer, and Thaler (1991), Chopra, Lee, Schleifer, andThaler (1993), Swaminathan (1996), and Neal and Wheatley (1998) argued thatthe discount (or premium) between net asset value (NAV) and the price closed-end fund shares were trading could be treated as an indicator of investor sentiment,where optimistic sentiment results in a price premium and conversely pessimisticsentiment results in shares selling at a discount to NAV. However, authors of morerecent studies question the validity of the closed-end fund discount as measure ofsentiment (e.g., Brown and Cliff, 2004; Baker and Wurgler, 2006; Lemmon andPortniaguina, 2006; Qiu and Welch, 2006; Canbas and Kandir, 2009).

Brown, Goetzmann, Hiraki, Shiraishi, and Watanabe (2003) and Randall, Suk andTully (2003) contend that mutual fund flows may be a proxy for investor sentimentin the stock market. Frazzini and Lamont (2008) find that mutual fund flowsshould be used as contrarian tools, with bullish investor sentiment evidenced byhigh fund flows capable of predicting low future returns. The authors attributethese findings to the idea that inflows into mutual funds push prices higher in theshort-run, as described by Wermers (2004) and Coval and Stafford (2007).

Trading activity-based measures have also been shown to proxy investorsentiment. Kumar and Lee (2006) find that the volume of retail investor trades asa proxy for retail investor sentiment and often plays a role in the pricing of stocks.Specifically, the authors find that retail trades are correlated across individuals,and that the summation of these individual retail trades, which are partiallymotivated by sentiment, is sufficient enough to influence stock returns. Similarly,Barber, Odean, and Zhu (2009) find that retail investors herd by systematicallypurchasing stocks based on strong recent pricing performance and high tradingvolume. Amihud (2002) and Jones and Lamont (2002) also use trading volumeas a measure of investor sentiment to illustrate that an increase in sentiment(trading volume) is associated with lower future stock returns.

In addition to the commonly applied indirect sentiment measures previouslydiscussed, there are several alternative indirect measures linking investor sentimentto derivative variables and IPO related variables. Dennis and Mayhew (2002) usethe put-call ratio as an indicator of sentiment. A high ratio indicating a largetrading volume in puts is indicative of bearish investor sentiment. Whaley (2009)

Page 5: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

T h e U . S . H o u s i n g M a r k e t a n d t h e P r i c i n g o f R i s k u 1 9 1

J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4

suggests that the market volatility index (VIX) generated from the daily spreadbetween options traded at the CBOE is a good indicator of sentiment. Kaplanskiand Levey (2009) find that the VIX may contain an irrational component of risk,which is inversely related to stock returns. Baker and Wurgler (2000, 2006) useIPO activity as a proxy for investor sentiment.

Lastly, and most relevant to this study, one of the most common sentiment proxiesare direct survey-based measures to gauge sentiment. These surveys typicallyattempt to measure either consumer, individual or institutional sentiment. Theliterature is rich with studies documenting the efficacy of direct sentiment-basedmeasures. Brown and Cliff (2004, 2005) and Kalotay, Gray, and Sin (2007) usethe Conference Board Consumer Confidence Index, the Investors’ IntelligenceSurvey Index, and the University of Michigan Consumer Sentiment Index(UMichCCI) as proxies of U.S. investor sentiment. Charoenrook (2003) and Chen(2008) contend that the inclusion of survey-based consumer confidence measuresenhance market forecast. Additionally, Qiu and Welch (2006) show that theUMichCCI has a strong positive relationship with investor sentiment (UBS/GallupSurvey of Investor Sentiments) but the closed-end fund discount did not exhibita significant relationship with investor sentiment measures.

A recent study by Verma, Baklaci, and Soydemir (2008) builds upon the existingliterature by dividing investor sentiment into rational and irrational components.They then examine the unique but simultaneous impact of both rational andirrational sentiment on U.S. stock market returns. They find that the rationalcomponent of investor sentiment has a larger impact on future stock market returnsthan the irrational component. However, irrational sentiment has a greaterimmediate positive impact on stock market returns, which, however, is correctedby negative responses in upcoming periods. These results provide evidence thatin a publicly traded securities market, periods of high irrational sentiment are oftenfollowed by low returns as market prices revert to intrinsic values. However,Hirschleifer, Subrahmanyam, and Titman (2006) provide a model in which‘‘irrational’’ (non-fundamental-based) investors can in some cases earn greaterrisk-adjusted returns than informed rational investors.

In the private real estate markets, typically characterized by relative marketinefficiencies when compared to the public security markets, one might expect therole of sentiment and particularly irrational sentiment to have a larger impact onpricing. Stein (1995) and Lamont and Stein (1999) find that collateral constraintsin the housing market, through the use of leverage, can lead to price changes thatare greater than fundamental market changes suggest. The use of leverage has asubstantial impact on market pricing because in the housing market the stabilizingeffects of arbitrage traders are lacking. Also, the ability of market participants toborrow is related to the value of the assets, so an increase (decrease) in assetvalues can result in an increase (decrease) in demand. Lamont and Stein (1999)refer to this notion as ‘‘key’’ to the amplifying effect of observed housing pricescompared to fundamental price expectations. Variations of this theme have beenexamined in a variety of markets including corporate asset sales (Shleifer andVishny, 1992) and the stock market (Garbade, 1992).

Page 6: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

1 9 2 u J i n , S o y d e m i r , a n d T i d w e l l

In addition to the constraining influence of leverage on the housing markets, Miller(1977) suggests that asset overpricing may increase with valuation dispersion, andDaniel, Hirschleifer, and Subrahmanyam (1998) contend that investors overact toprivately gathered information. Chen, Hong, and Stein (2002) and Diether, Malloy,and Scherbina (2002) examine stock returns and find supportive evidence ofMiller’s (1977) contention. Hong and Stein (1999) examine overreaction,underreaction, and momentum trading and find that as information diffuses slowlyacross market participants, prices underreact in the short run, but overreact in thelong run as momentum traders enter the market. Daniel, Hirschleifer, andSubrahmanyam (1998) find that investors are overly confident about their abilityto generate private information and thus overreact to private information signalsand underreact to publicly available information. In the real estate context,momentum traders may be thought of as speculators; additionally, valuationambiguity and privately gathered information is more prevalent in the housingmarkets than in more efficient securities markets due to heterogeneity and liquidityconstraints, often leading to greater opportunities for speculation and mispricing.

In the real estate domain there has been limited research on the role of sentimentand the pricing pattern of real estate. Barkham and Ward (1999) find that sentimentand United Kingdom’s property company’s discounts to NAV are inversely related.As sentiment increases (decreases), discounts decline (increase). Gallimore andGray (2002) examine the role of investor sentiment in the U.K.’s commercialmarket using a direct survey approach. They conducted a survey with activelyinvolved commercial real estate investors who are members of the U.K. investmentproperty forum and find that U.K. commercial property investors consider investorsentiment to be an important factor when analyzing property. The lack oftransparency and informational asymmetry in the private commercial real estatemarkets results in investors having incomplete (in some cases incorrect)information, which may compromise investment decision models. Therefore,property investors turn to non-fundamental indirect signals in the form of marketand or investor sentiment when formulating their investment decisions.

Clayton, Ling, and Naranjo (2009) examined the role of fundamentals and investorsentiment in determining capitalization rates, revealing a dynamic variation incapitalization rates across time. In the short-run, using an error correctionprocedure with fundamental control variables, the authors find a positiverelationship in investor sentiment and subsequent quarter returns. However, in thelong-run, sentiment-induced mispricing are eventually followed by price reversal.The findings document a distinction between the public and private commercialreal estate markets, with the private markets susceptible to a more prolongedsentiment-induced mispricing, presumably due to market inefficiencies. In thecontext of real estate investment trusts (REITs), Lin, Rahman, and Young (2009)examine the asset-pricing factors of REITs and find investor sentiment to be asignificant contemporaneous factor. The authors use changes in fund discountsbased on NAV as the proxy for sentiment, contending that when investors are

Page 7: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

T h e U . S . H o u s i n g M a r k e t a n d t h e P r i c i n g o f R i s k u 1 9 3

J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4

optimistic (pessimistic) the discount should be smaller (larger). They find that thesize of the discount is a significant variable in the pricing patterns of REITs. Morespecifically, a large (small) discount results in lower (higher) contemporaneousprice returns, thus indicating investor sentiment should be included in REITpricing models.

In the context of residential housing prices, in a series of studies Clayton (1996,1997, 1998) investigates the efficiency of both the single-family and multi-familycondominium residential real estate markets using data collected from the city ofVancouver, British Columbia. He finds that models of fundamental variables failedto adequately explain observed house price dynamics. These studies provideevidence that residential price movements are predictable and not always rational.Periods of sharply increasing values signaled a looming correction in which pricesreverted to values supported by market fundamentals. The author attributed thesesharp run-ups in housing prices to ‘‘irrational expectations.’’

More recently, Case and Shiller (2003) investigate whether the U.S. residentialreal estate markets were experiencing a market bubble. They empirically analyzedresidential housing market prices at the state level and a vector of fundamentaleconomic variables, with per capital income being the most influential, during the1985 to 2002 time period. They find fundamental variables only partially explainthe recent increase in housing prices in the years after 2000. More specifically,the fundamental model under forecasts home prices in 2000 and 2002. The authorsattribute the excess return exceeding the predicted values to an increase insentiment. This is verified through a survey conducted in 2003 on subjects in fourmetropolitan areas (Los Angeles, San Francisco, Boston, and Milwaukee) whopurchased houses in 2002. They find that homebuyers who recently purchased ahome in 2002 expect home prices to increase 7%–11% annually. At that time,people generally viewed housing investment as an escalator, meaning that if theydo not buy now, they will not be able to buy later. This mentality is consistentwith prior periods (e.g., 1980s) of exuberant market sentiment, which eventuallyresulted in price declines.

Archer and Smith (2010) construct an integrated model of residential mortgagedefault risk and empirically test the model based on mortgages originated in 20Florida counties from 2001 to 2008. They suggest that lenders’ and borrowers’interpretation of risk is subject to ‘‘euphoria’’ generated by past price appreciation.The authors recommend that because residential real estate pricing is subject toeuphoric behavior, underwriting standards should be standardized and consistentacross varying market conditions.

The results from the extant literature in real estate and elsewhere provide evidencethat lenders, borrowers, and investors are subject to sentiment-based behavioralinfluences. And, as a result studies examining the pricing patterns of various assetsshould consider sentiment as a potential exogenous variable. While recognizingthat the proxy for sentiment may vary depending on the purpose of the study, the

Page 8: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

1 9 4 u J i n , S o y d e m i r , a n d T i d w e l l

Exhibi t 1 u Descriptive Statistics for Annual Home Price Change

City Mean Median Max Min Std. Dev. Skewness Kurtosis

Boston 0.07 0.09 0.17 20.08 0.07 20.73 2.10

Chicago 0.05 0.07 0.10 20.15 0.06 21.90 5.69

Denver 0.05 0.04 0.14 20.06 0.06 0.07 2.16

Los Angeles 0.08 0.11 0.29 20.33 0.14 21.59 5.10

Las Vegas 0.05 0.05 0.43 20.40 0.16 20.49 4.70

Miami 0.07 0.10 0.28 20.34 0.15 21.38 4.66

New York 0.08 0.11 0.16 20.10 0.07 21.22 3.23

San Diego 0.08 0.12 0.29 20.31 0.14 21.33 4.14

San Francisco 0.07 0.11 0.27 20.37 0.14 21.42 4.92

Washington D.C. 0.07 0.10 0.24 20.22 0.11 20.86 3.18

Note: December 1998 as beginning and December 2008 as an ending period.

literature offers that direct measures of sentiment, especially those based on theUMichCCI, have a high degree of fidelity to actual sentiment. The present studyparcels the effect of sentiment into rational sentiment based on fundamentals andirrational sentiment; therefore, we are able to examine the unique effect ofirrational sentiment on the residential real estate markets. This allows us toempirically examine the impact of sentiment and explain whether the causal effectof sentiment on housing market returns can be attributed entirely to rational riskfactors, noise, or to a mixture of both.

u D a t a

H o m e P r i c e D a t a

This study employs the 10-City Standard and Poor’s Case-Shiller Home PriceIndices as a proxy of residential housing market prices for the January 1998 toDecember 2008 time period. The indices cover housing markets in 10 metropolitanregions across the United States2: Boston, Chicago, Denver, Las Vegas, LosAngeles, Miami, New York, San Diego, San Francisco, and Washington D.C. Theindices are calculated monthly, using a repeat sales methodology employing athree-month moving average of paired sales, and is published monthly.

Exhibit 1 contains descriptive statistics on the annual return for the respectivehousing markets for the 10 MSAs in this study. The annual returns indicate that

Page 9: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

T h e U . S . H o u s i n g M a r k e t a n d t h e P r i c i n g o f R i s k u 1 9 5

J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4

Los Angeles, San Diego, and New York experienced the highest average (median)annual returns while Denver, Chicago, and Las Vegas exhibited the smallestreturns. Las Vegas had the highest annual increase in home prices, along with thehighest decrease of housing prices. To control for differences in volatility betweencities, we divided excess housing returns by a city-specific standard deviation tostandardize regional home price changes on a risk-adjusted basis.

E x c e s s H o u s i n g M a r k e t R e t u r n p e r R i s k

We calculated excess housing market return per risk (MPR), , a SharpeMPRR i,t

measure with a specification in equation 1:

r 2 ri,t fMPRR 5 . (1)i,t si,t

First, we calculated annual home price change (ri,t) by subtracting the annual homeprice in time t-12 from the annual home price at time t for each regional marketi at time t. Second, we calculated a moving average standard deviation (s2) forthe period t-1 to t-12. Then we subtracted the annual risk-free rate from eachregional market return at each month t. The MPR for each region can be definedby dividing the excess return with the moving average standard deviation attime t.

Exhibit 2 shows the relationship between housing market MPR, U.S. average GDPexcess return per risk, and market sentiment. We derived the U.S. average GDPexcess return per risk similarly to MPR, as shown in equation 1. The housingmarket had a positive MPR from 1998 to 2005, while GDP excess return remainedgenerally lower than the housing market, even exhibiting negative adjusted returnsfrom late 2002 to the end of 2003, declining again in January 2008. Interestingly,GDP excess return per risk exhibited negative growth from 2000 to 2003; however,the U.S. average housing market experienced positive growth in that same period.The peaks in housing market excess return occurred around September 1999 andJuly 2005.

Consumer sentiment remained relatively level between 110 and 120 prior to 2000and although it fluctuated from 2003 to 2007, sentiment generally was increasing.Consumer sentiment is our primary variable of interest, as we utilized theConference Board Consumer Sentiment Index as a proxy for market sentiment.The Conference Board publishes this composite index as a representative sampleof U.S. households. The confidence index provides a measurement of consumerperception on market conditions such as business conditions, employment,inflation rate, purchasing willingness, interest rate, and expectation of stock price.We consider, from a structural perspective, the explanatory power of current

Page 10: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

1 9 6 u J i n , S o y d e m i r , a n d T i d w e l l

Exhibi t 2 u Market Sentiment and Housing Market Excess Return

Notes: The study includes monthly observations from January 1998 to December 2008. The S&P/Case-ShillerHome Price Index is used as a proxy for U.S. home prices at the national level (10 CMSAs). The GDP wasderived from the growth in each state income where these 10 CMSAs are located. The 3-month T-bill was usedas a proxy for the risk-free rate. The excessive return per risk for GDP and housing markets is similar to Sharperatios where excessive return is determined by subtracting the risk-free rate return at time t and moving averagestandard deviation for the period t -1 and t -12. We divided the excess return by the moving average standarddeviation at time t to derive excessive market return per risk for both GDP and housing market, respectively.

consumer market sentiment on future housing returns. The Conference Boardprovides nine regional confidence indexes; so, this study adopts the confidenceindex of the region closest to the city being analyzed as to increase the fidelityof measuring market sentiment.

N o n - f u n d a m e n t a l M a r k e t S e n t i m e n t

The volatility in consumer confidence reflects changes in market fundamentals, aswell as more expectations that may or may not be perfectly rational based givenmarket fundamentals. So, we can think of the consumer’s beliefs as reflecting theconsensus that can be explained by fundamental variables as rational marketsentiment with the remaining variance attributed to irrational market sentiment.Using an approach similar to Baker and Wurgler (2006), we attempt to decompose

Page 11: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

T h e U . S . H o u s i n g M a r k e t a n d t h e P r i c i n g o f R i s k u 1 9 7

J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4

market sentiment (Senttt) into two components, a rational market expectation andan irrational (non-fundamental) market expectation:

n

Sentt 5 g 1 r Fund 1 j , (2)Ot 0 j jt tj51

where g0 is constant and gj are the parameters to be estimated. The term jt is therandom error term, and Senttt represents the relative changes in sentiment onmarket. This sentiment can be explained by Fundjt , a vector of fundamentalvariables that represent rational expectations based on market risk factors.Therefore, the predicted value in equation 2 captures the rational component ofsentiment, while the residual, jt , captures the irrational component of sentiment.We adopt the residual as a proxy for the irrational market component used in ouranalysis.

F u n d a m e n t a l Va r i a b l e s

In our analysis we include both the 1-year adjustable-rate mortgage (ARM)interest rate and the 30-year mortgage interest rate. We also calculate the spreadbetween the 30-year mortgage interest rate and 1-year ARM; both interest rateswere obtained from Fannie Mae. The spread serves as a measure for the relativeattractiveness for ARMs with a lower ‘‘teaser’’ rate and can be viewed as a proxyfor many alternative loans (e.g., subprime loans and option ARMs).

In Exhibit 3, 30-year fixed-rate mortgages (FRMs) and 1-year ARMs are depicted,along with the percentage of ARM market share. We obtained both sets of datafrom the Federal Home Loan Mortgage Corporation (Freddie Mac). The 30-yearFRM began at 7% and increased to 8.5% in early 2000, prior to settling in the5% to 7% range from 2000 to 2008. ARM rates followed a similar pattern before2001 but diverged substantially, making them more attractive, beginning inDecember 2000. The percentage of ARM market share peaked from 2004 to 2006.However, the interest rate gap between FRM and ARM rates began to close in2005, when ARM rates increased sharply. The spread was largest from 2003 to2004. By mid-2006, just before MPR became negative, the spread was the lowestin our analysis.

We used the Engineering News Record’s national Construction Cost Index tomeasure change in construction costs. The construction costs include material andlabor costs for construction. The data covers 20 major cities across the U.S. Weinclude state income from the Bureau of Economic Analysis and CMSA monthlyunemployment rates from the Bureau of Labor Statistics. Additionally, extendedmodels include two national economic variables: term structure (spread between10-year Treasury notes and 3-month T-bill rates) and default risk (spread between

Page 12: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

1 9 8 u J i n , S o y d e m i r , a n d T i d w e l l

Exhibi t 3 u Mortgage Interest Rates and Percentage of ARM Share

0

10

20

30

40

50

60

70

80

90

100

0

1

2

3

4

5

6

7

8

9

Percentage of ARM Share 30-Year FRM Rate 1-Year ARM Rate

Inte

rest

Rat

e (%

)

Percen

tage

of

AR

MS

hare

(19

98

:1=

10

)

Note: The study includes monthly observations from January 1998 to December 2008. The mortgage rate hasbeen derived from Fannie Mae.

AAA-rated bonds and BAA-rated bonds). We also include three new real estateregional variables: market volatility, building permits, and trading volume.

u M e t h o d o l o g y

In this study, we adopted error correction models (ECMs) to identify both thelong-run relationships and the short-run adjustment process in which the modelshows how the long-run relationship is achieved through error correction (Harvey,1990). We followed the Engle-Granger two-step method, in which we specify along-run model using variables in a level series and a short-run adjustment modelusing the first difference of the variables. Our methodology is similar toHendershott and MacGregor (2005) and Clayton, Ling, and Naranjo (2009).Regression of time series data may produce spurious results if the explanatoryvariables are nonstationary. Under certain conditions, we may expect nonspuriousregression results of level series, if modeled with a long-run error correction term.These conditions are satisfied if the variables are integrated to the order of I(1)or first-difference stationary, and have a cointegrating relationship. In order toinvestigate our posited relationship, we applied the cointegration test suggestedby Johansen (1991).

Page 13: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

T h e U . S . H o u s i n g M a r k e t a n d t h e P r i c i n g o f R i s k u 1 9 9

J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4

S t r u c t u r a l M o d e l i n g f o r L o n g - r u n a n d S h o r t - r u n M o d e l s

We adopt the Engle-Granger two-step method in which a long-run model isspecified in levels, and a short-run adjustment model is specified in firstdifferences, but includes a long-run error correction term derived from theestimation of the long-run equilibrium model. In the first stage, a long-run modelcan be specified in levels as:

n

y 5 b 1 bX 1 v , (3)Ot 0 it ti51

where yt is the dependent research variable and Xit are theoretical-based researchvariables i at time t. From this regression, residuals can be estimated as thedifference between the actual and estimated equilibrium between testing factors.If the residuals from equation 3 are stationary, they can be used as an errorcorrection term in the short-run model as follows:

n

Dy 5 a 1 a DX 2 g v 1 « , (4)Ot 0 i it t21 ti51

where Dyt 5 yt 2 yt21 is the first difference of the dependent variable, residentialhome prices, DXit , are the first differences of the explanatory variables, and vt21

is the error correction term (lagged residuals from the long-run regression).Estimation of equation 4 provides evidence of a short-run residential marketdynamic and adjustments to the previous disequilibrium in the long-run relation,g (the speed of adjustment parameter). If g 5 1, then there is full adjustment,while g 5 0 indicates no adjustment. A more general specification of the short-run model can also include multiple lags of the explanatory and dependentvariables.

u E m p i r i c a l R e s u l t s

A unit root test was conducted as the first stage in investigating the relationshipbetween the variables. The unit root results are reported in Exhibit 4.

We test for stationary using the Augmented Dickey-Fuller (ADF) procedure,which tests for the presence of a unit root (Exhibit 4). The hypothesis test showsevidence of ‘‘nonstationary’’ in the level series, thus we could not reject the nullof nonstationary. When examining the first difference in the data series, there is

Page 14: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

2 0 0 u J i n , S o y d e m i r , a n d T i d w e l l

Exhibi t 4 u Unit Roots Test

Level First Difference

Variables T-Stat. P -Value T-Stat. P -Value

ERR (Housing Market Return) 21.624 0.467 23.690 0.005

FRM Rate 21.136 0.701 27.917 0.001

ARM rate 21.213 0.667 26.703 0.001

GDP Growth 22.176 0.216 22.456 0.015

ARM Share 21.624 0.467 23.690 0.005

Construction Cost 22.170 0.218 28.479 0.000

Rate Spread (FRM-ARM) 21.418 0.571 24.514 0.003

Consumer Sentiment 0.044 0.960 26.627 0.001

Notes: We adopt the augmented Dickey-Fuller test statistic (ADF). We only report national leveldata where we generate an equally-weighted state or CMSA average to create national level data(Market Excessive Return Per Housing and Market Sentiment Index).*Significant at the 10% level.**Significant at the 5% level.

strong evidence that all fundamental variables modeled are stationary, thereforerejecting the null of nonstationarity. This result confirms that the research variablesmay have a cointegrated, nonspurious long-run relationship (Exhibit 5).

Since we cannot reject the hypothesis of the number of cointegrating vectors atr # 2, there can be at most two integrating vectors (i.e., r # 2).

L o n g - r u n F u n d a m e n t a l M o d e l

As outlined previously, we can specify a long-run model by the Engle-Grangertwo-step method in levels, since the data series are cointegrated. The long-runequilibrium relationship is embedded into the specification so that it restricts thelong-run behavior of the endogenous variables to converge to their cointegratingrelationships, while allowing for short-run adjustment dynamics. The cointegrationterm is known as the error correction term since the deviation from long-runequilibrium is corrected gradually through a series of partial short-run adjustments.One condition for long-run equilibrium is that the relationship between excessreturn, fundamental variables, and market sentiment is not changing.The fundamental long-run model with irrational sentiment is specified asfollows:

Page 15: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

T h e U . S . H o u s i n g M a r k e t a n d t h e P r i c i n g o f R i s k u 2 0 1

J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4

Exhibi t 5 u Johansen’s Cointegration Test Fundamental Model

# of Cointegrating Vectors

(r ) at Null Hypothesis Eigenvalue Trace Statistics Critical Value at 5% P -Value

r 5 0 0.29 116.93 95.75 0.00

r # 1 0.23 73.14 69.81 0.02

r # 2 0.13 39.31 47.85 0.24

r # 3 0.08 20.17 29.79 0.41

r # 4 0.03 8.34 15.49 0.42

r # 5 0.02 3.42 3.84 0.64

Notes: Trace test indicates two cointegrating eigenvalues. The parameter estimates are from theJohansen’s test using monthly data over the December 1998 to December 2008 time period. Thevariables are excess housing return per risk, U.S. average state income, FRM rate, rate spreadbetween ARM and FRM, unemployment rate, and construction cost.

MPRR 5 b 1 b FRM 1 b CC 1 b Spreadi,t 0 1 t 2 i,t 3 t,FRM ARM

1 b GDP 1 b Unempl 1 b Irr Sentt 1 v , (5)4 i,t 6 i,t 7 i,t t

where is the excess housing market return in monthly home prices; FRMt isMPRR i,t

the 30-year FRM rate; is the natural logarithms of construction cost;CCi,t

is a rate spread between the 30-year FRM and the 1-year ARMSpreadt, FRM ARM

rate, corresponding to the measure ‘‘relative attractiveness for variable interestloans’’; GDPi,t , is a growth rate on state income; Unempli,t is the unemploymentrate in each region; and Irr Sentti,t represents irrational market sentiment and isthe orthogonalized market sentiment residual, jt , from equation 2. b0 is a constant,and vt is the error term. Exhibit 6 contains these fundamental variables.

In Model 1, we find that the estimated coefficient on the interestSpread ,t,FRM ARM

rate spread between a 30-year FRM and a 1-year ARM, was positive andstatistically significant in nine of ten regions. The parameter estimate for the FRMrate was negative and statistically significant in four of the ten cities studied. Theestimated coefficient on construction costs, CCi,t , was negative and statisticallysignificant in all ten regions. The estimated coefficient on GDPi,t , the state income,was positive and statistically significant in Chicago and Denver, and negative andsignificant in four regions, all of which were regions where home prices hadsharply increased while state income remained stable or decreased. Otherwise,state income marginally increased after excess return on the housing market

Page 16: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

20

2u

Ji

n,

So

yd

em

ir

,a

nd

Ti

dw

el

l

Exhibi t 6 u Model 1: Long-run Residential Market Excess Return Level Models with Market Sentiment

Boston Chicago Denver Las Vegas LA Miami New York San Diego San Francisco Wash. D.C.

Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff.

Constant 4.43*** 1.32** 4.82*** 0.71** 3.69*** 1.57** 6.50*** 3.53*** 2.36*** 1.39**(6.55) (2.38) (7.50) (2.37) (4.89) (2.05) (7.34) (7.22) (5.22) (2.35)

FRM Rate 20.73 21.80** 2.10** 20.23 21.45 23.58*** 23.34** 20.84 22.81 21.04*(20.21) (22.09) (2.07) (20.49) (21.29) (24.16) (22.70) (21.22) (4.03) (21.18)

Construction Cost 20.49*** 20.14** 20.55*** 20.07** 20.38*** 20.14** 20.65*** 20.39*** 20.23*** 20.15*(26.52) (22.49) (27.93) (22.33) (24.91) (21.90) (27.24) (27.52) (24.86) (22.29)

ARM Spread (FRM-ARM) 2.54** 4.24*** 4.45*** 2.21*** 3.95*** 8.92*** 20.01 2.59*** 3.36** 7.48***(2.15) (3.48) (3.13) (3.83) (2.71) (7.06) (20.02) (2.95) (3.26) (5.84)

State Income 0.02 0.73** 0.38** 20.39** 20.60** 0.26 21.08*** 0.26 0.21 20.71***(20.19) (2.45) (2.21) (24.07) (23.21) (0.87) (24.02) (0.01) (1.49) (24.41)

CMSA Unemployment Rate 21.29 0.73 0.14 20.99*** 22.32*** 22.02** 27.38*** 20.03 23.26*** 21.82(21.27) (0.82) (0.17) (23.12) (23.08) (22.25) (25.61) (20.05) (25.33) (21.25)

Ortho. Sentiment (Lag) 20.01 0.14*** 20.07** 20.07*** 20.29*** 20.10*** 20.13** 20.05** 20.05** 0.13***(0.10) (4.65) (21.87) (23.76) (26.24) (2.19) (22.36) (22.01) (22.12) (4.36)

Adj. R2 0.68 0.35 0.75 0.41 0.35 0.59 0.44 0.65 0.62 0.55

Notes: T-statistics are in parentheses. The parameter estimates are from the equation 10. Using monthly data over the January 1998 to December 2008 timeperiod, the dependent variable is the monthly excessive return per risk in 10 CMSA housing markets. The independent variables are monthly ARM rate,percentage of ARM shares, State GDP Growth, Construction Cost, and Rate Spread between FRM and ARM. Although we have not reported in the resultssection, Johanson conintegration test shows that fundamental variables and irrational market sentiment has two cointegrating vectors at the 5% significantlevel. We used state income data and the unemployment rate is level of CMSA obtained from census. We use 9-month lag in sentiment index. We alsoestimate the model with 1, 3, and 6 lags for orthogonalized consumer sentiment and found 9-month lag has been confirmed by SBC criterion. We alsoconfirm that other different lags provide a similar coefficient with 9-month lag.*Significant at the 10% level.**Significant at the 5% level.***Significant at the 1% level.

Page 17: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

T h e U . S . H o u s i n g M a r k e t a n d t h e P r i c i n g o f R i s k u 2 0 3

J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4

significantly decreased. The estimated coefficient on region unemployment rate,Unempli,t , was negative and statistically significant in five regions: Las Vegas, LosAngeles, Miami, New York, and San Francisco. The extent of relative changes inFRM rate, construction costs, and unemployment rates is inversely related toexcess housing return per risk.

The orthogonalized (irrational) market sentiment used in this analysis representslagged market sentiment not supported by market fundamentals. Exhibit 6 presentsthe results with a 9-month lag in the irrational sentiment component.3 Theparameter estimate on orthogonalized irrational sentiment is negative andsignificant in 7 of the 10 regions. The negative irrational (orthogonalized)sentiment coefficients observed in the markets indicates that an increase inirrational market sentiment resulted in a future decrease in excess return per risk,implicating that the historic irrational sentiment component is an additionalexplanatory factor in predicting contemporary excess return per risk at the MSAlevel. The adjusted R2 for the long-run model ranged from 35% to 75% acrossregions.4

As a matter of practical consideration, we modify our model to examine housingprice returns instead of excess risk-adjusted returns. In general, the findings forthe long-run model as well as the short run error correction model are robustacross MSAs when substituting the risk-adjusted housing returns with unadjustedhousing returns.

S h o r t - r u n E r r o r C o r r e c t i o n

Error correction models (ECM) use a combination of first differenced and laggedlevels of cointegrated variables. Here, the ECM is specified as follows:

MPRDR 5 b 1 Db FRM 1 b DCCi,t 0 1 t 2 i,t

1 b DSpread 1 b DGDPi,t3 t,FRM ARM 4

1 b DUnempl 1 b DIrr Sentt 1 gv , (6)6 i,t 7 i,t t21

where, D represents the first difference of each research factor and g is the speedadjustment parameter. The change in excess housing return per risk is associatedwith the change in fundamental variables and orthogonalized market sentiment,but it also acts with regard to adjust or ‘‘correct’’ for any disequilibrium observedin the previous period.

Exhibit 7 presents the results for the short-run model from December 1998 toDecember 2008. In general, g in equation 6 describes the speed of adjustment tothe long-run fundamental model. In our research, the change in excess return isassociated with prior changes in fundamental variables and unexplained market

Page 18: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

20

4u

Ji

n,

So

yd

em

ir

,a

nd

Ti

dw

el

l

Exhibi t 7 u Model 1: Short-run Residential Market Excess Return Level Models with Orthogonalized Sentiment Effect

Boston Chicago Denver Las Vegas LA Miami New York San Diego San Francisco Wash. D.C.

Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff.

Constant 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00(20.33) (20.48) (0.03) (20.21) (0.03) (20.18) (20.19) (0.09) (20.82) (0.05)

DFRM Rate 1.87** 0.28 0.47 0.46 2.25 22.99** 0.47 23.52*** 20.52 21.77*(1.67) (0.34) (0.31) (0.88) (0.96) (22.03) (20.23) (23.44) (20.60) (21.74)

DConstruction Cost 20.11 0.01 20.57 0.01 20.62 0.36 20.24 20.06 0.05 20.21(20.39) (0.05) (21.33) (0.14) (0.98) (0.89) (20.44) (22.04) (0.23) (20.77)

DARM Spread 22.40** 0.03 1.29 1.14* 21.83 4.43** 0.95 5.33*** 20.19 2.40*(21.66) (0.03) (0.53) (1.61) (20.57) (2.19) (0.34) (3.73) (20.16) (1.75)

DState Income 20.43** 0.05** 0.49* 0.13 20.30 0.96** 0.03 0.05** 20.26* 20.14(22.09) (1.85) (0.16) (1.18) (20.63) (2.03) (0.10) (2.23) (21.46) (20.83)

DCMSA Unemployment Rate 20.01 20.00 20.03 0.00 20.01 20.05 20.03 20.01 20.02 0.01(21.28) (20.25) (1.10) (0.38) (20.33) (21.53) (20.68) (20.71) (21.92) (0.18)

Error Correction Term 20.09*** 20.02 20.10** 20.02 20.09** 20.08** 20.06** 20.07** 20.06** 20.02*(23.04) (20.99) (2.31) (21.09) (22.00) (22.22) (21.69) (22.01) (21.91) (21.71)

Page 19: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

Th

eU

.S

.H

ou

si

ng

Ma

rk

et

an

dt

he

Pr

ic

in

go

fR

is

ku

20

5

JR

ER

uV

ol

.3

6u

No

.2

–2

01

4

Exhibi t 7 u (continued)

Model 1: Short-run Residential Market Excess Return Level Models with Orthogonalized Sentiment Effect

Boston Chicago Denver Las Vegas LA Miami New York San Diego San Francisco Wash. D.C.

Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff.

DOrtho. Sentiment (Lag) 0.001 0.02* 20.03* 0.01 0.01 0.00 0.02 20.02* 0.02 20.01(0.67) (1.62) (21.70) (0.12) (0.83) (0.08) (0.87) (1.78) (0.24) (20.24)

Adj. R2 0.12 0.04 0.11 0.11 0.06 0.11 0.05 0.20 0.10 0.06

Notes: T-statistics are in parentheses. We use the data over the January 1998 to December 2008 time period. The dependent variable is the monthlyexcessive return per risk in 10 CMSA Housing Markets. The independent variables are change (D) in monthly ARM rate, State GDP Growth, ConstructionCost, and Rate Spread between FRM and ARM. We adopt state income data and the unemployment rate is level of CMSA. Orthogonalized Sentiment is theresidual calculated from regressing fundamental variables on market sentiment index. In order to match each CMSA cities with regional sentiment measure,we use U.S. Consumer Confidence Index, which provides the ten standard Federal Regions. For example, we use regional confidence index: U.S. ConfidenceIndex for New England (Boston), Pacific (Los Angeles, San Francisco, and San Diego), East North Central (Chicago), South Atlantic (Miami and District ofColumbia), Mid-Atlantic (New York) and Mountain (Las Vegas, and Denver), respectively.*Significant at the 10% level.**Significant at the 5% level.***Significant at the 1% level.

Page 20: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

2 0 6 u J i n , S o y d e m i r , a n d T i d w e l l

sentiment but also acts in part to ‘‘correct’’ for any disequilibrium that existedin the previous period. Broadly, g describes the speed adjustment back to theequilibrium measuring the proportion of the last period’s equilibrium error that iscorrected. We find that the estimated coefficients generally shared the same signas presented in the long-run model; however, the effect was subdued in the short-run model as the differences in the variables do not appear to be as rich as thelevel series of the variables. Additionally, the short-run model generally lacksexplanatory power with adjusted R2s ranging from 0.04 to 0.20. In fact, most ofthe explanatory power is attributed to the error correction term g. The errorcorrection term has a statistically significant coefficient estimate in eight of tenregions. The short-run model confirms that returns per risk exhibit a short-termdynamic adjustment toward long-run equilibrium between variables. The errorcorrection term (constrained to 21 to 1) can be interpreted as the speed our modelreturns to equilibrium one month following an exogenous shock. The adjustmentback towards equilibrium ranges from 10% to 2% one time period after a shock.5

Estimated coefficients on the change of the lagged orthogonalized irrationalsentiment variable were negative and statistically significant across only two ofthe regions, indicating that the first differenced model lacks explanatory powerrelative to the long-run model.

C a u s a l i t y Te s t f o r S e n t i m e n t a n d F u n d a m e n t a l Va r i a b l e s

Based on the previously reported results, we test to see if changes in ‘‘sentiment’’will affect ‘‘fundamental’’ variables. In this case, we would expect that changesin fundamental variables will lead to changes in sentiment. If the direction werereversed and changes in sentiment precede changes in fundamental variables thenwe might conclude that the negative coefficients produced for laggedorthoganalized sentiment in prior models might be spuriously calculated. In thisscenario, sentiment levels would result in a subsequent change in fundamentalvariables, which would affect house price returns. To test for this inverserelationship, we conduct a test for relative sensitivity of changes in sentiment tothe changes in fundamental economic variables using the Granger causality test.Using the Granger causality test, we attempt to answer the question of whethervariable X ‘‘Granger causes’’ Y. Thus, Y is Granger-caused by X if X improves theprediction of Y, or equivalently if the coefficients on the lagged Xs are statisticallysignificant. We can specify the Granger causality test as:

Dy 5 a 1 a Dy 1 z z z 1 a Dy 1 b Dx 1 z z zt 0 1 t21 l t2l 1 t21

1 b Dx 1 « (7)l 2l t

Dx 5 a 1 a Dx 1 z z z 1 a Dx 1 b Dy 1 z z zt 0 1 t21 l t2l 1 t21

1 b Dy 1 u (8)l 2l t

Page 21: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

T h e U . S . H o u s i n g M a r k e t a n d t h e P r i c i n g o f R i s k u 2 0 7

J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4

Exhibi t 8 u Granger Causal Tests for the Change in Sentiment Index and Fundamental Variables

Panel Null Hypothesis F-Stat. P -Value

Panel 1 DFRM does not Granger Cause D Sentiment 3.82 0.01DSentiment does not Granger Cause D FRM 1.03 0.39

Panel 2 DARM Spread does not Granger Cause D Sentiment 2.55 0.04DSentiment does not Granger Cause D ARM Spread 0.52 0.68

Panel 3 DConstruction Cost does not Granger Cause D Sentiment 2.54 0.04DSentiment does not Granger Cause D Construction Cost 0.42 0.79

Panel 4 D Income does not Granger Cause D Sentiment 2.75 0.03DSentiment does not Granger Cause D Income 0.21 0.93

Panel 5 DBuilding Permit does not Granger Cause D Sentiment 0.27 0.89DSentiment does not Granger Cause D Building Permit 0.35 0.84

Panel 6 DUnemployment does not Granger Cause D Sentiment 1.64 0.10DSentiment does not Granger Cause D Unemployment 2.48 0.04

Panel 7 Ddefault risk does not Granger Cause D Sentiment 0.49 0.74DSentiment does not Granger Cause D Unemployment 0.91 0.45

Panel 8 DTerm risk does not Granger Cause D Sentiment 0.74 0.56DSentiment does not Granger Cause D Term risk 0.45 0.76

Notes: Using monthly data over the January 1998 to December 2008 time period. We include theConference Board Sentiment Index as proxy for sentiment variable and also include fundamentalvariables; Income, Building Permit, Unemployment Rate, Term Structure (a spread between 10-yearTreasury note and 3-month T-bill rate), Default Risk (a spread between 10 AAA-rated bond andBAA-rated bond). As a robustness check of sentiment index, we also compare the MichiganSentiment Index with the Conference Board Index. However the results are consistent with theConference Board Sentiment Index and thus omit for a concise manner of presentation. Results areavailable based upon a request.

for all possible pairs of (x, y) series in the study. The estimates of p-value arefrom the Wald statistics for the joint hypothesis:

b 5 b 5 z z z 5 b 5 0. (9)1 2 l

Therefore, the null hypothesis is that X does not Granger-cause Y in equation 7and that Y does not Granger-cause X in the equation 8.

In Exhibit 8, each panel (panels 1–8) indicates that the change in fundamentalvariables relates significantly to our research variable sentiment, indicatingfundamental variables Granger cause changes in sentiment. However, we were notable to find the inverse relationship where the sentiment variable precedesfundamentals except for the unemployment rate; in the case of unemployment,

Page 22: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

2 0 8 u J i n , S o y d e m i r , a n d T i d w e l l

sentiment and unemployment are tightly connected. Although not reported inExhibit 8, as a robustness check of the sentiment index, we compare the MichiganSentiment Index with the Conference Board Index and find consistent results withthe Conference Board Sentiment Index.

The previous model (Model 1) is used to explain the role of irrational sentimentat the MSA level using a combination of real estate related local and nationalexogenous variables. We extend the existing model by including additionalpossible explanatory variables, including real estate variables collected for abroader regional use. Model 2 includes all fundamental variables adopted inExhibit 8 and equation 5, and two additional national economic variables: termstructure (spread between 10-year Treasury notes and 3-month T-bill rates) anddefault risk6 (spread between AAA-rated bonds and BAA-rated bonds), and twoadditional real estate regional variables: market volatility and building permits.Considering data availability related to the additional variables, we classify eachMSA into four unique regions for our variables based on regional data: Northeast,West, Midwest, and South. Similarly to the previous model, we report the 9-monthlag in the orthogonalized (non-fundamental) sentiment index. See Exhibits 9 and10 for Model 2 results.

In the long-run equation, we find that the estimates from many of the explanatoryvariables, FRM, Construction Cost, ARM Spread, Income, and UnemploymentRate, previously modeled are robust to our respecified model. The two newnational ‘‘economic’’ variables, Default Risk and Term Structure, generallyproduce significant parameter estimates. As expected, an increase in both thedefault risk and term spread will negatively impact house price returns (dependentvariable). The two new regional ‘‘real estate’’ variables, Regional Housing MarketVolatility and Regional Building Permits issued, both produce negativecoefficients; however, the number of buildings permits issued is not significant.Real estate market volatility tends to have a significant negative impact on houseprices across all MSAs. The coefficients produced by the lagged orthogonalizedsentiment variable are negative and statistically significant in six of the ten MSAs.Thus it appears that the influence of non-fundamental or irrational sentiment isrobust to the inclusion of the additional ‘‘economic’’ and ‘‘real estate’’ variablesin the long-run model. Compared to the prior more parsimonious model, therespecified model is a substantial improvement with the adjusted R2 for the long-run model ranging from 67% to 92% across MSAs.

In the short-run model, the error correction term is generally significant acrossthe regions; however, it is positive, which indicates that housing price returns donot exhibit a short-term adjustment toward a long-run equilibrium betweenvariables. This is perhaps an artifact to the time period of this analysis and theinteraction of house prices with market volatility. The newly added marketvolatility variable is significant and negative across eight of the ten CMSAs. Thecoefficients produced by the lagged orthogonalized (non-fundamental) sentimentvariable are generally not significant in the short-run. It appears that the influenceof non-fundamental or irrational sentiment is less invasive in the short-run model

Page 23: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

Th

eU

.S

.H

ou

si

ng

Ma

rk

et

an

dt

he

Pr

ic

in

go

fR

is

ku

20

9

JR

ER

uV

ol

.3

6u

No

.2

–2

01

4

Exhibi t 9 u Model 2: Long-run Residential Market Annual Return Level Models with Market Sentiment

Boston Chicago Denver Las Vegas LA Miami New York San Diego San Francisco Wash. D.C.

Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff.

Constant 0.16** 20.01 20.06 0.52*** 0.69*** 0.76*** 0.23*** 0.46*** 0.04 0.24**(2.18) (20.27) (21.29) (3.14) (9.40) (6.79) (5.34) (3.38) (0.27) (2.36)

FRM Rate 1.86** 1.45*** 4.15*** 25.55*** 28.22*** 27.69*** 21.38** 23.09** 2.89 22.74***(2.27) (5.46) (7.79) (24.10) (29.11) (27.62) (22.51) (22.08) (1.60) (2.82)

Construction Cost 0.01* 20.01 0.01 20.02 20.02 20.01 20.01*** 20.01 20.01 20.01(0.17) (20.95) (0.03) (21.01) (1.12) (20.78) (20.63) (20.51) (20.51) (20.99)

ARM Spread (FRM-ARM) 7.92*** 4.24*** 2.46** 3.44*** 18.84*** 9.26*** 12.01*** 14.79*** 11.31*** 16.23***(5.08) (6.42) (2.55) (0.37) (8.36) (3.13) (9.18) (4.46) (3.01) (6.68)

State Income 21.08*** 20.52** 20.59** 1.65*** 0.55** 0.63 20.51*** 20.01 20.30 21.15***(27.90) (23.90) (5.38) (3.13) (2.54) (1.26) (23.58) (20.05) (20.77) (25.36)

CMSA Unemployment Rate 24.18*** 21.03 21.95*** 21.97 22.52** 23.80*** 21.14* 24.46*** 23.26*** 24.04*(24.92) (0.82) (24.69) (20.99) (22.03) (22.97) (21.68) (22.74) (25.33) (21.94)

Term Structure 20.06 20.02*** 20.01** 0.02 20.04*** 20.03*** 20.04*** 0.01 20.05** 0.13***(0.10) (27.06) (0.83) (1.60) (23.40) (22.51) (25.71) (0.12) (22.12) (4.36)

Default Risk 20.06*** 20.01*** 20.04*** 20.14*** 20.06** 20.01 20.03** 20.15*** 22.64 20.15***(23.75) (27.70) (24.87) (23.62) 22.54 (20.51) (22.33) (24.38) (21.49) (26.72)

Regional Building Permit 0.01 20.01 0.01 20.04 20.05 0.01 0.01 20.03 20.01 20.04(0.10) (20.50) (0.45) (20.83) 21.28 (0.13) (0.25) (20.61) (20.06) (21.20)

Market Volatility 20.02*** 20.04** 20.08*** 20.02*** 20.03*** 20.03*** 20.03*** 20.01** 20.01*** 20.02***(3.87) (21.80) (23.64) (25.48) (25.19) (27.97) (29.56) (22.46) (22.40) (24.83)

Page 24: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

21

0u

Ji

n,

So

yd

em

ir

,a

nd

Ti

dw

el

l

Exhibi t 9 u (continued)

Model 2: Long-run Residential Market Annual Return Level Models with Market Sentiment

Boston Chicago Denver Las Vegas LA Miami New York San Diego San Francisco Wash. D.C.

Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff.

Ortho. Sentiment (Lag) 20.10*** 0.04** 20.08*** 20.05 20.04 20.27*** 20.12*** 20.12* 20.19** 0.03(23.87) (2.34) (23.64) (20.74) (20.97) (24.30) (24.04) (21.69) (22.38) (0.78)

Adj. R2 0.75 0.92 0.83 0.78 0.86 0.84 0.78 0.71 0.67 0.76

Notes: T-statistics are in parentheses. The parameter estimates are based on a set of fundamental variables adopted in Exhibit 7 and also include nationallevel variables; default risk and term structure risk. The market volatility is measured as a moving average market standard deviation of the last 12 months.We also include a supply side variable; building permit available at four regional levels, Northeast, South, West, and Midwest. We also estimate the modelwith 1, 3 and 6 lags for orthogonalized consumer sentiment and found 9-month lag has been confirmed by SBC criterion. We also confirm that otherdifferent lags provide a similar coefficient with 9-month lag.*Significant at the 10% level.**Significant at the 5% level.***Significant at the 1% level.

Page 25: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

Th

eU

.S

.H

ou

si

ng

Ma

rk

et

an

dt

he

Pr

ic

in

go

fR

is

ku

21

1

JR

ER

uV

ol

.3

6u

No

.2

–2

01

4

Exhibi t 10 u Model 2: Short-run Residential Market Annual Return Level Models with Market Sentiment

Boston Chicago Denver Las Vegas LA Miami New York San Diego San Francisco Wash. D.C.

Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff.

Constant 20.01 20.01*** 20.01*** 20.00* 20.01** 20.01* 20.01** 20.01 20.01** 20.01(22.27) (22.58) (23.16) (21.77) (22.80) (21.88) (22.21) (23.15) (21.91) (22.13)

D FRM Rate 0.13 0.56 1.64*** 1.16 0.12 20.62 0.15 0.36 2.64** 20.01(0.25) (1.41) (4.78) (0.94) (0.14) (20.78) (0.31) (0.43) (2.31) (20.12)

D Construction Cost 0.01 20.01 0.00 20.00 0.00 0.01 0.15 0.01 0.00 0.01(1.10) (20.26) (20.09) (20.08) (20.02) (0.27) (0.37) (0.29) (0.36) (0.19)

D ARM Spread (FRM-ARM) 0.23 0.15 2.13** 21.79 0.92 2.10* 0.58 20.65 22.59 0.86(0.31) (0.25) (4.26) (20.95) (0.72) (1.73) (0.83) (20.53) (21.53) (0.83)

D State Income 20.23** 20.09 0.01 0.87*** 0.28 0.61** 0.02 0.31* 0.22 20.04(22.31) (20.63) (0.11) (3.17) (1.62) (2.36) (0.31) (1.80) (0.95) (20.40)

D CMSA Unemployment Rate 20.08 20.01 20.01 20.01 20.01 20.01 0.01 20.01 20.01 20.01(21.18) (20.14) (20.42) (20.69) (20.36) (20.46) (0.06) (20.12) (20.23)) (20.06)

D Term Structure 20.01 20.01** 0.01 0.00 20.01 20.01 20.01 0.01 20.01** 20.01*(21.17) (21.98) (1.52) (0.81) (20.62) (21.34) (21.94) (0.33) (22.19) (21.81)

D Default Risk 20.01** 20.01*** 20.04 20.01 20.01 20.01 20.01* 20.01 20.02* 20.01**(22.21) (23.04) (21.15) (20.62) (20.66) (21.34) (21.81) (20.88) (21.81) (22.59)

D Regional Building Permit 20.00 20.01 20.01 20.01 20.01 20.01 20.01 20.00 20.01 20.01(20.89) (20.60) (20.07) (20.09) (20.13) (20.23) (20.03) (20.03) (20.51) (20.79)

D Market Volatility 20.01*** 20.01* 20.08*** 0.01 20.01** 20.01*** 20.01** 20.01 20.01*** 20.01***(22.78) (21.86) (23.50) (1.36) (22.40) (24.53) (22.37) (20.14) (23.41) (23.05)

Error Correction Term 0.01 0.10** 0.04** 0.04* 0.05*** 0.03* 0.03 0.04*** 0.02 0.01(0.90) (2.52) (2.59) (1.85) (2.73) (1.73) (1.37) (3.16) (1.29) (0.64)

Page 26: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

21

2u

Ji

n,

So

yd

em

ir

,a

nd

Ti

dw

el

l

Exhibi t 10 u (continued)

Model 2: Short-run Residential Market Annual Return Level Models with Market Sentiment

Boston Chicago Denver Las Vegas LA Miami New York San Diego San Francisco Wash. D.C.

Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Coeff.

D Ortho. Sentiment (Lag) 0.01 0.01** 20.01 0.01 20.01 20.01 0.01 20.01 0.01 0.01(0.12) (1.99) (20.41) (0.28) (20.32) (20.75) (0.21) (20.78) (0.62) (1.06)

Adj. R2 0.18 0.21 0.30 0.14 0.15 0.28 0.13 0.13 0.31 0.17

Note: T-statistics are in parentheses. The parameter estimates are based on a set of fundamental variables adopted in Exhibit 7 and also include nationallevel variables: default risk and term structure risk. The market volatility is measured as a moving average market standard deviation of the last 12 months.We also include a supply side variable; building permit available at four regional levels: Northeast, South, West, and Midwest. We also estimate the modelwith 1, 3, and 6 lags for orthogonalized consumer sentiment and found 9-month lag has been confirmed by SBC criterion. We also confirm that otherdifferent lags provide a similar coefficient with a 9-month lag.*Significant at the 10% level.**Significant at the 5% level.***Significant at the 1% level.

Page 27: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

T h e U . S . H o u s i n g M a r k e t a n d t h e P r i c i n g o f R i s k u 2 1 3

J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4

than the previous more parsimonious model might confer, but remains influentialin the long-run model.7

S u m m a r y o f R e s u l t s

In summary of the ECM models, the long-run models indicate that non-fundamentally induced market sentiment does indeed impact housing prices.Further, this sentiment is inversely related to future housing prices, indicating thathigh non-fundamentally-based sentiment will put downward pressure on futurehousing prices. The estimates in the long-run model cannot be interpreted similarlyto partial derivatives; however, the combined variables in Model 1 explain between35% and 75% of the variability in housing returns per risk. Extending the modelto include two national variables (term structure and default risk) and two regionalreal estate variables (market volatility and building permits) improves ourunderstanding of real estate returns, with variance ranging from 67% to 92% inModel 2. Variables such as the CMSA unemployment rate, CMSA marketvolatility, and regional non-fundamentally-based market sentiment, along with thenational ARM spread, default risk, and term structure seem to impact housingreturns the most consistently across Models 1 and 2. Local economic, sentiment,and real estate variables cannot fully explain local housing price patterns, whilenational financial variables are important to the model.

The short-run equations provide us with the error correction term (constrained to21 to 1), which can be interpreted as the speed our model returns to equilibriumfollowing an exogenous shock.8 Model 1 has an error correction term rangingfrom 20.10 to 20.02, indicating 10% to 2% movement back towards equilibriumone month subsequent to an exogenous shock to the model. The error correctionterm in Model 2 across all regions is positive (10% to 1%), suggesting slightmovement away from equilibrium following a shock to the model during the timeperiod studied. This is likely attributed to the market volatility variable, notincluded in Model 1, impacted by the run-up in house prices followed by thecredit crises.

A third model, results presented in Exhibit 11, is derived to control for the impactof the internal dynamics of the housing markets, particularly changes in laggedhouse price returns and house volume, and to also provide greater economicinterpretability of the partial coefficients. The parameter estimates are derived froma set of regional fundamental variables as well as national level variables: defaultrisk and term structure risk. Due to limited data availability, regional variablespreviously in 10 CMSA levels were transformed into four regions, Northeast,South, West, and Midwest. The D Housing Returnt-1 measures one lag of regionalaverage housing market return. Our four regional models indicate that the internalhousing dynamics, particularly lagged return on residential real estate markets,explains most of the short-run variation observed at a regional level. The laggedresidential real estate market returns show highly significant estimates, 0.80, 0.95,0.55, and 0.88, respectively. This indicates that the expected impact of a one dollar

Page 28: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

2 1 4 u J i n , S o y d e m i r , a n d T i d w e l l

Exhibi t 11 u Model 3: Regional Residential Market Return Models with Market Sentiment Model Controlling

for Internal Housing Dynamics

Northeast West Midwest South

DHousing Return DHousing Return DHousing Return DHousing Return

Constant 20.00 20.00 20.00 20.00(0.00) (0.00) (0.00) (0.00)

DHousing Returnt21 0.80*** 0.95*** 0.55*** 0.88***(0.05) (0.04) (0.08) (0.04)

D Income 20.04 20.06 20.02 0.22***(0.06) (0.09) (0.12) (0.12)

DUnemployment rate 20.03 0.11 20.02 0.16(0.12) (0.14) (0.11) (0.16)

DBuilding Permit 20.01 20.00 0.00 20.00(0.01) (0.01) (0.00) (0.00)

DTrading Volume 20.01 0.01 0.02*** 20.01(0.01) (0.08) (0.01) (0.01)

DFRM rate 0.34 0.80*** 0.32 20.20(0.25) (0.34) (0.33) (0.35)

DARM Spread 0.43 1.06*** 20.16 1.24***(0.37) (0.53) (0.51) (0.55)

DConstruction cost 0.01*** 0.00 20.00 0.00(0.01) (0.00) (0.00) (0.00)

DDefault Risk 20.01 0.01 20.01*** 0.00(0.01) (0.01) (0.00) (0.00)

DTerm Structure 20.01*** 0.00 20.00 20.01***(0.00) (0.01) (0.00) (0.00)

DSentiment 20.01 0.01 20.00 0.01(0.01) (0.01) (0.00) (0.00)

R2 0.710 0.840 0.394 0.825Adj. R2 0.680 0.83 0.337 0.809

Notes: T-statistics are in parentheses. The parameter estimates are based on a set of regionalfundamental variables adopted in Exhibit 8 and also include national level variables: default riskand term structure risk. Our regional variables previously in 10 CMSA level have transformed intofour regions: Northeast, South, West, and Midwest. We transform a set of CMSA levelfundamental variables into regional equal-weight fundamental variables including consumersentiment. The D Housing Returnt21 measures one lag of regional average housing market return.We also include a supply side regional variable and construction cost.*Significant at the 10% level.**Significant at the 5% level.***Significant at the 1% level.

Page 29: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

T h e U . S . H o u s i n g M a r k e t a n d t h e P r i c i n g o f R i s k u 2 1 5

J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4

change in housing returns from the previous month has a contemporaneous impacton housing returns ranging from $0.55 to $0.95. We also find that trading volumehas a marginal impact on returns in the Midwest. The spread between ARM andFRM has a positive impact on the residential market in the West and Southregions, as a one unit increase in the spread results in an expected 1.06 and 1.24increase in housing price changes, respectively. While, again, sentiment does notseem to impact housing returns in the short-run.

u C o n c l u s i o n

This study builds on existing fundamental residential real estate market modelsby implementing a systematic procedure to account for irrational (non-fundamental-based) market sentiment. Specifically, we investigate the role ofirrational market sentiment and excess return per risk in the U.S. residential realestate market over a 10-year period (January 1998 to December 2008) coveringperiods of high and low sentiment and corresponding housing prices. We generallyfind responding information on each fundamental variable with the expecteddirectional relationship in the long-run models. The inclusion of our derivedirrational (orthogonalized) sentiment measure to the vector of fundamentalvariables has important structural and predictive implications, and increases theunderstanding of residential real estate pricing.

These results empirically support Shiller’s (2007) argument that marketfundamentals do not fully explain the recent housing market price movement, butrather an approach incorporating consumer psychology may describe it better. Thisstudy is a novel approach to empirically investigate this relationship within thecontext of the residential real estate market. The findings support the existence ofa long-run relationship between irrational market sentiment and the pricing patternof residential real estate during the 1998 to 2008 time period. Our findings indicatethat consumer irrational (non-fundamental-based) sentiment does indeed impactsubsequent housing prices and can lead to euphoric behavior, therefore real estatepricing models should include a variable capable of measuring irrationalsentiment.

u E n d n o t e s1 This study denotes the fundamental analysis as a model based on a microeconomic

framework that often uses macroeconomic variables. Chen, Roll, and Ross (1987) usearbitrage models as premised on cash flow analysis using systematic aggregated variables.Fundamental variables often combined to simulate cash flow measures include inflationmeasure, volatility in mortgage interest rates, regional population and income growth,and changes in money supplies and other capital market activities. The validation ofspecification on fundamental analysis is challenged by the selection of the rightfundamental exogenous variables. To avoid controversy, economic variables are includedin our parsimonious fundamental model.

Page 30: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

2 1 6 u J i n , S o y d e m i r , a n d T i d w e l l

2 The description of area coverage is available at the S&P/Case-Shiller website: http: / /www.standardandpoors.com.

3 We also estimate the model with 1, 3, and 6 month lags for irrational consumer sentimentand found that the 9-month lag was confirmed by SBC criterion.

4 The adjusted R2 for the long-run model with the orthogonalized market sentiment variableincreased in all regions compared to the fundamental model without the orthogonalizedmarket sentiment variable (not presented).

5 Negative one indicates full adjustment to equilibrium one time period later, while 0indicates no adjustment towards equilibrium. A positive error correction coefficientindicates movement away from equilibrium.

6 In the extended model, the authors have added term spread and default spread, as inYang, Zhou, and Leung (2012).

7 Considering data availability, we also classified each state into four different regions:Northeast, West, Midwest, and South and regressed the monthly return in regional(Northeast, West, Midwest, and South) housing markets on the regional (we combinedstate income and unemployment rate data respective of region to proxy for regional leveldata) and national variables, along with a broad regional irrational sentiment index. Wefind that the coefficients produced by the lagged orthogonalized (non-fundamental)sentiment variable generally are not significant in this regional analysis. These resultscombined with the previous MSA level results indicate that the impact of irrationalsentiment is highly localized. Where local irrational sentiment has a substantial impacton subsequent local housing price patterns, we were unable to detect a regional effect asregional irrational sentiment levels do not substantially impact subsequent regional housepricing.

8 Negative one indicates full adjustment to equilibrium one time period later, while 0indicates no adjustment towards equilibrium. A positive error correction coefficientindicates movement away from equilibrium.

u R e f e r e n c e s

Amihud, Y. Illiquidity and Stock Returns: Cross-Section and Time-Series Effects. TheJournal of Financial Markets, 2002, 5:1, 31–56.

Archer, W.R. and B.C. Smith. Residential Mortgage Default: The Roles of House PriceVolatility, Euphoria and the Borrower’s Put Option. Federal Reserve Bank of Richmond,Working Paper Series, 2010.

Baker, M. and J. Wurgler. The Equity Share in New Issues and Aggregate Stock Returns.Journal of Finance, 2000, 55:5, 2219–57.

——. Investor Sentiment and the Cross-section of Stock Returns. Journal of Finance, 2006,61:4, 1645–80.

Barber, B., T. Odean, and N. Zhu. Do Retail Trades Move Markets? Review of FinancialStudies, 2009, 22, 151–186.

Barkham, R.J. and C.W.R. Ward. Investor Sentiment and Noise Traders: Discount to NetAsset Value in Listed Property Companies in the U.K. Journal of Real Estate Research,1999, 18:2, 291–312.

Brown, G.W. and M.T. Cliff. Investor Sentiment and the Near-term Stock Market. Journalof Empirical Finance, 2004, 11:1, 1–27.

Page 31: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

T h e U . S . H o u s i n g M a r k e t a n d t h e P r i c i n g o f R i s k u 2 1 7

J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4

——. Investor Sentiment and Asset Valuation. Journal of Business, 2005, 78, 405–40.

Brown, S.J., W.N. Goetzmann, T. Hiraki, N. Shiraishi, and M. Watanabe. Investor Sentimentin Japanese and U.S. Daily Mutual Fund Flows. NBER Working Paper No. 9470, February2003.

Canbas, S. and S.Y. Kandir. Investor Sentiment and Stock Returns: Evidence from Turkey.Emerging Markets Finance & Trade, 2009, 45:4, 36–52.

Case, K.E. and R.J. Shiller. Is There a Bubble in the Housing Market? Brookings Papers

on Economic Activity, 2003, 2, 299–362.

Charoenrook, A. Does Sentiment Matter? Vanderbilt University working paper, 2005.

Chen, J., H. Hong, and J. Stein. Breadth of Ownership and Stock Returns. Journal of

Financial Economics, 2002, 66, 171–205.

Chen, K. Investor Sentiment and Use of Accounting Information. University of SouthernCalifornia working paper, 2008.

Chen, N., R. Roll and S. Ross. Economic Forces and the Stock Market. Journal of Business,1986, 59, 383–403.

Chopra, N., C.M. Lee, A. Shleifer, and R. Thaler. Yes, Discounts on Closed-end Funds area Sentiment Index. Journal of Finance, 1993, 48:2, 801–08.

Clayton, J. Rational Expectations, Market Fundamentals and Housing Price Volatility. Real

Estate Economics, 1996, 24:4, 441–70.

——. Are Housing Price Cycles Driven by Irrational Expectations? Journal of Real Estate

Finance and Economics, 1997, 14:3, 341–63.

——. Further Evidence on Real Estate Market Efficiency. Journal of Real Estate Research,1998, 15:1, 41–57.

Clayton, J., D.C. Ling, and A. Naranjo. Fundamentals Versus Investor Sentiment. Journal

of Real Estate Finance & Economics, 2009, 38:1, 5–37.

Coval, J.D. and E. Stafford. Asset Fire Sales (and Purchases) in Equity Markets. Journal

of Financial Economics, 2007, 86, 479–512.

Daniel, K., D. Hirshleifer, and A. Subrahmanyam. Investor Psychology and Security MarketUnder- and Overreactions. Journal of Finance, 1998, 53:6, 1839–85.

De Bondt, W. A Portrait of the Individual Investor—Heuristics and Biases. European

Economic Review, 1998, 42:3, 831–44.

De Long, J.B., A. Shleifer, L.H. Summers, and R.J. Waldmann. Noise Trader Risk inFinancial Markets. Journal of Political Economy, 1990, 98:4, 703–8.

Dennis, P. and S. Mayhew. Risk-neutral Skewness: Evidence from Stock Options. Journal

of Financial and Quantitative Analysis, 2002, 37:3, 471–93.

Diether, K., C. Malloy, and A. Scherbina. Differences of Opinion and the Cross Sectionof Stock Returns. Journal of Finance, 2002, 57:5, 2113–41.

Diaz III, J. How Appraisers Do Their Work: A Test of the Appraisal Process and theDevelopment of a Descriptive Model. Journal of Real Estate Research, 1990, 5:1, 1–16.

Frazzini, A. and O.A. Lamont. Dumb Money: Mutual Fund Flows and the Cross-Sectionof Stock Returns. Journal of Financial Economics, 2008, 88:2, 299–322.

Gallimore, P. and A. Gray. The Role of Investor Sentiment in Property InvestmentDecisions. Journal of Property Research, 2002, 19:2, 111–20.

Page 32: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

2 1 8 u J i n , S o y d e m i r , a n d T i d w e l l

Garbade, K.D. Federal Reserve Margin Requirements: A Regulatory Initiative to InhibitSpeculative Bubbles. In P. Wachtel (ed.), Crises in Economic and Financial Structure.Lexington, MA: Lexington Books, 1982.

Hansz, J.A. and J. Diaz III. Valuation Bias in Commercial Appraisal: A Transaction PriceFeedback Experiment. Real Estate Economics, 2001, 29:4, 553–65.

Harvey, A.C. The Econometric Analysis of Time Series. Second edition. Hertfordshire, UK:Philip Allan, 1990.

Hendershott, P.H. and B. MacGregor. Investor Rationality: Evidence from U.K. PropertyCapitalization Rates. Real Estate Economics, 2005, 33:2, 299–322.

Hirschleifer, D., A. Subrahmanyam, and S. Titman. Feedback and the Success of IrrationalTraders. Journal of Financial Economics, 2006, 81, 311–38.

Hong, H. and J. Stein. A Unified Theory of Underreaction, Momentum Trading andOverreaction in Asset Markets. Journal of Finance, 1999, 54:6, 1939–2406.

Johansen, S. Estimation and Hypothesis Testing of Cointegrating Vectors in GaussianVector Autoregressive Models. Econometrica, 1991, 59:6, 1551–80.

Jones, C.M. and O.A. Lamont. Short Sale Constraints and Stock Returns. Journal of

Financial Economics, 2002, 66:2–3, 207–39.

Kahneman, D., J. Knetsch, and R. Thaler. Anomalies: The Endowment Effect, LossAversion, and Status Quo Bias. The Journal of Economic Perspectives, 1991, 5:1, 193–206.

Kalotay, E.P. Gray, and S. Sin. Consumer Expectations and Short-horizon ReturnPredictability. Journal of Banking and Finance, 2007, 31:10, 3102–24.

Kaplanski, G. and H. Levy. Seasonality in Perceived Risk: A Sentiment Effect. SSRNworking paper, 2009.

Kumar, A. and C.M.C. Lee. Retail Investor Sentiment and Return Comovements. Journal

of Finance, 2006, 61, 2451–85.

Lamont, O. and J. Stein. Leverage and House-price Dynamics in U.S. Cities. RAND Journal

of Economics, 1999, 30:3, 498–514.

Lee, C.M.C., A. Shleifer, and R.H. Thaler. Investor Sentiment and the Closed-End FundPuzzle. Journal of Finance, 1991, 46:1, 75–109.

Lemmon, M. and E. Portniaguina. Consumer Confidence and Asset Prices: Some EmpiricalEvidence. Review of Financial Studies, 2006, 19:4, 1499–1529.

Lin, C., H. Rahman, and K. Yung. Investor Sentiment and REIT Returns. Journal of Real

Estate Finance & Economics, 2009, 39:4, 450–71.

Miles, W. Housing Investment and the U.S. Economy: How Have the RelationshipsChanged? Journal of Real Estate Research, 2009, 31:3, 329–49.

Miller, E. Risk, Uncertainty, and Divergence of Opinion. Journal of Finance, 1977, 32:4,1151–68.

Neal, R. and S.M. Wheatley. Do Measures of Investor Sentiment Predict Returns? Journal

of Financial and Quantitative Analysis, 1998, 33:4, 523–47.

Nneji, O., C. Brooks, and C. Ward. Intrinsic and Rational Speculative Bubbles in the U.S.Housing Market: 1960–2011. Journal of Real Estate Research, 2013, 35:2, 121–52.

Qiu, L. and I. Welch. Investor Sentiment Measures. Brown University working paper, 2006.

Randall, M.R., D.Y. Suk, and S.W. Tully. Mental Fund Cash Flows and Stock MarketPerformance. Journal of Investing, 2003, 12:1, 78–81.

Page 33: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

T h e U . S . H o u s i n g M a r k e t a n d t h e P r i c i n g o f R i s k u 2 1 9

J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4

Shiller, R.J. Market Volatility. Cambridge, MA: The MIT Press, 1989.

——. Understanding Recent Trends in House Prices and Home Ownership. Paper presentedat the Federal Reserve Bank of Kansas City’s Jackson Hole Symposium, August 31–September 1, 2007.

Shleifer, A. and R.W. Vishny. Liquidation Values and Debt Capacity: A Market EquilibriumApproach. Journal of Finance, 1992, 47:4, 1343–66.

Stein, J. Prices and Trading Volume in the Housing Market: A Model with Down-PaymentEffects. Quarterly Journal of Economics, 1995, 110:2, 379–406.

Swaminathan, B. Time-varying Expected Small Firm Returns and Closed-end FundDiscounts. Review of Financial Studies, 1996, 9:3, 845–87.

Verma, R., H. Baklaci, and G. Soydemir. The Impact of Rational and Irrational Sentimentsof Individual and Institutional Investors on DIJA and S&P 500 Index Returns. Applied

Financial Economics, 2008, 18:16, 1303–17.

Wheaton, C.W. and G. Nechvey. The 1998–2005 Housing ‘Bubble.’ Journal of Real Estate

Research, 2008, 30:1, 1–26.

Wermers, R. Is Money Really ‘Smart’? New Evidence on the Relation Between MutualFund Flows, Manager Behavior, and Performance Persistence. SSRN working paper, 2003.

Whaley, R.E. Understanding VIX. Journal of Portfolio Management, 2009, 35:3, 98–105.

Yang, J., Y. Zhou, and W.K. Leung. Asymmetric Correlation and Volatility Dynamicsamong Stock, Bond, and Securitized Real Estate Markets. Journal of Real Estate Finance

and Economics, 2012, 45:2, 491–521.

The authors gratefully acknowledge Paul Gallimore, and 2010 American Real EstateSociety participants; especially discussant Fabrice Barthelemy, and the anonymousreferees for their many insightful comments.

Changha Jin, Hanyang University, Sangnok-gu, Ansan-si, Gyeonggi-do 426-791,Korea or [email protected].

Gokce Soydemir, California State University, Turlock, CA 95382 or [email protected].

Alan Tidwell, Columbus State University, Columbus, GA 31907 or [email protected].

Page 34: The U.S. Housing Market and the Pricing of Risk ... · J R E R u V o l . 3 6 u N o . 2 – 2 0 1 4 The U.S. Housing Market and the Pricing of Risk: Fundamental Analysis and Market

This page intentionally left blank