Public hedge funds measure risk down to the decimal.
Private equity still measures it with adjectives.
Sit in on a hedge fund risk meeting and you get factor exposures, value-at-risk, and a fresh stress number on every position, rerun overnight.
Now sit in on a PE update and you get “great management team” and “diversified across 6 industries.”
I have taught hedge fund desks how to use AI for stats work, and I lecture in Harvard’s stats department. So when I watch how many private equity funds handles risk, I wince a little.
It got me wondering....
Could you take the statistical toolkit a pod shop run and point it at a private book?
You can.... and I'll show you our experiment this week.
Let me show you on a real one.
As always, AI outputs are only as good as your data. For this run, I used Gladstone Investment ($GAIN).
It is a publicly traded BDC, so its whole portfolio is sitting right there in the SEC filings. It behaves like a buyout shop, holding debt and equity in 30 private businesses. The point is, the data is rich here so if you are looking to try this out with your own fund, make sure you have the right data.
We're going to answer a few things with our demo below:
How do you run a factor model on a private book?
How truly diversified is this portfolio?
What are the true underlying bets?
Where are hidden risks in this portfolio?
....and most importantly "Can we run this without a data science background?"
Here’s the plan we're laying out to Claude Code:
Pull every holding out of the SEC filing
Match each one to how it trades in public markets
Blend them into a public-market mirror of the book
Run the same risk analysis a hedge fund would
The Prompt
I ran the whole thing in Claude Code.
It all rests on one trick in your prompt logic. A private company does not have a price, but it does have a personality.
Its debt acts like high-yield credit. Its equity acts like a small, leveraged version of the public companies in its sector.
So you find each holding a public twin, size them by how big the position is, and blend them into one portfolio. This gives you a proxy for the data needed.
Once you have that stand-in, every tool on a hedge fund desk suddenly works on it, from factor models to value-at-risk to stress tests.
First, on diversification and what the 'underlying bet' is...
GAIN holds 30 companies across 16 industries.
Run the math and those 30 names crunch down to about 3 big underlying bets: (1) Long cyclicals.... which is 73% of the portfolio (!) (2) Energy (3) Healthcare
Bet 1 — Cyclicals (73% of all the movement); Bet 2 — An energy / commodity swing (9%); Bet 3 — A healthcare wobble (5%)
And one company, SFEG Holdings, is 39 percent of the entire net asset value. The top five names are 44 percent of the book.
Sector + individual holdings concentrations
Next I ran a factor model on it, the same Fama-French setup that sits under every quant desk.
The “private equity” book from a factor perspective is a levered small-cap value trade, with a market beta of 0.88 and clear tilts toward small and cheap companies.
One last one: Can we model stress tests on portfolio volatility across historical periods?
Once the public-market mirror exists, you can drop the portfolio into any crash in history and watch how it would have moved.
A few things jump out.
In the COVID crash, the book fell harder than the market, down 38.5 percent against the S&P's 33.7. That is the leverage and the small-cap tilt doing exactly what you would expect in a broad sell-off.
The 2022 rate shock went the other way. The book held up better than the S&P, down 20.9 versus 24.5, because it leans toward value and energy rather than the expensive tech that got hit hardest that year.
And when Silicon Valley Bank failed in 2023, the portfolio barely moved, down 3.3 percent while the market finished flat. That one was a bank problem, and this book has almost no bank exposure.
The Big Picture
Private equity has always described its risk.
"We're diversified. We back great teams. We like the space."
Now you can measure it.
The same numbers a public desk lives by: what you're actually betting on, how concentrated you really are, and what breaks you when the market turns.
And it gets more powerful the more positions you hold.
If you allocate capital or run a fund of funds, this is the whole point. You aren't holding one private book, you're holding a dozen of them, with public positions stacked on top. Run all of it through the same look-through and you finally see what no one can see today: what your entire portfolio, public and private together, is really betting on.
You might find your twelve managers are twelve versions of the same cyclical trade. You might find your private book and your public book are leaning on the exact same factor.
The big point is ..... with AI, you can be so much more imaginative on the stats and "quant" side because the manual labor of data pre-processing and having to understand coding is now negligible.
Personal
Everything in today's issue ran on Claude Code.
If you want to learn live the techniques behind this build....I'm teaching all of it live.
We're running Cohort 2 of Claude Code for Investors, our 5-day online bootcamp, July 13-17, at 7pm London / 2pm New York / 11am San Francisco.
Last bootcamp had investment professionals from HIG Capital, OMERS, Goldman, and more.