Goldman Sachs Home Prices and Credit Losses Projections and Policy Options

Please download to get full document.

View again

of 28
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Information Report
Category:

Music

Published:

Views: 0 | Pages: 28

Extension: PDF | Download: 0

Share
Related documents
Description
Global Economics Paper No. 177 GS GLOBAL ECONOMIC WEBSITE Goldman Sachs Global ECS Research at http://360.gs.com Home Prices and Credit Losses: Projections and P
Transcript
olicy Options We Introduce Two New Models to Track Metro Area Home Prices and Loan-Level Mortgage Credit Performance National Housing Valuations Have Largely Normalized, but Excess Supply Is Likely to Result in Further Price Declines Loan-Level Model Shows that Home Prices Are the Primary Driver of Mortgage Credit Performance Risk of Feedback from Foreclosures to Further House Price Declines Justifies Aggressive Policy Response Bulk Loan Modification: Microeconomic Costs, Macroeconomic Benefits A Government Subsidy for Writedowns of Mortgage Principal? Important disclosures appear at the back of this document. We are grateful to our colleagues in economic research, banks equity research, and mortgage strategies for generously sharing their data and insights. All opinions and remaining errors are our own. Jan Hatzius Michael A. Marschoun January 13, 2009 Goldman Sachs Global ECS Research Global Economics Paper No. 177 Table of Contents Highlights I. Supply Overhang Pushes Home Prices Below Fair Value Box 1: FHFA versus CS House Price Indexes Box 2: Modeling House Prices II. Projecting Aggregate Mortgage Credit Losses Default and Loss Models for Private-Label Securitizations Box 3: Loss and Default Models Box 4: Classifying Whole Loans Held at Banks Extrapolating Losses to the Broader Market III. Loss Recognition Still Has a Way to Go IV. The Risk of Adverse Feedback Loops V. Policy Options for Reducing Defaults The Benefits and Costs of Bulk Modification Policymakers Could Subsidize Principal Writedown The Devil Is in the Details VI. Concluding Remarks 1 2 4 6 8 8 10 11 12 14 15 18 18 21 22 23 Recent Global Economics Papers 24 Home Prices and Credit Losses January 13, 2009 Goldman Sachs Global ECS Research Global Economics Paper No. 177 Highlights In this paper, we introduce two disaggregated econometric models to track home prices at the metropolitan area level and mortgage credit performance at the loan level. We then use these models to evaluate different types of broad mortgage modification programs. Our models yield two basic conclusions. First, home prices depend primarily on the supply/demand balance in the local housing market as measured by the inventory of existing homes at the metropolitan area level, as well as housing valuations, local unemployment rates, and past local home price trends. Second, mortgage credit performance depends primarily on local home prices, as well as a host of loan-level characteristics such as FICO scores, debt-to-income ratios, and owner occupancy status. Regarding home prices, the good news is that housing valuations at the national level have largely normalized following the price declines of the past 2 1/2 years. Nevertheless, our model suggests that the current level of excess supply and the persistence of past home price trends is consistent with a further price decline through mid-2010 of 5%-10% in terms of the FHFA (formerly OFHEO) index and 20%-25% in terms of the CaseShiller index. Regarding mortgage credit performance, feeding the predictions from the home price model into the mortgage loss model results in a projection of $1.1 trillion in lifetime credit losses on the currently outstanding $11.3 trillion stock of US residential mortgage debt. This includes losses of $422 billion on private label securities and $402 billion on whole loans held at depository institutions, with the GSEs' book of business, FHA loans, and other smaller mortgage holders making up the remainder. Our results imply a strong case for aggressive foreclosure mitigation efforts. While we believe that our loss model correctly pegs the likely losses given the home price path implied by the current level of excess supply, failure to stem foreclosures could result in a further increase in excess supply and thus push up mortgage losses even beyond our baseline estimates. The recent stability in existing home inventories holds out hope that we may avoid such a worst-case scenario, but it would be a mistake for policymakers to ignore the risks. We therefore use our model to estimate the benefits and costs of different types of bulk mortgage modifications. We find that modifications are more cost-efficient if they focus on nonprime rather than prime loans, and if they involve principal writedowns rather than note rate reductions. While most bulk modification programs still have a negative "private" net present value (NPV), their broader "public good" benefits can be large because excess supply is such an important driver of home prices. This may justify significant public outlays on foreclosure prevention efforts, even in cases where the private NPV is negative. For example, the government might offer to pay for a certain percentage of the cost of any large-scale principal writedown program. Home Prices and Credit Losses 1 January 13, 2009 Goldman Sachs Global ECS Research Global Economics Paper No. 177 I. Supply Overhang Pushes Home Prices Below Fair Value The main goal of this paper is to estimate credit losses on the currently outstanding stock of mortgage debt, and to propose policy options designed to reduce the associated downside risks to the housing market and the broader economy. However, a sensible loss forecast needs to start with a sensible home price forecast because home prices are by far the most important macroeconomic determinant of losses. They have substantially higher explanatory power than unemployment rates, interest rates, or rate resets on adjustable-rate mortgages (ARMs). While the lifetime losses on alt-A or subprime loans far exceed those on prime loans, the proportional increase in lifetime losses as a function of HPA is relatively stable across different product categories and vintages. Exhibit 1 clearly shows the strength of the relationship of house prices to losses for several product categories and vintages. For example, for the 1992 vintage of prime loans, lifetime losses average only 0.01% in metropolitan statistical areas where home price appreciation (HPA) in the first three years averaged 6%, but this number rises to 1.00% in metro areas where home price appreciation averaged -6%. While losses on alt-A or subprime loans far exceed those on prime loans, Exhibit 1 shows that the proportional increase in losses as a function of HPA is relatively stable across different product categories and vintages. Hence, the remainder of this section gauges the outlook for US home prices at the metropolitan statistical area (MSA) level. Exhibit 1: Tight Links Between HPA and Loan Losses Percent 100.00 10.00 Product Type: Prime 1992 Prime 2000 Alt A 2000 Subprime 2000 Lifetime Loss 1.00 0.10 0.01 -10 -5 0 5 10 15 20 Average Annual HPA over First Three Years of Life of Loan (Percent) Note: Each dot represents a different metropolitan area. Each color represents a different product type and/or period. HPA was calculated over a 3 year period after loan origination. Source: LoanPerformance. FHFA. Goldman Sachs. Most of the home price valuation excess has been corrected, at least at the national level. Before presenting our econometric model, we take a look at housing valuations at the national level. Our reading is that most of the valuation excess has been corrected, at least at the national level, though prices probably remain somewhat above sustainable levels. This assessment is based on two measures that compare mortgage payments--which depend on home prices and nominal interest rates--with rents and household incomes, respectively. Exhibit 2 shows the cost of owning versus renting, calculated as principal and interest payments (P&I) relative to rents for both the Federal Housing Finance Agency (FHFA, former OFHEO) index and the CaseShiller (CS) index. (The differences between the two house price indexes are discussed in Box 1 on page 4.) Exhibit 3 shows our measure of mortgage affordability, defined as the ratio of mortgage payments to household income. In both cases, we use 1993-2003 as a reference period because inflation during that decade was low and stable and the housing "bubble" had not yet gathered steam.1 Both charts show that the price decline over the past 2 1/2 1 The series shown in Exhibits 2 and 3 are only meaningful for periods of relatively low and stable inflation. In a period of higher inflation, we would 2 January 13, 2009 Home Prices and Credit Losses Goldman Sachs Global ECS Research Global Economics Paper No. 177 years and the recent sharp decline in mortgage rates have eliminated most of the valuation excess in the broad US housing market. Admittedly, the affordability chart suggests that prices are still modestly above fundamental values if we use the FHFA measure of house prices, but broadly speaking, we no longer see a large-scale valuation problem in the US housing market. Prices are still modestly above fundamental values if we use the FHFA measure of house prices. Exhibit 2: Cost of Owning vs. Renting Back to Pre-Bubble Norm Percent 180 Ratio of P&I to Rent 160 140 120 100 80 60 92 94 96 98 00 02 04 06 08 Note: 1993-2003 average is set equal to 100%. Source: FHFA. S&P/Case-Shiller. Freddie Mac. Department of Labor. Goldman Sachs. FHFA S&P/Case-Shiller 140 120 100 80 60 160 Percent 180 Exhibit 3: Affordability Has Also Improved Significantly Percent 30 Mortgage payment on median-priced home in percent of mean family income 25 FHFA S&P/Case-Shiller 25 Percent 30 20 20 15 15 10 92 94 96 98 00 02 04 06 08 Note: Data for Q4 2008 are our estimates. Source: FHFA. S&P/Case-Shiller. Goldman Sachs. 10 expect higher levels of home prices than suggested by this analysis because homebuyers face a much larger incentive to "stretch" and incur larger payments in the early years of a mortgage in the expectation that the real value of their debt will be inflated away quickly in subsequent years. Home Prices and Credit Losses 3 January 13, 2009 Goldman Sachs Global ECS Research Global Economics Paper No. 177 The bad news is that our formal house price models suggest that it may not matter all that much how close house prices are to fundamental value. So much for the good news. The bad news, unfortunately, is that our formal house price models suggest that it may not matter all that much how close house prices are to fundamental value. It shows that while fundamental values act as an anchor for house prices in the long run, their short term behavior is dominated by "technical" factors such as the extent of oversupply and self-fulfilling expectations of further house price declines. These factors are likely to push down home prices considerably further over the next two years. Our econometric house price model predicts metro area level house prices over short and medium term horizons for both the FHFA and CS indexes. However, because our loss model uses the FHFA index as an input we will focus on the forecasts for this index. While fundamental values act as an anchor for house prices in the long run, their short term behavior is dominated by "technical" factors such as the extent of oversupply and self-fulfilling expectations of further house price declines. The model combines "fundamental" variables such as housing affordability and unemployment rates with "technical" measures such as housing inventories, lagged changes in home sales volumes, and short term house price momentum. The estimation technique is a panel regression of current house price changes on lagged values of the fundamental and technical variables. National forecasts are obtained by aggregating the MSA level forecasts. For details on the variables and the estimation technique see Box 2 on page 6. Box 1: FHFA versus CS House Price Indexes Both the Federal Housing Finance Agency (FHFA) and the CaseShiller (CS) national house price index exhibit several biases. On net, the FHFA index paints too optimistic a picture while the CS index paints too pessimistic a picture of house prices. The FHFA index currently understates the rate of house price 2 decline for three reasons. First, it is based on transactions involving conforming mortgages only and thus leaves out the worse performance of prices in the nonprime sector. Second, it is aggregated using unit weighting, although total mortgage risk is better measured using value weighting. Since the highest priced metro areas have been seeing the biggest price drops, this understates the rate of decline. Third, it includes appraisals, which tend to be inflated and "sticky." Meanwhile, the national CS index currently exaggerates the price declines for two reasons. First, its more limited geographic coverage excludes most of the mid-sized and smaller metro areas that have performed better during the downturn than larger metro areas. Second, because of the inclusion of subprime and alt-A loans, a substantial amount of all transactions are distressed sales. The associated "foreclosure discount" depresses the CS index further. Although inclusion of distressed sales is the correct choice if the objective is to accurately measure true transaction values, an index that contains few distressed sales is better suited for loss modeling because it allows an apples-to-apples comparison with historical house price downturns during periods of better overall loan quality and fewer distressed sales. Time series analysis on the two indexes at the MSA level shows that each index "Granger causes" the other. This means that in a time series regression of one index on lagged values of both indexes, the other index is statistically significant. At the local level, the CS index predicts subsequent FHFA HPA more strongly than the other way around, suggesting that it is the better index. Nevertheless, its limited geographical coverage distorts the picture it paints at the national level. Historically the CS indexes have been more volatile at the MSA level than their FHFA counterparts. This, together with the aggregation properties, suggests that CS will continue to underperform FHFA in the current downturn. 2 See the excellent analysis by Andrew Leventis (FHFA) for further details. "Revisiting the Differences between the OFHEO and S&P/Case-Shiller House Price Indexes: New Explanations," January 2008. Home Prices and Credit Losses 4 January 13, 2009 Goldman Sachs Global ECS Research Global Economics Paper No. 177 As shown in Exhibit 4, our model predicts cumulative house price declines over the 2008Q3 to 2010Q3 period of 9% for the FHFA index and 27% for the CS index. This implies declines of about 6% for the FHFA index and 23% for the CS index from the (estimated) yearend 2008 level. The projected declines are due to three main factors: (1) a large excess inventory, (2) very strong persistence in downward HPA momentum (especially in the short run), and (3) falling sales volumes. All three have a large amount of predictive power and look very weak at present. In contrast, the underlying economic factors are less important for our projections. Valuations have at least partly normalized, and the unemployment rate has less predictive power than generally believed. Exhibit 4: House Price Projections across Metro Areas National or Metro Area Fundamental Factors Technical Factors House Price Forecasts %1Qtr %2Yr HPA %2Yr HPA HPA Forecast Forecast Forecast (FHFA) (CSW) (FHFA) 3Q08-3Q10 3Q08-3Q10 3Q08(4) (4) 4Q08 -9 -9 -16 7 3 8 -21 -26 -36 -7 -14 -10 -18 -15 -20 -29 -21 -26 -8 -26 -20 -15 -10 -2 -32 -39 -51 -21 -24 -19 -32 -26 -36 -41 -30 -34 -19 -37 -32 -31 -21 -7.9 -7.3 -8.3 -3.3 -2.4 -1.6 -7.2 -5.2 -7.1 -9.4 -5.6 -6.3 -1.7 -5.7 -4.6 -7 -1.2 -27 -19 -30 -1.9 -3.7 % Unemploy -ment 9M08 All MSAs (6) Atlanta-Sandy Springs Chicago-Naperville-Joliet Dallas-Plano-Irving Denver-Aurora Houston-Sugar Land Las Vegas-Paradise Los Angeles-Long Beach Miami-Miami Beach-Kendall Minneapolis-St. Paul Nassau-Suffolk New York-White Plains Oakland-Fremont-Hayward Orlando-Kissimmee Phoenix-Mesa-Scottsdale Riverside-San Bernardino San Diego-Carlsbad Santa Ana-Anaheim-Irvine Seattle-Bellevue-Everett Tampa-St. Petersburg Washington-Arlington Top 20 MSAs MSAs (n=99) with CS&FHFA not in top 20 (5) MSAs (n=262) with FHFO Indices Only (6) 4.7 4.9 5.4 4.0 4.3 3.8 5.5 5.7 4.7 4.5 3.9 4.5 5.3 4.4 3.2 6.9 5.1 4.4 3.2 4.9 3.2 4.7 5.1 4.4 % DTI (1) 21 16 18 13 15 12 19 37 24 14 23 26 23 24 26 33 25 38 26 18 22 24 20 17 % DTI Demeaned (2) -2 -6 -2 -10 -5 -9 -6 4 3 -3 2 1 -2 -2 -2 0 -3 1 2 -3 -1 -1 -2 -3 Months Supply 2Q08 (3) 11 15 15 6 6 6 3 10 55 9 11 11 4 10 10 10 15 9 12 12 9 % Change % YoY in HPA Sales Volume (FHFA) 3Q07-3Q08 2Q07-2Q08 -4 0 -1 2 0 4 -20 -16 -12 -4 -4 -2 -17 -12 -12 -27 -16 -18 0 -15 -10 -8 -6 1 -15 -24 -29 -17 -12 -17 16 4 -14 -11 -12 -12 4 -14 -4 4 4 4 -46 -14 -33 -11 -16 -22 Note: 1) Estimated average 2008 front end debt to income ratio calculated using average 2007 household income, loan amounts estimated using 2007 HMDA data projected to 2008 using FHFA HPI, Freddie Mac PMMS interest rates, and estimated option ARM and IO shares. 2) DTI demeaned calculated as 2008 DTI ratio minus MSA specific long term average. 3) Months' supply (homes listed for sale divided by average monthly sales volume) Source: NAR. 4) National fundamental balance weighted averages of the MSA figures; national historical and forecasted HPA calculated as the loan count weighted average for OFHEO and loan balance weighted average for CSW; MSA loan count and loan balance obtained from HMDA 2007. 5) Average of 99 mostly larger MSAs for which both CS and FHFA indexes are available, excluding top 20 MSAs 6) Average of 262 mostly smaller and midsize MSAs for which only FHFA indexes are available, but not CS. Source: FHFA. Fiserv Case Shiller Weiss. NAR. Department of Labor. Department of Commerce. HMDA. Freddie Mac. Inside Mortgage Finance. Goldman Sachs. Our analysis suggests that large metro areas are likely to see much worse performance than smaller ones. Our house price projections vary widely across metro areas. In Miami, we expect house prices over the next two years to decline by 36% (using the FHFA index), with extremely high housing inventory as the main driver. In Los Angeles, we expect prices to drop by 26%, because of downward price momentum, high housing inventory, and poor affordability. In contrast, in Houston we project an 8% house price increase because of good fundamentals and moderate inventory. In general, our analysis suggests that large metro areas are likely to see much worse performance than smaller ones. As shown in the table, the two-year forecast for the top 20 metro areas is -15% HPA, but for the typically smaller 262 metro areas for which no CS indexes are available the two year forecast is -2% HPA. Home Prices and Credit Losses 5 January 13, 2009 Goldman Sachs Global ECS Research Global Economics Paper No. 177 Box 2: Modeling House Prices The two-year FHFA house price model is estimated using a panel regression of the general form 1990s. Does this mean that bad affordability caused the downturn? Not necessarily, because the fact that house prices dropped after 1990 improved average affordability for LA and makes 1990 look worse than it would otherwise have. So the causality found with this method (which is effectively how fixed effects regression works) is partially spurious. On the other hand, a finding that LA had poor affordability in 1990 compared with periods prior to that point, and that house prices dropped subsequently contains real information. Thus, using the particular variable specification chosen in our models gives us a more credible estimate of how fast house prices mean revert from extreme affordability levels. All the predictive variables are used in lagged form, which means the HPA forecast over the next two years uses current values of unemployment, affordability, and other variables as inputs. This technique avoids the need to forecast any of the explanatory variables. To make forecasts over different time horizo
Recommended
View more...
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks