CBA Comment Letter re CFPB Use of Alternative Data and Modeling Techniques RFI

 

May 19, 2017

 

Ms. Monica Jackson

Office of the Executive Secretary

Bureau of Consumer Financial Protection

1700 G Street, NW

Washington, DC 20552

 

Via electronic submission

 

Re: Response of the Consumer Bankers Association to the Request for Information Regarding Use of Alternative Data and Modeling Techniques in the Credit Process (Docket No. CFPB-2017-0005)

 

Dear Ms. Jackson,

 

The Consumer Bankers Association (“CBA”)[1] appreciates the opportunity to respond to the Consumer Financial Protection Bureau’s (“Bureau”) “Request for Information Regarding Use of Alternative Data and Modeling Techniques in the Credit Process” (the “RFI”).  Through this RFI, the Bureau seeks to learn more about current and future market developments in the alternative data space, including existing and emerging benefits and risks posed to consumers, and how these developments may alter the marketplace and consumer experience.

 

The use of alternative data and modeling techniques (collectively, “alternative approaches”) presents significant potential benefits for consumers by enabling financial institutions to serve more consumers and better assess the creditworthiness of consumers, especially those with thin or no file at consumer reporting agencies.  CBA members have found the use of alternative approaches can improve their predictive power in the credit process, facilitate operational improvements within their institutions and reduce fraud, and can be used by consumers spanning the entire credit spectrum.  Alternative approaches can also lead to more effective pricing and marketing techniques to fulfill consumer needs based on their individual financial situations.  These benefits may help financial institutions better serve those 26 million Americans the CFPB estimates are credit invisible, as well as the 19 million Americans with too thin a credit file to produce a credit score[2] by increasing credit availability through mainstream financial institutions.

However, there are impediments to consumers’ realization of these benefits as the use of alternative approaches do not come without various risks and regulatory concerns.  In particular, CBA members are concerned that regulatory or supervisory actions regarding alternative approaches could inadvertently stifle innovation in this space.  CBA members are conscientious of the various fair lending laws and requirements, and will not use alternative approaches if it would run afoul of these concerns.

 

Given CBA members’ enthusiasm to continue innovating in this space, we respectfully ask the Bureau publish research insights highlighting what the Bureau has gained from the collection of data through its RFI.  We further believe that new regulatory policy on the use of alternative data and modeling techniques is not necessary, and may create barriers between CBA’s members, and the credit invisible consumers they hope to serve.  In particular, we believe that for CBA members, risk management programs created pursuant to model risk management and fair lending guidance issued by prudential regulators provide proper clarity and guardrails for compliant use of alternative data.  We look forward to working with the CFPB as it studies this important space, and we respectfully request that the agency take a transparent approach to the development of its views.

 

  1. Response to Questions 1 & 2: Alternative Data Products and Services Used by CBA Members

 

The CFPB defines “alternative data” for purposes of the RFI to be “any data that are not ‘traditional.’”[3]  The RFI states that “traditional” data are data “in the core credit files of the nationwide consumer reporting agencies” and information consumers provide on applications.[4]  Therefore, it appears that “alternative data” means anything other than tradeline data, public records, inquiries on a credit file, and application information.  The CFPB explains that it uses “alternative” “in a descriptive rather than normative sense and recognize(s) there may not be an easily definable line between traditional and alternative data.”[5]  Nevertheless, we believe this description is too broad, and therefore fails to identify important differences between types of data.

 

By labeling virtually any type of information outside of credit bureau data as “alternative,” the CFPB suggests that this data must somehow be outside of the ordinary,[6] and therefore outside the existing regulatory construct. This may be true for some data, but it is not true for many of the types of data termed “alternative” in the RFI.

 

Many of CBA’s members are already using data and modeling techniques deemed “alternative” under the RFI.  This data includes rent, telecommunications, and other sources of payments and similar data.  Third-party data providers also manage additional forms of alternative data that is yet to be utilized in a widespread manner by CBA members.  Such data includes social network information, internet browser history, and online purchase history.  As such, lumping both of these types of data together under the label of “alternative” in the CFPB’s RFI could lead to confusion and invalid outcomes by the Bureau as it considers this issue.  CBA members remain confident about the use of alternative data such as telecommunications, rental, and internal bank data, and are also optimistic about the potential of other forms of alternative data to expand access to credit; however, CBA members remain wary of using such data as currently analyzed through sound risk management processes. 

 

  1. What Alternative Data CBA Members Use

 

The “alternative data” CBA members use includes internal bank data, and data supplied by third parties that often is not found on a traditional credit report.  More specifically, CBA members use data they have collected on their own customers, including payment behaviors, asset and transaction data, channel preference (e.g., online, in person, mobile), and other measures of customer engagement.  CBA members also have begun utilizing data from third parties, including telecommunications, utility, rental, and insurance payments, as well as change of addresses and phone numbers.  However, the majority of alternative data CBA members use is collected from customers through internal bank systems.  Some CBA members have also found level of educational attainment to be a helpful and predictive source of data, using data solely on what level of education consumers have attained, and not which school they have attended.  CBA members believe each of these sources can act as useful tools to help get a better understanding of a consumer’s creditworthiness, in addition to assisting with fraud detection techniques.

 

Data CBA members have not yet widely adopted include data sourced from consumer’s social media accounts, internet browser history, and online purchases.  Given the fairly new and untested nature of this data, CBA members are yet to widely adopt its use in their various credit decisions.  These data sources also increase the risk of inadvertent violations of the law, and increase institutional risks, while in some cases lacking clearly defined benefits to consumers.

 

  1. What Alternative Modeling Techniques CBA Members Use

 

In its RFI,[7] the Bureau has labeled the following modeling techniques as “alternative,” though CBA and its members may not always consider them as such:

 

  • Decision trees (or sets of decision trees such as “random forests”);
  • Artificial neural networks;
  • Genetic programming;
  • “Boosting” algorithms; and
  • K-nearest neighbors

 

CBA members are already using many of the methods listed above, and most financial institutions, servicers, and credit reporting agencies (collectively, “market participants”) consider these methods mainstream, as opposed to the CFPB’s label of “alternative”.  In addition to the methods listed above, CBA members use traditional regression models and continue to explore other types of modeling techniques, such as machine learning.  CBA members engage in these forms of modeling under the principles of model development, validation, and monitoring that presently exist under model risk governance guidance of prudential bank regulators.[8]  These guidance, in conjunction of applicable law, provide a foundational framework that ensures risks will be well controlled and well managed, and CBA members believe that this framework is both appropriate for current algorithmic practices as well as flexible to changes in practices and supervisory expectations.  CBA members support the use of such modeling techniques that provide safe and sound predictive power to aid decision processes, balanced with the need to provide consumers transparency in how they are being evaluated and the specific principal reasons for any adverse actions.

 

  1. How CBA Members Use Alternative Data

 

CBA members have begun to use the types of alternative data outlined above in conjunction with traditional data sources to drive the credit process.  Alternative data is often used:

 

  • directly within the credit process to segment consumers, such as those CBA members have a pre-existing relationship with, versus those whom they do not, including when CBA members have transaction data with customers using credit products; 
  • to create a second scorecard for consumers when they fail their first;
  • as an input to internally and externally developed custom credit models;
  • for client verification and fraud evaluation to help protect consumers;
  • by third party scores, which are then used as part of the CBA member’s credit process; and
  • to create operational improvements within a bank, such as using the data towards the automation of employment verification.

 

CBA members remain optimistic about the potential benefits use of such data can provide to market participants, and those credit invisible individuals looking to enter the credit stream.

 

  1. Response to Questions 9-12: Benefits Alternative Data and Modeling Techniques Offer to Consumers and Market Participants

 

CBA members are well aware of the various benefits utilizing many alternative approaches can provide both to consumers and market participants.  Most notably, the use of alternative approaches can help to improve the predictive power within the credit process, and help CBA members reach the “credit invisible” population.[9]  The use of alternative approaches can help market participants better assess the creditworthiness of those who are completely new to credit, have limited credit history, and even those with mature credit history.  Alternative approaches also provide promising paths to better fraud prevention and detection, saving consumers and market participants money and aggravation.

 

The use of alternative approaches helps market participants better automate their underwriting process, while managing fraud concerns.  They can also provide operational improvements for CBA member institutions, which can increase speed to market on new products and services.  Additionally, the use of alternative approaches can help increase credit to unbanked and underbanked communities, as the alternative approaches can help with the engagement of communities in need of credit, and assist market participants in providing credit to those communities.  Finally, alternative approaches can lead to more effective pricing and marketing techniques to better fulfill consumer needs.

 

Specifically, CBA members have found that particular sources of alternative data can greatly benefit consumers. For example, rental data has proven to be extremely predictive, and can help get younger consumers who do not own homes, and low-income consumers who are unable to own, enter the credit process.  Additionally, as typically only negative data associated with utility payments is reported (as a missed payment results in a collection action, but typically payments made on time are not reported), CBA members estimate that including utility payment data in the credit decision results in benefits to the consumer 70% of the time.

 

Although CBA members have not widely incorporated the use of more experimental data (browser history, social media profiles, and online purchases) in their credit decisions, many have begun using this data to market to their customers, and to those not currently in the credit process.  While the risks of using such data has largely kept CBA members from using this data in their credit decisions, the potential benefits of using it in marketing procedures is just beginning to be tapped.

 

Alternative approaches can provide many benefits, both to consumers and marketplace participants, by improving the credit processes used by CBA members, and increasing access to credit for many of those consumers currently left out of the credit process.  Given the widespread demographic of consumers alternative approaches can help serve, it is important for the Bureau not to isolate the use of alternative approaches to any one credit demographic.

 

  1. Response to Questions 13 and 15: Risks Alternative Data Presents to Consumers and Market Participants

 

While CBA members understand the many benefits alternative data can provide, its use does not come without potential risks both for consumers and market participants.  As much as CBA members wish to continue using alternative data throughout the credit process, uncertainty in how alternative data may raise various safety and soundness concerns, privacy issues, and regulatory requirements remains our largest concern. As such, we ask the Bureau to consider these issues as they collect this data, and publish insights gained by the collection to the public.

 

  1. Risks to Consumers

 

While CBA members are increasingly using alternative approaches at all stages of the credit process, concerns about the various risks it poses will raise some challenges.  In March, the CFPB released a report focusing on the various issues consumers have in correcting and adjusting errors on their credit reports.[10]  The report details how often, many consumers are unaware of what data goes in to their credit reports, and thus left with no way to dispute the errors they find.  Increasing the sources of data that go in to credit reports may further muddy this process for consumers, and leave them more lost when attempting to correct or clarify information on their reports.  For this reason, market participants should do their best to treat alternative data with the same protections and error reporting procedures as standard credit bureau data when applicable.

 

Additionally, certain sets of alternative data may pose significant challenges for those consumers looking to enter the credit process.  The use of more experimental forms of data (browser history, social media profiles, and online purchases) may provide benefits to consumers, but carries unknown risks.  CBA members are engaging existing risk management processes to determine whether and how such data can be used in credit processes through the employment of proper risk mitigation.  However, the risk of inadvertent fair lending violations – discussed below – particularly through the use of more experimental data that have not been adequately tested, needs to be taken into consideration as the industry begins to use these new approaches to expand access.

 

  1. Risks to Market Participants

 

Many of the risks facing consumers also face the market participants that serve them.  CBA members remain hesitant to tap in to some of the more experimental sources of data listed above due to the risk of privacy violations.  Inadvertently violating fair lending laws with the use of more experimental sources of alternative data is a particular concern for all parties involved.  For instance, a financial institution using browser history or social media in credit decisions needs to be aware of its potential impact on protected classes that may have less access to these capabilities. 

 

Alternative data also presents various Fair Credit Reporting Act (“FCRA”) concerns, as much of the data provided may be by those who have not previously adhered to FCRA standards.  For example, many of the landlords who operate in the rental market have never supplied data on their tenants for this purpose, and may be unfamiliar or unwilling to comply with the FCRA.  Telecommunications companies who are not currently furnishing this data may have similar concerns.  These data providers would greatly benefit from the Bureau publicly stating how they will treat data furnishers attempting to provide this data, especially if the Bureau indicates some leeway or flexibility will be given to those furnishers making a good faith effort to provide the data to increase credit access.

 

In general, alternative data sources should meet rigorous enterprise data management standards, and comply with all existing regulatory requirements, especially those listed above, or risk violating Unfair, Deceptive or Abusive Acts or Practices (UDAAP) or other laws.  Further, alternative data used in the credit process should adhere to safety and soundness principles, including:

 

  • complying with laws, rules, and regulations;
  • ensuring the data is effective at demonstrating a consumer’s creditworthiness and willingness to repay;
  • confirming the data is accurate; and
  • ensuring the data is transparent to the client

 

Failure of market participants to provide existing protections and safeguards could limit consumer access to credit, frustrate consumers, and expose them to financial stress.  Market participants must continue to use a disciplined approach to help consumers establish credit and have the ability to repay.  Further, alternative modeling techniques, as described above, need to adhere to the existing regulatory requirements and provide transparency on the principal reasons for denial. 

 

To that end, it is important to recognize there is a tangible risk some market participants that are not supervised by prudential regulators may fail to provide existing protections and safeguards to consumers.  This could limit consumer access to credit, frustrate consumers, and expose them to financial stress.  For instance, alternative data presents various FCRA questions, and these issues may not be adequately understood by entities who have not typically adhered to FCRA standards, as outlined above.  However, CBA members believe that all market participants, irrespective of type or nature, must apply existing legal protections to ensure that consumers have the same rights, regardless of the type of institution they are facing.  In light of the fact that the consumer rights actually provided by institutions may differ drastically by the type of credit being solicited, or the type of institution providing the credit, we respectfully implore the Bureau to ensure consumers are not in effect penalized for choosing to use one type of credit or credit provider over another, even if the choice is truly voluntary.

 

Additionally, CBA members often use vendors that use alternative data for marketing purposes, but sometimes the vendors do not always know the inputs or sources of their own data.  Without more detailed information on this alternative data, CBA members will not similarly use the data in their credit decisions, as it may be too complex to fully understand the risks it carries, and it may be difficult to provide reasons for denial.  Finally, the Bureau should remain cognizant that alternative data may not be historically available for performance evaluation through the most recent economic cycle, so that data should be carefully considered when integrated into the credit process.  In addition to allowing flexibility to CBA members as they attempt to innovate in this space, publically disclosing how the Bureau will treat good faith, commercially reasonable efforts to use alternative approaches moving forward will greatly benefit both market participants and consumers looking to enter the credit stream.

 

**********

 

CBA remains enthusiastic about the potential for alternative data and modeling techniques to increase consumer access to credit for individuals across the entire credit spectrum, yet remains cognizant of the various risks these new sets of data inherently provide.  While CBA members will continue to comply with the existing rules and regulations as they look to expand credit possibilities, research insights from the Bureau, highlighting the findings of the RFI would be greatly appreciated.  In addition to allowing flexibility to CBA members and other market participants as they attempt to innovate in this space, publically disclosing how the Bureau will treat good faith, commercially reasonable efforts to use alternative approaches moving forward will greatly benefit both market participants and consumers looking to enter the credit stream.  CBA does not believe that new regulatory policy is necessary in this space, as it may create barriers between CBA’s members and the consumers they hope to serve.

 

CBA greatly appreciates the opportunity to respond to the Bureau’s RFI in this developing and vital space. Should you need further information, please do not hesitate to contact the undersigned directly.

 

Sincerely,

 

Stephen Congdon

Regulatory Counsel

Consumer Bankers Association

scongdon@consumerbankers.com

 

 

[1] The Consumer Bankers Association is the only national trade focused exclusively on retail banking. Established in 1919, the association is now a leading voice in the banking industry and Washington, representing members who employ nearly two million Americans, extend roughly $3 trillion in consumer loans, and provide $270 billion in small business loans.

[2] See, Consumer Financial Protection Bureau, Data Point: Credit Invisibles, at 6, available at  https://www.consumerfinance.gov/about-us/newsroom/cfpb-explores-impact-a....

[3] Request for Information Regarding Use of Alternative Data and Modeling Techniques in the Credit Process, Docket No. CFPB-2017-0005 (Feb. 14, 2017), at 5.

[4] Id.

[5] Id.

[6] The issue is somewhat less pronounced with respect to the use of the term “alternative modeling” which is vaguely defined to include modeling techniques not “traditionally used in automated credit processes.”

[7] See, Request for Information Regarding Use of Alternative Data and Modeling Techniques in the Credit Process.

[8] See Federal Reserve Board SR Letter 11-7 (“Guidance on Model Risk Management”) (April 4, 2011); Office of the Comptroller of the Currency (OCC) Bulletin 1997-24 (“Credit Scoring Models”) (May 20, 1997); OCC Bulletin 2000-16 (“Risk Modeling”) (May 30, 2000); OCC Bulletin 2011-12 (“Sound Practices for Model Risk Management”) (April 4, 2011); Federal Deposit Insurance Corporation (FDIC) Supervisory Insights (“Model Governance”) (last updated December 5, 2005); FDIC Supervisory Insights (“Fair Lending Implications of Credit Scoring Systems”) (last updated April 11, 2013).

[9] See, Consumer Financial Protection Bureau, Data Point: Credit Invisibles, at 6.

[10] See, e.g., Consumer Financial Protection Bureau, Supervisory Highlights Consumer Reporting Special Edition, available at, https://www.consumerfinance.gov/about-us/newsroom/cfpb-oversight-uncover....