Portal de Periódicos da CAPES

The Determinants of Home Price Appreciation Among Community Reinvestment Homeowners

2007; Taylor & Francis; Volume: 22; Issue: 3 Linguagem: Inglês

10.1080/02673030701254152

1466-1810

Michael A. Stegman, Roberto G. Quercia, Walter R. Davis,

Housing, Finance, and Neoliberalism

Abstract Homeownership is considered an effective wealth creation mechanism for low-income households. This study examines the appreciation of homes purchased with community reinvestment loans in a national pilot in the USA called the Community Advantage Program (CAP). Homes purchased between 1998 and 2002 are found to have appreciated at a median annual rate of 5.4% between the time of purchase and spring 2003. This is less than the national house price appreciation index of 7.0% (covering 1998–2003) but higher than other types of investments such as the Dow Jones Index (2.78% annual growth rate) and the average rate on a 6-month CD (4.34%) over the same time period. The median increase in net housing wealth of the Community Advantage homeowners is $17 492. Returns have been particularly impressive for the lowest-income borrowers. Borrowers with less than $20 000 in annual household income at the time of purchase had a median down-payment of $1790 and have experienced a median equity increase of $24 724. Price and equity appreciation rates for African Americans are solid, but are about 10% lower than for whites. Keywords: Home price appreciationlow-income homeownershipnegative Notes 1 Fannie Mae's Flexible 100 mortgage allows for no down-payment but does require a contribution of 3 per cent to the closing costs. 2 Community Reinvestment Act (CRA) loans refer to those loans that a depository institution can use to meets its CRA obligations. Under CRA, depository institutions are evaluated on a periodic basis for the degree to which they meet the credit needs of the communities in which branches are located, particularly the credit needs of low-income borrowers and borrowers in underserved communities. The regulators administering CRA exams then reference these evaluations when considering an institution's application for deposit facilities, particularly applications for merger and acquisition. While the examinations differ by regulatory body and include multiple means for banks to show reinvestment in their communities, most banks develop and implement affordable mortgage products and programs as one component of their CRA strategy. 3 These requirements mirror the definition of affordable mortgage used by Fannie Mae and Freddie Mac. 4 Although not quite national in scope, CAP's geographic reach is impressive; CAP includes loans from 47 states and the District of Columbia. Because the program originated in Self-Help's home state, North Carolina lenders have originated 34 per cent of all CAP loans, with lenders in California accounting for 18 per cent and those in South Carolina a distant third at 5 per cent (Figure 10). Virginia, Ohio and Oklahoma each account for about 4 per cent of all loans in the 1998–2002 sample. It should be noted that selection into the CAP program is not a random assignment process. It is likely that the over-sampling of North Carolina relative to the population, if anything, understates the average appreciation rate; however, the study does not attempt to correct for such sampling issues, as the mechanism of selection into CAP is not random assignment. 5 Post-purchase joblessness among spouses includes those who are not in the labor force. 6 Scores are grouped into five buckets that categorize the confidence level of predictions, where each successive bucket is characterized by a flatter and more widely dispersed distribution. Because of substantially higher mean and median rate and variance of price appreciation for the least reliable confidence category, it was decided to omit those loans from the analysis. 7 Each of these models depends both on Fannie Mae's own proprietary loan data as well as public tax record and purchased deed data, and each was tested out-of-sample in ongoing Fannie Mae research efforts. Testing the fitness of a proposed model by verifying that it works well on a separate data sample not used in its development, that is, out-of-sample data, is a common practice to mitigate the over-fitting problem. In other words, it is possible that factors and models perform well on a particular sample of data by chance alone, and optimizing parameters of the models on that sample increases this probability. Verifying that a proposed model works well on out-of-sample data mitigates this risk. 8 To estimate changes in the Dow Jones Index, the Dow value was taken on 1 January 1998 and 30 September 2003, and the appreciation rate (on an annual basis, compounded monthly) was calculated. The CD annualized rate of return from January 1998 to September 2003 assumes a 6-month CD purchase in January 1998 and rolled over every six months at the national average CD rate in that month (CD rates downloaded from http://mortgage-x.com/general/indexes/default.asp). 9 Because of time and resource constraints, it was not possible to run the AVM separately for each prepaid loan marking to market all house values on their respective loan termination dates. 10 For the sake of notational simplicity, measures based on 10-year T-bill rates were assigned to the ‘property-level’ variables although these are in fact characteristics of the month in which the property was purchased. 11 Historical T-bill rates were downloaded from the ‘Data Buffet’ at economy.com on 13 June 2004 (Economy.com, Citation2004). The original source is the Federal Reserve Board. 12 Historical monthly county-level unemployment rates, non-seasonally adjusted, were downloaded from the ‘Data Buffet’ at economy.com on 13 June 2004 (Economy.com, Citation2004). The original source is the Bureau of Labor Statistics. 13 As presented, Model 4 is a trimmed model. Other zip-level variables in the full model included the percentage of housing units with incomplete kitchen and/or plumbing, percentage of crowded housing units, percentage of population without a high school diploma, percentage of population with a college degree, and percentage of the population receiving some public assistance. None were significant, and they were removed due to concerns with collinearity and model estimation time. 14 The reliability-specific error terms are imposed to correct for heteroskedasticity caused by different levels of precision in the AVM scores. In addition to producing the AVM score, Fannie Mae's AVM model assigns each score a reliability value of 1, 2, 3 or 4. Accordingly, the variance of the AVM scores increases with each group. Model 4 includes a fixed effect for each reliability level in order to normalize each group to the same mean value. Because the variance of the error term clearly varies with these fixed effects, a correction is necessary. The reliability-specific error terms refer to the structure imposed on the variance-covariance matrix; observations are grouped by reliability score, but otherwise the matrix is left unstructured. This choice is consistent with the known heteroskedasticity caused by the reliability score fixed effects, but can also be compared to other potential covariance matrix structures using the Akaike Information Criterion (AIC) or a similar measure of fit. 15 Kreft & DeLeeuw (Citation1998) survey simulation studies addressing the issue of power in mixed linear models. They report that for data with 150 macro-units with 5 micro-units each, the model has a power of 0.90 for detecting cross-level effects. They also report that variance component estimates show no bias with 300 or more macro-units. The data used for the model consist of 11 524 properties in 821 zip codes. 16 The R2 is approximated as the square of the correlation between the predicted value and the observed value of the dependent variable. 17 See footnote 10. 18 Without the interaction of Hispanic and foreign born, the percentage of the population that is Hispanic does not have a significant effect, while the percentage of the population that is foreign born has a positive and significant effect.