1 Economic Determinants of Foreclosures Amanda Freche December 9, 2017 ECO4935
2 I. INTROD UCTION Although mortgages are essential to providing majority of the population looking to purchase a house with the ability to purchase a house, mortgages can also cause significant applicants to complete credit checks, lo ok at outstanding debt in comparison to assets, and compare annual and monthly income to ensure that the applicant will be financially sound enough to continue with payments until the mortgage has been released and all the payments have been fulfilled. Ho wever even after a successful inspection of an applicant, deeming them financially stable enough to not default on their mortgage, there are still variables that can affect whether a homeowner will default on their mortgage and have to go through fo reclosure. This do not take into consideration when viewing applicants, and look at their effect on the foreclosure rate in counties in Florida. II. SAMPLE The sample used for this study are 53 1 counties in Florida from 2007 to midyear 2008. The housing market during this period within the United States was very tumultuous due to the financial crisis, causing significant variation and larger levels of fo reclosure rates to study. III. DEPENDENT VAR IABLE Estimated Foreclosure Rate by County est_foreclosure_rate 1 Calhoun County, DeSoto County, Dixie County, Gilchrist County, Glades County, Hamilton County, Hardee County, Holmes County, Jefferson County, Lafayette County, Liberty County, Madison County, Union County, and Washington County are all excluded due to lac k of Single Family Home Sales Prices data.
3 The dependent variable for this study is the estimated foreclosure rate by county, or the rate of estimated number of foreclosures divided by the estimated numbe r of mortgages in eac h county. These data are obtained through the U.S. Department of Housing and Urban Development (HUD) 2 throughout the entire 2007 calendar year through October 2008. To col l ect these data the HUD synthesized data from The Mortgage Bankers Association, Equifax, and information from the United States Postal Service. IV. INDEPENDENT VARIABLES Estimated High Cost Loan Rate est_hicost_loan_rate The Estimated High Cost Loan Rate is the estimated percentage of loans made between Data ave raged from 2004 through 2006 are used because it is estimated that the average household will own their home between two to four years be fore possibly f oreclosing, this would be the appropriate lag time before the using 2 008 foreclosure rates. These loans are considered high cost based off the Home Mortgage Disclosure Act, where the rate spread is 3 percentage points above U.S Treasury securities with comparable maturity. These high cost loans differ from traditional conventional loans because they are riskier and often have special protective measures to ensure that lenders are not being predatory against homeowners. The Estimated High Cost Loan Rate is gathered from data from the U.S. Department of Housing and Urban Development 2 within their local level data. With an increase of high cost loan rates, it is hypothesized that there should be a strong positive correlation with the increase of foreclosure rates with in a county. These high cost, riskier loans are more likely to default often because the homeowner has lower credit, irregular 2 U.S. Department of Housing and Urban Development, https://www.huduser.gov/portal/datasets/nsp_foreclosure_data.html
4 income, or other factors that make them less likely to be financially sound. It is also hypothesized that there should be a neg ative correlation between the high cost loan rates and the per capita income because these high cost loans are usually given to people with lower income. Unemployment Rate bls_unemployment_rate // bl s_unemployment_rate_2004 The unemployment rate is the level of unemployment during June 2008 using the official U3 unemployment rate. This counts people who currently do no t have a job, have looked for employment within the last four weeks, and are currently available to work. These data are taken from the U.S. Department of Housing and Urban Development 2 gathered through the U.S. Bureau of Labor Stati stics. In the second regression annua l unemployment rate from 2004 is used to acco unt for a possible lag time in losing jobs and income to foreclosing It is hypothesized that as unemployment rates increase, there should be an increase in foreclosure rates. When a homeowner is unemployed, they will be less likely to afford payments on their home, therefore increasing the possibility of foreclosing on their home. Median Household Income med_household_income M edian household income represents the middle point of the income distribution for that county. Half of the income in the county is above this point, and half is below this point. These data are obtained through the Bureau of Economic and Business Research, collected by the Bureau of Economic Analysis (BEA) 3 using data from the Small Area Income a nd Poverty Estimates (SAIPE) program. SAIPE provides single year estimates of median household income for school districts and federal purposes and is known as one of the most accurate estimates of 3 U.S Census Bureau and Bureau of Economic Analysis, https://www.bebr.ufl.edu/data/localities/207/county
5 income 4 Median household income provides a more accurate measure of income, not skewing an average for large earners or those who are significantly below the poverty line 5 With an increase in per capita personal income, it is hypothesized that there should be a negative correlation with the increase of foreclos ure rates. This is assuming that with a higher personal income, a homeowner is more likely to afford their mortgage loan and continue with their payments. If the median single family home value is a higher proportion compared to the per capita personal i ncome, the homeowner in that area may be more likely to default on their mortgage loan, therefore increasing the foreclosure rate in that county 6 New Construction Value building_permits _pop // building_permits_ pop_2004 T he annual building permits for single family units is the number of building permits handed out within a year for single family units. Although these data include fully detached, semi detached, row houses, and town houses, it can stil l be used as a proxy for new construction of single family homes during the 2007 and 2008 period 7 These data are then divided by the population of each county 8 to account for larger amounts of construction activity in more populated counties and vice versa. These data are obtained through the Bureau of Economic and Business Research collected by the U.S. Census Bureau 9 In the second re g ression, single family building permits d ivided by populati on during 2004 is used to account for construction activity 4 U.S Census Bureau, h ttps://www.census.gov/programs surveys/saipe/about.html 5 Originally Per Capita Personal Income was used, but was eliminated due to the inaccurate averages for extremely high or low earners within the county. 6 This does not consider mortgage lenders offering disproportionally larger loans due to the increased income of individuals, assuming they are financially sound enough to cover these larger loans. Although issuing faulty loans was a major cause of th e housing crisis, it is too complicated with a short amount of time to fully analyze this. 7 Previous ly, a percentage of houses buil t still in exist ence was used, but this was not effective in finding the number or percentage of construction acti vity during the time period. 8 Gathered through Bureau of Economic Business Research, https://www.bebr.ufl.edu/data/localities/9085/county 9 U.S Cen sus Bureau and Bureau of Economic Business Research, https://www.bebr.ufl.edu/data/localities/345/county
6 occurring before the foreclosure period. These data are obtained through the U.S Department of Housing and Urban Development 10 There is significant research pointing to high amounts of construction and influx of new homes as being one o f the causes of the housing bubble and eventually the economic downturn. Hedberg and Krainer from the San Francisco Federal Reserve are producing a series of working papers on the effects of housing supply on foreclosures 11 y true that construction fell more in states where foreclosure rates were higher, but construction activity According to the paper, one of the reasons why new c onstruction homes were particularly affected by foreclosures is that young new homebuyers that are more susceptible to unstable income and employment were more likely to purchase new homes (Hedberg & Krainer, 2012, p. 2). If this is assumed to be true, it is hypothesized that there will be a positive correlation with an increase in percentages of new home construction, an increase in high cost loan rates and unemployment rates, causing an increase in foreclosure rates within each county. M edian Sales Pr ice and Home Value for Single Family Homes med_home_value The median sales price and home value for single family homes represents the middle point of the prices of single family homes sold within each county of Florida. Half of all the homes sold are above this point, and half of all the homes sold are below this point. These median sales price and home value data are collected from the Bureau of Economic and Business Research (BEBR) 12 The median value is used instead of an average because extremely high or extremely low 10 U.S Department of Housing and Urban Development https://socds.huduser.gov/permits/index.html 11 Hedberg, W., & Krainer, J. (September 2012). Housing Supply and Foreclosures. FED ERAL RESERVE BANK OF SAN FRANCISCO. Retrieved from http://www.frbsf.org/economic research/files/wp12 20bk.pdf 12 Bureau of Economic and Business Research Housing and Real Estate Data base, https://www.bebr.ufl.edu/data/localities/3555/county
7 sale pric es or home values could skew the data and push the average home value in the county up or down. A single family home is a detached, free standing residential building. This paper specifically observes data of single family homes, because they have the highest rate of homeownership compared to condominiums, duplexes, and other owned forms of multi family housing. With an increase in single family home values and sales prices, it is hypothesized that there will be an increase in the size of a mortgage loan to cover the value of the house. With an increase in the mortgage loan and higher payme nt amounts, it is more likely that a homeowner will default on their loan payments, correlating to an increase in foreclosure rates. V. RESULTS Correlation Matrix Regression 1 Co rrelati on Matrix Regression 2 To account for collinearity issues, a correlation matrix was made to ensure that no independent variables were colinear. No two variables were highly correlated 13 meaning they could all be used within the regression. 13 In Regression 1, High Cost Loan Rate and Unemployment were largely correlated since unemployment is one of the criteria of the loans. Median Household Income and Median Home Value were also largely correlated since an increase in income can lead to an increase in what one can afford for housing. The threshold for determining whether something is too correlated is the absolute value of 0.6.
8 Impact Summary Regression 1 Im pact Summary R egression 2 Descriptive Statistics of Data Regression 1 Descri ptive Statistics of Data Regression 2 Regression 1 with All Variables est_foreclosure_rate = 0 + 1 ( est_hicost_loan_rate) + 2 (bls_unemployment_rate) + 3 4 (building_permits _pop 5 (med_home_value) + error Regression 2 with All Variables est_foreclosure_rate = 0 + 1 ( est_hicost_loan_rate) + 2 (bls_unemployment_rate _2004 ) + 3 4 (building_permits _pop _2004 5 (med_home_value) + error
9 Excel Regression Output Regression 1 E xcel Regression Output Regression 2 Estimated High Cost Loan Rate est_hicost_loan_rate In the first regression t he t stat for estimated high cost loan rate is 6.3664 meaning that it is statistically significant at the 9 9.9 % confidence level. In the second regression, the t stat is 7.3099, still statistically significant at t he 99.9% confidence level The impact summary is 0.011 in the first regression meaning that for every standard deviation increase in the loan rate, there is a 0.011 increase in the foreclosure rate. In the second regression, the impact summar y only increases by 0.008. This is consistent with the hypothesis that an increase in riskier loans would cause an increase in the fore closure rate. When looking at the correlation matrix, this is also consistent with the hypothesis that high cost loan rates would be negatively correlated with
10 median income, because these loans are specifically given to people with lower than average inc omes. Unemployment Rate bls_unemployment_rate // bls_unemployment_rate_2004 In the first regression, t he t stat for unemployment rate is 8. 5771 meaning that it statistically significant at the 9 9.9 % confidence level. The impact summary for this regression is 0.013 6 meaning that for every standard deviation increase in the unemployment rate, there is a 0. 0 13 6 increase in the foreclosure rate. This is consistent with the hypothesis that an increase in the unemployment rate can be asso ciated with increase in the foreclosure rate However, in the second regression using data from 2004 the t st at is 1.0190 meaning that it is not statistically significant at any co nfidence le vel. This is not consistent with the hypothesis that ther e could be a lag time between loss o f job or income, and foreclos ing It is pos sibl e that losing income would have a more immed iate effect on mort gage payments and f oreclos ures Median Household Income med_household_income In the first regression, t he t stat for median household income is 3.5121 meaning that it is statistically significant at the 9 9 % confidence level. The second regression s t stat is 2.5604, still statistically si gnificant at the 98% confidence level. The impact summary for the first regression is 0.0 0 636 meaning that for every standard deviation increase in the median household income, there is 0.0 0 636 increase in the foreclosure rate. In the second regression, the impact summary only increases by 0.0009. This is not consistent with the hypothesis that an increase in income wo uld mean that one is more likely to be able to afford their loans and therefore not default. A possible explanation for this is the larger the income, the larger loan an applicant would be approved for, making it riskier if something would happen to their income, an d therefore more likely to default on their loan. New Construction Value building_permits _pop // bu ilding_permits_pop_2004
11 In the first regression, t he t stat for the number of building permits representing new construction is 1. 256 0 meaning that it is not statistically significant within any confidence level s There is no linear relationship between construction levels and the foreclosure rate. The coefficient and impact summary were negative however, indicating that as the foreclosure rate increased, the number of building permits, or value of new construction decreased during 2008 In the secon d regression using the 2004 levels of bu ilding permits, t he t stat is 1.6628, statistically significant at the 90% con fidence level indicating that there is a linear relations hip between 2004 construction levels and 2008 f oreclosures. Th is impact summary is 0.00372 meaning that for every standard deviation increase in the 2004 bu ilding permi t levels there is 0.0 0 372 increase in the 2008 foreclosure rate The coefficient is also positive, indicating that as the const r uction levels increa sed, the forec l o sure rate inc re ased as hypothesized This confi rms that there is a lag time for the construction levels to have an effect on foreclosure rates. Median Sa les Price and Home Value for Single Family Homes med_home_value The t stat for median sales price and home value for single family homes is 0. 7939 in the fir s t regression and 0.5375 in the second regression meaning that it was not statistically significant at any con ventional level s in either regression There is no linear relationship between the price of homes and the rate of foreclosures. VI. CONCLUSION The results of this study show that outside variables that would not be accounted for during screening for loans by banks and other financial institut ions are highly influential on the rate in which people foreclose on their homes. The first study achieved an adjusted R Squared of 0.815, indicating that 81.5% of the variation of the estimated foreclosure rate can be explained by this model and the independen t variables. The second study achieved an adjusted R Squared of 0.5437, in dica ting tha t 54.4% of the variation is explained by the ind ependent variables
12 The strongest independent variable was high cost loan rate, having th e strongest t statistic i n both regressions This is mostly because therefore high cost to the financial institutions due to their likeli hood of default. Therefore, there would be a strong correlation between these high risk, high cost loans and the foreclosure rate. This is also something that financial institutions would gauge for when administering loans, but could prove that these institutions administer these loans knowing they will fail and default due to their strong correlation with the foreclosure rate. The next strongest t statistic in both regressions was Median Household Income A lthough statistically sign ificant, the positive coefficient and impac t summary contra dicted the original hypothesis that an increase in inc ome will r esult in less likelihood of default. This could be because a lar ger income mig ht lead to a larger mortgage lo an, and with a loss of income or fina ncial issues, more likely to foreclose. The ad dition of the se cond regression had conflicting results with t he unemployment rate and new c onstruction values. In the first regression, unemployme nt was the strongest t statistic and was the most statistically significant This was consistent with the hypothesis that if someone does not have a job or steady income, they will not be able to afford their loan on their home and therefore default and foreclose. Although unemployment would be something that financial institutions would check for at the time of administering the loan, this variable can account for people changing their employment status during the life of the loan after they have received the initial approval How ever in the second regression u sing 2004 unemploy ment rates, it w as no longer statistically significant. Th is does not support the hy pothesis that 2004 unempl oyment would have a larger effect on 2008 foreclosure rates Because unem ployment was statistically significant during the first regression a nd not statis tically significant in the second regression, this proves that unemployment affects foreclosure rates im m ediately and there is not a possible lag effect Construction values al so had a d ifferent outcome w hen addin g in the second regression wi t h o lder 2004 data. In the first regression, new con struction levels were not
13 stat istically signifi cant meaning that an increase in construction did not affect foreclosure rate. However, when using 2004 construction levels in comparison to 2008 foreclosur e rate, this t stat was statistically significa nt T his could be because there is a possible la g time with the effects of increased const ruction levels on foreclosure rates. This stat istical significan ce not o nl y support s the h ypothesis, but also confirms t he r esearch that increased construction can be attributed as one of the causes of the housi ng bu b ble and financial crisis. When looking at the housing bubble and foreclosures during this tim e, the increase in foreclosures can mostly be attributed to macroeconomic forces such as unemployment and median income, rather than market specific variables such as median sales price of homes. It is po ssible that there would be a change in these variables if studying other locations, such as areas that were severely affected by the recession, however most likely the same results would occur Overall, these large outside forces that financial institutio ns could not account for or prepare for, had the greatest effect on the foreclosur e rate
14 W orks Cited BEBR Population Estimates All Races Total. (n.d. ). Retrieved December 08, 2017, from https://www.bebr.ufl.edu/data/localities/9085/county Hedberg, W., & Krainer, J. (September 2012). Housing Supply and Foreclosures. FEDERAL RESERVE BAN K OF SAN FRANCISCO. Retrieved from http://www.frbsf.org/economic research/files/wp12 20bk.pdf Home Value and Home Prices Single Family Homes Median Sales Price ($). (n.d.). Retr ieved December 08, 2017, from https://www.bebr.ufl.edu/data/localities/3555/county U.S. Census Bureau. (n.d. ). Housing Monthly Building Permits Single Family Units. Retrieved December 08, 2017, from https://www.bebr.ufl.edu/data/localities/345/county U.S. Census Bureau. (n.d.). Income Media n Household Income (SAIP). Retrieved December 08, 2017, from https://www.bebr.ufl.edu/data/localities/207/county U.S. Census Bureau. (n.d. ). Small Area Income and Poverty Estimates (SAIPE) Program. Retrieved December 09, 2017, from https://www.census.gov/programs surveys/saipe/about.html U.S. Department of Housing and U rban Development. (2008, October 20). HUD Provided Local Level Data. Retrieved December 08, 2017, from https://www.huduser.gov/portal/datasets/nsp_foreclosure_data.html U.S. Department of Housing and U rban Development. (200 9 ). State of the Cities Data Systems (SOCDS) Retrieved March 1 8, 201 8 from https://socds.huduser.gov/permits/index.html Unite d States, U.S. Department of Housing and Urban Development. (2008, October 20). Neighborhood Stabilization Program: Methodology and Data Dictionary for HUD
15 Provided Data Retrieved December 8, 2017, from https://www.huduser.gov/portal/datasets/nsp_foreclosure_data.html