What 500 institutional portfolios told us this quarter

The quarter in one number

Every quarter we point Souppe at a large set of real institutional portfolios and ask one question. If a manager owned nothing but the book they disclosed, how well built is it, and can the model make it sturdier without discarding what the manager believes?

For the fourth quarter of 2025 we ran 500 sole-managed institutional portfolios through the model. Together they hold about $1.34 trillion in disclosed US equity. The most useful result is a single line. Souppe's first suggested change lifted the overall structure score for 88% of them, and it raised downside protection for 91%.

That is the headline. The rest of this post explains what was weak, what the model changed, and why no two portfolios got the same answer.

Most books shared the same blind spot

When we line up all 500 books and ask which single dimension scores lowest in each one, the answers cluster around a few places.

• Systematic Coverage was the weakest point in 44% of books (221 of 500). A low score here means a large share of the book's movement comes from individual company surprises the broad model cannot see in advance, rather than from the broad drivers it can. Risk the model can see is risk you can plan around. Risk it cannot see is the kind that surprises you.
• Sector Diversity was weakest in 25% (123 books). The book leans on too few parts of the market.
• Downside Protection was weakest in 19% (96 books). The book is built to fall harder than the market when the market falls.
• The remaining 12% split across Diversification Quality, Position Diversity and a thin tail of other dimensions.

Weakness is not random. In about 69% of books the lowest score sat in one of two places, Systematic Coverage or Sector Diversity, and it was rarely the dimension a brokerage screen puts in front of you.

The fix held across the sample

Across the 500 books the average overall score rose from 68 to 77, a gain of about nine points, and 88% of books improved. Souppe gets there by proposing a small set of changes aimed at each book's weakest point, then re-scoring the book as if those changes were made. The clearest movement was exactly where most books were thin.

• Downside protection rose from an average of 50 to 62, and it improved for 91% of books. By the usual statistical yardstick this was a large effect, not a fluke of rounding.
• The tendency to fall with the market in a downturn dropped from 1.01 to 0.74 on average. In plain terms, the typical book went from falling about as hard as the market to falling noticeably less.

Most of the gain arrived early. The first suggested change delivered about a third of the total improvement, the first three changes delivered most of it, and the fifth change added little on top. A manager does not need to rebuild the book to move the number. A few targeted changes carry the weight.

Of the 13 things we measured before and after, 12 moved by a statistically meaningful amount.

No two portfolios got the same answer

This is the result we care about most.

Across 500 books, the single top suggestion was 276 different names. Look at the top five suggestions per book and the count rises to 389 different names. The average overlap between any two managers' suggested sets was close to zero.

The model is not handing every investor the same shopping list. It reads the actual book in front of it and answers the weakness it actually finds. A concentrated growth book and a sprawling value book get different advice, because they have different problems. With 276 different names landing as the single top pick, the model is reading structure rather than copying a model portfolio.

It worked across very different managers

The benefit did not depend on the kind of manager.

• Hedge fund managers improved in every case we tested, with an average gain of nine points.
• Family offices that also run funds improved the most, an average of 16 points, because their disclosed books started with the most room to improve.
• Private equity and venture managers gained about 13 points on average.
• Registered advisers and family offices gained about eight points on average, from a higher starting base.

It is worth being honest about the other side. About one book in eight did not improve, and a few large managers came out a point or two lower after the suggested change. That is by design. The model will not chase a higher number with changes it cannot justify on the structure of the book.

Reading the sample

A few things to keep in mind, the same way we frame every case study.

• A 13F shows long US equity only. It does not show cash, shorts, derivatives, private holdings or most non-US positions. The read describes the disclosed book, not the manager's full picture.
• The before and after scores are the model's structural estimate of the book if the suggested changes were made. They describe how the book is built, not a promise about future return.
• The sample is sole-managed books above $0.5 billion with focused position counts, so it leans toward higher-conviction managers rather than broad index-like holdings.

Reading the Q4 2025 filings

The sample

Books analysed 500

Disclosed equity $1.34T

• Composite score improved 88%
• Average composite gain +9.4
• Measures that moved significantly 12

Where the fix landed

Downside protection improved 91%

Crash sensitivity improved 92%

• Downside protection, average after 62 / 100
• Market-drop sensitivity, average after 0.74

How specific the advice was

Different top suggestions 276

Books compared 500

• Different names across the top five 389
• Measures tested 13

Every number sourced from the Q4 2025 institutional case study sample. Derived from SEC 13F filings.

Putting it together

Across 500 institutional books holding $1.34 trillion, the visible holdings were usually fine and the hidden structure usually carried one weak point that 88% of managers could improve with a few targeted changes. Counting holdings says a portfolio is a list of good decisions. Reading the structure shows whether those decisions hold together when the market turns.