Robust small business lending data is vital -- but if it isn't easy for the public to use, little will change.
In early 2023, the Consumer Financial Protection Bureau (CFPB) plans to publish its final regulation implementing the collection and disclosure of small business and farm loan data, as authorized by Section 1071 of the Dodd-Frank Wall Street Reform and Consumer Protection Act of 2010. This vital public disclosure law promised to inform the public about whether or not women-owned and minority-owned small businesses and farms are receiving an equitable share of loans, in much the same way that Home Mortgage Disclosure Act (HMDA) data illuminates the housing finance market.
Advocates have been pushing for this data for decades and are excited about the final rule’s imminent release. However, in the euphoria accompanying a final rule, we must remember that the rule is only as good as its implementation. If the rule requires lenders to submit a robust amount of data to the CFPB but the agency does not publicly disseminate the data in a manner that is easy for the public to use, the data will lose a good deal of its power to help stakeholders increase access to credit for underserved small businesses and farms.
The CFPB’s dissemination of HMDA data is a cautionary tale that must not be repeated in the dissemination of the Section 1071 data. By examining how the agency disseminates the HMDA data in tables, charts and maps, we can highlight promising features of the agency’s dissemination methods but also warn against aspects that needlessly limit how the public can use the data. This examination can help regulators, advocates and lenders improve performance.
Current limitations of CFPB’s data dissemination
In order to illustrate the promise and limitations of the CFPB’s current dissemination methods, let’s suppose that a community group in Montgomery County, Maryland, meets with a large bank based in the Southeast part of the United States about its fair lending record. This example is based on actual bank data but the name of the bank is being withheld because the focus here is not the bank’s performance but the usefulness of the public data.
The group has only a couple of hours to prepare since the bank’s officials just called the meeting for later that afternoon. Ideally, the group uses the CFPB’s website to pull a few revealing statistics about the bank’s performance and discuss weaknesses and strengths about the performance with the bank. If the community group hits this link, it can summarize or download HMDA data for all lenders or for a particular bank in a geographical area. In a box called “Select a Filter” the group can choose up to two variables for which to produce data outputs in a table. For example, if the group selects “race” and “loan purpose” and in the “loan purpose” option selects home purchase, a table is produced that reports the number of applications for home purchase loans from each racial category. In this example, the bank received just 49 applications from African Americans out of 1,680 total applications or just 2.9% of all applications in 2021. In contrast, all lenders in the county received 10.9% of their applications for home purchase loans from African Americans.
This example certainly provides evidence for the community group that this bank needs to improve its performance with African Americans, but the evidence is limited. The bank representative could counter that the community group considered all applications but did not focus on actual loans or originations that could tell a different story. Moreover, the community group would not have information on conventional home purchase lending compared to loans guaranteed by the Federal Home Loan Administration (FHA). The group could be caught flat footed in any nuanced conversation and lack information to either confirm or counter the bank’s assertions.
But this would not be the fault of the community group; the shortcoming is the CFPB’s query engine. The query engine only allows comparisons between two variables at a time, which prevents a community group from telling the query engine to focus on subsets of activity such as conventional home purchase loans only. The CFPB could reply that the website allows for relatively easy downloads of data, especially on a county level. However, a single bank’s data could easily contain thousands of rows of data, which would defeat most novice and intermediate level users from quickly sorting the data in the few hours before this impromptu meeting with the bank. One should not need to be a master of Excel or other sophisticated software to obtain tables that meaningfully display lending activity.
A few years ago, the community organization could have gone to another website maintained by the Federal Financial Institutions Examination Council (FFIEC) to produce the tables necessary to answer the bank representative’s questions. The FFIEC’s tables were discontinued after the CFPB took over dissemination of HMDA data. NCRC had asked the CFPB to either preserve some core FFIEC tables or to improve its query engine to produce a variety of tables like the FFIEC’s. The CFPB responded by making some improvements to downloading the raw HMDA data but opted against bolstering the capability of users to select variable choices in the query engine. The agency did not have compelling explanations for that choice, but hinted at resource limitations.
CFPB’s online HMDA tool compared to NCRC’s
The limited nature of the CFPB’s data dissemination prompted NCRC to create an online tool for our member organizations. While our tool has some limitations, it is considerably more robust than the CFPB’s query engine. For example, it allows a user to focus on originations only or consider denials and other lender decisions on applications. Not only does it produce tables with rows and columns but it can produce vivid charts that rapidly compare lenders’ performance against each other for different loan purposes such as home purchase or refinance in serving low- and moderate-income borrowers or people of color. A user can tell quickly that lenders in the lower left of a chart are performing poorly in terms of the percentage of their loans to underserved borrowers, or that lenders in the upper right of the chart are performing well. The fair lending tool uses three years’ worth of data in order to reflect performance over a considerable time period.
While NCRC appreciates the CFPB’s meaningful steps to strengthen fair lending oversight, the ongoing lack of a robust data tool for data novices or intermediate users remains an unfortunate gap with concrete repercussions. As the example of NCRC’s Fair Lending Tool shows, an agency as resourceful as the CFPB surely has the capacity to create such a tool. It would not take too much more effort to fix the deficiencies in the CFPB’s data query engine. The engine could be altered readily to allow a user to conduct calculations using more than two variables.
Another limitation is the absence of data on the newer aspects of HMDA data that were mandated by Congress in the Dodd-Frank Act. For example, NCRC’s Fair Lending Tool allows a user to calculate median interest rates and fees by loan type for lenders. This is not possible on the CFPB’s website. HMDA data now has a robust range of variables which is challenging for any public website but the CFPB should create an advisory group of data users and conduct periodic surveys to figure out which additional data tables would be useful and valuable. The CFPB also provides an interactive map for users in order to determine on a state or county level the extent to which states or counties are being served by lending institutions that report HMDA data. The map allows users to look at loans per one thousand people, denials per thousand people and other indicators of credit access. It also shows loans by race or ethnicity – but does not allow users to know whether the map displays loans to a specific racial or ethnic group per 1,000 people of that group, or per 1,000 people of the overall population. It would be more helpful if the comparator was per 1,000 people of that racial or ethnic group.
The CFPB map is commendable and shows at a glance how a user’s county or state compares against others in terms of credit access. It could be improved by providing a choice to select a specific lender or group of lenders – but overall, it is an impressive example of an agency retrofitting an existing system to accommodate new complexities. As the agency builds out the brand-new Section 1071 public data tools, however, it has a chance to design something clean and robust from scratch.
Lessons for Section 1071 data
After the CFPB releases its Section 1071 final rule, it should then develop a plan for disseminating the data. We hope that this blog provides guidelines about how to avoid the pitfalls with its current HMDA data dissemination and how it can improve its HMDA dissemination as well as start with a robust dissemination of Section 1071 data. Specifically, the data dissemination tool should include a query engine that is nimbler than the current HMDA one and can allow a user to use more combinations and choices. For example, the query engine should separately assess the number of credit card and non-credit card loans and applications for businesses and farms by race and gender. In addition, an interactive map can help CRA and fair lending analyses identify geographical areas that are well served and those that need more access to credit for small businesses or farms.
Lastly, a user should understand clearly what the strengths and limitations of any publicly available data tool are. The CFPB HMDA tool currently lacks videos, webinars and guide books that clearly describe what variables are available in the data and how analysts can use the data in clear and nuanced ways as discussed above in the case example. The Section 1071 data dissemination should start right out of the gate with these user tools.
The real-world impact of a data law depends on how well an agency publicly disseminates the data and makes it understandable. The CFPB needs to make improvements in order to realize the full potential of the data for identifying and addressing disparities by race and gender.
Josh Silver is a Senior Fellow at NCRC.
Photo by CafeCredit via Flickr.