The release of the Paycheck Protection Plan (PPP) loan data was intended to bring transparency to the US’ $517 billion loan program to support small businesses during the coronavirus pandemic. But mistakes from some banks may have caused more transparency than the Small Business Administration (SBA) had planned for.
A Quartz analysis of the data shows that there are at least 842 occasions where the name of a loan applicant appears in a place it shouldn’t. In a few cases that means that the data about an organization’s loan contain the name of a person involved in applying for it. In most cases it’s the result of an applicant’s name finding its way into the field for the city of the recipient’s mailing address.
Of these 842 loans, 792 were for less than $150,000, which should have entitled the recipient to more confidentiality under SBA’s release policies. The data files for those loans don’t even contain a field to name the recipient. The data lists loans over $150,000 as a range rather than a precise figure, and the issue affects loans for between $36.9 million and $54.2 million in total that claim to retain about 6,000 jobs.
This error appears almost exclusively on loans prepared by Bank of America. The bank declined to comment on this story.
In the fine print on the PPP loan application, applicants were warned that their name could be released publicly through records requests, so the release of this information shouldn’t be too concerning from a privacy standpoint. However, the fact that the errors are so heavily skewed toward one bank should give Bank of America’s clients pause. These loans represent just 0.25% of the banks loans, but it was making the error at a rate 337 times higher than JPMorgan, which had 0.0007% of its loans with the name-for-city mistake.
To find these loans we compared the listed city with those that the US Postal Service associates with the zip code on the loan. We then reduced the list to only those with city fields that contained both a name from a list of 98,000 American first names and a name from a list of 162,000 American last names. To eliminate common misspellings we reduced the list further by only looking at possible names that appear less than 10 times in the data. Finally we checked the resulting list by hand to remove uniquely misspelled or misattributed city names.