Demystifying the Refi-Share Mystery

Demystifying the Refi-Share Mystery Authors Yan Chang and Frank E. Nothaft Abstract The refinance shares reported by Freddie Mac s Primary Mortgage Market Survey (PMMS), the Mortgage Bankers Association (MBA), the Home Mortgage Disclosure Act (HMDA), and the National Mortgage News (NMN) have differed by up to 21 percentage points between 1990 and 2005. If a lender s refinance share varies with loan volume, then weighting by origination volume could explain the observed discrepancies. Based on HMDA data for 2000 (a low refinance year) and 2003 (a refinance boom), lender size was positively related to refinance share, after controlling for institution type, cultural affinity, and location. HMDA provides the best national measure of refinance share, and weighting by lender volume explains most of the difference among the four measures. Accurate measurement of refinance volume is critical to understanding the financial health of the household sector and the effect of consumer spending on economic activity. Estimates of current refinance activity serve as an ingredient into estimating the amount of interest payment reductions or the volume of homeequity cash-out, which can fuel consumer spending and home improvements. Former Federal Reserve Chairman Greenspan (2002) recognized the contribution of homeowner refinance to bolstering the economy: Especially important in the United States have been the flexibility and the size of the secondary mortgage market. Since early 2000, this market has facilitated the large debt-financed extraction of home equity that, in turn, has been critical in supporting consumer outlays in the United States throughout the recent period of economic stress. Zandi (2002) estimated that housing and mortgage activity accounted for nearly one-third of U.S. economic growth between 2000 and 2002. A Federal Reserve study that covered refinancing in 2001 and early 2002 found that about 61% of the monies went toward home improvements and the repayment of other debts; the use of the remaining funds was approximately split between consumer expenditures and various financial or business investments (Canner, Dynan, and Passmore, 2002, Table 6). The refinance share also serves as a critical input in estimating overall home mortgage originations in the model advanced by Greenspan and Kennedy (2005). Their approach was to decompose the net change in home mortgage debt outstanding into its constituent parts: gross credit flows in the home lending JRER Vol. 29 No. 4 2007

512 Chang and Nothaft market (capturing aggregate home-equity extraction through mortgage activity) and repayments of existing debt. Their estimates of gross home mortgage originations depend, in part, on an accurate assessment of the refinance share of home mortgage lending by lender type. Thus, accurate refinance share measurements have helped government officials in their deliberations over the appropriate course of monetary policy and the financial health of the household sector. Refinance volume has been more variable over time than purchase-money lending, and extended refinance booms and busts have led to substantial expansion and contraction in employment in the mortgage industry. 1 An accurate measure of refinance share is important for lenders, servicers, and others in the industry to gauge their employment needs. Currently there are four sources that independently collect and report the refinance share of home mortgage activity: Freddie Mac s Primary Mortgage Market Survey (PMMS), the Mortgage Bankers Association (MBA) weekly Mortgage Applications Survey, the Home Mortgage Disclosure Act (HMDA) data aggregated by the Federal Financial Institutions Examinations Council (FFIEC), and the mortgage volume reports from the National Mortgage News (NMN). 2 The PMMS has canvassed 125 lenders on a monthly basis since August 1987, and of the four sources is the longest time series of consistently measured refinance shares. Lenders are asked about the proportion of applications they received in the previous month that are for refinance. Some lenders do not respond to the refinance survey; the resulting sample size has typically varied between 70 and 100 lenders per month. Both large and small lenders are included in the survey, with a representative mix of thrifts, commercial banks, and mortgage companies, roughly proportional to the level of mortgage business each type commands nationwide. The refinance shares reported by each lender are averaged for the reported market refinance share. The MBA has surveyed about 20 large lenders each week since January 1990, including mortgage bankers, commercial banks, and thrifts and covers approximately 50% of the market. 3 Both shares of number of applications and dollar volumes are reported. HMDA data collection was authorized by Congress and implemented by the Federal Reserve Board s Regulation C. Amendments in 1988 expanded reporting and coverage and led to the collection of application-level data beginning in 1990; these amendments served several purposes, including making public the lending institutions efforts in extending housing credits, identifying potential areas where additional efforts were needed, and discouraging discriminatory lending. Institutions that are required to report under HMDA must meet the minimum level of assets or origination volume requirement and have a home or branch office in a metropolitan statistical area (MSA), or, in the case of nondepository institutions, have lending activity in an MSA (Federal Financial Institutions Examination

Demystifying the Refi-Share Mystery 513 Council, 2003). Micro-level HMDA data go back to 1990 and include both application and origination data, as well as loan purpose (home purchase, refinance, home improvement), the latter permitting calculation of the refinance share. The Federal Reserve Board has also made available aggregate quarterly HMDA origination volumes by refinance, home purchase, and home improvement, thereby allowing computation of a quarterly refinance share. The Federal Reserve Board estimates that HMDA covers about 80% of home lending; in 2005, 8,690 lenders reported single-family originations under HMDA. 4 The NMN has surveyed 100 of the largest mortgage originators each quarter since 1998. Their reported market share is near 85%, according to the latest survey as of the second quarter of 2006. From the surveyed lenders, a portion would not disclose their refinance shares; therefore, their population for the refinance calculation is smaller than the total, usually between 60 and 100 lenders. During the fourth quarter of 2005, 79 lenders reported refinance shares. NMN adds up the total refinance dollar volume across these lenders and divides by the total origination volume of the same group of lenders to arrive at the refinance share. Of the four data sources, HMDA is likely to present the most accurate assessment of the national refinance share in the single-family origination market because of the large number of lenders that are covered. However, HMDA is neither a census of single-family lending nor a random sample of activity, but a mandatory regulatory report for a prescribed set of loans and institutions; Berkovec and Zorn (1996) found substantial variability in coverage across census tracts and lenders. Further, refinance shares differ across the four data sources. This paper addresses the accuracy and national representativeness of HMDA by comparing the refinance share across the four data sources and identifying the causes of any differences. For comparison purposes, the PMMS and MBA refinance share series were converted into quarterly series by taking the average of monthly (for the PMMS) or weekly (for the MBA) refinance shares to comprise the quarter. In general, the four series track each other closely. 5 To illustrate the historical refinance trends, Exhibit 1 plots the calculated quarterly refinance share from HMDA for the period of 1990 to 2005. The refinance booms of 1992 1993, 1998, and 2001 2003 are clear in the time series. Exhibit 2 graphs the quarterly difference between the HMDA refinance share and the share in each of the other three data sources. Close examination reveals some systematic differences in refinance shares across the data sources. The refinance share observed in HMDA was consistently higher than that reported by the PMMS or MBA. For example, based on quarterly data between 1990 and 2005, the difference between HMDA and the PMMS ran between 9 to 17 percentage points, with an average 5.4 percentage point higher share in HMDA; the PMMS refinance share was higher in only 11 of the 64 quarterly observations. The difference between HMDA and the MBA series had a range of 7 to 21 percentage points, the HMDA share averaged 7.4 percentage points higher, and the MBA was larger in only 4 quarters. NMN also reported a lower share than HMDA in the late 1990s, but after 2001 the difference had practically disappeared and the two series were almost identical; the average JRER Vol. 29 No. 4 2007

514 Chang and Nothaft 90 80 70 60 50 40 30 20 10 0 Exhibit 1 HMDA Refinance Series by Quarter: 1990 2005 1990 Q1 1990 Q4 1991 Q3 1992 Q2 1993 Q1 1993 Q4 1994 Q3 1995 Q2 1996 Q1 1996 Q4 1997 Q3 1998 Q2 1999 Q1 1999 Q4 2000 Q3 2001 Q2 2002 Q1 2002 Q4 2003 Q3 2004 Q2 2005 Q1 2005 Q4 Percent

Demystifying the Refi-Share Mystery 515 25 20 15 10 5 0-5 -10 Exhibit 2 Difference in Reported Refinance Shares between HMDA and Other Series by Quarter: 1990 2005 1990 Q1 1990 Q3 1991 Q1 1991 Q3 1992 Q1 1992 Q3 1993 Q1 1993 Q3 1994 Q1 1994 Q3 1995 Q1 1995 Q3 1996 Q1 1996 Q3 1997 Q1 1997 Q3 1998 Q1 1998 Q3 1999 Q1 1999 Q3 2000 Q1 2000 Q3 2001 Q1 2001 Q3 2002 Q1 2002 Q3 2003 Q1 2003 Q3 2004 Q1 2004 Q3 2005 Q1 2005 Q3 Percent HMDA - PMMS HMDA - MBA HMDA - NMN JRER Vol. 29 No. 4 2007

516 Chang and Nothaft difference between the two series from the fourth quarter of 1998 to the fourth quarter of 2005 was 2.2 percentage points, but after the fourth quarter of 2001, the average difference was only 0.1 percentage points. These differences are large enough to substantially effect estimates of home equity cash-out volume and overall gross credit flows. For example, Greenspan and Kennedy (2005) rely on HMDA s estimates of refinance share by lender type to estimate overall origination patterns. If HMDA s refinance shares are as much as 20% too high, then their estimates of gross home mortgage flows may be more than 10% too high during a refinance boom (a period where refinance shares of lending exceed 50%); for 2003, their origination estimate could be $0.5 trillion too large (about one-quarter too large) if HMDA significantly overstated the refinance share. HMDA s refinance share is also an important component to the estimation of cash-out volume in Freddie Mac s Cash-Out Refinance Report. A 20% overstatement in the refinance share could result in $90 billion overestimation of total home equity cashed out in 2005 (an estimate that could be one-third too large). Alternative refinance shares also affect estimates of changes in cash-out volume as well: the quarterly change in cash-out volume was reduced by as much as $9.6 billion in one quarter (which would change the original estimate from a $9.4 billion growth to a $0.2 billion decline), or increased by as much as $6.6 billion in another quarter (nearly three times the original estimate) if the PMMS refinance share was used in place of the HMDA refinance share. In the case that refinance shares are used as an input to measure mortgage industry labor needs, an estimation error in the range of 20% could result in an over- or under-employment of nearly 10,000 in the industry. 6 It is the purpose of this paper to investigate the various sources that may lead to the observed differences. The study contributes to a better understanding of the different refinance series, reconciliation between reported shares, and provides a guide for the proper application of the appropriate series for various purposes. Analysis The analysis evaluates the methodological differences in refinance share calculation, from the sample universe to the computation of aggregate refinance shares. This section examines hypotheses that may explain the disparity in refinance share estimates, tests the validity of each hypothesis, and quantifies the effect of each factor toward explaining the refinance share differences. Application vs. Origination Series The most apparent discrepancy lies in the stage of the mortgage lending process captured by the different series. The PMMS and MBA survey the refinance percentage of mortgage applications. The HMDA and NMN focus on the refinance percentage of mortgage originations. Home purchase and refinance mortgage applications may be subject to different loan approval and reporting processes,

Demystifying the Refi-Share Mystery 517 thus resulting in a difference in refinance shares measured between applications and originations. 7 Specifically, three types of difference exist: (1) difference in fallout rates of purchase versus refinance applications, or the probability of an application resulting in origination; (2) difference in the timing of report, because a mortgage is included in the calculation of refinance share of applications before it is counted in the origination refinance series, usually one or two months earlier; and (3) difference in the length of time from application to closing for purchase and refinance applications, which could mean that some purchase and refinance loans are originated in the same quarter, yet with their applications made in different quarters (or vice versa). Fallout Rates. One hypothesis is that purchase money and refinance applications have a different fallout rate from application to origination. If refinance applications are more likely to proceed to loan settlement, then the refinance share of applications will be lower than the refinance share of originations. There are at least three causes for different fall-out rates: (1) the home-purchase contract may fall through (e.g., the home fails inspection or buyer changes mind; for refinance, there is no home purchase contract); (2) purchase-money applications may be rejected at a higher or lower rate than those for refinance due to underwriting considerations; 8 and (3) some borrowers may put in multiple applications with different lenders before choosing one before settlement (more likely during periods of heightened rate volatility and for refinance). On net, it is uncertain whether refinance or purchase-money applications have a higher fallout rate before closing. To test this hypothesis, HMDA data is used to compare the fallout rates between home purchase and refinance loans by year of application (Exhibit 3). There is no consistent pattern, although the fallout rate for refinance applications tends to be higher in most years. This suggests that the refinance share of applications should, if anything, be higher than the refinance share of originations. HMDA data tend to support this conclusion. To illustrate, Exhibit 4 reports the refinance share of originations and the refinance share of unsuccessful applications (i.e., applications that do not result in an origination). While the pattern varies by year, generally the refinance share of applications tends to be greater than the refinance share of originations. Thus, differential fallout rates cannot explain the observed discrepancy between the PMMS and MBA series and the HMDA and NMN series. Report Timing. Since it usually takes one to two months for a mortgage to complete the process from application to origination, the loan application refinance series should lead the loan origination series by approximately the same amount of time. In the case of quarterly time series, this timing difference should result in a lead of perhaps one half quarter. This should be more pronounced right before and after a change in the refinance environment, such as a refinance boom, with the application series showing elevated refinance rates relative to the origination series in the early stage of the boom, and a drop in refinance application levels before the origination series as the boom nears its end. JRER Vol. 29 No. 4 2007

518 Chang and Nothaft Exhibit 3 Fallout Rates from Application to Origination (Percent by Dollar Volume) All Loans Conventional Prime Loans Year Refinance Purchase Refinance Purchase 1990 37 30 37 29 1991 34 30 34 29 1992 29 28 28 26 1993 24 25 24 26 1994 35 27 32 23 1995 38 30 31 24 1996 42 32 35 25 1997 44 33 31 24 1998 36 33 24 23 1999 49 34 34 25 2000 58 35 41 26 2001 39 30 27 24 2002 36 28 25 24 2003 36 31 26 26 2004 50 35 36 30 2005 51 39 38 34 The higher frequency of the PMMS (monthly) and MBA (weekly) data allow construction of quarterly refinance shares that incorporate either a one-month or a two-month lead of applications over originations to test the hypothesis that the relatively large quarterly differences shown in Exhibit 2 may be attributable to this factor. This results in a mitigation of the differences between HMDA and the two application series in some quarters, such as the first quarter of 2002 and the third quarter of 2003, where a large drop in applications (but not originations) occurred. The third quarter of 2003 showed the most significant improvement: With a two-month lead, the difference is reduced from 15 to 0 percentage points with the PMMS data and from 19 to 1 percentage point with the MBA data. However, in some other quarters, leading the application series magnifies the differences instead of reducing them. While the application series lead the originations in decline at the end of a refinance boom, the two series tend to go up by about the same magnitude at the same time from the start to the peak of the boom. Therefore, leading the application series causes it to trail the origination series when there is a rapid increase in refinance share.

Demystifying the Refi-Share Mystery 519 Exhibit 4 Refinance Share of Loan Originations and Applications (Percent by Dollar Volume) All Loans Conventional Prime Loans Year Originations Unsuccessful Applications Total Applications Originations Unsuccessful Applications Total Applications 1990 25 32 27 28 36 31 1991 40 45 41 46 51 47 1992 61 62 62 66 68 66 1993 65 60 64 68 64 68 1994 39 48 42 40 51 43 1995 31 39 33 33 41 35 1996 37 47 41 39 51 42 1997 38 51 43 39 47 41 1998 58 61 59 61 62 61 1999 43 58 49 45 56 48 2000 28 50 38 27 42 31 2001 61 70 64 64 67 65 2002 66 74 69 68 70 69 2003 72 77 74 74 73 74 2004 52 68 59 51 57 53 2005 48 61 54 48 52 49 Overall, adjusting for the lead time does not account for the differences in refinance share between the application and origination series. The average quarterly difference from 1990:Q1 to 2005:Q4 between the HMDA and PMMS refinance series was 5.42 percentage points. When the PMMS series was led by one month, the average difference from the resulting series was 5.52 percentage points. Leading the PMMS by two months resulted in an average difference of 5.59 percentage points. Repeating the same exercise with the MBA refinance shares, the average quarterly differences between the HMDA and MBA refinance percentages, MBA led by one month, and MBA led by two months, and were 7.35%, 7.44%, and 7.57%, respectively. 9 Therefore, the effect of the mismatch of report timing is inconsistent through time, insignificant overall, and cannot explain the observed differences in refinance share between the PMMS and MBA series and the HMDA and NMN series. Time to Close. Purchase-money and refinance applications may exhibit different lengths of time from application to settlement. Thus, the origination series will include purchase and refinance loans whose applications were initiated in different JRER Vol. 29 No. 4 2007

520 Chang and Nothaft quarters. Specifically, refinance applications are hypothesized to take less time to settle than purchase money applications for several reasons: (1) refinance applicants may have more complete and better qualifying documentation as a result of their previous purchase or refinance experience; (2) streamline refinance programs are designed to shorten the time a refinance application spends waiting for underwriting approval and thus closing (e.g., the program may allow use of a recent appraisal); (3) home sellers and buyers may negotiate longer periods between contract and settlement to accommodate moving and other schedules; and (4) home builders sometimes offer extended rate-lock programs to assist buyers in locking in favorable interest rates. If refinance applicants do settle more quickly, the difference between the origination and application series should be most prominent in the fourth quarter because, according to HMDA reporting regulations, loans whose decisions are still pending at year end should be excluded from the current year (the calendar year the application was placed) but included in the next calendar year s report. If the underwriting decisions for purchasemoney loans are disproportionately delayed to the next year s reporting, higher refinance percentages should be observed in the fourth quarter with HMDA than MBA or PMMS. A regression with the HMDA series as dependent variable and the PMMS (or MBA) series as independent variable, together with quarterly dummy variables, can test this hypothesis. As shown in Exhibit 5, none of the indicators for first, second, or third quarters were significant at the 5% level when HMDA and PMMS were compared, while a comparison between HMDA and MBA found the first and second quarter dummy variables were significant and positive, meaning that HMDA exceeded MBA by less in the fourth quarter than in the first or second quarters. 10 This contradicts the assumption about the effect of time-to-close between home purchase and refinance loans. Overall, the findings suggest that either the time it takes from loan application to origination does not make a Exhibit 5 Regression Tests for Quarterly Differences: 1990 2005 Estimate p-value Estimate p-value Intercept 0.076 0.001 Intercept 0.084 0.001 PMMS Refi 0.924 0.000 MBA Refi 0.926 0.000 1 st Quarter Dummy 0.022 0.258 1 st Quarter Dummy 0.040 0.050 2 nd Quarter Dummy 0.024 0.225 2 nd Quarter Dummy 0.046 0.027 3 rd Quarter Dummy 0.014 0.452 3 rd Quarter Dummy 0.012 0.553 R 2 0.901 R 2 0.891 Note: The dependent variable is the HMDA Refi Series. The number of observations is 64.

Demystifying the Refi-Share Mystery 521 difference in refinance share calculations, or it actually takes longer for a refinance loan to close than a home purchase loan, which results in lower fourth quarter differences than first and second quarter differences. Sample Coverage There are inherent differences between each series in lender coverage. PMMS samples both national and local lenders in compiling its results, weights regional averages by dollar volume of lending to compute a national average, but does not weight individual lenders (i.e., within a region, each respondent is assigned the same weight). NMN reports the cumulative refinance percentage, based on total refinance dollar volume divided by total origination volume summed across the 60 to 100 lenders that report their refinance originations from the list of top 100 lenders in origination volume for each quarter. HMDA is neither a census nor a random sample but a regulatory report required of covered lenders for covered loans; the coverage rules exempt small lenders and lenders that do not lend in MSAs, and covers purchase money loans, home-improvement loans, and refinances of purchase money and home improvement loans. 11 The HMDA coverage requirement that a lender needs to have a branch in an MSA makes it likely that HMDA has a higher percentage of lending activity inside MSAs. The geographic location of borrowers will affect their likelihood of refinance. Home values in MSAs generally are much higher than in rural areas, with higher loan balances; higher balance loans typically refinance more frequently, in part because the payment savings are more substantial for a given decline in interest rates (Deng, Quigley, and Van Order, 2000). In addition, the greater number of lenders and competition in MSAs may increase borrower awareness of refinance opportunities. The national representativeness of refinance shares computed from HMDA will be affected if there is a substantial difference in coverage of MSA and non-msa areas and if there are significantly different refinance rates inside and outside MSAs. First, there is a test of whether HMDA over-represents MSA originations relative to national surveys based on random samples. Second, there is an examination of whether the refinance shares differ in MSA and non-msa areas. The Residential Finance Survey (RFS) and the American Housing Survey (AHS) were used to compare MSA versus non-msa origination volumes. The 2001 RFS was used to measure the percentage of loans originated inside an MSA for 1998, 1999, 2000, and 2001. These were compared with the percentage of MSA originations from HMDA. Single-family residences that are owner-occupied or rental (or vacant) are included in the RFS calculations. The findings reveal that there is no significant difference between the MSA shares in HMDA and the RFS, as shown in Exhibit 6. The AHS differs from the RFS in that mortgage information is only available for owner-occupied homes and is collected more frequently (every two years). JRER Vol. 29 No. 4 2007

Panel A: Results from the 2001 RFS Year All: Owner First Lien First, Second and Third Combined Exhibit 6 Percentage of Loan Originations Inside MSAs by Origination Year All: Owner and Renter First Lien First, Second and Third Combined Prime Conventional: Owner First Lien First, Second and Third Combined Prime Conventional: Owner and Renter First Lien 1998 88.9 88.8 88.9 88.8 88.9 89.0 88.8 88.9 1999 90.7 90.7 89.7 89.8 91.5 91.4 90.2 90.2 2000 90.1 89.9 89.8 89.5 90.5 90.3 90.2 89.9 2001 89.3 89.2 89.2 89.1 88.6 88.6 89.0 89.0 First, Second and Third Combined 522 Chang and Nothaft Panel B: Results from the 2001, 2003, and 2005 AHS 2001 AHS 2003 AHS 2005 AHS Year First Lien First and Second Combined First Lien First and Second Combined First Lien First and Second Combined 1999 87.2 87.1 N/A N/A N/A N/A 2000 86.0 86.0 N/A N/A N/A N/A 2001 89.1 89.0 85.7 85.8 N/A N/A 2002 N/A N/A 87.6 87.5 N/A N/A 2003 N/A N/A 89.6 89.7 85.5 85.5 2004 N/A N/A N/A N/A 81.5 81.8 2005 N/A N/A N/A N/A 78.6 79.0

Exhibit 6 (continued) Percentage of Loan Originations Inside MSAs by Origination Year Panel C: Results from HMDA JRER Vol. 29 No. 4 2007 Year All Originations Applications Conventional Prime Originations Applications Owner Occupied Only Originations Applications 1992 86.9 84.9 86.4 82.5 90.2 87.0 1993 86.7 83.0 86.9 79.8 90.1 84.8 1994 85.3 80.4 85.4 78.2 89.2 82.8 1995 84.0 76.5 84.2 74.5 88.6 78.9 1996 88.4 81.9 88.0 79.0 90.0 80.6 1997 89.1 81.3 88.5 78.9 89.4 79.2 1998 89.9 83.8 89.5 80.7 90.4 80.8 1999 89.7 86.6 89.2 81.6 90.1 81.7 2000 89.8 86.9 89.4 81.0 90.4 81.2 2001 90.6 89.1 90.7 85.8 91.7 86.1 2002 91.5 90.2 91.6 88.7 92.7 89.1 2003 91.4 90.5 91.5 89.5 92.5 90.0 2004 92.4 89.7 92.3 90.6 93.2 89.9 2005 92.4 87.8 92.3 86.7 93.3 88.1 Demystifying the Refi-Share Mystery 523

524 Chang and Nothaft Questions regarding whether the current mortgage is a refinance of a prior mortgage were only collected starting in the 2001 survey and was asked for first and second liens only. According to the RFS, the MSA percentage of originations can have a wide difference between the owner-occupied and non-owner occupied homes (in part, because homeownership rates are higher in non-metro areas). Therefore, the AHS results were compared with the subset of owner-occupied loans reported in HMDA. AHS data for 2001, 2003, and 2005 were used to find the MSA share of total originations for owner-occupied properties from 1999 to 2005. Comparison with HMDA results reveals a higher percentage of non-msa loans in the AHS than in HMDA (see Exhibit 6). However, confounding the AHS analysis is that the publicly available AHS microdata use 1983 definitions of metropolitan area boundaries. These boundaries are based on population and commuting patterns reported in the 1980 census. While the effect of old (and smaller) MSA boundaries on the MSA share of originations is uncertain, it is possible that the MSA share is lower using the 1983 designations rather than current definitions. The difference of MSA originations as a percentage of total originations runs between 3% and 6%. Over-representation of MSA lending in HMDA (as suggested by the AHS) could make HMDA refinance shares higher than other sources if refinance shares are significantly higher within MSAs. To examine this, the refinance percentages of mortgages originated inside and outside an MSA were calculated separately using both the RFS and AHS data (Exhibit 7). In general, the refinance percentages reported by the RFS were lower and had less variation across years compared to those reported by the AHS and HMDA. No consistent pattern emerges from the RFS: Higher refinance percentages were found outside MSAs in 2000 and 2001; 1998 and 1999 show the reverse. In contrast, results from the AHS show that refinance percentages are consistently higher inside MSAs, compared to loans on homes outside of MSAs. The difference in refinance percentage runs from 0.6% in 2001 to 28% in 2004. The effect of HMDA s coverage regulations should be to over-represent MSA lending. If refinance shares are the same within and outside of MSAs, as suggested by the RFS, then the coverage rule will have no material effect on HMDA s reported refinance share for the U.S. If refinance shares are higher within MSAs, as indicated by the AHS, then it is possible that some of the higher refinance share in HMDA reflects the coverage rule. Even so, the refinance share differences found in the AHS for 1999 2003, if used to re-weight HMDA, would reduce the HMDA refinance share by less than one percentage point (since about 85% to 90% of lending was within MSAs). Thus, this implies that differences in MSA and non-msa coverage, if any, is unlikely to explain the refinance share differences between HMDA and other data sources. In fact, the analysis supports the conclusion that the refinance share computed from HMDA is nationally representative.

Exhibit 7 Comparison of Refinance Shares Inside- and Outside-MSAs Panel A: Results from the 2001 RFS JRER Vol. 29 No. 4 2007 All: Owner First Lien First, Second and Third Combined All: Owner and Renter First Lien First, Second and Third Combined Year Inside MSA Outside MSA Inside MSA Outside MSA Inside MSA Outside MSA Inside MSA Outside MSA 1998 27.6 23.9 27.8 24.0 26.2 24.4 26.4 24.5 1999 25.4 21.8 25.6 22.4 25.9 19.7 25.9 20.3 2000 17.4 24.1 18.5 25.7 18.8 20.9 19.6 22.2 2001 39.1 41.3 38.3 40.8 38.1 37.5 37.4 37.3 Prime Conventional: Owner First Lien First, Second and Third Combined Prime Conventional: Owner and Renter First Lien First, Second and Third Combined Year Inside MSA Outside MSA Inside MSA Outside MSA Inside MSA Outside MSA Inside MSA Outside MSA 1998 29.5 26.9 29.7 26.6 27.8 27.1 28.0 26.8 1999 26.7 24.9 26.8 25.4 27.0 21.5 27.0 22.1 2000 17.5 24.7 18.5 26.5 19.4 21.6 20.0 23.0 2001 40.6 45.3 39.9 44.7 39.3 42.3 38.7 42.2 Demystifying the Refi-Share Mystery 525

Exhibit 7 (continued) Comparison of Refinance Shares Inside- and Outside-MSAs Panel B: Results from 2001, 2003, and 2005 AHS First Lien 2001 AHS 2003 AHS 2005 AHS 526 Chang and Nothaft Year Inside MSA Outside MSA Inside MSA Outside MSA Inside MSA Outside MSA 1999 32.4 28.2 N/A N/A N/A N/A 2000 20.3 18.9 N/A N/A N/A N/A 2001 44.2 42.9 41.2 40.7 N/A N/A 2002 N/A N/A 53.7 42.7 N/A N/A 2003 N/A N/A 73.3 69.4 57.9 41.3 2004 N/A N/A N/A N/A 49.9 21.5 2005 N/A N/A N/A N/A 46.5 20.9

JRER Vol. 29 No. 4 2007 Exhibit 7 (continued) Comparison of Refinance Shares Inside- and Outside-MSAs Panel B: Results from 2001, 2003, and 2005 AHS (continued) First and Second Combined 2001 AHS 2003 AHS 2005 AHS Year Inside MSA Outside MSA Inside MSA Outside MSA Inside MSA Outside MSA 1999 31.0 27.0 N/A N/A N/A N/A 2000 19.5 18.2 N/A N/A N/A N/A 2001 42.6 40.5 40.3 39.5 N/A N/A 2002 N/A N/A 52.5 42.7 N/A N/A 2003 N/A N/A 72.4 68.0 56.8 40.8 2004 N/A N/A N/A N/A 48.1 21.0 2005 N/A N/A N/A N/A 43.5 20.1 Demystifying the Refi-Share Mystery 527

528 Chang and Nothaft Lender Size and Computation Methods If refinance share varies significantly by lender size, then data sources that weight lenders refinance shares by volume weights will provide better estimates of the national refinance share. Even though the PMMS weights regional averages by dollar volume weights, within region each lender is assigned the same weight; this can impart a downward bias to the PMMS share if large lenders have higher refinance shares. Likewise, HMDA coverage rules that exempt small lenders can result in a refinance share that is higher than in the overall market, although this effect is likely very limited because small lenders account for relatively little origination volume. In HMDA and NMN, the refinance percentage was calculated by dividing the total refinance dollars by the total origination dollars, or: Re ƒii Prefi, (1) Originations i where i represents individual lenders, and Refi and Originations refer to the dollar amounts for the period calculated. For the PMMS, Freddie Mac computes the unweighted average of the refinance percentages across lenders within each region, then dollar weights each regional average. In other words, the sum of reported refinance percentages is divided by the total number of lenders within the region: P refi,i Prefi, (2) n where P refi,i is the i th lender s refinance share and n is the number of respondents for that region, and then each regional average is weighted by the region s share of U.S. dollar originations; the five weighted means are summed to arrive at the national average. In the PMMS, large and small lenders have equal weight within region. If there is positive correlation between lender volume and refinance percentage, then the PMMS will understate refinance percentages even if the composition of lender sizes is the same as in HMDA. Refinance Share and Lender Size. HMDA is used to identify the relationship between lender origination volume and refinance percentage. As shown in Exhibit 8, there is substantial variation in refinance share by major segment of the singlefamily origination market: prime conventional conforming, jumbo, governmentinsured (FHA, VA, and RHS), subprime, and manufactured housing. 12 Over 1990

Exhibit 8 HMDA Refinance Percentage by Loan Type JRER Vol. 29 No. 4 2007 100 90 80 70 60 50 40 30 20 10 0 Percent Jumbo Conventional Conforming Government Insured Subprime Manufactured Housing 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Demystifying the Refi-Share Mystery 529

530 Chang and Nothaft 2005, the prime conventional conforming and jumbo refinance shares traced a similar pattern, averaging about 50% with clear peaks (1992 1993, 1998, and 2001 2003 refinance booms) and troughs. The government-insured segment also has a refinance share that is driven by the level of interest rates, but the share averaged 26% because of the home purchase and low down payment emphasis of the programs. Subprime has historically been a refinance product with an average 63% refinance share and is less interest rate sensitive. Likewise, the manufactured housing segment also is less market rate sensitive, and the refinance share averaged only 22% over the period. The refinance share of total originations by five types of financial institutions: commercial banks, savings institutions, credit unions, independent mortgage companies, and mortgage company subsidiaries of depositories was also examined. The refinance shares by year were highly correlated with small differences across lender types. The average refinance percentage over 1990 2005 was 53% for credit unions, 51% for savings institutions, 49% for commercial banks, and 46% for independent mortgage companies and subsidiary mortgage companies. Since the MBA survey excludes credit unions, this could also explain some of the difference between HMDA refinance rates and the MBA percentages, though the effect may be limited due to the small volume of credit union mortgage originations. Refinance shares by Census Bureau regions, as shown in Exhibit 9, have a similar pattern over time and clearly reflect the major refinance booms. However, the refinance share levels differ. Shares were generally the highest in the West and averaged 53%. Shares were consistently lowest in the South and averaged 41%. These differences partly reflect the substantially higher home prices and loan balances in the West, and smaller loan balances and greater manufactured housing activity in the South. In addition, within region, some states and local jurisdictions have significant mortgage recordation fees that tend to reduce refinance shares. Because a lender s refinance share is affected by type of product segment, lender type, and location of activity, an OLS regression with a set of explanatory variables including lender size is used to test for the relationship between size and refinance percentage across individual lenders. To preserve homogeneity in the sample, lenders identified as specialized in subprime or manufactured housing lending were excluded. Both the refinance percentage of originations and the logit transformation of the refinance percentage of originations by lender were used as the dependent variable, as follows: P refi,i yi Log. (3) 1 P refi,i The results obtained were similar. Only the results using the refinance percentage as the dependent variable are reported in the tables.

Exhibit 9 HMDA Refinance Percentage by Region JRER Vol. 29 No. 4 2007 100 90 80 70 60 50 40 30 20 10 0 (Percent) South MidWest NorthEast West 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Demystifying the Refi-Share Mystery 531

532 Chang and Nothaft Two extreme years were modeled: 2000 because it had the lowest refinance percentage in a decade, and 2003 because it had the highest over the 16-year period. Two alternative dependent variables were used: one is refinance share by lender, defined as individual HMDA filers, and the second is refinance share for lender-state combinations (i.e., when a lender originates in several states, it was considered as a separate lending organization in each state). A distinction is made between the two dependent variables because while corporate characteristics and policies affect the originations at an overall lender level, individual markets differ in state laws and restrictions, local housing market conditions, strength of lender representation in the particular area, and even borrower attitude toward refinance. The hypothesized relationship between lender size and refinance percentage across lenders on the national level is examined first. The explanatory variables include dummy variables designating lender type: savings institution, commercial bank, credit union, or independent mortgage company; minority- or women-owned institution; and lender size proxied by the natural log of total origination volume in the year. 13 Because of the different refinance shares between prime conventional and government-insured mortgages, OLS is performed on the two types of loans separately. Excluded from the sample are subprime and manufactured housing loans. The focus is on loans made inside MSA areas only. The mean values of the variables included can be found in Exhibit 10. Regression results for conventional prime mortgages for both 2000 and 2003 are in Exhibit 11. 14 For 2003, refinance shares of both the conventional and government-insured mortgage originations are positively correlated with lender size, indicating that larger lenders, defined by higher origination volume on a national level, do have higher refinance percentage than smaller lenders in both conventional and government-insured markets. While credit unions have a significantly positive coefficient in both regressions as well, meaning that they issue a significantly higher percentage of refinance mortgages than mortgage company subsidiaries, no other factors are significant in the government-insured regression. For conventional loans, savings banks have significantly higher refinance percentages than the mortgage company subsidiaries, and independent mortgage companies have significantly lower percentages. Institutions owned by minorities or women have lower refinance percentages than others. Results from the same regression applied to 2000 HMDA data show that for conventional originations, the lender size effect is smaller but significantly positive; for government-insured originations, larger lenders are associated with lower refinance percentages. No other explanatory variables are significant in the government-insured refinance percentages. Commercial banks and credit unions are shown to have significantly higher refinance rate than the subsidiary mortgage companies in the conventional mortgage originations. Switching the dependent variable to lender-state combinations, a series of state dummy variables were created to designate each of the 50 states and the District of Columbia; the California dummy was excluded as a regressor. Other variables

JRER Vol. 29 No. 4 2007 Exhibit 10 Means of Regression Variables 2000 2003 Lender Volume ($ in millions) 108.9 Lending Institution Volume ($mil) 35.1 Lender Volume ($ in millions) 425.6 Lending Institution Volume ($mil) Savings Bank 15% Savings Bank 15% Savings Bank 14% Savings Bank 15% Commercial Bank 49% Commercial Bank 36% Commercial Bank 48% Commercial Bank 36% Independent Mortgage Co. 10% Independent Mortgage Co. 22% Independent Mortgage Co. 10% Independent Mortgage Co. 23% Credit Union 23% Credit Union 15% Credit Union 25% Credit Union 16% Minority or Women Owned 1% Minority or Women Owned 1% Minority or Women Owned 1% Minority or Women Owned Refi Share 29% Refi Share 29% Refi Share 61% Refi Share 63% # of Observations 7,277 Number of Observations 22,237 # of Observations 6,942 # of Observations 26,457 110.7 1% Demystifying the Refi-Share Mystery 533

Panel A: 2000 HMDA Conventional Prime Mortgages Exhibit 11 Regression Results Testing for Lender Size Effect on Refinance Share Dependent Variable: Lender Refi Share Dependent Variable: Lender-State Refi Share Dependent Variable: Lender-State Refi Share Variable Estimate p-value Variable Estimate p-value Variable Estimate p-value Intercept 0.22 0.0001 Intercept 0.24 0.0001 Intercept 0.22 0.0001 Size: Total Origination 0.004 0.01 Size: Total Origination 0.002 0.01 Size: State Origination 0.01 0.0001 Savings Bank 0.02 0.32 Savings Bank 0.01 0.08 Savings Bank 0.01 0.07 Commercial Bank 0.04 0.00 Commercial Bank 0.02 0.01 Commercial Bank 0.02 0.00 Independent Mrtg Co. 0.02 0.15 Independent Mrtg Co. 0.05 0.0001 Independent Mrtg Co. 0.05 0.0001 Credit Union 0.09 0.0001 Credit Union 0.01 0.35 Credit Union 0.01 0.20 MWO* 0.03 0.17 MWO 0.10 0.0001 MWO 0.10 0.0001 # of Observations 7,277 # of Observations 22,237 # of Observations 22,237 R 2 0.0206 R 2 0.0379 R 2 0.0397 F-test for All State Dummies 16.13 0.0001 F-test for All State Dummies 16.87 0.0001 534 Chang and Nothaft * MWO: Minority or women-owned institution.

Exhibit 11 (continued) Regression Results Testing for Lender Size Effect on Refinance Share JRER Vol. 29 No. 4 2007 Panel B: 2003 HMDA Conventional Prime Mortgages Dependent Variable: Lender Refi Share Dependent Variable: Lender-State Refi Share Dependent Variable: Lender-State Refi Share Variable Estimate p-value Variable Estimate p-value Variable Estimate p-value Intercept 0.07 0.01 Intercept 0.30 0.0001 Intercept 0.45 0.0001 Size: Total Origination 0.05 0.0001 Size: Total Origination 0.03 0.0001 Size: State Origination 0.02 0.0001 Savings Bank 0.06 0.00 Savings Bank 0.04 0.0001 Savings Bank 0.01 0.33 Commercial Bank 0.02 0.40 Commercial Bank 0.03 0.0001 Commercial Bank 0.08 0.0001 Independent Mrtg Co. 0.11 0.0001 Independent Mrtg Co. 0.01 0.17 Independent Mrtg Co. 0.03 0.00 Credit Union 0.16 0.0001 Credit Union 0.10 0.0001 Credit Union 0.03 0.0001 MWO* 0.00 0.91 MWO 0.02 0.49 MWO 0.02 0.35 # of Observations 6,942 # of Observations 26,457 # of Observations 26,457 R 2 0.222 R 2 0.083 R 2 0.082 Note: * MWO: Minority or women-owned institution. F-test for All State Dummies 18.16 0.0001 F-test for All State Dummies 16.08 0.0001 Demystifying the Refi-Share Mystery 535

536 Chang and Nothaft are the same as those in the lender regressions. Two indicators are used as proxies for lender size: one is the natural log of total originations for the lender at national level, the other is the natural log of total originations made by a lender in the particular state considered. Thus, the separate effects of lender size are examined as a whole, along with its intensity of activities in the local market on the refinance share of a lender-state. The results for the primary regressors and prime conventional originations are reported in Exhibit 11. 15 For lending organizations in the regression using 2003 HMDA data, both the size variables still have a significantly positive effect on the lender-state refinance shares for both conventional and government-insured mortgages. Both the nationwide origination volume and state-specific origination volume are significantly positive in the regression using 2000 data as well, but to a much less degree (Exhibit 12). The effects on the government insured mortgages are negative for both measures for year 2000. The results confirm the hypothesis that larger institutions, proxied by origination volumes, have a higher percentage of refinance mortgages in their total originations. Though the exact reason behind the observed relationship between refinance share and loan origination volume remains to be explored, it is hypothesized that the institutions that have larger holdings of mortgage-retained portfolios or mortgageservicing portfolios have more incentives to maintain the size of their portfolio and thus devote more resources to encourage their existing customers to refinance with them or pursue other refinance borrowers to supplement their portfolio. To test the hypothesis that lender mortgage portfolio size is positively correlated with their refinance share, the dollar volume of the single-family mortgage holdings and the dollar amount of total single-family mortgages serviced were obtained from the Reports of Condition and Income filed by commercial banks and savings institutions. Using each institution s Federal Reserve System ID, the portfolio information as of the prior year s end for each individual bank was then matched to loan origination information from HMDA. This match results in 3,441 institutions with non-missing or non-zero mortgage portfolios and 1,045 institutions with non-zero or non-missing servicing portfolios in 2000, and 3,700 institutions with non-missing or non-zero mortgage portfolios and 992 institutions with non-missing or non-zero servicing portfolios in 2003. 16 Similar regression models are estimated as before with total refinance shares and lender-state combinations as dependent variables. The models include the log transformation of total mortgage-holding portfolio and mortgage-servicing portfolio as independent variables. The results on the main variables are presented in Exhibit 13. 17 The sign and significance on both the mortgage-holding and mortgage-servicing portfolio sizes for both 2000 and 2003 confirm the hypothesis that lenders with larger mortgage portfolios originate larger shares of refinance loans. To further separate the portfolio composition effect from the size effect, the regression analysis was repeated for prime conventional conforming loans with the results

Exhibit 12 Regression Results Testing for Lender Portfolio Size Effect on Refinance Share Panel A: 2000 HMDA Conventional Prime Mortgages JRER Vol. 29 No. 4 2007 Dependent Variable: Lender Refi Share Dependent Variable: Lender Refi Share Dependent Variable: Lender-State Refi Share Dependent Variable: Lender-State Refi Share Variable Estimate Variable Estimate Variable Estimate Variable Estimate Intercept 0.282 Intercept 0.212 Intercept 0.108 Intercept 0.136 (0.001) (0.001) (0.001) (0.001) Total Serving 0.001 Total Mortgage 0.005 Total Serving 0.007 Total Mortgage 0.005 Portfolio (0.73) Portfolio (0.02) Portfolio (0.001) Portfolio (0.001) Savings Bank 0.023 Savings Bank 0.001 Savings Bank 0.019 Savings Bank 0.014 (0.28) (0.98) (0.32) (0.37) Commercial Bank 0.039 Commercial Bank 0.032 Commercial Bank 0.066 Commercial Bank 0.040 (0.05) (0.07) (0.001) (0.001) Minority or 0.022 Minority or 0.043 Minority or 0.018 Minority or 0.037 Women Owned (0.56) Women Owned (0.07) Women Owned (0.69) Women Owned (0.21) # of Observations 3,441 # of Observations 1,045 # of Observations 8,533 # of Observations 4,147 R 2 0.03 R 2 0.01 R 2 0.07 R 2 0.05 F-test for All State 1.44 F-test for All State 1.44 Dummies (0.025) Dummies (0.025) Demystifying the Refi-Share Mystery 537

Panel B: 2003 HMDA Conventional Prime Mortgages Dependent Variable: Lender Refi Share Exhibit 12 (continued) Regression Results Testing for Lender Portfolio Size Effect on Refinance Share Dependent Variable: Lender Refi Share Dependent Variable: Lender-State Refi Share Dependent Variable: Lender-State Refi Share Variable Estimate Variable Estimate Variable Estimate Variable Estimate Intercept 0.639 Intercept 0.221 Intercept 0.590 Intercept 0.437) (0.001) (0.001) (0.001) (0.001 Total Serving 0.008 Total Mortgage 0.041 Total Serving 0.010 Total Mortgage 0.019 Portfolio (0.001) Portfolio (0.001) Portfolio (0.001) Portfolio (0.001) Savings Bank 0.039 Savings Bank 0.023 Savings Bank 0.074 Savings Bank 0.061 (0.12) (0.34) (0.001) (0.001) Commercial Bank 0.025 Commercial Bank 0.107 Commercial Bank 0.055 Commercial Bank 0.099 (0.28) (0.001) (0.001) (0.001) Minority or 0.092 Minority or 0.041 Minority or 0.001 Minority or 0.038 Women Owned (0.03) Women Owned (0.13) Women Owned (0.98) Women Owned (0.22) # of Observations 3,700 # of Observations 992 # of Observations 11,315 # of Observations 4,991 R 2 0.04 R 2 0.12 R 2 0.05 R 2 0.09 F-test for All State 3.05 F-test for All State 3.05 Dummies (0.001) Dummies (0.001) 538 Chang and Nothaft