The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Federal Reserve Bank of Chicago
Does the Community Reinvestment Act
Influence Lending?
An Analysis of Changes in Bank Low-Income
Mortgage Activity
By: Drew Dahl, Douglas D. Evanoff and
Michael F. Spivey
WP 2000-06
Does the Community Reinvestment Act Influence Lending?
An Analysis of Changes in Bank Low-Income Mortgage Activity
Drew Dahl *
Department of Business Administration
Utah State University
Logan, UT 84322-3510
Douglas D. Evanoff
Research Department
Federal Reserve Bank of Chicago
Chicago, IL 60068
Michael F. Spivey
Department of Finance
Clemson University
Clemson, S.C. 29634
May, 2000
* The authors wish to thank Dan Aaronson, Eli Brewer, Marsha Courchane, Mark Flannery, Jim
Moser, Wayne Passmore, Owi Rvivivar, Alicia Williams, Lorrie Woos and participants at the
Federal Reserve Bank of ChicagoZs 1999 Conference on Bank Structure and Competition and at
the 1999 Western Economic Association Meetings for constructive comments on earlier drafts.
The views expressed, however, are those of the authors and do not necessarily reflect the views
of our colleagues mentioned above, or others including the Federal Reserve Bank of Chicago or
the Federal Reserve System.
Does the Community Reinvestment Act Influence Lending?
An Analysis of Changes in Bank Low-Income Mortgage Activity
Abstract. Anecdotal evidence that the Community
Reinvestment Act (CRA) influences the lending behavior of financial
institutions has not been uniformly supported by empirical research.
We revisit this issue by evaluating changes in low-income mortgage
lending at commercial banks over the 1992 –96 period. Our empirical
results fail to support a hypothesis that banks respond to public and
regulatory pressure exerted as a result of a downgrade in CRA rating
by increasing low-income mortgage lending. The findings are
consistent with the contention that during this period regulators
stressed adjustments in the lending process of banks (e.g.,
documentation of lending programs and efforts directed at targeted
markets) more than lending performance. The findings underscore the
importance of regulatory efforts made later in the decade to more
closely link enforcement of the CRA to lending outcomes.
2
Does the Community Reinvestment Act Influence Lending?
An Analysis of Changes in Bank Low-Income Mortgage Activity
1.
INTRODUCTION
Since its inception more than twenty years ago, the Community Reinvestment Act (CRA) has
played a controversial role in encouraging banks and other financial institutions to make loans to
borrowers in neighborhoods with low or moderate incomes. The legislation reflects the notion that
banking is fundamentally a local business and depositories should reinvest in local markets. It has been
criticized, simultaneously, for both lenience in the assessment of institutional performance and rigor in
the allocation of credit (see, among others, Garwood and Smith, 1993). These criticisms underscore the
ongoing debate concerning the effectiveness of the CRA in redistributing financial resources to lowincome neighborhoods.
Empirical research on the impact of the CRA on bank lending is mixed. Cross-sectional studies
of banks which vary by CRA rating find that banks with good, or improving, CRA ratings have high, or
increasing, ratios of total loans to assets (Gunther, 1999; Harrison and Seiler, 1999; Bierman, Fraser and
Zardkoohi, 1994). This appears to suggest that supervisors consider overall lending in assessing CRA
performance. These studies, however, do not directly address the issue of whether or not the CRA causes
lending behavior to change, nor do they isolate lending targeted under the CRA from lending generally.
Other studies have examined the effect of the CRA on bank lending within low-income areas or
to minorities (Schill and Wachter, 1994; Evanoff and Segal, 1996). These studies compare mortgage loan
3
screening by banks, which are subject to the CRA, and mortgage loan screening by mortgage companies,
which are not subject to the CRA. Neither study is able to confirm that the CRA alters mortgage credit
allocation within low-income or minority areas. A drawback of these studies is that mortgage companies
may be influenced by potential regulation even though they are not currently subject to the CRA.
Other studies examine the CRA and bank profitability in tests of competing theories that (1) the
CRA, by forcing financial institutions to make unprofitable loans, hurts profitability, or (2) the CRA, by
directing lenders to potentially overlooked markets, helps profitability. Canner and Passmore (1997) find
that the profitability of mortgage lending in low-income areas generally is comparable to mortgage
lending in other areas, while Harrison (1999) concludes that CRA lending at banks leads to decreased
bank profitability.1
In this study, we revisit the issue of how the CRA affects bank lending. An empirical advantage,
relative to prior research, is that we provide an intra-industry test of the hypothesis that banks respond
to public and regulatory pressure under the CRA by increasing the quantity of mortgage loans to lowincome borrowers. In this regard, our empirical approach to studying the CRA compares banks under
supervisory pressure with other banks, rather than with mortgage companies, focuses on lending activities
which are explicitly targeted under the CRA and does not require assumptions concerning relationships
between lending and profitability.
Our methodology analyzes changes in low-income mortgage lending for 170 banks that
experienced a CRA rating downgrade between 1991 and 1995. We hypothesize that these banks, relative
to a control sample of several thousand banks which did not experience a ratings downgrade, would
increase their quantities of low-income mortgage lending after being downgraded. Such a response would
be predicated, presumably, on the potentially adverse consequences of public criticism and/or possible
4
regulatory denials of applications for future mergers or other organizational changes (U.S. General
Accounting Office, 1995).
Our findings do not support the hypothesis that banks responded to supervisory pressure for CRA
compliance during this period by significantly increasing low-income mortgage lending. This should be
of interest to policymakers who, judging from ongoing revisions of the guidelines under which the CRA
is implemented, struggle to achieve the ActZs original lending objectives.2 In particular, during the period
analyzed regulators did not explicitly tie CRA ratings to lending performance, but rather to the lending
process including documentation of lending programs and efforts directed at targeted markets. Although
one might hope that process and output are correlated, they need not be. In addition to having
implications for the ongoing CRA debate, the results are also relevant to the more general literature
concerning how firms respond to regulatory enforcement mechanisms.
The paper is organized as follows. A discussion of the CRA, with an emphasis on its potential
implication for lending behavior, is presented in the next section. Our sample and methodology are
described in Section 3, which also includes descriptive statistical information. Section 4 discusses our
empirical results and section 5 concludes.
2.
BACKGROUND
The CRA was passed by Congress and signed into law by the President in 1977. It directs Federal
regulatory agencies to encourage banks and savings associations ^to help meet the credit needs of local
communities] in a manner ^consistent with safe and sound operations.] The major impetus for the CRA
was a concern by community groups that banks and thrifts were not responding adequately to the credit
needs of local communities. Banks argued that the legislation ran counter to existing safety and soundness
5
regulation.
The CRA, as originally passed, did not mandate how the objective of serving community needs
was to be accomplished.3 Implementation was left to regulatory agencies, which established 12
assessment factors intended to reflect the performance of a financial institution in determining community
credit needs, marketing credit, participating in community development, maintaining branch offices and
avoiding discriminatory credit policies. Supervisory agencies were required explicitly to assess an
institutionZs record in meeting the credit needs of low- and moderate-income neighborhoods.
Supervisors assess CRA performance in formal examinations from which a composite rating is
determined. The ratings from this process initially were confidential. Additionally, there were no explicit
sanctions imposed on institutions that were found not to have adequately served the needs of their
communities. Instead, regulators considered the CRA rating along with other safety and soundness factors
when ruling on applications for geographic expansion through mergers and acquisitions, new branches
or related operational changes.
During this early period, bankers frequently complained that the CRA was both costly to
administer and vague in application. In response, regulatory agencies periodically issued policy
statements delineating how the assessment criteria were scored and discussing elements of effective CRA
programs. These statements emphasized the importance of CRA compliance effort and the documentation
of that effort (Garwood and Smith, 1993). Following heightened public attention to the volume and
complexity of CRA issues, Congress amended the act in 1989 to require the public release of CRA
ratings. Supervisory guidelines continued to emphasize effort and documentation.
In 1995, amid increasing complaints (by bankers) about excessive reliance on paperwork and
increasing complaints (from community groups) that the CRA was ineffective, the regulatory agencies
6
issued new guidelines which based CRA evaluation primarily on how well an institution was helping to
meet the credit needs of local communities. In a conscious movement away from effort and
documentation, the evaluation process introduced three specific performance tests--a lending test, an
investment test and a service test. The lending test assesses the number, amount and distribution across
income groups and geographic areas of a financial institutionZs mortgage loans, small business loans,
small farm loans and consumer loans.
The new guidelines were put into place in July 1997 for smaller banks (less than $250 million)
and in July 1998 for large banks. The general consensus from discussions with bank supervisory
personnel is that the lending test is by far the most important element of the new assessment process, and
mortgages loans the most important type of loan. The other performance elements in the evaluation tend
to have marginal impacts on the overall grade once the lending activity is taken into account (see Federal
Reserve Bank of Chicago, 1999).
The preceding overview of the CRA underscores difficulties encountered in developing regulation
that achieves appropriate policy objectives in the banking industry without adversely distorting bank
behavior.4 Indeed, there are numerous examples of regulation having unintended, and in some cases,
undesirable, effects. Specific examples include unintended responses to deposit insurance (Brewer and
Mondschean, 1994), to reserve requirement provisions (Evanoff, 1990), to price and geographic barriers
(Evanoff, 1988), and to credit allocation programs (Boyes, Hoffman and Low, 1986; Williamson, 1994;
Thakor and Beltz, 1994).
Our study extends this line of research insofar as banks may have attempted to convince
supervisors that they were adhering to the CRA guidelines by providing voluminous documentation and
evidence of compliance effort. As a result, the presumed intent of the regulation, to increase lending in
7
low-income markets, may not have been achieved--and, moreover, may not even have been a planned
objective of the regulated banks if, in a conscious effort to satisfy regulators, banks may have become
more involved in process than performance. Thus, the stated objectives of CRA may not have been
realized.
3.
DATA AND METHODOLOGY
We investigate the response of banks to regulation imposed under the CRA. We hypothesize that
banks experiencing a ratings downgrade in the early to mid 1990s attempted to reestablish an acceptable
rating. In question is whether the response is in the form of increased lending, as is typically associated
with the intent of CRA, or an alternative means aimed at satisfying regulators.
We focus on low-income mortgage lending because it is the category of lending most commonly
associated with the CRA (our definition, in fact, directly reflects those delineated by the CRA). From a
practical standpoint, reporting requirements under Home Mortgage Disclosure Act (HMDA) provide
extensive information on borrower characteristics for these types of loans. This offers an opportunity to
isolate a form of lending specifically targeted by the CRA.5
3.1
The sample
We obtained CRA ratings from 1990 to 1995 for commercial banks supervised by the Federal
Reserve, the Federal Deposit Insurance Corporation (FDIC) or the Office of the Comptroller of the
Currency (OCC). From these ratings, we identified banks that experienced a ratings downgrade in any
year over this period. Downgrades include those from ^outstanding] to ^satisfactory] as well as those from
^satisfactory] to ^needs to improve] or ^substantial noncompliance.]
8
Our sample of banks experiencing downgrades over this period is limited by the relatively
infrequent incidence of changes in ratings, a failure to match entities across the HMDA data base and the
Reports of Condition and Income (which is the source of our financial data) and the relative scarcity of
banks with inferior ratings. For instance, in 1995, less than 5 per cent of the ratings in our sample were
assigned to categories lower than ^outstanding] or ^satisfactory.]=
Our control sample consists of commercial banks that did not experience a downgrade in any year
of the sample period and downgraded banks observed in years prior to their downgrade. Adjustment for
missing observations and related data inconsistencies narrowed our sample to 170 downgraded banks.
Pooling across time, 1991 to 1996, yielded a final total pooled sample of 14,776 annual bank
observations.6
3.2
Modeling low-income mortgage lending
Following a specification for bank lending used by Brewer et al. (2000) and Sharpe and Acharya
(1992) we evaluate adjustments in bank CRA targeted lending by relating the dollar change in the volume
of low-income mortgage loans to a set of variables representing supply and demand factors. By
considering changes in low-income lending, as opposed to levels of low-income lending, we avoid biases
produced by bank-specific propensities which banks may have in favor of, or against, these loans.
Our specification is based on the perceived determinants of changes in low-income mortgage
lending. Low-income mortgage loans include the sum of mortgage loans in low-income areas and
mortgage loans to low-income individuals.7 We standardize changes in a bankZs low-income loans by
its lagged total assets to allow for the substitution of loans across all asset categories. ∆TLOAN(i,t),
therefore, denotes the change in CRA targeted loans: the ratio of low-income mortgage loans to assets
9
for bank i in year t.
To identify the effects of a CRA downgrade, we create a binary variable, CRA_DG(i,t-1), which
equals one for observations on a downgraded bank in year t-1, zero otherwise. A significant response to
CRA enforcement, in the year after the downgrade, would result in a positive and significant coefficient
on this variable if banks increased low-income mortgage lending relative to other banks.
Because lending assessments under the CRA compare an institutionZs lending relative to other
lenders, it is crucial that our specification allows for variation in lending potential across market areas.
We include a variable, ∆MARKET(i,t), representing the dollar change in the volume of low- or moderateincome mortgage loans for all financial institutions in a bankZs market, normalized by lagged total
mortgage lending in the market. Markets are defined to be Metropolitan Statistical Areas (MSAs).8 The
hypothesized sign on the coefficient for this variable is positive.
Other explanatory variables include the lagged ratio of low-income mortgage loans to total assets,
TMORTG/ASSET(i,t-1), which will be negatively correlated with the change in targeted mortgage lending
to the extent that bank asset allocations revert over time toward some optimum level. The lagged ratio
of equity capital to assets, ECAP(i,t-1), may be important if low-income mortgage lending is impacted by
a bankZs financial condition. A positive coefficient on this variable would reflect the notion that
capitalization creates the capacity for loan growth (Brewer et al., 2000). A negative effect may be
associated with the retention of cash flows, as opposed to reinvestment of cash flows in low-income
lending initiates (Harrison, 1999).
The log of lagged bank assets, ASSET(i,t-1), accounts for possible effects of bank size on lowincome mortgage lending.9 A binary variable, HOLD(i,t-1), accounts for possible impacts of affiliation
status (HOLD = 1 if a bank is affiliated with a holding company, 0 otherwise). The signs on the
10
coefficients for these variables are indeterminate.
A binary variable indicating supervisory status controls for differences that may exist among
regulators in their approaches toward low-income mortgage lending. This variable, FDIC(i,t-1), is one if
a bank is regulated by the Federal Deposit Insurance Corporation (FDIC) and zero if it is regulated by the
Federal Reserve or the Office of the Comptroller of the Currency (OCC). The sign on the coefficient for
this variable will be positive if banks regulated by the FDIC expand low-income mortgage lending
relative to other banks. Such a finding could be attributed to a greater emphasis by the FDIC on lowincome mortgage lending as argued by Thomas (1998). It also could be attributed to the existence of
omitted variables from our model which render banks supervised by the FDIC more capable of expanding
low-income lending.
Thus, we propose that changes in a bankZs low-income mortgage lending are influenced by (1)
the inherited level of low-income mortgage loans; (2) the change in low-income mortgage loans in a
bankZs market; and (3) lagged bank-specific variables. The basic equation is:
∆TLOAN(i,t) = a + d1CRA_DG(i,t-1) + d2ASSET(i,t-1) + d3∆MARKET(i,t) + d4ECAP(i,t-1)
+ d5TMORTG/ASSET(i,t-1) + d6FDIC(i,t-1) + d7HOLD(i,t-1) + e(i,t)
(1)
where di are coefficients associated with each parameter and e(i,t) is an error term.
The final econometric issue associated with equation (1) concerns sample selectivity. If CRA
downgrades occur randomly, then the effect of a downgrade would be correctly impounded in the OLS
coefficient on CRA_DG(i,t-1). However, this specification suffers from an omitted variables bias if the
regulator’s bank downgrade decision is not random; and few would argue this is a random decision. We
11
account for this potential bias by estimating the ^treatment effects] model described by Greene (1993,
p. 713) and Maddala (1983, p. 260).
3.3
The treatment effects model
To isolate the effect of the CRA downgrade, equation (1) can be rewritten as:
∆TLOAN(i,t) = d1 (CRA_DG) + β’ X + e
(2)
where X is a vector of the explanatory variables other than the regulatory downgrade, and β a vector of
the corresponding coefficients. Since the downgrade decision is not random, but rather the result of a
conscious decision by regulators, the value of CRA_DG may reflect selection bias causing parameter
estimates to be inconsistent. To correct for this potential, we first model the downgrade decision
explicitly. Assume that regulators consider a bank’s characteristics when deciding whether to downgrade
it for CRA compliance and there exists an unobservable factor, CRA_DG*, that drives the downgrade
decision. We model CRA_DG* as a function of a vector of variables, W. The bank is downgraded if
CRA_DG* > 0, thus:
CRA_DG = 1 if CRA_DG* > 0,
0 otherwise
(3)
and
CRA_DG* = γ W + µ
(4)
where e and µ have bivariate normal distribution with means zero and the covariance matrix:
1
ρσ
ρσ
σ 2
(5)
12
Given the downgrade specification, one can show that 10
E [e | CRA_DG = 1 ] = ρσ φ(γW) / Φ(γW)
and
(6)
E [e | CRA_DG = 0 ] = - ρσ φ(γW) / (1-Φ)(γW)
(7)
where φ(.) and Φ (.) are the probability density function and the cumulative distribution function,
respectively, of the standard normal distribution. Equations (6) and (7) can be combined to generate the
expression:
E [ e | CRA_DG ] = ρσ
ϕ (γ W) {CRA_DG - ΦF(γ W)}
Φ (γ W) {1- Φ(γ W)}
= λ ( CRA_DG, W; γ)
(8)
To estimate the parameters in (1) or (2) consistently, we first estimate the probit model given by
(4), and calculate λ ( CRA_DG, W; γ ) for each observation using the estimate for γ.11 This calculated
value is the observation’s “Inverse Mills Ratio.” If
λ̂ = λ ( CRA_DG, W; γˆ ) represents this estimate
then consistent parameter estimates for the targeted lending model can be obtained by estimating:
∆TLOAN(i,t) = d1 (CRA_DG) + β’ X + βλ λ̂ + …
(9)
via OLS. The new disturbance term … is given by {e – E [ e | DR ]} and is uncorrelated with CRA_DG.
The OLS estimate of βλ gives a consistent estimate of ρσ. This is the approach taken.
If the correction, λ, is not accounted for in equation (9), the specification will over- or under-state
the role of the CRA downgrade depending on the sign of βλ . For example, it may be that banks that are
prone to downgrades are relatively poor managers of their loan portfolios. Thus, for these banks, targeted
loan growth will be relatively low whether or not a CRA downgrade is imposed. This would correspond
to a negative sign for β̂
λ.
Our priors are that this could be a plausible result. The somewhat less plausible
13
scenario would be if downgraded banks were more prone to have greater targeted loan growth regardless
of the presence or absence of a downgrade. Then excluding βλ from equation (9) would tend to overstate
the impact of the downgrade and would correspond to βλ > 0.
3.4
The probit model
Our probit estimation explains the likelihood of a bank experiencing a CRA ratings downgrade
in year t-1 on the basis of lagged (year t-2) bank-specific information. We will use this information for
estimating targeted lending growth in the year following the downgrade (year t). In modeling the
likelihood of a downgrade, we include the ratio of total loans to assets, LOAN/ASSET(i,t-2), to account
for the possibility that supervisors consider a bankZs overall lending activity in their decisions on
downgrades. Bierman et al. (1994) and Gunther (1999) both found evidence that banks with high CRA
ratings tended to have high ratios of total loans to assets. Harrison (1999) similarly concluded that banks
with active loan departments were more likely to receive high CRA ratings. Our hypothesized sign on
the coefficient for this variable, therefore, is negative--i.e., banks with high loan to asset ratios are less
likely to be downgraded.
MARKET(i,t-2), which denotes the ratio of low-income mortgage loans to total mortgage loans for
the market in which a bank operates, is included to account for the possibility that banks in markets with
significant low-income mortgage lending may be more likely to be downgraded than other banks. This
may be related to the imposition of a higher standard under the CRA for banks in markets with greater
potential for low-income mortgage lending.
The ratio of low-income mortgage lending to total loans, TMORTG/ASSET(i,t-2) for bank i in year
t, is included to account for the possibility that loans within specific areas targeted by the CRA legislation
14
play a role in whether or not a bank is downgraded. We hypothesize a negative relationship between this
variable and likelihood of downgrade. We note, however, that previous research linking overall lending
with CRA compliance has failed to link CRA compliance and lending which directly served the local
community (Bierman et al., 1994).
Since the affiliation status, and size, of banks may influence incentives to avoid downgrades, we
include HOLD(i,t-2) and ASSET(i,t-2), as previously defined. Previous research has found that larger banks,
or affiliated banks, are more likely to receive higher CRA ratings (Bierman et al., 1994; Gunther, 1999;
Harrison, 1999; Hylton, 1999). The hypothesized signs on the coefficients for these variables are
therefore negative. However, this effect may be dampened to the extent that supervisors are more critical
of the lending behavior of larger banks.
The ratio of net income to total assets, ROA(i,t-2), is included to account for the role of profitability
on CRA downgrades. The sign on the coefficient for this variable is indeterminate. Bierman et al. (1994)
find that banks with poor profitability have higher CRA ratings, which suggests that these banks have
adopted suboptimal portfolios in order to conform to the CRA.12 Gunther (1999), however, finds that
profitability is negatively related to CRA rating. He states that financial losses may require a redirection
of resources away from CRA objectives.
The ratio of equity capital to total assets, ECAP(i,t-2), also accounts for financial condition.
Harrison (1999) and Gunther (1999) find that higher levels of bank capital increase the chance of a
low CRA rating. Harrison argues that highly-capitalized banks retain cash flows rather than
reinvesting the funds into community development initiatives. The hypothesized sign on the
coefficient for this variable, therefore, is positive--as capital increases, so will the likelihood of a
downgrade.
15
The final variable, FDIC(i,t-2), is an indicator for supervision by the FDIC (as opposed to the
Federal Reserve or the OCC). This variable is intended to reflect different practices of bank
supervisors in determining which banks to downgrade [see U.S. General Accounting Office (1995)
and Thomas (1998)]. A positive relationship between downgrades and regulation by the FDIC could
be attributed to a greater emphasis by the FDIC in identifying CRA inadequacies. It also could be
attributed to the existence of omitted variables from our model which render banks regulated by the
FDIC less capable of maintaining CRA compliance.
Our probit model is therefore:
PR(CRA_DG)(i,t-1) = a + d1HOLD(i,t-2) + d2(LOAN/ASSET)(i,t-2) + d3ASSET(i,t-2)
+ d4MARKET(i,t-2) + d5ROA(i,t-2) + d6ECAP(i,t-2)
+ d7TMORTG/ASSET(i,t-2) + d8FDIC(i,t-2) + u(i,t)
3.5
(10)
Descriptive statistics
Table 1 presents descriptive information on the independent variables used in our analysis.
The sample consists of 14,776 bank observations. We divide the variables into two groups, those that
are used in the probit model and those which are used in the treatment effects model. Differences
exist in both the variables included in each model and in the lag structures when a variable is used in
both models.
The annual change in low-income mortgage lending for bank markets, expressed as a
percentage of lagged market mortgage loans, ∆MARKET(i,t), is .0331 indicating that low-income
mortgage lending increased over our sample period. The data also indicate that a majority of banks in
16
our sample are supervised by the FDIC and are affiliated with a bank holding company.
Table 2 presents a comparison of variable means for downgraded and non-downgraded banks.
These univariate measures indicate that downgraded banks are more likely to be regulated by the
FDIC and are relatively better capitalized institutions. Additionally, downgraded banks exhibit
declines in growth for low-income mortgage lending in the year following the downgrade. While the
decline is not statistically different from zero, the difference in targeted lending between downgraded
and non-downgraded banks is significant, with non-downgraded banks showing slight increases. This
is inconsistent with the notion that banks aggressively respond to downgrades by increasing lending
relative to other banks in categories targeted by the CRA. However, the difference in means is only
significant at the 10 percent level, and no additional influences are being controlled for.
4.
EMPIRICAL RESULTS
In table 3, we present results of our probit model explaining the likelihood of a bank
experiencing a ratings downgrade in year t-1. In table 4, we present results of our treatment effects
model of changes in targeted lending in year t. Two subsamples are considered. The first is the
overall sample and compares downgraded banks to banks that were not downgraded.13 The second
sample compares downgraded banks to banks that were not downgraded, as before, but excludes all
banks that were not in ^merger-mode] over the sample period. A merger-mode bank is one that
experienced a merger at any time during our sample period. This subsample focuses on banks that
were more likely to be sensitive to pressure under the CRA because of outstanding regulatory
applications. In this regard, Sarkar and Johnson (1996) found banks subject to CRA protests--which
usually arise when an institution applies for a merger, acquisition or branch relocation--experienced
17
negative average excess returns. Litan et al. (2000) stressed the potential aggressiveness of merger
active banks in CRA targeted lending.
For the treatment effects model we also redefine our DOWNGRADE variable to equal one for
banks that were downgraded, and subsequently recovered to at least a satisfactory rating by 1995,
zero otherwise. This is useful in isolating a group of banks that were successful in reversing observed
declines in CRA compliance and evaluating whether the recovery was associated with changes in
lending behavior. Thus, we evaluate our entire sample, a subsample of banks which have more to
gain from aggressively responding to a CRA downgrade, and a subgroup which did recover from the
downgrade.
4.1
The probit model
Results for the probit models relating the likelihood of a bank experiencing a CRA downgrade
to bank-specific factors are statistically significant. The signs on the coefficients for individual
variables are generally consistent across the three samples.
The coefficient on HOLD(i,t-2) is negative in both specifications. This suggests that affiliated
banks are less likely to experience a downgrade than independent banks and is consistent with the
conclusions of earlier research (Bierman et al., 1994; Harrison, 1999).
The coefficient on ratio of total loans to assets, LOAN/ASSET(i,t-2), is negative and significant,
as hypothesized, which indicates that banks with high ratios of loans-to-assets are less likely to be
downgraded. The coefficient on TMORTG/ASSETS(i,t-2), however, is insignificant. These twin results
are consistent with the conclusion of Bierman et al. (1994) that CRA ratings are influenced by overall
lending, but not by lending with categories typically associated with the CRA.
The coefficient on MARKET(i,t-2) is positive and significant in two of the specifications. This
18
indicates that lagged ratios of low-income mortgage loans to total mortgage loans, by all financial
institutions in a particular market, positively affect the likelihood of a bank being downgraded. It
suggests that regulators apply higher standards to banks in markets with active low-income mortgage
lending.
Capitalization and profitability are not statistically related to the odds of a bank experiencing a
downgrade. The log of the asset size of a bank, surprisingly, is positively associated with downgrades.
Supervision by the FDIC is found to increase the odds of a bank being downgraded.
4.2
The treatment effects model
All three versions of the treatment effects models are statistically significant. They explain a
substantial minority of the variation in the dependent variable, defined to be the percentage change in
the ratio of low-income mortgage loans to total assets.
Among the independent variables, the change in a marketZs low-income loans, ∆MARKET(i,t) ,
is positively associated with low-income loan growth for banks in all three samples, and is
statistically significant in one of the three samples. This result, as anticipated, indicates that banks
find it easier to expand targeted lending within markets characterized by greater growth potential.
The lagged level of targeted loans for a bank, TMORTG/ASSET(i,t-1) , is negatively and
significantly associated with low-income lending growth, as hypothesized, in all three samples. This
indicates that banks revert over time to optimum levels through a partial adjustment process. The
coefficient on ASSETS(i,t-1) indicates that smaller banks expand targeted lending at faster rates than
larger banks. Banks supervised by the FDIC have slower growth rates in targeted lending (this despite
the fact that they were found to be more apt to be downgraded in the probit model). This may suggest
that FDIC-supervised banks are less capable of expanding low-income lending because of omitted
19
variables from our model, which are associated with supervisory status.
The crucial relationship in our analysis is that between changes in low-income lending and the
binary variable indicating bank observations in the year after the CRA downgrade. The sign on the
coefficient for this variable is positive, but is always insignificant.14 This is inconsistent with the
contention that banks significantly increase low-income lending in the year after being downgraded
relative to banks that were not downgraded. The fact that this relationship is observed among mergermode banks, which would likely be more sensitive to CRA enforcement than other banks, is
particularly striking. Low-income lending outcomes apparently were not a factor in the lending
decisions of these banks following a downgrade.
The finding that downgraded banks that recovered within the time period analyzed were
indistinguishable from other banks in low-income mortgage lending suggests that they recovered due
to changes that were unrelated to this category of targeted lending. Our empirical approach assumes
that banks feel the regulatory scrutiny from CRA enforcement when they are downgraded, and
respond in the next time period to avoid public criticism and avoid having bank regulators deny future
merger applications. To the extent that downgrades are considered signs of increased regulatory
scrutiny, our results are inconsistent with greater low-income lending growth following periods of
heightened CRA enforcement.15
The negative coefficient on the inverse Mills ratio is consistent with our priors that CRA
downgraded banks may have had slower targeted loan growth irrespective of any regulatory action.
However the coefficient is not significant suggesting that the samples are not significantly distorted
by sample selection bias.16 The findings were robust to alternative specifications.
Our results must be qualified by two factors. First, it is possible that bank-specific factors are
20
less closely correlated with low-income mortgage lending, which is relatively small in relation to a
bankZs overall assets, than they have been shown to be in prior lending studies using larger loan
categories (Brewer et al., 2000). This may impair the ability of our model to identify an effect of the
CRA on lending. Second, it is possible that banks are expanding low-income lending in areas other
than mortgages, which our methodology is incapable of discerning. Given our approach to the
problem, however, we find no evidence that banks responded to CRA downgrades by increasing
targeted lending.
5.
SUMMARY AND CONCLUSIONS
We compare the lending behavior of banks experiencing a CRA downgrade to the lending
behavior of several thousand banks that do not experience a downgrade. We hypothesize that banks
would respond to regulatory and public pressure exerted as a result of the downgrade by increasing
the volume of mortgage loans made in CRA targeted markets: in low-income areas or to low-income
individuals. Our results are inconsistent with the hypothesis that downgraded banks, in an attempt to
reestablish an ^acceptable] CRA rating, significantly increase their low-income mortgage lending.
This result is robust to various specifications.
We note that our conclusions are based on a sample period during which regulators did not
take an aggressive approach in tying CRA ratings to lending performance but rather to the lending
process (including documentation of lending programs and efforts directed at targeted markets).17
This suggests that recent regulatory adjustments that more closely link CRA enforcement and lending
outcomes are warranted. Future research over time periods subsequent to our sample period, when the
lending test was imposed, is needed to determine the effectiveness of these revisions.
21
LITERATURE CITED
Barnow, B.S., G.G.Cain, and A.S. Goldberger. (1981) Issues in Analysis of Selectivity Bias,
Evaluation Studies Review Annual 5, 43-59.
Bierman, L., D. Fraser and A. Zardkoohi. (1994) The Community Reinvestment Act: A preliminary
empirical analysis, Hastings Law Journal 45, 383-412.
Boyes, W. J., Dennis Hoffman and Stuart Low. (1986) Lender Reactions to Information Restrictions:
The Case of Banks and the ECOA, Journal of Money, Credit and Banking 18, 211-219.
Brewer, E., and T.H. Mondschean. (1994) An empirical test of the incentive effects of deposit
insurance: The case of junk bonds at savings and loan associations. Journal of Money, Credit, and
Banking, 26, 146-64.
Brewer, E., B. Minton and J. Moser. (2000) Interest rate derivatives and bank lending. Journal of
Banking and Finance, 21, 353-379.
Calomiris, CW., C.M. Kahn and S.D. Longhofer. (1994) Housing-finance intervention and private
incentives: helping minorities and the poor. Journal of Money, Credit, and Banking, 26, 634-74.
Canner, G. and W. Passmore. (1997) The Community Reinvestment Act and the profitability of
mortgage-oriented banks. working paper, Federal Reserve Board of Governors, Washington, D.C.
Community Credit Needs, (1977) Hearings before the Senate Committee on Banking, Housing and
Urban Affairs, 95 Congress, 1 Session (Government Printing Office).
Evanoff, D. (1988) Branch banking and service accessibility. Journal of Money, Credit, and Banking,
20, 191-202.
Evanoff, D. (1990) An empirical examination of bank reserve management behavior Journal of
Banking and Finance 14, 131-143.
Evanoff, D. and L. Segal. (1996) CRA and fair lending regulations: Resulting trends in mortgage
lending. Economic Perspectives, Federal Reserve Bank of Chicago, 19-42.
Evanoff, D. and L. Segal. (1997) Strategic responses to bank regulation: Evidence from HMDA data.
Journal of Financial Services Research 11, 69-92.
Federal Register. (1995) Community Reinvestment Act Regulations and Home Mortgage Disclosure;
Final Rules. May 4, Washington, D.C.
Federal Reserve Bank of Chicago. (1999) A banker’s quick reference guide to CRA. FRB-Chicago,
22
Chicago.
Garwood, G.L. and D.S. Smith. (1993) The Community Reinvestment Act: Evolution and current
issues. Federal Reserve Bulletin 79, 251-67.
Greene, W.H. (1993) Econometric Analysis. New York: MacMillan Publishing Company.
Gunther, J.W. (1999) Between a rock and a hard place: the CRA--safety and soundness pinch.
Economic and Financial Review, Federal Reserve Bank of Dallas (Q2), 32-41.
Harrison, D. and M. Seiler. (1999) Is Community Reinvestment Act (CRA) lending profitable?
Evidence from ratings revisions, Managerial Finance 25, 3-18.
Harrison, D. (1999) Bank specific characteristics as a determinant of Community Reinvestment Act
(CRA) performance, working paper, University of Vermont.
Heckman, J. (1979) Sample Selection Bias as a Specification Error. Econometrica 47, 153-61.
Hylton, Keith N. (1999) Banks and Inner Cities: Market and Regulatory Obstacles to Development
Lending. working paper, Boston University School of Law, No. 99-15.
Johnson, S. and S. Sarkar. (1996) The valuation effects of the 1977 Community Reinvestment Act
and its enforcement. Journal of Banking and Finance 20, 783-803.
Litan, Robert E., Nicolas P. Retsinas, Eric S. Belsky and Susan White Haag. (2000) The Community
Reinvestment Act After Financial Modernization: A Baseline Report, Department of the Treasury,
Washington, D.C.
Maddala, G.S. (1983) Limited-Dependent and Qualitative Variables in Econometrics. Cambridge:
Cambridge University Press.
Schill, M. and S. Wachter. (1994) Borrower and neighborhood racial and income characteristics and
financial institution mortgage application screening, Journal of Real Estate Finance and Economics 9,
223-239.
Sharpe, S.A. and S. Acharya. (1992)Loan Losses, Bank Capital, and the Credit Crunch. Working
Paper, Federal Reserve Board of Governors, Washington, D.C.
Thakor, T.A. and J. Beltz. (1994) A ‘Barter’ Theory of Bank Regulation and Credit Allocation,
Journal of Money, Credit and Banking 26, 679-705.
Thomas, K.H. (1998) The CRA Handbook. McGraw-Hill: New York.
23
U.S. General Accounting Office. (1995) Community Reinvestment Act: Challenges Remain to
Successfully Implement CRA (Washington, D.C.).
Van Order, Robert. (1994) Comment on Housing-finance intervention and private incentives: helping
minorities and the poor. Journal of Money, Credit, and Banking, 26, 675-78.
Williamson, S.D. (1994) Do Informational Frictions Justify Federal Credit Programs?
Journal of Money, Credit and Banking, 26, 523-544.
24
Table 1 - Descriptive Statistics, Overall Sample
Panel A -- Independent variables used in the probit model
VARIABLE
MEAN
STD. DEV.
MINIMUM
MAXIMUM
HOLD(t-2)
0.7529
0.4313
0.0000
1.0000
LOAN/ASSET(t-2)
0.5669
0.1377
0.0041
0.7957
ASSETS(t-2)
11.522
1.314
8.601
19.164
MARKET(t-2)
0.1113
0.1046
0.0007
0.6712
ROA(t-2)
0.0096
0.0092
-0.1427
0.5569
ECAP(t-2)
0.0866
0.0265
-0.0118
0.5403
TMORTG/ASSET(t-2)
0.0030
0.0206
0.0000
0.6599
FDIC(t-2)
0.6244
0.4842
0.0000
1.0000
Panel B _ Independent variables used in the treatment effects model
ASSET(t-1)
11.611
1.32
8.545
19.209
ECAP(t-1)
0.0876
0.0262
-0.0118
0.5403
FDIC(t-1)
0.6244
0.4842
0.0000
1.0000
HOLD(t-1)
0.7529
0.4313
0.0000
1.0000
∆MARKET(t-1)
0.0331
0.0952
-0.7760
0.9999
TMORTG/ASSET(t-1)
0.0030
0.0185
0.0000
0.6222
Note: Year (t-1) denotes the year in which a bank is downgraded.
25
Table 2 - Comparison of Means, Downgraded Banks Versus Other Banks
Downgraded
Banks
Other
Banks
VARIABLE
MEAN
MEAN
∆TLOAN(t)
-.0013
.0003*
ASSETS(t-1)
11.636
11.610
ECAP(t-1)
.092
.087*
FDIC(t-1)
.941
.621***
HOLD(t-1)
.712
.753
∆MARKET(t-1)
.032
.033
TMORTG/ASSET(i, t-1)
.005
.003
Notes: . Year (t-1) denotes the year in which a bank is downgraded. . *** indicates statistical
significance at the 1% level, ** at the 5% level, and * at the 10% level in the difference in
means for the two groups.
26
Table 3 - Probit Model Estimates of the Likelihood of a Bank Being Downgraded
Dependent variable: PR (CRA_DG)(t-1)
All Banks
Merger-Active
Banks
Intercept
3.206***
(-8.77)
3.143***
(-4.06)
HOLD(i, t-2)
-0.057
(-0.79)
0.007
(0.04)
0.952***
(-4.47)
1.339***
(-2.94)
0.057**
(2.15)
0.063
(1.22)
1.096***
(4.48)
0.097
(0.15)
ROA(i, t-2)
0.568
(0.17)
-0.348
(-0.04)
ECAP(i, t-2)
-0.183
(-0.15)
-0.244
(-0.08)
TMORTG/ASSET(i, t-2)
1.027
(0.87)
2.504
(1.16)
FDIC(i,
0.876***
(8.09)
1.102***
(4.73)
LOAN/ASSET(i,
ASSETS(i,
t-2)
t-2)
MARKET(I,
t-2)
t-2)
Log Likelihood
-856
-217.13
Number of Bank
Observations
14,776
3,668
170
45
Downgraded Bank
Observations
Notes: The dependent variable is a binary variable equal to 1 (0) if a bank was (not) downgraded
for CRA performance during a particular year. t statistics are in parentheses. *** indicates
statistical significance at the 1% level, ** at the 5% level, and * at the 10% level.
27
Table 4 - Treatment Effects Model Estimates
Dependent Variable: ∆TLOAN(i,
t)
Recovered
Banks
All
Banks
Merger
Active
Banks
0.003***
(3.20)
0.002
(1.42)
0.002***
(3.15)
0.003
(0.57)
0.001
(0.02)
0.015
(0.18)
-0.001**
(-2.40)
-0.001
(-0.22)
-0.001***
(-2.34)
ECAP(i, ,t-1)
-0.002
(-0.53)
-0.003
(-0.63)
-0.002
(-0.52)
FDIC(i,
-0.001**
(-2.29)
-0.001
(-0.53)
-0.001***
(-2.32)
-0.001
(-0.05)
-0.001**
(-2.01)
-0.001
(-0.11)
0.003***
(4.22)
0.004***
(3.85)
0.003***
(4.24)
-0.075***
(-18.59)
-0.053***
(-6.40)
-0.075***
(-18.59)
MILLS
-0.002
(-0.83)
-0.001
(-0.27)
-0.006
(-1.61)
Number of
Observations
14,776
3,668
14,776
Adjusted R2
.0240
.0153
.0239
Intercept
CRA_DG (i,
ASSETS(i,
t-1)
t-1)
t-1)
HOLD(i,
t-1)
∆MARKET (i,
t)
TMORTG/ASSET(i, t-1)
Notes: For the first two columns, downgrade (CRA_DG ) is as defined in Table 3. For the
last column, it is defined as a bank that is downgraded and recovers during our sample period.
t-statistics (from a consistent covariance matrix corrected for heteroskedasticity) are
shown in parentheses. ***Indicates statistical significance at the 1% level, ** at the 5%
level, and * at the 10% level.
28
1
Johnson and Sarkar (1996) examine the effect of passage of the CRA on the stock market
valuations of banks and savings and loan associations. They conclude that the expected costs of
the CRA exceed the expected benefits among small institutions only.
2
For example, the Financial Services Modernization Act of 1999 (Gramm-Leach-Bliley Act)
specifically directs the U.S. Treasury ^to study the extent to which adequate services are being
provided as intended by the CRA.]=Concerning the Act’s lending objectives, in describing the
original intent of the act its principal sponsor said it was to “have banks be more sensitive…to
their responsibilities to provide for local community needs;” see Community Credit Needs
(1977).
3
The potential conflict between CRA performance and bank safety is probably one reason
the regulators did not originally require explicit CRA performance criteria. Gunther (1999)
provides evidence consistent with ^a supervisory process in pursuit of conflicting goals,
particularly at smaller-sized banks.]
4
For a discussion of the appropriate role of government intervention in housing markets see
Calomiris, Kahn and Longhofer (1994), and Van Order (1994).
5
We recognize that low-income mortgage lending does not capture the entirety of potential changes
in bank behavior resulting from CRA enforcement, such as low-income lending in non-mortgage
areas, partnerships with community groups, ^layered] lending or additional small business lending.
Mortgage activity, however, is the lending typically associated with the CRA and there have been
increased interests recently expressed in evaluating the relationship between CRA and mortgage
lending [e.g., Litan et al. (2000) and Hylton (1999)].
6
We excluded a small number of observations (less than two per cent) identified as outliers on
the basis of our independent variable, which measures changes in low-income mortgage lending
(see section 3.2). We repeated the analysis with the unrestricted sample. The results were
qualitatively the same as those reported in the paper.
7
"Low-income areas] are census tracts where the median income is less than 80% of the MSA
median income (technically, this includes what is considered low- and moderate-income areas
under the CRA).
8
For banks operating in multiple markets this was calculated as a weighted average of its
activity across markets. Market mortgage lending includes lending by non-bank institutions (this
is appropriate because the variable is included to capture the overall market lending activity).
9
Assets are measured in thousands.
10
See Greene (1993) or Maddala (1983).
29
11
See Greene or Barnow, Cain, and Goldberger (1981). The general approach is attributable to
Heckman (1979).
12
It is also possible that this variable reflects managerial talent--if so, bad managers may be less
profitable and less able to satisfy CRA requirements.
13
The latter category includes observations on downgraded banks in years prior to the
downgrade.
14
Various tests of robustness were preformed. We altered our lag structure on the time allowed
for bank response after the regulatory downgrades. One specification focused on changes in
mortgage lending during the year of the downgrade as well as the following year, and another
specification focused on changes in mortgage lending during the two years after the year of
downgrade. Our conclusions were unaffected by these modifications.
15
We differentiated with respect to the type of downgrade to see if this affected the results
insofar as banks may be less concerned about maintaining an “outstanding” rating than they are
with insuring that they do not fall below the “satisfactory” rating. To test for this we employed
more detailed measures of downgrades. Again, we found no impact on lending following the
downgrade.
16
Results in column 3 of Table 4 were generated with the same Mills ratio as used in the first
column—i.e., based on the probit using all downgraded banks. While this is the appropriate
means to account for potential sample selection bias, for completeness we reestimated the model
defining DOWNGRADE as a bank that was downgraded and also recovered (i..e., the model in
column 3) and using a Mills ratio based on a probit using this definition for the binary variable.
The results were nearly identical. We also estimated the targeted-lending relationship [equation
(1)] with OLS to insure we were not introducing distortions with the treatment effects model
without gaining any of the potential benefits. Again, the basic findings were unchanged.
Hylton (1999) questions whether recent changes will have much of an impact on bank lending
behavior.
17
30
@FedFRASER