University of Southern California Dissertations and Theses

Housing market crisis and underwater homeowners

(USC Thesis Other)

Housing market crisis and underwater homeowners

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content HOUSING MARKET CRISIS AND UNDERWATER HOMEOWNERS By Jung Hyun Choi _____________ A Dissertation Presented to FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirement for the Degree DOCTOR OF PHILOSOPHY (PUBLIC POLICY AND MANAGEMENT) August 2015 Dissertation Committee Members: Gary Painter, Chair USC, Sol Price School of Public Policy Richard Green USC, Sol Price School of Public Policy & Marshall School of Business John Matusaka USC, Marshall School of Business Copyright 2015 Jung Hyun Choi i Acknowledgments I would like to express my deepest gratitude to my advisor, Gary Painter, for his guidance throughout this research. As a doctoral student at USC, I have always been thankful that I was able to meet Gary, who has provided both intellectual and emotional support during the past five years. I would also like to extend my sincere appreciation to two of my committee members: Richard Green and John Matsusaka. For the past three years, I was very fortunate to meet them almost every week while writing my dissertation. All three of my committee members are outstanding scholars and my lifelong role models and I will not forget their whole hearted concern towards their students, including myself, as a teacher. I also thank my family and friends back in Korea and my family here in Los Angeles, at the Power of Praise Church and the Korean Bible Study. I would not have been able to complete this long journey without all of their love and support. Finally, I would like to thank the One who created me, who guided and helped each and every step of my life and brought such wonderful people into my life. ii Table of Contents Page Abstract ............................................................................................................................ vi I. Underwater Homeowners: Race, Ethnicity and Immigrant Status ..................... 1 I.1. Introduction .................................................................................................. 1 I.2. Previous Literature – Reason for Being Underwater .................................... 4 I-3. Data .............................................................................................................. 7 I.4. Methodology ................................................................................................. 13 I-5. Empirical Findings ....................................................................................... 17 II. Housing Market Shock and Mobility of Underwater Homeowners: Why Do Underwater Homeowners Move? ....................................................................... 35 II-1. Introduction ................................................................................................. 35 II-2. Previous Literature ...................................................................................... 38 II-3. Data ............................................................................................................. 43 II-4. Methodology ............................................................................................... 51 II-5. Empirical Findings ...................................................................................... 57 III. Measurement, Perceptions and Negative Equity-Mobility Relationship ............ 72 III-1. Introduction ............................................................................................... 72 III-2. Previous Literature .................................................................................... 75 III-3. Data ............................................................................................................ 79 III-4. Methodology .............................................................................................. 86 III-5. Empirical Findings ..................................................................................... 90 IV. Conclusion .......................................................................................................... 110 References ....................................................................................................................... 114 Appendix ......................................................................................................................... 124 iii List of Tables Page Chapter I. Table I-1. Summary Statistics ........................................................................................ 9 Table I-2. Model (1): Likelihood of Being Underwater ................................................ 20 Table I-3. Model (2): Likelihood of Being Underwater since 2007 .............................. 22 Table I-4. Likelihood of Being Underwater: Immigrant-Race Groups ......................... 26 Table I-5. Likelihood of Being Underwater: Immigrant by Years in the U.S. ............... 28 Table I-6. Model (3): Likelihood of Being Underwater-Neighborhood Composition Effect .............................................................................................................. 30 Table I-7. Model (4): Likelihood of Being Underwater since 2007- Neighborhood Composition Effect ........................................................................................ 34 Chapter II. Table II-1. Summary Statistics ....................................................................................... 47 Table II-2. Difference in Differences: Total Moves ....................................................... 57 Table II-3. Difference in Differences: In-State & Out-State Moves .............................. 60 Table II-4. Extent of Home Equity and Mobility ........................................................... 61 Table II-5. Double Trigger Effect: 1999-2011 ............................................................... 63 Table II-6. Multinomial Logit: Trading Up, Trading Down and Switching to Rental Housing ......................................................................................................... 65 1 Table II-7. House Price Expectation ............................................................................... 67 Table II-8. Moving to a Better Job Market? ................................................................... 70 Chapter III. Table III-1. Summary Statistics ...................................................................................... 81 Table III-2. No. of Underwater Homeowners: Rep. LTV vs. Est. LTV (MSA) ............. 83 Table III-3. Likelihood of being in the Same LTV Category ......................................... 92 Table III-4. Difference between Rep. & Est. House Value: Est. LTV>100% ................ 94 iv Table III-5. Difference between Rep. & Est. House Value: 8 LTV Categories ............. 95 Table III-6. Matching Est. and Rep. LTV ........................................................................ 97 Table III-7. Difference between Rep. & Est. House Value: 4 Homeowner Groups ....... 98 Table III-8. Difference in Differences: Likelihood of Moving ....................................... 101 Table III-9. Difference in Differences: Likelihood of Moving: 4 Homeowner Groups .. 103 Table III-10. Likelihood of Switching to Rental Housing .............................................. 106 Table III-11. Likelihood of being Delinquent.................................................................. 108 Appendix Table II-A1. Summary Statistics of Underwater Homeowners B & A 2007 ................. 124 Table II-A2. Difference in Differences: In-State & Out-State Moves: Controls ............ 125 Table II-A3. Likelihood of Trading Up, Down and Switching to Rental Housing ........ 126 Table II-A3. House Price Expectation: MSA HPI .......................................................... 128 Table III-A1. No. of Underwater Homeowners: Rep. LTV vs. Est. LTV (ZIP) ............ 129 Table III-A2. No. of Underwater Homeowners: Est. LTV (MSA) vs. Est. LTV (ZIP) . 129 Table III-A3. Likelihood of being in the Same LTV Category (ZIP) ............................ 130 Table III-A4. Difference between the Rep. & Est. House Value (ZIP) ......................... 131 v List of Figures Page Abstract Figure I-1. Three Topics of the Dissertation ................................................................... vi Chapter I. Figure I-1. AHS Underwater Homeowners by Race & Ethnicity .................................. 10 Figure I-2. AHS Underwater Homeowners by Immigrant Status .................................. 11 Figure I-3. AHS Underwater Homeowners by Race & Immigrant Status ..................... 12 Figure I-4. Marginal Effects by Race & Ethnicity: Before & After 2007 ...................... 23 Figure I-5. Marginal Effects by Immigrant Status: Before & After 2007 ...................... 24 Figure I-6. Marginal Effects by Race and Immigrant Status: Before & After 2007 ...... 27 Chapter II. Figure II-1. Changes in the Total Mobility Rate between 1999-2005 and 2007-2011 ... 49 Figure II-2. Changes in the In-State Mobility Rate between 1999-2005 and 2007-2011 50 Figure II-3. Changes in the Out-State Mobility Rate between 1999-2005 and 2007-2011 50 Chapter III. Figure III-1. Percentage of Underwater Homeowners .................................................... 82 Figure III-2. Four Groups of Homeowners: based on Rep. & Est. LTV ......................... 85 vi Abstract The onset of the housing market crisis left a considerable proportion of U.S. households underwater. According to Corelogic, 11.4 million U.S. households had negative home equity as of 2012:Q1, accounting for 23.7% of all residential properties with a mortgage. This increase has gained particular interest from policy makers who became concerned that underwater homeowners may have negative spillover effects on other parts of the economy. Responding to this increased, my dissertation features three independent essays related to the topic of ―negative equity‖ (Figure I-1). 1 Using the Panel Study of Income Dynamics and the American Housing Survey, I examine who underwater homeowners are, how and why they move, and the impact of their mobility decisions on the labor market. I also compare the two most commonly used proxies of negative equity and examine whether the empirical outcomes are sensitive to how the underwater homeowners are defined. [Figure I-1] Three Topics of the Dissertation 1 A mortgage is considered to be underwater when the amount of owed debt exceeds the value of the underlying property. This is status also referred to as having negative equity. vii The first essay investigates who the underwater homeowners are with a special focus on race, ethnicity and immigrant status. So far, not many studies have closely looked into this question and have examined whether the difference in the likelihood of being underwater can be fully explained by observable differences across households with different race and immigrant status. In Chapter II, I first lay out the possible reasons why minorities or immigrants may more likely to be underwater. Not only do the differences across households matter, but the differences in the racial composition across neighborhoods may also have an impact on the probability of household‘s becoming underwater. For example, neighborhoods with greater share of minorities may have experienced a greater boom and the bust as the relaxation of credit during the early 2000s increased the access to homeownership for many people living in these neighborhoods. If so, households dwelling in black or Hispanic neighborhoods may have higher likelihood of being underwater. The results show that prior to the crisis, black and Hispanic households are more likely to be underwater compared to whites. On the other hand, the likelihood of being underwater for Asians is similar to whites when controlling for the demographic and socioeconomic characteristics. Immigrants also have higher likelihood of being underwater relative to the native born. These results hold even after controlling for the time period of purchase as well as the percentage of initial down payment. Before 2007, however, the results show that minorities and immigrants were less likely to be underwater in comparison to their reference groups, although during this period the percent of underwater homeowners was significantly lower than the post- crisis period. Furthermore, even after controlling for the head‘s race, I find that households living in neighborhoods with greater proportion of blacks and Hispanics were more likely to be viii underwater in the post crisis period. In addition, non-Hispanics in Hispanic neighborhoods have higher probability of becoming underwater than Hispanics in the same neighborhood. The findings of the study, suggest that minorities and immigrants have been more adversely affected by the recent housing market collapse which cannot be fully explained by their observable characteristics. The second chapter investigates the causal relationship between negative equity and residential mobility. This question gained particular interest as several media outlets highlighted that negative equity may hamper homeowners from moving to a better job market and as a consequent increase the structural unemployment rate. The recent surge in the unemployment rate has provoked greater concerns about whether negative equity would further delay the recovery of the sluggish labor market. Numerous studies have looked at this issue since the outbreak of the crisis, but most did not find evidence that negative equity reduces mobility. The existing studies, however, have three major limitations. First, they do not properly control for the possible existence of endogeneity that occurs from the households‘ unobserv ed characteristics which affect both their likelihood of moving and their likelihood of being underwater. Second, they do not examine theoretical reasons behind the negative equity-mobility relationship. Finally, they have not identified how the mobility of underwater homeowners actually affects the labor market by looking at where these homeowners move to. I address each of these three issues in chapter II. First, I use the difference in differences framework to examine whether the likelihood of moving changes for households that exogeneously became underwater due to the housing market shock. The results show that the mobility of those who became underwater due to the unexpected house price collapse increased significantly compared to those with positive equity. I also find that the increase in the mobility ix of underwater homeowners is more prominent for the out-state moves. Second, among the existing theories that provide competing reasons for the negative equity-mobility relationship, I find the double trigger and the house price expectation theories best fit the result of the findings. Those who are significantly underwater are more likely to move if they are at the same time unemployed, in accordance with the double trigger theory. Also, as the house price expectation theory suggests, underwater homeowners are more likely to move when they expect their house prices to go down, to avoid further losses. When they expect prices to go up, they are more likely to stay at their current housing, to wait until house prices exceed their mortgage debts. The linkage between negative equity and labor market efficiency, however, still remains uncertain as the findings show that since 2007, underwater homeowners have not moved to states with lower unemployment rates. Thus, how the increase of mobility of those in negative equity influences the macro labor market still needs further exploration. The final chapter looks at the measurement of negative equity. In order to identify those who are underwater, we need the exact value of the house price. However, it is difficult to accurately measure house prices, as the housing market is thin due to infrequent transactions and the heterogeneity of housing units. This study compares the two most commonly used proxies of negative equity: one calculated using the self-reported house value and the other calculated using the house price indices. The first part of the study looks at differences in the two house values and examines whether the estimated value of home equity affects how households report their house values. Next, I investigate whether using different measures of negative equity affect its association with three of the following household behaviors: 1) moving, 2) moving to rental units, and 3) falling behind their mortgage payment. x The results show that many households who are estimated to be underwater using the house price indices, report their house value higher than their mortgage debt, and thus do not classify themselves as underwater homeowners. Also, the gap between the reported and the estimated house price is the greatest for those who are estimated to have negative equity, but report that they have positive equity. The findings can be explained by the loss aversion theory which suggests that people‘s disutility from losses is greater than the utility from the same amount of gains. Next results show that those who are estimated have in negative equity but do not report so have similar likelihood of moving compared to those who are both estimated and reported to have positive equity. On the other hand, those who admit that they are underwater have significantly higher likelihood of moving than those who do not. This again accords with the loss aversion theory, suggesting that those who did not report themselves as underwater homeowners may not be moving to avoid realization of losses. To identify whether the households are correctly reporting their house prices, I further examine households‘ likelihood of swi tching to rental units and their likelihood of being delinquent. Conditional on moving, I find those estimated to be underwater are more likely to move to rental units compared to those whose reported and estimated home equity are both positive. Even those who do not report that they are underwater have significantly higher likelihood of moving to rental units than those estimated to have positive equity. Those estimated to be underwater are also more likely to be behind their mortgage payment, regardless of how they report their home equity value. These results suggest that many of the homeowners who do not report themselves as underwater homeowners are facing greater financial stress due to the drop in their home equity level, and some of them are actually aware of their situation. 1 I. Underwater Homeowners: Race, Ethnicity and Immigrant Status I.1. Introduction The increase of underwater homeowners have gained great attention from policy makers, media and researchers as there have been growing concerns that negative equity may hamper residential mobility 2 and therefore prevent the labor market from making faster recovery. 3 Following the crisis, numerous studies 4 have investigated the relationship between the level of home equity and mobility to identify the existence of the ―lock -in effect‖. Few studies, however, have attempted to identify the characteristics of underwater homeowners and the reason why they became underwater. Since the housing bust, existing studies 5 have found racial disparities in the foreclosure and default rates. These studies show that minorities have higher probability of default or foreclosure compared to whites, even after controlling for households‘ socioeconomic and demographic characteristics. However, not all people who are underwater go through foreclosure or walk away from their home. Although the amount of home equity is suggested as the key determinant of default (e.g., Quigley and Van Order 1995; Quercia et al. 1995), negative equity alone is not a sufficient condition for default or foreclosure (Bhutta, 2010; Elul, 2010). In fact, studies have pointed out that the amount of foreclosures and defaults are significantly lower relative to the number of households that are underwater. This means that many households 2 This phenomenon is called the ―lock -in effect‖. 3 For example, many media outlets, including the Economist have suggested that underwater homeowners may be locked in their current homes. This lock-in effect may further aggravate the labor market conditions by preventing the underwater homeowners from moving to a better job location (Economist, 2010, p. 68). 4 e.g., Ferrier et al., (2010, 2012), Schulhofer-Wohl (2011), Donovan and Schnure (2011), Coulson and Grieco (2013), Demyanyk et al. (2013) and Choi (2014). 5 Bocian et al. (2010), Luea et al. (2011), Bayer et al. (2013) and Chan et al. (2013). 2 continue to pay their mortgage bills even when their mortgage values exceed their current house prices. Why many underwater homeowners chose to stay at their current housing is an interesting question, but I leave this for future research. 6 Before looking at how and why underwater homeowners make certain decisions or prior to designing and implementing policies to help them, it is necessary to examine the homeowners who have been most negatively affected by the recent house market crash. This is the focus of this chapter. Building upon the existing studies, this study is the first to test whether the likelihood of being underwater differs across race and ethnicity groups, before and after the housing market crash. I also investigate whether immigrants are more likely to be underwater. Compared to the native born, immigrants have less resources and information, and thus may be more vulnerable to negative impacts of the recent economic crisis (Painter and Yu, 2013). To examine the differences in the home equity level, I use two data sets, the American Housing Survey (AHS) and the Panel Study of Income Dynamics (PSID) during periods between 2001 and 2011. This period covers the recent boom and the bust, and thus enables me to identify which subgroups of households were most negative affected by sudden changes in the market conditions. The difference-in-differences results show that disparities in the likelihood of being underwater across race, ethnicity and immigrant status mostly occurred during the period after 6 According to the LPS (Lender Processing services), in 2010, the delinquency rate was at its peak, reaching 10 percent. The foreclosure rate was slightly above 4 percent. Some studies have tried to identify the reason why not all underwater homeowners continue to hold on to their mortgages. For example, Foote et al. (2008) argue that it is the rational expectation of future house prices lowers the default rate. When people expect future house prices to increase, they hold on to their current house even when they are underwater. This argument, however, cannot explain why relatively small number of households committed default following the crisis when the expected future house prices turned negative. Meanwhile, White (2010) argues that emotional feelings attached to committing default such as shame, guilt and fear explain why underwater homeowners defaulted less than what the rational economic theory would predict. 3 the crisis. Compared to whites or the native born, the likelihood of being underwater increased significantly for blacks, Hispanics and immigrants since 2007. These differences remain even after controlling for households‘ socioeconomic and demographic characteristics. The Asian- white gap, however, becomes insignificant when the observable variables are included. It is unclear whether the differences in the likelihood of being underwater are due to the unobserved individual characteristics or whether these differences can be explained by the unobserved neighborhood characteristics. At the individual level, there could be unobserved household behaviors or unobserved discrimination in the home purchasing process that affect the household‘s likelihood of being underwater. On the other and, the racial composition of neighborhoods may also affect household‘s level of home equity. For example, minorities and immigrants could have resided in areas that experienced a greater house price drop since 2007. If so, blacks (Hispanics) living in black (Hispanic) neighborhoods would have higher likelihood of being underwater than blacks (Hispanics) living outside of these neighborhood. By controlling for the tract level racial composition, this study further examines whether individual or neighborhood unobservables better explain the gap between the households‘ level of home equity. The findings shows that while households living in black or Hispanic neighborhoods are more likely to underwater since the housing market crisis. However, the likelihood of being underwater for blacks and Hispanic households still remains positive even after controlling for the share of blacks and Hispanics in each census tract, suggesting that both individual and neighborhood effect exists. Compared to black neighborhoods, those residing in Hispanic neighborhoods have greater likelihood of being underwater. Furthermore, I find that non-Hispanics in Hispanic neighborhood have higher likelihood of being underwater than the Hispanics residing in the same neighborhood. 4 There are several explanations of why minorities and households residing in neighborhoods with greater share of minorities are more likely to be underwater, especially since the crisis. By reviewing prior studies, the following section provides possible reasons that explain this phenomenon both at the individual and the neighborhood level. I.2. Previous Literature - Reasons for Being Underwater Even after controlling for socioeconomic and demographic differences, households there may be unobserved factors that lead to differences in the likelihood of being underwater according to race, ethnicity and immigrant status. These variables could exist in both individual and neighborhood level. I first discuss the individual level differences that may affect the likelihood of being underwater. These are (1) differences in the level of initial down payment, (2) differences in the home purchasing prices, and (3) differences in the mortgage costs. Since minorities, on average, have less wealth than the whites, they put lower down payment and thus have lower level of home equity. During the housing boom many minority households were able to buy homes with little or no down payment as credit conditions became less stringent. Homeowners with lower level of initial home equity would more likely to be underwater when house prices fall, thus initial down payment may explain the racial discrepancies in the likelihood of being underwater. Most datasets do not provide any information about the initial level of down payment, and thus this variable is mostly unobserved. Since 2007, however, the AHS started to collect information about the percentage of down payment for every owner occupied household. Thus, in my empirical analysis, I can directly control of level of down payment. 7 The regression results will show whether the differences in 7 In the AHS sample, blacks and Hispanic s place lower down payment while Asians put more down than whites. 5 the level of down payment can fully explain the differences in the likelihood of being underwater across different race and ethnic groups. If minorities are still more likely to be underwater after controlling for down payment, it could be due to differences in the purchasing price. Minorities or immigrants may purchase similar house for a more expensive price compared to whites or the native born. This could be related to either the timing of purchase or discrimination. Many of the minority and immigrant households gained access to the mortgage market during the housing boom when credit conditions relaxed. Thus, these households may have bought houses at more expensive prices than those who bought houses at an earlier time period. Indeed, studies including Schwartz (2010) show that greater share of minorities became homeowners during the period of housing boom. However, since the data provide information about the period of buying, I can control for the differences in the likelihood of being underwater due to the time of purchase. On the other hand, it is difficult to control for the level of discrimination in the housing market. Numerous studies showed that minorities pay higher premium on houses but the empirical findings are still mixed. Earlier studies including King and Mieskowski (1973), Kain and Quigley (1975), and Yinger (1978) find that in comparison to whites, minorities pay a premium for a comparable housing. On the other hand, Follain and Malpezzi (1981) and Kiel and Zabel (1996) show that black buyers receive discounts relative to white buyers. The major challenge for these studies is the difficulty of finding comparable houses of which the only difference is the race of the buyer. If not, the results are likely to be biased. More recently, Bayer et al. (2012) use a unique panel data on repeated-sales housing transactions to find that blacks However, the difference s in the level of down payment between immigrants and the native born is less significant. 6 and Hispanics, on average, pay 3 percent premium on housing, after controlling for the house and neighborhood quality. Finally, minorities and immigrants may be purchasing more expensive mortgages. The existing studies have shown that minority households are more likely to buy subprime mortgages (Mayer & Pence, 2008) which are costly compared to the prime mortgages. Discrimination can also exist in the mortgage market. Some research finds even within the subprime mortgage market, that black and Hispanic households pay higher price relative to non-Hispanic whites (Bocian et al. 2007, Ghent et al. 2014). Households that pay higher mortgage cost have greater chance of becoming underwater when house prices fall. Where individuals live can also affect their likelihood of being underwater. Although many neighborhoods are becoming less segregated, still, people seem to prefer being located in neighborhoods with similar race or ethnicity (Rugh and Massey, 2010). If neighborhoods with greater share of minorities experience greater fall in the house price following the crisis, households living in these neighborhoods would have greater chance of becoming underwater. If so, minorities living in neighborhoods with greater share of minorities would have higher likelihood of being underwater compared to minorities living in neighborhoods with a lower share of minorities. Existing studies have provided some evidence that the house prices in black and Hispanic neighborhoods have experienced a greater fall since the crisis. During the housing market boom, Sufi and Mian (2009) find that mortgage credit expanded in subprime neighborhoods despite the average fall in income. Compared to the prime neighborhoods, these subprime neighborhoods show a higher increase of house prices during the boom and greater fall following the bust. Although the study does not provide information on race or immigrant status, other studies show 7 subprime neighborhoods are likely to have greater share of minority households (Mayer and Pence, 2008). In addition, studies find that households in subprime neighborhoods or black and Hispanic neighborhoods are more likely to default or go through foreclosure (Sufi and Mian, 2009; Chan, Gedal, Been & Haughwout; 2013). The increase of default and foreclosure may further lower the house prices in these neighborhood due to contagion effect (Harding et al., 2009). All these research suggest that households residing in neighborhoods with greater share of minorities are more likely to become underwater. I.3. Data This study uses two data sets, the American Housing Survey (AHS) and the Panel Study of Income Dynamics (PSID). The AHS is conducted by the Census Bureau and is sponsored by the Housing and Urban Development (HUD). The first survey was executed in 1973. Between 1973 and 1981, surveys were conducted annually. But from 1997 they have been conducted biannually. The PSID is directed by the University of Michigan and has followed U.S. households since 1968. The PSID also switched from an annual to a biannual survey since 1997. This study covers time period from 2001 to 2011 which includes periods of unprecedented housing boom and a subsequent bust. A major advantage of both datasets is that they contain extensive information on individual and family characteristics, and thus it is possible to look into various factors associated with the level of home equity. By using the two datasets, I can also complement limitations of each dataset. The AHS is 9 to 10 times larger than the PSID and includes greater number of Asian and immigrant households. Thus, for the main part of the analysis, I use the AHS. Meanwhile, I have obtained the geo-coded PSID, so I can merge tract level variables to the 8 PSID sample. 8 Therefore, I use the PSID for regressions testing whether neighborhoods racial composition matters to household‘s likelihood of being und erwater. For years 2000 and 2010, when the Census offers tract level data, I calculate the percent of black and Hispanic households in each tract. For other years, I smoothed the variables to match it with the PSID sample. For 2011, I use the 2010 data. Table I-1 provides the summary statistics of owner occupied households for the AHS and the PSID sample. All variables are weighted by the sample weights provided in the datasets. The percent of underwater households in the AHS sample is more than twice greater than the percent of underwater homeowners in the PSID sample. One possible reason for this difference is the sample composition. The PSID has greater proportion of white households and less proportion of households who bought their house after 2001. Also, the heads in the PSID, on average, are older and thus has more income. Greater share of household heads completed college education in the PSID sample, while more than half of the household heads in the AHS sample received some level of college education. Also, greater proportion of households refinanced their mortgages in the PSID sample. 8 There are other differences between the two data sets. The AHS sample is closer to the U.S. distribution of homeowners by race and ethnicity, while the PSID over sa mples blacks and low income households. In the PSID, the amount of remaining principle is reported by households. However, the variable is not provided in the AHS and needs to be calculated using the interest and mortgage payment data. The AHS provides codes for calculating the variable. When calculated, the remaining principle can only be obtained for one third of the homeowners in the AHS sample. Thus, in the final dataset, the AHS sample is about 3 times and not 9 to10 times larger than the PSID. 9 [Table I-1] Summary Statistics Variable AHS PSID Mean Std. Dev. Mean Std. Dev. % Underwater 0.080 0.271 0.034 0.181 White 0.748 0.434 0.819 0.385 Black 0.103 0.304 0.084 0.277 Hispanic 0.101 0.301 0.061 0.240 Asian 0.037 0.190 0.018 0.134 Immigrant 0.137 0.344 Age 45.231 13.257 53.885 15.863 Female Head 0.405 0.491 0.239 0.426 Number of Family 2.900 1.487 2.490 1.347 Married 0.645 0.478 0.642 0.479 Single 0.132 0.339 0.090 0.287 Widowed 0.057 0.231 0.110 0.313 Divorced or Separated 0.166 0.372 0.137 0.344 Less High School 0.098 0.297 0.113 0.317 High School 0.247 0.431 0.336 0.472 Some College 0.532 0.499 0.227 0.419 College 0.123 0.329 0.324 0.468 Family Income 67533 99747 85601 121978 Refinance 0.078 0.269 0.317 0.465 Years in House 7.532 7.745 12.461 12.115 First Time Owner 0.452 0.498 0.019 0.135 Own after 2001 0.500 0.500 0.350 0.477 Observation 68797 31421 The descriptive statistics show the average percentage of underwater homeowners of total households in the sample period. This percentage, however, may differ by race, ethnicity and immigrant status and also by the time period of estimation. Thus, I further examine the percentage of underwater homeowners by race and ethnic groups and also by immigrant status before and after the 2007 housing market crisis. Figure I-1 shows the proportion of underwater homeowners in the AHS sample. During the period between 2001 and 2011, Hispanics had the highest share of underwater homeowners, followed by Asians, blacks and whites. Prior to the crisis, however, Asians had the highest proportion of homeowners in negative equity while 10 blacks had the lowest share of underwater homeowners. Following the crisis, the racial share of underwater homeowners changed considerably. From 2007 to 2011, Hispanics had the greatest proportion of negative equity households. Also, the share of black underwater homeowners exceeded the share of white underwater homeowners. [Figure I-1] AHS Underwater Homeowners by Race & Ethnicity Figure I-2 classifies underwater homeowners by immigrant status. The graph shows that before the crisis, immigrants and the native born had similar percentages of underwater homeowners. Since the crisis, however, a greater share of immigrants became underwater, resulting in a noticeable gap in the percentages of underwater homeowners between the two groups. 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 White Black Hispanic Asian 2001-2011 2001-2005 2007-2011 11 [Figure I-2] AHS Underwater Homeowners by Immigrant Status In Figure I-3, I further categorize immigrants by race and ethnicity. 9 The white immigrants in Figure I-3 show similar percentage of underwater homeowners compared to the native born in Figure I-2. For those in other race-immigrant categories, the patterns are similar to the patterns shown in Figure I-1. Hispanic immigrants have the highest share of underwater homeowners during the total sample period and also show the greatest increase in the share of 9 Studies including Painter et al. (2001) find considerable gaps across the white, black, Asian and Hispanic homeownership rates even after controlling for the immigrant status. After controlling for the observables such as income and education, Painter et al. (2001) finds that the Asians immigrants have similar likelihood of being a homeowner than whites and also than the Asian native born. The Hispanic immigrants have lower likelihood of being a homeowner than whites and the native Hispanics, but the Hispanic-white homeownership differential disappears when the endowment differences are controlled for. On the other hand, the endowment-adjusted homeownership gap between whites and blacks remains sizable. These differences across race and immigrant status could also appear in the likelihood of being underwater as they are observable and unobservable differences among the race groups within the same immigrant category. Thus, I further classify immigrants into white, black, Hispanic and Asian immigrants and compare the share of underwater homeowners by the race-immigrant category. 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 Native Born Immigrant 2001-2011 2001-2005 2007-2011 12 underwater homeowners since 2007. Black immigrants had the lowest share of those with negative equity prior to the crisis, but the share of underwater homeowners grew almost as much as the Hispanic immigrants. Again, Asian immigrants had the highest share of underwater homeowners before 2007, but since 2007, the percentage increased less than the other two minority groups, although the increase was still larger than the white immigrants. [Figure I-3] AHS Underwater Homeowners by Race & Immigrant Status Overall, the three figures suggest that minorities and immigrants are more likely to become underwater since the 2007 housing market crisis compared to the whites and the native born. Also, racial differences exist among the immigrant households. In the following section, I will explain the models that examine whether these differences still exist after controlling for the observables as well as time and locational fix effects. 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 White_Immigrant Black_Immigrant Hispanic_Immigrant Asian_Immigrant 2001-2011 2001-2005 2007-2011 13 I.4. Methodology While simple comparison shows that minorities and immigrants are more likely to be underwater compared to whites and the native born, these differences could be explained by the differences in the observable characteristics. For example, blacks, on average, receive less education compared to whites. If the level of education is associated with the likelihood of being underwater, the difference in the propensity of being underwater between blacks and whites may disappear when the level of education is controlled for. The first model tests whether the likelihood of being underwater differs depending on race, ethnicity or the immigrant status after including the control variables. ( ) (1), where ( ) is the probability that the household i is underwater at time t. equals 1 if the household is underwater. RI is a dummy variable indicating race and ethnicity or immigrant status. White households are the omitted base category for the race dummies and the native born households equals zero for the immigrant dummy. represents variables related to the demographic and socioeconomic characteristics of households that may be associated with their likelihood of being underwater. I also control for the percentage of down payment which is available in the AHS from 2007. If is significant even after controlling for the period of buying and the percentage of down payment, it suggests that minorities and immigrants are more likely to be underwater because they purchased houses or mortgages at a greater cost, possibly due to unobserved discrimination in the housing market. I also include the region fixed effect ( ) in the 14 regression. 10 Finally, is the random error term which is assumed to be normally distributed. Since is binary, I use the logit model. 11 Before proceeding to the other models, it is worthwhile discussing the measure of the loan to value (LTV) ratio which determines whether the household is underwater or not. There are two major ways of determining the LTV ratio which is calculated by dividing the remaining mortgage principal by the current house price. While the amount of the remaining principal can be obtained any time, the actual house price is only realized at the point of sale. Thus, the house value needs to be predicted during the period of no transaction. In this case, either the homeowner‘s self -reported house price or the estimated house price could be used to measure the LTV ratio. The estimated house value is calculated by inflating the initially reported home sale price with the repeated-sale price index. The two different measures of house value result in different LTV ratios. Previous studies use both measures of LTV. For example, Demyanyk et al. (2013) use the estimated LTV while Coulson and Grieco (2013) use the reported LTV. Following Coulson and Grieco (2013), I use the self-reported house value for the LTV calculation. 12 10 More ideally, I would like to use the state fixed effects. However, the National AHS only provides MSA information and not the state informa tion. Thus, I do not know the state which houses in the non -MSAs are located. An alternative is to regress only the houses in the MSAs, and include the MSA or the state fixed effects. This does not lead to noticeable changes in the results. 11 For all regressions, I present odds -ratio, and also provide marginal effects for some of the results. I provide the odds-ratios rather than the marginal effects as it is not doable to calculate marginal effects for the interaction term. This is because interaction terms cannot change interdep endently with the each components used in the interaction term (Williams, 2012). To calculate the marginal effects, I regress separate models for period before and after the crisis rather than regressing a single model using the diff -in-diff framework. 12 I can only calculate the estimated house value for th e PSID sample as I only have the PSID geocoded data. With the PSID sample, I also use the estimated house value instead of the self -reported house value to calculate the LTV ratio.. I find that main resu lts are not affected when using the estimated house value. Blacks and Hispanics are more 15 The second model transforms model (1) into a difference-in differences framework. Using this model, I estimate whether the change in the likelihood of being underwater following the crisis differs for different race and ethnicity groups or for immigrants and non-immigrants. The key assumption of this method is that without the housing market shock, the difference in the likelihood of being underwater would have remained unchanged for all groups. ( ) (2) The above model includes a year dummy which equals one for period after 2007. An interaction term is also added to estimate whether the likelihood of being underwater changed following the crisis for different groups of race, ethnicity and immigrant status. Here, is the key coefficient of interest. I also use model (1) and (2) to test whether likelihood of being underwater differs within immigrants with different race groups and different years of residence in the U.S. As mentioned in the data section, I further categorize immigrants by the four different race groups. I include the years of residence as studies, including Myers and Lee (1998), suggest that immigrants gradually assimilate in the US as years of living in the US increases. If recent immigrants receive less favorable mortgage conditions or pay more for the house due to language barriers, insufficient information or networks, then more years they spent in the U.S. would lower their probability of likely to be underwater even after adding the control variables. Also , those living in Hispanic and black neighborhoods have higher likelihood of being underwater. Mea nwhile, in another paper I compare the reported and the estimated valu e of the house price and investigate how different measures affect the empirical results (Choi, 2014). 16 becoming underwater. In order to test this hypothesis, I interact the immigrant dummy with the number years they lived in the U.S. 13 The third model uses the PSID to test whether households located in neighborhoods with more percentage of blacks or Hispanics are more likely to be underwater. This will show up if these neighborhoods experienced greater fall in the house prices since 2007, as suggested by Sufi and Mian (2009). As minorities are more likely to live in places with greater share of minorities, the individual and neighborhood effects cannot be perfectly disentangled. The third model, however, tests whether the share of minorities in neighborhood mitigates the unobserved differences across different race groups. If in model (3) becomes less significant after including the share of minorities in the neighborhoods, this suggests that neighborhood racial composition has an influence on minority households‘ likelihood of being underwater. ( ) (3) Since I only have the tract level data for the PSID, I do not include Asian dummy or immigrant dummy in this model. is the variable which indicates the percent of black and Hispanic households in each census tract. If the neighborhood racial composition effect exists, the size of the coefficient will decrease when is included. I also include the interaction term to identify whether the neighborhood racial composition has a different association across the race groups. For example, if living in black dominated neighborhood increases blacks‘ likelihood of being underwater compared to other race groups, 13 The native born is the reference group, when the immigrants are categorized by ra ce or years in the U.S. 17 then odds ratio would be greater than 1 when R indicates black. When the interaction term is included, will show how the neighborhood racial composition affects individuals of other race. Finally, I modify model (3) to test whether underwater homeowners increased more in tracts with greater share of black and Hispanics since 2007. As in model (2), I use the difference-in- differences framework. I.5. Empirical Findings Individual Effect The first column in Table I-2 shows that minority households, on average, are more likely to be underwater compared to white households before adding any control variables. The relative likelihood of being underwater is the highest for Hispanics, followed by Asians and blacks, as in Figure I-1. Immigrants are also more likely to be underwater than the native born. However, when control variables are included, I find that blacks are most likely to be underwater (column (2)). While the odd-ratio of Hispanics still remains to be statically significant, Asians are not more likely to be underwater than whites after controlling for households‘ demographic and socioeconomic characteristics. The result is similar to Painter et al. (2001) who find that the unexplained homeownership gap between blacks and whites remains while this gap disappears for Asians and Hispanics after adjusting for the endowment factors. The odd-ratio of immigrants also decreases when the control variables are included. As for other control variables, I find that the likelihood of being underwater increases with age and the number of family members. Meanwhile, marital status do not show a significant relationship with the propensity of being in negative equity. Households with higher level of education and higher income are less likely to be underwater, indicating that those with lower 18 income and less education were more adversely affected by the housing market crisis. Households that refinanced their mortgages or those who lived in their current house for a longer period of time are also less likely to be underwater. As expected, those who purchase their home after 2001 have significantly higher likelihood of being underwater compared to those who bought their house prior to 2001. Even after controlling for the period of buying, however, the gap in the likelihood of being underwater remains significant across difference groups of race, ethnicity and immigrant status. This suggest that the period of buying does not fully explain why whites and native born are less likely to be underwater. Lastly, the first time homeowners, mostly whom have little wealth, are also more likely to be underwater. This supports the hypothesis that households with less wealth may put less initial down payment which increases their likelihood of becoming underwater homeowners when house price falls. Columns (3) and (6) directly controls for the percentage of initial down payment which the household paid at the time of purchasing their house. This data is only available in years 2007, 2009 and 2011. So for other years, I can only obtain the percentage of down payments for those households who remained in the same house. For these households, I input households‘ down payment data in 2007 for previous years. This extrapolation reduces the sample size due to moving, and also due to attrition of existing houses. Within each year, this would not be a problem unless the sample omission is systematically related to the initial level of down payment. For the whole sample period, however, this could be problematic since the sample size is greater for the later years which could bias the estimation. If minorities and immigrants were relatively more likely to become underwater after the crisis than whites and the native born, then the odd- ratio for minority and immigrant dummies will increase when greater share of sample is obtained from the period after the crisis. Despite this problem, I include the down payment variable to 19 examine whether difference in the level of down payment fully explains the difference in the likelihood of being underwater. If so, then the gaps in the likelihood of being underwater will disappear onve the level of down payment is included. After including the down payment variable, 14 however, the odd-ratios for the three minority race groups (column (3)) and immigrant households (column (6)) become larger than those in column (2) and (5). The reason for the increases in the odds ratios is most likely the fact that the sample is weighted towards the latter year. But the findings do indicate that initial down payment also do not fully explain the gap in the likelihood of being underwater. It is also interesting to observe that the odd-ratio of Asians increases the most when the down payment variable is included. This is likely to be due to the fact that Asians pay greatest proportion of down payment among all race groups. 15 As expected, the likelihood of being underwater proportionally decreases as the percentages of initial down payment increases. 14 Here the omitted category is no down payment. 15 According to the AHS 2011, 10.8 percent of white households answered that they put no down payment when they bought their current home. The percent of households with no down payment were 15.45 percent for blacks, 12.38 percent for Hispanics and 5.91 percent for Asians. Also, the percent of immigrants who answered they put no down payment was 8.62 percent, 2.98 percent lower than that of the native born. 20 [Table I-2] Model (1): Likelihood of Being Underwater VARIABLES (1) (2) (3) (4) (5) (6) Black 1.188*** 1.262*** 1.315*** (0.079) (0.089) (0.103) Hispanic 1.428*** 1.159** 1.218*** (0.075) (0.067) (0.079) Asian 1.269*** 1.137 1.314*** (0.101) (0.091) (0.120) Immigrant 1.324*** 1.092* 1.186*** (0.061) (0.055) (0.068) Age 1.022** 1.010 1.023** 1.011 (0.010) (0.012) (0.010) (0.011) Age2 1.000** 1.000 1.000** 1.000 (0.000) (0.000) (0.000) (0.000) Female Head 1.054 1.051 1.062 1.058 (0.042) (0.047) (0.042) (0.048) # of Family 1.028* 1.008 1.036** 1.016 (0.015) (0.016) (0.014) (0.016) Single 0.874 0.906 0.890 0.941 (0.086) (0.107) (0.087) (0.110) Widowed 1.044 1.037 1.066 1.063 (0.062) (0.071) (0.063) (0.071) Divorced/Separated 1.005 1.069 1.040 1.117* (0.060) (0.073) (0.062) (0.075) High School 0.839** 0.813** 0.838** 0.822** (0.068) (0.082) (0.067) (0.082) Some College 0.740*** 0.773*** 0.735*** 0.776*** (0.058) (0.075) (0.056) (0.073) College 0.687*** 0.729*** 0.676*** 0.728*** (0.064) (0.081) (0.061) (0.078) Log Family Income 0.912*** 0.943* 0.911*** 0.945* (0.020) (0.030) (0.020) (0.030) Refinance 0.785*** 0.770*** 0.787*** 0.774*** (0.061) (0.065) (0.060) (0.065) Years in House 0.964*** 0.966*** 0.965*** 0.966*** (0.004) (0.006) (0.004) (0.006) Own After 2001 2.610*** 2.819*** 2.615*** 2.799*** (0.158) (0.244) (0.157) (0.242) 21 [Table I-2] (Continued) VARIABLES (1) (2) (3) (4) (5) (6) First Own 1.136*** 0.992 1.162*** 1.012 (0.050) (0.050) (0.050) (0.050) Down Payment 0-2 percent 0.754*** 0.756*** (0.058) (0.058) 3-5 percent 0.801*** 0.804*** (0.056) (0.056) 6-10 percent 0.652*** 0.648*** (0.046) (0.045) 11-15 percent 0.589*** 0.579*** (0.057) (0.056) 16-20 percent 0.493*** 0.480*** (0.039) (0.038) 21-40 percent 0.401*** 0.390*** (0.035) (0.034) 41-99 percent 0.287*** 0.276*** (0.043) (0.041) Midwest 1.112* 1.012 1.059 1.135** 1.016 1.061 (0.063) (0.059) (0.072) (0.064) (0.059) (0.072) South 1.198*** 1.096* 1.067 1.240*** 1.118** 1.090 (0.065) (0.060) (0.069) (0.066) (0.061) (0.069) West 1.694*** 1.545*** 1.683*** 1.785*** 1.579*** 1.731*** (0.097) (0.090) (0.114) (0.100) (0.090) (0.115) Constant 0.0648*** 0.1000*** 0.139*** 0.0648*** 0.0949*** 0.128*** (0.003) (0.035) (0.066) (0.003) (0.033) (0.060) Observations 67,888 65,338 44,855 68,797 66,214 45,449 Pseudo R 2 0.008 0.058 0.076 0.007 0.058 0.077 Note: Robust seeform in the parenthesis * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. Individual Effect before and after 2007 Table I-3 shows the result of model (2) which tests whether minorities and immigrants disproportionately became underwater following the crisis. Prior to the crisis, I find that minorities and immigrants did not have higher likelihoods of having negative equity. In fact, after adding the control variables, the likelihoods of being underwater for Hispanic households 22 and immigrants were lower than whites and the native born. Following the crisis, however, the likelihood of being underwater increased significantly for black and Hispanic households and also for the immigrants relative to the reference groups. On the other hand, compared to whites, Asians do not show a significant increase in the likelihood of being underwater since 2007. The coefficients for control variables are similar to Table I-2, and thus are not reported. 16 [Table I-3] Model (2): Likelihood of Being Underwater since 2007 VARIABLES (1) (2) (3) (4) Black 0.926 0.974 (0.116) (0.123) Hispanic 0.899 0.700*** (0.090) (0.074) Asian 1.072 0.928 (0.156) (0.139) Black*Years≥2007 1.458*** 1.458*** (0.196) (0.198) Hispanic*Years≥2007 1.822*** 1.934*** (0.203) (0.220) Asian*Years≥2007 1.204 1.300 (0.196) (0.217) Immigrant 0.970 0.773** (0.091) (0.078) Immigrant*Years≥2007 1.466*** 1.565*** (0.153) (0.169) Years≥2007 2.050*** 1.966*** 2.170*** 2.079*** (0.092) (0.099) (0.089) (0.099) Control N Y N Y Region FE Y Y Y Y Observations 67,888 65,338 68,797 66,214 Pseudo R 2 0.030 0.073 0.028 0.072 Note: Robust seeform in the parenthesis * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. 16 The result of the control variables can be provided upon request. 23 Using the results of model (1) and (2), I calculate the marginal effects for each race and ethnic group for clearer interpretation. Figure I-4 presents the marginal effects of being underwater for each race groups after controlling for the demographic and socioeconomic factors. 17 Before 2007, the Hispanics had the lowest likelihood of being underwater. Their probability of being underwater was 1.2 percent lower than whites. Blacks had 0.5 lower probability of being underwater than whites, while Asian and white households had the same probability of holding negative equity prior to the crisis. Since 2007, the probability of being underwater increased for all race groups. The increase was most significant for Hispanic households followed by blacks and Asians. Following the crisis, Hispanics had 2.9 percent higher probability of being underwater than whites, and blacks had 3.6 percent higher probability than whites. Asian also had higher likelihood of holding negative equity than whites but only by 1.3 percent. [Figure I-4] Marginal Effects by Race & Ethnicity: Before & After 2007 17 I do not include the percent of down payment when calculating the marginal effects. 0.045 0.083 0.040 0.119 0.032 0.112 0.045 0.096 0.000 0.020 0.040 0.060 0.080 0.100 0.120 0.140 2001-2005 2007-2011 White Black Hispanic Asian 24 Figure I-5 shows that prior to 2007, immigrants also had lower probability of being underwater compared to the native born. However, greater proportion of immigrants fell underwater following the crisis. From 2007-2011, switching from native born to immigrant increased the probability of being underwater by 1.8 percent. The results of the two figures agree with the findings in the regressions, confirming that greater share of minorities and immigrants became underwater following the crisis. [Figure I-5] Marginal Effects by Immigrant Status: Before & After 2007 Immigrants by Race The next two regressions look into immigrants. The first regression classifies immigrants based on their race and ethnicity and examines their likelihood of being underwater before and after the crisis. The omitted category is the native born households. Column (1) in Table I-4 shows that during the period between 2001 and 2011, Hispanic immigrants were most likely to be underwater. Black and Asian immigrants were also more likely to be underwater than the native 0.045 0.088 0.035 0.106 0.000 0.020 0.040 0.060 0.080 0.100 0.120 2001-2005 2007-2011 Non Immigrant Immigrant 25 born, but the white immigrants did not have a statistically higher likelihood of being underwater compare to the U.S. born households. When the control variables are included, the statistical significance only remains for the Hispanic immigrants (Column (2)). The likelihood of being underwater also differs by the period of estimation. The results from columns (3) and (4) show that since 2007, the likelihood of being underwater for black and Hispanic immigrants increased significantly compared to the native born households. In fact, prior to the crisis, both black and Hispanic immigrants have odd-ratios less than one, although the coefficient is insignificant for blacks. 18 These results show that even within the immigrants, the likelihood of being underwater differs by race. Hispanic and black immigrants are more likely to be underwater, especially following the crisis. On the other hand, Asian and white immigrants do not show noticeable differences in the likelihood of being underwater compare to the native born households. 18 The reason for this insignificance is because the sample size of the black immigrants is only one fifth of the Hispanic households resulting in a greater standard error. 26 [Table I-4] Likelihood of Being Underwater: Immigrant-Race Groups VARIABLES (1) (2) (3) (4) White Immigrant 1.025 0.951 1.069 0.983 (0.099) (0.096) (0.217) (0.206) Black Immigrant 1.452** 1.272 0.811 0.733 (0.229) (0.199) (0.264) (0.239) Hispanic Immigrant 1.543*** 1.162** 0.923 0.664*** (0.0973) (0.085) (0.120) (0.0922) Asian Immigrant 1.276*** 1.102 1.086 0.903 (0.107) (0.093) (0.169) (0.145) Years≥2007 2.170*** 2.080*** (0.089) (0.099) White Immigrant*Years≥2007 0.901 0.953 (0.205) (0.223) Black Immigrant*Years≥2007 2.222* 2.080* (0.941) (0.892) Hispanic Immigrant*Years≥2007 2.135*** 2.134*** (0.562) (0.574) Asian Immigrant*Years≥2007 1.319 1.364 (0.370) (0.392) Years≥2007 2.170*** 2.080*** (0.089) (0.099) Control N Y N Y Region FE Y Y Y Y Observations 68,616 66,045 68,616 66,045 Pseudo R 2 0.008 0.058 0.029 0.072 Note: Robust seeform in the parenthesis * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. Figure I-6 shows that prior to 2007, white and Asian immigrants have similar probability of being underwater compared to the native born while Hispanic and black immigrants have 1.4- 1.6 percent lower probability of being underwater than white immigrants. However, the probability of being underwater increased the most for black and Hispanic households following the crisis. Since 2007, black immigrants have 4.3 percent higher probability of being underwater 27 than the native born while the probability difference between Hispanic immigrants and the native born is 3.3 percent. [Figure I-6] Marginal Effects by Race and Immigrant Status: Before & After 2007 Immigrants by Years in the U.S. Table I-5 further examines whether the likelihood of being underwater decreases with immigrants‘ length of stay in the U.S. When control variables are included, the immigrant who came to the U.S. most recently have the lowest likelihood of being underwater. This is likely to be related to the fact that these homeowners are less affected by the housing bust due to the relatively short length of time as a homeowners. It is also likely that the homeowners who were able to buy home in a short period of time has relatively more wealth to put down as the initial down payment. Without including the control variables, the odd-ratio in column (1) decreases as the years the immigrants resided in the U.S. increase. After adding the control variables, however, only the immigrants who have lived in the U.S. for 5-9 years or over 30 years have higher 0.045 0.088 0.044 0.083 0.028 0.131 0.030 0.121 0.043 0.100 0.000 0.020 0.040 0.060 0.080 0.100 0.120 0.140 2001-2005 2007-2011 Native Born White Immigrant Black Immigrant Hispanic Immigrant Asian Immigrant 28 likelihood of being underwater compared to the native born. The results are inconsistent with the assimilation hypothesis, as it shows that more years living in the U.S. does not necessarily decreases the likelihood of being underwater for immigrant households. [Table I-5] Likelihood of Being Underwater: Immigrant by Years in the U.S. VARIABLES (1) (2) Years In US: Immigrants 0-4 years 1.005 0.646* (0.238) (0.155) 5-9 years 1.930*** 1.203* (0.214) (0.134) 10-14 years 1.536*** 1.087 (0.141) (0.105) 15-19 years 1.238** 0.935 (0.120) (0.0954) 20-24 years 1.347*** 1.105 (0.136) (0.115) 25-29 years 1.149 0.989 (0.130) (0.116) 30- years 1.128 1.273** (0.101) (0.121) Control N Y Region FE Y Y Observations 68,485 65,919 Pseudo R 2 0.008 0.059 Note: Robust seeform in the parenthesis * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. So far, the AHS results show that minorities and immigrants are more likely to be underwater, from 2001 to 2011. Moreover, following the crisis, greater proportion of minorities and immigrants became underwater compared to the whites and the native born. 29 Neighborhood Racial Composition Effect Next, I use the PSID to test whether this phenomenon can be explained entirely by the unobserved individual level differences or whether this could also be related to racial composition of each neighborhood. Using the geo-coded PSID data, I will test whether including the neighborhood‘s share of black and Hispanic households change the racial and ethnic differences in the likelihood of being underwater. 19 The first column in Table I-6 shows that blacks and Hispanic households are more likely to be underwater which is similar to the AHS results. The next column shows that households residing in tracts with more percentage of black or Hispanic households have greater likelihood of being underwater. The odd-ratio for the percent of Hispanic is almost 3 times as high as the odd-ratio for the percent of blacks. Column (3) includes both individual race dummies and the share of blacks and Hispanic households in each tracts. The odd-ratios for both black and Hispanic dummies decrease compared to those in column (1) indicating the existence of neighborhood composition effects. However, although the odd-ratio of the black neighborhoods is greater than 1, the coefficient is insignificant. This suggests that blacks living in black neighborhoods do not have higher likelihood of being underwater than blacks living in other neighborhoods 20 . In other words, no matter where you are located, blacks have higher likelihood of being underwater. 19 Since the PSID provides the c urrent state of residence for all individuals, state fixed effects are included in all regressions. 20 One caveat for this interpretation is the strong correlation between black and the percent of black households in the tract. The correlation between the two variables is 71.38 percent while correlation between Hispanic and the percent of Hispanic households is 50.29 percent. 30 [Table I-6] Model (3): Likelihood of Being Underwater-Neighborhood Composition Effect VARIABLES (1) (2) (3) (4) (5) Black 2.250*** 2.071*** 2.432*** 1.567 (0.384) (0.491) (0.675) (0.613) Hispanic 3.755*** 2.785*** 5.067*** 3.620*** (0.678) (0.644) (1.318) (1.205) % Black (Tract) 2.373*** 1.137 (0.538) (0.387) % Hispanic (Tract) 6.798*** 2.509** (2.013) (0.917) Black: %Black 0.363 0.447 (0.236) (0.390) Non-Black: % Black 2.134* 1.898 (0.977) (1.178) Hispanic: % Hispanic 0.124*** 0.115*** (0.0746) (0.0871) Non-Hispanic: % Hispanic 7.163*** 4.671*** (3.237) (2.787) Age 1.034 (0.032) Age2 0.999* (0.000) Female Head 1.148 (0.289) # of Family 1.035 (0.052) Single 1.358 (0.677) Widowed 1.181 (0.322) Divorced/Separated 1.232 (0.295) High School 0.658* (0.144) Some College 0.675* (0.158) College 0.547** (0.137) Log Family Income 1.295** (0.148) 31 [Table I-6] (Continued) VARIABLES (1) (2) (3) (4) (5) Refinance 1.463*** (0.184) Years in House 1.019* (0.011) Own After 2001 2.822*** (0.538) Log Wealth w/o Equity 0.780*** (0.029) Constant 0.024*** 0.022*** 0.023*** 0.021*** 0.012*** (0.010) (0.010) (0.010) (0.009) (0.016) State FE Y Y Y Y Y Observations 22,280 21,860 21,860 21,860 17,638 Pseudo R 2 0.058 0.047 0.056 0.060 0.141 Note: Robust seeform in the parenthesis * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. Meanwhile, the odd-ratio for the Hispanic neighborhood is statistically significant. This explains why the odd-ratio of the Hispanic dummy decreases more than the odd-ratio of the black dummy when the neighborhood variables are included. This indicates that the neighborhood racial composition effect seems to be larger in the Hispanic neighborhoods than the in the black neighborhoods. This also suggests the house prices in Hispanic neighborhoods may have fallen more than the house prices in black neighborhoods. This may be related to the fact that the increase of homeownership was higher for Hispanic households compared to the Black households during the housing boom. 21 This could have led to greater increase of house prices in the Hispanic neighborhood during the boom but also subsequently higher price fall 21 In 2000, the census data shows that the homeownership rate was 47.2 percent for blacks and 46.3 percent for Hispanic. However, in 2006, the Hispanic homeownership rate reached 49.7 percent while that of black only increased by 0.7 percent. 32 during the bust. The relationship between racial composition and house price changes needs further investigation. Columns (4) and (5) include the interaction term of black (Hispanic) and percent of blacks (Hispanics). These variables are added to identify whether the size of the neighborhood racial composition effect differs for different race groups. When the interaction term is added %Black*Black estimates how black neighborhood is associated with the likelihood of being underwater for black household and the %Black estimates the black neighborhood effect for other households. The same holds for the %Hispanic*Hispanic and %Hispanic variable. The result in column (4) shows that non-black households are more likely to be underwater in black neighborhoods compared to the blacks in black neighborhood. However, both odd-ratios for %Black and %Black*Black become insignificant when the control variables are added (column (5)). These results holds and is stronger in Hispanic neighborhoods. The odd ratio of %Hispanic is significant and greater than 1, while the odd-ratio for the interaction term is significantly less than 1 in both columns (4) and (5). This indicates non-Hispanic households are more likely to be underwater in neighborhoods with larger share of Hispanic population. Hispanic households, however, have lower likelihood of being underwater in these neighborhoods. One possible explanation for this finding is the benefit of networks among Hispanic households in areas with greater Hispanic population. The Hispanic households residing in an area with more Hispanics may be able to share more resources and information which lowers the likelihood of being underwater compared to non-Hispanics living in the same neighborhood. 22 22 In the PSID sample, I find that the likelihood of being underwater increases with age, but at a marginally decreasing rate. Households with female heads are more likely to be underwater in the PSID sample. Marital status 33 Finally, it is notable that the black dummy in column (5) becomes insignificant when control variables are added. These results differ from the AHS results. One possible reason for this difference is the fact that the PSID oversamples low income households. Thus, the white households, who are the reference group in the regression model are likely to be less off than those in the AHS sample. Also the sample size is smaller in the PSID, which increases the standard error for all variables including the black dummy. Thus, the standard error is larger in the PSID sample, and therefore the statistical power is lower. Neighborhood Composition Effect before and after 2007 The final table tests the neighborhood racial composition effect before and after the crisis. The first column shows that before the crisis, greater share of Black and Hispanic population in a neighborhood did not increase the likelihood of being underwater. However, since 2007, the likelihood of being underwater increased significantly for households residing in areas with higher share of minority population. The result is in line with Sufi and Mian (2009) who shows house prices in subprime neighborhoods fell more than the house prices in prime neighborhoods following the crisis. The odd ratios for the interaction term of %Black *(%Hispanic) and Year2007 are greater than 1 in all three columns. However, the size of the odd-ratio for the % Hispanic variable after 2007 is greater than that of the % Black variable after 2007. This again does not seem to show a significant relationship with the propensity of holding negative equity. Households with higher level of education and higher income have lower likelihood of being underwater in accordance with the AHS result. The years in the house variable is insignificant in the PSID result. However, this variable becomes statistically significant when I omit the dummy variable which equals 1 when the household purchase the home after 2001. As expected, this dummy variable has a positive sign in both data, suggesting that the recent buyers are more likely to be underwater. Finally, those who have greater non-housing wealth have lower probability of being underwater. 34 suggests that house prices in Hispanic neighborhoods fell more than the house prices in black neighborhoods. [Table I-7] Model (4): Likelihood of Being Underwater since 2007- Neighborhood Composition Effect VARIABLES (1) (2) (3) Black 2.059*** 1.297 (0.500) (0.414) Hispanic 2.460*** 1.873** (0.586) (0.563) 1.362 0.651 0.804 (0.428) (0.260) (0.419) 1.380 0.657 0.174* (0.984) (0.467) (0.159) % Black * 2007 2.063** 2.071** 1.789 (0.739) (0.739) (0.873) % Hispanic * 2007 5.573** 4.670** 10.39*** (3.729) (3.022) (8.884) Years≥2007 2.420*** 2.426*** 2.200*** (0.324) (0.324) (0.353) Control N N Y State FE Y Y Y Observations 21,860 21,860 17,638 Pseudo R 2 0.080 0.088 0.163 Note: Robust seeform in the parenthesis * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. 35 II. Housing Market Shock and Mobility of Underwater Homeowners: Why Do Underwater Homeowners Move? II.1. Introduction Since the 2007 housing market crisis, the question of whether negative equity 23 decreases residential mobility — referred to as the lock-in effect — was highlighted by several media outlets (e.g. Economist, 2010, p. 68). The proponents of the lock-in effect also argued that negative equity prevents homeowners from moving to their optimal job location and thus consequently increases the structural unemployment rate. The recent surge in the unemployment rate has provoked more concerns about whether negative equity would further delay the recovery of the sluggish labor market. Previous research provides competing theories that influence the relationship between negative equity and households‘ mobility. Earlier theories predict that underwater homeowners would move less than those with positive equity due to financial constraints (Quigley, 1987; Stein, 1995; Chan, 2001) and psychological aversion towards losses (Engelhardt, 2003). However, the double trigger theory expects underwater homeowners to move more when the labor market conditions turn negative as many of these homeowners default (Riddiough, 1991 & Kau et al. 1993). Meanwhile, the house price expectation theory expects the mobility of underwater homeowners to either increase or decrease depending on how households view the future house prices (Foote et al., 2008). So far, the existing studies which examined negative mobility-relationship show mixed results. 24 The results of these studies, however, may be biased as households‘ unobserved 23 A mortgage is considered to be underwater when the amount of owed debt exceeds the value of the underlying property. Being underwater is identical to holding negative home equity. 24 Refer to Stein (1995), Chan (2001), Ferreira et al. (2010).Donovan and Schnure (2011), Molloy et al. (2011), 36 characteristics may be correlated with both their were likelihood of moving and their likelihood of falling underwater. For example, households can choose the initial level down payment or the level of equity extraction both of which affect their level of home equity. During the period of housing boom, homeowners were able to and willing to extract considerable amount of mortgage equity 25 as they expected house prices to continue rising. While the unobserved expectation behind these choices increases homeowners‘ likelihood of being underwater, it may also be correlated with households‘ decision to move . The underwater homeowners who expect the house price to go up are likely to stay in their current housing until the price moves according to their expectation. Thus, the possible existence of endogeneity makes it difficult draw any causal inferences from the previous research. To overcome this problem, this study conducts a quasi-experiment using the recent housing market crisis. The exogenous shock in the house market left many households underwater, and thus enables me to carve out the possible bias that occur from the unobserved behaviors that are associated with households‘ likelihood of being underwater. The paper is first to use a difference and differences framework to test the causal relationship between negative equity and residential mobility. The results from the Panel Study of Income Dynamics (PSID) show that the mobility of those who became underwater due to the unexpected housing market shock increased significantly compared to those with positive equity. Also the underwater homeowners‘ propen sity to move show a greater increase for the out-state moves compared to Farber (2012), Bucks and Bricker (2013), and Demyanyk et al. (2013). 25 According to the Kennedy -Greenspan estimates of mortgage equity extraction, mortgage extraction as of percent of disposable per sonal income rose rapidly since 2002, from 4 percent in 2002 to 10 percent in 2006 (Greenspan & Kennedy, 2008). This number fell significantly following the crisis. 37 the in-state. As out-state moves are more likely to be job-related moves, the findings contradict the lock-in hypothesis. 26 While the difference-in-differences results suggest that negative equity increases residential mobility, the question of why still remains which also has been understudied by the existing literature. Thus, I further identify the reason for the increase in the mobility of underwater homeowners following the crisis. Two of the existing theories can explain the positive relationship between negative equity and mobility. First is the double trigger hypothesis. In addition to the unexpected drop in the house prices, the subsequent increase in the unemployment rate may have imposed further financial stress on many households, increasing their likelihood of moving. The empirical results show that being unemployed significantly increases the mobility of deeply underwater households. I also find that a greater share of severely underwater homeowners gave up their homeownership and switched to rental housing following the crisis, indicating that many of the severely underwater households faced greater financial difficulties. However, since the number of underwater homeowners who are also unemployed only accounts for a small proportion of the sample, the double trigger theory alone is insufficient to explain the substantial increase in the underwater homeowner‘s mobility rate. Another theory which supports this phenomenon is the house price expectation theory. When house prices are rising, underwater homeowners may choose to stay in their current residence and wait until their house value exceeds their mortgage debt. Meanwhile, those with a similar level of home equity 26 The PSID asks households their reason for moving. During the sample period, I find that only 2.83 percent of in- state movers responded that they moved because of job or school changes while 21.28 percent of out-state movers said they moved due to the same reason. 38 but who expect house prices to fall may decide to sell their homes and move. The results of this study support this hypothesis. Finally, while the empirical findings do not support the lock-in hypothesis, it is still questionable whether the increase in mobility of underwater homeowners has a positive impact on the macro labor market. While almost all of the existing studies seem to conclude that greater mobility of underwater homeowners is positive to the job market, it is difficult to make this conclusion without carefully examining where these homeowners moved to. The final result of the paper finds that underwater homeowners were more likely to move to states with a higher unemployment rates following the crisis, which indicates that the increase in their moves are unlikely to have enhanced labor market efficiency. II.2. Previous Literature Theoretical Explanation: Negative Equity and Mobility Existing theories provide competing explanations for the negative equity and mobility relationship. Earlier studies suggested that underwater households may stay in their current residence due to financial constraints. Moving requires transaction costs and down payment. During the housing bust, homeowners may not have enough financial resources to pay for their initial down payment. In addition, rising interest rates could require owners to put up extra cash beyond standard closing costs or restrict them from obtaining new loans to finance the purchase of a new home (Quigley, 1987). These factors result in households, especially those with little or no home equity, to remain in the current residence (Stein, 1995; Chan, 2001). Prospect theory (Kahneman, & Tversky, 1979) offers another mechanism which may affect the moving decision of underwater homeowners. The theory suggests that people tend to 39 dislike losses more than they like the same amount of gain, due to greater psychological aversion towards losses. According to this view, underwater homeowners may hold on to their current housing, even when moving is the most economically optimal decision. In line with this theory, Genesove and Mayer (2001) find that the nominal loss aversion cause households to place a higher weight on capital losses than on equivalent gains, which leads to longer spells in housing units in declining markets. Engelhardt (2003) suggests that it is nominal loss aversion and not financial constraints that deters households‘ decision to move in a weak housing market. Several other studies provide other psychological reasons that prevent underwater households from moving, such as fear of defaulting and sentimental attachment to their homes. White (2010) suggests that feelings of shame, guilt and fear associated with defaulting explain the relatively low rates of defaults compared to the amount of underwater mortgages. Guiso et al. (2011) use survey methods to show that the non-pecuniary factors such as views on fairness and morality also influence the default decision. Their survey finds that the 82 percent of people in the sample who answered that it is morally wrong to engage in a strategic default were less likely to walk away from their homes. All of the above factors suggest that underwater homeowners are less likely to move than those in positive equity. On the other hand, negative equity is one of the important factors associated with default. Sharp house price decline may lead homeowners to default and move, countervailing the ―lock - in‖ effect. Previous studies show that negative equity indeed is a strong factor that explains default (Schwartz and Torous, 2003; Mayer, Pence, and Sherlund, 2009). However, recent studies suggest that negative equity per se, does not result in default. In fact, the default probability increases significantly when adverse life events such as job losses are combined with negative equity (Foote, et al. (2008), Bhutta et al. (2010), Elul et al. (2010)). This is referred to as 40 the ―double trigger‖ effect. As underwater homeowners are more likely to be unemployed in the recent labor market, the double trigger effect could have increased the mobility of underwater homeowners following the recent crisis. Finally, Foote et al. (2008) provide a simple mathematical model to explain the relationship between negative equity and default that is consistent with the economic theory. The paper shows that choosing not to default despite being underwater could be an optimal decision, if expectations of future house price is considered in the decision making process. If households expect future house prices to rise, they may hold onto the mortgages with the belief that current losses could be recovered by future gains. On the other hand, when future house price expectations are negative, households are more likely to give up their current home and move to a new location. Empirical Findings: Negative Equity and Mobility The competing explanations make it difficult to predict whether negative equity increases or decreases households‘ likelihood of moving. Previous studies provide mixed results which reflects this difficulty. Chan (2001) and Engelhardt (2003) are two earlier studies which investigate the negative equity-mobility relationship. Using 30 year mortgages originated in New York, New Jersey and Connecticut from 1989 to 1994, Chan (2001) finds that the Loan-to-Value (LTV) is negatively associated with mobility. This suggests that households with less home equity are less likely to move, consistent with the lock-in hypothesis. On the other hand, using the National Longitudinal Survey of Youth (NLSY) for a similar time period, Engelhardt (2003) finds that higher LTV is not associated with lower likelihood of moving. 41 The negative equity-mobility relationship recently gained greater interest as the housing market crash left many households underwater. Numerous studies have investigated the relationship between the home equity and mobility to identify the existence of the lock-in effect which may have negative influences on regional labor market. Ferriera et al. (2010) reinitiated this research question, though their estimation period ended just before the occurrence of the financial crisis. Using the American Housing Survey (AHS) from 1985 to 2007, the study strongly supports the lock-in hypothesis, finding that negative equity lowers the 2-year mobility rate by one third. However, Schulhofer-Wohl (2011) criticizes these results, pointing out that the study excludes houses that become rental units or vacant after the household move. If these households are included, Schulhofer-Wohl finds that negative equity results in higher mobility. Extending the sample to 2009, Ferreira et al (2012) respond to Schulhofer-Wohl (2011)‘s critique by pointing out that his study doe s not differentiate between permanent and temporary moves, and thus his classification leads to incorrect results. The 2012 study corroborates their former research finding that negative equity reduces household mobility by 30 percent. However, their study does not provide a clear explanation why a temporary move is not related to job decisions. Another major limitation comes from the data itself as the AHS follows houses and not households. Thus, it is difficult to track households after they move to a new residence. Meanwhile, Donovan and Schnure (2011) use 2007-2009 panel data from the American Community Survey (ACS) and find that the lock-in effect exists for movements within the same county. However, counties with greater house price falls observe more out-state moves. Using the same data, Molloy et al. (2011) also show that states with greater proportion of underwater homeowners do not have lower mobility. However, these studies use county or state as their unit 42 of analysis so they cannot discern whether movers in the region are actually those who are underwater. Coulson and Grieco (2013) directly test the lock-in effect by using the PSID, a panel data which follows households over time. Their study finds that people with negative equity have a slightly higher probability to move both in and out of state than those with positive equity, also showing that there is no lock-in effect. Since distant moves tend to be more related to job changes, the results indicate that negative equity does not create labor market inefficiency. More recent studies have closely examined how negative equity affects unemployment. Bucks and Bricker (2013) test both the lock-in effect hypothesis and the double trigger theory using the Survey of Consumer Finances (SCF) panel in 2007 and 2009. The results show that the propensity to move is higher for underwater homeowners than those with positive home equity. Furthermore, homeowners who are both underwater and economically insecure have the highest likelihood of moving. The reasons for them moving are mostly involuntary such as foreclosure. The findings imply that it is likely that many underwater homeowners have moved not to get better jobs but because they could not afford to stay in their current residence. Farber (2012) compares the changes in mobility of unemployed homeowners and renters over the past 25 years. Using the Current Population Survey (CPS) and the Displaced Workers Survey (DWS), the study finds that the mobility of unemployed homeowners fell less than mobility of the unemployed renters since 2007. With these results, Farber (2012) concludes that the collapse of the housing market did not impede unemployed homeowners to move to find new jobs. Valletta (2012) also finds similar results using the CPS data from 2008 to 2011. The paper finds that homeowners have similar mobility rate, unemployment rate and the unemployment duration as renters. However, the data used by Farber (2012) and Valletta (2012) do not provide 43 information on home equity level. Thus, it is not possible to compare the mobility decision of underwater homeowners to those with positive equity who may have different reasons and incentives to move. Finally, Demyanyk et al. (2013) merge Trans Union and Equifax data to Core Logic to test the lock in effect with a large sample. The paper adds labor market conditions of the geographic area to examine whether the mobility rate of underwater homeowners differs in areas with negative labor market shock to the places with positive labor market shock. The results show that during 2007-2009 homeowners with negative equity moved more than other homeowners regardless of the relative labor market condition, finding no evidence of the lock-in effect. However, while the study covers the largest population, it is unable to control for the households demographic and socioeconomic characteristics due to the lack of data availability. Except Ferrier et al. (2010, 2012), most of the recent studies do not support the lock-in hypothesis. All of the existing studies, however, have a major drawback as they do not properly control for the possible selection bias. The level of home equity is likely to be affected by household‘s unobserved behavior in the housing market which may also be correlated with their tendency to move. Thus, in order to identify the causal relationship between negative equity and mobility, this study uses the external housing market shock which led many households to randomly fall underwater. The paper also investigates which of the competing theories best explain the negative equity-mobility relationship. II.3. Data This study uses the Panel Study of Income Dynamics (PSID), which has followed U.S. households since 1968. The major advantage of the PSID is that it contains extensive 44 information on individual and family characteristics. This enables me to control for various factors that is associated with the household‘s mobility decisions. Be tween 1968 and 1997, surveys were conducted annually. Since 1997, however, they have been conducted biannually. In this study, I chose the sample period from 1999 to 2011. 27 The period covers the time when the U.S. housing market experienced an unprecedented price increase and a subsequent collapse, providing an opportunity to observe how households react in different market conditions. Before looking at the summary statistics, it is worthwhile to discuss the measure of the LTV ratio which determines whether the household is underwater or not. Thus, a precise measure of the LTV ratio is critical for accurate results. There are two major ways of determining the LTV ratio which is calculated by dividing the remaining mortgage principal with the current house price. Unless households purposely report their remaining mortgage principal 28 incorrectly, the amount of reported mortgage debt should be quite accurate. However, as the actual house price is only realized at the point of sale, the house value needs to be predicted during the period of no transaction. In this case, either the homeowner‘s self -reported house price or the estimated house price could be used to measure the LTV ratio. The estimated house value is calculated by inflating the initially reported sale price of the home with the chosen price index. The two different measures of house value results in different LTV ratios. Previous studies have used both measures of LTV. For example, Demyanyk et al. (2013) use the estimated LTV while Coulson and Grieco (2013) use the reported LTV. Although the 27 Another reason for choosing this period is due to the fact that the PSID survey has been conducted biannially since 1997. Thus, the percentage of households who answered they moved from the previous survey year increases significantly from 1999. Thus, including the period prior to 1999 can be problematic when using the difference -in- differences method. 28 The PSID asks the amount of remaining principal for both the first and the second mortgage. The remaining principal is the combined value of the two. 45 studies provide justifications for why they chose their LTV measures, it is still unclear whether households make their decision based on the reported or the estimated house price. Ferreira et al. (2010) argues that the self-reported house prices are often misestimated which causes an attenuation bias. However, this claim only holds when the measurement errors are random. In fact, previous literature suggest that the bias in the reported house value is not random but is mostly systematic (Kain and Quigley,1972; Follain and Malpezzi,1981; Goodman and Ittner, 1992; Bení tez-Silva et al.,2010). On the other hand, studies which use the self-reported house price claim that households use their own estimated house prices when they decide to move (Coulson and Grieco, 2013). In line with these studies, this study also compares the mobility decision of those who perceive to be underwater and those who do not. 29 Table II-1 presents the summary statistics of the variables used in the estimation. I keep the panel data unbalanced by including new households that enter the sample after 1999. 30 Following Coulson and Grieco (2013), I break the total sample into homeowners and renters. I further classify homeowners into two groups: households that are underwater and households that are not. As expected, renters have higher mobility rate than homeowners. Almost 49 percent 29 Meanwhile, in Chapter III, I compare the reported and the estimated value of the house price and discuss how it affects the mobility decisions of underwater homeowners. The findings show that many households who fall in the underwater category according to the estimated house value, report their house price above their current mortgage debt. These households are less likely to move compared to those who report that they are underwater. The results suggest that loss aversion is already incorporated in the households‘ reporting behavior. 30 As in any panel data, in the PSID, new households enter the sample while some drop out. Coulson and Grieco (2013) removed households that first entered the sample and thus the number of samples decreases over the years due to attrition. As the newly formed households tend to be younger, this makes the average age go up. As I am interested in the relationship between the level of home equity and mobility, I do not think it is necessary to follow only the households that were in the sample at initial year of the estimation. Thus, I choose to use the unbalanced panel. However, I also replicate the regressions with the sample of Coulson and Grieco (2013) and find that the results do not change. 46 of renters moved during the sample period, which is almost four times higher than the mobility rate of the homeowners. Between 1999 and 2011, underwater homeowners have a higher mobility rate compared to households with positive equity. The percentage of households that traded up is higher than those who traded down and similar to those who moved to rental housing. Compared to those with positive equity, a greater proportion of underwater homeowners switched to rental housing while the differences in the percentages of those who traded up or down are smaller between the two groups of homeowners. The underwater homeowners are, on average, almost 10 years younger than those with positive equity. This gap is similar to the average age difference between the homeowners and renters. Owners live in a larger family than the renters. Within owners, those underwater live in a slightly larger family than those with positive equity. Compared to whites, higher percentages of blacks and Hispanics are renters. Also, greater share of black and Hispanic homeowners are underwater, indicating that whites are more likely to be owners with positive equity. Those who are unemployed also are more likely to be underwater homeowners compared to those with jobs. In addition, positive equity holders are more likely to receive college education compared to those underwater, while greater proportion of underwater homeowners have not completed high school. On the other hand, the mean family income does not show a noticeable difference between those with positive and negative equity. As expected, the average year in the house is significantly lower for renters compared to owners. Also, the average length of tenure for those with positive equity is almost twice higher than those with negative equity. 47 [Table II-1] Summary Statistics (1) (2) (3) Variables Own LTV≤100% LTV>100% Mean Std. Dev. Mean Std. Dev. Mean Std. Dev. Move 0.123 0.328 0.120 0.325 0.199 0.399 In-State Move 0.100 0.300 0.097 0.296 0.165 0.372 Out-State Move 0.023 0.149 0.023 0.149 0.034 0.181 In-MSA Move 0.087 0.282 0.085 0.278 0.143 0.351 Out-MSA Move 0.029 0.168 0.029 0.167 0.052 0.223 Trade Up 0.045 0.207 0.044 0.206 0.056 0.230 Trade Down 0.029 0.166 0.029 0.169 0.030 0.170 Own-Rent 0.041 0.199 0.039 0.193 0.089 0.285 Age 54.338 15.730 54.451 15.618 44.737 12.170 Female Head 0.236 0.425 0.230 0.421 0.246 0.431 No. of Family 2.508 1.360 2.495 1.344 2.958 1.548 Increase in # of Children 0.059 0.235 0.058 0.234 0.100 0.300 Decrease in # of Children 0.089 0.284 0.088 0.283 0.113 0.317 Single 0.086 0.280 0.084 0.277 0.116 0.320 Widowed 0.113 0.316 0.110 0.313 0.036 0.186 Divorced/Separated 0.136 0.343 0.135 0.342 0.167 0.373 Black 0.087 0.282 0.081 0.273 0.142 0.349 Hispanic 0.059 0.236 0.053 0.224 0.169 0.375 Refinance 0.254 0.435 0.260 0.438 0.249 0.433 Log(Family Income) 10.951 0.890 10.972 0.874 10.998 0.773 Less than High School 0.122 0.327 0.115 0.319 0.145 0.353 High School 0.338 0.473 0.338 0.473 0.356 0.479 Some College 0.224 0.417 0.224 0.417 0.239 0.427 College 0.316 0.465 0.323 0.468 0.260 0.439 Years in House 12.782 12.190 12.895 12.212 7.593 7.101 Observations 30,764 28,406 1144 The descriptive statistics show that the underwater homeowners have higher mobility rate than households with positive equity. This trend, however, may differ from the period before and after the housing market crisis, since prior to 2007 some homeowners may have selectively increased their likelihood of being underwater. 31 During the housing boom, greater proportion of 31 Appendix Table II-1 presents the summary statistics for underwater homeowners before and after the crisis. I also 48 homeowners were able to obtain mortgages with lower level of initial down payment and were also able to easily extract home equity through refinancing. While these behavior increases the likelihood of being underwater, this self-selection may also be related to their likelihood of moving. Thus, I examine the mobility rate before and after the housing market shock to identify whether the unobserved characteristics of homeowners affect their likelihood of moving, and thereby change the negative equity-mobility relationship. Figure II-1 shows the changes in the mobility rate of owners and renters from 1999 to 2005 and from 2007 to 2011. I further classify owners into those with positive and negative equity. The total mobility in the PSID sample increased following the crisis. However, this is due to the increase in the renters‘ mobility. In fact, the mobility rate of homeowners barely changed during this period. 32 Prior to the crisis, underwater homeowners moved slightly less than those above water. But following the housing run t -tests to identify whether observable characteristics differ between the two groups. The results show that that those who were underwater prior to the cri sis were on average younger, with less income and received lower level of education. They also lived in their house for a shorter length of time. Black homeowners account for a greater share of underwater homeowners prior to the crisis, while the share of Hispanic underwater households increased significantly since 2007. Also prior to the crisis, a larger proportion of underwater homeowners refinanced their mortgages. Meanwhile, there were no observable differen ces for variables including sex, marital and e mployment status of the head. The observable differences between the two groups of underwater homeowner s suggest that there are likely to be unobserved differences. 32 Many data, including the CPS (Current Population Survey) and the IRS (Internal Revenue Survey), show that average mobility fell after the crisis. The increase of mobility in this sample may be related to the fact that the PSID consist more renters and low income households than the national average. Also, the decrease in the mobility may be affected by housing formation. In this paper, I exclude the moves that occurred from moving out of their parent‘s households and moves from rental to owner occupied units. If I include first times owners, the mobility of homeowners drop by 1.8 percent following the crisis. This is related to the study by Lee and Painter (2013) which shows that household formation falls during the recession. 49 bust, significantly greater proportion of underwater households moved while the mobility of those with positive equity slightly fell. [Figure II-1] Changes in the Total Mobility Rate between 1999-2005 and 2007-2011 Next, I divide the moves into in-state and out-state moves. Figure II-2 shows that the trend in the in-state moves is similar to that of the total moves. On the other hand, greater changes are shown for the out-state moves (Figure II-3). During the 1999 to 2005 period, the out of state mobility rate of underwater homeowners was only a half of the positive equity households‘ mobility rate. Following the crisis, however, the out -state moves for those with negative equity increased substantially. In contrast to the in-state moves, the out of state mobility rate of positive equity households also increased slightly but at a much lower rate compared to underwater homeowners. 0 0.1 0.2 0.3 0.4 0.5 0.6 Total Own Rent LTV≤100% LTV>100% Move: Before & After 2007 1999-2005 2007-2011 50 [Figure II-2] Changes in the In-State Mobility Rate between 1999-2005 and 2007-2011 [Figure II-3] Changes in the Out-State Mobility Rate between 1999-2005 and 2007-2011 Figure II-2 and II-3 show that those who randomly became underwater following the crisis are more likely to move compared to the positive equity holders, suggesting that negative equity increases the likelihood of moving. However, the significant increase in the mobility of underwater homeowners following the crisis needs further investigation. In the following section, 0 0.1 0.2 0.3 0.4 0.5 Total Own Rent LTV≤100% LTV>100% In State Move: Before & After 2007 1999-2005 2007-2011 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 Total Own Rent LTV≤100% LTV>100% Out State Move: Before & After 2007 1999-2005 2007-2011 51 I will describe methods to examine whether this increase is still significant after controlling for observable factors that may also be associated households decision to move. II.4. Methodology The main purpose of this study is to identify the casual relationship between negative equity and mobility by controlling for both the observable and the unobservable differences across homeowners. The standard model of the literature is the following logit model: ( ) (1), where the dependent variable ( ) is the probability to move for household i between year t-2 and t. is 1 if households move. is a dummy variable which equals 1 if households are underwater (LTV≥100%) at the previous period. represents variables related to the demographic and socioeconomic characteristics of households that may influence their mobility decision. and are the state and year fixed effects and is the random error term which is assumed to be normally distributed. All other models of the paper modifies model (1) by adding appropriate variables to test each hypothesis. Housing Market Shock & Mobility of Underwater Homeowners While model (1) controls for homeowners observable differences, the unobservable differences still remains which can bias the results. An ideal experiment is to compare the mobility decision of the same person keeping everything unchanged except his/her equity status. However, this is not feasible. Thus, I use external housing market shock to control for the unobservable 52 differences using a difference-in-differences framework. The key assumption of this method is that without the housing market shock, the difference in the mobility rate between those with positive equity and those with negative equity would have remained the same. This model is estimated by the following equation: ( ) (2) where is 1 from years 2007 to 2011 and captures the overall trend of mobility for positive equity households following the crisis. 33 The interaction term is added to examine whether the change in the mobility of households with negative equity differs from change in the mobility of positive equity households. Thus, in this model ‗ ‘ is the core coefficient of interest and measures the unbiased relationship between negative equity and mobility. Since coefficients show odds ratios, a greater than 1 means that in response to the crisis, the likelihood of moving has increased relatively more for those in negative equity than those above water. A negative means the opposite. I further alter the model by classifying moves( ) into in-state moves and out-state moves to identify whether the relationship 33 Choosing when to divide the data is a critical issue for the study. Since the PSID implements the survey biannually, the move variable in year 2007 indicates whether the household moved between 2005 and 2007. Thus, it is reasonable to argue that the time period should be divided into periods between 1999 -2007 and 2009 -2011. However, one of my main hypotheses is that un derwater homeowners‘ mobility decision is affected by the house price expectation. These homeowners may have started to move as the house price growth slowed from the fourth quarter of 2005 which is the reason that I included the year 2007 for the later pe riod. For robustness, I als o excluded the year 2007 and ran the same regressions for the periods between 1999 -2005 and 2009 -2011. Another method is to assign different dummy values (0 or 1) for the states that experience a price increase from 2005 to 2007 period and those that experience a price decrease over the same period. I find little changes in both the size of the coefficients and the statistical power in the regression results of the two specifications. 53 between the level of home equity and the two different categories of moving. As suggested by prior studies (Donovan and Schnure, 2011; Coulson and Grieco, 2012), out-state moves are more likely to be related to job changes. Next, I further investigate whether the extent of being underwater have different association with household‘s likelihood of moving. According to Schulhofer-Wohl (2011), households that are severely underwater may have higher propensity to move due to default. I further look into this by classifying LTV ratios into four categories ( ): LTV than 80%, 80 to 100%, 100% to 120% and above 120%. I use homeowners with loan-to-value ratio just below water (80-100%) as the base category to better compare those just above and below the water. Although the observed variables in the model may capture the possible factors that can affect homeowner‘s mobility decision, there may still be substantial differences in the u nobserved characteristics between those with large proportion of home equity and those underwater. This unobserved difference is likely to be greater than the unobserved gap between those near the cutoff point (LTV=100%). This regression is also tested with a difference in differences framework. Possible Reasons for the Changes in Mobility After verifying whether the mobility patterns of underwater homeowners changed following the crisis, I further test for possible factors which influence the negative equity-mobility relationship. Following the recent house price boom and the bust, we have observed drastic changes in the market environment. The stringent market condition including the increased interest rate and stricter lending standards should have inhibited homeowners including those with negative equity to move. The psychological aversion towards losses or the feeling of guilt and shame 54 would have further hampered underwater homeowners moving from their current housing as media outlets and lenders purported that it was morally wrong to default (White, 2010). Two theories, however, support the positive relationship between negative equity and mobility. The first is the double trigger hypothesis. Double trigger effect predicts that job losses combined with negative equity further increase the financial and economic difficulties of underwater homeowners, and thus these homeowners are more likely to default and move. Since the PSID provides information for the individual employment status, I can I test the double trigger effect by including a variable interacting the four different LTV categories with the individual‘s employment status ( ). According to the double trigger effect, the odd ratio for those who are unemployed and underwater should be significantly greater than 1. To better investigate whether double trigger effect increased moves due to default, it would be ideal to directly use the default data. However, the PSID only provides data for delinquencies and foreclosures from 2009, and the proportion of these households is too small to obtain robust results. Thus, I estimate how the negative equity is associated with the decision to trade up, trade down 34 or move to rental property. In this equation the dependent variable, ( ) is replaced to four categories of mobility decisions: (1) do not move (base), (2) trade up, (3) trade down and (4) switch to rental housing. According to Chan (2001), underwater homeowners are unlikely to switch to rental units as it is difficult to find a desirable rental unit at a desirable location. However, when financial difficulties exceed a certain threshold, underwater households may eventually move to rental housing. If a greater share of underwater households 34 The trade up and trade down variables are c reated by subtracting the CPI adjusted previous house value from the house value of the newly moved house. If the difference is positive, the move is categorized as trading up, while if the difference is negative the move is defined as trading down. 55 switch tenure since 2007, this also indicates that more households with negative equity moved due to financial difficulties. Another possible explanation for the positive relationship between negative equity and household mobility is the house price expectation hypothesis suggested by Foote et al.(2008). However, a major problem with testing this hypothesis is the fact that the expectations are not directly observable. Many studies have used lag price appreciation as the current expectation (Sinai & Souleles, 2005; Himmelberg, Mayer & Sinai, 2005) or assumed that households form econometric forecasts based on lagged price appreciation (Han (2013)). Especially for a short term forecast, these studies find that households depend highly on the most recent house price trends. According to Case et al. (2012), homeowners are generally well informed about the trend of in home prices and the one year lag changes in the house prices is highly correlated with the one year forecast. Thus, I use lag change in the state house price indices as a proxy of the short term house price expectation. Under this assumption, I include and to model (1). estimates the change of HPI from the former period. For underwater homeowners, a one period forecast may significantly impact in their mobility decision. If the housing price is rising, it is likely that the underwater households will wait until the house price exceeds their mortgage debt. Thus is expected be less than 1. This suggests that underwater homeowners are more likely to move, when house prices start to stagnate or fall. Meanwhile, a greater than 1 indicates that positive equity households move less in the falling market, in accordance with the loss aversion theory. Finally, I test how the mobility of underwater homeowners is related to labor market efficiency. In order to do this, I use the out-state moves and compare the unemployment rate of 56 the origin and the destination states, and investigate whether underwater households move to locations with lower unemployment rates. If so, this result will strengthen the argument that negative equity has no impact on labor market inefficiency. On the other hand, if the underwater homeowners do not move to areas with lower unemployment, it is questionable whether the increase in the mobility of underwater homeowners has a positive impact on the local unemployment rate. If these homeowners chose to move out of state due to default or foreclosure, than showing that a greater share of underwater homeowners moved compared to those above water is insufficient to confirm that negative equity is unrelated to the labor market inefficiency. To test this hypothesis I use a triple difference framework using the following equation 35 : ( ) (3), where is the difference in the unemployment rate of destination state and the lag unemployment rate of the origin state. If is less than 1, this indicates that those with positive equity are more likely to move to states with lower unemployment rate prior to the crisis. Odds ratio shows whether this relationship have changed following the crisis. Meanwhile, a less than 1 means that underwater homeowners were more likely to move to a state with a lower unemployment rate compared to those with positive equity before 2007. Likewise, shows whether this relationship has changed in the post-crisis period. I also replace state‘s 35 An i deal data to test this hypothesis would be a sample of households who moved out of state. For underwater homeowners and those with positive house value, I would compare the unemployment rate of destination and the origin state. Due to the small sample size, I use model (3) to identify whether different mobility pattern is observed for households who moved fo llowing the crisis. I leave further investigations for future studies 57 unemployment rate with state‘s median ho use prices, 36 to examine whether underwater homeowners moved to states with less expensive house price. In the final regression, I include both unemployment rate and house price data as households are likely to consider both house prices and the labor market conditions when they decide to move. II.5. Empirical Findings Difference in Differences: Change in the Mobility Table II-2 presents the odds-ratios of the difference-in-differences model. Result in column (1) confirms Figure II-1, showing that the underwater homeowners‘ likelihood of moving increased significantly compared to the positive equity households following the housing market shock. Prior to 2007, the underwater homeowners‘ likelihood of moving was not statistically diff erent from the positive equity households. There is little change in the results even after adding the control variables. The findings show that those who exogenously became underwater are more likely to move than those with positive equity. [Table II-2] Difference in Differences: Total Moves VARIABLES (1) (2) LTV>100% 1.218 0.890 (0.228) (0.210) Year≥2007 0.957 0.942 (0.0403) (0.0504) LTV>100%*Year≥2007 1.931*** 1.895** (0.446) (0.542) Age 0.871*** (0.00986) Age2 1.001*** (0.000109) Female head 0.795** (0.0911) 36 The median house price is obtained from Zillow. I use the median house value per square foot in January. 58 [Table II-2] (Continued) VARIABLES (1) (2) # of family 0.984 (0.0244) increase in # of children 1.263*** (0.102) increase in # of children 1.073 (0.0973) Single 1.050 (0.135) Widowed 1.363* (0.241) Divorced/Separated 1.562*** (0.171) Black 0.854 (0.106) Hispanic 1.028 (0.130) Log(family income) 0.982 (0.0401) Unemployed 1.332** (0.194) Refinance 0.910 (0.0537) High school 1.011 (0.103) Some college 1.043 (0.112) College 1.167 (0.126) Years in house 0.972*** (0.00382) Constant 0.139*** 12.78*** (0.00421) (6.984) State FE No Yes Observations 28,929 21,075 Pseudo R-Sq 0.002 0.061 Note: Robust seeform in the parenthesis * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. 59 As for other variables, I find that the average likelihood of moving falls as people become older, but at a marginally decreasing rate. Households with female heads are less likely to move. As expected, increase in the number of children significantly increases the propensity to move. Those who are divorced or widowed are also more likely move than those who are married. Meanwhile, the likelihood of moving do not show any statistical differences according to race, ethnicity, income and education. Finally, the length of tenure reduces the probability of moving to a new house, consistent with the previous studies (Ferrier et al., 2010; Coulson and Grieco, 2013). Table II-3 categorizes the moves into in-state and out-state moves. Without including the control variables, the results show the likelihood of moving both in and out of states increased more for the underwater homeowners compared those with positive equity. When the control variables are added, however, the in states moves of underwater homeowners do not show a statistical difference from those with positive equity in both periods before and after the crisis. On the other hand, prior to 2007, underwater homeowners were significantly less likely to move out of state than those with positive equity. The negative equity and mobility relationship changed from negative to positive following the crisis. The results suggests that prior to the crisis, many negative equity homeowners who may have selectively increased their likelihood of falling underwater had little incentive to move out of state. However, those who became exogenously underwater due to the housing market shock are more likely to move out of state contradicting the lock-in hypothesis. Most of the control variables show the similar signs and magnitude to those in Table II-2, and is presented in Appendix Table II-A1. 60 [Table II-3] Difference in Differences: In-State & Out-State Moves In_State Out_State VARIABLES (1) (2) (3) (4) LTV>100% 1.383* 1.005 0.484 0.227** (0.269) (0.245) (0.262) (0.144) Year ≥2007 0.941 0.946 1.030 0.928 (0.043) (0.055) (0.095) (0.108) LTV>100%*Year ≥2007 1.668** 1.445 5.251*** 12.53*** (0.406) (0.438) (3.229) (8.771) Control No Yes No Yes State FE No Yes No Yes Observations 28,929 21,091 28,929 21,091 Pseudo R-Sq 0.002 0.068 0.002 0.068 Note: Robust seeform in the parenthesis * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. Level of Negative Equity & Mobility To further investigate the association between home equity and mobility, I classify homeowners into four groups according to their LTV ratio. I also include renters in the estimation. The reference category in the regression is households with LTV ratio between 80% and 100%. I differentiate the underwater homeowners into two groups: (1) slightly underwater homeowners with LTV ratio between 100% and 120%, and (2) severely underwater homeowners with LTV ratio greater than 120%. These two groups of underwater homeowners may have different motivations to move. For example, those severely underwater are more likely to move due to default. Thus, if moving is mostly concentrated to those who are deeply underwater, than the increase in the underwater homeowners‘ mobility rate may have little or no impact on labor market efficiency as many of these moves would be involuntary and unrelated to job changes. Table II-4 presents regression results of total, in-state and the out-state moves for homeowners in each categories of LTV ratio. As expected, renters have higher mobility rate than 61 homeowners with the LTV ratio between 100-120%. Compared to those in the base category, renters‘ likelihood of moving within states increased , while the likelihood of moving out-of state does not show a statistically significant change. The likelihood of moving both within and out of states for those with home equity below 80% are not statistically different from those in the base category in both periods of boom and bust. [Table II-4] Extent of Home Equity and Mobility (1) (2) (3) VARIABLES Move In-State Out-State Renter 3.522*** 3.729*** 3.432*** (0.314) (0.367) (0.606) LTV≤80% 1.056 1.080 0.971 (0.0929) (0.105) (0.173) LTV100-120% 0.920 1.028 0.285* (0.292) (0.336) (0.209) LTV>120% 0.879 1.032 0.086*** (0.323) (0.387) (0.0737) Year≥2007 0.978 1.000 0.842 (0.109) (0.122) (0.190) Renter*Year≥2007 1.318** 1.318** 1.379 (0.164) (0.176) (0.351) LTV≤80%*Year≥2007 0.973 0.933 1.173 (0.122) (0.128) (0.304) LTV100-120%*Year≥2007 1.564 1.138 10.77*** (0.617) (0.478) (8.900) LTV>120%*Year≥2007 2.274* 1.780 32.57*** (1.020) (0.832) (32.55) Control Yes Yes Yes State FE Yes Yes Yes Observations 31,917 31,917 31,917 Pseudo R-Sq 0.186 0.176 0.176 Note: Robust seeform in the parenthesis * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. Meanwhile, the mobility patterns of underwater homeowners changed considerably following the crisis. The change is greater for the out of state moves and for those who are severely underwater homeowners. Prior to the crisis, the severely underwater homeowners were significantly less likely to move out of state compared to those in the reference group. However, following the crisis, the likelihood of moving for those with LTV ratio greater than 120% 62 increased significantly. Similar patterns are also shown for those in the LTV 100-120% category but the extent of change is much smaller than the severely underwater homeowners. The results so far shows that negative equity caused many underwater homeowners to move out of states following the crisis. The increase in the mobility was greater for the severely underwater homeowners. However, the reason why negative equity increases mobility still needs further investigation. Double Trigger Effect According to this view, negative equity alone does not lead households to default strategically. However, when negative equity is combined with other financial and economic difficulties such as unemployment, households are more likely to walk away from their current housing. The probability of committing default increases, especially, for those who are significantly underwater. Following the housing market crisis, the unemployment rate went up by more than 5 percent. Thus, more households could have chosen to walk away from their home due to the combination of negative equity and unemployment. Table II-5 shows that during the 1999-2011 period, renters and those with LTV ratio between 80 and 100 percent are more likely to move when they are unemployed. The increase in the likelihood of moving is more significant for the in-state moves. 37 However, the severely underwater homeowners are more likely to move out of states when they are unemployed. Most of the odd-ratios of the interaction terms show that being unemployed increases the propensity to 37 For this analysis, I also extended the data to 1985 , as the number of households in each interaction category is too small to obtain robust results . When the data is extended, most of the coefficients for those who are unemp loyed becomes significant due to the decrease in the standard errors. Also, those who are slightly underwater and unemployed are not less likely to move out of states than the employed homeowners with LTV between 80 and 100%. Still, severely underwater hom eowners who are unemployed are most likely to move out of state. 63 move both in-state and out-state. Those who are unemployed with LTV between 100-120%, however, are less likely to move out of states compared to those with jobs. The cost of moving provides one possible explanation of this result. As moving associates costs, a slightly underwater homeowner without a job may not be able to afford the cost to move to a new house. However, when the home equity falls further, unemployed underwater homeowners may eventually choose to default and move. The fact that those in the LTV>120% * unemployed category has the highest odds-ratio for both total and out-state moves fits well with the double trigger hypothesis. [Table II-5] Double Trigger Effect: 1999-2011 (1) (2) (3) VARIABLES Move In -State Out-State Renter 4.497*** 4.662*** 3.780*** (0.300) (0.338) (0.478) LTV≤80% 1.108 1.112 1.086 (0.074) (0.081) (0.147) LTV100-120% 1.176 1.061 1.698 (0.215) (0.207) (0.643) LTV>120% 1.323 1.374 1.113 (0.304) (0.334) (0.631) Renter*Unemployed 1.266** 1.275** 1.129 (0.132) (0.135) (0.234) LTV≤80*Unemployed 1.035 1.052 0.977 (0.186) (0.203) (0.385) LTV80-100%*Unemployed 1.944** 1.885** 2.263 (0.607) (0.598) (1.571) LTV100-120%*Unemployed 1.366 1.824 0.000*** (0.794) (1.066) (0.000) LTV>120%*Unemployed 3.160* 1.586 12.97*** (1.929) (1.315) (10.98) Control Yes Yes Yes State FE Yes Yes Yes Year FE Yes Yes Yes Observations 33,820 33,820 33,820 Pseudo R-Sq 0.210 0.182 0.182 Note: Robust seeform in the parenthesis * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. 64 The results in Table II-5 suggest that the rise in unemployment since 2007 have triggered the severely underwater homeowners to move, especially to move to another state. The findings also show that the relationship between being unemployed and moving differs depending on the level of negative equity. Those who face financial constraints due to being unemployed but are only slightly underwater may choose to stay in their current housing as they are reluctant to downgrade their overall level of housing quality. However, when the level of negative equity passes below a certain threshold, households eventually give up on consuming a similar level of housing and move to a lower quality housing or switch to rental units. In other words, for each household, there is a minimum home equity threshold of which the household finally decides to reduce housing consumption. The following regression looks further into this point. Moving to Rental Housing Unemployment of the head may not fully capture households‘ financial difficulty following the crisis. There may be unmeasured difficulties presumably due to the unemployment of other family members which may not be fully captured by the PSID data. Therefore, investigating changes in the housing consumption could provide better understanding of why people move. Table II-6 presents the multinomial results which estimate household‘s likelihood of trading up, trading down or switching to renters. The sample only consists of homeowners. The results show that the likelihoods of trading-up or trading down do not show statistical differences across households with different level of home equity in both the period before and after the crisis. We might expect that greater level of home equity will encourage households to move to a more expensive house than their current housing and negative equity will induce households to move to a cheaper house. However, the results do not support this. Perhaps, the level of home 65 equity may not be a key determinant when households decide to expand or downsize their current housing. In other words, trading up and trading down decisions may be more related to other factors than the level of home equity. In fact, Appendix Table II-A4 shows that households are more likely to trade up when the number of children increases. Those with greater income also have greater propensity to move to a more expensive housing. Meanwhile, the likelihood of trading down decreases with the family size. [Table II-6] Multinomial Logit: Trading Up, Trading Down, and Switching to Rental Housing Trade Up Trade Down Own to Rent VARIABLES (1) (2) (3) LTV≤80% 1.001 1.092 0.783 (0.129) (0.192) (0.125) LTV100-120% 0.706 0.538 1.295 (0.293) (0.347) (0.650) LTV>120% 0.961 1.289 0.544 (0.423) (0.881) (0.344) Year≥2007 0.806 0.679 1.250 (0.135) (0.163) (0.226) LTV≤80%*Year≥2007 1.122 0.867 1.056 (0.208) (0.226) (0.221) LTV100-120%*Year≥2007 1.428 3.437 1.120 (0.834) (2.694) (0.673) LTV>120%*Year≥2007 1.748 0.344 5.390** (1.150) (0.349) (3.787) Control Yes Yes Yes State FE Yes Yes Yes Observations 20,968 20,968 20,968 Pseudo R-Sq. 0.076 0.076 0.076 Note: Robust seeform in the parenthesis * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. Although I do not find differences in the likelihood of trading up and down for underwater homeowners after the crisis, I do find substantial changes in the own-to-rent transition. According to the result, greater proportion of severely underwater homeowners have switched to rental housing following the housing market shock. This is somewhat similar to 66 Molloy and Shan (2011), who find that post-foreclosure individuals are more likely to become renters. The result suggests that underwater homeowners finally choose to give up owning when their level of home equity falls below a certain threshold, especially when the economic situation is negative. This result is also in line with the double trigger hypothesis. The results for the other control variables are presented in Appendix Table II-A3. House Price Expectations Although the double trigger hypothesis provides some explanation for the positive relationship between negative equity and mobility, it does not fully explain the increase in the mobility of underwater homeowners following the crisis. During the sample period, only 6.02 percent of underwater homeowners with negative equity were unemployed in the PSID data. Also, while households that said that they moved due to involuntary reason almost doubled from 16.44 percent to 32 percent following the crisis, still a considerable number of underwater homeowners stated that they moved due to other reasons. The change in the house price expectation provides another possible explanation for the increase in underwater homeowners‘ mobility. According to this theory, underwater homeowners‘ decision to move depends on the expectation of future house prices. For example, during the period of house price bubble, underwater households are likely to delay their move, as they anticipate their house price to exceed their mortgage debt. However, once the growth stagnates and house prices fall, underwater households are likely to sell their houses before prices fall further. 38 38 Michigan Consumer Survey asks households whether they think it is a good or a bad time to buy a house and reason why they think so. Prior to 2005, the number respondents who believed that house prices would go up continued to increase, reaching 20 percent by 2005 (Piazzesi and Schneider, 2009). This proportion dropped from 67 Table II-7 provides results which test whether mobility of underwater homeowners is associate with house price growth. The results show that households with positive equity move more in states that experience higher house price growth. This result are in line with Engelhardt (2003) who finds that positive equity households move more during the housing boom and less in the housing bust. In contrast to those with positive equity, the odds-ratio of the interaction term shows that underwater households are less likely to move in states with positive HPI growth. 39 [Table II-7] House Price Expectation Move In-State Out-State VARIABLES (1) (2) (3) (4) (5) (6) LTV>100% 1.318** 1.309** 1.262 1.244 1.591 1.639* (0.177) (0.176) (0.180) (0.180) (0.482) (0.483) State HPI 2.736*** 3.180** 2.290** 2.232 5.449** 8.063** (0.922) (1.658) (0.861) (1.343) (3.917) (8.262) State HPI*LTV>100% 0.008*** 0.008*** 0.008** 0.007** 0.007* 0.010* (0.015) (0.015) (0.018) (0.016) (0.018) (0.024) State FE No Yes No Yes No Yes Year FE No Yes No Yes No Yes Observations 22,023 22,023 22,023 22,023 22,023 22,023 Pseudo R-Sq. 0.055 0.063 0.052 0.069 0.052 0.069 Note: Robust seeform in the parenthesis * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. States are large and thus house price growth may vary in smaller regions within the same state. In Appendix Table II-A4, I show the result of the same analysis using MSA moves and MSA HPI. I find that the size of the coefficient is almost identical, although coefficients become slightly less significant due to the small sample size. Overall, the findings offer an explanation of 2006 and plummeted in 2007, falling to 2 percent. This shows that the optimistic view towards the housing market turned pessimistic following the crisis. 39 Note that the statistical significance for odds-ratios of both in and out of state moves decrease compared to total moves. This is because standard errors increase as sample size becomes smaller. 68 why homeowners who became underwater due to the housing market shock moved considerably more compared to those who were underwater during the boom. Moving to a Better Job Market The final regression further examines the lock in hypothesis by focusing on homeowners who moved out of states. According to the lock-in hypothesis negative equity households are financially constrained and thus are unable move to a better job market. The lack of mobility results in labor market inefficiency and increases structural unemployment. While most of the recent studies including this one find no evidence of the lock-in effect, it is still questionable whether underwater homeowners do or are able to move to locations with better employment conditions. If not, it is also questionable whether the mere increase in the mobility of underwater homeowners is sufficient to prove that negative equity is unrelated to the labor market inefficiency. Table II-8 40 presents the result of the triple difference regression. The first column shows that prior to the crisis, households were more likely to move to a state with a lower unemployment rate. Compared to those with positive equity, greater proportion of underwater homeowners moved to a state where the unemployment rate is lower than where they moved from. For those with positive equity, this trend did not show a different pattern following the crisis. During the housing bust, however, those with negative equity were more likely to move to states with higher unemployment rates than the state they have moved from,. This findings suggest that the despite the increase in the mobility of underwater homeowners, it is difficult to conclude that this phenomenon had a positive impact on the labor market. Column (2) shows that 40 Note that I do not include the year fixed effect in the regression. This is because state and year fixed effects explain most of the variability in the state unemployment rate (refer to Choi & Green, 2014). 69 during the 1999-2005 period, those with negative equity were more likely to move to states with higher house prices. Since these states are likely to have more jobs, this finding also suggests that underwater homeowners were able to move to locations with better job opportunities prior to 2007. But since the crisis, underwater homeowners were less likely to move to states with higher house prices, although the coefficient is statistically insignificant. These results suggest that some underwater homeowners could have moved to regions where house prices are cheaper, yet the labor market conditions are worse than where they have moved from. If so, it is even possible that the increase in the mobility of underwater homeowners actually increased unemployment rate in some regions. This statement needs further analysis. 70 [Table II-8] Moving to a Better Job Market? Out-State VARIABLES (1) (2) (3) LTV>100% 0.226** 0.172*** 0.171*** (0.141) (0.0978) (0.0989) Year≥2007 0.989 0.854 1.016 (0.181) (0.190) (0.247) LTV>100%*Year≥2007 5.895** 10.39*** 8.630*** (4.487) (7.317) (6.223) Origin State Unemployment 0.943 1.000 0.926 (0.0398) (0.0434) (0.0530) Origin State House Price 1.235 1.274 1.187 (0.322) (0.358) (0.321) D(State Unemployment) 0.791** 0.777** (0.0885) (0.0916) D(State Unemployment)*LTV>100% 0.602** 0.693 (0.152) (0.176) D(State Unemployment)*Year≥2007 1.190 1.180 (0.143) (0.156) D(State Unemployment)*LTV>100%*Year≥2007 2.062** 1.890** (0.620) (0.588) D(State House Price) 1.093 0.799 (0.596) (0.455) D(State House Price)*LTV>100% 12.21** 7.029 (14.51) (9.103) D(State House Price)*Year≥2007 1.213 0.874 (0.837) (0.727) D(State House Price)*LTV>100%*Year≥2007 0.213 0.780 (0.352) (1.367) Control Y Y Y State Fixed Y Y Y Observations 18,752 18,752 18,752 Pseudo R-Sq. 0.0885 0.0851 0.0889 Note: Robust seeform in the parenthesis * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. Column (3) finds that coefficients for differences in the states‘ unemployment rates remain significant even after including the house price variables, but the coefficients for differences in the states‘ house prices becomes less significant. The reason that the house price 71 variables do not show statistical significance could be due to the small sample size as by adding multiple interaction terms decrease the degree of freedom. Also, using the median house price per square foot may not be an appropriate proxy of house value, since households may be comparing rental prices or may choose to move to smaller housing units. While, this study do show that underwater homeowners did not move to a better job market following the housing market crisis, further studies are needed to investigate aggregate the impact of how the increase in the mobility of underwater homeowners following the crisis affected the labor market. Further implications for the future studies are provided in Chapter IV. 72 III. Measurement, Perceptions and Negative Equity-Mobility Relationship III.1. Introduction As the number of underwater homeowners increased following the housing financial crisis, these subgroup of homeowners have received greater attention from both the researchers and the policy makers. According to Corelogic, 11.4 million U.S. households had negative home equity as of 2012:Q1, accounting for 23.7 percent of all residential properties with a mortgage. One of the major concerns was whether the underwater homeowners have a negative spillover effect on other parts of the economy. For example, media outlets including the Economist (May, October, 2010) reported that negative equity could be limiting labor mobility (which is referred to as the ―lock-in effect‖) and contributing a structural element to the high unemployment following the 2007 crisis. In response to the growing interest, numerous studies 41 have investigated the relationship between negative equity and mobility. While most studies do not find evidence of the ―lock -in‖ effect, there are some discrepancies in the findings. 42 One of the important differences across the studies that examine the lock-in effect is the proxy of negative equity, which is the most critical variable when identifying the negative equity-mobility relationship. The negative equity is determined by the loan-to-value (LTV) ratio, which divides the remaining mortgage principal over the current house value. The household is underwater when the LTV ratio exceeds 100 percent. Unless households purposely misreport their remaining mortgage principal, the amount of reported mortgage debt should be quite 41 Refer to Ferreira et al. (2010), Donovan and S chnure (2011), Molloy et al. (2011), Farber (2012), Coulson and Grieco (2013), Bucks and Bricker (2013), Demyanyk et al. (2013) and Choi (2014). 42 For example, while Ferreira et al. (2010) and Modestino and Dennett (2013) find that underwater homeowners move less than those with positive equity, Couldson and Grieco (2013) and Demyanyk et al. (2013) find that they move more. Choi (2014) finds that the negative equity -mobility relationship differs according to the estimation period. Prior to the crisis, under water homeowners move significantly less than those with positive equity. Following the crisis, however, their likelihood of moving increased significantly compared to those above water. 73 accurate. An accurate current house value, however, is more difficult to obtain as the actual house price is realized only at the point of sale. Unlike other commodities, houses are more heterogeneous and traded less frequently. Thus, when calculating the LTV ratio, researchers either use the household‘s self -reported house values or estimate the house value using the house price indices. The estimated house price is calculated by inflating the initially reported sale price of the home with the chosen price index. For example, Demyanyk et al. (2013) use the LTV based on the estimated house value while Coulson and Grieco (2013) use the reported house value. 43 The existing studies provide some justifications of why they chose one LTV measure over the other. Those which use the estimated house value point out that the inaccuracy of the reported house value (Ferreira et al., 2010) while others claim that households make decisions to move based on their self-reported house value. Unless we can observe individuals‘ actual thought process, it is difficult to identify which of the two values they use when they make their mobility decision. We can, however, compare whether the different measures of LTV lead to different results which enables us to better understand how people make decisions in the housing market. Using the Panel Study of Income Dynamics (PSID), this study finds that households that are categorized as underwater homeowners according to the estimated house value are more likely to report their house values higher than those who are categorized as positive equity households. In other words, this shows many households tend to report their house value higher than their mortgage balance when the market indices suggest otherwise. This behavior can be explained by the nominal loss aversion theory proposed by Kahneman and Tversky (1979). 43 For simplicity, I name the LTV calculated using the HPI index as t he estimated house value and the LTV calculated using the self -reported house value as the reported LTV. 74 Among those who are estimated to be underwater homeowners, however, those who also report themselves to be underwater do not show a show statistically different gap between the reported and the estimated house value compared to those with positive equity. In addition, using the difference in differences framework, I find that the patterns of mobility differ according to the two measures of negative equity. When using the reported house value, the mobility of underwater households show a significant increase following 2007, but I find no statistical changes in the likelihood of moving when using the LTV calculated by the house price index. The reason for the differences in the results is because the mobility rate of those who report themselves as underwater homeowners is different from those who do not report themselves to be underwater but are estimated to be underwater. The latter group show similar mobility rates to those in positive equity by both LTV measures, while those who report themselves as underwater homeowners show a significant increase in their likelihood of moving following the crisis. There can be three possible explanations for these results. First, some households that are estimated to be underwater may actually have positive home equity and thus behave similarly to those above water. On the other hand, some of these homeowners may actually be underwater. One group may acknowledge their home equity status but may be holding on to their houses as they do not wish to realize their losses if they sell their house at a lower value than their current mortgage. Due to loss aversion, these households also avoid reporting their house value lower than their outstanding mortgage debt. The other group may be ignorant about their home equity status and behave like those with positive equity, because they actually think they are above water. To further examine which of the two LTV ratios is closer to the real value, I look at households‘ likeli hood of moving to a rental unit. Also, to identify whether homeowners are 75 aware of their true home equity status, I look at their likelihood of being delinquent on their mortgages. If the reported house value is more accurate or if households believe this value is true, then the two decisions for those who report they have positive equity, on average, will be similar, regardless of how they are categorized by the estimated LTV. The results, however, show that compared to those who are estimated to have positive equity, those who are estimated be underwater are more likely to switch to rental housing, even for those who do not admit they are underwater. The likelihood of being delinquent is also higher for those estimated to be underwater but report otherwise. In fact, this group of homeowners has similar likelihood of becoming renters and falling behind their mortgage payment as those who report that they are underwater. This suggests that there are considerable proportion of underwater households that report their house values higher than their mortgage price. Their lack of home equity, however, leads to difficulty or reluctance in making their mortgages payments and also many of them reduce housing consumption. III.2. Previous Literature Starting from Kish and Lansing (1954), numerous studies have investigated whether homeowners accurately report their house value by measuring how the reported house value differs from the market proxy of house value. 44 The two most widely used proxies are the initial sales value adjusted by the house price indices (ex) Buck and Pence, 2006; Agarwal, 2007) or the 44 Underlying assumption of these studies is that the assessed or estimated house value is a better proxy of the true house value, which could, by itsel f, be a questionable assumption. For instance, individual households may have greater information about the status of their houses, such as recent renovation or particular amenities associated with the surrounding environment which may not be well captured in the assessed or estimated house prices. Further investigation is required to identify which measure is closer to the actual market value. Although I do not assume the estimated house value to be a better estimate of the market value, I regard the repor ted house value to be less accurate when its difference with the estimated house value is larger. 76 value measured by the appraisers (Kish and Lansing, 1954; Ihlanfeldt and Martinex-Vasquez, 1986). Others have used the estimated house value using the hedonic pricing method (ex) Follain and Malpezzi, 1981; Kuzmanko & Timmins, 2011). As the assessed or the appraised house values can also be bias, some studies have compared the self-reported house prices to the subsequent or previous sale price (Goodman and Ittner, 1992; Kiel and Zabel, 1999). On average, the studies find that the homeowners‘ self -reported house prices are 2-15 percent higher than the market estimated prices although few studies that find that households understate their home price (Follain and Malpezzi, 1981). Not only have studies tried to estimate the accuracy of the reported house value, many have also tested whether some demographic or socioeconomic characteristics are related to the extent of the bias. For example, Kain and Quigley (1972) do not find evidence that race, age, income, employment and marital status are associated with the accuracy of owner‘s reported house value. The study, however, do find that household with a head that received more years of education are less likely to over reported their house value. On the other hand, Agarwal (2007) finds that the house price misestimation is correlated with house tenure, income, borrower‘s credit quality, age, and employment status. Goodman and Ittner (1992) find no correlation between any of these variables and the house value estimates. In addition to Agarwal (2007), some studies have suggested that the length of tenure is an important variable which explains the accuracy of the reported house value. Along with Argarwal (2007), Kiel and Zabel (1999) also find that relatively recent movers are more likely to overstate their house value, resulting in a greater upward bias compared to those with a longer tenure. They interpret their results by suggesting that short term homeowners tend to be over confident about their house value. On the other hand, Kumanko and Timmins (2011) find 77 opposite results and argue that long term homeowners are likely to have relatively outdated information about their home prices. How households report their house value is related to how they acquire and process information. As learning and processing new information takes time, the reported house value is likely to lag behind the market price. Kumanko and Timmins (2011) find evidence that support this hypothesis, showing that the difference between the market and the self-reported house price is negatively correlated with the appreciation rate. In a rising market households tend to under report their house value but when house prices are falling, households are more likely to over report the value. Bení tez-Silva et al. (2010) also relates the external market condition to the accuracy of the reported house value. This study looks at the market condition at the period of purchasing and finds that those who bought their house during the housing bust tend to be more accurate, and in some cases even underestimate their house value. The study, however, does not provide a clear explanation of why the timing of buying affects the accuracy of the reported house value. While the existing research examined the accuracy of the reported house value after controlling for various demographic, socioeconomic variables and market conditions, not many studies have looked at how the level of home equity affects how household‘s report their house value. One exception is Chan et al. (2014). This study uses the American Housing Survey and the Health and Retirement Study and finds that homeowners with little or no equity in their homes are more likely to overstate their house value than the market value to a greater degree. The underlying theory that explains their results is the nominal loss aversion, first suggested by Kahneman and Tversky (1979). According to this theory, most people obtain greater disutility from losses than utility received from the same about of gains. If the loss aversion theory holds, 78 than those who are categorized as underwater according to the estimated house market value will have higher likelihood to report their house price above their mortgage debt. This will lead to a greater gap between the estimated and the reported house value. Psychological inclination not only affects how people report their house value but can also influence the actual transactions. Genesove and Mayer (2001) and Engelhardt (2003) find that the nominal loss aversion cause households to place a higher weight on capital losses than on equivalent gains, which potentially leads to longer spells in housing units, especially in declining markets. In addition to looking the relationship between the home equity and the difference in the two house price measures, this study also looks at whether different measures of LTV ratio changes the relationship between negative equity and mobility. As mentioned in the introduction, different studies have used different LTV measures when examining negative equity-mobility relationship. If the proxies of negative equity do not affect its relationship with the likelihood of moving, then using either one of them would be less problematic. If not, researchers should be more cautious about using different proxies of LTV. When looking at the negative equity-mobility relationship, the study uses the difference and difference framework used in Chapter I. One of the major limitations of the previous studies that examined the lock-in effect is that they have not properly controlled for the possible endogeneity that can bias the results. Homeowners can self-select their level of home equity by choosing the initial level of down payment, or by extracting home equity. This behavior is unobserved in most datasets but is likely to be associated with their decision to move. Choi (2014) tackles this identification problem by using the 2007 housing market shock. The paper finds that the mobility of those who became underwater due to the unexpected house price shocks increased significantly compared to those with positive equity. While Choi (2014) uses 79 the reported house value, this study uses both proxies of house values to further identify whether the results differ according to the two measurements. Furthermore, this study also examines how the proxies of negative equity is related to household‘s likelihood to switch to rental housing and their likelihood to be behind in their mortgage payment. III.3. Data This study uses the Panel Study of Income Dynamics (PSID), which has followed U.S. households since 1968. Between 1968 and 1997, surveys were conducted annually and biannually thereafter. The major advantage of the PSID is that it contains extensive information on individual and family characteristics. This enables me to control for various demographic and socioeconomic factors which could affect the gap between the self-reported house value and the estimated house value. This study covers period from 1985 to 2011. The period includes the time when the housing market experienced an unprecedented boom and bust. As I have obtained the geo-coded PSID, I am able to calculate the estimated house value using House Price Indices provided by the Federal Housing Finance Agency (FHFA) and Zillow. Both indices have advantages and disadvantages. The FHFA index only covers Metropolitan Statistical Areas (MSAs) but is available from 1975. The FHFA uses the repeated sales method to obtain the index. The Zillow index is offered at the ZIP code level and thus covers larger geographical areas but the data starts from 1996. Thus, I can only calculate the estimated house price for those who bought their homes after 1996. Zillow uses their own automated valuation models (AVM) to measure the home value index. Instead of using the actual sale price, Zillow estimates the price for every home in their sample. While the appropriate geographical area and the method of calculating the index can be questioned, the correlation between the estimated 80 house value using the MSA HPI and the ZIP code HPI is 97.41 percent. 45 Thus, although I present the result using both estimations in most of my analyses, the results do not show much differences. Table III-1 presents the summary statistics. All the values in the table are weighted by the family weights provided by the PSID to improve the external representativeness of the households. In order to make the data comparable, the sample is restricted to homeowners in MSAs in which both the reported and the estimated LTV is available. During 1985 to 2011, 2.64 percent of homeowners are underwater water according to the reported LTV 46 while 5.05 percent of homeowners are underwater according to the estimated LTV using the MSA HPI. Those who are classified as underwater homeowners according to the estimated LTV using the ZIP code HPI accounts for 8.23 percent of homeowners. 47 Among the homeowners, 11.48 percent moved. Those who moved within the same state accounts for 9.4 percent, while 2.1 percent moved to another state. Those who moved to an owner occupied household accounts for 7.0 percent of the total homeowners while 3.5 percent moved to a rental unit. 48 The average age of the head is close to 53 years old. Black and Hispanic households account for 13 percent of the total sample of homeowners. The average household size is 2.64. More than 60 percent of homeowners received more than some level of college education. Those who never married accounts for 8.5 45 The correlation between the self -reported house value and the estimated house value using the MSA HPI is 79.36 percent while the correlation between the self -reported and the estimated house value using the ZIP code HPI is 79.99 percent. 46 If the househo lds living outside of MSAs are included, the percent of underwater homeowners falls to 2.2, indicating greater proportion of underwater homeowners reside in urban areas. 47 Note that the estimated LTV using the ZIP code HPI is only available from 1996. Since many homeowners became underwater following the crisis, the percent of underwater homeowners using the ZIP code HPI is considerably higher than the percent using the MSA HPI. 48 Note that the sum of those who move to owner occupied units and rental housing do not add up to the total moves. This is because there are households that do not report there tenure. 81 percent, while those who became single due to divorce, separation or death of spouse accounts for 16.4 percent. The average house value is around 175000 dollars and the average family income is around 78000 dollars. About 2.6 percent of household heads reported that they were unemployed. The average number of years that households lived in the same house is 7.1 years. [Table III-1] Summary Statistics Variable Mean Std. Dev. Reported LTV>100% 0.026 0.161 Estimated LTV(MSA HPI)>100% 0.050 0.219 Estimated LTV (ZIP HPI)>100% 0.082 0.275 Move 0.115 0.319 Own-Own 0.070 0.255 Own-Rent 0.035 0.185 Age 52.688 16.134 Black 0.083 0.276 Hispanic 0.045 0.207 No. of Family 2.617 1.354 High School 0.300 0.458 Some College 0.239 0.427 College 0.377 0.485 Single 0.085 0.279 Widowed 0.047 0.212 Divorced/Separated 0.117 0.321 House Value 174875 201055 Family Income 78033 108718 Unemployed 0.026 0.160 Years in House 7.095 6.656 No. of Observations 35222 Note: The summary statistics are weighted by the family weight provided by the PSID. The summary statistics show that the percentage of underwater homeowners are greater using the estimated house value, but it does not show whether this trend has changed over time. Figure III-1 displays the percentage of underwater homeowners using the three LTV ratios from 1985 to 2011. The blue line shows the percentage of homeowners with negative equity calculated by the estimated house price, the orange and the gray line indicate the percentage of 82 underwater homeowners using FHFA MSA HPI and the Zillow ZIP code HPI, respectively. In the graph, the orange and the gray line is always higher than the blue line. This indicates that many households report their house prices higher than the estimated house value. The graph also shows that the gap between the percent of underwater homeowners categorized by estimated and the reported LTV has increased following the 2007 housing market crisis. [Figure III-1] Percentage of Underwater Homeowners In order to further examine whether the LTV ratio affects how households report their house value, I classify the LTV ratio into eight categories and calculate the number of homeowners in each category. The estimated LTV is calculated using the MSA HPI. Consistent with Figure III-1, Table III-2 shows that the number of underwater homeowners almost doubles when the LTV ratio is calculated using the estimated house value compared to the number calculated by the reported house value. Column (3) shows the number of observations which falls into the same LTV category according to both LTV measures. Columns (4) and (5) divide 0.00 5.00 10.00 15.00 20.00 25.00 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1999 2001 2003 2005 2007 2009 2011 LTV Comparison: % Underwater Reported LTV Estimated LTV: MSA Estimated LTV: ZIP 83 the number of matched homeowners in column (3) by the numbers in columns (1) and (2) to show the percentage of underwater homeowners that are in the same LTV category. Among the total homeowners in the sample, 68.27 percent are in the same LTV category. Greatest percentage of households is classified into the same category for those with LTV ratio less than 20 percent. In this category, 80.99 percent of homeowners are matched when divided by the number of homeowners in column (1) and 84.42 percent are matched when divided by the number of homeowners in column (2). For positive equity homeowners in other LTV categories, the percent matched ranges between 62.01 and 76.32 percent. For these homeowners, the absolute difference between the percentages in column (4) and (5) ranges from 0.68 to 7.37 percent. This suggests that for considerable proportion of positive equity households, the gap between the reported and the estimated house value is small. [Table III-2] No. of Underwater Homeowners: Rep. LTV vs. Est. LTV (MSA) LTV Category (%) (1) (2) (3) (4) (5) Reported LTV Estimated LTV (MSA) # Matched % Matched (3)/(1) % Matched (3)/(2) 0<LTV≤20 2,094 2,009 1,696 80.99% 84.42% 20<LTV≤40 4,300 4,259 3044 70.79% 71.47% 40<LTV≤60 6,196 6,621 4,106 66.27% 62.01% 60<LTV≤80 9,277 8,321 5,955 64.19% 71.57% 80<LTV≤100 7,422 7,104 5,422 73.05% 76.32% 100<LTV≤120 683 1,140 345 50.51% 30.26% 120<LTV≤140 232 386 82 35.34% 21.24% 140<LTV 298 662 173 58.05% 26.13% Total 30,502 30,502 20,823 68.27% 68.27% As for underwater homeowners, however, the differences in percentage in columns (4) and (5) are substantial. Among those who are classified as underwater homeowners according to the reported house value, the percent of households that are in the same LTV category, using the reported LTV categorization ranges from 35.41 to 58.05 percent. When using the estimated LTV 84 categorization, less than 31 percent of households are in the same negative equity category. 49 Table III-A1 in the appendix shows the same results for the estimated LTV using the ZIP code HPI. The findings are similar from those in Table III-3. Appendix Table III-A2 compares the two estimated LTV using the MSA and the ZIP code LTV, and finds that 86 percent of those who are categorized as underwater homeowners according to the LTV estimated by the MSA HPI are also in the negative equity category using the ZIP code HPI. This suggests that there are less differences between the two estimated LTV ratios compared to the difference between the reported and the estimated LTV. Figure III-2 classifies the homeowners into four groups. The numbers in the figure shows that among the 30,502 homeowners in the total sample, 27,099 of them are categorized as positive equity households according to both proxies of LTV. The number of homeowners who report they are underwater is 1,213, while 2,188 homeowners are estimated to be underwater. Those who are both reported and estimated to be underwater are 847. This indicates that over 60 percent of households who are estimated to be underwater report their house value higher than their remaining mortgage principal. While the study of Chan et al. (2014) only compares how the reported and the estimated house value differs between those with LTV≥80% and those with LTV<80%, this study uses the four classification in Table III-3 to further identify how the level of reported and the estimated equity is simultaneously associated with how households make decisions in the housing market. 49 Note that the final category of the LTV has no upper bound. Thus, there are more households in this category compared to the underwater homeowners in other two negative equity categories. 85 [Figure III-2] Four Groups of Homeowners: based on Reported & Estimated LTV Although Table III-1, III-2, and Figure III-1, III-2 suggest that those who are estimated to be underwater homeowners are more likely to report their house value higher than those with positive equity, it is still questionable whether these results still hold after controlling for individual‘s demographic and socio economic conditions as well as controlling for the time and location fixed effects. The method section presents empirical models to answer this question. I also describe models that examine how the level of home equity is related to households‘ likelihood of moving, their likelihood of switching to rental housing and their likelihood of being delinquent. The paper is especially interested in examining whether those who are estimated to be underwater but report that they have positive equity make different choices from those who report themselves as underwater homeowners. 86 III.4. Methodology This paper examines three questions: 1) Do the level of home equity affect how homeowners report their house value? 2) Does the estimated mobility of underwater homeowners differ according to the two measures of negative equity? 3) How does the two measures affect household‘s likelihood of switching to rental housing or falling behind mortgage payment? The first question investigates whether there is a systematic association between the level of home equity and households‘ self -assessment of house values. The second question looks at whether homeowners‘ perception of house value affect their decision to move. While homeowners may not want to admit their losses, they may be facing financial constraints due to the reduction of home equity. Moreover, some may be aware of their situation although they are reluctant to report their house value below their mortgage price. The last question looks into this by examining households‘ tenure choices and their decision to continue making their mortgage payment. In order to answer the question (1), I use the following two regressions: P( ) (1) ( ) ( ) (2) The first model use the logit regression to test whether the results in Table III-2 hold after controlling for various households‘ characteristics and state and time fixed effects. The dependent variable P(Same LTV t ) is the probability that the household is in the same LTV category. The LTV is divided into 8 categories as in Table III-2. The equals to 1 if household is underwater according to the estimated LTV. I use the estimated LTV calculated by 87 both the MSA and ZIP code HPI. A negative indicates that those who are estimated to be underwater are less likely to be in the same LTV category compared to those who are estimated to be above water. represents all control variables related to the household‘s demographic and socioeconomic characteristics. I also control for state ( ) and year ( ) fixed effects. is the random error term which is assumed to be normally distributed Regression (2) further examines how the level of home equity is related to the discrepancy between the reported and the estimated house value. The dependent variable is the log difference between the two house prices. I use the OLS regressions which include time and location fix effects. A positive indicates that the gap between the reported and the estimated house value is higher for those who are estimated to be underwater compared to those who are not. While model (1), classifies homeowners into two groups based on their estimated home equity status, in model (2), I use two additional classifications of homeowners. First, I categorize the estimated LTV ratio into 8 categories as in Table II to further investigate the relationship between the level of home equity and the gaps in the two house values. If a person‘s tendency of avoiding losses are incorporated in their house value reports, we will observe that the difference between the reported and the estimated house value will increase as households become more underwater, as many of these homeowners will report their house values higher than their remaining mortgage principal. Finally, I categories the home owners into four categories ( ) shown in Figure II. The four categories are 1) homeowners who have positive equity according to both LTV measures, 2) homeowners who are underwater by the self-reported but not the estimated LTV, 3) homeowners who are underwater only by the estimated LTV and 4) homeowners who are underwater according to both LTV measures. If the loss aversion theory 88 holds, the gap between the two house values will be the greatest for homeowners who report themselves to be above water but are estimated to be underwater. Next, I examine whether the negative-equity mobility relationship differs according to the two proxies of LTV. I use the difference in differences framework to mitigate endogeneity following Choi (2014). The model assumes that without the housing market shock, the difference in the mobility rate between those with positive equity and those with negative equity would have remained unchanged. I use the following logit model to test this hypothesis: ( ) (3) Where the dependent variable ( ) is the probability of household i moving between year t-2 and t. 50 equals 1 if households moved since the last survey. is a dummy variable which equals 1 if households are underwater (LTV≥100%) at the previous period. I run two separate regressions using two classifications of the underwater homeowners and compare the results. is a dummy variable which equals to 1 from years 2007 to 2011. The variable captures the changes of mobility for positive equity households following the crisis. The interaction term examines whether the changes in the likelihood of moving differs between underwater homeowners and those with positive equity since the housing market shock. Thus, ‗ ‘ measures the unbiased relationship between negative equity and 50 For mobility regression, I use the period from 1999 to 2011 to compare the change in the mobility during the boom and the bust period. The PSID survey was conducted biannuall y since 1997, and thus the percentage of households who answered they moved from the previous survey year increases significantly from 1999. Therefore, when using the difference in differences method, including the data prior to 199 7 can be problematic. F or most of the regressions in the paper, I use the sample period from 1985 to 2011, as greater sample size enhances the statistical preciseness. 89 mobility. If the value is significantly positive it indicates that negative equity triggers mobility. represents variables for the demographic and socioeconomic characteristics of households that may influence their mobility decision. To further examine whether homeowners‘ mobility differs according to how they report their house values, I also categorize the homeowners into four categories as in model (2). The result will show whether those who do not report themselves as underwater homeowners but are estimated be underwater show similar mobility patterns compare to the homeowners in other three categories. Final two regressions examines whether households‘ self -perception of their home equity status affect their tenure choices and their mortgage payment. Quigley (1987) and Stein (1995) point out that underwater homeowners face greater financial constraints compared to positive equity households. Moving requires transaction costs and down payment and underwater homeowners may not have enough financial resources to pay for these costs. While the financial constraints may deter underwater homeowners from moving, as negative equity also triggers default, underwater homeowners may move more than those with positive equity, especially when the market condition is negative. If they choose to move, underwater homeowners are more likely to move to rental units compared to positive equity households due to financial constraints (Choi, 2014). If all the households correctly report their house value, then their likelihood of moving to a rental unit would not be affected by differences in the two measures of home equity. If, however, those who do not report themselves as underwater homeowners but estimated to be underwater show higher likelihood of moving to a rental housing than those who are estimated to be above water, this suggests that the reported house value for some households are inaccurate. In order to test this, I use the following multi-logit model: ( ) (4) 90 Where, categorizes the moves into three categories: 1) do not move (reference), 2) move to owner occupied units and 3) switch to rental housing. The key dependent variable ( ) classifies the homeowners into four groups as in model (2). The final regression replaces in model (4) with a dummy variable that indicates whether the household is behind their mortgage payment or not. From 2009, the PSID has asked households the following question ―Some people have had difficulties recently making their mortgage or loan payments. Are you (or anyone in your family living there) currently behind on (your/their) (mortgage/loan) payments for this (mortgage/loan)?‖ Using this variable , I further identify whether people are aware of their equity status. If all homeowners who state they are not underwater believe that they have positive equity, then they will be more likely to continue making their mortgage payment compared to those who believe that their home price has fallen below their mortgage debt. If, however, the homeowners who avoid admitting their negative equity status are actually aware that they are in negative equity according to the market house value, these homeowners would also have greater reluctance to make their mortgage payment compared to those who are estimated to have positive equity. III.5. Empirical Findings Reported and Estimated LTV Comparison Table III-3 presents the results of model (1). The dependent variable equals 1 if the homeowner is in the same LTV category according to the reported and the estimated house price. The results show that even after controlling for the socioeconomic and demographic characteristics of households, those who are underwater by the estimated LTV are less likely to 91 be in the same LTV category. This result holds whether I calculate the estimated LTV using the MSA or the ZIP code HPI. 51 As for the control variables, I find that households with older heads or black heads are less likely to be in the same LTV category while those living in a more expensive house or those who experienced divorce or separation are more likely to be in the same LTV category. Consistent with Kumanko and Timmins (2011), years in the house is negatively associated with the likelihood of being in the same LTV category. In all regressions, the size of the coefficient is the greatest for the dummy variable indicating whether households are estimated to be underwater. This suggests that the level of home equity is an important factor that affects how people report their house value. 51 Appendix Table III-A3 and III-A4 show the results using the ZIP code level HPI. 92 [Table III-3] Likelihood of being in the Same LTV Category VARIABLES (1) (2) (3) Est. LTV (MSA) >100% -2.116*** -2.172*** -2.128*** (0.088) (0.106) (0.108) Age -0.058*** -0.062*** (0.014) (0.014) Age2 0.001*** 0.001*** (0.0002) (0.0002) Black -0.353*** -0.340*** (0.082) (0.088) Hispanic -0.066 -0.006 (0.124) (0.136) High School 0.068 0.072 (0.094) (0.092) Some College 0.033 0.041 (0.099) (0.099) College 0.093 0.071 (0.100) (0.100) Single 0.055 0.038 (0.087) (0.087) Widowed 0.040 0.056 (0.212) (0.210) Divorced/Separated 0.192** 0.168** (0.084) (0.084) Log(House Value) 0.212*** 0.263*** (0.041) (0.050) Log (Family Income) -0.031 -0.043 (0.041) (0.042) Unemployed 0.083 0.113 (0.114) (0.115) Years in House -0.061*** -0.062*** (0.005) (0.005) Constant 0.975*** 0.449 0.169 (0.023) (0.487) (0.629) Year FE N N Y State FE N N Y Observations 30,015 27,205 27,199 Pseudo R2 0.049 0.071 0.077 Note: Robust standard errors in parentheses * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. 93 The next three regressions examine the degree of which the reported house value differs from the estimated house value. The dependent variable in all three regressions is the gap between the reported and the estimated house value. A positive coefficient means households report their house value higher than the estimated house value. Table III-4 52 shows that those who are estimated to be underwater report their house value higher than the estimated house value compared to those who are estimated to be above water. The result is consistent with the loss aversion theory. The result also shows that the gap between the reported and the estimated house value is slightly larger for older households but the size of the coefficient is close to zero. On average, those with a college degree report their house value closer to the estimated value compared to those with less than high school education. While Table III-4 shows that those who live in a more expensive house are more likely to be in the same LTV category, the gap between the reported and the estimated house value is greater for these people. The years lived in the current house is positively associated with the gap between the reported and the estimated house value. Again, among all variables, the dummy variable for the estimated underwater homeowners show the largest size of coefficient. 52 In all titles of tables Rep. inducates ―Reported‖ and Est. indicates ―Estimated‖. 94 [Table III-4] Difference between Rep. & Est. House Value: Est. LTV>100% Note: Robust standard errors in parentheses * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. MSA Estimated LTV VARIABLES (1) (2) (3) Est. LTV>100% 0.452*** 0.511*** 0.522*** (0.059) (0.071) (0.071) Age -0.006* -0.008** (0.003) (0.003) Age2 0.000 0.000** (0.000) (0.000) Black 0.021 0.032 (0.020) (0.022) Latino -0.025 0.008 (0.017) (0.021) High School -0.025 -0.027 (0.020) (0.020) Some College -0.037 -0.043* (0.024) (0.024) College -0.056** -0.080*** (0.026) (0.027) Single 0.008 0.040 (0.027) (0.025) Widowed 0.027 0.026 (0.046) (0.045) Divorced/Separated 0.010 0.031* (0.016) (0.017) Log(House Value) 0.111*** 0.161*** (0.014) (0.019) Log (Family Income) -0.024** -0.014 (0.011) (0.012) Unemployed -0.012 -0.003 (0.015) (0.014) Years in House 0.002** 0.004*** (0.001) (0.001) Constant -0.009** -0.865*** -1.399*** (0.003) (0.180) (0.250) Year FE N N Y State FE N N Y Observations 29,891 27,140 27,140 R-squared 0.123 0.192 0.230 95 In Table III-5, I classify estimated LTV into eight categories as in Table III-2. The reference group is those with LTV between 80 and 100 percent. The results show that the gap between the reported and the estimated house value increases as the level of home equity decreases. Also, the absolute size of the coefficient is greater for those with negative equity than those with positive equity. Those with LTV greater than 140 percent has the largest coefficient value, suggesting that those who are significantly underwater report their house value most differently from the estimated house value. [Table III-5] Difference between Rep. & Est. House Value: 8 LTV Categories MSA Estimated LTV VARIABLES (1) (2) (3) Est. LTV≤20% -0.135*** -0.182*** -0.228*** (0.018) (0.019) (0.021) 20%<Est. LTV≤40% -0.084*** -0.130*** -0.169*** (0.008) (0.010) (0.011) 40%<Est. LTV≤60% -0.055*** -0.086*** -0.112*** (0.006) (0.007) (0.008) 60%<Est. LTV≤80% -0.032*** -0.046*** -0.061*** (0.004) (0.005) (0.005) 100%<Est. LTV≤120% 0.135*** 0.124*** 0.128*** (0.013) (0.013) (0.013) 120%<Est. LTV≤140% 0.288*** 0.295*** 0.296*** (0.027) (0.030) (0.029) 140%<Est. LTV 0.940*** 1.026*** 1.005*** (0.146) (0.160) (0.158) Year FE N N Y State FE N N Y Observations 29,891 27,140 27,140 R-squared 0.219 0.306 0.350 Note: Robust standard errors in parentheses * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. 96 For further investigation, in Table III-6, I match all estimated and the reported LTV categories. Table III-6 shows that 63.44 percent of homeowners with estimated LTV is greater than 140 percent report themselves as having positive equity. This explains why the gap between the reported and the estimated house value is largest for these homeowners. Almost half the homeowners who are categorized as underwater homeowners according to the estimated LTV either report they are in the LTV 80-100 or in the LTV 60-80 category. This suggests that most underwater homeowners report their house value higher than their mortgage value but they do not report the value too high so that their LTV ratio falls below 60 percent. 53 53 The coefficients of the control variables in Table III-4 and III-5 show insignificant changes from those in Table III-3 and thus are not reported. 97 [Table III-6] Matching Est. and Rep. LTV Estimated LTV Reported LTV 0<LTV≤20 20<LTV≤40 40<LTV≤60 60<LTV≤80 80<LTV≤100 100<LTV≤120 120<LTV≤140 140<LTV 0<LTV≤20 84.42% 8.34% 0.38% 0.10% 0.08% 0.09% 0.00% 0.45% 20<LTV≤40 13.69% 71.47% 12.05% 1.24% 0.51% 0.79% 3.11% 3.47% 40<LTV≤60 0.95% 16.51% 62.01% 12.47% 2.52% 3.86% 6.48% 12.39% 60<LTV≤80 0.50% 2.93% 23.11% 71.57% 17.38% 17.81% 17.10% 23.11% 80<LTV≤100 0.35% 0.49% 1.95% 13.50% 76.32% 40.18% 26.68% 24.02% 100<LTV≤120 0.05% 0.05% 0.17% 0.67% 2.36% 30.26% 16.58% 5.44% 120<LTV≤140 0.05% 0.05% 0.12% 0.23% 0.53% 4.30% 21.24% 4.98% 140<LTV 0.00% 0.16% 0.21% 0.23% 0.28% 2.72% 8.81% 26.13% 98 Table III-7 classifies homeowners into four groups: (1) those who have positive equity according to both reported and estimated LTV ratio (2) those with negative equity by the reported LTV but with positive equity by the estimated LTV (3) those with positive equity by the reported LTV but with negative equity by the estimated LTV and (4) those with negative equity according to both LTV ratios. The reference group is category (1). Table III-7 presents several interesting findings. First, compared to the reference group, those who report themselves as underwater homeowners but are estimated to have positive equity show smaller gap between the reported and estimated house value. On the other hand, the gap between the two house values is greatest for those who report that they have positive equity but are estimated to be underwater. The coefficient for households who are categorized as underwater homeowners by the two LTV measures are insignificant in column (1), but when control variables are added, the coefficient becomes significant. The size of the coefficient, however, is close to zero. [Table III-7] Difference between Rep. & Est. House Value: 4 Homeowner Groups VARIABLES (1) (2) (3) Rep. LTV >100% Est. LTV≤100% -0.495*** -0.429*** -0.397*** (0.041) (0.046) (0.046) Rep. LTV ≤100% Est. LTV>100% 0.689*** 0.732*** 0.726*** (0.080) (0.093) (0.092) Rep. LTV >100% Est. LTV>100% 0.013 0.042*** 0.070*** (0.013) (0.016) (0.018) Age -0.006** -0.008*** (0.003) (0.003) Age2 4.92e-05 7.31e-05** (3.09e-05) (3.17e-05) Black 0.018 0.027 (0.018) (0.020) Hispanic -0.010 0.015 (0.014) (0.018) 99 [Table III-7] (Continued) VARIABLES (1) (2) (3) High School -0.026 -0.027 (0.019) (0.019) Some College -0.040* -0.044* (0.022) (0.023) College -0.058** -0.077*** (0.024) (0.027) Single 0.005 0.030 (0.023) (0.023) Widowed 0.015 0.014 (0.043) (0.042) Divorced/Separated 0.006 0.024 (0.015) (0.017) Log(House Value) 0.099*** 0.139*** (0.013) (0.019) Log (Family Income) -0.018* -0.010 (0.011) (0.011) Unemployed 0.002 0.006 (0.013) (0.013) Years in House 0.001 0.003*** (0.001) (0.001) Constant 0.040*** -0.906*** -1.698*** (0.004) (0.174) (0.263) Year FE N N Y State FE N N Y Observations 29,891 27,140 27,140 R-squared 0.217 0.265 0.294 Note: Robust standard errors in parentheses * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. 100 These results indicate that those who admit they are underwater report their house value closely to the estimated house value. Among these homeowners, those who are estimated to have positive equity show even smaller gap between the two house values. These homeowners may be more accurate about their house value due to the private information they hold about their homes or may generally be more pessimistic. On the other hand, those who do report that they have positive equity but are estimated to be underwater have greatest tendency to over report their house value reflecting their reluctance to admit their losses. The findings suggest that the level of home equity and households‘ attitude towards losses simultaneously affect how households report their house values relative to the estimated house value. Negative Equity-Mobility: Difference between the Two LTVs The following regressions test whether using different proxies of LTV affect the relation between negative equity and mobility. Table III-8 presents the result of difference in difference model where the dependent variable equals 1 if the household moved to a different house from the previous survey year. In the first three columns, the underwater homeowners are classified using the reported LTV, while the in following three columns the homeowners are classified by the estimated LTV. The results show that the negative equity-mobility relationship differs depending on the proxy of LTV. Prior to the crisis, the coefficient for those with reported LTV above 100 percent shows opposite signs from the coefficient for those with estimated LTV above 100 percent, although both coefficients are statistically insignificant. Following the crisis, the likelihood of moving significantly increased for those who reported themselves as underwater 101 homeowners but no such patterns are observed for those who are estimated to be underwater homeowners. 54 [Table III-8] Difference in Differences: Likelihood of Moving Reported LTV Estimated LTV VARIABLES (1) (2) (3) (4) (5) (6) Rep. LTV>100% -0.054 -0.361 -0.397 (0.317) (0.346) (0.346) Est. LTV>100% 0.269 0.146 0.0762 (0.211) (0.220) (0.219) Year≥2007 -0.114* -0.083 -0.088 -0.104* -0.069 -0.078 (0.0589) (0.067) (0.067) (0.060) (0.068) (0.069) Rep. LTV>100%*Year≥2007 0.978*** 0.957** 0.978** (0.358) (0.395) (0.395) Est. LTV>100%*Year≥2007 0.252 0.155 0.252 (0.244) (0.260) (0.259) Age -0.124*** -0.126*** -0.124*** -0.126*** (0.0153) (0.0154) (0.015) (0.015) Age2 0.001*** 0.001*** 0.001*** 0.001*** (0.000) (0.000) (0.000) (0.000) Female Head -0.042 -0.036 -0.031 -0.026 (0.145) (0.147) (0.145) (0.147) Black -0.204 -0.218 -0.217 -0.229 (0.132) (0.141) (0.134) (0.142) Hispanic -0.062 -0.115 -0.050 -0.108 (0.136) (0.145) (0.138) (0.146) High School 0.007 0.023 0.007 0.023 (0.029) (0.029) (0.029) (0.029) Some College -0.073 -0.049 -0.075 -0.051 (0.149) (0.152) (0.149) (0.152) College 0.067 0.092 0.059 0.085 (0.259) (0.263) (0.259) (0.263) Single 0.404*** 0.398*** 0.397*** 0.391*** (0.140) (0.141) (0.141) (0.142) 54 As for other control variables, I find that the age of the head is negatively associated with the likelihood of moving but the change in the likelihood also decreases as heads gets older. Those who are divorced or are unemployed are more likely to move while those who lived in their house for a long er period are less likely to move. 102 [Table III-8] (Continued) Reported LTV Estimated LTV VARIABLES (1) (2) (3) (4) (5) (6) Widowed -0.0441 -0.044 -0.040 -0.040 (0.146) (0.148) (0.147) (0.148) Divorced/Separated -0.0821 -0.084 -0.082 -0.083 (0.148) (0.152) (0.148) (0.152) Log(House Value) 0.015 0.045 0.019 0.0480 (0.147) (0.148) (0.147) (0.148) Log (Family Income) -0.074 -0.064 -0.072 -0.063 (0.052) (0.051) (0.052) (0.052) Unemployed 0.393** 0.442** 0.398** 0.447*** (0.172) (0.173) (0.172) (0.173) Years in House -0.042*** -0.038*** -0.043*** -0.039*** (0.007) (0.007) (0.007) (0.007) Constant -1.830*** 2.635*** 2.940*** -1.841*** 2.609*** 2.913*** (0.044) (0.594) (0.652) (0.044) (0.594) (0.654) State FE N N Y N N Y Observations 13,314 12,429 12,400 13,314 12,429 12,400 Pseudo R2 0.00333 0.0499 0.0569 0.00201 0.0491 0.0561 Note: Robust standard errors in parentheses * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. Next table (Table III-9) examines why different proxies of LTV ratio lead to different results by classifying homeowners into four groups, as in Table III-7. Prior to the crisis, those who report themselves as underwater homeowners are less likely to move than those who do not report themselves as underwater homeowners but are estimated to be so. During this period, however, the coefficients for all three categories of homeowners are not statistically different from zero. Following the crisis, however, those who report themselves as underwater homeowners are significantly more likely to move than the reference group who are both estimated and reported to have positive equity. Those who report themselves as underwater homeowners but are not estimated to be underwater also show an increase in the likelihood of moving, but the statistical power is small. On the other hand, those who are underwater only by 103 the estimated LTV ratio show slightly negative, although insignificant, decrease in the likelihood of moving compared to those in the reference group. [Table III-9] Difference in Differences: Likelihood of Moving: 4 Homeowner Groups VARIABLES (1) (2) (3) Before 2007 Rep. LTV >100% Est. LTV≤100% -0.143 -0.224 -0.265 (0.560) (0.565) (0.564) Rep. LTV ≤100% Est. LTV>100% 0.362 0.369 0.286 (0.256) (0.259) (0.260) Rep. LTV >100% Est. LTV>100% 0.0241 -0.446 -0.484 (0.373) (0.419) (0.419) After 2007 Rep. LTV >100% Est. LTV≤100% 0.510 0.219 0.188 (0.712) (0.812) (0.812) Rep. LTV ≤100% Est. LTV>100% -0.297 -0.441 -0.306 (0.313) (0.333) (0.332) Rep. LTV >100% Est. LTV>100% 1.010** 1.131** 1.170** (0.417) (0.465) (0.466) Year≥2007 -0.108* -0.070 -0.079 (0.060) (0.068) (0.069) State FE N N Y Observations 13,314 12,429 12,400 Pseudo R2 0.00388 0.0505 0.0573 Note: Robust standard errors in parentheses. * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. This means that among those who are estimated to be underwater in Table III-8, the likelihood of moving only increased for those who report themselves as underwater homeowners. For those who said they had positive equity --who accounts for more than 60 percent of those 104 who are estimated to be underwater -- the mobility rate slightly decreased following the crisis. This explains why the coefficient for the estimated underwater homeowners in Table III-8 shows no statistical significance following the crisis. Why those in the negative equity category only by the estimated LTV ratio have lower likelihood of moving compared to those who report themselves as underwater homeowners has two possible explanations. First, those who are in this category may actually have positive equity as they incorporate their own information when assessing their house value that is not captured by the house price index. Thus, they may behave similarly to those in the positive reported- positive estimated equity category. Another explanation is that these people are in fact underwater but do not want to face financial losses. This psychological tendency not only influences them to report their house price upwards but also affects their decisions to move, as by moving they will realize losses. These two explanations will be further looked into in the final two regressions. Why those who are underwater only by the reported LTV ratio also show lower likelihood of moving than those who are classified as underwater by both LTV measures is less clear. One possible explanation is the depth of negative equity. The mean value of the reported LTV ratio is greater for those in the negative equity category according to both LTVs compared to those in the negative equity category only by the reported LTV. 55 Since the likelihood of moving due to default increases with the level of negative equity (Buttha et al., 2010), those in the negative reported-positive estimated equity category may have enough financial resources to hold onto their current housing. On the other hand, those in the negative reported-negative 55 The mean reported LTV value is 126.37 for those who are underwater only by the reported house value, while it is 142.35 for those who are underwater according to both proxies. The mean estimated LTV for the former group is 81.05 while the value is 146.66 for the latter group. 105 estimated equity category may choose to or be forced to default as they face greater financial stress to continue making their mortgage payments. Another possible explanation for the difference across the two groups may be due to the fact that only small number of homeowners is in the negative reported-positive estimated equity category. 56 Due to the small sample size, the standard error for these households are larger compared to households in other three categories. Thus, the statistical significance is smaller for homeowners in this category. The final two regressions examine which of the two proxies of LTV is closer to reality. If homeowners in the positive reported-negative estimated equity category are in fact above water, then they will not face greater financial constraints when they move compared to those are in the positive equity category according to both LTV measures. On the other hand, if they are in fact underwater, they will be more likely to move to rental housing as they may not be able to afford transaction costs and make initial down payment when purchasing a new house (Choi, 2014). Thus, in Table III-10, I categorize that dependent variable into three categories: (1) no moves (reference), (2) moving to another owner occupied housing, and (3) moving to rental housing. 57 The result of the multinomial model suggests that households in all three categories have similar likelihood of moving to an owner occupied housing compare to those in the reference category. The likelihood of moving to a rental unit, however, is significantly larger for those in the three categories of underwater homeowners compared to those in the reference group. Even after including control variables and year and state fixed effects, size of the coefficient for all 56 During 1999 -2011, only 106 are in this category. 57 In this regression, I use the data from 1985 -2011. I find similar results when I use data from 1999 -2011 but since there are not many people who move to rental housing, the standard error for each category of homeowners increases, decreasing the statistical power. 106 three classification of underwater homeowners remains similar. 58 This indicates that considerable number of households are likely to be underwater although they do not report they have negative equity. While these homeowners move less than those who report that they have negative equity, when they do move they are more likely to move to rental housing than homeowners in the reference group. 59 [Table III-10] Likelihood of Switching to Rental Housing Own to Own Own to Rent VARIABLES (1) (2) (3) (4) (5) (6) Rep. LTV >100% Est. LTV≤100% -0.239 -0.195 -0.290 1.098*** 0.659** 0.523 (0.336) (0.345) (0.358) (0.289) (0.332) (0.342) Rep. LTV ≤100% Est. LTV>100% 0.0145 -0.013 -0.031 0.658*** 0.571*** 0.533*** (0.172) (0.176) (0.178) (0.178) (0.187) (0.187) Rep. LTV >100% Est. LTV>100% 0.172 -0.148 -0.221 1.068*** 0.784*** 0.541** (0.208) (0.208) (0.210) (0.213) (0.222) (0.230) Age -0.087*** -0.085*** -0.147*** -0.155*** (0.0144) (0.0145) (0.0214) (0.0207) Age2 0.001*** 0.001*** 0.001*** 0.001*** (0.000) (0.000) (0.001) (0.000) Female Head -0.350** -0.409*** 0.0056 -0.053 (0.143) (0.142) (0.173) (0.173) Black -0.567*** -0.669*** 0.254* 0.250* (0.167) (0.169) (0.142) (0.151) 58 The likelihood of moving to rental housing is not statistically significant for those who are categorized as underwater homeowners usi ng the reported LTV and not by the estimated LTV. Again, this may be due to fact that they are less underwater than those who are estimated to be underwater, or because of the small sample size. 59 The results also show that age is negatively associated with the likelihood of moving to both owner occupied and rental units. Households with female or black heads are less likely to move to owner occupied house, while households with black heads have greater likelihood of moving to a rental unit. Number of family members reduces the likelihood of moving to an owner occupied house but increases the likelihood of becoming renters, family income shows opposite relationship. As expected, being divorced or separated, increases the likelihood of moving to both owner occupied and rental units. Finally, those who have lived in their house for longer years are less likely to move to both owner occupied households or rental units. 107 [Table III-10] (Continued) Own to Own Own to Rent VARIABLES (1) (2) (3) (4) (5) (6) Latino -0.086 -0.164 0.0399 -0.275 (0.133) (0.136) (0.193) (0.213) No. of Family -0.091*** -0.071*** 0.061* 0.081** (0.025) (0.025) (0.036) (0.036) Single -0.058 -0.141 0.494*** 0.385** (0.144) (0.147) (0.175) (0.178) Widowed 0.091 0.130 0.322 0.356 (0.247) (0.250) (0.282) (0.290) Divorced/Separated 0.476*** 0.408*** 0.880*** 0.838*** (0.133) (0.134) (0.174) (0.175) High School -0.0553 -0.0459 -0.136 -0.154 (0.135) (0.135) (0.162) (0.161) Some College -0.0768 -0.0734 -0.107 -0.184 (0.135) (0.139) (0.169) (0.171) College -0.008 0.0526 -0.264 -0.296* (0.132) (0.133) (0.173) (0.173) Log Family Income 0.285*** 0.165*** -0.053 -0.181*** (0.048) (0.053) (0.050) (0.040) Unemployed 0.114 0.149 0.576*** 0.578*** (0.169) (0.171) (0.189) (0.190) Years in House -0.038*** -0.040*** -0.053*** -0.057*** (0.007) (0.007) (0.011) (0.011) Constant -2.560*** -2.731*** -0.812 -3.337*** 1.021 3.209*** (0.031) (0.528) (0.635) (0.044) (0.633) (0.711) Year FE N N Y N N Y State FE N N Y N N Y Observations 32,028 30,664 30,664 32,028 30,664 30,664 Pseudo R2 0.002 0.041 0.065 0.002 0.041 0.065 Note: Robust standard errors in parentheses * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. Whether the homeowners are aware of the true equity status is the final question to be addressed in this chapter. The homeowners could be reporting their house value above their mortgage debt because they truly believe so, or they could be aware of their true equity status but could be reporting their house value higher as they do not want to admit their losses. Whether 108 homeowners pay their mortgage debt on time could provide some evidence about the households‘ awareness of their home equity status. Table III-11 finds that compared to the reference category, all three categories of homeowners have higher likelihood of being behind their mortgage payment. 60 In fact, among those who are estimated to be underwater, the size of the coefficient for those who do not report that they have negative equity is similar to those who report themselves as underwater homeowners. 61 This suggests that even among those who report their house price higher than their mortgage value, many are likely to know that their house price worth less than what they have reported and thus are less willing make their mortgage payment on time. 62 [Table III-11] Likelihood of being Delinquent VARIABLES (1) (2) (3) Rep. LTV >100% Est. LTV≤100% 0.630 0.059 0.307 (0.604) (0.773) (0.781) Rep. LTV ≤100% Est. LTV>100% 1.333*** 1.236*** 1.178*** (0.302) (0.299) (0.287) Rep. LTV >100% Est. LTV>100% 1.691*** 1.123*** 1.107*** (0.252) (0.304) (0.328) 60 One caveat is that the people in the negative reported-negative estimated equity category have higher likelihood of moving. If these homeowners had not moved, it is likely that the likelihood of being delinquent would be the highest for these homeowners. Also, the reason why the coefficient for those in the negative reported-positive estimated equity category is insignificant similar to the reason why their coefficient was insignificant when the dependent variable was whether the homeowner moved or not; Their sample size is small and the average home equity is higher than the other two groups of underwater homeowners. 61 Also, from 2009 survey, the PSID also asked the following question. ―How likely is it that (you/they) (will continue to be behind/will fall behind) on (your/their) (mortgage/loan) payments in the next 12 months?‖ I have also used this variable and found s imilar results from Table III-11. The results can be provided upon requests. 62 The results also find that older people are more likely to be delinquent. Hispanics, households with greater number of family members or with unemployed heads also have higher likelihood of being behind their mortgage payment. Meanwhile, family income is negatively associated with the l ikelihood of being delinquent . 109 [Table III-11] (Continued) VARIABLES (1) (2) (3) Age 0.225*** 0.222*** (0.067) (0.066) Age2 -0.002*** -0.002*** (0.000) (0.000) Female Head 0.175 0.230 (0.395) (0.367) Black 0.599** 0.713** (0.305) (0.296) Hispanic 0.892*** 0.987*** (0.290) (0.339) No. of Family 0.214*** 0.219*** (0.071) (0.074) Single 0.250 0.278 (0.433) (0.447) Widowed -0.852 -0.670 (1.137) (0.974) Divorced/Separated 0.278 0.101 (0.413) (0.400) High School -0.201 -0.380 (0.378) (0.391) Some College -0.531 -0.729* (0.420) (0.417) College -0.580 -0.725* (0.437) (0.421) Log Family Income -0.372*** -0.393*** (0.130) (0.138) Unemployed 1.119*** 1.160*** (0.326) (0.335) Years in House -0.040* -0.035 (0.023) (0.022) Constant -3.460*** -4.321** -3.135 (0.116) (2.203) (2.340) State FE N N Y Observations 4,485 4,159 3,879 Pseudo R2 0.042 0.159 0.201 Note: Robust standard errors in parentheses * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. 110 IV. Conclusion My dissertation focuses on underwater homeowners. Chapter I identifies whether minorities and immigrants are more likely to be underwater than whites and the native born. Even after controlling for households‘ socioeconomic and demographic characteristics, the study finds that black, Hispanic households are more likely to be underwater than white households. Asians, however, do not show a higher likelihood of being underwater in comparison to white, once the control variables are included. The immigrants also have higher likelihood of being underwater compared to the native born. The differences in the likelihood of being underwater also exist across the immigrants in different race and ethnic groups. However, the number of years that immigrants lived in the U.S. do not lower their likelihood of being underwater. Chapter I also shows that the differences in the likelihood of being underwater across race groups and immigrant status have mostly occurred following the 2007 housing market crisis. The results show that the gap remains after controlling for the period of buying and the level of initial down payment which indicates that minorities and immigrants may be paying relatively higher house price or mortgage cost. Further investigation is needed to pin down exact reasons behind this phenomenon, especially if we are interesting in reducing the likelihood of becoming underwater for minorities and immigrants. In addition, the first chapter finds evidence that racial composition of neighborhoods matters but it does not fully eliminate the differences in the likelihood of being underwater across race and ethnicity. Even after including the share of blacks and Hispanic households in each neighborhood, black and Hispanic are more likely to be underwater. On the other hand, the size of the coefficients for black and Hispanic dummies becomes smaller when the neighborhood variables are included, suggesting that the location is related to the likelihood of being 111 underwater. In fact, households living in neighborhoods with greater share of black and Hispanic households have greater propensity to be underwater, especially since 2007. Finally, non- Hispanic households in neighborhoods with greater share of Hispanic population are more likely to be underwater compared to the Hispanic households, indicating the existence of a positive network effect among Hispanics. Overall, the first study show that minorities and immigrant households have been more adversely affected by the recent housing market crisis. The second study in Chapter II finds that negative equity increases mobility. Using a difference-in differences framework, the paper shows that the mobility of those who became underwater due to the recent housing market crisis increased significantly than those with positive equity. Compared to the existing studies, this study better controls for endogenity using the housing market shock. It also examines theoretical reasons behind the negative equity- mobility relationship and finds evidence that supports the double trigger effect and the house price expectation theory. Meanwhile, although negative equity is found to be a triggering factor of residential mobility, it is important to acknowledge that in the post-crisis period, the shifts in the labor market conditions and expectation of house prices also affected the mobility decision of underwater homeowners. In fact, although negative equity may have been a major cause of moving, without the changes in the external environment, the mobility of underwater homeowners may have been less prevalent. This suggests that when we examine the negative equity-mobility relationship, we should also consider the external market conditions. In addition, while this study supports most of the recent studies that find no evidence of a lock-in effect, this evidence is insufficient to prove that the increase is the mobility of underwater homeowners enhances labor market efficiency. The final regression provide some evidence that 112 greater share of underwater homeowners have been moving states with higher unemployment rates since the crisis. This suggests that while the negative conditions of labor market may have stimulated the mobility of underwater household, the impact of house market debacle on the labor market is still uncertain, and thus needs further exploration. Chapter III investigates whether the psychological aversion towards losses affect how households report their house prices and also their decision to move. In accordance the nominal loss aversion theory, I find that many homeowners who are estimated to be underwater report their house value above their remaining mortgage value. These homeowners also show lower likelihood of moving, also reflecting their psychological tendency to avoid losses. When they do move, however, they are more likely to become renters. The study also finds that these homeowners are more likely to be delinquent than those who are estimated to be above water. These results indicate that considerable number of households who do not report themselves as underwater homeowners are in fact underwater. Also, some of these households are likely to be aware of their true equity status but may be reluctant to admit that they have negative equity. The findings suggest that policy makers and researchers should be cautious when using different proxies of negative equity, as households‘ view towards their level of home equity affect their behavior in the housing market. For example, while many studies use the estimated house value to examine households‘ likelihood of moving or their likelihood of default, the market value LTV might not have a direct impact on these choices. Rather households may process market value but react differently based on their psychological tendency. The study shows that using both proxies of LTV can enhance our understanding about how households‘ perceptions affect their decision making. Future studies may also benefit from comparing the different measures of house value when connecting this variable with households‘ choices. This 113 could help us to better understand how households process and respond to market information on house prices based on their own perception and their attitude towards losses. 114 References Chapter II. Bayer, P., Casey, M., D., Ferreira, F., & McMillan, R. (2012). Price Discrimination in the Housing Market. NBER Working Paper 18069. Bhutta, N., Bokko, J., & Shan, H. (2010). The Depth of Negative Equity and Mortgage Default Decisions, Working Paper. Bocian D.G., Ernst, K.S., Li W. (2008) Race, Ethnicity and Subprime Home Loan Pricing, Journal of Economics and Business 60, 110–124 Bocian D.G., Li W., & Ernst, K.S. (2010). Foreclosures by Race and Ethnicity: The Demographics of a Crisis, Center for Responsible Lending Research Report Choi, J.H. (2014). Measurement, Perceptions and Negative Equity-Mobility Relationship, Working Paper Coulson, N. E., & Grieco, P. (2013). Mobility and Mortgages: Evidence from the PSID. Regional Science and Urban Economics, 43, 1–7. Demyanyk, Y., Hryshko, D., José Luengo-Prado, M., & Sørensen, B. E. (2013). Moving to a Job: The Role of Home Equity, Debt, and Access to Credit. Federal Reserve Bank of Cleveland, Working Paper Series 13-05. Donovan, C., & Schnure, C. (2011). Locked in the House: Do Underwater Mortgages Reduces Labor Market Mobility. Working Paper. Elul, R., Souleles, N. S., Chomsisengphet, S., Glennon, D., & Hunt, R. (2010). What ―Triggers‖ Mortgage Default? Federal Reserve Bank of Philadelphia Working Paper, No. 10-13. Ferreira, F., Gyourko, J., & Tracy, J. (2010). Housing Busts and Household Mobility. Journal of Urban Economics, 68, 34–45. 115 Ferreira, F., Gyourko, J., & Tracy, J. (2011). Housing Bust and Household Mobility: An Update. NBER Working Paper 17405. Foote, C. L., Gerardi, K., & Willen, P. S. (2008). Negative Equity and Foreclosure: Theory and Evidence. Journal of Urban Economics, 64(2), 234–245. Follain, J. R., & Malpezzi, S. (1981). Another Look at Racial Differences in Housing Prices. Urban Studies, 18(2), 195–203. Ghent, A.C., Hernández-Murillo R., & Owyang M.T. (2014) Differences in Subprime Loan Pricing Across Races and Neighborhoods, Federal Reserve Bank of St. Louis Working Paper Harding, J. P., Rosenblatt, E., & Yao, V. W. (2009). The Contagion Effect of Foreclosed Properties. Journal of Urban Economics, 66, 164–178. Kain, J. F., & Quigley, J. M. (1975). Housing Markets and Racial Discrimination: A Microeconomic Analysis. NBER, Cambridge MA. Kiel, K. A., & Zabel, J. (1996). Housing Markets and Racial Discrimination: A Microeconomic Analysis. Journal of Housing Economics, 5(2), 171–189. King, A. T., & Mieszkowski, P. (1973). Racial Discrimination, Segregation, and the Price of Housing. Journal of Political Economy, 81(3), 590–608 Luea , H., Reicchenberger, A. and Turner, T. (2011) Mortgage Default by 2009: Effects of Race, Ethnicity and Economic Standing During the Boom Years, Working Paper. Mayer, C. J. and K. Pence (2008). Subprime mortgages: What, where, and to whom? Working Paper 14083, National Bureau of Economic Research. Mian A.M. & Sufi A., (2009). The Consequences of Mortgage Credit Expansion: Evidence from the U.S. Mortgage Default Crisis, Quarterly Journal of Economics, 124 (4), 1449-1496. 116 Myers, D., & Lee, S. W. (1998). Immigrant Trajectories into Homeownership: A Temporal Analysis of Residential Assimilation. International Migration Review, 32(3), 593–625. Painter, G., Gabriel, S., & Dowell, M. (2001). Race, Immigrant Status, and Housing Tenure Choice. Journal of Urban Economics, 49(1), 150–167. Painter, G., & Yu, Z. (2013). Caught in the Housing Bubble: Immigrants‘ Housing Outcomes in Traditional Gateways and Newly Emerging Destination. Urban Studies, 51(4), 781-809. Quercia, R G., McCarthy G.W., & Stegman M.A. (1995). Mortgage Default among Rural, low- Income Borrowers, Journal of Housing Research, 6(2), 349–369. Quigley, J.M. & Van Order, R. (1995). Explicit Test of Contingent Claims Modes of Mortgage Defaults, Journal of Real Estate Finance and Economics, 22(3), 99–117. Rugh J.S. & Massey D.S. (2010) Racial Segregation and the American Foreclosure Crisis American Social Reviews, 75(5), 629–651. Schulhofer-Wohl, S. (2011). Negative Equity Does Not Reduce Homeowners‘ Mobility. NBER Working Paper 16701. Schwartz, A. F. (2010). Housing Policy in the United States (2nd ed.). Routledge. Williams, R. (2012). Using the Margin Command to Estimate and Interpret Adjusted Predictions and Marginal Effects. Stata Journal, 12(2), 308–331. White, B. T. (2010). Underwater and Not Walking Away: Shame, Fear and the Social Management of the Housing Crisis. Wake Forest Law Review, 45, 971–1023. Yinger, J. (1978). The Black-White Price Differential in Housing: Some Further Evidence. Land Economics, 54(2), 187–206. 117 Chapter II. Bení tez-Silva, H., Eren, S., Heiland, F., & Jiménez-Martí n, S. (2010). How Well do Individuals Predict the Selling Prices of their Homes? Working Paper. Bhutta, N., Bokko, J., & Shan, H. (2010). The Depth of Negative Equity and Mortgage Default Decisions, Working Paper. Bucks, B., K., & Bricker, J. (2013). Household Mobility over the Great Recession: Evidence from the. U.S. 2007-09 Survey of Consumer Finances Panel. Federal Research Board Working Paper 2013-53. Case, K. E., Shiller, R. J., & Thompson, A. (2012). What Have They Been Thinking? Home Buyer Behavior in Hot and Cold Markets, Cowles Foundation Discussion Paper, No. 1876 September 2012. Chan, S. (2001). Spatial Lock-in: Do Falling House Prices Constrain Residential Mobility? Journal of Urban Economics, 49(3), 567–586. Choi, J.H. (2014). Measurement, Perceptions and Negative Equity-Mobility Relationship, Working Paper Choi, J.H., Green, R.K. (2014) Human Capital Spillovers and Local Unemployment, Working Paper Coulson, N. E., & Grieco, P. (2013). Mobility and Mortgages: Evidence from the PSID. Regional Science and Urban Economics, 43, 1–7. Demyanyk, Y., Hryshko, D., José Luengo-Prado, M., & Sørensen, B. E. (2013). Moving to a Job: The Role of Home Equity, Debt, and Access to Credit. Federal Reserve Bank of Cleveland, Working Paper Series 13-05. 118 Donovan, C., & Schnure, C. (2011). Locked in the House: Do Underwater Mortgages Reduces Labor Market Mobility. Working Paper. Elul, R., Souleles, N. S., Chomsisengphet, S., Glennon, D., & Hunt, R. (2010). What ―Trigge rs‖ Mortgage Default? Federal Reserve Bank of Philadelphia Working Paper, No. 10-13. Engelhardt, G. (2003). Nominal Loss Aversion, Housing Equity Constraints, and Household Mobility: Evidence from the United States. Journal of Urban Economics, 53, 171–195. Farber, H., S. (2012). Unemployment in the Great Recession: Did the Housing Market Crisis Prevent the Unemployed from Moving to Take Jobs. American Economic Review, 102(3), 520–525. Ferreira, F., Gyourko, J., & Tracy, J. (2010). Housing Busts and Household Mobility. Journal of Urban Economics, 68, 34–45. Ferreira, F., Gyourko, J., & Tracy, J. (2011). Housing Bust and Household Mobility: An Update. NBER Working Paper 17405. Follain, J. R., & Malpezzi, S. (1981). Are Occupants Accurate Appraisers? Review of Public Data Use, 9, 47–55. Foote, C., Gerardi, K., & Willen, P. (2008). Negative Equity and Foreclosure: Theory and Evidence. Journal of Urban Economics, 64, 234–245. Genesove, D., & Mayer, C. (2001). Loss Aversion and Seller Behavior: Evidence from the Housing Market. Quarterly Journal of Economics, 106(4), 1233–1260. Goodman, J. L., & Ittner, J. B. (1992). The Accuracy of Home Owners‘ Estimates of House Value. Journal of Housing Economics, 4, 339–357. Greenspan, A., & Kennedy, J. (2008). Sources and Uses of Equity Extracted from Homes. Oxford Review of Economic Policy, 24(1), 120–144. 119 Guiso, L., Sapienza, P., & Zingales, L. (2011). The Determinants of Attitudes towards Strategic Default on Mortgages. Han, L. (2013). Understanding the Puzzling Risk-Return Relationship for Housing. Review of Financial Studies, 26, 877–928. Himmelberg, C., Mayer, C., & Siani, T. (2005). Assessing High House Prices: Bubbles, Fundamentals and Misperceptions. Journal of Economic Perspectives, 19, 67–92. Kahneman, D., & Tversky, A. (1979). Prospect Theory: An Analysis of Decision under Risk. Econometrica, 47(2), 263–292. Kain, J. F., & Quigley, J. M. (1972). Note on Owners Estimate of Housing Value. Journal of the American Statistical Association, 67, 803–806. Kau, J. B., Keenan, D., & Kim, T. (1993). Transaction Costs, Suboptimal Termination and Default Probabilities. Real Estate Economics, 21(3), 247–263. Mayer, C., Pence, K., & Sherlund, S. (2009). The Rise in Mortgage Defaults. Journal of Economic Perspectives, 23(1), 27–50. Mollay, R., & Shan, H. (2012). The Post-Foreclosure Experience of U.S. Households. Real Estate Economics, Forthcoming. Mollay, R., Smith, C. L., & Wozniak, A. (2011). Internal Migration in the United States. Journal of Economic Perspectives, 25(3), 173–196. Painter, G., & Lee, K. O. (2013). What happens to household formation in a recession? Journal of Urban Economics, 76, 93–109. Piazzesi, M., & Schneider, M. (2009). Momentum Traders in the Housing Market: Survey Evidence and a Search Model. American Economic Review: Papers & Proceedings, 99(2), 406–411. 120 Quigley, J. M. (1987). Interest Rate Variations, Mortgage Prepayments and Household Mobility. Review of Economics and Statistics, 69, 636–643. Riddiough, TJ. (1991). Equilibrium Mortgage Default Pricing with Non-optimal Borrower Behavior. Dissertation Paper, University of Wisconsin. Schulhofer-Wohl, S. (2011). Negative Equity Does Not Reduce Homeowners‘ Mobility. NBER Working Paper 16701. Schwartz, E., S., & Torous, W., N. (2003). Mortgage Prepayment and Default: A Poisson Regression Approach. Real Estate Economics, 21(4), 431–449. Siani, T., & Souleles, N. S. (2005). Owner-Occupied Housing as a Hedge Against Rent Risk. Quarterly Journal of Economics, 120, 763–789. Stein, J. (1995). Prices and Trading Volume in the Housing Market; a Model with Down- Payment Effects. Quarterly Journal of Economics, 110, 379–406. Valletta, R., G. (2012). House Lock and Structural Unemployment. Federal Reserve Bank of San Francisco Working Paper Series, Working Paper Series 2012-25. White, B. T. (2010). Underwater and Not Walking Away: Shame, Fear and the Social Management of the Housing Crisis. Wake Forest Law Review, 45, 971–1023 Chapter III. Agarwal, S. (2007). The Impact of Homeowners‘ Housing Wealth Misestimation on Consumption and Saving Decisions. Real Estate Economics, 35(2), 135–154. Bení tez-Silva, H., Eren, S., Heiland, F., & Jiménez-Martí n, S. (2010). How Well do Individuals Predict the Selling Prices of their Homes? Working Paper. Bhutta, N., Bokko, J., & Shan, H. (2010). The Depth of Negative Equity and Mortgage Default 121 Decisions. Working Paper. Bucks, B., & Bricker, J. (2013). Household Mobility over the Great Recession: Evidence from the. U.S. 2007-09 Survey of Consumer Finances Panel. Federal Research Board Working Paper 2013-53. Bucks, B.K., & Pence, K. (2006). Do Homeowners Know Their House Values and Mortgage Terms? Working Paper, Federal Reserve Board of Governors, 2006. Choi, J. H. (2014). Housing Market Shock and Mobility of Underwater Households: Why do Underwater Homeowners Move? Working Paper. Coulson, N., & Grieco, P. (2013). Mobility and Motgages: Evidence from the PSID. Regional Science and Urban Economics, 43, 1–7. Demyanyk, Y., Hryshko, D., José Luengo-Prado, M., & Sørensen, B. E. (2013). Moving to a Job: The Role of Home Equity, Debt, and Access to Credit. Federal Reserve Bank of Cleveland, Working Paper Series 13-05. Donovan, C., & Schnure, C. (2011). Locked in the House: Do Underwater Mortgages Reduces Labor Market Mobility. Working Paper. Economist. (2010 May). American Joblessness: Structural Unemployment. Retrieved from http://www.economist.com/blogs/freeexchange/2010/05/american_joblessness_0 Economist. (2010 Oct). Structural Unemployment: Can‘t Move, Can‘t Move Up. Retriev ed from http://www.economist.com/blogs/freeexchange/2010/10/structural_unemployment Engelhardt, G. (2003). Nominal Loss Aversion, Housing Equity Constraints, and Household Mobility: Evidence from the United States. Journal of Urban Economics, 53, 171–195. Farber, H.S. (2012). Unemployment in the Great Recession: Did the Housing Market Crisis Prevent the Unemployed from Moving to Take Jobs. American Economic Review, 102(2), 122 520–525. Ferreira, F., Gyourko, J., & Tracy, J. (2010). Housing Busts and Household Mobility. Journal of Urban Economics, 68, 34–45. Follain, J. R., & Malpezzi, S. (1981). Are Occupants Accurate Appraisers? Review of Public Data Use, 9, 47–55. Genesove, D., & Mayer, C. (2001). Loss Aversion and Seller Behavior: Evidence from the Housing Market. Quarterly Journal of Economics, 106(4), 1233–1260. Goodman, J. L., & Ittner, J. B. (1992). The Accuracy of Home Owners‘ Estimates of House Value. Journal of Housing Economics, 4, 339–357. Ihlanfeldt, K.R., & Martinez-Vazquez, J. (1986). Alternative Value Estimates of Owner- Occupied Housing: Evidence on Sample Selection Bias and Systematic Errors. Journal of Urban Economics, 20(3), 356–369. Kahneman, D., & Tversky, A. (1979). Prospect Theory: An Analysis of Decision under Risk. Econometrica, 47(2), 263–292. Kain, J. F., & Quigley, J. M. (1972). Note on Owners Estimate of Housing Value. Journal of the American Statistical Association, 67, 803–806. Kiel, K.A., & Zabel, J.E. (1999). The Accuracy of Owner-Provided House Values: The 1978- 1991 American Housing Survey. Real Estate Economics, 27(2), 263–298. Kish, L., & Lansing, J. B. (1954). Response Errors in Estimating the Value of Homes. Journal of the American Statistical Association, 49, 520–538. Kuzmenko, T., & Timmins, C. (2011). Persistence in Housing Wealth Perceptions: Evidence from the Census Data. Working Paper. Durham: Duke University. Modestino, A. S., & Dennett, J. (2013). ―Are American Homeowners Locked Into Their Houses? 123 The Impact of Housing Market Conditions on State‐to‐state Migration. Regional Science and Urban Economics, 43(2), 322‐337. Mollay, R., Smith, C.L., & Wozniak, A. (2011). Internal Migration in the United States. Journal of Economic Perspectives, 25(3), 173–196. Quigley, J. M. (1987). Interest Rate Variations, Mortgage Prepayments and Household Mobility. Review of Economics and Statistics, 69, 636–643. Stein, J. (1995). Prices and Trading Volume in the Housing Market; a Model with Down- Payment Effects. Quarterly Journal of Economics, 110, 379–406. 124 Appendix Chapter II. [Table II-A1] Summary Statistics of Underwater Homeowners B & A 2007 (1) (2) Variables LTV <2007 LTV≥2007 T-test Mean Std. Dev. Mean Std. Dev. Age 40.267 0.605 42.236 0.387 *** Female Head 0.242 0.021 0.214 0.014 No. of Family 3.277 0.080 3.192 0.050 Married 0.688 0.023 0.688 0.016 Single 0.100 0.015 0.111 0.011 Widowed 0.050 0.011 0.020 0.005 *** Divorced/Separated 0.117 0.016 0.139 0.012 Black 0.349 0.024 0.265 0.015 *** Hispanic 0.050 0.011 0.169 0.013 *** Refinance 0.312 0.023 0.200 0.014 *** Family Income 56947 1945 86940 2473 *** Unemployed 0.057 0.012 0.066 0.009 Less than High School 0.183 0.020 0.110 0.011 *** High School 0.427 0.026 0.337 0.017 *** Some College 0.236 0.022 0.268 0.016 College 0.154 0.019 0.285 0.016 *** Years in House 5.124 0.341 6.367 0.211 *** Observations 401 858 Note: *, ** and *** indicate that the mean of the two values are statistically difference at 10%, 5% and 1% respectively. 125 [Table II-A2] Difference in Differences: In-State & Out-State Moves: Controls In_State Out_State VARIABLES (1) (2) LTV>100% -0.0225 -1.541** (0.233) (0.634) Year≥2007 -0.040 -0.076 (0.057) (0.112) LTV>100%*Year≥2007 0.383 2.527*** (0.291) (0.701) Age 0.868*** 0.887*** (0.011) (0.021) Age2 1.001*** 1.001*** (0.000) (0.000) Female head 0.796* 0.806 (0.095) (0.210) # of Family 0.990 0.954 (0.027) (0.053) Increase in # of Children 1.236** 1.403** (0.110) (0.235) Increase in # of Children 1.076 1.034 (0.108) (0.193) Single 1.102 0.844 (0.149) (0.242) Widowed 1.399* 1.221 (0.249) (0.498) Divorced/Separated 1.669*** 1.155 (0.192) (0.261) Black 0.900 0.652 (0.113) (0.197) Hispanic 1.035 0.986 (0.146) (0.288) Log(family income) 0.961 1.068 (0.043) (0.092) Unemployed 1.344* 1.275 (0.208) (0.424) Refinance 0.907 0.936 (0.059) (0.119) High school 1.012 0.958 (0.107) (0.268) Some college 0.960 1.505 (0.109) (0.427) 126 [Table II-A2] (Continued) In_State Out_State VARIABLES (1) (2) College 1.004 2.059** (0.115) (0.586) Years in house 0.974*** 0.964*** (0.004) (0.009) Constant 15.53*** 0.461 (9.487) (0.507) State FE Yes Yes Observations 22,005 22,005 Pseudo R-Sq. 0.068 0.068 Note: Robust seeform in the parenthesis * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. [Table II-A3] Likelihood of Trading Up, Down and Switching to Rental Housing Trade Up Trade Down Own to Rent VARIABLES (1) (2) (3) LTV≤80% 1.001 1.092 0.783 (0.129) (0.192) (0.125) LTV100-120% 0.706 0.538 1.295 (0.293) (0.347) (0.650) LTV>120% 0.961 1.289 0.544 (0.423) (0.881) (0.344) Year≥2007 0.806 0.679 1.250 (0.135) (0.163) (0.226) LTV≤80%*Year≥2007 1.122 0.867 1.056 (0.208) (0.226) (0.221) LTV100-120%*Year≥2007 1.428 3.437 1.120 (0.834) (2.694) (0.673) LTV>120%**Year≥2007 1.748 0.344 5.390** (1.150) (0.349) (3.787) Age 0.889*** 0.944** 0.815*** (0.016) (0.022) (0.016) Age2 1.001*** 1.000** 1.002*** (0.000) (0.000) (0.000) Female head 0.820 0.845 0.791 (0.153) (0.189) (0.130) # of family 0.945* 0.800*** 1.151*** (0.032) (0.040) (0.048) Increase in # of children 1.464*** 1.137 1.079 (0.156) (0.211) (0.154) Decrease in # of children 0.843 1.228 1.358* (0.121) (0.192) (0.213) 127 [Table II-A3] (Continued) Trade Up Trade Down Own to Rent VARIABLES (1) (2) (3) Single 0.762 0.589** 1.705*** (0.149) (0.159) (0.341) Widowed 0.699 1.013 1.710** (0.224) (0.331) (0.443) Divorced/Separated 1.040 1.189 2.683*** (0.173) (0.239) (0.438) Black 0.595** 0.569** 1.436** (0.122) (0.150) (0.246) Hispanic 1.027 1.643** 0.801 (0.183) (0.411) (0.185) Log(family income) 1.235*** 1.140* 0.784*** (0.087) (0.089) (0.042) Unemployed 0.611 1.115 1.967*** (0.204) (0.325) (0.404) Refinance 0.886 0.931 0.839 (0.075) (0.101) (0.094) High school 0.879 1.122 1.080 (0.145) (0.227) (0.175) Some college 1.015 1.046 1.080 (0.174) (0.224) (0.190) College 1.261 1.131 1.039 (0.213) (0.249) (0.186) Years in house 0.985** 0.972*** 0.962*** (0.006) (0.007) (0.006) Constant 0.437 0.173* 76.97*** (0.365) (0.180) (68.97) State FE Yes Yes Yes Observations 20,968 20,968 20,968 Pseudo R-Sq. 0.076 0.076 0.076 Note: Robust seeform in the parenthesis * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. 128 [Table II-A4] House Price Expectation: MSA HPI Move In_MSA Out_MSA VARIABLES (1) (2) (3) (4) (5) (6) LTV>100% 1.517*** 1.511** 1.316 1.308 2.161** 2.258** (0.243) (0.243) (0.228) (0.228) (0.690) (0.731) MSA HPI 1.627** 1.560 1.537* 1.385 2.267** 2.178 (0.313) (0.436) (0.356) (0.455) (0.844) (1.229) MSA HPI*LTV>100% 0.207* 0.213 0.120* 0.128* 0.324 0.254 (0.198) (0.201) (0.142) (0.151) (0.493) (0.378) State FE No Yes No Yes No Yes Year FE No Yes No Yes No Yes Observations 14,469 14,469 14,332 14,332 14,332 14,332 Pseudo R-Sq. 0.054 0.056 0.052 0.054 0.052 0.054 Note: Robust seeform parenthesis * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. 129 Chapter III. [Table III-A1] No. of Underwater Homeowners: Rep. LTV vs. Est. LTV (ZIP) LTV Category (%) (1) (2) (3) (4) (4) Reported LTV Estimated LTV(ZIP) # Matched % Matched (3)/(1) % Matched (3)/(2) 0<LTV≤20 331 313 274 82.78% 87.54% 20<LTV≤40 833 811 609 73.11% 75.09% 40<LTV≤60 1,455 1,611 1,052 72.30% 65.30% 60<LTV≤80 2,750 2,399 1,907 69.35% 79.49% 80<LTV≤100 2,628 2,542 2,070 78.77% 81.43% 100<LTV≤120 390 537 222 56.92% 41.34% 120<LTV≤140 126 213 48 38.10% 22.54% 140<LTV 139 226 89 64.03% 39.38% Total 8,652 8,652 6,271 72.48% 72.48% [Table III-A2] No. of Underwater Homeowners: Est. LTV (MSA) vs. Est. LTV (ZIP) LTV Category (1) (2) (3) (4) (4) Estimated LTV(MSA) Estimated LTV(ZIP) # Matched % Matched (3)/(1) % Matched (3)/(2) 0<LTV≤20 193 207 179 92.75% 86.47% 20<LTV≤40 554 584 498 89.89% 85.27% 40<LTV≤60 1,237 1,235 1,077 87.07% 87.21% 60<LTV≤80 1,948 1,919 1,684 86.45% 87.75% 80<LTV≤100 2,164 2,158 1,933 89.33% 89.57% 100<LTV≤120 464 427 309 66.59% 72.37% 120<LTV≤140 160 174 93 58.13% 53.45% 140<LTV 165 181 131 79.39% 72.38% Total 6,885 6,885 5680 82.50% 82.50% 130 [Table III-A3] Likelihood of being in the Same LTV Category (ZIP) VARIABLES (1) (2) (3) Est. LTV (ZIP)>100% -1.854*** -1.877*** -1.869*** (0.096) (0.114) (0.121) Age -0.025 -0.033 (0.025) (0.024) Age2 0.000 0.000 (0.000) (0.000) Black -0.035 -0.029 (0.144) (0.153) Hispanic 0.026 0.158 (0.150) (0.160) High School -0.349* -0.383* (0.208) (0.207) Some College -0.334 -0.327 (0.213) (0.215) College -0.180 -0.263 (0.215) (0.214) Single 0.009 0.050 (0.141) (0.134) Widowed 0.488 0.364 (0.432) (0.381) Divorced/Separated -0.260* -0.262** (0.144) (0.132) Log(House Value) 0.090 0.218** (0.075) (0.090) Log (Family Income) -0.065 -0.074 (0.074) (0.075) Unemployed -0.346 -0.310 (0.221) (0.220) Years in House -0.209*** -0.204*** (0.014) (0.015) Constant 1.189*** 2.692*** 2.424** (0.037) (0.895) (1.060) Year FE N N Y State FE N N Y Observations 8,580 6,644 6,643 Pseudo R2 0.060 0.126 0.148 Note: Robust standard errors in parentheses * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id. 131 [Table III-A4] Difference between the Rep. & Est. House Value (ZIP) VARIABLES (1) (2) (3) Est. LTV>100% 0.249*** (0.034) Est. LTV≤20% -0.166*** (0.026) 20%<Est. LTV≤40% -0.127*** (0.014) 40%<Est. LTV≤60% -0.086*** (0.009) 60%<Est. LTV≤80% -0.038*** (0.006) 100%<Est. LTV<120% 0.059*** (0.013) 120%<Est. LTV<140% 0.200*** (0.032) 140%<Est. LTV 0.583*** (0.091) Rep. LTV >100% Est. LTV≤100% -0.231*** (0.035) Rep. LTV ≤100% Est. LTV>100% 0.448*** (0.049) Rep. LTV >100% Est. LTV>100% 0.029 (0.018) Year FE Y Y Y State FE Y Y Y Observations 6,614 6,614 6,621 R-squared 0.166 0.277 0.276 Note: Robust standard errors in parentheses * indicates P < 0.10, **: P < 0.05 and ***: P < 0.01. All regressions are weighted by sample weights. Standard errors are clustered by id.

Abstract (if available)

Abstract The onset of the housing market crisis left a considerable proportion of U.S. households underwater. According to Corelogic, 11.4 million U.S. households had negative home equity as of 2012:Q1, accounting for 23.7% of all residential properties with a mortgage. This increase has gained particular interest from policy makers who became concerned that underwater homeowners may have negative spillover effects on other parts of the economy. Responding to this increased, my dissertation features three independent essays related to the topic of “negative equity"". Using the Panel Study of Income Dynamics and the American Housing Survey, I examine who underwater homeowners are, how and why they move, and the impact of their mobility decisions on the labor market. I also compare the two most commonly used proxies of negative equity and examine whether the empirical outcomes are sensitive to how the underwater homeowners are defined. ❧ The first essay investigates who the underwater homeowners are with a special focus on race, ethnicity and immigrant status. So far, not many studies have closely looked into this question and have examined whether the difference in the likelihood of being underwater can be fully explained by observable differences across households with different race and immigrant status. In Chapter II, I first lay out the possible reasons why minorities or immigrants may more likely to be underwater. Not only do the differences across households matter, but the differences in the racial composition across neighborhoods may also have an impact on the probability of household’s becoming underwater. For example, neighborhoods with greater share of minorities may have experienced a greater boom and the bust as the relaxation of credit during the early 2000s increased the access to homeownership for many people living in these neighborhoods. If so, households dwelling in black or Hispanic neighborhoods may have higher likelihood of being underwater. ❧ The results show that prior to the crisis, black and Hispanic households are more likely to be underwater compared to whites. On the other hand, the likelihood of being underwater for Asians is similar to whites when controlling for the demographic and socioeconomic characteristics. Immigrants also have higher likelihood of being underwater relative to the native born. These results hold even after controlling for the time period of purchase as well as the percentage of initial down payment. Before 2007, however, the results show that minorities and immigrants were less likely to be underwater in comparison to their reference groups, although during this period the percent of underwater homeowners was significantly lower than the post-crisis period. ❧ Furthermore, even after controlling for the head’s race, I find that households living in neighborhoods with greater proportion of blacks and Hispanics were more likely to be underwater in the post‐crisis period. In addition, non-Hispanics in Hispanic neighborhoods have higher probability of becoming underwater than Hispanics in the same neighborhood. The findings of the study, suggest that minorities and immigrants have been more adversely affected by the recent housing market collapse which cannot be fully explained by their observable characteristics. ❧ The second chapter investigates the causal relationship between negative equity and residential mobility. This question gained particular interest as several media outlets highlighted that negative equity may hamper homeowners from moving to a better job market and as a consequent increase the structural unemployment rate. The recent surge in the unemployment rate has provoked greater concerns about whether negative equity would further delay the recovery of the sluggish labor market. Numerous studies have looked at this issue since the outbreak of the crisis, but most did not find evidence that negative equity reduces mobility. ❧ The existing studies, however, have three major limitations. First, they do not properly control for the possible existence of endogeneity that occurs from the households’ unobserved characteristics which affect both their likelihood of moving and their likelihood of being underwater. Second, they do not examine theoretical reasons behind the negative equity-mobility relationship. Finally, they have not identified how the mobility of underwater homeowners actually affects the labor market by looking at where these homeowners move to. ❧ I address each of these three issues in chapter II. First, I use the difference in differences framework to examine whether the likelihood of moving changes for households that exogeneously became underwater due to the housing market shock. The results show that the mobility of those who became underwater due to the unexpected house price collapse increased significantly compared to those with positive equity. I also find that the increase in the mobility of underwater homeowners is more prominent for the out-state moves. Second, among the existing theories that provide competing reasons for the negative equity‐mobility relationship, I find the double trigger and the house price expectation theories best fit the result of the findings. Those who are significantly underwater are more likely to move if they are at the same time unemployed, in accordance with the double trigger theory. Also, as the house price expectation theory suggests, underwater homeowners are more likely to move when they expect their house prices to go down, to avoid further losses. When they expect prices to go up, they are more likely to stay at their current housing, to wait until house prices exceed their mortgage debts. The linkage between negative equity and labor market efficiency, however, still remains uncertain as the findings show that since 2007, underwater homeowners have not moved to states with lower unemployment rates. Thus, how the increase of mobility of those in negative equity influences the macro labor market still needs further exploration. ❧ The final chapter looks at the measurement of negative equity. In order to identify those who are underwater, we need the exact value of the house price. However, it is difficult to accurately measure house prices, as the housing market is thin due to infrequent transactions and the heterogeneity of housing units. This study compares the two most commonly used proxies of negative equity: one calculated using the self-reported house value and the other calculated using the house price indices. The first part of the study looks at differences in the two house values and examines whether the estimated value of home equity affects how households report their house values. Next, I investigate whether using different measures of negative equity affect its association with three of the following household behaviors: 1) moving, 2) moving to rental units, and 3) falling behind their mortgage payment. ❧ The results show that many households who are estimated to be underwater using the house price indices, report their house value higher than their mortgage debt, and thus do not classify themselves as underwater homeowners. Also, the gap between the reported and the estimated house price is the greatest for those who are estimated to have negative equity, but report that they have positive equity. The findings can be explained by the loss aversion theory which suggests that people’s disutility from losses is greater than the utility from the same amount of gains. ❧ Next results show that those who are estimated have in negative equity but do not report so have similar likelihood of moving compared to those who are both estimated and reported to have positive equity. On the other hand, those who admit that they are underwater have significantly higher likelihood of moving than those who do not. This again accords with the loss aversion theory, suggesting that those who did not report themselves as underwater homeowners may not be moving to avoid realization of losses. ❧ To identify whether the households are correctly reporting their house prices, I further examine households’ likelihood of switching to rental units and their likelihood of being delinquent. Conditional on moving, I find those estimated to be underwater are more likely to move to rental units compared to those whose reported and estimated home equity are both positive. Even those who do not report that they are underwater have significantly higher likelihood of moving to rental units than those estimated to have positive equity. Those estimated to be underwater are also more likely to be behind their mortgage payment, regardless of how they report their home equity value. These results suggest that many of the homeowners who do not report themselves as underwater homeowners are facing greater financial stress due to the drop in their home equity level, and some of them are actually aware of their situation.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

A better method for measuring housing affordability and the role that affordability played in the mobility outcomes of Latino-immigrants following the Great Recession

PDF

Insights into residential mobility and pricing of rental housing: the role of gentrification, home-ownership barriers, and market concentrations in low-income household welfare

PDF

Reshaping Los Angeles: housing affordability and neighborhood change

PDF

Three essays on aging, wealth, and housing tenure transitions

PDF

The impact of mobility and government rental subsidies on the welfare of households and affordability of markets

PDF

The role of the timing of school changes and school quality in the impact of student mobility: evidence from Clark County, Nevada

PDF

Three essays on housing demographics: depressed housing access amid crisis of housing shortage

PDF

Risks, returns, and regulations in real estate markets

PDF

Environmental justice in real estate, public services, and policy

PDF

Competing across and within platforms: antecedents and consequences of market entries by mobile app developers

PDF

Emissions markets, power markets and market power: a study of the interactions between contemporary emissions markets and deregulated electricity markets

PDF

Essays on congestion, agglomeration, and urban spatial structure

PDF

Talent migration: does urban density matter?

PDF

The history of autonomous vehicle development and its likely futures and consequences

PDF

Household mobility and neighborhood impacts

PDF

Curating gastronomy: restaurants and social media in the cultural economy

PDF

The economic and political impacts of U.S. federal carbon emissions trading policy across households, sectors and states

PDF

Active travel, outdoor leisure, and neighborhood environment: path analysis, Los Angeles County

PDF

Essays on the empirics of risk and time preferences in Indonesia

PDF

Essays on the economics of cities

Asset Metadata

Creator Choi, Jung Hyun (author)

Core Title Housing market crisis and underwater homeowners

School School of Policy, Planning and Development

Degree Doctor of Policy, Planning & Development

Degree Program Policy, Planning, and Development

Publication Date 06/24/2015

Defense Date 03/04/2015

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag housing market crisis,mobility,OAI-PMH Harvest,underwater homeowner,Unemployment

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Painter, Gary Dean (committee chair), Green, Richard Kent (committee member), Matsusaka, John G. (committee member)

Creator Email choijung@usc.edu,lookatjesus@paran.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-583084

Unique identifier UC11301247

Identifier etd-ChoiJungHy-3513.pdf (filename),usctheses-c3-583084 (legacy record id)

Legacy Identifier etd-ChoiJungHy-3513.pdf

Dmrecord 583084

Document Type Dissertation

Format application/pdf (imt)

Rights Choi, Jung Hyun

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA