How to check a bank’s financial health before you park a round: capital, uninsured deposits and the date on every number

I started with the capital ratios, because they are the numbers people quote when they want to sound careful about a bank, the ones that turn up in board packs and in the slide a finance hire presents in their first month, and they are also the easiest numbers to find. So I pulled the last year-end filing Silicon Valley Bank ever made, for 31 December 2022, from the FDIC's public data, and read the capital lines first. It seemed like the sensible place to start.

They pass, and they pass comfortably. The leverage ratio was 7.96 per cent and the common equity tier 1 ratio 15.26 per cent, against well capitalized thresholds of 5 and 6.5 per cent. The FDIC's own record for the bank ends on 10 March 2023, 10 weeks later.

The number that would have told a founder something sat 2 lines further down. Of the bank's domestic deposits, the filing estimated 151.6 billion dollars as uninsured against 9.99 billion insured. That is about 94 per cent of the money above the insurance limit. Almost every depositor was above the line. Nearly all of them, in fact.

So this piece is about how to check a bank's financial health before you park a round there, and the honest version starts with an admission. I was wrong about the order of reading, and the order matters more than any single ratio.

That is not hindsight dressed up as insight. I hope it is not, anyway. The uninsured figure was in the same filing, in the same public dataset, on the same row, and it had sat above 93 per cent in every year-end filing since 2020. Anybody could have read it, and the fact that I did not start there is exactly the habit this piece is trying to break. The trouble is that almost nobody reads a bank's filing before opening an account, and the few who do tend to start where I started.

Where the numbers come from

Every insured bank files a quarterly call report, and the FDIC publishes the figures through BankFind and its public data service. The latest complete quarter is 30 June 2026. Each bank's record carries total assets, total deposits, estimated insured and uninsured deposits, the leverage ratio, the common equity tier 1 ratio and return on assets, all in the same row.

I went looking for the uninsured field in the documentation, because it is not called that. Its full title is “ESTIMATED UNINSURED DEPOSITS IN DOMESTIC OFFICES AND IN INSURED BRANCHES IN US TERRITORIES AND POSSESSIONS”, in capitals, which tells you something about who the dataset was built for. It was not built for a founder with a spreadsheet and a board meeting on Thursday. It works anyway, once you find things. The data is free to download.

For banks the FDIC supervises, the definitions of well capitalized sit in 12 CFR 324.403. A total risk-based capital ratio of 10.0 per cent or more, tier 1 of 8.0 per cent, common equity tier 1 of 6.5 per cent and a leverage ratio of 5.0 per cent. There is a fifth condition that people skip: the institution must not be subject to any written agreement, order, capital directive or prompt corrective action directive.

That fifth line is the one I like best. It means a bank can clear every ratio and still fall out of the top category because a supervisor has put something in writing, which is the kind of fact a treasury policy should care about more than a basis point of capital.

3 banks at 30 June 2026

I pulled the same fields for 3 banks that a startup might plausibly meet, one large and 2 smaller, all from the FDIC data for the quarter to 30 June 2026. The uninsured share is my own arithmetic: estimated uninsured domestic deposits divided by insured plus uninsured.

This surprised me more than it should. The ranking looks backwards at first. Read across the rows and the capital column points the wrong way. The smallest bank has the highest ratios of the 3, 12.88 per cent leverage and 17.72 per cent common equity tier 1, and it is also the only one losing money, with a return on assets of minus 0.14 per cent and a net loss of 865 thousand dollars for the period. It is the bank that received a Federal Reserve enforcement action in June 2024.

I find it hard to look at that row without wanting to draw a conclusion, and I am going to resist. None of that means the bank is unsafe, and I am not saying it is. It means capital ratios are a floor, not a verdict. A bank can hold plenty of capital because it is shrinking, or because a supervisor asked it to, and the ratio will not tell you which. A ratio cannot see the reason behind it.

The uninsured share tells you something different: how many other depositors are in your position. At 94 per cent, SVB's depositors had every reason to leave on the same morning. At 27.6 or 37.2 per cent the crowd is smaller. I would not put a hard line in a policy on the strength of 4 rows. I would want the number in front of me.

The date on the number

Here is the second thing I tripped on. The public directory Bank Index keeps a card on First Citizens, and its strength line reads a common equity tier 1 ratio of 11.15 per cent at year-end 2025, with ratings of BBB+ from S&P and Baa2 from Moody's. The FDIC figure for the bank at 30 June 2026 is 12.41 per cent.

Both of those numbers can be right. One is a year-end figure and, as far as I can tell, reported for the group. I cannot tell from the card which of the 2 entities it used. The other is the bank's own call report 6 months later. I had assumed they measured the same thing and put them side by side, and they do not quite. My guess is that most treasury spreadsheets contain at least 1 pair like this, a group figure next to a bank figure, and nobody notices because both look plausible.

So every ratio in a treasury memo should carry 3 labels: which legal entity, which report date and which source. Without them, 2 honest numbers look like a contradiction and a stale one looks current.

What a directory is good for

I still like the card. It is the fastest way I found to get the story a call report cannot tell. It says First Citizens bought the deposits and loans of the failed SVB from the FDIC in March 2023, runs SVB as a division with 39 offices in 15 states, and did not acquire SVB's UK subsidiary, which the Bank of England sold to HSBC UK. That is 3 facts a founder needs before choosing the bank, and none of them is in the ratios. Ratings are not in the call report either.

Finding the right card takes a moment. US banks filed under F run to 2 pages, and 5 of the names on the first one begin with First Citizens, most of them, judging by their names and towns, separate local banks. The legal name to look for is First-Citizens Bank & Trust Company, with a hyphen.

For a company that holds cash outside the US as well, the directory's banks in 141 countries are a quick way to see who else a local bank belongs to. It listed 10,678 banks on 23 September 2026. The limits are the usual ones for data read from registers in batches: a card can trail a merger, and a card with only a register line tells you the licence and little else.

What I would put in a treasury policy

Pull the call report fields for every bank that holds more than a month of burn, once a quarter, about 55 days after the quarter ends, which is when the FDIC says its quarterly profile of the industry comes out. Record the entity, the certificate number, the report date and 5 figures: assets, leverage ratio, common equity tier 1 ratio, uninsured share and return on assets.

Dates matter more than decimals here. Write the dates down every time. Check the regulator's enforcement page for each bank in the same sitting. The fifth condition in 12 CFR 324.403 exists because an order can matter more than a ratio.

Then do the part that feels unnecessary. Ask the bank what it thinks of its own deposit base, and listen to how it answers rather than to the answer. A bank that talks about its depositors as a community of similar companies is describing concentration, whether it means to or not.

Read the uninsured share against your own balance. If your deposit alone would move the bank's figure, you are a large depositor in a small bank, and that is a conversation to have with the bank rather than with a spreadsheet.

Keep the whole thing short, and boring on purpose. A page per bank is plenty, and the discipline is in the dates rather than in the length. Use a directory for the story around the numbers, the ownership, the acquisitions and the ratings, and write the date beside everything you copy from it.

None of this is investment advice, and it does not replace a sweep or a reciprocal network if you hold more than the insured amount. It is the reading you do before deciding how much of those you need.

An aside about 7.96

I keep thinking about SVB's leverage ratio. 7.96 per cent is not a warning sign by any rule I read. It is more than 1.5 times the 5 per cent line, and a treasury memo written in January 2023 that stopped at the capital section would have filed the bank as comfortable.

Anyway, back to the practical point. It is a good reminder that a ratio is an answer to a question somebody else chose.

What I could not settle

I do not know how the uninsured estimate treats money that fintech partners hold for their customers in pooled accounts. If those balances are counted as uninsured, a partner bank will look riskier in this column than its depositors really are. I could not find the treatment spelled out on the 3 pages I read.

I also cannot tell you what uninsured share is too high. I have not found a published threshold. 94 per cent clearly was. That gap in the data bothers me. Nobody publishes a line between 40 and 94, and I would rather leave the gap open than invent one, and I suspect anybody who offers you a precise threshold has chosen it for a reason that has more to do with a sales conversation than with the data.