Wholesale Price Monitoring in the Age of Tough Enforcement

Transcription

1 Wholesale Price Monitoring in the Age of Tough Enforcement Melanie H. Brody, Partner, K&L Gates LLP Ric Pace, Principal, PricewaterhouseCoopers LLP Copyright 2010 by K&L Gates LLP. All rights reserved Copyright 2010 PricewaterhouseCoopers LLP. All rights reserved.

2 Overview Deciding whether to monitor Legal framework applicable to wholesale pricing Wholesale enforcement cases Structuring and implementing a wholesale monitoring program Addressing results 1

3 Wholesale Pricing Overview Categories of Lending Retail Loan officers are employees of the lender 100 percent of a loan officer s loans are funded by the lender Lender can cap and monitor discretionary pricing Wholesale Brokers are independent third parties Typically, a relatively small percentage of a broker s loans go to any one lender More difficult to meaningfully monitor pricing 2

4 To Monitor or Not To Monitor? Many lenders monitor retail pricing Historically, fewer have monitored wholesale pricing Pros of wholesale monitoring: Can help identify brokers who are creating potential fair lending risk Demonstrates commitment to fair lending Cons of wholesale monitoring Expense and business disruption Limited volume from individual brokers Within vs. across broker differences 3

5 Example of Within Broker vs. Across Broker Broker A Sends 5 white loans All loans: 1 point Broker B Sends 5 African-American loans All loans: 1.5 points Within Broker: No issue Across Broker: When Broker A s loans are combined in the lender s portfolio with Broker B s, there is a difference between what the white and African-American borrowers were charged 4

6 Legal framework Federal Antidiscrimination statutes Equal Credit Opportunity Act Fair Housing Act Prohibit lenders from charging minority borrowers more than similarly situated non-minority borrowers But are lenders legally responsible for pricing differences caused by third party mortgage brokers? 5

7 Generally, loan pricing is a function of 2 main components 1. Objective factors: loan product, borrower credit and transaction characteristics 2. Discretionary factors*** Retail loans: underage/overage Wholesale loans: broker compensation Differences in wholesale loan pricing across borrower groups is usually a function of differences in broker compensation 6

8 Are Lenders Responsible for Broker Compensation Differences? Legal vs. Policy Answer Legal Answer 1. FHA: Meyer v. Holley, 537 U.S. 280 (2003): FHA does not impose a nondelegable duty on persons to ensure that the parties with whom they do business do not discriminate. Liability should be determined under principles of vicarious liability. 2. ECOA: A person is not a creditor for purposes of another creditor s ECOA violation, unless the person knew of the violation before becoming involved in the transaction. Policy Answer Many government authorities believe lenders should be liable 7

9 Wholesale Pricing Enforcement Cases 1996: DOJ Settlement with Long Beach Mortgage Included compensation for wholesale borrowers No wholesale monitoring 2000: DOJ Settlement with Delta Funding Included wholesale price discrimination claims 2006: NYAG Settlement with Countrywide Included wholesale price discrimination claims Included wholesale enhanced monitoring requirements 2008: FTC Settlement with Gateway Funding Included wholesale price discrimination claims, suspended judgment and price monitoring program 8

10 Wholesale Pricing Enforcement Cases 2010: DOJ Settlement with AIG DOJ alleged African-American borrowers paid higher broker fees than similarly situated non-hispanic white borrowers Settlement included: 6.1 Million in restitution and $1 Million toward consumer education Broker fee monitoring, disclosures, manager review of broker compensation if AIG reenters wholesale lending business 9

11 Keep in mind There is currently no formal information available on how the federal bank regulatory agencies perform their own statistically based fair lending tests of mortgage loan pricing. There is no uniform testing methodology or approach used by these agencies that a lender can reliably leverage in designing its own monitoring program. There is no silver bullet approach only different alternatives none of which is perfect or beyond regulatory critique. 10

12 In general, the development of a wholesale price monitoring program involves the following steps: 1) Develop an external view of wholesale fair lending risks through statistical analysis of HMDA data. 2) Select appropriate non-hmda loan price measure(s) for effective fair lending compliance monitoring. 3) Identify baseline wholesale loan pricing differences between protected class and control group borrowers and use these baseline pricing differences to riskfocus the fair lending regression analysis. 4) Develop regression models to estimate demographically neutral loan prices for each wholesale loan. 5) Test for statistically significant unexplained loan price differences between protected class and control group borrowers at various geographic levels and by originating broker. 6) As necessary, implement appropriate corrective actions. 11

13 1) Develop an external view of wholesale fair lending risks through statistical analysis of HMDA data: While not a crucial part of a monitoring program, this step can provide an important third-party lens through which to view your lending outcomes and identify elevated fair lending risk areas. Typically designed similar to the following example: Loan Price Metric Frequency of APR Rate-Spread Reportable Loans Average APR Rate-Spread Amount Hispanic Borrowers White Non- Hispanic Borrowers Difference Statistically Significant? 23.1% 15.6% 7.5% No 3.78% 3.32% 0.46% Yes Elevated fair lending risk may be present when differences are: Adverse to one or more protected classes, and Statistically significant, and Relatively large in magnitude 12

14 2) Select appropriate non-hmda loan price measure(s) for effective fair lending compliance monitoring: APR Rate Spread has certain limitations for fair lending monitoring: It is only reported for a small subset of a lender s originations Where it is reported, it is expressed as a relative price (as opposed to an absolute price such as APR) For wholesale originations, it is not a direct measure of the discretionary fees paid by borrowers that create most, if not all, of the fair lending risk in loan pricing Most lenders, and many regulators, instead use the following two loan price metrics: Total Broker Fees ( TBF ) Annual Percentage Rate ( APR ) 13

15 Total Broker Fees: Measured as the sum of: Any up-front broker points paid directly by the borrower to the broker Any yield spread premium paid directly by the lender to the broker at closing based on the loan s note rate Expressed in terms of up-front points (% of loan amount) Advantages Reflects purely discretionary loan pricing differences across borrowers Reflects the totality of discretionary broker fees as well as borrower choices on how to pay these fees Challenges Materially complete and accurate TBF data is needed TBF data on the separate liens associated with piggyback transactions may not be individually meaningful Does not capture any lender-based discretionary pricing differences 14

16 Annual Percentage Rate: Represents the measure of total loan price disclosed to the borrower per Regulation Z Expressed in terms of annual interest rate Advantages Standardized measure of loan price that is readily available Reflects the loan s annual interest rate as well as any finance charges Captures the effects of both TBF and lender-based discretionary price differences Challenges Differences across borrowers reflect a number of discretionary and nondiscretionary rate and price differences Best suited to detect pricing differences driven either by note rate or by relatively large up-front point / fee differences APR is sensitive to borrower trade-offs between up-front points and annual interest rate and it is difficult to capture these trade-offs through regression analysis 15

17 3) Identify baseline wholesale loan pricing differences between protected class and control group borrowers: While not a crucial part of a monitoring program, this provides a useful starting point for subsequent regression analysis Captures the types and magnitudes of loan pricing differences that exist and, therefore, need to be explained. Loan Price Metric Hispanic Borrowers White Non- Hispanic Borrowers Difference Statistically Significant? Average APR Average TBF 6.89% 6.42% 0.47% Yes 2.89% 2.02% 0.87% Yes 16

18 4) Develop regression models to estimate demographically neutral loan prices for each wholesale loan: A regression model is a statistical technique that allows a lender to estimate a demographically neutral APR or TBF for each loan. The demographically neutral loan price is based on statistically derived relationships between certain transaction characteristics and observed loan-level prices across all borrowers. These loan price estimates are considered demographically neutral since they will be exactly the same for each similarly situated borrower regardless of each borrower s race, ethnicity, or gender. These loan price estimates can be used to identify unexplained price differences that represent the amount of each group s average price that cannot be attributed to legitimate nondiscretionary transaction factors contained in the regression model. Hispanic Borrowers White Non- Hispanic Borrowers Average TBF 2.89% 2.02% Average Demographically Neutral TBF Average Unexplained TBF 2.84% 2.08% 0.05% -0.06% 17

19 Simplified Regression Model Example: Total Broker Fees (TBF) Observed Relationship Between TBF and Loan Amount 7.00% 6.00% 5.00% 4.00% 3.00% 2.00% 1.00% 0.00% $0 $200,000 $400,000 $600,000 Loan Amount We want to explain observed TBF differences across borrowers We identify loan amount as a potential transaction characteristic that may explain some of the TBF differences We observe that there is a pronounced negative relationship between TBF and loan amount for all loans in our analysis Our statistical analysis indicates that this negative relationship is statistically significant that is, it is not due to random chance This negative relationship also makes business sense Regression Equation TBF = x Loan Amount 18

20 Calculating Unexplained Price Differences: The estimated regression equation is applied to each loan to estimate a demographically neutral loan price The average demographically neutral loan price is calculated for both the protected class and control group Hispanic Borrowers White Non- Hispanic Borrowers Difference Average TBF 2.89% 2.02% 0.87% Average Demographically Neutral TBF Average Unexplained TBF 2.84% 2.08% 0.76% 0.05% -0.06% 0.11% We started with a baseline TBF difference of 87 bps The regression analysis was able to explain 76 bps of this baseline difference Hispanic borrowers have a 5 bps unexplained TBF difference and White Non-Hispanic borrowers have a -6 bps unexplained TBF difference Collectively, there is an 11 bps unexplained TBF differential for Hispanic borrowers relative to the control group 19

21 Regression Model Tips: In practice, TBF regression models contain more factors than just Loan Amount. Regression model factors should have statistically significant relationships to TBF and these relationships should be intuitive, consistent with known business practices, and reviewed with legal counsel. Regression models should be developed by staff with training and expertise in statistical modeling. APR regression analysis is similar to the TBF example; however, These models have a significantly larger number of factors since they must also control for non-discretionary pricing differences driven by the lender s rate sheets Lenders may develop separate APR regression models for different product segments Consult with legal counsel before including Originating Broker into the regression model. Based on our recent experience, certain regulatory agencies are objecting to the use of this factor. 20

22 5) Test for statistically significant unexplained loan price differences between protected class and control group borrowers at various geographic levels and by originating broker: We can now use the regression model estimates to finalize our statistical testing Identify cases where the average unexplained price difference for the protected class group is greater than the average unexplained price difference for the control group. For these cases, determine whether there is a statistically significant difference between these group-level average unexplained differences. For those cases where the difference is statistically significant, determine whether the magnitude of the difference exceeds a minimum threshold for corrective action. Hispanic Borrowers White Non- Hispanic Borrowers Difference Statistically Significant? Average TBF 2.89% 2.02% 0.87% Yes Average Demographic ally Neutral TBF Average Unexplained TBF 2.84% 2.08% 0.76% 0.05% -0.06% 0.11% Yes 21

23 Other Testing Considerations: Minimum Disparity Thresholds Given the imperfection of regression modeling, it is common for lenders to establish a minimum disparity threshold to reflect a reasonable margin of error No formal or informal regulatory guidance associated with these thresholds, and the regulatory agencies appear to be pursuing ever smaller disparities Use cautiously and set conservatively Units of Analysis In addition to performing these tests at an aggregate channel level, it is important to drill down into various geographies and brokers to identify fair lending risk areas that may be masked when aggregated together 22

24 Addressing Monitoring Results No statistical disparities: lender will be in a good position to explain raw disparities If analyses reveal unexplained disparities, responses may include: Further research to identify root causes Action to address brokers with pricing differences Counseling Training Fee caps Terminating relationships Borrower refunds 23

25 Questions? Melanie Brody K&L Gates LLP Ric Pace PricewaterhouseCoopers LLP