key: cord-0533815-wkuq5z3q
authors: Alam, Masud
title: Time Varying Risk in U.S. Housing Sector and Real Estate Investment Trusts Equity Return
date: 2021-07-22
journal: nan
DOI: nan
sha: 5cf3fab1c1ee981c0130ec76fbda2f3d572e6110
doc_id: 533815
cord_uid: wkuq5z3q

This study examines how housing sector volatilities affect real estate investment trust (REIT) equity return in the United States. I argue that unexpected changes in housing variables can be a source of aggregate housing risk, and the first principal component extracted from the volatilities of U.S. housing variables can predict the expected REIT equity returns. I propose and construct a factor-based housing risk index as an additional factor in asset price models that uses the time-varying conditional volatility of housing variables within the U.S. housing sector. The findings show that the proposed housing risk index is economically and theoretically consistent with the risk-return relationship of the conditional Intertemporal Capital Asset Pricing Model (ICAPM) of Merton (1973), which predicts an average maximum of 5.6 percent of risk premium in REIT equity return. In subsample analyses, the positive relationship is not affected by sample periods' choice but shows higher housing risk beta values for the 2009-18 sample period. The relationship remains significant after controlling for VIX, Fama-French three factors, and a broad set of macroeconomic and financial variables. Moreover, the proposed housing beta also accurately forecasts U.S. macroeconomic and financial conditions.

Real Estate Investment Trusts (REITs) are tradeable income-generating real estate equity investment vehicles related to the housing sector and are listed on major exchanges with more than $3 trillion in gross assets and more than $1 trillion equity market capitalization across a range of U.S. real estate sectors. In 2019, approximately 87 million U.S. investors invested in REITs stocks through their 401(k) 2 and these portfolio investments are directly associated with more than 500,000 properties. Therefore, understanding what variables drive the fluctuation of real estate investment trusts (REIT) equity return is of paramount importance for REIT investors aiming to make optimal investment decisions.

Existing studies mainly focus on monetary and fiscal policy shocks, firm-specific idiosyncratic risk, financial, and standard equity risk variables in explaining real estate stock and REIT returns. For example, Bond and Xue (2017) consider firm-specific investment and profitability factors to predict the expected REIT returns, Liow, Ibrahim, and Huang (2006) consider conditional volatilities of six macroeconomic variables and their expected risk premia on property stock returns, and McCue and Kling (1994) examine the response of REIT returns to shocks related to four macroeconomic variables. While the dynamics in conditional correlations of REIT returns and the volatility of macroeconomic predictors is well examined in the empirical literature, studies on the role of predictors directly related to the U.S. housing market is limited.

Recent theoretical studies show that the volatility associated with the housing sector is not adequately identified as a distinct risk factor in the capital asset pricing model (CAPM) or in Fama-French three factors model (DeLisle, Price, & Sirmans, 2013; Ooi, Wang & Webb, 2009; Sun, Titman, &Twite, 2015) when applied to REIT return estimation. Research by Morck, Yeung, and Yu (2000) and Roll (1988) also observe that the factors related to economic and market fundamentals can only explain a small portion of the asset return volatility and suggest sectorspecific factors can better explain the remaining unexplained variance of asset returns. Given the lack of literature on housing volatilities and the theoretical call for more sector-specific factors in the asset pricing model, this study focuses on identifying relevant factors within the U.S. housing sector and examines if the conditional covariance of volatilities in the housing sector with those factors and REIT returns are significantly correlated.

This study proposes a housing risk index as an additional measure of the volatilities within the U.S. housing sector. A two-step methodology is used to estimate the conditional volatilities of these ten housing variables. In the first step, common factors are extracted using a dynamic factor model (Alam, 2021; Bernanke, Boivin, & Eliasz, 2005) from an extensive macroeconomic data set, made up of 76 U.S. monthly macroeconomic series. Next, we estimate ten separate bivariate Threshold-Generalized Autoregressive Conditional Heteroscedastic (T-GARCH) models (Glosten, Jagannathan, & Runkle, 1993) , which include the first common factor 3 and one of the ten housing variables to account for asymmetric responses of the volatility of housing variables.

This yields a conditional volatility series for the ten housing variables from which a risk index is constructed by estimating the first principal component of the conditional volatility series of the ten housing variables.

The conditional volatility of the ten housing variables is chosen after being identified in the previous literature as fundamental variables of the U.S. housing market. For instance, Fratantoni and Schuh (2003) use regional housing starts and mortgage rate, Davis and Heathcote (2005) examine mortgage markets and Case-Shiller house prices, Saks (2008) uses newly constructed housing units and the number of houses sold, and Mian and Sufi (2009) examine the role of subprime mortgage credit to estimate the sources of aggregate housing market volatility.

The results of the univariate regressions support the positive risk-return relationship of the intertemporal capital asset pricing model (ICAPM) (Merton, 1973) when the constructed housing risk index is included as a state variable. The estimated housing risk betas are economically and theoretically consistent across the cross-section of regressions of ICAPM models and predict, on average, a maximum 5.6 % risk premium for REIT equity returns. This risk-return relationship remains stable and significant when regression models include controls for a set of macroeconomic and financial variables. Consistent with Chiang and Li (2012) , the quantile risk betas show a higher positive risk-return relation when the distribution of returns is above the 10% quantile. After confirming the housing risk index's economic and statistical significance across univariate and multivariate cross-section regressions, this study examines the predictive power of the housing risk index for the U.S. business cycles fluctuations and finds that the proposed housing risk index accurately forecasts the U.S. macroeconomic and financial conditions and significantly predicts regional macroeconomic fluctuations.

These findings are robust across models after controlling for a set of standard macroeconomic and financial variables from Welch and Goyal (2008) . The model further controls for the Fama-French three factors and find economically and statistically significant evidence of housing risk premia in all regression specifications. The maximum and minimum housing risk premia across all multivariate OLS models are an average of 3.7 percent and 1.9 percent, respectively. The housing risk premium remains almost stable in quantile regression with a maximum of 6.7 percent for the 95 th percentile and 5.6 percent for the 75 th percentile of the return distribution. 4 In addition to in-sample predictive analysis, the analysis includes an out-of-sample and sub-sample forecasting analysis. Over the out-of-sample periods, the estimated coefficient of housing risk beta remains positive and consistent with the in-sample estimates and deliver lower out-of-sample predictive errors for the Fama-French three-factor model. The sub-sample analysis also supports the robust relationship between the housing risk index and the expected REIT returns.

The analysis uses April, 2009 as a breakpoint and shows that the positive relationship remains unaffected by sample periods' choice but shows higher values of housing risk beta for the May, 2009 -December, 2018 sample period 4 Springer and Cheng (2006) estimate firm-specific risk premium ranging from 2% to 9% for individual REITs, Edelstein and Magin (2013) show U.S. REIT equity risk premium is likely in between 4.3 and 6.3 percent. The historical REITs risk premium in excess of the 10-year treasury bond is approximately 6% for 1972-2015 period (Van Nieuwerburgh, 2019).

This study complements and contributes to three parts of the asset price literature. First, this paper is related to the literature concerning the construction of the new risk factors by combining a high-dimensional factor model with the GARCH model (Bali, Brown, & Caglayan, 2014; Ghysels, Plazzi, Valkanov, & Torous, 2013; Plazzi, Torous, & Valkanov, 2010) and accounting for asymmetric responses of the volatility of housing variables, as in Kwan, Li, and Ng, (2005) and Liow, Zhou, and Ye, (2015) . In this context, this study contributes to a large strand of REIT literature that has explored the empirical applications of principal component and latent factor approach in arbitrage pricing theory (Lizieri, Satchell, & Zhang, 2007; Mühlhofer, 2013; and Kozak, Nagel, & Santosh, 2018) . Second, the factor-based bi-variate T-GARCH approach employed in this study contributes to the improvement of existing multivariate GARCH estimates of the sector-specific risk factors and examining the role these risk factors in predicting REIT returns (Bali et al., 2014; Baillie and Myers, 1991; Chuang, Liu, Susmel, 2012; Devaney, 2001; Kaiser, 1996) . Finally, the empirical examination of ICAPM and the Fama-French model in REIT returns lies at the confluence of broader asset price literature that examines the risk-return relationship and estimates expected risk premia of new risk factors (Chan, Hendershott, & Sanders, 1990; Devaney, 2001; Ewing and Payne, 2005; Liow et al., 2015; and Naranjo and Ryngaert, 1998) .

The construction of housing risk factors of this study and the application of the factor-GARCH approach are closely related to Bali et al. (2014) , Ghysels et al. (2013) , Liow et al. (2006) , and Plazzi et al. (2010) . Bali et al. (2014) propose a broad index of a risk factors from eight macroeconomic variables and tests the proposed risk factor's predictive ability for hedge fund returns. They argue that unexpected changes in macroeconomic variables can be a source of aggregate risk, and the first principal component extracted from the volatilities of macroeconomic variables can predict the expected asset returns. Liow et al. (2006) apply principal component analysis with a GARCH (1, 1) process to estimate the risk premia of macroeconomic risk factors in predicting the variation of REIT returns. They applied GARCH (1, 1) approach to a set of estimated principal components of six macroeconomic variables and find that exposure to the six 5 macroeconomic risk factors provides better predictions than the financial risk factors commonly used to explain REIT returns.

Evidence for predictability in REIT returns and commercial real estate returns is provided by Ghysels et al. (2013) and Plazzi et al. (2010) using principal components extracted from a set of macroeconomic risk factors. Lizieri et al. (2007) examine multifactor asset price models for the U.S. REIT returns, where macroeconomic risk factors are estimated using principal component and independent component analysis. They suggest that the multi-factor arbitrage pricing (APT) model with estimated macroeconomic risk factors outperforms other factor-based statistical models. 5 While a substantial volume of studies related to the construction of macroeconomic risk factors, no attention has been devoted to the U.S. housing sectors' risk factors. Furthermore, Leamer (2007) argues that housing is a business cycle considering its contribution to the U.S.

GDP. This study fills this gap by considering variables from the U.S. housing sectors and estimates a housing risk index. The housing sector's risk is vital because the massive spike of volatility in housing sectors has identified as a main driving force in declining investment, prices, and employment in the U.S. economy during the great recession.

The bivariate T-GARCH approach employed in this study is related to econometric methodologies employed by Bali et al. (2014) , Baillie and Myers (1991) , Chuang et al. (2012) . Devaney (2001) , Kaiser (1996) , Kwan, Li, and Ng (2005) , and Liow et al. (2015) . Liow et al. (2015) use the multivariate T-GARCH approach to model the macroeconomic risk variables in predicting property stock excess returns and Bali et al. (2014) use the same approach to estimate macroeconomic risk factors. Unlike Bali et al. (2014) , who consider cross-correlation of eight correlated macroeconomic variables in the T-GARCH model, Kaiser (1996) employs a factor structure in the GARCH model with the assumption that the factors are orthogonal. Baillie and Myers (1991) focus on the joint distribution of the volatility of cash flow and the future prices of six commodities using the bivariate GARCH model, while Chuang et al. (2012) use the bivariate T-GARCH model to examine the covariance of ten Asian stock market's trading volume and return volatilities.

Among these examples, the methodology most similar to this study is that used by Bali et al. (2014) , Baillie and Myers (1991) , and Chuang et al. (2012) . However, while Bali et al. (2014) estimate the model using a pre-specified eight macroeconomic variables, this study includes variables from U.S. housing sectors and a factor extracted from a large panel of macroeconomic variables. This study also estimates time-varying conditional mean volatility in contrast to the constant mean volatility estimate of Baillie and Myers (1991) . Chuang et al. (2012) estimate the bivariate T-GARCH model without considering the asymmetric effects of macroeconomic innovations on the volatilities of asset returns, while this study incorporates time-varying effects of macroeconomic factors on REIT returns' volatilities. Additionally, the empirical procedure of this study differs from Devaney (2001) and Kwan et al. (2005) in selecting the lag of factors and housing variables. This paper also contributes to the empirical literature examining multifactor asset pricing models linked with the volatility of macroeconomic variables. These include Merton's (1973) original intertemporal capital asset pricing models (ICAPM), a variant version of ICAPM by Cox, Ingersoll, and Ross (1985) and Rubinstein (1976) , the conditional asset pricing model by Cochrane (1996) , the consumption-based model by Campbell (1993) , and the arbitrage pricing theory (APT) of Ross (1976) . This literature group postulates a theoretical connection between expected stock returns and the default risk premium to one or more volatility measures of macroeconomic fundamentals. This connection motivates this study to hypothesize a positive relationship between the housing risk factors and excess REIT returns. Chan et al. (1990) employ a multifactor asset pricing model to test how risk exposure of various factors influence the variation of U.S. stock, and REIT returns. The findings suggest that unexpected variations in macroeconomic factors significantly impact asset returns. The risk premium of macroeconomic factors in predicting expected REIT returns is also documented by Chen et al. (1997) , Devaney (2001) , and Naranjo and Ryngaert (1998) in various specifications of Fama-French asset pricing models. The consensus in these empirical studies is that the conditional variance of macroeconomic factors is priced into asset pricing models and predicts the higher expected future returns. The empirical validity of multifactor asset pricing models in the previous literature supports this study for assessing REIT returns exposures to housing risk factors. 7 On the predictability of U.S. REIT returns, Ewing and Payne (2005) and Liow et al. (2015) investigate whether equity REIT returns are related to the pre-specified macroeconomic factors.

They find that those factors significantly predict the variation of REIT equity returns. In contrast to the pre-specified set of macroeconomic variables used in previous studies, this paper uses orthogonal macroeconomic factors from 76 macroeconomic variables and then estimate a housing risk factor by considering the joint dynamics of macroeconomic factors and the ten housing variables.

The construction of the housing risk index for this study uses ten variables related to U.S. (Belviso and Milani, 2005; Karolyi and Sanders, 1998; Leamer, 2007; McCue and Kling, 1994) consider these leading housing indicators, not only because they directly signal the dynamics of housing markets, but their volatilities also potentially affect the investment opportunities.

A bivariate Threshold Generalized Autoregressive Conditional Heteroscedasticity model with a time-varying parameter (T-GARCH) is used for these ten housing variables to estimate individual housing risk factors. In empirical implementation, I first estimate the conditional mean equation for an individual housing series ( it X ) and the macroeconomic factor ( t f ) using a standard vector auto-regression (VAR) model of order one: 

, and 1  +  The dummy variable D imposes a condition on the model that the news, regional events, economic turmoil have a strong and significant influence on the regional housing sector. The dummy value of D is equal to one for 0 1  − t  , when there is bad news that affect economy negatively and D is zero otherwise. For instance, the regional business cycles tend to show a tranquil state when positive shocks influence the economies output and employment favorably.

Using equation (2)-(4), we can define the residual vector ) ( and the time varying conditional variance-covariance matrix ) ( as:

where t  is a symmetric and positive definite matrix, and

. For an estimated positive and statistically significant value of  , the negative shocks have larger effects on 2  than positive shocks in equation (2)- (4), and their impact on the volatilities of housing sectors is also considered differently.

Estimation of the factor augmented bivariate T-GARCH model follows a two-step approach. In the first stage, the unobservable factors t f are estimated using the principal component analysis method, and in the second stage, the first of those factors t f 1 and the housing variable X are jointly used for estimating T-GARCH model parameters through the following maximum likelihood function:

 denotes the vector of parameters in the GARCH models and n is the monthly observation of housing variables and the macroeconomic factor.

The factor-GARCH specification originally developed by Engle (1987) and applied to the prediction of stock returns by Ng, Engle, and Rothschild (1992) as a multivariate application to volatility transmission by Lanne and Saikkonen (2007) , and in bond returns by Engle et al. (1990) .

In recent REIT literature, Asteriou and Begiazi (2013) and Chang and Chen (2014) to any other alternative models, such as OLS, 6 auto-regressive integrated moving average (ARIMA), and auto-regressive conditionally heteroscedastic (ARCH) model. The standard GARCH model, as well as several parameterized extensions like exponential-GARCH (Nelson,1991; Nelson and Cao, 1992) , GARCH-in-mean, threshold-GARCH (Glosten et al.,1993) , and GARCH with exogenous factors (factor-GRACH), are among some of the most widely applied methods for modeling heteroscedastic variances and forecasting volatilities.

The individual housing risk factors are the conditional volatility of the ten housing variables derived from the factor augmented bivariate T-GARCH model above. Similarly, volatility in real house prices, houses sold, and the ratio of houses for sales to houses sold is higher during 1980-83. The higher volatility in the early 1980s is related to the rise in the number of bank failures related to the savings and loan crises, which was continued steadily through 1982 because several regional and sectoral recessions hit financial and housing sectors.

Since then, the volatility of all housing variables has declined significantly reached their historical lower values in early 2000.

Panel A of Table 3 representation of the common comovement of all ten individual risk factors (Stock and Watson, 2016 ). In the current context, this correlation implies the possibility of constructing an alternative single measure of housing risk based on the existence of a common comovement across these ten measures and the assumption that this common factor is the primary source of volatilities in housing sector.

This new single volatility index or housing risk index is generated as linear combinations of PCA loadings 8 and the original ten volatility series. The first principal component in equation (8) explains a maximum of 78 percent of variation of these ten housing volatilities 9 and could be seen as a weighted sum of the volatilities of all ten housing variables. 7 The level of significance is estimated using the corresponding p-values for all Pearson correlation coefficients. p < .001 '***', p < .01 '**', p < .05 '*' 8 As Hastie, Tibshirani, and Friedman (2009) and Skinner, Holmes, and Smith (1986) show If

are the original series of n number of correlated variables, then a weighted variable y can be generated from a linear combination of these n variables using n n x x x y

) are the are the corresponding eigenvectors or principal component coefficients. 9 PCA identifies how these 10 housing volatility series can be summarized in some fashion to construct a single index of meaningful aggregate volatility measures for further analysis and which captures maximum possible of the variation of these ten series. (25) Pctl (75) Max 

The construction of the proposed housing risk index in equation (8) is based on the factor loading matrix in Table 3 .2. As shows in panel B of Table 3 .1, volatility of all these ten housing variables in the correlation matrix are positively correlated with each other. Therefore, all the loadings in the first principal component are also positive in Table 3 .2. The equation shows that the risk index has a higher weight (also refer as higher factor loadings) of 0.98 on the conditional volatility of the ratio of the house for sales to houses sold, whereas relatively smaller loadings of 0.01 on the conditional volatility of house price index. Housing permit has the second highest optimal weight, whereas conditional volatility of four housing starts variables have a sizeable role in constructing the housing risk index.

The proportion of variation explained by principal components in Table 3 Cox et al. (1985) and Ross (1976) propose a macro-factor based arbitrage Pricing Theory (APT) model that incorporates a set of macroeconomic variables, with the assumption that returns r on assets N − are related with a number of k factors as in the following model: where, where, I is a vector of ones, and [.] is vector of constant prices of risk. Using equation (9) and (10) ) 11 ( [.] [.]

[.] ER is a vector of excess return. Equation (11) suggest that the excess return of assets N − are generated by the constant risk premium and the covariance of returns with the random factors.

Intuitively, the APT model shows that the variation of excess return can be decomposed into the contribution related to constant risk premium of the each factors in [.] and the covariance of return to the set of factors 1 kx f . Risk associated with the expected asset returns in the APT model are derived from two sources of risk factors. One is macroeconomic and financial risk factors that are non-diversified and the second source is asset specific risk that can be diversified.

The factors 1 kx f in equation (11) Chen (1983) and Chen, Roll, and Ross (1986) find the volatility of real industrial production, yield curve, inflation, and other sectorspecific macroeconomic variables are statistically significant observed risk factors. Following Chen et al. (1986) , a plethora of empirical tests of the APT Antoniou, Garrett, & Priestley, 1998; Burmeister and McEIroy, 1988; Connor and Korajczyk, 1988; French, 2017; Priestley, 1996; Shukla and Trzcinka, 1990 ) also propose a range of macroeconomic variables as common risk factors, with the confirmation that expected asset returns are significantly related to observed macroeconomic and financial factors.

In the spirit of the APT model and following Chen (1983) and Chen et al. (1986) , this study estimates a single risk factor from a set of housing variables and argues that the proposed housing risk factor is a proxy for a sector-specific macroeconomic risk factor in equation (11). This factor can explain the time-series behavior of expected REIT returns because it captures the dynamics and the pervasive information of the volatility of the U.S. housing sector. According to the APT model, regressing expected REIT equity returns on the housing risk factor follows the specification of the linear factor model, and the estimated coefficients should be positive. 10 This positive relationship provides the theoretical support to incorporate the housing risk factor to evaluate the empirical performance of the APT model.

The APT model suggests a positive relationship between all risk variables and the expected REIT equity returns. However, macroeconomic factors in the APT model are non-diversifiable.

Also, the factor-based model's statistical nature does not provide any economic intuition of why factors deserve compensation. A further investigation of the equilibrium risk-return relationship in Merton's (1973) ICAPM validates the role of aggregate housing risk index in predicting REIT equity return. The ICAPM shows the following relationship: 

The proposed housing risk index developed in section (3) can be considered a proxy measure of a factor in the set of state variables X in equation (13) and the estimated beta can define as the housing risk beta. In the ICAPM specification, volatility in the housing sector is then a source of non-diversifiable systematic risk. Therefore, the model predicts that the investors holding REIT portfolios must be compensated by a risk premium proportional to the degree of housing risk exposure. Campbell (1996) also provides a multifactor intertemporal asset pricing model where asset returns are related to the covariance of the market returns and the innovations in the future investment opportunity set. In the context of asset risk premia, a risk-averse investor must compensate by the risk premium, which leads to a positive correlation of asset returns with the innovations in the future investment opportunity set. Recent theoretical (Anderson, Ghysels, & Juergens, 2009; Bekaert, Engstrom and Xing, 2009; Bloom, Bond, and Reenen, 2007; Pramanik, 2020; Polansky and Pramanik, 2021) and empirical studies (Bali et al. 2014; Caldara, Fuentes-Albero, Gilchrist, & Zakrajšek, 2016) also provide evidence that the volatility in macroeconomic variables can be a relevant proxy measurement of risk in investor's investment opportunity set.

Therefore, an increase in macroeconomic volatility implies a reduction of asset return through an unfavorable shift in the investment set, and investors require a higher expected return for bearing higher risk.

Overall, the ICAPM model provides a theoretical foundation that the proposed housing risk factor is a relevant risk factor in REIT investor's investment set, with the assumption that the estimated beta should positively predict the future REIT equity returns. So, equation (13) suggests that the housing risk index is a proxy measure of broader macroeconomic factor in state variable 21 set X , and the expected market risk premium

at time t contains compensation for the risk in the U.S. housing sector.

The data set consists of monthly return on three REIT equity portfolios spanning from The REIT equity return index has the highest average return of 1.033, and the higher return accompanied by the lowest standard deviation of 4.86. The distribution of the monthly REIT returns shows fat tails and nonzero skewness coefficients. The Jarque-Bera (JB) test suggests that the normality assumption of return distribution holds in all REIT returns. Finally, a portmanteau Q-test of order one rejects the presence of autocorrelation. 

The baseline univariate regression model is

is the monthly excess REIT returns at time t , in which i represents one of the three REIT equity returns, t H is the housing risk index in month 1 − t , and i  is the housing risk beta. The Newey and West (1987) t-statistic is used to test the statistical significance of estimated coefficients.

The OLS and the quantile regression model test whether estimated coefficients of housing risk index satisfy standards risk-return relationship for forecasting the REIT equity return. Table   3 .4 presents the OLS and quantile regression results. The first row shows that the OLS estimates of i  for all REIT returns are positive and statistically significant, suggesting that an increase in risk in housing sectors predicts risk premium for the REIT equity market.

Since the housing risk index is standardized, the estimated  implies that a one standard deviation increase in housing risk factor predicts a 7.61 percent increase in all REIT excess return and a 6.01 percent for REIT equity excess return. Looking at historical REIT return data, the average annual returns of REIT lies between 7 percent and 8.6 percent over the 1977-2012 periods.

For the period of 2012-17, the average annual all REIT equity total return is an average of 11.64 percent (REIT, 2018).

The adjusted 2 R values of the OLS regression model lies in between 4.02 percent and 5.6 percent. While the relatively smaller values of 2 R is mainly due to the absence of a large set of explanatory variables, the comparison of model performance in section 5.6 provides superior outof-sample forecasting performance of univariate and multivariate models. Next, following Huang, Jiang, Tu, and Zhou (2015) , I further investigate the predictive power of housing risk beta by adding NBER recession dummy and the Chicago Board of Options Exchange's (CBOE) implied volatility index (VIX) to the equation (14):

where NBER is a dummy variable that takes a value one for NBER recession and zeroes for expansion. The VIX index is a popular measure of aggregate market volatility that commonly referred as the forward looking barometer of market's risk. Panel A of Table 3 .4 reports the In addition to the conditional mean forecast, the empirical examination also employs the quantile regression approach (Koenker and Bassett, 1978) that focuses on the REIT return predictability at different quantiles of the return distribution. The approach examines the relationship between REIT excess returns and the housing risk for the full range of distribution of returns. Compared to the least-squares regression model, the quantile estimation removes the impact of the left-and right-tailed error distribution and can provide a robust estimation of housing risk betas (Chiang and Li, 2012; Hua, Polansky, & Pramanik, 2019) . While both the univariate and multivariate regression model considers a wide range of quantiles, for the brevity of space, Table   3 .4 only reports 25%, 50%, 75%, and 99% points of the return distribution.

Consistent with results in Panel A, the estimated betas are positive and significant for the quantiles above 10 %, providing evidence that the housing risk beta significantly predicts risk premiums over the various distributions points. Among the four quantile estimates of i  , 50%, 75%, and 99% quantile beta are highly significant, where the i  's are much higher in higher quantiles. This finding is also consistent with Chiang and Li (2012) , who also finds higher positive betas above the median quantile of the distribution of returns.

The baseline OLS and quantile regression findings show a strong positive risk-return relation and the economically meaningful predictive ability of the housing risk index. This section investigates whether the risk-return relationship of housing beta remains unchanged when regression models include standard control variables widely used in asset price literature.

In financial economics, controlling Fama-French three factors and Carhart's momentum factor (MOM) are the standards for predicting conditional mean returns (Fama and French, 1989; Ferson and Harvey, 1991) . Following Welch and Goyal (2008) , recent empirical studies Alam, Khondker, & Molla, 2013; Bollerslev, Marrone, Xu, and Zhou, 2014; Goh, Jiang, Tu, and Wang, 2013; Hossain and Ahamed, 2015; Pramanik and Polansky, 2021; Rapach, Strauss, and Zhou, 2013) also use a set of financial and macroeconomic variables including the book-to-market ratio, earnings-price ratio, the dividend yield default premia, yield spread between BAA and AAA bonds, short and long interest rates, stock variance, net equity expansion, consumer price index, and inflation to investigate the risk-return hypothesis. Also, Conrad, Dittmar, and Ghysels (2013) However, many of the standard predictor variables found to be statistically insignificant (as also document by Cenesizoglu and Timmermann (2007) and Welch and Goyal (2008) ) for predicting the left and right tail of the return distribution. While adding more predictors provide better goodness of fit of the data to the model, too many predictors also penalize models by overfitting the data. Also, Campbell (1987) and Ghysels, Santa-Clara, and Valkanov (2005) suggest that the non-robust estimation of a positive risk-return relationship in many empirical studies is partly due to lack of guidance of the selection of predictive variables in the multivariate asset price models.

A Bayesian variable selection approach is applied to overcome this limitation, which compares model performance using log posterior odds and selects the model's covariates accordingly. In general, the approach selects the best set of predictors and suggests the best model with the smallest log posterior odds. There are twenty competing models for each predictor, where all models are ordered according to the log of posterior probabilities. The selection of predictors starts with the model that considers the full set of twelve predictors. Next, the approach uses the log posterior odds values to compare alternative models with selected predictors. Table 3 .5 reports the multivariate OLS regression results, where the Bayesian log posterior odds select a relevant set of model covariates. After controlling for a selected predictor, the housing risk beta remains positive and statistically significant for the three REIT portfolios.

While the estimated housing beta for all REIT total return and equity returns are smaller than the univariate estimates, statistically significant beta at 5 percent level of significance suggests the robust predictive ability of the housing risk index. The R 2 -values of all three models are higher than the univariate models, ranging from 48 percent to 55 percent, suggesting the validity of inclusion of housing risk index with the standard predictors in the models.

Panel B of Table 3 .5 provides evidence for a positive and significant housing risk beta for the 50%, 75%, and 99% quantile of the return distribution. After controlling for the same set of predictors, the housing risk index appears to be highly significant for the quantiles above 50 % of the return distribution. For the real estate50 equity return, housing risk beta remains insignificant for most of the quantiles. Consistent with the univariate regression, the quantile coefficient of housing risk betas are also relatively higher for the higher quantile of the return distribution.

The robustness analysis of the results focuses on the housing risk index's predictive ability obtained from the OLS and quantile estimation of the Fama-French factor model. This section reestimates the Fama-French three-factor model using two sub-samples to examine if the estimated housing risk beta is sensitive to the sample periods. The estimation assumes multiple unknown structural breakpoints over the sample periods and applies the endogenous identification approach proposed by Bai and Perron (1998) . The approach detects four corresponding breakpoints in figure 3.7: 1984:M3, 1993:M7, 2002:M4, and 2009:M4 . The estimation of the model using structural breaks allows the study to examine the cases if the risk-return relationship changes due to an expected policy change. periods, reflecting the effect of higher volatiles of U.S. housing sectors on the expected REIT returns. As the findings in previous sections, housing betas are positive and significant for the 50%, 75%, and 99% quantile of the return distribution. Therefore, the positive relationship between the housing risk index and REIT returns is not affected by choice of two sample periods, which validates the robustness of the findings of this study.

The out-of-sample forecasting analysis splits the total sample into a training set (in-sample) and a validation set (out-of-sample). Following Welch and Goyal (2008) 

Std. error Accuracy The Fama-French model also appears the best performing out-of-sample model according to the MSFE and correlation accuracy criteria. Looking over the out-of-sample portion of panel A, the Fama-French model's directional accuracy is 64%, which is the highest among the models. The model is also the best predictive model considering its lowest value of MSFE. Panel B reports the in-sample and the out-of-sample evaluation of the Fama-French models with the housing risk index. The model exhibits a superior forecasting accuracy in one period ahead prediction for all REIT portfolios, where the average directional accuracy lies between 67% and 74%.

After confirming the robustness of housing beta and the best predictive model, we explore the relationship between the housing risk index and measure of U.S. economic and financial conditions. The aim is to examine if the proposed housing risk index can predict the U.S. business cycle fluctuations using six macroeconomic and financial indexes as the dependent variables in OLS regression, in which the model estimates the alternative predictive ability of the proposed housing risk index. There are a significant number of empirical studies investigating the role of U.S. housing sectors and the REIT related variables to measure the U.S. business cycles, as in Christidou and Fountas (2018) , Cauley, Pavlov, and Schwartz (2007) , Davis and Heathcote (2005) , Leamer (2015) , and Strobel, Nguyen Thanh, & Lee (2018) . Leamer (2015) shows that the housing market accurately predicts eight U.S. economic recessions in the post-war II periods. Chauvet (1998) . The SRP index relies on four monthly coincident variables, such as non-farm payroll employment, industrial production index, real personal income excluding transfer payments, and real manufacturing and trade sales.

Finally, the analysis looks beyond the aggregate U.S. economic conditions by considering the regional economic fluctuations. The regional counterpart to the Chicago Fed National Activity Index (CFNAI) is the Midwest Economy Index (MEI), which measures the growth in nonfarm business activity in Illinois, Indiana, Iowa, Michigan, and Wisconsin. The final proxy is the Kansas City Financial Stress Index (KCFSI), which captures the critical aspects of eleven U.S. financial market variables. Developed by Hakkio and Keeton (2009) , positive values of KCFSI coincide with periods above the long-run average of financial stress, while negative values indicate the financial stress is below the long-run average.

The predictive analysis estimates the following regression: (15) for each of the six macroeconomic and financial condition indexes. As findings show, housing risk beta is significantly related to future business cycles and financial conditions.

The positive coefficient of housing risk betas predicts official periods of NBER economic recessions and the likelihood for the periods of higher financial stress in the KCFSI. The positive beta also predicts higher U.S. recession probabilities. For the CFNAI and MEI, the negative beta implies the below-average growth, lower nonfarm business activities, and the economic contraction. The negative beta is also related to the periods of lower production in IPG. The predictive power of housing risk beta for all cases remain consistent as predicted. 

This chapter estimates Merton's (1973) intertemporal capital asset pricing model and the Fama-French three-factor model using a proposed housing risk factor. The housing risk index is the first principal component of the ten time-varying conditional volatilities, which serves as an alternative aggregate measure of the U.S. housing sector volatilities. A higher value of the U.S. housing risk index is a significant leading indicator of downturns in U.S. business cycles, leaves investors in higher-risk positions, which requires a higher risk premium in their investment decisions. Due to the vital role of the housing sectors in the U.S. economy, REIT investors are likely to focus more on the risk associated with the aggregate macroeconomic condition and the various components of housing sectors, which potentially provide a guide to their optimal investment decisions. 

Determinants of Liquidity Risk in the Commercial Banks in Bangladesh

Macroeconomic Impact of Covid-19: A case study on Bangladesh

Output, Employment, and Price Effects of US Narrative Tax Changes: A Factor-Augmented Vector Autoregression Approach

Current Account Dynamics, Adjustment and Capital Mobility in Bangladesh

Does systemic risk in the financial sector predict future economic downturns

The impact of risk and uncertainty on expected returns

Macroeconomic variables as common pervasive risk factors and the empirical content of the arbitrage pricing theory

Modeling of daily REIT returns and volatility

Estimating and testing linear models with multiple structural changes

Macroeconomic risk and hedge fund returns

Bivariate GARCH estimation of the optimal commodity futures hedge

A multivariate analysis of United States and global real estate investment trusts

Risk, uncertainty, and asset prices

Structural factor-augmented VAR (SFAVAR) and the effects of monetary policy

Measuring the effects of monetary policy: a factoraugmented vector autoregressive (FAVAR) approach. The Quarterly journal of economics

Uncertainty and investment dynamics. The review of economic studies

Property portfolio allocation: a multi-factor model

Stock return predictability and variance risk premia: statistical inference and international evidence

The cross section of expected real estate returns: Insights from investment-based asset pricing

Joint estimation of factor sensitivities and risk premia for the arbitrage pricing theory

The macroeconomic impact of financial and uncertainty shocks

Stock returns and the term structure

Intertemporal asset pricing without consumption data

Understanding risk and return

Homeownership as a constraint on asset allocation

Predictability of stock returns: a quantile regression approach

Evidence of contagion in global REITs investment

Risk and return on real estate: evidence from equity REITs

An econometric characterization of business cycle dynamics with factor structure and regime switching

Some empirical tests of the theory of arbitrage pricing

Some empirical tests in the arbitrage pricing theory: Macro variables vs. derived factors

Real estate and the arbitrage pricing theory: macrovariables vs. derived factors

Economic forces and the stock market

Uncertainty in the housing market: evidence from US states

Stock returns and risk: Evidence from quantile

The bivariate GARCH approach to investigating the relation between stock returns, trading volume, and return volatility

The time-varying nature of the link between REIT, real estate and financial asset returns

A cross-sectional test of an investment-based asset pricing model

Ex ante skewness and expected stock returns

Risk and return in an equilibrium APT: Application of a new test methodology

A theory of the term structure of interest rates

Housing and the business cycle

Pricing of volatility risk in REITs

Time varying risk premia for real estate investment trusts: A GARCH-M model

The dynamics of exchange rate volatility: a multivariate latent factor ARCH model

The equity risk premium for securitized real estate: The case for US Real Estate Investment Trusts

Multivariate ARCH with factor structures-cointegration in variance

Asset pricing with a factor-ARCH covariance structure: Empirical estimates for treasury bills

The response of real estate investment trust returns to macroeconomic shocks

Business conditions and expected returns on stocks and bonds

The variation of economic risk premiums

Monetary policy, housing, and heterogeneous regional markets

Macroeconomic forces and arbitrage pricing theory

On the relation between the expected value and the volatility of the nominal excess return on stocks

Can US economic variables predict the Chinese stock market

There is a risk-return trade-off after all

Forecasting real estate prices

Financial stress: what is it, how can it be measured, and why does it matter

Conditional skewness in asset pricing tests

The elements of statistical learning: data mining, inference, and prediction

Time-varying correlation between stock market returns and real estate returns

Determinants of bank profitability: A study on the banking sector of Bangladesh

Assessing bivariate tail non-exchangeable dependence

Investor sentiment aligned: A powerful predictor of stock returns

Average skewness matters

A multivariate threshold GARCH model with timevarying correlations

One-factor-GARCH models for German stocks-estimation and forecasting

The variation of economic risk premiums in real estate returns

Regression quantiles

Interpreting factor models

A multivariate generalized orthogonal factor GARCH model

Housing is the business cycle

Housing really is the business cycle: what survives the lessons of 2008-09?

Macroeconomic risk influences on the property stock market

Correlation dynamics and determinants in international securitized real estate markets

The Underlying Return-Generating Factors for REIT Returns: An Application of Independent Component Analysis

The international REIT's time-varying response to the US monetary policy and macroeconomic surprises

FRED-QD: A quarterly database for macroeconomic research

Real estate returns and the macroeconomy: some empirical evidence from real estate investment trust data, 1972-1991

An intertemporal capital asset pricing model

The consequences of mortgage credit expansion: Evidence from the US mortgage default crisis

Volatility spillover between the RBI's intervention and exchange rate

The information content of stock markets: why do emerging markets have synchronous stock price movements?

Why do REIT returns poorly reflect property returns? Unrealizable appreciation gains due to trading constraints as the solution to the short-term disparity

Stock returns, dividend yields, and taxes

Conditional heteroskedasticity in asset returns: A new approach

Inequality constraints in the univariate GARCH model

Hypothesis testing with efficient method of moments estimation

A multi-dynamic-factor model for stock returns

Idiosyncratic risk and REIT returns

Expected returns and expected growth in rents of commercial real estate

A motif building process for simulating random networks

Optimal Estimation of Brownian Penalized Regression Coefficients

Optimization of market stochastic dynamics

The arbitrage pricing theory, macroeconomic and financial factors, and expectations generating processes

Out-of-sample equity premium prediction: Combination forecasts and links to the real economy

REITs: A Look Back at 2018 and a Look Forward to 2019

Presidential address: R2

The Arbitrage Theory of Capital Asset Pricing

The strong case for the generalized logarithmic utility model as the premier model of financial markets

Job creation and housing construction: Constraints on metropolitan area employment growth

Sequential tests of the arbitrage pricing theory: a comparison of principal components and maximum likelihood factors

Time-varying betas of US REITs from 1972 to 2013

The effect of sample design on principal component analysis

Real Estate Property Portfolio Risk: Evidence from REIT Portfolios

Dynamic factor models, factor-augmented vector autoregressions, and structural vector autoregressions in macroeconomics

Effects of macroeconomic uncertainty and labor demand shocks on the housing market

Reit and commercial real estate returns: A postmortem of the financial crisis

Risk and the performance of real estate investment trusts: A multiple index approach

Why are REITs currently so expensive?

A comprehensive look at the empirical performance of equity premium prediction

Multiscale analysis of international linkages of REIT returns and volatilities

Consumption Real Personal Consumption Expenditures (Billions of Chained 2012 Dollars)

Real Personal Consumption Expenditures: Services (Billions of 2012 Dollars), 5. Real Personal Consumption Expenditures: Nondurable Goods

Investment Real Gross Private Domestic Investment, (Billions of Chained

Real private fixed investment (Billions of Chained 2012 Dollars)

Real Gross Private Domestic Investment: Fixed Investment: Nonresidential: Equipment (Billions of Chained 2012 Dollars

Real private fixed investment: Nonresidential (Billions of Chained 2012 Dollars)

Real private fixed investment: Residential (Billions of Chained 2012 Dollars)

Shares of gross domestic product: Gross private domestic investment: Change in private inventories (Percent)

Real Government Consumption Expenditures & Gross Investment (Billions of

Real Government Consumption Expenditures and Gross Investment: Federal (Percent Change from Preceding Period)

Real Gov Receipts Real Federal Government Current Receipts (Billions of Chained 2012 Dollars) 15. Real government state and local consumption expenditures

Nonfarm Business Sector: Real Output

Manufacturing Sector: Real Output

Group 2: Industrial Production

Industrial Production: Final Products (Market Group

Consumer Goods

Durable Materials

Nondurable Materials

Durable Consumer Goods

Nondurable Consumer Goods

Business Equipment

Consumer energy products

Capacity Utilization: Total Industry (Percent of Capacity)

Capacity Utilization: Manufacturing (SIC) (Percent of Capacity)

Industrial Production: Manufacturing (SIC)

Residential Utilities

Industrial Production: Fuels

Group 3: Employment and Unemployment

All Employees: Total nonfarm (Thousands of Persons)

All Employees: Total Private Industries (Thousands of Persons)

All Employees: Manufacturing (Thousands of Persons)

All Employees: Service-Providing Industries (Thousands of Persons)

All Employees: Goods-Producing Industries (Thousands of Persons)

All Employees: Durable goods (Thousands of Persons)

All Employees: Nondurable goods (Thousands of Persons)

All Employees: Construction (Thousands of Persons)

All Employees: Education & Health Services (Thousands of Persons)

All Employees: Financial Activities (Thousands of Persons)

All Employees: Information Services (Thousands of Persons)

All Employees: Professional & Business Services (Thousands of Persons)

Civilian Employment (Thousands of Persons)

Civilian Labor Force Participation Rate (Percent)

Civilian Unemployment Rate (Percent)

Business Sector: Hours of All Persons

Manufacturing Sector: Hours of All Persons

Nonfarm Business Sector: Hours of All Persons

Average Weekly Hours of Production and Nonsupervisory Employees: Manufacturing (Hours)

Weekly Hours of Production And Nonsupervisory Employees: Total private (Hours)

Gross Domestic Product: Chain-type Price Index

Gross Private Domestic Investment: Chain-type Price Index

Consumer Price Index for All Urban Consumers: All Items (Index 1982-84=100)

Consumer Price Index for All Urban Consumers: All Items Less Food & Energy (Index 1982-84=100)

Producer Price Index for All Commodities

Group 5: Interest Rates

Effective Federal Funds Rate (Percent) 2. 10-Year Treasury Constant Maturity Rate (Percent)

30-Year Conventional Mortgage Rate© (Percent)

Moody's Seasoned Aaa Corporate Bond Yield© (Percent)

Moody's Seasoned Baa Corporate Bond Yield© (Percent)

Group 6: Money and Credit

Adjusted Monetary Base (Billions of 1982-84 Dollars)

Real Institutional Money Funds (Billions of 2012 Dollars)

Total Reserves of Depository Institutions (Billions of Dollars)

Reserves of Depository Institutions, Nonborrowed (Millions of Dollars)

Group 7: Stock Markets

S&P's Common Stock Price Index: Composite

S&P's Common Stock Price Index: Industrials