The refund scoreboard banks grade themselves

The scoreboard

The Consumer Financial Protection Bureau (CFPB) publishes the complaints it forwards to companies, along with the company's own description of how it ended. The measure here is simple: of the checking- and savings-account complaints filed about a bank in a year, what share did the bank close "with monetary relief"? The CFPB defines that as money the company reports giving back that is "objective, measurable, and verifiable."

It is the closest thing the public has to a refund rate. It is also self-reported. The company picks the category; the CFPB says it reviews responses, but we found no published audit of the labels.

What changed

2019 was not a fluke year. Chase's rate held between 30.2% and 31.7% from 2017 to 2019. Then it slid: 22.2% in 2020, 16.6% in 2021, 10.0% in 2023 and 8.2% in 2025 (Exhibit 1). So far this year it is 12.8%. Over the same years Bank of America stayed between 30.4% and 40.7%. Wells Fargo barely moved (12.7% in 2019, 12.9% in 2025), and Citi stayed between 18.9% and 25.6%.

Meanwhile, complaints about Chase checking accounts rose from 3,049 to 8,937, nearly three times as many. Money-back outcomes fell from 921 to 732. More customers complained, and fewer got cash.

The obvious objection is mix: maybe Chase's customers started complaining about things banks don't owe money for, such as scams a customer authorized. The data say no. Chase and Bank of America get almost the same mix of complaints. "Managing an account" is 55.5% of Chase's and 57.5% of Bank of America's. The gap shows up inside every type (Exhibit 2). On disputed or unauthorized charges, Chase reported money back on 10.0% and Bank of America on 42.6%. When an account was closed, it was 1.1% against 40.6%. If Chase had Bank of America's mix of complaints, its rate would still be 8.2%.

There is one place where the two banks look alike: overdraft fees. There, Chase reported money back on 33.0% of complaints and Bank of America on 36.2%. A refunded fee is a clear number on a statement. Where the money is that obvious, the banks behave, or at least report, the same.

The drop is also mostly a checking story. Chase's credit-card complaints ended with money back 13.0% of the time in 2019 and 11.3% in 2025.

The bridge: a known way to fill in the box

The CFPB itself has pointed to how this can happen. In its 2025 annual report, it said that when a company's chosen category didn't match its written response, the company most often chose "closed with explanation" even though the customer had received money or another fix. In other words, the most common labeling error hides relief rather than inflating it.

At Chase, the share of checking complaints closed with an explanation alone rose from 69.3% in 2019 to 88.5% in 2025. That is consistent with tougher decisions. It is equally consistent with a change in how refunds get labeled. On the public evidence, it is a roughly even chance which of the two is the bigger part of the gap. We found no public explanation from Chase.

The strongest case against

Bank of America may simply be the outlier. Its rate is high against every peer, so the real story could be one generous bank, not one stingy one. Fair, but Chase is the bank that moved: it sat level with Bank of America in 2019.

A complaint is not a random sample of customers either. People who complain to a regulator are the angriest ones, and their number depends on social media waves and on how hard a bank pushes people to resolve things in-house first.

And scholars who have studied this database with the CFPB's confidential files find little evidence that differences between firms drive who gets restitution (Haendler and Heimer, 2025). Across thousands of firms that is likely right. But it is a finding about the whole market. It doesn't rule out two of the biggest banks drifting apart within one product.

So the honest claim is narrow. Chase reports far fewer refunds than it used to and than its closest rival does. Whether it pays fewer, we can't tell from outside.

What should change

Research on public scorecards suggests they change behavior most when someone other than the seller sets the score. Los Angeles restaurants improved their hygiene after 1998, when grades set by county inspectors had to be posted in the window (Jin and Leslie, Quarterly Journal of Economics, 2003). The standard review of this literature asks the two questions that matter here: when should disclosure be mandated, and can you trust the party doing the reporting (Dranove and Jin, Journal of Economic Literature, 2010)? Finance researchers already treat complaints in this database as a measure of service quality, in their case for mortgages (Begley and Purnanandam, Journal of Financial Economics, 2021). If the measure is soft, so are their conclusions.

The fix is cheap, because the number already exists. Researchers working with the CFPB's internal data note that it includes the dollar amount of any monetary relief, averaging about $1,470 per successful complaint. The public file only has the box.

Three moves would help:

• The CFPB should publish relief amounts, rounded and anonymized, next to the category, and audit a random sample of each large bank's labels every year.
• Bank boards and the consultants who advise them should stop treating "monetary relief rate" as a customer-outcome measure until it is audited. A rate that falls because of labeling is a compliance risk, not a win.
• Chase could clear this up in a paragraph by saying whether its refund policy or its labeling changed after 2019.

Had Chase matched Bank of America's rate inside each complaint type last year, and assuming complaints of the same type are alike, about 2,550 more of its checking complaints would have ended with money back. That is a Theoria estimate, and it could be wrong in either direction. Until refunds are counted in dollars, a bank's refund rate measures how it fills in a form.