The Effects of Securitization, Foreclosure, and Hotel Characteristics on Distressed Hotel Prices, Resolution Time, and Recovery Rate

639124CQXXXX10.1177/1938965516639124Cornell Hospitality QuarterlySingh research-article2016 Article The Effects of Securitization, Foreclosure, and Hotel Characteristics on Distressed Hotel Prices, Resolution Time, and Recovery Rate Cornell Hospitality Quarterly 2017, Vol. 58(1) 39 52 The Author(s) 2016 Reprints and permissions: sagepub.com/journalspermissions.nav DOI: 10.1177/1938965516639124 journals.sagepub.com/home/cqx Amrik Singh 1 Abstract This study investigates the effects of securitization, foreclosure, and hotel characteristics on the sale prices of distressed hotels as well as their influence on resolution time and recovery rate. Using a sample of 4,763 financially distressed hotels between 2010 and 2014, this study provides evidence that hotel size, securitization, foreclosure, and disposal methods are important predictors of distressed property prices, resolution time, and recovery rate. Keywords hotel; distress prices; securitization; foreclosure; resolution time; recovery rate The financial crisis and ensuing recession of 2008 2009 had a devastating effect on the U.S. lodging industry. Historical data from Smith Travel Research s (STR) Host Almanac (STR Analytics, 2015) show that the lodging industry suffered a 14 percent decline in revenue in 2009, which led to a 33 percent decline in net operating income (NOI). As a result, many hotels fell into financial distress and subsequently disposed of at heavily discounted sale prices (STR Analytics, 2015). Even though the crisis is over for most hotel owners and operators, there are still many hotels in financial distress whose condition has yet to be resolved. According to Trepp, LLC. (2015), as of December 2014, up to 200 hotels that were securitized by commercial mortgage-backed securities (CMBS) were still in various states of financial distress including foreclosure, bankruptcy, and real estate owned (REO) distress, which raises two fundamental research questions: Research Question 1: How was financial distress resolved in the aftermath of the financial crisis and what factors could have influenced the distressed sale prices? Research Question 2: Did the determinants of distress have an effect on the time to resolution and the recovery rate of these hotels? To the author s knowledge, no academic research to date has attempted to investigate whether hotel characteristics that were found to be associated with non-distressed transaction prices in previous research can also explain distressed hotel prices, resolution time, and recovery rates. Hence, the purpose of this study is to investigate the effect of securitization, foreclosure, and hotel characteristics on resolution prices, resolution time, and recovery rate of distressed hotels. 1 According to Downs and Xu (2015), prior research on distressed real estate has largely focused on the residential sector with limited analysis of the effect of securitization on distressed outcomes in commercial real estate (Downs & Xu, 2015). Although existing academic research on the hotel industry has investigated the effects of various property characteristics on hotel values (Corgel, 2008; Corgel & deroos, 1993; O Neill & Xiao, 2006), the focus of these prior studies has been on non-distressed hotels. Moreover, the evidence is mixed on the factors influencing hotel prices (Corgel, 2008; O Neill, 2004). Given the stigma of foreclosure and evidence that distressed properties are sold at a discount (Pennington-Cross, 2006), it is reasonable to expect that the empirical findings on some of these hotel characteristics and their influence on distressed sales prices are likely to differ from prior research in the hotel industry. The empirical evidence on the determinants of distressed prices is currently lacking, which suggests the need for further research that can provide further insights into the effects of these determinants on distressed prices. 1 University of Denver, Denver, CO, USA Corresponding Author: Amrik Singh, Fritz Knoebel School of Hospitality Management, University of Denver, 2044 E. Evans Ave., Denver, CO 80208, USA. Email: amrik.singh@du.edu

40 Cornell Hospitality Quarterly 58(1) Exhibit 1: Conditions and Resolution of Financial Distress. Note. REO = real estate owned. This study contributes to the existing commercial real estate and lodging academic literature in several ways. First, this study provides descriptive evidence on the magnitude and scope of financial distress in the U.S. hotel industry from 2008 through 2014. By consolidating data from three major providers of commercial real estate data (CoStar, RCA, and Trepp), this study constructs a unique dataset of 4,763 financially distressed hotels within a multiple regression framework to provide new and distinct insights into the characteristics of distressed hotels and how these characteristics influence various financial resolution outcomes as shown in Exhibit 1. Second, it provides timely evidence on the type of distressed hotel condition and the effects of various hotel characteristics on liquidation prices as well as the length of time to resolve distress and the recovery rate. The focus on financial distress in the hotel industry distinguishes it from prior research, which emphasized the residential mortgage market and investigated the likelihood of default behavior in the commercial mortgage sector. For example, Pennington-Cross (2006) estimated the value of a foreclosed residential property, while Quan and Lebret (2006) estimated a structural model to explain the delinquency and default decision of hotels. Third, the results of this study complement and extend prior research by Downs and Xu (2015) that examined the marginal effect of portfolio lending versus securitization in the resolution of distressed commercial real estate. Although the current study is similar to Downs and Xu s (2015) in examining the

Singh 41 resolution time and recovery rate, it differs from the prior study in terms of its industry-specific emphasis on hotels, the methodology used, and its examination of distressed sale prices. Literature Review Financial Distress and Commercial Mortgage Research The existing literature on financial distress in commercial real estate is relatively limited. Many studies on commercial real estate mortgage loans examined the incidence of default risk and used the various structural or hazard models to estimate delinquency and default probability. Moreover, studies of delinquency and default find the hotel sector to be different from other commercial real estate property types. The stages and model of the causes and consequences of financial distress have been proposed by Brown, Ciochetti, and Riddiough (2006), which include empirical evidence of endogenous borrower default, significant underinvestment in distressed assets, and delayed asset sales by the lender. The results of their study revealed that foreclosure occurred more frequently during the height of the recession when it was difficult for lenders to sell REO properties because they faced an illiquid and uncertain market with few willing buyers and sellers. Restructuring was the prevalent distressed outcome when market conditions improved and a thriving real estate market developed for REO properties (Brown et al., 2006). With the growth of the CMBS market and the availability of CMBS data, researchers turned their attention to testing theoretical relationships between indicators and probability of default in CMBS loans. For example, Gordon and Kizer (2004) examined the aggregate performance of CMBS loans and expected default losses. Their study found that healthcare and lodging properties showed increasing default over time unlike other property types. Grovenstein, Harding, Sirmans, Thebpanya, and Turnbull (2005) used a large sample of commercial mortgages to investigate how loan-to-value (LTV) and debt coverage ratios (DCRs) were associated with default. They provide strong evidence that LTV and DCRs are endogenous to the underwriting process. In addition, they noted that hotels were most severely affected by the aftermath of 9/11. The findings indicated that hotel loans had a greater than average risk of default and that loans backed by hotels had a higher propensity to default than other commercial property types. Lancaster and Cable (2004) also found healthcare and hotel loan default rates to be the highest among all commercial property types. They concluded that hotel and healthcare loans had five times greater probability of default relative to other property types consistent with their inherent risk characteristics. Loan losses as a percentage of the original loan balance were also found to be the highest for the hotel sector. In a more recent related study, Downs and Xu (2015) investigated the marginal effect of portfolio lending versus CMBS lending in the resolution of distressed commercial real estate. Portfolio loans are those loans that are originated and retained by the lender, whereas CMBS loans are pooled together and placed in a trust and the bonds backed by the loans are sold to investors. The results of their study showed substantial and significant differences between portfolio and securitized loans based on various loan and property characteristics. First, their findings showed that distressed portfolio loans were more likely to be resolved and experienced higher foreclosure rates than loans secured by CMBS. Similarly, loans secured by properties located in judicial foreclosure states were less likely to be resolved than properties located in non-judicial foreclosure states. Judicial foreclosure requires lengthy and costly court action to foreclose and dispose of a property, whereas non-judicial foreclosure avoids such court action, enabling the lender to quickly foreclose on a property. Second, an analysis of resolution outcomes in the abovementioned study revealed that distressed portfolio loans were more likely to be liquidated or foreclosed and less likely to be restructured than CMBS loans. In contrast, CMBS loans were less likely to be liquidated or foreclosed and more likely to be restructured. Third, results of the study by Downs and Xu (2015) indicated that portfolio loans experienced shorter time to foreclosure and resolution relative to CMBS loans. Moreover, properties located in judicial foreclosure states experienced longer time to foreclosure and resolution relative to properties located in non-judicial states. Finally, the researchers found portfolio loans had significantly higher recovery rates than CMBS loans, while properties in judicial foreclosure states had lower recovery rates than properties located in non-judicial foreclosure states. However, these results only hold when distressed loans were resolved swiftly within six months; otherwise, the overall results on recovery rates for portfolio loans and judicial foreclosure were insignificant. These insignificant findings on recovery rates provide the basis to further examine whether hotels collateralized by CMBS have higher recovery rates than hotels secured by portfolio loans. Although the current study does not investigate the likelihood of default or resolution, it does provide evidence for the effects of financial distress on resolution prices, time to resolution, and recovery rate using an industry-specific sample of distressed hotels. To provide a basis for comparison, the current study includes several of the determinants identified in prior research to assess their association with resolution prices, resolution time, and recovery rate within the lodging industry. Hospitality-Related Research There has been very little hospitality research using commercial mortgages and CMBS loan data because there is a

42 Cornell Hospitality Quarterly 58(1) limited amount of data availability. Nagpal and Sheel (2002) provided a basic explanation of the securitization process, the typical CMBS structure, and the various parties involved in the securitization process. However, their study is limited to comparing and contrasting various features of CMBS loans. Quan and Lebret (2006) estimated a structural model of securitized hotel mortgages to explain the delinquency and default decision of hotels. Their findings revealed that the level of distress was associated with a lower likelihood of resolving a loan. In another study, Corgel and Walls (2010) analyzed a sample of 365 hotels to determine the future delinquency of hotel CMBS loans. The analysis involved estimating NOI, DCRs, and debt service payments as the primary indicators of loan delinquency. Corgel and Walls (2010) concluded that deterioration in hotel DCRs would reach a peak in 2010 before gradually improving in 2011. Although Corgel and Walls s (2010) study was limited by sample size, Quan and Lebret s (2006) research was conducted using data prior to 2004, well before the current financial crisis of 2008-2009. A few hospitality studies have examined the predictors of hotel values using non-distressed hotel transaction prices. For example, Corgel and deroos (1993) compared various models for predicting property values. The results of their analysis indicated that property characteristics such as property age, number of rooms, and facilities had statistically significant relationships with room rates and sale prices. Surprisingly, location was an insignificant factor. In another study of 327 hotel sales transactions from 1990 through 2002, O Neill (2003) found average daily rate (ADR), NOI, occupancy, and date of sale to be significant predictors of hotel sale prices; however, age of the property was not. Subsequently, using the same dataset, O Neill (2004) confirmed that NOI, ADR, number of rooms, and occupancy rate were the most important determinants of hotel sale prices, explaining the 90 percent variation in sale price per room. However, hotel type, location, year open, property age, and year of sale were insignificant predictors. O Neill and Xiao (2006) further extended O Neill s (2004) study to conclude that brand affiliation was also significant in explaining hotel market values. In addition, the association between renovations and hotel prices is important. Using a dataset of 3,810 hotel real estate transactions between 1996 and 2004, Corgel (2008) concluded that hotel renovations countered obsolescence. In contrast to O Neill s (2004) study, published room rate, property age, location, chain-scale or property type, and year of sale were found to be important determinants of hotel prices in Corgel s (2008) study. In summary, the evidence on the determinants of nondistressed hotel prices appears to be mixed. Even though studies have shown that some hotel characteristics do influence the sale prices of non-distressed hotels and some factors do predict the likelihood of distress, it is unclear which factors predict the outcome of distressed hotels. By focusing on distressed hotels and using similar predictors as those used in prior studies, the current study extends the hotel real estate literature on the market values. As financial distress is costly for investors, owners, and lenders, it is important to empirically investigate the effects of securitization and foreclosure on distressed hotel prices. Given the scarcity of empirical studies, the current study makes an important contribution to the growing literature on distressed commercial real estate and its resolution of financial distress. Method and Procedure Definition of Variables There is no consensus on how financial distress or default is defined in the existing commercial mortgage literature. For the purpose of this study, financial distress is defined broadly as the financial challenges and difficulties hotels face in meeting their debt service payments. This broad definition of financial distress captures and describes the stages and sequence of events that occur prior to delinquency (e.g., short sale, distress sale) on through delinquency and default (e.g., transfer to special servicer, maturity default, foreclosure) to final resolution outcome or liquidation (e.g., REO sale, auction sale). Exhibit 1 outlines the sequence of events and resolution of financial distress. Based on prior research (Corgel, 2008; Downs & Xu, 2015), dummy variables were used to capture the effect of securitization, foreclosure, location in a judicial foreclosure state, resolution outcome (e.g., REO sale), and various hotel characteristics. Sample Data The sample for this study spans the period from the beginning of 2008 to year-end 2014. Lodging financial distress data were obtained from three major providers of commercial mortgage loan information: Real Capital Analytics (RCA), CoStar, and Trepp, LLC. RCA tracks commercial property investment trends and transactions including performance and resolution of distressed commercial mortgage loans from the time they are in distress until the time they are resolved. Resolved status is the resolution of the distress condition via disposition/liquidation such as an REO sale, auction/trustee sale, and note sale. Moreover, a loan could be resolved by a distress sale, receivership sale, bankruptcy sale, or short sale without going through the foreclosure process. Dummy variables were used to designate these distressed conditions. Transaction prices were obtained from CoStar, a major provider of comprehensive commercial real estate information while Trepp was the leading source for securitized hotel information. Information from Trepp was used as a

Singh 43 cross-check against RCA data to avoid any potential misclassification and to ensure all financially distressed hotels securitized by CMBS were accurately identified and classified. This study excluded multi-property loans in the empirical analysis due to the lack of individual property sale prices. A dummy variable equal to 1 was used for a hotel secured by CMBS and a variable equal to 0 was used to identify a hotel collateralized by a portfolio loan. A portfolio loan is originated and retained by a lender such as a bank, insurance company, or another third party unlike CMBS in which loans are pooled together and placed in a trust and the bonds backed by the loans are sold to investors. Similarly, a dummy variable that is equal to 1 was used to define a property located in a state with a judicial foreclosure process. In a judicial foreclosure, the lender files a lawsuit against the borrower in state court and obtains a judgment to foreclose and auction the property to recoup any unpaid mortgage debt. However, a non-judicial foreclosure does not require any court action for the lender to foreclose on a property. The power of sale provision in the mortgage or deed of trust authorizes the public trustee to sell the property in a public auction/trustee sale in the event of default. The nonjudicial process is thus quicker and less costly for the lender. Finally, distress information was matched with STR s census database to obtain information on hotel characteristics including type of operation, market, affiliation, location, and scale. The final sample comprised a total of 4,763 financially distressed hotels (3,821 observations from RCA and 942 from CoStar), representing 9.2 percent of an estimated 52,000 active U.S. hotels with more than twenty rooms that are tracked by STR. Of the total 4,763 properties, 488 hotels were still in financial distress as of year-end 2014, 446 hotels were foreclosed on and in the hands of lenders (lender REO), 344 hotel loans were modified or restructured, 381 distressed conditions were resolved though unknown reasons, and 44 were resolved through short sales. The remaining 3,060 financially distressed hotels were resolved through sale to third parties including REO sale (1,138), auction/trustee sale (1,702), distress sale (204), and note sale/debt assumption (16). Out of the total 4,763 distressed properties, 2,623 or 55.1 percent were also subject to foreclosure (2,522) and deed-in-lieu of foreclosure (101). Model Specification A fundamental question raised in this study is whether hotel characteristics that were found to be significant in explaining non-distressed hotel prices can also explain distressed hotel prices. In addition, this study investigates the effects of securitization and foreclosure on distressed prices. A multiple regression model is used in which distressed transaction prices are specified as a function of securitization, foreclosure, and various hotel characteristics such as hotel size, age, location, and scale (Corgel, 2008; O Neill, 2004). More specifically, the multiple regression model takes the following functional form: Price = β + β Rate + β Size + β Age + β FCL i 0 1 i 2 i 3 i 4 i + β CMBS + β State + β Location 5 i 6 + β Scale + ε, 8 i i i 7 i where price is the natural log of distressed sales price and rate is the single high average room rate as defined in the STR census database. Size is the hotel size measured as the natural log of the number of rooms. Age is the property age from the time hotel was built till year-end 2014. Foreclosure is a foreclosure dummy variable that is equal to 1 if foreclosure has been filed or completed or if it is a deed-in-lieu of foreclosure; otherwise, the dummy variable is equal to 0. Commercial mortgage-backed securities (CMBS) is a securitization dummy variable that is equal to 1 if a hotel is collateralized by CMBS; otherwise it is 0. State is a judicial foreclosure dummy variable that is equal to 1 if the hotel is located in a judicial foreclosure state; otherwise, it is 0. Location is a categorical variable as defined in the STR census database such as urban, suburban, airport, and the like. Scale is a categorical variable that defines the chain-scale of a hotel as defined by STR. The ε is error term. In addition, a categorical variable is included to represent how the hotel is operated (chain managed, franchised, independent) and a dummy variable to represent an REO sale as equal to 1 if the hotel was disposed via an REO sale; otherwise, it is 0 for an auction/trustee type sale. To further examine the effects of these determinants on resolution time and recovery rate, the dependent variable (distressed sale price) in Equation 1 is replaced with two additional variables, resolution time and recovery rates. Specifically, resolution time is defined as the length of time (months) required to final resolution of a distressed hotel, while recovery rate is measured as the distressed sales price divided by the outstanding mortgage loan balance at time of default. Results Hotel Frequency Statistics Active hotels accounted for 91 percent of the 4,763 hotels as shown in Exhibit 2. More than half of the sample comprised franchised hotels followed by independent and chainmanaged hotels, which accounted for almost equal proportions each. By combining franchised and chain-managed hotels, it can be observed that 76 percent of the sample of distressed hotels is composed of branded hotels with the remaining 24 percent representing independent hotels. Financial distress was evident across all hotel segments particularly in the economy and independent segments, which accounted for 53 percent of the distressed sample. Most of (1)

44 Cornell Hospitality Quarterly 58(1) Exhibit 2: Distressed Hotel Frequency Statistics. the independent hotels were relatively small properties with many of them ceasing operations immediately following the financial crisis. The relatively high proportion of economy hotels that were distressed is due to the bankruptcy and foreclosure of a few large hotel portfolios including Extended Stay America and Red Roof Inns. Half of distressed hotels were located in suburban areas followed by small metro towns and urban locations. The south Atlantic region was the most affected by the financial crisis followed by the east central region. The top five states with the most distressed hotels were Florida (12.95%), California (12.85%), Texas (7.12%), Georgia (5.31%), and Illinois (3.93%). Hotel Descriptive Statistics Hotels Percent Status Active 4,348 91.3 Closed 415 8.7 Operation Chain management 1,047 22.0 Franchise 2,569 53.9 Branded total 3,616 75.9 Independent 1,147 24.1 Chain-scale Economy 1,380 29.0 Mid-scale 553 11.6 Upper mid-scale 823 17.3 Upscale 457 9.6 Upper upscale 336 7.0 Luxury 67 1.4 Independents 1,147 24.1 Location Airport 376 7.9 Interstate 365 7.7 Resort 455 9.5 Small metro town 618 13.0 Suburban 2,386 50.1 Urban 563 11.8 Region New England 155 3.2 East central 951 20.0 West central 686 14.4 Middle Atlantic 322 6.8 South Atlantic 1,393 29.2 Mountain 476 10.0 Pacific 780 16.4 Note. Frequencies are based on a total sample size of 4,763 distressed hotels. Exhibit 3 provides the descriptive statistics of the distressed hotels. The median represented the appropriate measure for Exhibit 3: Descriptive Statistics for All Distressed Hotels. N M Median SD Size (rooms) 4,763 146 110 164 Age (years) 4,635 27 23 19 Distress sale price 2,089 12.659 3.1 55.804 (US$ in millions) Distress price/room 2,089 58,858 30,909 91,824 (US$) Troubled loan amount 3,670 16.787 5.642 56.146 (US$ in millions) Troubled loan 3,670 80,695 54,230 143,055 amount/room (US$ in millions) Recovery rate 1,328.950.754 1.199 Resolution time (months) 3,821 15.2 16 13.9 Note. Variation in the number of observations is due to missing data. Recovery rate is computed as the disposition sales price divided by the distressed mortgage loan balance at the time of default. Recovery rate can only be computed for 1,328 hotels where both a disposition price and troubled loan amount are available for a resolved loan. Resolution time is computed as the number of months from the loan becoming distressed to its final resolution. describing the data because of missing observations from non-disclosure of distressed prices. In Exhibit 3, the median distressed hotel was twenty-three years of age and with 110 rooms, reflecting the propensity of smaller and older hotels to be in financial distress during a downturn. Distressed hotels were sold at a median price of $3.1 million or $30,909 per room. The median outstanding troubled loan amount at the time of default was $5.64 million or $54,230 per room. Consequently, the median recovery rate (sales price/troubled loan amount) was approximately 75.4 percent based only on those cases where both the disposition price and distressed loan amount were available. The resolution time or the time when a hotel went into distress until the time it was resolved through liquidation was approximately sixteen months. Resolution Prices The regression results of the determinants of distressed prices are presented in Exhibit 4. The initial test of the models included a dummy variable to proxy for the year of sale but this variable was found to be insignificant and therefore excluded from the analysis. Hotel size was controlled using two alternative model specifications with natural log transformation, a common approach used in prior hospitality research for normalizing size and price (Corgel, 2008; O Neill & Xiao, 2006). The results using both specifications were qualitatively similar but one functional form was a better fit over the other. In Model 1, the dependent variable (disposition price) was transformed using the natural log of

Singh 45 Exhibit 4: Regression Analysis of Determinants of Distressed Prices. Dependent Variable Model 1 Ln (Price/Room) Model 2 Ln (Price) Model 3 Ln (Price) Model 4 Ln (Price) Constant 10.05*** 11.25*** 11.34*** 10.98*** Rate 0.005*** 0.005*** 0.005*** 0.005*** Size 0.726*** 0.725*** 0.807*** Age 0.004*** 0.004*** 0.004*** 0.005*** FCL 0.113** 0.114** CMBS 0.072 0.114*** 0.145*** 0.224*** State 0.177*** 0.143*** 0.168*** 0.157*** Location Suburban 0.218*** 0.254*** 0.329*** 0.395*** Airport 0.265*** 0.229** 0.314*** 0.385*** Interstate 0.493*** 0.568*** 0.686*** 0.837*** Resort 0.294*** 0.270*** 0.198* 0.127 Small metro town 0.240*** 0.377*** 0.447*** 0.527*** Scale Luxury 0.550** 0.742*** 0.717*** Upper upscale 0.570*** 0.846*** 0.801*** Upscale 0.489*** 0.663*** 0.636*** Upper mid-scale 0.305*** 0.357*** 0.313*** Mid-scale 0.076 0.123** 0.109* Economy 0.143** 0.134** 0.153** REO sale 0.139*** 0.136*** Operation Chain managed 0.434*** Franchise 0.169*** N 1,731 1,731 1,494 1,494 F value 56.63*** 74.97*** 151.01*** 158.52*** R 2.430.690.688.668 Mean VIF 1.54 1.57 1.56 1.54 Note. Dependent variable in Model 1 is the natural log of disposition price per room. Dependent variable in Models 2, 3, and 4 is the natural log of the unscaled disposition price. Rate is the single high average daily rate as defined in the STR census database. Size is measured as the log of the number of hotel rooms. Age is measured from the time property was built until 2014. FCL is equal to 1 if foreclosure proceedings have been initiated or completed or if it is a deed-in-lieu of foreclosure, otherwise 0. CMBS is a dummy variable that is equal to 1 if a property is securitized by CMBS, otherwise zero. State is equal to 1 if a property is located in a judicial foreclosure state, otherwise 0. Location is a categorical variable as defined in the STR census database with urban as the base reference group. Scale is a categorical variable that defines the chain-scale of hotels in the STR census database with independent hotels as the base reference group. REO sale is a dummy variable that is equal to 1 if a property is disposed via an REO sale, otherwise 0. Operation is a categorical variable as defined in the STR census database with independent hotels serving as the base reference group. VIF is an indicator of multicollinearity. FCL = foreclosure; CMBS = commercial mortgage-backed securities; REO = real estate owned; VIF = variance inflation factor; STR = Smith Travel Research. ***,**, * Significance at 1%, 5%, and 10% levels, respectively. selling price per room (selling price/number of rooms) in which hotel size entered the equation as a scaled dependent variable consistent with prior research (Corgel, 2008; O Neill & Xiao, 2006). However, the results of this specification in Model 1 of Exhibit 4 show that only 43 percent of the variation in the dependent variable is explained. All of the coefficients of the determinants, with the exception of the CMBS proxy, were found to be statistically significant at the 5 percent level. The explained variation in Model 1 is lower than the 50 percent explained variation in Corgel (2008) and substantially lower than the 90 percent explained variation documented by O Neill (2004). In Model 2, the dependent variable is expressed as the natural log of the distressed sales price, while hotel size is introduced as a predictor variable and measured as the natural log of the number of hotel rooms. This alternative functional form substantially increased the explained variation in the dependent variable from 43 percent in Model 1 to 69 percent in Model 2 with all explanatory variables significant at the 5 percent level. Corgel (2008) noted that hotel size often accounted for 30 percent or more of the variation in hotel sales prices. Given the non-availability of NOI, ADR, and occupancy rate that

46 Cornell Hospitality Quarterly 58(1) were used by O Neill (2004), the explained variance in this study is substantial considering the use of distressed hotel prices instead of non-distressed prices. Moreover, the results of this study suggest that variations in distressed prices are better captured by the regression model when size appears as a transformed predictor in the equation instead of a scaled dependent variable. The dependent variable in Model 2 is also consistent with the standard semi-log hedonic pricing models widely used in real estate for its ease of estimation in which the natural logarithm of the selling price is often modeled as a function of property and location characteristics. The main results of this study are presented in Models 2 to 4 in Exhibit 4. Model 3 differs from Model 2 by the exclusion of the foreclosure dummy variable and its replacement by an alternative dummy variable for REO sales to investigate whether REO disposal prices were different from auction/trustee sale prices. The foreclosure variable was excluded in Model 3 to control for multicollinearity because all REO sales included foreclosure by the lender. To test whether distressed prices were influenced by location, a categorical variable to proxy for location was introduced with an urban location serving as the base reference group. Model 4 differs from Model 3 by the substitution of the chain-scale categorical variable with an alternative categorical variable for hotel operation to test whether the type of hotel operation (franchise or chain managed) was associated with distressed property prices. Robust standard t statistics were used in all models to mitigate the presence of heteroskedasticity while multicollinearity was tested in each model using variance inflation factors. The regression analysis in Exhibit 4 revealed several interesting findings. First, hotel rate was significantly and positively related to distressed selling prices at the 1 percent level of confidence. If hotel room rates are increased by $1, the expectation is that the distressed price will increase by 0.5 percent. This result is consistent with previous research rate has a statistically positive relationship with hotel sales prices (Corgel, 2008; Corgel & deroos, 1993; O Neill, 2004). Thus, distressed hotels with higher average room rates will sell at higher disposal prices than hotels with lower room rates. The results are qualitatively similar when the published double room rate is used as an alternative variable to the published single room rate. Second, hotel size measured as the number of rooms was statistically significant and positively associated with distressed sales price. A 1 percent increase in the number of rooms is associated with a 0.73 percent increase in the distressed price. This result is also consistent with prior research that the number of rooms is an important predictor of sales price (Corgel & deroos, 1993; O Neill, 2004). Thus, a larger a hotel is likely to command a higher distressed sale price, which makes intuitive sense if the hotel is in good operating condition. Furthermore, the higher distressed sale price can be explained by the fact that a larger hotel tends to be more full-service oriented, provides more amenities (O Neill, 2004), costs more to build, and is inherently more complex to manage and operate. Third, age was significantly and negatively correlated with distressed sales prices. A one- year increase in the age of the hotel will decrease prices by 0.4 percent implying that older properties will sell at lower distressed prices. The older the hotel, the lower is the disposal price. Buyers are less likely to pay premium prices for older distressed hotels and more likely to pay higher amounts for newer hotels. This finding on age is also consistent with Corgel s (2008) finding that prices decline with age but differs from O Neill s (2004) study, which found age to be an insignificant predictor of sales prices. The empirical results were qualitatively similar even when age was measured to the date of sale. Fourth, foreclosure was significantly and negatively correlated with disposition prices. A foreclosure filing can lead to a distressed price that is 11 percent less than the absence of a foreclosure. The initiation of foreclosure proceedings could be a potential stigma as well as an indication of the poor physical and operating condition of the hotel. Fifth, hotel properties that are securitized by CMBS loans were valued significantly more than hotels secured by portfolio loans. The results indicated that distressed hotels securitized by CMBS were priced between 11 percent and 22 percent higher than hotels secured by portfolio loans. The quality, size, and asset values of hotels collateralized by CMBS tend to be higher than portfolio hotels. Furthermore, transparency and availability of loan-level information in CMBS transactions as well as the role of the special servicer could also contribute to higher resolution prices for hotels secured by CMBS. Sixth, the results showed that distressed hotels located in judicial foreclosure states were significantly more likely to be sold at distressed prices that are between 14 percent and 17 percent lower than hotels located in non-judicial foreclosure states. Court action is required in judicial states to initiate foreclosure proceedings, which may be perceived as a negative factor by buyers, considering the high cost and length of time involved in resolving financial distress in these states. Seventh, the results in Exhibit 4 show significant effects of location on distressed prices. The distressed prices of almost all locations, with the exception of resort properties, were significantly lower than urban properties. Hotels located along interstates and small metro towns are expected to sell at greater discounts than hotels located in urban locations. For example, in Model 2, the discount for an interstate hotel could be 57 percent lower than an urban hotel. Hotels located in urban settings tend to be larger full-service properties that are more complex to operate and consequently are expected to generate

Singh 47 higher distress sales proceeds than hotels in other locations. The results for the location proxy differ from O Neill s (2004) finding that location is an insignificant predictor of non-distressed hotel sales prices. O Neill (2004) explained that variations in sale prices due to location could be captured by monetary measures because hotels in larger metro areas tend to have higher ADRs, and consequently, sell at higher market prices. However, the results from this study provide contrasting evidence that location is a significant predictor of distressed sales prices even after controlling for the hotel room rate. Therefore, buyers of distressed properties must consider location to be an important factor if there is any possibility of turning the hotel around through a major renovation, an upgrade, or a brand conversion. The regression coefficient on resort properties was found to be weaker in Model 3 and insignificant in Model 4 due largely to the reduced number of observations, especially when REO sales were introduced as a predictor in the model. Eighth, the results indicate that chain-scale had a significant influence on prices consistent with previous research by Corgel (2008). It differs from O Neill s (2004) study that hotel type is an insignificant predictor of sales price. With independent hotels serving as the base reference group, the results in Models 2 and 3 of Exhibit 4 show that distressed hotels in the luxury, upper upscale, upscale, and upper mid-scale were resolved at significantly higher prices relative to independent hotels, while economy properties were disposed for significantly greater discounts than independent hotels, even after controlling for hotel size and location. For example, in Model 3, a distressed upper mid-scale hotel could be sold for a price that is 31 percent higher than an independent hotel, while an economy hotel could be disposed at a price that is 15 percent lower than an independent hotel. Even though distressed prices of mid-scale hotels could be higher than independent hotels, this finding is inconclusive in Model 3. To investigate the effect of the resolution outcome on distressed prices, the foreclosure proxy in Models 3 and 4 was replaced by an alternative dummy variable to designate an REO sale. The results in Models 3 and 4 provide evidence that resolution prices via an REO sale were significantly lower than an auction/trustee sale. The REO sale price was 14 percent lower than an auction/trustee sale. Finally, the effect of hotel operations on prices was tested using a categorical variable with independent hotels serving as the base reference group. The results of Model 4 in Exhibit 4 show distressed sale prices of both chain-managed and franchised hotels were significantly higher than independent hotels. For example, the distress price of a franchised hotel is 17 percent higher than an independent hotel, which suggests some value-added benefit from branding. Resolution Time The second part of the study involved an investigation of the effects of the determinants on resolution time and recovery rate. The regression results for resolution time (months) are presented in Exhibit 5. The results in Exhibit 5 reveal a low explained variance in resolution time. Property size is significantly and positively associated with resolution time. For example, a 10 percent increase in the number of rooms will prolong resolution time by an estimated 0.3 months (log 1.10*3.1) based on Model 3. Property age is marginally significant indicating an increase in resolution time for older hotels. The presence of foreclosure is also found to be positively significant. Holding all else constant, the presence of foreclosure is expected to take about a year longer than the absence of foreclosure. The effect of securitization is significantly and positively related to resolution time. A distressed hotel that is collateralized by CMBS is resolved between five to eight months longer than a portfolio hotel. Distressed mid-scale and upper mid-scale hotels are resolved about five months less than independent hotels. Similarly, a distressed franchised hotel is resolved in four months less time than an independent hotel. The results also show that an REO sale will take up to eight months longer to resolve than an auction/trustee sale. Recovery Rate The findings on the determinants of recovery rate in Exhibit 6 show a lower explained variance for the recovery rate. Hotel size is significantly and negatively related to recovery rate. A 10 percent increase in the number of rooms will lead to a 1 percentage point (log 1.1*0.10) decline in the recovery rate. The combined results of Exhibits 5 and 6 show an increase in hotel size not only prolongs resolution time but also reduces the recovery rate. The initiation of foreclosure is marginally related to the recovery rate implying that foreclosure will reduce the recovery rate by an estimated 15 percentage points relative to the absence of foreclosure. Taken together, findings in Exhibits 5 and 6 indicate that foreclosure not only increases resolution time but also decreases the recovery rate. Similarly, the effect of securitization on the recovery rate is significant in Models 1 and 2 but not in Models 3 and 4. Hotels securitized by CMBS had an 18 percentage point higher recovery rate than portfolio loans. The combined results indicate that although hotels collateralized by CMBS take a longer time to resolve than portfolio hotels, these hotels will also generate a higher recovery rate than portfolio loans. In general, the judicial foreclosure process is not associated with resolution time; however, it is negatively related to the recovery rate. Judicial foreclosure has the effect of reducing the recovery rate between 22 and 27 percentage points compared with nonjudicial foreclosure.

48 Cornell Hospitality Quarterly 58(1) Exhibit 5: Regression Analysis of Determinants of Resolution Time. Dependent Variable Model 1 (Months) Model 2 (Months) Model 3 (Months) Model 4 (Months) Constant 10.82*** 10.97***.852 1.08 Rate 0.001 0.001 0.004 0.006 Size 2.91*** 2.75*** 3.08*** 3.07*** Age 0.027* 0.030* 0.006 0.002 FCL 12.03*** 11.95*** CMBS 5.24*** 5.35*** 7.95*** 7.40*** State 1.10* 0.873 0.820 0.878 Location Suburban 1.95** 0.734 0.611 Airport 0.457 0.819 1.14 Interstate 0.723 0.217 0.197 Resort 1.93 1.03 0.841 Small metro town 0.084 0.669 0.788 Scale Luxury 0.576 Upper upscale 2.02 Upscale 1.64 Upper mid-scale 5.26*** Mid-scale 4.94*** Economy.006 REO sale 7.69*** 7.74*** Operation Chain managed 2.02 Franchise 3.54*** N 2,531 2,531 1,393 1,393 F value 88.95*** 51.80*** 15.28*** 19.15*** R 2.159.164.148.146 Mean VIF 1.07 1.46 1.58 1.52 Note. Dependent variable is the resolution time measured in months. Rate is the single high average daily rate as defined in the STR census database. Size is measured as the log of the number of hotel rooms. Age is measured from the time property was built until 2014. FCL is equal to 1 if foreclosure proceedings have been initiated or completed or if it is a deed-in-lieu of foreclosure, otherwise 0. CMBS is a dummy variable that is equal to 1 if a property is securitized by CMBS, otherwise zero. State is equal to 1 if a property is located in a judicial foreclosure state, otherwise 0. Location is a categorical variable as defined in the STR census database with urban as the base reference group. Scale is a categorical variable that defines the chainscale of hotels in the STR census database with independent hotels as the base reference group. REO sale is a dummy variable that is equal to 1 if a property is disposed via an REO sale, otherwise 0. Operation is a categorical variable as defined in the STR census database with independent hotels serving as the base reference group. VIF is an indicator of multicollinearity. FCL = foreclosure; CMBS = commercial mortgage-backed securities; REO = real estate owned; VIF = variance inflation factor; STR = Smith Travel Research. ***,**, * Significance at 1%, 5%, and 10% levels, respectively. In terms of location, only airport and resort properties had a significantly negative association with the recovery rate. Specifically, the recovery rate for resort hotels is expected to be 21 percentage points lower than urban hotels. With regard to chain-scale, recovery rates for upscale hotels were significantly higher (36 percentage point) than independent hotels. The unique locations of these properties, the uncertainty and sufficiency of demand generators, and the heavy focus on leisure guests could contribute to the lower recovery rates for airport and resort hotels. The significantly higher recovery rate for distressed upscale hotels is an indication of their high value to prospective buyers. These hotels can be converted into select service branded hotels that are more efficient and profitable. Finally, an REO sale is negatively and significantly associated with the recovery rate. An REO sale has a 14 percentage point lower recovery rate than an auction/trustee sale. The combined results from Exhibits 5 and 6 show that REO sales not only take a significantly longer time to resolve but also contribute to a lower recovery rate. The empirical results of this study are consistent with Downs and Xu s (2015) finding that securitized loans experienced a significantly longer time to resolution than portfolio loans. The prior study found that properties in judicial foreclosure states experienced significantly longer time to resolution and foreclosure; however, the results of

Singh 49 Exhibit 6: Regression Analysis of Determinants of Recovery Rate. Dependent Variable Model 1 (Recovery) Model 2 (Recovery) Model 3 (Recovery) Model 4 (Recovery) Constant 1.30*** 1.54*** 1.58*** 1.47*** Rate 0.001* 0.001 0.000 0.000 Size 0.093** 0.104** 0.108** 0.086** Age 0.001 0.001 0.001 0.000 FCL 0.153* 0.144* CMBS 0.180** 0.177** 0.075 0.093 State 0.233*** 0.217*** 0.265*** 0.268*** Location Suburban 0.192* 0.170* 0.177* Airport 0.294*** 0.230** 0.247** Interstate 0.142 0.112 0.134 Resort 0.218** 0.210** 0.210** Small metro town 0.237* 0.167 0.184 Scale Luxury 0.276 Upper upscale 0.096 Upscale 0.362*** Upper mid-scale 0.115 Mid-scale 0.235 Economy 0.011 REO sale 0.143** 0.140** Operation Chain managed 0.150 Franchise 0.146 N 1,099 1,099 949 949 F value 9.86*** 6.52*** 4.48*** 5.72*** R 2.028.033.033.028 Mean VIF 1.09 1.41 1.54 1.47 Note. Dependent variable is the recovery rate computed as the disposition sales price divided by the distressed mortgage loan balance at the time of default. Rate is the single high average daily rate as defined in the STR census database. Size is measured as the log of the number of hotel rooms. Age is measured from the time property was built until 2014. FCL is equal to 1 if foreclosure proceedings have been initiated or completed or if it is a deed-in-lieu of foreclosure, otherwise 0. CMBS is a dummy variable that is equal to 1 if a property is securitized by CMBS, otherwise 0. State is equal to 1 if a property is located in a judicial foreclosure state, otherwise 0. Location is a categorical variable as defined in the STR census database with urban as the base reference group. Scale is a categorical variable that defines the chain-scale of hotels in the STR census database with independent hotels as the base reference group. REO sale is a dummy variable that is equal to 1 if a property is disposed via an REO sale, otherwise 0. Operation is a categorical variable as defined in the STR census database with independent hotels serving as the base reference group. VIF is an indicator of multicollinearity. FCL = foreclosure; CMBS = commercial mortgage-backed securities; REO = real estate owned; VIF = variance inflation factor; STR = Smith Travel Research. ***,**, * Significance at 1%, 5%, and 10% levels, respectively. this study are inconclusive on this finding. Finally, the findings of this study differ from the prior study on the recovery rate. Downs and Xu (2015) reported that portfolio loans (and properties located in judicial foreclosure states) had significantly higher (lower) recovery rates, but only if they were resolved within six months of becoming troubled; otherwise their findings were insignificant. On the contrary, this study showed that hotel securitized by CMBS had higher recovery rates than portfolio hotels, while hotels located in judicial foreclosure states had significantly lower recovery rates than hotels located in nonjudicial foreclosure states. Summary and Conclusion The results of this study provided empirical evidence on the determinants of distressed hotel prices, resolution time, and recovery rate in a sample of distressed U.S. hotels. An estimated 4,763 total active U.S. hotels (greater than twenty rooms) fell into various states of financial distress immediately following the financial crisis. Three-fourths of these distressed hotels were branded properties, while the rest were independent hotels. Many of these distressed properties were concentrated within the lower priced segments in suburban locations, especially in the economy and budget