ArticlePDF Available

Abstract and Figures

U.S. House of Representatives Financial Services Committee considered many important banking reforms in 2009-2010 including the Dodd-Frank Act. We show that during this period, the foreclosure starts on delinquent mortgages were delayed in the districts of committee members even though there was no difference in delinquency rates between committee and non-committee districts. In these areas, banks delayed the start of the foreclosure process by 0.5 months (relative to the 12-month average). The total estimated cost of delay to lenders is an order of magnitude greater than the campaign contributions by the Political Action Committees of the largest mortgage servicing banks to the committee members in that period and is comparable to these banks’ lobbying expenditures.
Content may be subject to copyright.
The Politics of Foreclosures*
Sumit Agarwala
Gene Amrominb
Itzhak Ben-Davidc
Serdar Dincd
October 2 2017
Abstract
U.S. House of Representatives Financial Services Committee considered many important
banking reforms in 2009-2010 including the Dodd-Frank Act. We show that during this period,
the foreclosure starts on delinquent mortgages were delayed in the districts of committee
members even though there was no difference in delinquency rates between committee and non-
committee districts. In these areas, banks delayed the start of the foreclosure process by 0.5
months (relative to the 12-month average). The total estimated cost of delay to lenders is an
order of magnitude greater than the campaign contributions by the Political Action Committees
of the largest mortgage servicing banks to the committee members in that period and is
comparable to these banks’ lobbying expenditures.
JEL Codes: D72, G21, G01
Keywords: Political Economy, Real Estate Lending, Household Finance, Financial Crisis, Lobbying.
___________________________________
* Zachary Duey, Caitlin Kearns, Robert McMenamin, Michael Murto, and Felix Zhang provided excellent research
assistance. We thank Ken Singleton (editor), two referees, Gadi Barlevy, Jeff Campbell, Artyom Durnev (WFA
discussant), Jeffrey Frieden (NBER discussant) Luojia Hu, Brian Knight (NBER discussant), Chris Mayer, Brian
Melzer (FIRS discussant), Anna Paulson, Mark Pocock, Todd Sinai, Amit Seru, Dan Sullivan and seminar
participants at Chicago Fed, FIRS conference, NBER Political Economy, NBER Real Estate meetings, OCC,
Rutgers University, and WFA conference for comments. The views expressed here do not represent those of the
Federal Reserve Bank of Chicago or the Federal Reserve System.
a Georgetown University
b Federal Reserve Bank of Chicago
c The Ohio State University, and NBER
d Rutgers University
1
1 Introduction
The financial crisis of 2008 led to a sharp rise in foreclosures. As a highly visible symbol
of the crisis, foreclosures attracted considerable attention from the politicians, regulators, and the
public. In particular, large banksmany of which also service securitized mortgagesand their
foreclosure practices faced great public scrutiny. For example, during a congressional hearing on
February 10, 2009, Barney Frank, then-chairman of the Financial Services Committee in the U.S.
House of Representatives, called for a moratorium on new foreclosure starts.1 At least one very
large bank agreed within days to delay any new foreclosures.2 Foreclosures remained a focal
point in the political discussion in the following months.
Our paper focuses on whether the financial institutions’ decisions to start the foreclosure
process were systematically related to their political concerns in addition to economic concerns.
Congress took many major legislative actions in 2009-2010 regarding the financial sector
including the Dodd-Frank Act, as well as the issues of capital infusions into failing banks,
derivatives trading, and executive compensation. The financial industry thus had many reasons to
pay keen attention to legislative developments and to adjust some of its practices in anticipation.
The literature on campaign contributions and lobbying suggests many reasons why banks
might attempt to influence the political process through their foreclosure actions.3 First,
foreclosure delays might help decrease pressure on politicians from their voters and allow banks
1 Barney Frank said: “…while we wait for President Obama’s plan, I call on institutions that
hold or service mortgages to delay and stop any foreclosure proceedings…[I]n this situation
where the Obama Administration will have a specific plan shortly, a moratorium is clearly
called for.” http://democrats.financialservices.house.gov/press/PRArticle.aspx?NewsID=456
2 See the letter from Jamie Dimon, the CEO of JPMorgan Chase, to Barney Frank, dated February 12, 2009.
http://www.house.gov/apps/list/press/financialsvcs_dem/press021309.shtml
3 See Grossman and Helpman (2001), Stratmann (2005) and Leech (2010) for surveys of campaign contributions
and lobbying. Cooper et al (2010) shows the relationship between political contributions and stock returns.
2
to obtain more favorable legislative outcomes.4 Second, a lenient approach to foreclosure in the
districts of powerful politicians may help banks gain access to them.5 Third, these delays may
help politicians who have a reputation of being sympathetic to the banks’ perspective to get
reelected.6 Finally, the politicians themselves might pressure the banks for leniency on
delinquent borrowers in their district as the quote above suggests. Naturally, these reasons are
not mutually exclusive and possibly not exhaustive.
The foreclosure process on delinquent loans starts only when lender –or its agent, the loan
servicertakes explicit action; therefore, the start of the foreclosure process is largely
discretionary and can be delayed. To identify political motivations in the banks’ foreclosure
decision, we follow the political economy literature that emphasizes the importance of the
Financial Services Committee in the U.S. House of Representative for the laws related to the
financial sector.7 More specifically, we study whether banks delay foreclosure initiations on
delinquent mortgages in the districts of House Financial Services Committee members during
2009-2010 when important financial sector legislation was being considered. We use
institutional details of U.S. Congress in our test design. For example, given the importance of
seniority in Congressional committees, incumbents tend to stay on the same committee for
multiple terms. Hence, most members made the decision to be on the Financial Services
Committee long before the financial crisis.
4 For the ‘quid-pro-quo’ relationship between interest groups and politicians, see, e.g., Austen-Smith (1987), Baron
(1989), Baron (1994), Snyder (1990), Snyder (1991), and Grossman and Helpman (1994).
5 See, e.g., Austen-Smith (1995), Ansolabehere, Snyder, Tripathi (2002), Bertrand, Bombardini, and Trebbi (2014)
for the importance of access to politicians in influencing policy.
6 See Kroszner & Stratmann (2005) for the role of politicians’ reputation in the relationship between interest groups
and politicians.
7 See, e.g., Romer and Weingast (1991) who study the legislation around the saving and loans crisis, and Nunez and
Rosenthal (2004) who study the political economy issues surrounding the passage of the personal bankruptcy reform
bill of 2005. Kroszner and Stratmann (1998) provide a theory congressional committees and interest groups and test
their theory by analizing House Banking Committee, predecessor of the Financial Services Committee we study.
3
We find that mortgage-servicing banks indeed delayed the start of foreclosures on
delinquent loans if those loans were located in the electoral districts of House Financial Services
Committee members. Importantly, there was no difference in delinquency rates in committee
districts so this differential delay cannot be attributed to servicers’ capacity constraints under a
large volume of delinquent loans. These results are robust to many loan- and location-specific
controls, some of which are time-varying (e.g., zip code-level house price changes), as well as
any state-specific time effects.
The average time to foreclosure starts in non-committee districts is about 12 months,
taking into account the right-censoring at the sample end. In committee districts, however, this
average is about half a month longer. Based on the foreclosure cost estimates from the literature
and using only the aggregate value of delinquent loans in our sample, we conservatively estimate
the direct cost of delay to lenders to be about $30M. Although this cost may be small in the
context of the mortgage market, it should be judged relative to other political actions by the large
banks. After all, the importance of banks’ political activities is not captured by the direct impact
of their cost on bank earnings but rather through their potential to influence the political process.
For example, the top ten mortgage servicers, which include some of the largest financial
institutions like Bank of America, Citigroup, JPMorgan, and Wells Fargo, collectively spent
about $44M during this period for lobbying all of the legislative and executive branches, not just
committee members. The combined campaign contributions to the committee members by their
Political Action Committees were about $1M. In other words, the cost of foreclosure delays in
committee districts is comparable to the lobbying costs for the ten largest servicers and an order
of magnitude greater than their campaign contributions to committee members.
4
Delaying foreclosure starts is a novel channel for banks to influence the political process.
Despite the similarities of its cost to the lobbying expenses by these servicers, there are also
differences with lobbying and campaign contributions. By law, lobbying expenditures cannot be
channeled to politicians’ campaigns. These delays, on the other hand, directly benefit the
constituents of committee members. Campaign contributions can, of course, be used in the
politicians’ campaigns but the foreclosure delays may be able to target the politicians’ voters
even more precisely than some of the campaigning paid by contributions. For example,
television advertisements are used heavily in political campaigns but their coverage may
necessarily extend beyond the district borders especially in urban areas. By contrast, foreclosure
delays can target voters in a district accurately. Unlike campaign contributions, these delays are
not subject to campaign contribution limits and do not have to be disclosed.
We verify that our results are not spurious through a variety of tests. As placebo tests, we
check the membership in the Transportation & Infrastructure and Defense committees and find
no link between membership in either of those two committees and the timing of foreclosures.
We also do not find any effect of the Financial Services committee membership in the earlier
years when foreclosures did not attract as much attention. In addition, if the effect we detect were
due to spurious correlation, one might expect that the Financial Services committee membership
in 2009-2010 would influence foreclosure decisions before 2009. We find no such effect.
To address the concerns that legislators may self-select into the Financial Services
committee based on the delinquency rates of their constituents, we check the robustness of our
results by restricting our sample to legislators who were elected to the House and to the Financial
Services committee before 2005, well before the onset of the crisis. We find similar, even
stronger, effect of committee membership for this subsample as well. We also find similar results
5
when the non-committee districts are restricted to those near the committee districts.
Interestingly, we also check for the differences in delinquency rates but do not find any between
committee and non-committee districts. Given the consistency of our robustness results, we
interpret the delay in mortgage foreclosures as being due to loan servicing banks’ attempt to
influence the political process.
Our paper is related to three strands of the literature. First, there is a growing body of
work that explores various aspects of mortgage market practices. In contrast to a large set of
literature that evaluates pre-crisis market developments see, for instance, Keys, Mukherjee,
Seru, and Vig (2010, 2012), Mian and Sufi (2009), Adelino, Schoar, and Severino (2016),
Mayer, Pence, and Sherlund (2009), Jiang, Nelson, and Vytlacil (2014a, 2014b), and Agarwal,
Chang, and Yavas (2012) we focus on the aftermath of the crisis. As such, our paper is also
closely related to studies that explore lenders’ or their agents’ approaches to loss mitigation of
delinquent mortgages whose numbers surged during the crisis period. These studies include work
by Piskorski, Seru, and Vig (2010), Agarwal, Amromin, Ben-David, Chomsisengphet, and
Evanoff (2011), and Zhang (2013), which compare the start of foreclosures for portfolio loans
with that for securitized loans, focusing on potential agency problems. Agarwal, Amromin, Ben-
David, Chomsisengphet, Piskorski, and Seru (2017) evaluate the effects of the Home Affordable
Modification Program that offered mortgage modifications to millions of borrowers. Our focus is
the political motivations in banks’ approach to delinquent mortgages.
Second, our paper contributes to the literature on political influences by corporations that
largely focuses on campaign contributions and lobbying.8 However, by focusing on a non-
traditional political activity, it is similar to Bertrand, Kramarz, Schoar, and Thesmar (2007) who
8 See Grossman and Helpman (2001), Stratmann (2005) and Leech (2010) for surveys.
6
find that politically connected firms distort their labor practices to help favored politicians in
elections in France.9
Finally, our paper is related to the literature on the political economy of finance, in
particular to the studies that examine the role of politics in financial crisis and in its aftermath.10
Mian, Sufi, and Trebbi (2010) demonstrate that representatives of districts with high default rates
were more likely to vote for the Foreclosure Prevention Act in the U.S. House of
Representatives. Igan, Mishra, and Tressel (2009) document that lenders that lobbied more for
relaxation of rules pertaining to securitization and consumer protection subsequently increased
their mortgage lending more and originated loans with higher loan-to-income ratios. Duchin and
Sosyura (2010) argue that banks with headquarters located in the district of a member of the
House Financial Services committee were more likely to receive TARP funds. Adelino and Dinc
(2014) find that financially distressed firms lobbied more for the Stimulus Act and the firms that
had lobbied, and not necessarily the distressed firms per se, received larger stimulus funds.11 Our
paper instead studies the actions taken by banks, not by politicians.
The rest of the paper is organized as follows. Section 2 describes the institutional
background while Section 3 describes the data. Section 4 presents our empirical analysis and
Section 5 provides robustness analysis. Section 6 investigates whether delays in foreclosure
actions were associated with improvements in other aspects of borrower welfare. Section 7
concludes.
9 See also Purnanandam and Weagley (2016) who show how financial markets can discipline the government.
10 See Chinn and Frieden (2011) and Rajan (2012) for general studies of the interaction between politics and
economics that led to the financial crisis in 2008.
11 For earlier studies, see Romer and Weingast (1991) on how the environment in which the Savings and Loans
crisis developed during the 1980s was heavily influenced by lobbying; Nunez and Rosenthal (2004) for the
correlation between campaign contributions and congressional voting patterns on the Bankruptcy Reform bills in
2001; and Brown and Dinc (2005) for the role of the electoral cycle in the government’s decision to intervene in
failing banks in emerging markets.
7
2 Institutional Background
2.1 Political Background
We focus on foreclosures initiated during the 111th Congress, January 2009 through
December 2010. Following the literature on the importance of committees in Congress in
general, and that of the Financial Services Committee for financial institutions in particular,12 we
study the role of membership in the House Financial Services Committee.
The House Financial Services Committee was very busy in this period. The issues
discussed included Troubled Asset Relief Program (TARP) and similar crisis-related programs,
deposit insurance limits, restrictions on executive compensation, credit cards, derivatives
clearing houses, and consumer financial protection. The Committee also played a crucial role in
shaping the broadest financial sector reform since the Glass-Steagall Act of 1933 that culminated
in the Dodd-Frank Wall Street Reform and Consumer Protection Act.13 Moreover, the
Committee was actively involved in investigating various aspects of mortgage servicing markets,
including possible irregularities in the foreclosure process, commonly-referred to as “robo-
signing”.14
Financial institutions were active in this democratic process. The Center for Responsive
Politics reports that the financial sector as a whole spent about $459M on lobbying during 2009-
2010. Padovani and Gibson (2011) examine the lobbying efforts of banks in regards to the Dodd-
Frank bill and find that banks with less traditional businesses, such as securitization, spent more
money on lobbying, and increased their lobbying efforts after the financial reform proposal was
announced.
12 See, e.g., Shepsle and Weingast (1987), Kroszner and Stratmann (1998) for the former, and Stratmann (2002) for
the latter. Evans (2011) provides a recent survey of congressional committees.
13 See Appendix for a timeline of the legislative events that led to the passage of Dodd-Frank Act.
14 See http://archives.financialservices.house.gov/legis111.shtml for the legislative decisions of the House Financial
Services Committee during 111th Congress.
8
2.2 Foreclosure Laws and the Role of Servicers
Servicers of mortgage loans play multiple roles. First, they monitor and receive scheduled
periodic payments from a borrower pursuant to the terms of the mortgage contract. Second, they
transfer these payments to the owner of the loan. Finally, if the loan is in default, the servicer has
the right, subject to contractual limitations with the lender, to engage in a number of loss
mitigation activities. These activities may include loan modifications, pre-foreclosure
dispositions (e.g., short sales or deed-in-lieu arrangements), as well as foreclosures.
The foreclosure process works as follows. Typically, upon 90-day delinquency, the
lender or servicer issues a ‘Demand Letter’ or ‘Notice to Accelerate’. The borrower has typically
30 days to pay the full amount due. If the borrower fails to pay the full amount, the lender or
servicer may initiate the foreclosure process (either through a judicial, i.e., court-supervised,
process or a non-judicial process).15 Lenders are not likely to settle with borrowers for less than
the full amount due.16
It should be emphasized that the start of foreclosure is not triggered automatically for
delinquent loans; rather, it requires action by the loan servicer. Although the borrowers may have
some power to delay the resolution of the foreclosure process, especially in judicial foreclosure
states, they do not have much power in delaying the start other than making the missed
payments.17,18
Mortgage foreclosure laws are set at the state level, and there exists wide heterogeneity in
such laws across states (Pence, 2006; Mian, Sufi, and Trebbi, 2015). Some states require a
judicial foreclosure process, in which the repossession of the collateral property is overseen by
15 See http://www.nmacenter.org/foreclosureprocess.asp.
16 See http://portal.hud.gov/hudportal/HUD?src=/topics/avoiding_foreclosure/fctimeline.
17 See http://portal.hud.gov/hudportal/HUD?src=/topics/avoiding_foreclosure/fctimeline.
18 Delinquent borrowers may also try to persuade the servicer informally to delay the start of foreclosure. However,
to the extent that they also would try to persuade the servicer to accept a ‘short sale’, in which the servicer accepts
the sale of the collateral at a price lower than the outstanding balance without taking the ownership of the collateral,
the delinquent borrowers have not been very successful in these informal negotiations before the start of foreclosure.
Only two loans that meet our screening criteria had short sales before the start of foreclosure and our current sample
of nearly 370 thousand delinquent loans excludes them. The remaining short sales took place after the foreclosure
process started and, hence, after a loan exited from our analysis.
9
the court. Before the start of foreclosure, the servicer may try to negotiate forbearance options
with the borrower. However, if the servicer decides to initiate the foreclosure, the courts in the
judicial foreclosure states start the hearings process to determine the payoff structure for the
various lien holders on the mortgage subsequent to the auction of the property. Other states allow
lenders wide discretion in foreclosure actions although lender actions may still be subject to
judicial review to ensure their legality. In these states, mortgages include a ‘power of sale’
clause, which allows lenders or servicers to repossess and sell the property in the case of default.
Typically, the property is put up for auction and the proceeds go toward paying off the remaining
debt.
3 Data
Our main dataset is comprised of loan-level information provided by residential mortgage
servicers and collected by McDash Analytics. The dataset provides extensive information about
the loan, property and borrower characteristics at the time of origination as well as dynamically
updated loan information subsequent to origination. The servicer or the lender of a loan is not
identified, however. We focus on the loans that became 90-day delinquent for the first time
between January 2009 and December 2010, our sample period. Following the literature, we
restrict our sample to owner-occupied, single family, first lien loans that were originated in 2005
or after. We include all loans that were owned by one of the three major types of investors:
private securitization trusts, securitization trusts by government sponsored enterprises (GSEs)
such as Fannie Mae and Freddie Mac, and the banks (‘portfolio loans’). We drop the loans whose
investor type switches after delinquency; such loans may include the loans that are sold back to
the originator due to fraud or early delinquency as discussed in Piskorski, Seru, and Vig (2010).
We eliminate loans that are at the top or bottom one percent in appraisal amount, Loan-to-Value
ratio, and loan amount, all as of loan origination, to mitigate the effect of outliers.
In addition to loan-level servicer data, we make use of a unique dataset that is constructed
through a high-quality match between loan records and borrowers’ full credit reports. This
10
dataset is known as Equifax’s Credit Risk Insight™ Servicing McDash (CRISM). It is structured
as a borrower-level panel that matches every loan in the McDash servicing dataset with the
borrower’s Equifax credit file. The matching algorithm relies on a wide array of data available
only to the credit bureau such as detailed payment histories, resulting in a high-quality combined
dataset. The credit bureau data span the period during which a mortgage loan exists in the
servicer dataset. This period is further augmented by the 6 months prior to loan origination and
the 6 months following the loan termination (i.e., foreclosure sale or refinancing). For each
borrower associated with a given loan, the dataset contains time series that capture that
borrower’s credit score, total debt amount outstanding in various credit categories, the
breakdown of these aggregates into performing and non-performing components, and category-
specific required monthly payment. In particular, the dataset captures home equity lines of credit,
auto loans (both from banks and captive auto finance firms), and credit cards. Consequently, we
are able to observe performance of existing non-mortgage credit and originations of new credit
for all of the delinquent mortgage borrowers in our main sample.
The finest level of geographic data for loans available to us is their zip codes. We match
all loans to their congressional districts by zip code using the MABLE/Geocorr2K dataset from
the Missouri Census Data Center. We drop loans from the zip codes that match to multiple
congressional districts. We obtain party affiliation and committee assignments of the
representatives from the dataset by Stewart and Woon.19 Out of a total of 435 members of the
House of Representatives in the 111th Congress, 71 served on the Financial Services committee
(the Finance committee, for brevity)42 Democrats and 29 Republicans. Figure 1 provides a
map of committee member districts in the contiguous 48 states; Alaska and Hawaii had no
representation in the Finance committee. The map leads to several observations. First, the
committee members represent a broad geographic cross-section of the US. Second, states with
19 Charles Stewart III and Jonathan Woon. Congressional Committee Assignments, 103rd to 111th Congresses,
1993-2009, available at http://web.mit.edu/17.251/www/data_page.html#2.
11
high proportion of financial institutions such as New York, Massachusetts, and North Carolina
are well represented. Finally, some of the states with high appreciation of house prices before the
crisis such as Nevada and Arizona are not represented in the committee.
Loan-to-Value (LTV) of a loan, both at origination but also especially during the loan’s
life is likely to be an important factor in servicer’s decision to foreclose when the loan becomes
delinquent. Unfortunately, only LTV at origination is provided in our data set. To estimate LTV
during loan’s life, we use zip code-level Home Price Index provided by CoreLogic. This index is
available for at least one zip code for each congressional district except for 19 congressional
districts, one of which is represented in the Financial Services committee. Consequently, the
loans from those districts are omitted from the analysis. Table 1, Panel A gives the number of
loans and the number of districts by party affiliation and committee membership.20 One key
takeaway from this table is that the fraction of loans from Finance committee districts (17.2%) is
very similar to the fraction of the Finance committee member districts (16.9%).
Appendix A provides the description and source of our variables and Table 1 Panels B
and C present summary statistics for zip code-level demographic variables and for borrower-
level variables, respectively, with the latter provided by CRISM. The statistics are reported for
the overall sample as well as by the districts of committee and non-committee members. The
differences in subsample averages are also provided. The standard errors for the differences in
means are corrected for clustering at the congressional district level.
Table 1, Panel B reports sample statistics for demographic variables such as median
income, the fraction of African Americans and Hispanic households, both from American
Community Survey, unemployment rate from Bureau of Labor Statistics, and the fraction of
urban population from the 2000 Census. If the data are provided at the county level, we used the
county-level data. The average for demographic variables is not statistically significantly
20 One district that switched from Republican to Democratic midterm after special elections due to vacancy is
omitted from the analysis. One independent member of the U.S. House of Representatives caucuses with Democrats
so he is included as ‘Democrat’.
12
different from each other based on the Finance committee membership with the exception of the
median household income and the urbanization, both of which are significantly higher in
committee member districts at the 10% and 5% level respectively. The table also provides
sample statistics on the decrease in the home price index from the loan origination date to the
delinquency date using the zip code-level single-home residential price index from Corelogic.
This variable is updated monthly with the contemporaneous index in the regressions. Loans in
both non-committee and committee member districts experienced similar decline in house prices
at 33.5% and 31.6%, respectively, as of the 90-day delinquency date; the difference is not
statistically significant.
Table 1, Panel C provides sample statistics at the borrower level obtained from CRISM.
We use data on the borrower’s FICO score and total non-mortgage borrowing 6 months before
the onset of 90-day mortgage delinquency. We observe no statistically significant difference in
sample means between committee and non-committee districts.
Table 1, Panel D provides sample statistics on loan-level variables. As we only have data
on the first-lien Loan-To-Value (LTV) ratio as of loan origination, we estimate the
contemporaneous LTV by updating the house value using a zip code-level price index. Although
Table D reports this variable as of the month of delinquency, this variable is updated in
regression analysis with contemporaneous index values to capture the continuing changes in
house prices during the sample period. Not surprisingly for a sample restricted to delinquent
loans, the estimated first-lien LTV ratios are very high, averaging 109 percent.
Importantly, there is no statistical difference in terms of subsample averages based on
committee membership for any of our variables as indicated in the Difference column. Notice
that the standard errors reported for those differences are robust to clustering at the congressional
district level. Specifically, looking at loan amounts, FICO scores at origination, estimated LTV,
interest rate, and various measures of mortgage types (jumbo, interest only, and subprime) there
is no difference across non-committee and committee member districts.
13
Table 1, Panel E presents the mean time from the onset of 90-day delinquency to the start
of foreclosure. These are ‘restricted’ means in the sense that they ignore the fact that for many
loans the foreclosure process does not commence by the end of the sample period. In other
words, these simple calculations ignore the resulting right censoring and underestimate the mean.
Nevertheless, the average time to the start of foreclosures is longer in the committee districts and
the 95% confidence intervals do not overlap.21 Perhaps more informatively, we also test for the
equality of the distributions of time to foreclosure starts in committee and non-committee
districts. The last column in Table 1E provides the chi-square test and its p-value. The test
indicates that the distributions of time to the start of foreclosure in committee and non-committee
districts are different and the difference is highly statistically significant.
4 Regression Analysis
4.1 Empirical Approach
Our identification strategy relies on studying the impact of a political factor that is
unlikely to proxy for any economic factor that might affect the servicers’ foreclosure decision.
The Financial Services Committee membership of the congressperson in whose district the
mortgage is located is likely to be such a political, but not economic, factor for several reasons.
First, unlike party affiliation, which might proxy for many economic and social factors due to
gerrymandering of congressional districts, committee memberships are not tied to districts. If an
incumbent loses an election, the newly elected politician in that district does not inherit the
incumbent’s committee seat. Second, each member’s likelihood of becoming the committee
chair or ranking member depends on the member’s seniority within that particular committee, not
on their seniority in the House of Representatives. This rule discourages many members to self-
select themselves into the Finance committee from their previous committees once the crisis
started and the foreclosures increased in their districts. Third, because of the aforementioned
21 These statistics are estimated using the ‘stci, rmean’ command of Stata v14.2; no cluster robust standard errors are
available for this command.
14
seniority rules, most members in the Finance committee became members for the first time well
before the onset of the crisis. We perform further checks on this point in the robustness section.
Finally, the role of committee membership, especially the role of Finance committee
membership on the financial sector has already been well studied though not in this context.
Hence, any impact of Finance committee membership on foreclosure starts is very likely to be
due to political, not economic, reasons.
Focusing on the Finance committee membership also has disadvantages. First, our
identification of politically motivated (in)action by the servicers is relative to their actions in
non-committee districts. This does not rule out any political action by the servicers in those non-
committee districts. We only assume that any political motives by the servicers are likely to be
stronger in Finance committee districts than in non-committee districts because the former has
more power over the banks as shown in the literature in other contexts. To the extent that some
of the servicer behavior is also politically motivated in non-committee districts, our methodology
underestimates the politically-motivated foreclosure delays in the country. Second, some states
and their senators may have greater political power, which might also affect the servicer
behavior. As typical in political economy literature, we choose to use state fixed effects—or,
rather, state*time interaction fixed effectsto control for all state-level differences. In that
respect, our approach will again underestimate the political effect on foreclosures. Finally, the
majority party has more power in the House of Representatives. Unfortunately, congressional
districts are not drawn randomly and the party affiliation of the district representative may proxy
for unobserved demographic factors. Therefore, we control for the politician’s party affiliation in
the analysis but we choose not to interpret its coefficient in political terms. In other words, to the
extent that servicer behavior is different in the districts of majority party members, our analysis
will capture it but not interpret it as political delay.
Our null hypothesis is that the servicer decision to initiate the foreclosure process on a
delinquent loan does not vary according to whether the loan is in the district served by a member
of the Finance Committee. To test whether our null hypothesis can be rejected, we focus on the
15
rate of foreclosure starts of 90-day delinquent mortgages conditional on the foreclosure not being
initiated yet, namely, the hazard rate of foreclosure starts. Our main empirical specification is an
exponential hazard model of the servicers’ foreclosure decision for a loan that becomes 90-day
delinquent for the first time in month t during the period spanning January 2009 through
December 2010:
()
iTistiit tttmitteeFinanceComt
h...,*exp)
(0
=++=
θγβ
x
, (1)
where FinanceCommitteei is a binary variable that is set to one if the loan is located in a district
whose Representative is a member of the House Financial Services Committee; xit is the vector
of explanatory variables for loan i in month t. It includes both loan level and zip code level
variables that are as of origination or as of month t as well as the party affiliation of the district’s
representative; θst are state-specific month fixed effects.22
The loan i enters the study in month ti0, which is the first occurrence of the 90-day
delinquency status for that loan. The same loan exits the study in month tiT, which is the earliest
occurrence of the start of foreclosure or one of the exit events such as becoming current or no
longer being reported. Finally, since the servicer decision in a given congressional district may
not be independent from another decision within the same district, the error terms in the
regression above may be correlated within a district, which would lead to underestimated
standard errors for the coefficients. Hence, following Bertrand, Duflo, and Mullainathan (2004),
all the errors reported in this study are corrected for clustering at the congressional district level.
Loan-level controls are motivated by the literature. They include indicators for FICO
score above 680 or between 620 and 680 (omitted category: FICO<620), debt-to-income ratio at
origination, indicators for fixed rate and interest only loans (omitted category: Adjustable Rate
Mortgages), indicators for full documentation and unknown documentation (omitted category:
22 The Cox hazard model, which allows for arbitrary duration dependence, takes a prohibitively long time to
converge with state and month fixed effects for us to adopt it as the main model throughout the paper. We still
repeat the main regression using the Cox model and find our results to be very robust, as reported in the Robustness
section.
16
no/low documentation), and indicators for jumbo, low grade, refinance loans and securitization
status (GSE or private label, omitted category: portfolio loans). Following the literature, we also
include an indicator variable for LTV at origination equal to 80% as a proxy for the existence of
a second lien on the property. Continuous loan-level variables include (log of) loan amount, first
interest rate observed, elapsed time from origination to the first classification as 90-day
delinquent, and the decrease in the residential home price index for that zip code since the loan
origination.
Since our main hypothesis variable is based on politics and geography, our regressions
also include zip code-level demographic variables in addition to the party affiliation of the
congressperson representing that zip code. These variables include (log of) median household
income, shares of African Americans and Hispanic households, unemployment, urbanization and
the decrease in the home price index since the loan origination.
4.2 Main Result
Our main result, showing that foreclosure starts are delayed in the districts of Financial
Services committee members, is presented in Table 2. Column (1) serves as the benchmark; it
includes all our control variables, including party affiliation, but not the Financial Services
Committee membership.
The results from the baseline specification are largely intuitive and fall within the broader
context of the literature evaluating servicer decisions to modify or foreclose delinquent mortgage
loans. The discussion in this literature focuses on servicer ability to identify loans that are either
likely to self-cure in the absence of any actions or would redefault even in the event of
modification (e.g., Piskorski, Seru, and Vig, 2010; Agarwal, Amromin, Ben-David,
Chomsisengphet, and Evanoff 2011; Adelino, Gerardi, and Willen, 2013; Zhang, 2013). These
studies show also that foreclosure decisions are influenced by a number of agency frictions that
create conflicting incentives between owners of the loans (GSEs or private investors) and entities
making modification/foreclosure decisions (servicers). Consistent with these studies, we find that
17
securitized loans are foreclosed more quickly than portfolio loans. The same is true for low-
grade loans, loans underwritten without documentation, and non-fixed rate and non-amortizing
loans, all of which suggest ex ante higher risk of redefault. These findings are consistent with
Agarwal, Amromin, Ben-David, Chomsisengphet, and Evanoff (2011), who use a more detailed
dataset and find that these borrowers are more likely to experience foreclosure and less likely to
be offered loan modifications.
Column (2) of Table 2 includes only party affiliation and our main hypothesis variable,
Finance Committee, which is set to one if the congressperson representing that zip code is a
member of the House Financial Services Committee. The latter has a negative coefficient that is
statistically significant at the 1% level. This suggests that the loan servicers delay the start of
foreclosure for a delinquent loan if the loan is located in the district of a Financial Services
committee member. Note that the analysis also controls non-parametrically for all the time-
specific state-level factors through state*month fixed effects so this result is unlikely to be
capturing any state-level factors. Also, the errors are robust to clustering at the congressional
district level so the results are unlikely to be due to the large number of observations.
The regression in column (3) adds all the loan-level and demographic variables. The
Finance Committee variable again has a negative coefficient that is statistically significant at the
1% level, which confirms the foreclosure delays in committee districts. The magnitude of the
delay in committee members’ districts relative to non-committee members’ districts is about
3.7% (=1 exp(-0.038)). Finally, the regression in column (4) repeats that in column (3) with
only state and month fixed effects rather than state*month interaction fixed effects as some of
the estimation presented later did not converge with the latter set of fixed effects. The results are
again both quantitatively and qualitatively very similar.
Notice that the magnitude of the coefficient for the Finance Committee variable changes
little from columns (2) to (3) or (4) even though the latter specifications include many
demographic and loan-level variables suggested in the literature. This suggests that committee
membership of the politician is indeed orthogonal to many loan and borrower features as well as
18
zip code demographics. On the other hand, the coefficient of the majority party indicator, which
may proxy for demographic and economic factors unlike the committee membership indicator,
changes substantially and loses significance. In other words, both for the institutional reasons
that determine the committee membership and for the presence of many financial and
demographic control variables, the effect of committee membership we find is very unlikely to
be due to any economic reason. Instead, the delay we find in these districts is likely to be due to
political reasons.
4.3 Time to Foreclosure and Economic Significance
The hazard model presented above is the most common approach to survival time
analysis but its interpretation may not be very transparent. In this subsection, we focus directly
on the time it takes from the onset of 90-day delinquency to the start of foreclosure. More
precisely, we estimate the following regression:
,
(2)
where Ti is the time from the onset of 90-day delinquency to the start of foreclosure for loan i. xi
includes all the control variables in the previous hazard estimation. All of the time-varying
variables are measured as of the onset of delinquency. The model also includes fixed effects for
state and the first month of delinquency.
The OLS estimation of (2) cannot distinguish the exit reason of a loan after T. In
particular, unless the foreclosure process commences for all delinquent loans before the end of
our sample period, OLS will underestimate the time-to-the-start-of-foreclosure. However, a loan
may exit the analysis after T not only due to the start of foreclosure but also because a loan may
become current (self-cure) or because we do not follow the loans beyond the end of our sample
period in December 2010 (right censoring). Hence, the proper estimation needs to take the exit
reason into account. Consequently, regression (2) is estimated by Maximum Likelihood with the
likelihood function that incorporates the exit reason for a loan after T. Naturally, the likelihood
19
function depends on the assumed distribution for the error term
ε
i. We estimate (2) by using four
different distributional assumptions from the literature, namely, normal, logistic, extreme value,
and extreme value with a scale parameter.23,24 Note that the latter two distributions imply
exponential and Weibull hazard functions, which further relate this analysis to the one in the
previous section (Table 2). The regressions include state and month of delinquency fixed
effects25 and standard errors are again clustered at the congressional district level.
Table 3 presents the results of this analysis. The estimated coefficient for the Finance
Committee variable is positive and statistically significant at the 1% level, which indicates longer
time to foreclosure for delinquent loans in the districts of finance committee members. The
magnitude of the coefficient estimates range from 0.037 to 0.044. Given the logarithmic form of
our dependent variable, these estimates suggest about 3.7-4.4% increase in the time to
foreclosure in committee districts.
The estimated mean survival time to the start of foreclosure for the full sample ranges
from 12.1 to 33.8 months depending on the assumed distribution for the error term. The
estimated marginal increase in the mean survival time for the committee districts ranges from 0.5
to 1.5 months.26 Since the estimates of the mean may be sensitive to the distributional
assumptions for the right tail of the error term, we also report the predicted median time, which
is a more robust indicator of centrality. The estimated median survival time for the full sample
ranges from 7.7 to 8.5 months, while the estimated marginal increase for the median survival
time in the committee districts ranges 0.3 to 0.4 months – again a non-negligible marginal effect.
We also use these results to estimate the dollar amount for the cost of delay to the
financial sector and compare it to the campaign contributions to the Financial Services
23 See Klein and Moescherber (2013, pp. 45-49) and, for a more applied approach, Cleves et al. (2004, pp. 222-243).
24 The estimation was performed using Stata’s ‘streg’ command. Option ‘time’ was specified to indicate the use of
accelerated failure time metric for the latter two distributions.
25 Margin calculations to estimate the mean and median survival time in committee districts below did not converge
with state and delinquency month interaction fixed effects.
26 These are the sample mean and median of marginal effects at the loan level, not the ‘margin at the mean/median’.
They are estimated using Stata’s ‘margins’ command taking into account that Finance Committee is a discrete
variable.
20
committee members by the largest banks (servicers) that determine the delay. This estimate will
naturally be imprecise because we lack information on when the banks expected to complete the
foreclosure process and what amount they expected to recover from foreclosures. Their
expectations, unobservable to the econometrician, are related to their expectations about future
house price dynamics, which are also unobservable.
Any cost of delay to the lenders is also likely to be underestimated because our test
design is geared towards identifying the political delays in the foreclosure starts, not towards
quantifying the total effect of political motivations in the foreclosure starts. In particular, our
methodology measures delay in committee districts relative to non-committee districts and thus it
is silent about the possibility of political delays in the non-committee districts as well.
Despite these problems about precision and underestimation, a proverbial “ballpark
figure” may still be informative and we proceed by using the existing work on this topic. Cordell,
Geng, Goodman, and Yang (2015) focus on direct foreclosure costs that vary with the time the
loan remains in delinquency and foreclosure. These costs include hazard insurance, property
taxes, maintenance and repair, and increased depreciation in house value (Melzer, 2017). Their
calculation controls for fixed foreclosure costs and excludes any negative externalities to the
neighboring properties. These delay costs are ultimately borne by the lenders but the servicers
may have to make these expenditures until liquidation and recover them only at liquidation. They
estimate the total foreclosure delay costs to be, on average, about 18% of the unpaid loan balance
and the average time from the start of foreclosure to liquidation to be 32.1 months during the
period from November 2008 to August 2010. These figures rise to 20% of the loan balance and
33.6 months during the September 2010January 2012 period.27 We use these costs and length
of time as the basis for our calculation. For the average delay in committee districts, we use 0.5
months, the lowest estimate in Table 3. Assuming that the total costs of foreclosure delays in
27 See Tables 3 and 6 in Cordell, Geng, Goodman, and Yang (2015). The length of time spent in foreclosure is
derived by constructing a weighted average of Cordell et al. estimates for judicial and non-judicial states’ timelines
using our sample composition.
21
Cordell, Geng, Goodman, and Yang (2015) increases linearly by time, we arrive at the monthly
cost estimate of $280-$300 per $100,000 remaining loan balance on average.28 Even with an
aggressive assumption of 20% annual cost of capital for the lenders, the present value of monthly
delay costs as of the onset of delinquency and evaluated at the average time from delinquency to
foreclosure and to liquidation is about $170-$180 per $100K of delinquent loan balances.29
Although this may not be a large amount per loan, with $17.1B in outstanding delinquent
balances in districts of the Financial Services committee members in our sample, it implies a
total non-discounted cost of about $48M to $51M at the time of delinquency (or about $30M
using the 20% cost of capital assumption). As one benchmark for comparison, the largest ten
residential mortgage servicers,30 which service the large majority of the loans in our sample,
made campaign contributions of $980,000 to the Financial Services committee during our
sample period.31 Their total lobbying expenditures for the Legislative (both chambers in
Congress) and the Executive branch of the government was about $44M in 2009 and 2010. 32 In
other words, the total in-sample cost of delay is an order of magnitude larger than the campaign
28 (0.5*100000*0.18/32.1) and (0.5*100000*0.20/33.6), respectively. The lower figure is based on liquidation in the
earlier period while the higher figure is based on the liquidation in the latter period as discussed above.
29 We do not know the ultimate lenders except that about 10% of loans are portfolio loans owned by the servicers
themselves. We adopted a high discount rate to emphasize the fact that the qualitative comparison of foreclosure
delay costs to political expenses by these banks, as described below, is not very sensitive to cost of capital estimates
within the likely range.
30 As the mortgage servicing industry is fairly concentrated, the top ten servicers account for the vast majority of
loans. This is reflected in all loan-level servicer datasets. As reported by Mortgage Servicing News (2010), the ten
largest servicers are Bank of America, Wells Fargo, JPMorgan Chase, Citigroup, Residential Capital (formerly
GMAC), U.S. Bank, Sun Trust, PNC, PHH, and OneWest.
31 Based on the official disclosure reports as provided by the Center for Responsive Politics through its website
opensecrets.org. The size of campaign contributions of large banks may seem small relative to the size of the
mortgage market but they are in line with total corporate campaign contributions relative to the size of the U.S.
economy, see Ansolabehere, de Figueiredo, and Snyder (2003).
32 The lobbying disclosure filings are much less detailed than those for the campaign finance; the former do not even
disaggregate between the expenses for lobbying the legislative branch from those for the executive branch let alone
a specific politician. During the 2009-2010 two-yearly cycle, these ten servicers spent about $44M for lobbying
using the data from the Center for Responsive Politics through its website opensecrets.org. Just to obtain an estimate
for expenses the banks incurred lobbying the committee members, one may assume that none of their lobbying was
for the executive branch (probably a very strong assumption) and that the expenses incurred for lobbying the House
Finance committee members was 34.3%, the same fraction of their campaign contribution to the committee
members relative to their total contribution to all the House and Senate members. With these assumptions, the
expenses incurred by these servicers for lobbying the House Finance committee members can be estimated to be
about $15.1M or about half of the estimated cost of delaying the foreclosure starts.
22
contributions of the large servicing banks to the Financial Services committee members during
our sample period; in fact, it is more comparable to the lobbying expenses by these firms.
The fact that these costs are substantially greater than the campaign contributions has
important political economy implications. Corporate lobbying expenses faced little restrictions
but corporations were subject to limits in their campaign contributions during this period.
However, lobbying expenses could not legally be channeled to politicians. The delays in
foreclosure starts we document could, on the other hand, be targeted directly to the constituents
of particular politicians. In fact, in some respects, they could be even better targeted to their
constituents than some of the mass media political advertisements for which campaign
contributions are used to pay, as discussed in the Introduction.
The calculations above are necessarily imprecise. However, they also suggest that, if the
campaign contributions by the largest banks to the Financial Services committee members or
their lobbying are politically important as shown in the political economy literature reviewed in
the Introduction, the delay in foreclosure starts in the district of committee members is also
politically important.33
5 Robustness
We conduct several robustness tests for our results. First, we run a series of placebo tests
where we explore foreclosure patterns for members of other committees and for the Financial
Services committee members in past periods. Second, we test whether we see a similar impact
for the onset of delinquency. Third, we discuss whether committee members may be self-
selecting into financial services committee. Fourth, we check the robustness of our results to
alternative econometric procedures. Finally, we test whether the results hold in subsamples.
33 Although our sample includes all mortgages in McDash that became delinquent in 2009-10 and passed the screens
described in the Data section, the data coverage is not universal. The approach of comparing the cost of delay in our
sample to the total political expenditures by these servicers avoids extrapolation of results out of sample but it is
conservative.
23
5.1 Placebo tests: Other Congressional Committees
We conduct several falsification tests. In the first falsification test we explore whether
differences in the rate of foreclosure initiations are affected by the membership on Congressional
committees that have no ostensible link to housing markets. The three largest committees in the
U.S. House of Representatives in the 111th Congress were Transportation & Infrastructure,
Financial Services, and Armed Services, in that order. It is unlikely that banks would choose to
adjust their servicing practices in the districts of the Armed Services or the Transportation
Committee members. The membership in those committees can serve as a good placebo to our
analysis because they have very few common members with the Financial Services committee.34
Using the membership on each of these committees in turn, we re-estimate our main
specification. The results are presented in Table 4, Columns (1) and (2). We detect no
statistically significant effect of membership in the defense or the transportation committee on
servicers’ foreclosure initiation decisions.
5.2 Placebo tests: Past periods
Next, in Columns (3) and (4) of Table 4, we also use foreclosure choices for loans that
became delinquent in previous Congresses as placebos. Our focus is on the 109th (2005-2006)
and 110th (2007-2008) Congresses, as delinquency rates during those years were not as high as in
2009 and 2010 and had not yet become as political an issue as during our main study period.
Hence, any bank that was active in mortgage servicing was much less likely to adjust its
foreclosure actions to mitigate any legislative concerns it had. Moreover, in previous sessions of
Congress, the Financial Services committee did not consider any legislation that was nearly as
important as that taken up by the 111th Congress.
We repeat our main regression specification from Table 2 first for 2007 and 2008 years
focusing on loans that become 90-day delinquent for the first time during this period. We then
34 After eliminating the few members who did not stay in the House or in the Committee for the full term, Financial
Services committee has only 4 members who are also in the Transportation committee and only one in the Armed
Services committee.
24
repeat the analysis for the 2005-2006 period. In both cases, we reconstruct our political economy
variables to reflect contemporaneous party affiliation and committee membership. The results are
reported in Table 4, columns (3) and (4). As expected, the foreclosure decisions on delinquent
loans are not measurably different in districts of the Financial Services committee members. The
results show that coefficients are statistically insignificant and, for the 2005-2006 period—have
the opposite sign.
5.3 Self-Selection into Committee Membership? Analysis of Delinquency Rates
One possible concern is whether legislators self-select themselves into the Financial
Services committee based on foreclosure practices in their districts. The reverse causality is not
very plausible; legislators are unlikely to choose to be a member of the Financial Services just
because banks delay foreclosing on their delinquent voters. However, an unobservable factor
driving both the committee membership decision of legislators and the foreclosure decisions of
banks would present a bigger concern. For example, if delinquency rates are expected to be high
in a district, the legislator representing the district may want to be in the Financial Services
Committee. An increase in delinquencies in committee districts in itself is not sufficient to
observe the pattern we document because our analysis is conditional on the onset of delinquency.
However, if the foreclosures are delayed later due to the large volume of delinquencies in that
district, we can then see the effect we observe above.
To rule out this possibility, we extract all the loans that are current and have never
experienced delinquency as of December 2008 and follow them through December 2010. We
study their rate of reaching 30-day and 90-day delinquency in a hazard model.35 We use the same
control variables as in our main analysis. The results are reported in Table 5. We find no
statistical difference in the rate of delinquencies in districts of Financial Services committee
35 Note that our main sample of first-time 90-day delinquent loans is not a subset of this sample of current loans with
a three-month lag. Many of the delinquent loans in our main sample were 30-day delinquent for the first time more
than two months before their first 90-day delinquency; they became current again by making payments in arrears
before reaching 90-day delinquency that leads to the loan’s inclusion in our main sample. The loans with such
delinquencies in their past are excluded from the analysis in this subsection.
25
members. In other words, our results are unlikely to be driven mechanically by differential rates
of delinquencies in committee members’ districts or by the self-selection of legislators based on
expected delinquencies.
5.4 Self-Selection into Committee Membership? Analysis using Senior Committee
Members
The institutional features of congressional committees provide us with another test for the
possibility of self-selection by the legislators into the Financial Services committee. Senior
committee members tend to be more powerful and legislators receive seniority based on their
service tenure in a given committee. When legislators leave or switch their committee, they lose
their seniority in the committee even if they remain in the Congress. In fact, all 32 members of
the Financial Services committee in the 108th Congress (2003-2004) that were still in the U.S.
House in the 111th Congress remained on the Committee. In other words, many members of the
2009-2010 Financial Services committee made the decision to be in the committee well before
the foreclosure crisis. We use this feature to address the concerns of self-selection into the
Financial Services committee. Whatever the economic determinants of the banks’ foreclosure
decision might be in 2009 and 2010, they are likely to be orthogonal to the committee
membership decision made by the politicians years before.
The regression in column (1) in Table 6 repeats our main specification by excluding all
the Financial Services committee members who joined the committee in 2005 or later. The
second column repeats our main specification by excluding all the representatives who first took
office in the House in 2005 or after. Our main results remain robust to these exclusions; the
foreclosure of delinquent loans is delayed in the districts of Financial Services committee
members. In fact, the magnitude of this political effect becomes larger, which suggests that the
political effect we detect may be stronger in the districts of senior members.36
36 Notice that, in the second column, which focuses only on senior politicians, the coefficient of the majority party is
also negative and very significant. This indicates that foreclosure starts were delayed in Democratic districts and this
delay may also be, at least partially, due to political reasons because more senior members of the majority party,
26
5.5 Alternative Time Periods
Our sample lasts until the end of 2010 but in the latter part of that year the so-called
“robo-signing” practice came to widespread attention. This practice refers to cases where the
large servicing banks seemed to process large numbers of foreclosure documents in a short time
without legally required individual attention. Although it would be interesting in itself if the
servicing banks reacted differently in committee districts to the attention by the media and
politicians on this issue, it is also desirable to check the robustness of our results to the exclusion
of that period. We repeated our main regression by stopping the analysis period first at the end of
December 2009 and then at the end of March 2010, both of which were well before the robo-
signing attracted attention. We present the results in Table 7, Columns (1) and (2). Our earlier
results remain robust both quantitatively and qualitatively to excluding the latter part of our
sample period. It is also worth noting that all of our specifications include state-month
interaction fixed effects, which would pick up state-level delays caused by robo-signing.
5.6 Alternative Econometric Methods
We also check the robustness of our results to different econometric methods. First, we
estimate our main regression in Table 2, Column (3) using the Cox proportional hazard model.
The Cox hazard model has the advantage of allowing arbitrary duration dependence (baseline
hazard). We report the results in Table 7, Column (3). The coefficient of the finance committee
indicator is very similar to that reported in Table 2 both in magnitude and in statistical
significance. Next, we repeat that regression using Weibull distribution and report the results in
Column (4). We again obtain very similar results both quantitatively and qualitatively.
We also estimate a discrete hazard model based on the linear probability model given
below:
even outside the Finance committee, are likely to have more power over legislation of interest to the banks.
However, we choose not to interpret the statistical significance of the party affiliation coefficient as evidence of
political delay. Unlike the committee membership, party affiliation may proxy for demographic and economic
factors that may also affect bank decisions.
27
iT
istiitit tttmitteeFinanceComy ...,* 0
=++=
θγβ
x
, (3)
where yit is a binary variable that is one if the foreclosure starts for loan i in month t. This model
may not properly take right censoring into account and face the usual problems of estimating
probabilities using OLS but it also provides a robustness test for the distributional assumptions in
the previous analysis. The results, presented in Column (5) of Table 7, show that our previous
results are robust to this alternative econometric specification.
5.7 Nearby Zip Codes Only
A possible concern for our results is that we may be comparing the foreclosure rates in
non-committee member districts that may be rural to those in committee-member districts that
may be urban within the same state despite including the urbanization rate of the congressional
district in the control variables. Although we control for the proportion of urban population in the
regressions, since the housing crisis affected urban areas disproportionately, it is still important
to check the robustness of our results by limiting the comparison only to the zip codes that are
near one another. Hence, in Table 7, Columns (6), (7), and (8) we estimate our main regression
in Table 2, Column (3), but we now require all the non-committee zip codes to be within 10, 25,
and 50 miles of a committee zip code in the same state and vice versa. This significantly reduces
the sample size. For example, our 10-mile sample in Column (6) is less than a quarter of the
original sample we have for our main regression in Table 2, Column (3). However, in all three
specifications our results are both economically and statistically significant.
5.8 Subsamples by Loan Characteristics
We also examine subsamples by loan characteristics. In Table OA-1 in Online Appendix,
we purge subsets of the sample. These columns show that the results hold in all the subsamples.
In Column (1) we examine only mortgages that are classified as ‘grade A’ (or non-subprime) by
the servicers. In Column (2) we restrict the sample to loans with maturity of 15, 20, and 30 years
at origination. These are the most common maturities and the loans with different maturities may
be non-standard on other dimensions as well. In Column (3) we drop from the analysis October
28
2010 and afterwards. That period includes both the ‘lame duck’ months of November and
December 2010 as well as October 2010 when the news about the so-called “robo-signing” in the
foreclosure process broke out. Our results remain robust in all the subsamples; foreclosures of
delinquent loans are delayed in the districts of the Financial Services committee members and
the magnitudes are comparable across these specifications.
Our identification strategy is already based on a factor that has a very political nature but
little, if any, economic nature. These robustness checks allow us further to rule out economic
explanations for the delay in foreclosures. Hence, we interpret the delay in foreclosure initiations
as due to loan servicers' political concerns.
6 Welfare and Potential Economic Channels
It is conceivable that loan servicers granted foreclosure delays in the Financial Services
Committee member districts in order to allow the delinquent borrowers residing in such districts
to regain their financial footing and cure their delinquencies. If this were the case, we would
expect to find superior performance on existing non-mortgage credit obligations among
delinquent mortgage borrowers in such districts. Similarly, we would expect to find that these
borrowers were also more successful in obtaining new non-mortgage credit. The CRISM dataset
that uniquely links mortgage loans with the rest of the borrowers credit records allows testing of
this hypothesis. In particular, we analyze performance and origination of credit in several
categories: auto loans, credit cards, as well as student loans and retail (store card) credit for the
borrowers whose mortgages are 90-day delinquent during 2009-10 and thus form the sample for
our main analysis above.
6.1 Summary statistics
We begin our analysis by documenting the relative importance of various categories of
non-mortgage borrowing in Panel A of Table OA-7 in Online Appendix. Credit card balances
account for nearly half of outstanding non-mortgage debt among the delinquent borrowers in our
sample while auto loans make up about 29% of non-mortgage balances. More importantly, there
29
are no statistically significant differences between shares of various non-mortgage obligations
across member and non-member districts.
Panel B reports delinquency rates on non-mortgage credit for borrowers in our sample, all
of whom are delinquent on their mortgages at the time they enter the sample. The panel thus
documents incidence rates of non-mortgage delinquency during the two-year period following
mortgage delinquency. Starting with auto loans, we find an incidence rate of 2.54% for auto loan
delinquency where the incidence rate is the number of loans that fall into delinquency divided by
the total number of months loans remain in our sample until delinquency. The incidence rate is
very similar across committee member and non-member districts. The incidence rate of credit
card delinquencies is somewhat higher at 3.85% for the entire sample. However, the incidence
rates are virtually identical between the member and non-member subgroups. We find similar
results for all other non-mortgage loans, which span student debt, retail, and consumer finance,
as well as for the aggregate encompassing all non-mortgage loan obligations.
Panel C repeats the analysis of incidence rates but focuses on origination of new credit, as
opposed to performance of existing obligations. The data suggest that mortgage delinquency
does not necessarily exclude a borrower from participation in credit markets. The incidence rates
for new auto loans and credit card accounts stand at 11.4% and 15.0% of borrower-month
observations, respectively. The incidence rates of new instances of other non-mortgage debt are
higher still at 21.6%. This finding is not surprising, since this category includes student debt,
most of which is not underwritten and provided independent of borrower credit standing. The
differences in rates between member and non-member districts are mixed: member districts
display higher incidence of new auto loans, but lower incidence of new bank credit cards, and
almost identical incidence of other non-mortgage debt. However, none of these differences
surpasses conventional levels of statistical significance.
30
The summary statistics suggest that delinquent borrowers in Committee member districts,
which were shown to experience foreclosure delays in the earlier analysis, did not experience
favorable credit outcomes in non-mortgage categories. We next turn to a more rigorous
regression-based hazard analysis to evaluate the validity of these results.
6.2 Hazard regressions - Non-Mortgage Loan Delinquencies
Table OA-8 presents the results for non-mortgage performance utilizing the hazard
regression framework developed in section 4.1. In each of the specifications, we estimate the
hazard rate of being 90-day delinquent on a given type of existing non-mortgage obligations.
Each specification employs a set of loan, geography, and time controls that are identical to those
described in section 4.1. For all the loan categories except auto loans, credit cards the main
hypothesis variable, Finance Committee member, is a statistically insignificant factor in the
hazard analysis. Moreover, the point estimates of this variable change signs from one category to
the next. For auto loans, the coefficient is positive and statistically significant at the 10% level,
which indicates that these loans are more likely to become delinquent in committee districts; this
is the opposite of what might be expected if banks were delaying foreclosures to increase the
welfare. In other words, there is no evidence that the foreclosure delays in committee district
allowed delinquent mortgage borrowers in those districts to stay current in non-mortgage loans.
It is possible that any possible benefits from foreclosure delays may have been
concentrated among the most vulnerable borrowers so we repeat our analysis in that subgroup
and report our results in Table OA-9. Panel A only considers borrowers characterized by low
(below median) FICO score at the time of sample entry, while Panel B only looks at borrowers
who failed to fully document their income when they obtained their mortgage.37 Both of these
groups can be thought of as potentially most vulnerable, and might therefore have a different
response to foreclosure delays in committee member districts.
37 We also estimate a specification on a subsample of borrowers defined by below-median FICO scores at the time
of mortgage origination. The results are both qualitatively and quantitatively similar.
31
In either of these subsamples, we fail to find a beneficial (negative) effect of being in the
Committee member district. The only statistically significant results for low-documentation
borrowers – that for auto loan delinquency and for all other loans point in the opposite
direction. In other words, there is no evidence that foreclosure delays in committee districts
allowed delinquent mortgage borrowers to stay current in non-mortgage loans either in the total
sample or in the subsample of most credit-constrained borrowers.
6.3 Hazard regressions - Obtaining New Non-Mortgage Loans
Turning our attention to the origination of new credit, we present the results of hazard
analysis in Table OA-10. Once again, we fail to detect a statistically significant beneficial, this
time positive, effect of being in a Committee member district. We repeat this analysis for
borrower subsamples in the same fashion as was done for non-mortgage delinquencies. The
results in Table OA-11 also indicate the absence of any measurable effect of being in the
Committee member district.
To summarize, the results of the analysis in this section fail to provide support for the
hypothesis that foreclosure delays in Committee member districts led to welfare gains for the
affected borrowers. In particular, there is no evidence for improved performance on non-
mortgage credit or for obtaining new non-mortgage credit. Instead, the results confirm that
servicers were motivated by political, not economic, reasons in delaying the start of foreclosures
in the districts of Finance Committee members.
6.4 Potential Economic Channels
Having documented the delay in foreclosure initiations, we attempt to disentangle
potential economic mechanisms by exploiting several sources of cross-sectional variation.38 In
particular, we consider the role of banking market competitiveness, consumer protection laws,
degree of political competitiveness, and costs of foreclosure delays. Heterogeneity in these
38 We thank a referee for suggesting these tests.
32
measures across states may be reflected in potential costs and benefits of foreclosure delays. The
results of this investigation are summarized in the Online Appendix Table OA-12.
We start with a hypothesis that more competitive banking markets may motivate
servicers to delay foreclosures as they fight to enhance their relative competitive position. To test
this, we define the degree of banking market competitiveness by computing MSA-level
Herfindahl-Hirschman index values using FDIC Summary of Deposits data for 2008. We then
split the sample along the median HHI value and estimate our baseline specification (equation
(1) as implemented in regression (3) in Table 2) for each of the resulting subsamples. We find
that foreclosure delays are somewhat higher and more precisely estimated in less-competitive
markets, inconsistent with the null hypothesis.
To further exploit cross-sectional variation in costs and benefits from delaying, we
partition the sample between states with judicial foreclosure review processes and states with
non-judicial foreclosures. The former group has experienced much lengthier foreclosure
timelines, and so in relative terms, a politically-motivated delay is less costly, However, we find
statistically significant evidence of politically-motivated delays in both types of states (Table
OA-12, columns (3)-(4)). The strength of this effect is somewhat larger in the non-judicial states,
although the difference in point estimates is not statistically significant.
We next test the hypothesis that lenders might benefit more from targeted delays in states
where existing consumer-level protections are stronger. We identify the presence of anti-
predatory consumer protection laws (APL) at the state level using the classification in Ding,
Quercia, Reid, and White (2012). The subsample analysis shown in columns (5)-(6) of the Table
OA-12 suggests somewhat weaker effects in states with active APL statutes, which is not quite
consistent with the notion that servicer actions are more pervasive in states where they are
already under tighter scrutiny. Part of the explanation for the apparent lack of cross-sectional
variation in foreclosure delays with APL laws may be that all of the servicers in our data have a
large national footprint and that the legislation under consideration was national in scope as well.
As a final cross-sectional check, we contrast Congressional districts with varying levels
33
of political competitiveness, as measured by the margin of victory. Using the results of the
November 2008 election, only 2% of loans in our sample come from districts in which the
margin of victory was less than 2 percentage points. Raising the win threshold to 5 percent only
increases our measure of loans in “competitive districts” to 7% of the original sample. The last
two columns of Table OA-12 show the results using the 10 percent victory margin. In this case,
14% of the loans are in less-than-10-pecent districts. However, there is not enough statistical
power to identify the political foreclosure delay in a relatively small sample of quasi-competitive
districts. We conclude that the extreme skewness of electoral victory margins makes it very
difficult to evaluate the hypothesis of greater political benefits for foreclosure delays in more
competitive districts.
In sum, the data limitations of our paper make it difficult to take a stand on the
mechanisms through which banks link their foreclosure actions and their political concerns. For
example, politicians may pressure the banks for delays as reported in the Introduction.
Alternatively, banks may agree on the delay voluntarily, as again reported in the Introduction.
Our test design and the available data do not allow us to distinguish among the various
mechanisms. More importantly, multiple mechanisms may, in fact, result in actions taken
simultaneously by different agents in an equilibrium. For example, Kroszner and Stratmann
(1998) provide a theory of congressional committees and interest groups in which interest groups
provide campaign contributions for favorable legislation. Crucially, they also show that this
equilibrium exists only if the interest groups and the politicians interact repeatedly, which is
facilitated by the committee structure in the Congress. The authors also provide empirical
evidence from the House Financial Services Committee (then called Banking Committee). In our
case, foreclosure delays have a role akin to campaign contributions. As a repeated game
equilibrium is likely to break down if the chance of reelection is low for the politicians, we can
expect to see both of the mechanisms mentioned above. In other words, potential mechanisms
34
may not be competing explanations of how the delay we demonstrate is originated but may, in
fact, be complementary and simultaneous aspects of a single equilibrium phenomenon.39
7 Conclusion
This paper is the first to document an effect of political motivations on foreclosure
decisions by banks in the aftermath of the financial crisis of 2008. Our results show that
mortgage servicing banks delayed the start of foreclosures for delinquent mortgages in the
districts of U.S. House Financial Services Committee members. There was no difference in the
onset of delinquencies based on the committee membership, however. This result is robust to
controlling for many loan- and zip code-level factors, and time-specific state-level fixed effects.
We do not find a similar effect in our placebo tests using the membership in Transportation and
Armed Services committees or in previous periods where foreclosures were not a major political
issue. Our results are also unlikely to be driven by concerns about sharing the stimulus funds
because the House Finance committee did not have jurisdiction on how those funds were
allocated.
Our calculations suggest that the cost of delay to financial institutions is an order of
magnitude larger than the campaign contributions to the Financial Services committee by the ten
largest loan servicers during our sample period. Our calculations further imply that the cost of
delay is at par with their lobbying expenses. These estimates are based on the in-sample figures
and are thus fairly conservative.
In addition to contributing to the finance literature on mortgages and foreclosures, our
work has implications for the political economy literature as well. Most of the political economy
39 Although several mechanisms may be expected to operate simultaneously in equilibrium, data also suggest that
some possible explanations are unlikely. For example, the delays in foreclosures are unlikely to be the result of
delinquent borrowers having more bargaining power in the committee district. First, the institutional features of
foreclosures discussed above make this improbable. Second, if this were the case, we would expect to see many
short sales, which are advantageous to delinquent borrowers, before the start of the foreclosure process. In our data
set of more than 360 thousand loans, there are only two loans that resulted in short sales before any foreclosure
action (both of those loans are excluded in our analysis) with the rest of (still infrequent) short sales taking place
after the foreclosure process is initiated.
35
literature on political influences focuses on campaign contributions and lobbying, which depend
on explicit direct expenditures and which are subject to disclosure requirements. The political
delays we document indicate instead an ‘in-kind’ political activity by the firms through their day-
to-day operations. Yet, these delays may be more effective than campaign contributions in some
respects because the delays precisely target a politician’s constituents. However, some campaign
activities like election-time television advertisements, one of the biggest expenditures that
campaign contributions finance, are typically broadcast over a geographic area that also includes
neighboring districts and, thus, are less precisely targeted. Lobbying expenditures, on the other
hand, are not directly captured by the politicians and cannot be legally directed to election
campaigns.
One question might be why the politicians prefer a delay in the foreclosures as indicated
by the quotes in the Introduction. After all, they may also have non-delinquent constituents in
their district. Although non-delinquent homeowners may have different incentives in general,
they may also prefer avoiding large number of foreclosures taking place in their neighborhood
due to the local negative externalities of foreclosures (Campbell, Giglio, and Pathak (2011)). In
addition, even if non-delinquent borrowers prefer rapid foreclosures, the benefit they obtain from
an expeditious process is likely to be diffuse across the non-delinquent owners. However, the
benefit of delay in foreclosures will be concentrated on delinquent borrowers and the classic
interest group arguments of Olson (1965) suggest that the politicians are likely to favor the
delinquent borrowers.
We only focus on the equilibrium outcome instead of equilibrium strategies due to the
lack of detailed data. However, the basic premise of the paper only requires that elected officials
be aware of foreclosure practices in their districts and that lenders or loan servicers be aware of
such interest. Both of these assumptions are quite plausible during that period of financial crisis.
Politicians are likely to learn about foreclosures in their districts directly from their constituents,
36
or from a number of non-profit organizations that focus on housing-related issues.40 For their
part, large financial institutions are likely to be aware of publicity surrounding foreclosure
activities.
Our paper does not focus and does not provide any evidence on what the banks received
or whether they received anything in return for delaying foreclosures. One potential future
research area might be to study the returns to banks from their political activities.
40 For example, Woodstock Institute publishes its quarterly foreclosure analysis at the neighborhood level for the
Chicago area, see, e.g., Woodstock Institute (2010).
37
References
Adelino, Manuel, and I. Serdar Dinc, 2014, Corporate distress and lobbying: Evidence from the
Stimulus Act, Journal of Financial Economics 114(2), 256-272.
Adelino, Manuel, Kristopher Gerardi, Paul S. Willen, 2013, Why Don’t Lenders Renegotiate
More Home Mortgages? Redefaults, Self-cures and Securitization, Journal of Monetary
Economics 60(7), 835-853.
Adelino, Manuel, Antoinette Schoar, and Felipe Severino, 2016, Loan Originations and Defaults
in the Mortgage Crisis: The Role of the Middle Class, Review of Financial Studies 29(7),
1635-1670.
Agarwal, Sumit, Gene Amromin, Itzhak Ben-David, Souphala Chomsisengphet, and Douglas
Evanoff, 2011, The Role of Securitization in Mortgage Renegotiation, Journal of Financial
Economics 102(3), 559-78.
Agarwal, Sumit, Gene Amromin, Itzhak Ben-David, Souphala Chomsisengphet, and Tomasz
Piskorski, and Amit Seru, 2017, Policy Intervention in Debt Renegotiation: Evidence from
the Home Affordable Modification Program, Journal of Political Economy 125(3), 654-712.
Agarwal, Sumit, Yan Chang, and Abdullah Yavas, 2012, Adverse Selection in Mortgage
Securitization, Journal of Financial Economics 105(3), 640-660.
Ansolabehere, Stephen D., John M. de Figueiredo, and James M. Snyder, 2003, Why is there so
little money in politics? Journal of Economic Perspectives 17, 105-130.
Austen-Smith, David, 1987, Interest Groups, Campaign Contributions and Probabilistic Voting,
Public Choice 54(2), 123-140.
Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan, 2004, How Much Should We Trust
Differences-In-Differences Estimates? Quarterly Journal of Economics, 119, 249-275.
Bertrand, Marianne, Francis Kramarz, Antoinette Schoar, and David Thesmar, 2007, Politicians,
Firms and the Political Business Cycle: Evidence from France, Working Paper, University of
Chicago.
38
Brown, Craig O., and Serdar Dinc, 2005, The Politics of Bank Failures: Evidence from
Emerging Markets, Quarterly Journal of Economics 120 (4), 1413-1444.
Campbell, John Y., Stefano Giglio, and Parag Pathak, 2011, Forced Sales and House Prices,
American Economic Review 101(5), 2108-2131.
Chinn, Menzie D., and Jeffrey Frieden, 2011, Lost decades: The Making of Americas Debt
Crisis and the Long Recovery. New Norton: York/London.
Cleves, Mario, William W. Gould, and Roberto G. Gutiérrez, 2004, Introduction to survival data
analysis with Stata, revised edition, Stata Press, College Station, Texas.
Cordell, Larry, Liang Geng, Laurie Goodman, and Lidan Yang, 2015, The Cost of Foreclosure
Delay, Real Estate Economics 43(4), 916-956.
Ding, Lei, Roberto G. Quercia, Carolina K. Reid, and Alan W. White, 2012, The Impact of
Federal Preemption of State Antipredatory Lending Laws on the Foreclosure Crisis, Journal
of Policy Analysis and Management 31(2), 367-387.
Duchin, Ran, and Denis Sosyura, 2010, TARP investments: Financials and Politics, Working
paper, Ross Business School, University of Michigan.
Evans, C. Lawrence, 2011, Congressional Committees, in F. Lee and E. Schickler, ed., The
Oxford Handbook of Congress, Oxford University Press.
Fine, Jason P., and Robert J. Gray, 1999, A Proportional Hazards Model for the Subdistribution
of a Competing Risk, Journal of the American Statistical Association 94, 496-509.
Ghent, Andra C. and Marianna Kudlyak. 2011. Recourse and Residential Mortgage Default:
Evidence from U.S. States. Review of Financial Studies 24(9), 149–86
Grossman, Gene, & Elhanan Helpman, 2001, Special Interest Politics. Cambridge, MA: MIT
Press.
Igan, Deniz, Prachi Mishra, and Thierry Tressel, 2009, A Fistful of Dollars: Lobbying and the
Financial Crisis, Working Paper, International Monetary Fund.
Jiang, Wei, Ashlyn Nelson, and Edward Vytlacil, 2014a, Liar's Loan? Effects of Origination
Channel and Information Falsification on Mortgage Delinquency, Review of Economic
Studies 96(1), 1-18.
39
Jiang, Wei, Ashlyn Nelson, and Edward Vytlacil, 2014b, Securitization and Loan Performance:
Ex Ante and Ex Post Relations in the Mortgage Market, Review of Financial Studies 27(2),
454-483.
Kalbfleisch, J., Prentice, R., 2002, The Statistical Analysis of Failure Time Data, 2nd Edition,
New York, John Wiley and Sons.
Keys, Benjamin, Tanmoy Mukherjee, Amit Seru, and Vikrant Vig, 2010, Did Securitization
Lead to Lax Screening? Evidence from Subprime Loans, Quarterly Journal of Economics
125, 307-362.
Keys, Benjamin, Tanmoy Mukherjee, Amit Seru, and Vikrant Vig, 2012, Lender Screening and
Role of Securitization: Evidence from Prime and Subprime Mortgage Markets, Review of
Financial Studies 25(8), 2071-2108.
Klein, John P., and Melvin L. Moeschberger, 2005, Survival analysis: techniques for censored
and truncated data, Springer Science & Business Media.
Kroszner, Randall S., and Thomas Stratmann, 1998, Interest-Group Competition and the
Organization of Congress: Theory and Evidence from Financial ServicesPolitical Action
Committees, American Economic Review 88(5), 1163-1187.
Lancaster, Tony, 1990, The Econometric Analysis of Transition Data, Cambridge, Cambridge
University Press.
Leech, Beth L., 2010, Lobbying and Influence, in The Oxford Handbook of American Political
Parties and Interest Groups, Jeffrey M. Berry and Sandy Maisel, eds. Oxford University
Press.
Mayer, Christopher, Karen Pence, and Shane M. Sherlund, 2009, The Rise in Mortgage Defaults,
Journal of Economic Perspectives 23(1), 27-50.
McFarland, Andrew, 2010, Interest Group Theory, in The Oxford Handbook of American
Political Parties and Interest Groups, L. Sandy Maisel and Jeffrey M. Berry, eds.; Oxford,
U.K., Oxford University Press, 37-56.
Melzer, Brian T., 2017, Mortgage Debt Overhang: Reduced Investment by Homeowners at Risk
of Default, Journal of Finance 72(2), 575-612.
40
Mian, Atif, and Amir Sufi, 2009, The Consequences of Mortgage Credit Expansion: Evidence
from the U.S. Mortgage Default Crisis, Quarterly Journal of Economics 124(4), 1449-1496
Mian, Atif, Amir Sufi, and Francesco Trebbi, 2010, The Political Economy of the U.S. Mortgage
Default Crisis, American Economic Review 100(5), 1967-1998.
Mian, Atif, Amir Sufi, and Francesco Trebbi, 2015, Foreclosures, House Prices, and the Real
Economy, Journal of Finance 70(6), 2587-2634.
Mortgage Loan Servicers, 2010, Top Residential Servicers 3/31/2010, vol. 14(7), p.1 (July
2010).
Nunez, Steven, and Howard Rosenthal, 2004, Bankruptcy Reform in Congress: Creditors,
Committees, Ideology, and Floor Voting in the Legislative Process, Journal of Law,
Economics and Organization 20, 527-557.
Olson, Mancur, 1965, The Logic of Collective Action: Public Goods and the Theory of Groups
Cambridge, MA: Harvard University Press.
Padovani, Miret, and Gibson, Rajna, 2011, The Determinants of BanksLobbying Activities,
Working Paper, Swiss Finance Institute.
Pence, Karen M., 2006, Foreclosing on Opportunity: State Laws and Mortgage Credit, Review of
Economics and Statistics 88(1), 177-182.
Piskorski, Tomasz, Amit Seru, and Vikrant Vig, 2010, Securitization and Distressed Loan
Renegotiation: Evidence from the Subprime Mortgage Crisis, Journal of Financial
Economics 97(3), 369-397.
Purnanandam, Amiyatosh, and Daniel Weagley, 2016, Can Markets Discipline Government
Agencies? Evidence from the Weather Derivatives Market, Journal of Finance 71(1), 303-
334.
Rajan, Raghuram, 2012, Fault lines, HarperCollins Publishers.
Romer, Thomas, and Barry R. Weingast, 1991, Political Foundations of the Thrift Debacle, in A.
Alesina and G. Carliner (eds.), Politics and Economics in the Eighties, Chicago: University
of Chicago Press.
41
Shepsle, Kenneth A., and Barry R. Weingast, 1987, The Institutional Foundations of Committee
Power, American Political Science Review 81, 85-127.
Stratmann, Thomas, 2002, Can Special Interests Buy Congressional Votes: Evidence from
Financial Services Legislation, Journal of Law and Economics 45(2), 345-373.
Stratmann, Thomas, 2005, Some Talk: Money in Politics. A (Partial) Review of the Literature,
Public Choice 124, 135-156.
Woodstock Institute, 2010, Chicago City and Regional Foreclosure Activity First Quarter
2010, http://www.woodstockinst.org/publications/download/first-quarter-2010-foreclosure-
filings-and-auctions/
Zhang, Yan, 2013, Does Loan Renegotiation Differ by Securitization Status? A Transition
Probability Study, Journal of Financial Intermediation, 513-527.
42
Table 1. Sample Statistics
The table gives the sample statistics for our variables. The sample is all the portfolio, GSE, or
privately securitized loans that become 90-day delinquent for the first time between January
2009 and December 2010; see text for the details of the sample constructions. Finance
Committee are the loans that are located in the districts of members of the U.S. House of
Representatives Financial Services Committee during the 111th Congress (2009-2010);
congressional districts with no zip-code level residential price index data are omitted. See
Appendix A for variable descriptions. The difference column gives the difference of means for
each variable based on committee membership. Time to the Start of Foreclosures ignores the fact
that some loans were still delinquent but not under foreclosure at the end of our sample period
(right censoring); hence, the mean time to foreclosure reported in this table is underestimated.
Equality of Survivor Functions is a chi-square test for the equality of distributions for the length
of time from 90-day delinquency to the start of foreclosures for committee and non-committee
districts. Standard errors in parentheses are robust to clustering at the congressional district level
except for Table 1E for which no option for cluster robust standard errors exists.
Panel A. Number of Loans and Districts
Finance committee member
Non committee member
Total
Democrat
Loans:
38,104
157,736
195,840
10.31%
42.68%
53.00%
Districts:
41
203
244
Republican
Loans:
25,257
148,444
173,701
6.83%
40.17%
47.00%
Districts:
29
142
171
Total
Loans:
63,361
306,180
369,541
17.15%
82.85%
100.00%
Balance:
$17.14B
$83.74B
$100.88B
17.00%
83.00%
Districts:
70
345
415
43
Panel B. Zip code-Level Demographic Variables
Full
sample
Finance
committee
members
Non
committee
members
Difference
Percent African American
Mean
11.36
12.85
11.05
1.80
SD (se)
(10.42)
(9.12)
(10.65)
(1.25)
Percent Hispanic
Mean
20.95
19.72
21.21
-1.48
SD (se)
(16.55)
(15.49)
(16.75)
(2.92)
Median household income,
thousands
Mean
55.31
59.26
54.49
4.77*
SD (se)
(13.43)
(15.58)
(12.79)
(2.64)
Percent Urban Population
Mean
89.17
91.72
88.65
3.08**
SD (se)
(13.65)
(11.78)
(13.95)
(1.36)
Unemployment Rate
Mean
10.48
10.12
10.55
-4.29
SD (se)
(2.85)
(2.33)
(2.94)
(4.04)
Decrease in home price index (%)
Mean
33.51
31.61
33.91
-2.29
SD (se)
(23.30)
(20.80)
(23.77)
(3.83)
Number of Loans
Mean
369,541
63,361
306,180
Panel C. Borrower-Level Variables
Full
sample
Finance
committee
members
Non
committee
members
Difference
FICO score 6 months before
delinquency
Mean
626.29
623.03
626.97
-3.94
SD (se)
(97.47)
(97.12)
(97.53)
(3.21)
Non first mortgage loan
amount 6 months before
delinquency
Mean
65302.05
63653.58
65643.19
-1989.60
SD (se)
(85920.07)
(84892.49)
(86127.38)
(3049.19)
44
Panel D. Loan-level Variables
Full
sample
Finance
committee
members
Non
committee
members
Difference
FICO score at origination
Mean
687.1
684.3
687.7
-3.388
SD (se)
(55.7)
(56.1)
(55.6)
(2.400)
Loan-to-Value (LTV) ratio at
origination (%)
Mean
77.927
78.158
77.879
0.28
SD (se)
(11.413)
(11.648)
(11.362)
(.669)
Estimated Contemporaneous
(LTV) ratio (%)
Mean
109.2
106.7
109.8
-3.06
SD (se)
(43.7)
(26.8)
(46.4)
(3.821)
Loan amount, thousands
Mean
272.9
270.5
273.5
-3.006
SD (se)
(147.2)
(152.2)
(146.1)
(17.524)
First interest rate reported (%)
Mean
6.481
6.524
6.473
0.051
SD (se)
(1.251)
(1.243)
(1.252)
(0.043)
Fixed rate flag
Mean
0.688
0.71
0.683
0.027
SD (se)
(0.463)
(0.454)
(0.465)
(0.022)
Interest only flag
Mean
0.236
0.207
0.242
-0.035
SD (se)
(0.424)
(0.405)
(0.428)
(0.021)
Jumbo flag
Mean
0.144
0.148
0.143
0.005
SD (se)
(0.351)
(0.355)
(0.350)
(0.036)
Subprime flag
Mean
0.061
0.065
0.06
0.005
SD (se)
(0.239)
(0.246)
(0.237)
(0.004)
Refi flag
Mean
0.615
0.614
0.614
0.002
SD (se)
(0.487)
(0.487)
(0.487)
(0.015)
Full documentation flag
Mean
0.379
0.372
0.381
-0.009
SD (se)
(0.485)
(0.483)
(0.486)
(0.011)
Documentation unknown flag
Mean
0.372
0.375
0.372
0.003
SD (se)
(0.483)
(0.484)
(0.483)
(0.005)
Debt-to-Income ratio
Mean
28.017
27.818
28.058
-0.240
SD (se)
(19.867)
(20.020)
(19.840)
(0.483)
Missing DTI flag
Mean
0.262
0.266
0.261
0.005
SD (se)
(0.440)
(0.442)
(0.439)
(0.009)
Elapsed Time at First
Delinquency
Mean
36.420
36.497
36.405
0.092
SD (se)
(12.443)
(12.468)
(12.438)
(0.309)
Public Securitized Flag
Mean
0.599
0.603
0.598
0.005
SD (se)
(0.490)
(0.448)
(0.490)
(0.024)
Private Securitized Flag
Mean
0.284
0.277
0.285
-0.008
45
SD (se)
(0.451)
(0.489)
(0.452)
(0.020)
Portfolio Flag
Mean
0.117
0.119
0.117
0.003
SD (se)
(0.321)
(0.324)
(0.321)
(0.007)
Panel E. Time to the Start of Foreclosures (right censoring ignored)
Whole sample
Non-committee
Committee
Equality of
Survival
Functions
(Chi-sq)
Mean
10.718
10.690
10.853
16.51
Std. Err./p-value
(0.018)
(0.020)
(0.044)
[<0.0001]
95% Confidence Interval
(10.683-
10.754)
(10.651-
10.729)
(10.767-
10.939)
--
Number of borrowers
369,541
306,180
63,361
--
Number of borrower-months
2,063,579
1,707,690
355,889
--
46
Table 2. U.S. House of Representatives Finance Committee Membership and the
Foreclosure of Delinquent Mortgages
The table presents exponential hazard analysis for the start of foreclosure process for 90-day
delinquent loans. The sample period covers the 111th Congress (January 2009 through December
2010); the sample includes all the mortgages that become 90-day delinquent for the first time
during that period. Finance Committee Member is a binary variable that is one if the loan is for a
house located in a district whose U.S. House representative is a member of the Financial
Services Committee. All the regressions include state*month fixed effects except for the fourth
regression, which includes state and month fixed effects without their interaction. Standard errors
are robust to clustering at the congressional district level and are in parentheses. ***, **, and *
denote statistical significance at the 1%, 5%, and 10% levels, respectively.
(1)
(2)
(3)
(4)
Finance committee member
-0.040***
-0.038***
-0.039***
(0.015)
(0.013)
(0.013)
Majority Party
-0.013
-0.022*
-0.009
-0.010
(0.012)
(0.012)
(0.011)
(0.011)
Zip code level variables
Unemployment rate, by zip
0.010***
0.010***
0.001
(0.003)
(0.003)
(0.002)
Urban population (%), by zip
-0.068**
-0.068**
-0.103***
(0.033)
(0.032)
(0.032)
Log(median household income),
by zip
-0.054* -0.040 -0.042
(0.032)
(0.031)
(0.032)
% black/African American, by zip
-0.001
-0.001
-0.000
(0.001)
(0.001)
(0.001)
% Hispanic or Latino, by zip
-0.000
-0.000
0.000
(0.000)
(0.000)
(0.000)
Decrease in home price index
since origination -0.125*** -0.122*** -0.062***
(0.046)
(0.045)
(0.044)
Borrower-level variables
47
FICO 620-680 6 months pre
delinquency 0.058*** 0.058*** 0.059***
(0.007)
(0.007)
(0.007)
FICO>=680 6 months pre
delinquency
0.080*** 0.080*** 0.082***
(0.007)
(0.007)
(0.007)
Log(non first mortgage loan
amounts 6 months pre
delinquency)
0.006*** 0.006*** 0.007***
(0.002)
(0.002)
(0.002)
Loan-level controls
LTV at origination = 80% flag
0.002
0.001
0.004
(0.006)
(0.006)
(0.006)
LTV Ratio
0.025***
0.025***
0.024***
(0.001)
(0.001)
(0.001)
LTV Ratio Squared
-0.000***
-0.000***
-0.000***
(0.000)
(0.000)
(0.000)
LTV Ratio Cubed
0.000***
0.000***
0.000***
(0.000)
(0.000)
(0.000)
Missing debt to income flag
0.237***
0.237***
0.240***
(0.018)
(0.018)
(0.018)
Debt to income ratio
-0.000
-0.000
-0.000
(0.000)
(0.000)
(0.000)
Full documentation flag
-0.102***
-0.102***
-0.102***
(0.010)
(0.010)
(0.010)
Unknown documentation flag
-0.258***
-0.258***
-0.256***
(0.015)
(0.015)
(0.015)
FICO 620-680 at origination
0.193***
0.193***
0.198***
(0.010)
(0.010)
(0.010)
FICO>= 680 at origination
0.352***
0.352***
0.356***
48
(0.012)
(0.012)
(0.012)
Log(original loan amount)
-0.131***
-0.132***
-0.139***
(0.012)
(0.012)
(0.012)
Elapsed term at first delinquency
0.007***
0.007***
0.007***
(0.000)
(0.000)
(0.000)
First interest rate reported
-0.062***
-0.062***
-0.063***
(0.003)
(0.003)
(0.004)
Fixed rate flag
-0.156***
-0.156***
-0.153***
(0.010)
(0.010)
(0.010)
Interest only flag
0.127***
0.127***
0.129***
(0.008)
(0.008)
(0.008)
Jumbo flag
0.030**
0.031**
0.034***
(0.014)
(0.014)
(0.013)
Low grade flag
0.366***
0.366***
0.365***
(0.018)
(0.018)
(0.018)
Refi flag
-0.040***
-0.040***
-0.040***
(0.006)
(0.006)
(0.006)
Public securitized flag
0.296***
0.296***
0.295***
(0.012)
(0.012)
(0.012)
Private securitized flag
0.309***
0.309***
0.306***
(0.010)
(0.010)
(0.010)
Fixed effects
State*month
State*month
State*month
State, month
Clustering
Congressional
District
Congressional
District
Congressional
District
Congressional
District
# of Loan-months
2,063,579
2,063,579
2,063,579
2,063,579
# of Loans
369,541
369,541
369,541
369,541
49
Table 3. Finance Committee Membership and the Time to Foreclosures
The table presents the results of Maximum Likelihood Estimation where the dependent variable
is the logarithm of the time length from the onset of 90-day delinquency to the start of
foreclosure. Each regression is estimated under a different assumption for the error term
distribution and takes right censoring into account. The sample period covers the 111th Congress
(January 2009 through December 2010); the sample includes all the mortgages that become 90-
day delinquent for the first time during that period. The estimation uses single-observation
survival-time data, where each regressor is set to its value in the month of 90-day delinquency.
Finance Committee is a binary variable that is one if the loan is for a house located in a district
whose U.S. House representative is a member of the Financial Services Committee. Predicted
Mean (Median) is the average (median) time to the foreclosure starts predicted by the model.
Finance Committee Marginal Effect for Mean (Median) is the marginal effect of being in a
finance committee district on the predicted mean (median) for the time to the start of
foreclosures. Standard errors are robust to clustering at the congressional district level and are in
parentheses. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels,
respectively.
(1)
(2)
(3)
(4)
Error Term Distribution Normal Logistic Extreme Value
Extreme Value
with a scale
parameter
Finance committee
0.037***
0.043***
0.043***
0.044***
(0.014)
(0.015)
(0.014)
(0.014)
Majority Party
0.010
0.011
0.013
0.013
(0.010)
(0.011)
(0.010)
(0.011)
Borrower-level controls
Yes
Yes
Yes
Yes
Loan level controls
Yes
Yes
Yes
Yes
Zip level controls
Yes
Yes
Yes
Yes
Fixed Effects State, Month State, Month State, Month State, Month
Predicted mean (months)
18.714***
(0.606)
33.762***
(1.424)
12.122***
(0.247)
12.620***
(0.259)
Finance Committee Marginal
Effect on Mean (months)
0.698***
(0.256)
1.482***
(0.518)
0.522***
(0.168)
0.559***
(0.167)
Predicted Median (months)
7.687***
(0.186)
7.744***
(0.187)
8.402***
(0.171)
8.464***
(0.174)
Finance Committee Marginal
Effect on Median (months)
0.287***
(0.105)
0.340***
(0.119)
0.362***
(0.117)
0.375***
(0.112)
Clustering level
Congressional
district
Congressional
district
Congressional
district
Congressional
district
# of Loans
369,540
369,540
369,540
369,540
50
Table 4. Finance Committee Membership and Foreclosures: Placebo Tests
The table presents exponential hazard analysis for the start of foreclosure process for the loans
that become 90-day delinquent for the first time during the analysis period. Finance / Defense /
Transportation Committee Member are binary variables that are one if the loan is for a house
located in a district whose U.S. House representative is a member of the House Finance /
Transportation/ Defense Committee, respectively. All the regressions include state*month fixed
effects except for the fourth regression, which includes state and month fixed effects without
their interaction. Standard errors are robust to clustering at the congressional district level and
are in parentheses. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels,
respectively.
(1)
(2)
(3)
(4)
Defense committee
member, 2009-2010
-0.009
(0.015)
Transportation
committee member, 2009
- 2010
0.010
(0.013)
Finance committee
member, 2007-2008
-0.023
(0.017)
Finance committee
member, 2005-2006
0.037
(0.037)
Majority Party
-0.014
-0.013
0.021
-0.019
(0.012)
(0.012)
(0.013)
(0.034)
Borrower-level controls
Yes
Yes
Yes
Yes
Loan level controls
Yes
Yes
Yes
Yes
Zip level controls
Yes
Yes
Yes
Yes
State*month FE
Yes
Yes
Yes
No
Sample
90-day
delinquent in
(Jan-2009,
Dec-2010)
90-day
delinquent in
(Jan-2009,
Dec-2010)
90-day
delinquent in
(Jan-2007,
Dec-2008)
90-day
delinquent in
(Jan-2005,
Dec-2006)
Clustering Level
Congressional
district
Congressional
district
Congressional
district
Congressional
district
# of Loan-months
2,063,579
2,063,579
499,183
25,089
# of Loans
369,541
369,541
180,004
11,939
51
Table 5. Finance Committee Membership and Mortgage Delinquencies
The table presents exponential hazard analysis for the delinquency of loans that are current at the
start of the sample period. The delinquency is defined as 30-day delinquency in regression (1)
and 90-day delinquency in regression (2). The sample period covers the 111th Congress (January
2009 through December 2010). Finance Committee Member is a binary variable that is one if the
loan is for a house located in a district whose U.S. House representative is a member of the
House Financial Services Committee in the 111th Congress. All the regressions include
state*month fixed effects. Standard errors are robust to clustering at the congressional district
level and are in parentheses. ***, **, and * denote statistical significance at the 1%, 5%, and
10% levels, respectively.
(1)
(2)
30-day delinquency
90-day delinquency
Finance committee member
0.003
0.009
(0.024)
(0.032)
Majority Party
0.022
0.023
(0.019)
(0.025)
Borrower-level controls
Yes
Yes
Loan level controls
Yes
Yes
Zip level controls
Yes
Yes
State*month FE
Yes
Yes
Sample
Current in Dec-2008
Current in Dec-2008
Clustering level
Congressional district
Congressional district
# of Loan-months
39,597,857
42,231,510
# of Loans
2,319,364
2,322,181
52
Table 6. Finance Committee Membership and Foreclosures:
Senior Committee Members Only
The table presents exponential hazard analysis for the start of foreclosure process for 90-day
delinquent loans. The sample period covers the 111th Congress (January 2009 through December
2010); the sample includes all the mortgages that become 90-day delinquent for the first time
during that period. Finance Committee Member is a binary variable that is one if the loan is for a
house located in a district whose U.S. House representative is a member of the Financial
Services Committee. The first regression excludes committee members that were appointed to
the committee in 2005 or after. The second regression excludes the politicians who took office in
the U.S. House of Representatives for the first time in 2005 or after. All the regressions include
state*month fixed effects. Standard errors are robust to clustering at the congressional district
level and are in parentheses. ***, **, and * denote statistical significance at the 1%, 5%, and
10% levels, respectively.
(1)
(2)
Finance committee member
-0.044***
-0.054***
(0.016)
(0.016)
Majority Party
-0.013
-0.035**
(0.012)
(0.014)
Borrower-level controls
Yes
Yes
Loan level controls
Yes
Yes
Zip level controls
Yes
Yes
State*month FE
Yes
Yes
Sample
Exclude first-time
committee members
as of 2005 or after
Exclude first-time
House members as
of 2005 or after
Clustering level
Congressional
district
Congressional
district
# of Loan-months
1,878,804
1,373,386
# of Loans
334,298
241,368
53
Table 7. Finance Committee Membership and Foreclosures: Alternative Econometric Models
Columns (1) and (2) present exponential hazard analysis for the start of foreclosure process for 90-day delinquent loans using sample
periods from January 2009 to December 2009 and to March 2010, respectively. The third and fourth columns provide hazard analysis
for the start of foreclosure process for 90-day delinquent loans using the Cox proportional hazard and Weibull models. The fifth
column presents the discrete hazard analysis with the linear probability model for the start of foreclosures of 90-day delinquent loans;
the dependent variable is one if the loan is foreclosed that month conditional on not being in foreclosure before. Columns (6), (7), (8)
represent the committee and non-committee zip codes that are within 10, 25, and 50 miles of one another. The sample period covers
the 111th Congress (January 2009 through December 2010), except for columns (1) and (2). Finance Committee Member is a binary
variable that is one if the loan is for a house located in a district whose U.S. House representative is a member of the House Financial
Services Committee in the 111th Congress. Standard errors are robust to clustering at the congressional district level and are in
parentheses. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.
54
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Exponential
Hazard Exponential
Hazard
Cox
Proportional
Hazard
Weibull
Hazard
Discrete
Hazard,
Linear
Probability
Model
Exponential
Hazard Exponential
Hazard Exponential
Hazard
Finance committee
member
-0.035**
-0.042**
-0.037***
-0.038***
-0.003***
-0.046**
-0.033**
-0.032**
(0.014)
(0.013)
(0.012)
(0.013)
(0.001)
(0.019)
(0.016)
(0.016)
Majority Party
-0.005
-0.004
-0.009
-0.009
-0.001
-0.020
0.004
-0.011
(0.012)
(0.012)
(0.010)
(0.011)
(0.001)
(0.025)
(0.020)
(0.018)
Borrower-level
controls
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Loan level controls
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Zip level controls
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
State*month FE
State*month
State*month
State*month
State*month
State*month
State*month
State*month
State*month
Sample Jan. 2009 to
Dec. 2009 Jan. 2009 to
Mar. 2010 Full Full Full
Committee
and non-
committee zip
codes within
10 miles
Committee
and non-
committee zip
codes within
25 miles
Committee
and non-
committee zip
codes within
50 miles
Clustering Level
Congressional
district
Congressional
district
Congressional
district
Congressional
district
Congressional
district
Congressional
district
Congressional
district
Congressional
district
# of Loan-months
745,441
1,118,307
2,063,579
2,063,579
2,063,579
462,298
799,549
968,799
# of Loans
213,795
268,184
369,541
369,541
369,541
79,118
137,011
166,214
55
Figure 1. Districts of the Members of the Financial Services Committee
in U.S. House of Representatives during 111th Congress (2009-2010)
56
Appendix A. Variable Description and Source
Variable
Description
Source
Political Variables
Majority Party
Binary variable equal to one if a member of the majority party
represents the district in which the mortgaged home is located during the
111th Congress
Stewart and Woon
Finance committee
member
Binary variable equal to one if a member of the Financial Services
Committee represents the district in which the mortgaged home is
located during the 111th Congress
Stewart and Woon
Demographic Variables
Unemployment rate
ZCTA-level monthly unemployment rate
BLS
% Urban population
% of population living in urban areas based on 2000 census. Original
data is at the county level and aggregated to the zip code level using
Mable/Geocorr county-to-zip crosswalk population-based allocation
factors.
Census 2000 SF 1
Log median household
income
Natural log of annual median household income on annual basis at the
congressional district level
ACS
% black/African
American
% of population reported black or African American on an annual basis.
Original data is at the county level and aggregated to the zip code level
using Mable/Geocorr county-to-zip crosswalk population-based
allocation factors.
ACS
% Hispanic or latino
% of population reported Hispanic or Latino. Original data is at the
county level and aggregated to the zip code level using Mable/Geocorr
county-to-zip crosswalk population-based allocation factors.
ACS
Decrease in home price
index since origination
Negative change in log zip code-level home price index between month
of origination and current month
Corelogic Home Price
Index
57
Loan-Level Variables
FICO 620-680 at
origination
Binary variable equal to one if borrower FICO score at origination
greater than or equal to 620 and less than 680
Mortgage servicer
dataset
FICO >= 680 at
origination
Binary variable equal to one if borrower FICO score at origination
greater than or equal to 680
Mortgage servicer
dataset
LTV at origination =
80% flag
Binary variable equal to one if the loan-to-value ratio at origination is
80%
Mortgage servicer
dataset
Missing dti flag
Binary variable equal to one if the debt-to-income at origination is not
reported
Mortgage servicer
dataset
DTI ratio
Debt-to-income ratio at origination
Mortgage servicer
dataset
Estimated current LTV
Ratio of the principal remaining to the estimated current home value,
derived from the appraisal value at origination and the ratio of the
current home price index to the home price index at origination
Mortgage servicer
dataset
Log loan amount
Natural log of the original loan amount in dollars
Mortgage servicer
dataset
First interest rate
reported
Earliest current interest rate reported in LPS dynamic data
Mortgage servicer
dataset
Fixed rate flag
Binary variable equal to one if principal and interest are constant at loan
origination
Mortgage servicer
dataset
Interest only flag
Binary variable equal to one if loan is interest only at origination
Mortgage servicer
dataset
Jumbo flag
Binary variable equal to one if loan is jumbo at origination
Mortgage servicer
dataset
Low grade flag
Binary variable equal to one if loan is grade "B" or "C" at origination
Mortgage servicer
dataset
Refi flag
Binary variable equal to one if loan purpose is refinance at origination
Mortgage servicer
dataset
58
Full documentation flag
Binary variable equal to one if full documentation was presented at loan
origination
Mortgage servicer
dataset
Unknown
documentation flag
Binary variable equal to one if documentation type at loan origination
classified as unknown or other
Mortgage servicer
dataset
Elapsed term at first
delinquency
Months between loan origination and first 90-day delinquency
Mortgage servicer
dataset
Public securitized flag
Binary variable equal to one if loan is securitized by Fannie Mae or
Freddie Mac in the month of first 90-day delinquency
Mortgage servicer
dataset
Private securitized flag
Binary variable equal to one if loan is securitized by a private investor in
the month of first 90-day delinquency
Mortgage servicer
dataset
59
Appendix B. Key Dates in the Passage of the Financial Reform Act
June 17, 2009: Obama Administration releases its financial reform proposal
July – December 2009: Discussions and hearings in the Financial Services Committee on
the proposal.
December 11, 2009: U.S. House of Representatives passes the final version.
January-May 2010: Senate considers the financial reform proposal.
May 20, 2010: Senate passes its own version and the bill moves to the conference
committee.
June 29, 2010: The unified bill leaves the conference committee.
June 30, 2010: House passes the bill.
July 15, 2010: Senate passes the bill.
July 21, 2010: President signs the bill and it becomes law.
ONLINE APPENDIX
Online Appendix Page 1
Table OA1. Finance Committee Membership and Foreclosures: Robustness in Subsamples
The table presents exponential hazard analysis for the start of foreclosure proceedings for 90-day
delinquent loans. The sample period covers the 111th Congress (January 2009 through December
2010); the sample includes all the mortgages that become 90-day delinquent for the first time
during that period. Finance Committee Member is a binary variable that is one if the loan is for a
house located in a district whose U.S. House representative is a member of the Financial
Services Committee. All the regressions include state*month fixed effects. Standard errors are
robust to clustering at the congressional district level and are in parentheses. ***, **, and *
denote statistical significance at the 1%, 5%, and 10% levels, respectively.
(1)
(2)
(3)
Finance committee
member
-0.037***
-0.037***
-0.036***
(0.013)
(0.013)
(0.013)
Majority Party
-0.008
-0.010
-0.009
(0.011)
(0.011)
(0.012)
Borrower-level controls
Yes
Yes
Yes
Loan level controls
Yes
Yes
Yes
Zip level controls
Yes
Yes
Yes
State*month FE
Yes
Yes
Yes
Sample
Exclude subprime
Loan term 15,20,30
years
Excludes Oct-Dec
2010
Clustering level
Congressional District
Congressional District
Congressional District
# of Loan-months
1,964,577
2,045,095
1,798,311
# of Loans
347,129
366,707
338,579
Online Appendix Page 2
Table OA2 (Counterpart of Table 2 with Time to Foreclosure). U.S. House of
Representatives Finance Committee Membership and the Foreclosure of Delinquent
Mortgages
The table presents the results of Maximum Likelihood Estimation where the dependent variable
is the logarithm of the time length from the onset of 90-day delinquency to the start of
foreclosure. The assumed distribution for the error term is extreme value distribution, which
generates exponential hazard. The sample period covers the 111th Congress (January 2009
through December 2010); the sample includes all the mortgages that become 90-day delinquent
for the first time during that period. The table uses single-observation survival-time data, where
each regressor is set to its value in the month of 90-day delinquency. Finance Committee
Member is a binary variable that is one if the loan is for a house located in a district whose U.S.
House representative is a member of the Financial Services Committee. All the regressions
include state*delinquency month fixed effects except for the fourth regression, which includes
state and delinquency month fixed effects. Standard errors are robust to clustering at the
congressional district level and are in parentheses. ***, **, and * denote statistical significance
at the 1%, 5%, and 10% levels, respectively.
Online Appendix Page 3
(1)
(2)
(3)
(4)
Finance committee member
0.044***
0.044***
0.043***
(0.015)
(0.014)
(0.014)
Majority Party
0.016
0.020
0.012
0.013
(0.011)
(0.013)
(0.011)
(0.010)
Borrower-level controls
Yes
No
Yes
Yes
Loan level controls
Yes
No
Yes
Yes
Zip level controls
Yes
No
Yes
Yes
Fixed Effects
State*month
State*month
State*month
State, month
Sample
90-day
delinquent in
(Jan-2009, Dec-
2010)
90-day
delinquent in
(Jan-2009,
Dec-2010)
90-day
delinquent in
(Jan-2009,
Dec-2010)
90-day
delinquent in
(Jan-2009,
Dec-2010)
Clustering level
Congressional
District
Congressional
District
Congressional
District
Congressional
District
# of Loan-months
369,541
369,541
369,541
369,541
# of Loans
369,541
369,541
369,541
369,541
Online Appendix Page 4
Table OA3 (Counterpart of Table 4 with Time to Foreclosure). Finance Committee
Membership and Foreclosures: Placebo Tests
The table presents the results of Maximum Likelihood Estimation where the dependent variable
is the logarithm of the time length from the onset of 90-day delinquency to the start of
foreclosure. The assumed distribution for the error term is extreme value distribution, which
generates exponential hazard. The table uses single-observation survival-time data, where each
regressor is set to its value in the month of 90-day delinquency. Finance / Defense /
Transportation Committee Member are binary variables that are one if the loan is located in a
district of a member of the House Finance / Transportation/ Defense Committee, respectively.
All the regressions include state*delinquency month fixed effects. Standard errors are robust to
clustering at the congressional district level and are in parentheses. ***, **, and * denote
statistical significance at the 1%, 5%, and 10% levels, respectively.
Online Appendix Page 5
(1)
(2)
(3)
(4)
Defense committee
member, 2009-2010
0.010
(0.015)
Transportation
committee member, 2009
- 2010
-0.016
(0.013)
Finance committee
member, 2007-2008
0.027
(0.019)
Finance committee
member, 2005-2006
-0.020
(0.032)
Majority Party
0.018
0.017
-0.019
0.006
(0.011)
(0.011)
(0.014)
(0.032)
Borrower-level controls
Yes
Yes
Yes
Yes
Loan level controls
Yes
Yes
Yes
Yes
Zip level controls
Yes
Yes
Yes
Yes
State*month FE
Yes
Yes
Yes
Yes
Sample
90-day
delinquent in
(Jan-2009,
Dec-2010)
90-day
delinquent in
(Jan-2009,
Dec-2010)
90-day
delinquent in
(Jan-2007,
Dec-2008)
90-day
delinquent in
(Jan-2005,
Dec-2006)
Clustering Level
Congressional
district
Congressional
district
Congressional
district
Congressional
district
# of Loan-months
369,541
369,541
180,004
11,929
# of Loans
369,541
369,541
180,004
11,929
Online Appendix Page 6
Table OA4 (Counterpart of Table 5 with Time to Delinquency). Finance Committee
Membership and Mortgage Delinquencies
The table presents the results of Maximum Likelihood Estimation where the dependent variable
is the logarithm of the time length from December 2008 to the onset of delinquency, for all loans
that are current at the start of the sample period. The assumed distribution for the error term is
extreme value distribution, which generates exponential hazard. The delinquency is defined as
30-day delinquency in regression (1) and 90-day delinquency in regression (2). The sample
period covers the 111th Congress (January 2009 through December 2010). The table uses single-
observation survival-time data, where each regressor is set to its value in December 2008.
Finance Committee Member is a binary variable that is one if the loan is for a house located in a
district whose U.S. House representative is a member of the House Financial Services
Committee in the 111th Congress. All the regressions include state*origination month fixed
effects. Standard errors are robust to clustering at the congressional district level and are in
parentheses. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels,
respectively.
(1)
(2)
30-day delinquency
90-day delinquency
Finance committee member
-0.010
-0.066
(0.026)
(0.0235)
Majority Party
-0.025
-0.136
(0.021)
(0.180)
Borrower-level controls
Yes
Yes
Loan level controls
Yes
Yes
Zip level controls
Yes
Yes
State*month FE
Yes
Yes
Sample
Current in Dec-2008
Current in Dec-2008
Clustering level
Congressional district
Congressional district
# of Loan-months
2,108,212
2,108,212
# of Loans
2,108,212
2,108,212
Online Appendix Page 7
Table OA5 (Counterpart of Table 6 with Time to Foreclosure). Finance Committee
Membership and Foreclosures: Senior Committee Members Only
The table presents the results of Maximum Likelihood Estimation where the dependent variable
is the logarithm of the time length from the onset of 90-day delinquency to the start of
foreclosure. The assumed distribution for the error term is extreme value distribution, which
generates exponential hazard. The sample period covers the 111th Congress (January 2009
through December 2010); the sample includes all the mortgages that become 90-day delinquent
for the first time during that period. The table uses single-observation survival-time data, where
each regressor is set to its value in the month of 90-day delinquency. Finance Committee
Member is a binary variable that is one if the loan is for a house located in a district whose U.S.
House representative is a member of the Financial Services Committee. The first regression
excludes committee members that were appointed to the committee in 2005 or after. The second
regression excludes the politicians who took office in the U.S. House of Representatives for the
first time in 2005 or after. All the regressions include state and delinquency month fixed effects.
Standard errors are robust to clustering at the congressional district level and are in parentheses.
***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.
(1)
(2)
Finance committee member
0.042**
0.055***
(0.017)
(0.016)
Majority Party
0.018
0.041***
(0.011)
(0.013)
Borrower-level controls
Yes
Yes
Loan level controls
Yes
Yes
Zip level controls
Yes
Yes
State*month FE
No
No
Sample
Exclude first-time
committee members
as of 2005 or after
Exclude first-time
House members as
of 2005 or after
Clustering level
Congressional
district
Congressional
district
# of Loan-months
334,298
241,368
# of Loans
334,298
241,368
Online Appendix Page 8
Table OA6 (Counterpart of Table 7 with Time to Foreclosure). Finance Committee
Membership and Foreclosures: Alternative Econometric Models
Columns (1) and (2) present the results of Maximum Likelihood Estimation where the dependent
variable is the logarithm of the time length from the onset of 90-day delinquency to the start of
foreclosure, using sample periods from January 2009 to December 2009 and to March 2010,
respectively. Columns (5), (6), (7) represent the committee and non-committee zip codes that are
within 10, 25, and 50 miles of one another. The sample period covers the 111th Congress
(January 2009 through December 2010), except for columns (1) and (2). The table uses single-
observation survival-time data, where each regressor is set to its value in the month of 90-day
delinquency. Finance Committee Member is a binary variable that is one if the loan is for a house
located in a district whose U.S. House representative is affiliated with the majority party in the
U.S. House. All the regressions include state*delinquency month fixed effects. Standard errors
are robust to clustering at the congressional district level and are in parentheses. ***, **, and *
denote statistical significance at the 1%, 5%, and 10% levels, respectively.
Online Appendix Page 9
(1)
(2)
(3)
(4)
(5)
(6)
Error Term Distribution Extreme Value Extreme Value
Extreme
Value with
Scale
Parameter
Extreme
Value Extreme
Value Extreme
Value
Implied Hazard
Exponential
Exponential
Weibull
Exponential
Exponential
Exponential
Finance committee
member
0.037*** 0.046*** 0.045*** 0.058*** 0.041** 0.037**
(0.014)
(0.014)
(0.014)
(0.018)
(0.016)
(0.016)
Majority Party
0.008
0.009
0.012
0.006
-0.008
0.013
(0.012)
(0.012)
(0.011)
(0.025)
(0.019)
(0.017)
Borrower-level controls
Yes
Yes
Yes
Yes
Yes
Yes
Loan level controls
Yes
Yes
Yes
Yes
Yes
Yes
Zip level controls
Yes
Yes
Yes
Yes
Yes
Yes
State*month FE
State*month
State*month
State*month
State*month
State*month
State*month
Sample Jan. 2009 to
Dec. 2009 Jan. 2009 to
Mar. 2010 Full
Committee
and non-
committee zip
codes within
10 miles
Committee
and non-
committee zip
codes within
25 miles
Committee
and non-
committee zip
codes within
50 miles
Clustering Level
Congressional
district
Congressional
district
Congressional
district
Congressional
district
Congressional
district
Congressional
district
# of Loan-months
213,795
268,184
369,541
79,118
137,011
166,214
# of Loans
213,795
268,184
369,541
79,118
137,011
166,214
Online Appendix Page 10
Table OA7. Selected Statistics for Non-mortgage loans
The table provides the sample statistics and incidence rates of non-mortgage borrowings of the
borrowers whose mortgage became 90-day delinquent between January 2009 and December
2010. Panel A. provides the share of total non-mortgage borrowing that a given subcomponent of
borrowing represents as of the delinquency date. Panel B. and C. provide the incidence rates of
‘failure’ for the exponential hazard analysis of non-mortgage borrowings. ‘Failure’ is defined as
the delinquency of the nonmortgage borrowing in Panel B. and as the appearance of a new non-
mortgage loan in Panel C. The data on household non-mortgage liabilities are from the Equifax
CRISM dataset.
Panel A. Share of Non-Mortgage Borrowing by Type of Loan
Full
Sample
Finance
committee
members
Non-
committee
members
Difference
Auto loans share of
non-mortgage balance
Mean
28.57%
28.05%
28.68%
-0.63%
SD (se)
(32.81)
(32.63)
(32.85)
(0.65)
n
358,827
61,514
297,313
Bank card share of
non-mortgage balance
Mean
45.96%
45.64%
46.02%
-0.38%
SD (se)
(35.65)
(35.58)
(35.66)
(0.89)
n
358,827
61,514
297,313
All others (incl.
student loans, retail
loans, consumer
finance loans)
Mean
25.47%
26.31%
25.30%
1.01%
SD (se)
(30.58)
(30.90)
(30.51)
(0.62)
n
358,827
61,514
297,313
Ratio of non-mortgage
debt to total mortgage
balance outstanding
Mean
16.48%
16.73%
16.42%
0.31%
SD (se)
(32.01)
(21.90)
(33.73)
(0.92)
n
369,541
63,361
306,180
Online Appendix Page 11
Panel B. Incidence Rates for Delinquency of a Non-Mortgage Loan
Auto loans incidence rate
Whole
sample
Non-committee
Committee
=Committee -
nonCommittee
Incidence Rate
2.54%
2.52%
2.64%
0.12%
Incidence rate Std.Err. (%)
(0.05)
(0.06)
(0.10)
(0.11)
Number of 'failures'
32,392
26,698
5,694
Number of borrowers
177,858
147,614
30,244
Number of borrower-
months
1,274,449
1,058,721
215,728
Bankcard loans incidence rate
Whole
sample
Non-committee
Committee
=Committee -
nonCommittee
Incidence Rate
3.85%
3.86%
3.85%
-0.01%
Incidence rate Std.Err. (%)
(0.05)
(0.05)
(0.11)
(0.12)
Number of 'failures'
57,217
47,514
9,703
Number of borrowers
218,454
181,571
36,883
Number of borrower-
months
1,484,436
1,232,292
252,144
All other loans (student, retail, consumer finance, other)
Whole
sample
Non-committee
Committee
=Committee -
nonCommittee
Incidence Rate
3.45%
3.45%
3.47%
0.02%
Incidence rate Std.Err. (%)
(0.04)
(0.05)
(0.10)
(0.11)
Number of 'failures'
53,438
44,137
9,301
Number of borrowers
234,914
194,482
40,432
Number of borrower-
months
1,548,859
1,280,608
268,251
Non-mortgage loans
Whole
sample
Non-committee
Committee
=Committee -
nonCommittee
Incidence Rate
5.63%
5.63%
5.68%
0.05%
Incidence rate Std.Err. (%)
(0.07)
(0.08)
(0.16)
(0.18)
Number of 'failures'
116,884
96,669
20,215
Number of borrowers
324,456
268,955
55,501
Number of borrower-
months
2,074,414
1,718,485
355,929
Online Appendix Page 12
Panel C. Incidence Rates for the Appearance of a New Non-Mortgage Loan
Auto loans incidence rate
Whole
sample
Non-committee
Committee
=Committee -
nonCommittee
Incidence Rate
11.39%
11.35%
11.60%
0.25%
Incidence rate Std.Err. (%)
(0.09)
(0.09)
(0.24)
(0.26)
Number of 'failures'
70,312
58,229
12,083
Number of borrowers
129,936
107,836
22,100
Number of borrower-
months
617,200
513,005
104,195
Bankcard loans incidence rate
Whole
sample
Non-committee
Committee
=Committee -
nonCommittee
Incidence Rate
15.04%
15.08%
14.86%
-0.21%
Incidence rate Std.Err. (%)
(0.09)
(0.10)
(0.26)
(0.28)
Number of 'failures'
102,623
85,188
17,435
Number of borrowers
172,836
143,245
29,591
Number of borrower-
months
682,401
565,085
117,316
All other loans (student, retail, consumer finance, other)
Whole
sample
Non-committee
Committee
=Committee -
nonCommittee
Incidence Rate
21.60%
21.61%
21.54%
-0.07%
Incidence rate Std.Err. (%)
(0.08)
(0.09)
(0.21)
(0.23)
Number of 'failures'
125,422
103,548
21,874
Number of borrowers
173,222
143,008
30,214
Number of borrower-
months
580,671
479,142
101,529
Non-mortgage loans
Whole
sample
Non-committee
Committee
=Committee -
nonCommittee
Incidence Rate
24.22%
24.23%
24.17%
-0.06%
Incidence rate Std.Err. (%)
(0.11)
(0.12)
(0.28)
(0.31)
Number of 'failures'
214,338
177,564
36,774
Number of borrowers
281,609
233,276
48,333
Number of borrower-
months
884,942
732,778
152,164
Online Appendix Page 13
Table OA8. Financial Committee Membership and Non-Mortgage Delinquencies
The table presents exponential hazard analysis for the 90-day delinquency of non-mortgage loans
of the borrowers whose mortgages became delinquent between January 2009 and December
2010. Finance Committee Member is a binary variable that is one if the loan is associated with an
individual with a house located in a district whose U.S. House representative is a member of the
House Financial Services Committee. The control variables include those in Table 2 as well as a
6-month lag of the logged balance of the loan in question and the logged number of loans for the
type in question. All the regressions include state*month fixed effects. Standard errors are robust
to clustering at the congressional district level and are in parentheses. ***, **, and * denote
statistical significance at the 1%, 5%, and 10% levels, respectively. The data on household non-
mortgage liabilities are from the Equifax CRISM dataset.
(1)
(2)
(3)
(4)
Auto Loans
Bankcard Loans
All other loans
(student, retail,
consumer finance,
other)
Non-mortgage
Loans
Finance committee
member
0.0338*
0.0011
0.0046
-0.0035
(0.0201)
(0.0127)
(0.0110)
(0.0091)
Majority Party
-0.0533***
-0.0096
-0.0117
-0.0241**
(0.0193)
(0.0118)
(0.0109)
(0.0097)
Borrower-level
controls
Yes
Yes
Yes
Yes
Loan-level controls
Yes
Yes
Yes
Yes
State*month fixed
effects
Yes
Yes
Yes
Yes
Clustering
Congressional
District
Congressional
District
Congressional
District
Congressional
District
# of Loan-months
1,274,449
1,484,436
1,548,859
2,074,414
# of Loans
177,858
218,454
234,914
324,456
Online Appendix Page 14
Table OA9. Financial Committee Membership and Non-Mortgage Delinquencies in
Subsamples
The table presents exponential hazard analysis for the 90-day delinquency of non-mortgage loans
of the borrowers whose mortgages became delinquent between January 2009 and December
2010. The sample for panel (A) is loans associated with borrowers with a FICO score below the
median at the time of first 90-day mortgage delinquency. The sample for panel (B) is loans
associated with borrowers with low- or no documentation for their mortgage. Finance Committee
Member is a binary variable that is one if the loan is associated with an individual with a house
located in a district whose U.S. House representative is a member of the House Financial
Services Committee. The control variables include those in Table 2 as well as a 6-month lag of
the logged balance of the loan in question and the logged number of loans for the type in
question. All the regressions include state*month fixed effects. Standard errors are robust to
clustering at the congressional district level and are in parentheses. ***, **, and * denote
statistical significance at the 1%, 5%, and 10% levels, respectively. The data on household non-
mortgage liabilities are from the Equifax CRISM dataset.
Panel A. FICO score below median at the time of first 90-day mortgage delinquency
(1)
(2)
(3)
(4)
Auto Loans
Bankcard Loans
All other loans
(student, retail,
consumer finance,
other)
Non-mortgage
Loans
Finance
committee
member
0.0313
-0.0012
0.0036
-0.0108
(0.0219)
(0.0198)
(0.0152)
(0.0113)
Majority Party
-0.0356*
-0.0002
-0.0104
-0.0083
(0.0197)
(0.0166)
(0.0137)
(0.0107)
Borrower-level
controls
Yes
Yes
Yes
Yes
Loan-level
controls
Yes
Yes
Yes
Yes
State*month fixed
effects
Yes
Yes
Yes
Yes
Clustering
Congressional
District
Congressional
District
Congressional
District
Congressional
District
# of Loan-months
532,854
319,042
607,042
739,084
# of Loans
79,398
57,031
96,521
137,020
Online Appendix Page 15
Panel B. Mortgages with Low Documentation or Documentation Information Missing
(1)
(2)
(3)
(4)
Auto Loans
Bankcard
Loans
All other loans
(student, retail,
consumer finance,
other)
Non-mortgage
Loans
Finance committee
member
0.0459**
0.0073
0.0238*
0.0049
(0.0221)
(0.0148)
(0.0142)
(0.0125)
Majority Party
-0.0410*
-0.0098
-0.0077
-0.0248**
(0.0216)
(0.0134)
(0.0132)
(0.0116)
Borrower-level
controls
Yes
Yes
Yes
Yes
Loan-level controls
Yes
Yes
Yes
Yes
State*month fixed
effects
Yes
Yes
Yes
Yes
Clustering
Congressional
District
Congressional
District
Congressional
District
Congressional
District
# of Loan-months
793,324
940,988
966,040
1,307,336
# of Loans
109,823
136,798
145,515
201,311
Online Appendix Page 16
Table OA10. Financial Committee Membership and New Non-Mortgage Loans
The table presents exponential hazard analysis of obtaining a new non-mortgage loan by the
borrowers whose mortgage became delinquent between January 2009 and December 2010.
Finance Committee Member is a binary variable that is one if the loan is associated with an
individual with a house located in a district whose U.S. House representative is a member of the
House Financial Services Committee. The control variables include those in Table 2 as well as a
6-month lag of the logged balance of the loan in question and the logged number of loans for the
type in question. All the regressions include state*month fixed effects. Standard errors are robust
to clustering at the congressional district level and are in parentheses. ***, **, and * denote
statistical significance at the 1%, 5%, and 10% levels, respectively. The data on household non-
mortgage liabilities are from the Equifax CRISM dataset.
(1)
(2)
(3)
(4)
Auto Loans
Bankcard Loans
All other loans
(student, retail,
consumer finance,
other)
Non-
mortgage
Loans