Fill and Sign the Optometry Intake Form

OFFICE OF POLICY DEVELOPMENT & RESEARCH Analysis of Housing Finance Issues Using the American Housing Survey (AHS) Analysis of Housing Finance Issues Using the American Housing Survey (AHS) Prepared for: U.S. Department of Housing and Urban Development, PD&R Prepared by: Ken Lam Bulbul Kaul Abt Associates Inc. Cambridge, MA April 2003 Acknowledgements This study has been made possible with the help of numerous persons and organizations. Eric S. Belsky of the Joint Center for Housing Studies at Harvard University served as a consultant to this study. He has contributed to both the Research Design and Final Report phases of the study. In addition to the report authors, several staff members from Abt Associates were instrumental in the study design, data collection, analysis, and final report production. In particular, Chris Herbert, Meryl Finkel and Jill Khadduri, as Technical Reviewers, have provided thoughtful and constructive comments on early versions of the report. Deborah Gruenstein and Elizabeth Burns assisted with the design phase of the study. Jessica Bonjorni and Missy Robinson provided production support throughout the project. Various staff members from the Census Bureau and HUD’s Office of Policy Development & Research, including Bill Reeder and Harold Bunce, have provided valuable feedback on early versions of this report. The report benefited greatly from their insights. We are especially grateful to the Government Technical Monitor, Sue George Neal, for her guidance, assistance, and support throughout the study. The contents of this report are the views of the contractor and do not necessarily reflect the views or policies of the U.S. Department of Housing and Urban Development of the U.S. Government. Table of Contents Executive Summary ............................................................................................................................. v Overview ......................................................................................................................................v Replication Analysis................................................................................................................... vi Internal Consistency of AHS Mortgage Variables Across Surveys .......................................... vii Reliable AHS Mortgage Variables ............................................................................................ vii Unreliable or Uncertain AHS Mortgage Variables .................................................................... ix Areas Requiring Further Investigation ....................................................................................... xi Recommendations for Modifications to the Survey ................................................................... xi Chapter One - Introduction ................................................................................................................ 1 Background and Policy Context .................................................................................................. 1 Outline of the Report ................................................................................................................... 2 Chapter Two - The AHS and Research on Mortgages ..................................................................... 4 History and Basic Characteristics of the AHS............................................................................. 4 Mortgage-Related Variables in the AHS..................................................................................... 5 Geographic Coverage of the Survey.......................................................................................... 10 Significant Changes to the AHS Over Time ............................................................................. 11 Advantages and Disadvantages of the AHS for Mortgage-Related Research........................... 14 AHS Mortgage Variables Chosen for the Replication Analysis ............................................... 16 Chapter Three - Comparison Databases ......................................................................................... 19 Principal Sources of Comparison Databases ............................................................................. 19 Other Comparison Databases Considered ................................................................................. 22 Choice of Comparison Databases for the Replication Analysis ................................................ 23 Chapter Four - Replication Analysis................................................................................................ 26 Goals and Approach for the Replication Analysis .................................................................... 26 Mortgage Origination Volume .................................................................................................. 32 Loan Characteristics .................................................................................................................. 45 Borrower Characteristics ........................................................................................................... 68 Summary of Major Findings on AHS Mortgage Variables Tested in the Replication Analysis82 Chapter Five - Internal Consistency of AHS Mortgage Variables Across Surveys..................... 88 Goals and Approach .................................................................................................................. 88 Consistency of Mortgage Variables Across Time ..................................................................... 91 Summary of Major Findings ..................................................................................................... 99 Chapter Six - Conclusions and Recommendations ....................................................................... 101 Reliable AHS Mortgage Variables .......................................................................................... 101 Unreliable or Uncertain AHS Mortgage Variables ................................................................. 103 Potential Use of the AHS for Research on Mortgages ............................................................ 105 iii Areas Requiring Further Investigation .................................................................................... 110 Recommendations for Modifications to the Survey ................................................................ 111 Appendix A - GSE Conforming Loan Limits, 1980-2000............................................................. A-1 Appendix B - Summary Information of Comparison Databases................................................. B-1 Appendix C - Alternative Approaches to Adjusting the Mortgage Origination Volume Estimates in AHS ............................................................................................................................. C-1 Appendix D - Tabulations Supporting the Analysis in Chapter Four ........................................ D-1 Appendix E - Procedures for Tracking Mortgage Records Across AHS Survey Years............ E-1 Appendix F - Tabulations Supporting the Analysis in Chapter Five ...........................................F-1 Appendix G - Mortgage and Purchase Modules in the 1997 AHS Questionnaire ..................... G-1 Appendix H - Derivation of AHS Mortgage Market Attributes Tested in the Study………….H-1 iv Executive Summary Overview The American Housing Survey (AHS, formerly Annual Housing Survey) is the most comprehensive source of information about the characteristics and condition of the nation’s housing stock. Started in 1973, the AHS national sample data were collected by the U.S. Census Bureau for the Department of Housing and Urban Development (HUD) on a nationally representative sample of housing units every year until 1981, and they have been collected every other year since then. Over the years, AHS data have been used extensively by researchers and policy analysts to answer questions about the nation’s housing conditions and occupant characteristics. The longitudinal nature of the AHS also permits the analysis of dynamic changes in housing and occupancy characteristics of the housing stock. The AHS data contain detailed questions about mortgages asked of respondents for owner-occupied units in the survey. The questions cover most basic mortgage and housing finance topics. This wealth of mortgage-related variables, combined with the occupant demographic and property location information, could be a very powerful resource for answering many housing finance research and policy questions. These micro-data have the potential to provide crucial information to support analysis of issues of interest to policy makers and the mortgage industry. The principal advantage of using the AHS for mortgage market and housing finance analysis is its detailed household, housing, loan, and geographic characteristics. In addition to use in detailed cross tabulations, these variables can also be used as micro data to conduct multiple regressions on the cross-sectional files and other loan-level statistical analyses on the longitudinal panels. However, neither the housing research community nor HUD staff has made as much use of the mortgage variables of the AHS data as might be expected. Among the reasons for this under­ utilization is the fact that the reliability of these mortgage-related variables in the AHS has not been verified. Analysis is needed to establish the extent to which limitations associated with sample size, survey design, and interview response affect the accuracy and consistency of the mortgage data in the AHS. The research presented in this study is intended to meet this need. The goals of the analysis are to determine: 1) what types of mortgage market analysis can be supported by the AHS; 2) what areas of the AHS are problematic for mortgage research; and 3) what analysis techniques or changes in the survey could potentially compensate for the problems. The analysis of the reliability of the AHS is composed of two broad categories. First, to test the reliability of the AHS variables we replicate measures of mortgage activity from other reliable sources of data that serve as benchmarks for the AHS estimates. This analysis is referred to as the “replication analysis.” Second, we use the longitudinal nature of the AHS to determine whether answers to questions on mortgages are consistent and stable over time. Findings from each of these analyses are presented in turn below. The study concludes with recommendations for the topics for which the AHS can reliably be used, subjects that are problematic given the nature of the AHS data, areas where further investigation is needed to explore the potential usefulness of the AHS, and options for improving the quality of the mortgage-related variables in the survey instrument. v Replication Analysis The principal benchmark data sets used in the replication analysis are the FHA administrative data, data collected under the Home Mortgage Disclose Act (HMDA), the Mortgage Interest Rate Survey (MIRS), the Residential Finance Survey (RFS), the Survey of Consumer Finance (SCF) and a homebuyer survey conducted by Chicago Title & Trust Company. Our benchmarking focuses on the AHS national sample data collected in the 1990s for all primary mortgages of owner-occupied single-family units, defined as structures with one to four units. Since many of the benchmark data sets provide information on new mortgage originations, much of the replication analysis focuses on estimates of the volume and characteristics of newly originated mortgages. Two characteristics of the AHS impose significant constraints on how this analysis can be conducted. First, since the AHS only gathers information on the year in which a mortgage is originated and not the month, analysis of originations can only be done on a calendar year basis. Second, since the AHS is conducted over the second half of the calendar year, it does not provide complete information on the year in which the survey is conducted. Given these two constraints, new originations are defined as loans made in the two complete calendar years prior to the year in which the AHS is conducted. For example, the 1997 survey data are used to study the characteristic and volume of loans originated in 1995 and 1996. An important concern about this approach is that by the time the AHS survey is conducted, some share of the new originations will have terminated and thus no longer be reported in the AHS. To account for the number of loans that have terminated prior to the interview in each cohort, we have devised a method to adjust all the AHS volume estimates based on the historical termination rates derived from an independent source. Among the variables we have evaluated are mortgage origination volume, mortgage market segment, original loan amount, mortgage interest rate of fixed-rate loans, loan-to-value (LTV) ratio at origination, mortgage payment product type, first-time homebuyer status, borrower race/ethnicity, and income. The selection was based on the importance of these variables in mortgage market research and the availability of reliable benchmark measures. To the extent possible, comparisons to the benchmark estimates are stratified by loan purpose (home purchase vs. refinance) and mortgage market segment (FHA, VA, FmHA/RHS, conventional conforming, and conventional jumbo). Major findings of the replication analysis are summarized in Exhibit 0-1 at the end of the Executive Summary. One of the findings of this study is the importance of recognizing the sampling variability of estimates obtained from the AHS. We found many cases where the small sample sizes of particular segments of the mortgage market examined resulted in very wide confidence intervals surrounding individual point estimates. Researchers using the AHS for mortgage market analysis should take sampling variability into account by reporting confidence intervals in addition to the AHS point estimates. When the data are used to support loan-level econometric analysis, we recommend that statistical software packages such as STATA® and SUDAAN® be employed to address fully the sampling errors in such analysis. vi Internal Consistency of AHS Mortgage Variables Across Surveys One of the advantages of the AHS is that it not only provides information on newly originated mortgages but also allows researchers to examine the characteristics of the entire stock of outstanding mortgages and changes in the use of mortgage finance over an owner's tenure in a unit. However, given that the AHS asks owners to report on mortgage characteristics each time the survey is conducted, there is a concern about whether the accuracy of owner's responses might degrade as the length of time since the mortgage was originated increases. In order for analysis of the characteristics of the entire stock of mortgages to be valid, homeowners residing in the sample housing units must report consistent information about their mortgages across survey years. The analysis of the internal consistency of the AHS responses is meant to investigate this concern. The variables we have examined in this analysis are mortgage market segment, loan size, mortgage payment product type, loan term, interest rate, monthly payment amount, and first-time buyer status. We found that, for most variables, between 80 and 90 percent of the respondents report mortgage information consistently over time in the AHS surveys. As expected, the number of responses that are consistent with the base year survey tends to decrease in the later survey years regardless of the variable examined. Among the variables and survey years we examined, first-time homebuyer status and mortgage term were reported most consistently, with 90 percent of responses consistent over time. Most homeowners also provided consistent answers on questions about their loan’s mortgage insurance type and payment plan type. Information on origination amount and monthly payment amount were less reliable, with consistency rates of 65 to 70 percent several years following mortgage origination. Interest rate information was by far the least reliable over time, with only 40 percent of responses consistent with initial answers eight years after origination. Of note, these last three variables all relate to the mortgage payment type: that is, whether the loan is fixed or adjustable rate. The relatively low consistency rates in these variables may reflect errors in reporting loans as fixed rate when they were actually adjustable rate. Reliable AHS Mortgage Variables We conclude that the following AHS variables and AHS-based estimates are reliable and can be used for mortgage market analysis, subject to the caveats described in each section: • Mortgage Originations for Home-Purchase Loans. Using the adjustments and timeframes described in this report, the AHS can be used to support analyses of the total volume of recent originations of loans for home purchases. Since only surviving loans are observed in the AHS, it is necessary to adjust volume estimates upward to account for the number of loans that terminated prior to the interview. We recommend that all AHS volume estimates should be adjusted upward using the termination rates of the market segment to which the mortgage belongs, when that is feasible. Because of the limited data availability for this study of termination rates for other market vii segments, the analysis in the report used FHA historical termination rates to adjust the volume of loans in all segments of the market.1 The AHS is best used to estimate volumes of recent originations, but not of originations that occur in the same year as the survey. Because of the timing of AHS interviews, spanning the last quarter of the year, the AHS will miss a large number of loans originated during the survey year. Therefore, information collected in the AHS should be used to examine the volume and loan characteristics of mortgages originated in the two years prior to the interview year. For example, the 1997 survey should be used to derive volume estimates for loans originated in 1995 and 1996. • Volume and Share by Market Segment. Starting with 1993 originations, researchers can feel comfortable using the AHS to examine mortgage activity by market segment using the following division: FHA, VA/FmHA/RHS, and conventional. Volumes and share estimates for these separate market segments were found to be reliable. The AHS slightly undercounts the share of jumbo loans. It is possible that some borrowers in the conventional segment of the market underestimate their original mortgage amounts. An alternative explanation is that homeowners with jumbo mortgages are underrepresented in the survey as a result of nonparticipation. Given the fact that participation in the AHS survey commits one to repeated long interviews, more wealthy homeowners may be less willing to participate. It is also likely that, in a declining interest rate environment, those with jumbo loans are particularly quick to refinance their mortgages multiple times, which would not be captured in the AHS. The AHS can be used to support research on mortgage activity by market segment as long as all conventional loans are grouped together. • Original Mortgage Amount. Original loan amounts for mortgages in the FHA segment generally are accurately reported. There is a slight underestimate of the mortgage amounts for borrowers in the VA/FmHA and conventional conforming segments. Users should consider applying an adjustment factor to inflate the loan amounts for such mortgages. In addition, the loan amount variable provided in the public-use file is top-coded for confidentiality reasons. Since the top-coded value has been above the conventional conforming loan limit, the truncation of the loan amount primarily affects analysis of jumbo loans. Therefore, the AHS cannot be used to evaluate jumbo loan amounts. • Interest Rate. Interest rates reported by borrowers in the AHS track the benchmarks for home-purchase fixed-rate mortgages in the FHA, conventional conforming, and conventional 1 Ideally, the loan volume estimates from the AHS should be adjusted using historical termination rates from the corresponding market segments and mortgage product types, since default and prepayment speeds are likely to be different across market segments. Borrowers with conventional mortgages are more likely to refinance their mortgages than their FHA counterparts, given the same interest rate environment. On the other hand, FHA loans have higher claim rates. We expect that the termination rates for FHA loans can serve as a reasonable proxy for the other market segments because, on average, the low FHA prepayment rates should be offset by the high claim rates of FHA loans. The proxy adjustment rate is probably most problematic in a survey year such as 2001 with big refinancing boom. viii jumbo segments of the market. Because of the lack of benchmark information, we have not been able to verify the reliability of interest rates for non-fixed-rate loan types reported in the AHS. • First-Time Homebuyer. The AHS is generally consistent with the RFS and Chicago Title Surveys regarding the share of borrowers who are first-time homeowners. The AHS does consistently report a lower share of first-time buyers than the FHA administrative data. But the cause of this discrepancy is most likely a difference in how this information is defined and solicited in the two data sets. While AHS asks whether any household members have previously owned a home, the FHA mortgage application form inquires whether the buyer has owned a home in the past three years. The AHS definition and estimates are thus more consistent with a strict notion of first-time homebuyers. • Borrower Race/Ethnicity. Racial and ethnic characteristics reported in the AHS track the benchmark database estimates reasonably well, although there is a tendency to report a slightly higher market share for African Americans and a lower share for Hispanics. These discrepancies most likely result from differences between the AHS and the administrative data of FHA and HMDA both in definitions of race/ethnicity categories and in the manner in which the race/ethnicity information is gathered. Unreliable or Uncertain AHS Mortgage Variables Users should avoid using the AHS (or use caution in some cases) for mortgage market research that would require the following: • Refinances. Refinance originations are not captured adequately in the AHS. Across most of the survey years and all mortgage market segments, the AHS consistently undercounts the number of refinance loans originated. Surprisingly, the loan purpose questions (REFI variable) added in the 2001 survey did not provide any noticeable improvement in this underreporting problem. Researchers thus should not use the surveys to produce volume estimates of such mortgages and should not attempt other analysis for which accurate identification of the market share of refinance loans is critical. However, the distributions of certain loan characteristics such as mortgage market segment, original mortgage amount, and interest rate among refinances that are captured in the AHS appear to be consistent with the benchmarks for most of the years. Therefore, users may be able to use the data for these loans for analysis pertaining to such loan characteristics. In particular, these records may be used as micro data to conduct regression modeling and other loan-level analyses. The reason for this underreporting is unknown. One possible explanation is that the questionnaire does not provide a clear guidance to the homeowners regarding the distinction between lump-sum home equity loans and mortgages that are used to refinance an existing loan in order to take out housing equity (so-called cash-out refinances). As a result, some ix survey respondents might have incorrectly classified their refinance mortgages as home equity loans. The mediocre performance of the loan purpose question (REFI variable) in the 2001 survey could be because 2001 was such a big refinancing year that the proxy adjustment rates have failed to account for many of the loans that had terminated prior to the interview. In particular, some borrowers might be able to refinance their mortgages multiple times. The REFI variable may perform better in future survey years. • Jumbo Loans. The AHS generally understates the share of mortgages that exceed the conforming loan limit, although it is fairly reliable at identifying conventional loans overall. Possible explanations for this underreporting are offered in the previous section titled “Volume and Share by Market Segment.” • Payment Product Types. Payment product types are not identified accurately in the AHS. In general, ARMs are underreported, and the survey instrument is incapable of identifying borrowers with more sophisticated payment product types such as hybrid ARMs and two-step mortgages. The share of fixed-rate mortgages is generally over estimated, although when measured as a percent of the overall share, the magnitude of the discrepancy is smaller than for ARMs. Thus, the AHS should not be used for analyses of mortgage product choice, and analysts should use caution when including mortgages other than fixed rate mortgages in other types of analyses. • Loan-to-Value (LTV) Ratios. Although average LTV ratios track well the benchmark estimates for both FHA and conventional loans, the distributions of the ratios do not track as well. Of particular note for policy makers, it appears that the AHS estimates of the share of mortgages with very high LTVs may be overstated. This may result from differences in the way house values are recorded in different databases. Discrepancies may also be caused by the extreme values of the loan amount and house value variables that were inaccurately reported in the AHS. Further investigation is needed. Thus, researchers should be cautious in their use of LTV estimates based on the AHS. • Borrower Income. Borrowers’ income in AHS does not track the values reported in HMDA and the FHA data well, mainly because the AHS variable includes different components of income. The AHS measure is more complete and thus yields estimates that are consistently higher. On the other hand, when incomes captured in the Current Population Survey (CPS) and Social Security Administration data are used as benchmarks, studies by the Census Bureau have shown that the AHS respondents have a tendency to underreport incomes. Researchers need to make a judgment independent of the benchmarking to HMDA and FHA data conducted for this study on whether income as identified by the AHS is an appropriate measure of income for their mortgage market research questions. x Areas Requiring Further Investigation • The substantial underreporting of refinance mortgages in the AHS deserves further examination. Given the longitudinal nature of the AHS survey design, linked AHS files might be used to explore whether refinances can be identified more accurately from AHS panel data. Looking at refinances from a panel setting might also shed new light on the structural cause of underreporting of such mortgages and suggest modifications that could be made to the survey. In particular, responses linked across surveys could be used to examine when home equity loans were originated to determine if this method might be able to provide a more complete accounting of refinance loans. For example, this method might be able to identify cases where a previous mortgage was retired and a new home equity loan was originated, which could be counted as a refinance. • For longitudinal analysis, additional research is needed to examine whether there are systematic ways to delete problematic cases from the AHS files to produce a core of internally consistent records with mortgage variables that are reliable. Would the remaining sample be too small to conduct any meaningful analysis? Would deletion of such cases distort the composition of the remaining sample relative to the benchmarks? • LTVs are another area where further study is needed before determining that the AHS is incapable of supporting LTV research. We suspect that some of the discrepancies in the LTV estimates between the AHS and other mortgage market data sets are caused by outliers in loan amounts and house values reported to the AHS. The distribution of loans by LTV category might be usable if problematic outlier cases were removed by the analyst. • The micro data for the new round (2001) of RFS will be available to researchers in the summer of 2003. Extensive benchmarking of the 2001 AHS to the new RFS data should be undertaken to gain insight into the impact of post-1995 AHS survey changes on the accuracy of the AHS data. The RFS can be used to test mortgage market measures of recent cohorts as well as the entire stock of mortgages that were in place in 2001. It will be especially valuable if the internal version of the RFS file can be used. The RFS is particularly valuable as a point of comparison since information on mortgages is collected directly from lenders. These data elements should be quite accurate, since the lenders’ information is likely supported by administrative records (rather than the homeowners’ memory as in data based only on household surveys). Recommendations for Modifications to the Survey Based on the research conducted for this study, we have identified several potential modifications to the AHS that would make it substantially more useful for research on mortgage market issues: • Questions should be added to identify the number of refinances that have taken place during the interval since the last survey. The home equity questions also should be changed to ask xi about lines of credits rather than “home equity loans,” since there is no real distinction between lump-sum home equity loans and standard mortgages. • In the current AHS questionnaire, for homeowners who obtained the mortgage the same year the home was purchased, the survey does not obtain the year of loan origination from the respondents; the origination year information can only be retrieved from the WHNGET variable in the “purchase” module of the AHS survey, which reports the year when the house was obtained. For other homeowners, the survey asks explicitly what year the owner obtained the mortgage and the information is stored in the YRMOR variable. HUD and the Census Bureau should consider asking all homeowners the month and year of mortgage origination explicitly and storing the information in one variable. • HUD and the Census Bureau should evaluate changes to specific questions that might elicit more accurate reporting of mortgage payment product types. One of the issues to be considered is the cost-effectiveness of making such changes, given the rapid evolution of mortgage products. • New questions should also be considered to collect information on mortgage payment status (delinquency and default) and the termination of mortgages that were previously in place. • HUD and the Census Bureau should consider asking an explicit question on the unpaid principal balance (UPB). UPB is useful for estimating the current home equity and LTV of each homeowner. Currently, UPB is not collected in the AHS. Users can calculate an estimate using information pertaining to interest rate, original amount and mortgage term. Given the lack of details on adjustment terms for non-fixed rate loans and the instability of the interest rates and mortgage amounts reported across waves of the survey, this method is not likely to yield reliable UPB estimates. Although it is likely that some borrowers may not be able to report the UPB accurately if asked in the survey, the self-reported estimates can nonetheless provide an internal validity check against the user-calculated amounts. • Given our findings on the instability of several key variables over time, we recommend that the “dependent interviewing” technique be extended to include mortgage insurance type, payment plan, interest rate (of fixed-rate loans), and principal payment amount. This technique is used to avoid repeating questions to the household if the answer should not have changed since the previous survey. For example, the first-time buyer status question was not asked in the 1999 survey if the same household occupied the housing unit and a valid answer had been obtained in a previous survey. xii Exhibit 0-1: Major Findings for AHS Mortgage Market Measures Tested in the Replication Analysis MORTGAGE MARKET MEASURE BENCHMARK DATA & ORIGINATION COHORTS EXAMINED Mortgage Origination Volume • • FHA: 1989-2000 for FHA HMDA: 1993-2000 for VA/RHS, Conventional Conforming and Conventional Jumbo STRATIFICATION MAJOR FINDINGS • Home-purchase loans: • Mortgage Market a Segment b Loan Purpose • • • Generally accurate for volume estimates of origination cohorts. Over the 1993 to 2000 period, the average coverage ratio was 99 percent for all loans combined (Exhibits 4-3, 4-4, and 4-5). Coverage ratio declined for the 1998 to 2000 cohorts. Average coverage ratio was in the 83-89 percent range for all loans combined during that period (Exhibit 4-4). Relatively low volume estimates for conventional jumbo originations (Exhibit 4-5). xiii Refinance loans: • • • Low volume estimates across all mortgage market segments. Average coverage ratio was 48 percent for 1993 to 2000 (Exhibit 4­ 7, 4-8, and 4-9). Especially low coverage ratio for cohorts since 1993 (1995 survey) probably because of question wording (a lack of clear distinction between cash-out refinance mortgages and lump-sum home equity loans) or coding errors in the survey (Exhibit 4-8). Loan purpose questions added in the 2001 survey did not significantly improve the survey’s ability to capture refinance loans (Exhibit 4-8). MORTGAGE MARKET MEASURE Loan Characteristics Mortgage Market Segment BENCHMARK DATA & ORIGINATION COHORTS EXAMINED • • • RFS: 1989-1991 HMDA: 1993-2000 SCF: 1993-1994 STRATIFICATION MAJOR FINDINGS • Home-purchase loans: Loan Purpose • • • For loans originated in 1989-1991, slight underestimate of FHA share and overestimate of conventional conforming share (Exhibit 4­ 10). Accurate market shares for loans in the FHA, VA/RHS and conventional segments, 1993-2000. Discrepancies are mostly in the 1 to 2 percent range (Exhibit 4-11). Slight underestimate of market share of conventional jumbo loans and overestimate of conventional conforming loans for most of the cohorts (Exhibit 4-11). Refinance loans: • xiv Origination Loan Amount • • FHA: 1989-2000 for FHA HMDA: 1993-2000 for VA/RHS, Conventional Conforming • • Mortgage Market Segment Loan Purpose Small sample sizes result in wide confidence intervals that make it difficult to draw statistically reliable comparisons. • Accurate shares for loans in the 1989-1991 cohorts (Exhibit 4-10). • Underestimate of shares for conventional jumbo loans for most cohorts (Exhibit 4-12). Home-purchase loans: • • • Accurate for loans in the FHA market segment (Exhibit 4-13). Slight underestimate, by about 10 percent, of loan amounts in the VA/RHS segment (Exhibit 4-15). Accurate for 1995, 1997, 1998, 1999 and 2000 conventional conforming loans. Underestimate, by about 10 percent, of those in 1993, 1994 and 1996 cohorts (Exhibit 4-15). Refinance loans: • • • Underestimate of loan amounts of FHA loans originated in 1995 and 1996 (Exhibit 4-14). Small sample sizes for VA/RHS loans (Exhibit 4-16). Accurate for 1995, 1997 and 1998 conventional conforming loans. Underestimate for 1993, 1994 and 1996 cohorts. Overestimate for 1999 and 2000 cohorts (Exhibit 4-16). MORTGAGE MARKET MEASURE BENCHMARK DATA & ORIGINATION COHORTS EXAMINED Mortgage Interest Rate (Fixed-Rate Loans) • • LTV at Origination • • STRATIFICATION MAJOR FINDINGS FHA: 1989-2000 for FHA MIRS: 1992-2000 for Conventional Conforming and Conventional Jumbo Home Purchases • Mortgage Market Segment Loan Purpose Payment Product c Type • FHA: 1989-2000 for FHA Home Purchases MIRS: 1992-2000 for Conventional Conforming Home Purchases • Mortgage Market Segment Loan Purpose • • • • • • • xv Mortgage Payment Product Type • • Borrower Characteristics First-time Homebuyer • Status • • FHA: 1989-2000 for FHA MIRS: 1992-2000 for Conventional Conforming and Conventional Jumbo Home Purchases • Chicago Title: 1995-1999 RFS: 1989-1991 FHA: 1991-2000 for FHA • • Mortgage Market Segment Loan Purpose • • • • Mortgage Market Segment Loan Purpose • • • Accurate averages for both home-purchase and refinance loans in segments of the market examined. Discrepancy in the averages is generally one to two tenths of a percent. Overall distribution appears to be more dispersed (Exhibits 4-17 and 4-18). Compared to MIRS, the AHS reports a higher proportion of conventional conforming purchases with relatively high interest rates, probably because subprime loans are included in the AHS but not in MIRS (Exhibits 4-19 and 4-20). Reasonably accurate mean LTVs by cohort for FHA home purchases. Distribution of loans by LTV categories does not track well with benchmarks (Exhibits 4-21 and 4-22). Reasonably accurate mean LTVs by cohort for conventional conforming purchases. Distribution of loans by LTV categories does not track well with benchmarks (Exhibits 4-23 and 4-24). LTVs in the other segments of the mortgage market cannot be tested because of the lack of comparable benchmarks. Understates ARM shares for FHA loans (Exhibits 4-25 and 4-26). Overstates shares of loans other than fixed-rate and ARMs in the FHA segment (Exhibits 4-25 and 4-26). Understates ARM shares for conforming and jumbo conventional home purchases (Exhibits 4-27 and 4-28). Estimate of first-time buyer shares for all home purchases consistent with RFS and Chicago Title data (Exhibits 4-29 and 4­ 30). While differences with RFS are not statistically significant, the share of first-time buyers in AHS is lower for FHA and conventional jumbo borrowers but about equal for VA/RHS and conventional conforming (Exhibit 4-30). Discrepancies between the AHS and FHA data are due to how firsttime buyer status is defined. The AHS estimates are more consistent with a strict notion of first-time buyers. MORTGAGE MARKET MEASURE BENCHMARK DATA & ORIGINATION COHORTS EXAMINED Borrower Race/Ethnicity • • Borrower Income Notes: a • FHA: 1989-2000 for FHA HMDA: 1993-2000 for VA/RHS, Conventional Conforming and Conventional Jumbo HMDA: 1993-2000 for FHA, VA/RHS, Conventional Conforming and Conventional Jumbo STRATIFICATION MAJOR FINDINGS • • • • • Mortgage Market Segment Loan Purpose • Mortgage Market Segment Loan Purpose • • Different race/ethnicity questions and method of gathering information between AHS and administrative benchmark data sets make interpretation of differences difficult. Overall, distributions are reasonably close to the benchmarks (Exhibits 4-32 to 4-40). Incomes are defined differently between AHS and benchmark data, making it difficult to interpret discrepancies. Household incomes in AHS are consistently higher than in HMDA, probably because in HMDA some borrowers do not need to report all income components to lenders to qualify for loans (Exhibits 4-41 to 4-44). Morgage market segments are FHA, VA/FmHA/RHS, conventional conforming, and conventional jumbo. Loan purposes are home purchase and refinance mortgages. c Payment product types are fixed-rate and adjustable rate mortgages. b xvi Chapter One Introduction This research was conducted under Task Order 7 of HUD Contract C-OPC-18571 for an “Analysis of Housing Finance Issues Using the American Housing Survey (AHS).” The purpose of the study is to evaluate the quality of mortgage data in the AHS for analyzing issues of interest to policy makers and the mortgage industry. To test the reliability of the AHS variables we replicate measures of mortgage activity from other reliable sources of data that serve as benchmarks for the AHS estimates. We also use the longitudinal nature of the AHS to determine whether answers to questions on mortgages are consistent and stable over time. The goals of the analysis are to determine: 1) what types of mortgage market analysis can be supported by the AHS; 2) what areas of the AHS are problematic for mortgage research; and 3) what techniques or changes in the survey could potentially compensate for the problems. Background and Policy Context The American Housing Survey (AHS, formerly Annual Housing Survey) is the most comprehensive source of information about the characteristics and condition of the nation’s housing stock. The AHS consists of two components, a national sample and a series of metropolitan area samples. Started in 1973, the AHS national sample data were collected by the U.S. Census Bureau for the Department of Housing and Urban Development (HUD) on a nationally representative sample of housing units (about 55,000 homes) every year until 1981, and they have been collected every other year since then. The survey has contained the same basic sample of housing units since 1985, with units added to each wave of the survey to reflect additions to the housing stock. As for the Metropolitan Statistical Area (MSA) samples, surveys for 46 selected MSAs are collected every 4 to 6 years, with an average of 12 MSAs included in each survey year. Each of the metropolitan area samples covers about 3,000 housing units. The metropolitan area survey includes all of the questions from the national survey, with some additional questions related to commuting patterns and the location of previous residence. Over the years, AHS data have been used by researchers and policy analysts to answer questions about the nation’s housing conditions (e.g., unit quality, unit and building size, repairs, structure age and neighborhood quality) and occupant characteristics (e.g., tenure choice, race, income, and housing costs). The longitudinal nature of the AHS also permits the analysis of trends for certain housing and occupancy characteristics of the housing stock. The AHS data contain detailed questions about mortgages asked of respondents for owner-occupied units in the survey. The questions cover most basic mortgage and housing finance topics, including loan amount, mortgage product type (fixed rate versus adjustable rate), mortgage term, contract interest rate, year of origination, type of mortgage insurance, and whether the owner is a first-time or repeat home owner. This wealth of mortgage-related variables, combined with the tenant demographic and property location information, could be a very powerful resource for answering many housing finance research and policy questions. These micro-data could provide crucial information to support HUD’s regulatory and programmatic responsibilities, which include regulating 1 the government-sponsored-enterprises (GSEs), evaluating the role of FHA in the mortgage market, and setting performance standards for HUD programs. However, neither the housing research community nor HUD staff has made as much use of the mortgage variables of the AHS data as might be expected. Among the reasons for this under­ utilization is the fact that the reliability of these mortgage-related variables in the AHS has not been verified. Analysis is needed to establish the extent to which limitations associated with sample size, survey design, and interview response affect the accuracy and consistency of the mortgage data in the AHS. The research presented in this study is intended to meet this need. Outline of the Report The rest of the report is organized as follows: • Chapter Two presents an overview of the AHS, a detailed description of the mortgagerelated variables, changes in the survey over time that may have affected the mortgage variables, a summary of the potential advantages and disadvantages of using the AHS for mortgage-related research, and the set of variables that were tested in the replication analysis. • Chapter Three describes the mortgage market data chosen to provide benchmark estimates for the AHS variables, highlighting the strengths and weaknesses of each database. The chapter provides the rationale for our choice of each data set used in the replication analysis. • Chapter Four presents the findings of the replication analysis. The AHS estimates are compared to the independent benchmarks derived from the comparison databases. The variables we have examined are loan volume, loan purpose, mortgage market segment, loan amount, contract interest rate, loan-to-value (LTV) ratio, mortgage payment product type, first-time buyer status, borrower race/ethnicity, and income. • Chapter Five evaluates the internal consistency and stability of the AHS mortgage data elements reported across time. Sample housing units are linked across different survey years. This investigation focuses on a few key mortgage variables. • Chapter Six synthesizes the findings. It serves as a guide for HUD staff and other researchers on the potential use of the AHS variables for different types of research on mortgages and housing finance. Topics related to the reliability of AHS mortgage variables that require additional investigation are described. • Appendix A documents the GSE conforming loan limits used to distinguish between conforming and jumbo conventional mortgages. • Appendix B presents detailed information about the comparison databases. 2 • Appendix C discusses alternative methods that can be used to account for mortgage activity between survey waves that is not captured by the AHS. • Appendix D provides additional tabulations to support the analyses presented in Chapter Four. It includes some additional estimates that may be of interest but are not discussed in the text. • Appendix E describes the steps taken to link observations on mortgage activity from multiple years of AHS surveys. • Appendix F presents additional tabulations to support the analyses presented in Chapter Five. • Appendix G provides the text of the questions asked in the “mortgage” and “purchase” modules of the 1997 AHS. • Appendix H presents a user’s guide for deriving the mortgage market attributes in the AHS that we have examined. For each mortgage market measure, it identifies the AHS variable(s), as well as the computations and selection criteria involved in the derivation. This information differs across survey years for some of the variables. 3 Chapter Two The AHS and Research on Mortgages This chapter examines the characteristics of the AHS that make it a potentially useful source of information to support analysis of issues related to residential mortgage finance. The first section provides a brief description of the history and basic characteristics of the survey. The next section describes the variables that are collected by the AHS that relate to mortgage finance. The third section describes changes over time to aspects of the survey that potentially affect the use of the data for mortgage market analysis. The fourth section summarizes the advantages and disadvantages of the AHS for mortgage related research. Specific mortgage variables to be investigated in the replication analysis are presented in the final section. History and Basic Characteristics of the AHS Besides the decennial census, the American Housing Survey contains the most comprehensive information available on the nation’s housing stock. The AHS consists of two components, a national sample of housing units and a series of metropolitan area samples. Begun in 1973 as the Annual Housing Survey, the national AHS survey was conducted annually through 1981. Since 1983 the national survey has been conducted every other year and has been known as the American Housing Survey. The AHS is a panel survey of about 55,000 units that collects information on the same housing units in each wave. A new sample was drawn for the 1985 national survey based on the 1980 decennial census. The same basic sample of housing units has been followed since 1985, with units added to each wave of the survey to reflect additions to the housing stock. Each unit surveyed is identified by a unique control number which allows the linking of data for individual housing units from different survey years to analyze changes in housing conditions over time. The metropolitan area surveys are conducted in 46 selected MSAs. Each area is surveyed every 4 to 6 years, with an average of 12 MSAs included in each survey year. Each of the metropolitan area samples covers 3,000 or more housing units. The metropolitan area survey includes all of the same questions from the national survey with some additional questions related to commuting patterns and the location of previous residence. Like the national survey, the metropolitan area samples include a panel of housing units surveyed each time supplemented by additional units that reflect new construction. From 1974 to 1994 the sample for the metropolitan area surveys was based on the 1970 census. All but six metropolitan area samples were redrawn in 1995 based on the 1990 census. Beginning in 1995 six of the largest metropolitan areas were covered as part of the national survey so, like the national survey, the sample for these areas is based on the 1980 census.2 As a result of these changes in the metropolitan area samples, analysis that relies on the panel nature of the AHS metro surveys must either focus on the 1973 to 1994 period or the 1995 period and later. 2 The metropolitan areas that are included as part of the national survey include Chicago, IL PMSA, Detroit, MI PMSA, Los Angeles-Long Beach, CA PMSA, New York-Nassau-Orange County, NY PMSA, Northern New Jersey PMSA, and Philadelphia, PA-NJ, PMSA. 4 The AHS covers a broad range of topics, including: • The general characteristics of the structure (size, age, type, tenure, lot size, parking, number and type of rooms); • The equipment used for appliances, heating and cooling; • The quality of the unit in terms of interior and exterior conditions and reliability of equipment; • Recent renovations to the unit; • Neighborhood characteristics; • All housing related costs (including rent, mortgage characteristics and costs, utilities and services, insurance, and property taxes) and the use of housing subsidies; • Household composition including the age, sex, and race of all household members and their relationship to the respondent; • Income for each household member; and • The previous residence of household members and the reasons for moving. The broad and detailed set of information covered by the survey supports a wide range of housing research. The AHS is a particularly important source of information on the cost and quality of housing over time and the housing characteristics of different demographic groups. Mortgage-Related Variables in the AHS In each AHS survey, a series of housing finance and mortgage-related questions is posed to respondents residing in owner-occupied housing units.3 Although many of the questions have remained the same over the years, a few have changed. For instance, questions related to home equity loans have been added since the 1995 survey, while questions concerning the existence of reverse mortgages were added in 1997 and then dropped from the 1999 survey. Appendix G provides the text of the questions asked in the “mortgage” and “purchase” modules of the 1997 AHS.4 Only minor changes have happened in these modules in the subsequent two survey years. The following description in this section applies to the questionnaire of survey years that we examined in this study. (A thorough discussion of relevant changes in the survey questions or methodology is presented in a later section of this chapter.) 3 These questions are grouped under the “Mortgage” module in the Codebook for the American Housing Survey, Volume 3: 1997 SAS Files and Questionnaire, draft, February 16, 1998. 4 Text of questions for the 1999 and 2001 surveys cannot be included here because the Census Bureau has not published them. 5 The mortgage module of the survey begins by identifying how many mortgages are on the property.5 There is some attempt to distinguish between regular mortgages and home equity loans. The respondent is asked explicitly how many mortgages are on the property excluding home equity loans as well as whether there is a home equity loan in place. However, no clear distinction is made between lump-sum home equity loans and mortgages that are used to refinance an existing loan in order to take out housing equity (so-called cash-out refinances). The owner is then asked whether a mortgage was obtained through a State or local government program that provides lower cost mortgages, although the survey does not link this question to a specific mortgage. Following these introductory questions, the survey asks a series of questions on the first mortgage on the property. If there is more than one mortgage, the survey then asks the same set of questions for the second mortgage. Exhibit 2-1 identifies the variables that are gathered in this series of questions and describes questions that are used to elicit this information. As the exhibit notes, some of the questions are worded indirectly as an attempt to elicit more reliable and consistent answers. For example, the questionnaire does not ask directly whether the borrower has an adjustable rate mortgage (ARM) loan; rather, the respondents are asked whether the payments remain the same over the life of the loan and, if they do not, why they change. Exhibit 2-1 Characteristics of First and Second Mortgages Collected in the AHS VARIABLE Year Mortgage Originated New or Assumed Mortgage Term of Mortgage Amortization Period SOURCE QUESTIONS FOR THE VARIABLE The survey first asks whether the owner got the mortgage the 6 same year the home was purchased. If not, it then asks explicitly what year the owner obtained the mortgage. The survey asks whether the owner got a new mortgage or assumed someone else’s mortgage. For new mortgages, the survey asks how many years the mortgage was for when it was acquired. If the mortgage is assumed, the survey asks how many years remained on the mortgage when assumed. If the reported term is less than 15 years, the respondent is asked how many years it would take to pay off the loan, given the current monthly payments. This question is intended to elicit the amortization period used to estimate payments on balloon loans. If the loan term is 15 years or more, the amortization and term are assumed to be the same. 5 In the 1997 survey, the module begins by asking the respondents whether there is a reverse mortgage on the property. If there is, the survey assumes there is no other mortgage in place. If there is no reverse mortgage, questions for the mortgage module continue. 6 For homeowners who obtained the mortgage the same year the home was purchased, the survey does not obtain the year of loan origination from the respondents; this information can be retrieved only from the WHNGET variable in the “purchase” module of the survey, which reports the year the house was obtained. For other homeowners, the YRMOR variable contains the year of mortgage origination. 6 VARIABLE Amount of Mortgage Current Interest Rate Current Monthly Payment Government-Insured Mortgage Financing Provided by Seller or Other Individual Mortgage Payment Features SOURCE QUESTIONS FOR THE VARIABLE For new mortgages, the survey asks how much was borrowed. If the mortgage is assumed, the question is how much was left to pay off when it was assumed. The respondent is asked whether the mortgage also covers other homes or apartments, farm land, or a business on the property. If so, the respondent is then asked how much of the mortgage applies to the home in the survey. The survey asks for the current interest rate rounded down to the nearest ¼ percent. The survey asks for the current monthly payment. Separate questions are then asked to determine if the payment includes property taxes, homeowner insurance, or “anything else” in addition to principal and interest. The survey taker is instructed that “anything else” may include payments for disability or life insurance. If “anything else” is reported to be part of the monthly payment, the survey asks for the annual amount of these payments. In the 2001 survey, a question is added to determine whether the mortgage payment includes private mortgage insurance (PMI). If it does, the survey inquires how much the PMI payment is. The survey asks if the mortgage is an FHA, VA, FmHA or some other type. The survey asks if the loan was provided by a bank or other organization or by an individual. If the loan was made by an individual, the survey then asks if the individual was the former owner of the property. The survey asks if the payments on the loan stay the same during the whole length of the mortgage. If not, the survey then asks how they change. Any of the following payment plan features that apply will be recorded: • Change in taxes or insurance, or due to decline in the principal balance; • Change based on interest rates; • Rise at fixed schedule during part of the loan term; • Rise at fixed schedule during the whole length of the loan; • Last payment is the biggest; or • Other. These payment features do not allow the AHS users to easily and unambiguously classify each loan into a mutually exclusive payment product type. After gathering this detailed information on the first and second mortgages, the survey then inquires about the amount borrowed under all other mortgages and the monthly payments for these loans. The sum of the amounts borrowed and the total monthly payments are reported. Finally, the survey asks a series of questions about any home equity loans on the property. The same set of questions is asked for up to three home equity loans. Exhibit 2-2 summarizes the information gathered on home equity loans. Much less information is gathered about home equity loans than about mortgages. Perhaps most significantly, the survey does not ask when the home equity loan was originated. 7 Exhibit 2-2 Characteristics of Home Equity Loans Collected in the AHS VARIABLE SOURCE QUESTIONS FOR THE VARIABLE Type of Home Equity Loan The survey asks whether the owner is allowed to borrow against the home as often as he or she wishes up to a fixed amount (a line of credit) or if he or she obtained a one-time, lump sum payment that is repaid over a period of time. For line of credit loans, the survey asks for the total credit limit. For lump sum loans, the survey asks for the amount of the lump sum. The survey asks for the current outstanding balance. The survey asks for the current interest rate rounded down to the nearest ¼ percent. The survey asks for the amount of the last monthly payment. Credit Limit Lump Sum Amount Outstanding Balance Current Interest Rate Monthly Payment In addition to the mortgage-specific variables, the AHS also contains other information relevant to the analysis of mortgages and housing finance. These include:7 • Amounts paid for other housing costs, including property taxes, homeowner’s insurance, homeowner association or condominium fees, and land rents; • The year the home was purchased; • The type of property (single-family, condominium, or mobile home); • Sales price of the property when purchased; • The owner-estimated current house value at the time of the interview;8 • Whether a down payment was used to purchase the home and, if so, the major source of the down payment: sale of previous home, sale of other investment, savings or cash on 7 Questions related to these data elements can be found in the “Housing Costs,” “Numbers of Units in Building,” “Rooms, Type of Building, Size, Vehicles,” and “Purchase, Inspection, Value, Year-round Use” modules in the Codebook for the American Housing Survey, Volume 3: 1997 SAS Files and Questionnaire, draft, February 16, 1998. 8 There is a sizable body of research literature testing the accuracy (against external data) and consistency (across survey years) of the owner-estimated house value in the AHS. For a review of the related research, see Katherine A. Kiel and Jeffrey E. Zabel, “The Accuracy of Owner-Provided House Values: The 1978­ 1991 American Housing Survey,” Real Estate Economics, summer 1999, v.27(2): 263-98; John L. Goodman and John B. Ittner, “The Accuracy of Home Owners’ Estimates of House Value,” Journal of Housing Economics, v.2(4): 339-57. 8 hand, borrowing other than a mortgage on the property, inheritance or gift, land where property was built was used for financing, or other; and • Whether the owner previously owned a home or is a first-time homebuyer. Finally, there are a number of mortgage-related characteristics that are not collected directly by the survey but can be derived by the users from the information that is collected. Exhibit 2-3 identifies the mortgage-related variables that can be derived from the AHS and indicates the process used to estimate these variables. Exhibit 2-3 Mortgage-Related Variables Derived from AHS Variables DERIVED VARIABLE HOW IT IS DERIVED Mortgage Market Segment Whether the loan is government insured, conventional conforming or a jumbo can be derived from the type of mortgage (FHA, VA, FmHA/RHS, or other), the original mortgage amount, and the year 9 of origination. Before the 2001 survey, home purchase vs. refinance can only be identified by comparing the mortgage origination year with the year of house purchase. Loans originated in years after the home was acquired are assumed to be refinancings. In the 2001 survey, refinances can be identified directly using the REFI variable. The survey also ask the reason for the refinance: • Get a lower interest rate; • Increase the payment period for the mortgage; • Reduce the payment period for the mortgage; • Renew or extend a loan that has fallen due (without increasing the outstanding balance); • Receive cash or increase the outstanding balance of the loan; or • Other reason. Then, the borrowers are asked how much cash they received as a result of the refinancing, and what percentage of the cash was used for additions, improvements or repairs to the house. Loan Purpose 9 “Conventional” refers to loans that are not insured or guaranteed by the government. “Conforming” refers to loans that are below the limits placed on residential loans that can be purchased or guaranteed by Fannie Mae and Freddie Mac. Jumbo loans are mortgages that exceed this limit. Since the loan limits are set for calendar years, the year of origination is needed to identify the appropriate limit. The loan limits are documented in Appendix A. One obstacle to accurately characterizing loans as conforming loans using the AHS is that the conforming loan limit is 50 percent higher in Hawaii and Alaska. The AHS does identify observations from the Honolulu MSA, which can be used to identify cases where this higher limit prevails. However, observations in Hawaii outside of Honolulu and in Alaska generally may not be identifiable. But since these areas represent a very small share of the U.S. mortgage market, this issue should not cause a significant distortion in the identification of the conforming loan market. 9 DERIVED VARIABLE HOW IT IS DERIVED Loan-to-Value Ratio at Origination Ratio of original mortgage amount to purchase house price (or estimated home value if it appears to be a non-arms length transaction or a refinance loan). Ratio of estimated unpaid principal balance (UPB) to estimated home value. The Census Bureau staff has written a SAS computer program to 10 calculate the estimated UPB based on : • Original mortgage amount • Mortgage term • Interest rate • Mortgage Payment Plan • Number of years since origination Given the lack of details on adjustment terms for non-fixed rate loans, the estimated unpaid balance may be difficult to calculate accurately for these loans. Information collected on mortgage payment features can be consolidated by running a SAS computer program provided by the 11 Census Bureau. The computer program classifies each mortgage into one the following seven mutually exclusive product types: • Fixed payment, self-amortizing mortgage • Adjustable rate mortgage (ARM) • Adjustable term mortgage • Graduated payment mortgage • Balloon mortgage • Combination of the above • Other Ratio of payments for the mortgage plus property taxes, insurance, homeowner association fees, and land rents to household income. Loan-to-Value Ratio Over Time Mortgage Payment Product Type Housing Cost-to-Income Ratio Geographic