Who is most likely to buy a microcap next

01The test

Krypton AI built a ranking of which of about 5,700 institutions would buy a given company next, from Form 13F data. We tested it on 6,000 real purchases it had never seen, across the four quarters to March 2026. For each purchase we asked one question: was the fund that actually bought in the ranking's top five?

The baseline matters more than the ranking. Simply naming the funds that open the most new positions every quarter puts a real buyer in the top five 26.6% of the time. A few managers buy hundreds of small positions a quarter, so a list that looks smart can be nothing more than those names. Any ranking of likely buyers should be compared with that list, not with a random draw.

02Former holders come first

The strongest single signal was a fund that had owned the stock before and sold it. It was worth about four times every other signal combined. How long the fund held it, how many times, and how large the position got sharpened the ranking further. A fund that held a full position for two years is a different prospect from one that held a starter position for a quarter.

It makes sense once you see it. A fund that held a company for two years already knows the management, the filings and the story. It sold for a reason. When the reason changes, it is the cheapest possible buyer to bring back, because it does not need to learn the company from the start.

The next signals were the company's neighbourhood, meaning the other stocks its holders own, and what the company is.

03What did not work

We expected funds that buy a company's stock offerings to be likely buyers in the open market. Across 205,882 prospectuses, it added nothing. A fund that buys a placement already shows up as a holder, so the offering says nothing new. It was the idea we most expected to work, which is why it is worth saying.

04How much to trust a ranking

On all companies, the ranking put a real buyer in its top five 45.5% of the time. On the smallest companies, with under $25 million of institutional money, that fell to 37.0%. In quarters where only one institution bought, it was 24.7%. Any honest list of likely buyers should say which of these numbers applies.

A score is a ranking, not a probability. The top of the list is where to start calling, not a forecast of who will buy.

05Reading 13F data without the traps

Every list of former holders starts from Form 13F, and the data has traps that produce numbers that look fine. The value column changed units: it was in thousands of dollars until the filings for the fourth quarter of 2022 and in whole dollars after. Even that rule is broken by 3.9% of filings, which is enough to make one small company look like it has tens of billions of dollars of institutional owners.

The fix is to ignore the value column. Take the reported share count and multiply it by the quarter-end price. Every filer holds the same stock at the same price, so a filing whose implied price disagrees with everyone else's is the one that is wrong. After that correction, 97.7% of positions agreed with the consensus price to within about a quarter.

06What a company can do with this

• Pull every Form 13F that has ever held your CUSIP and list the funds that sold out. Keep the ones that still run money by picking stocks.
• Take index funds and quantitative funds off the list. They do not take meetings.
• Measure positions from reported share counts times the quarter-end price, never from the value column.
• Rank what is left by how long each fund held you and how large the position got.
• Tell each former holder what changed since they sold. That is the one piece of news they have not heard.