Before there was a Medallion Fund, before Renaissance Technologies existed as a name, there was a piece of mathematics almost nobody outside a narrow academic circle had heard of, and a mathematician who had never set out to solve a market problem. The mathematician's name was Leonard Baum.
Leonard Baum spent his career on a problem that sounds abstract to the
point of uselessness: how do you infer the hidden state of a system when all
you can see is a noisy, partial signal generated by it? Working through the
1960s alongside colleagues like Lloyd Welch, Baum helped develop what's now
called the Hidden Markov Model — a mathematical framework for figuring out
what's probably happening beneath the surface of data, using only the
messy, incomplete observations that surface produces.
It was built for speech recognition and codebreaking, not stock charts.
And that's precisely why it mattered so much when Jim Simons brought Baum into
the fold.
The Insight Nobody in Finance Was Using
Markets don't announce their own state. A stock's price doesn't come
with a label saying "trending" or "mean-reverting" or
"about to break down." All a trader ever sees is the observable data
— price, volume — while the actual underlying condition driving that data stays
hidden. That is exactly the problem Baum had spent his career solving in other
domains.
A Hidden Markov Model doesn't try to predict the future by explaining
the present. It asks a narrower, more disciplined question: given what we can
observe, what hidden state is most statistically likely to be producing this
pattern of observations right now — and what does that imply about what tends
to follow? No story. No causal explanation. Just probability, inferred from
data.
Baum joined the research effort in the early days, well before Medallion
existed in its later form, and brought this exact framework with him. It became
one of the mathematical foundations underneath Renaissance's entire approach to
markets — not the only tool in the toolbox, but one of the earliest, and one
that shaped how everyone who came after him was expected to think about the
problem.
Why This Distinction Still Matters
Most trading approaches — even sophisticated ones — implicitly assume
you can know what "state" the market is in by looking directly at it.
A Hidden Markov approach assumes the opposite: you can never know the state
directly. You can only ever infer it, probabilistically, from what's observable
— and that inference should be revised constantly as new data arrives, never
treated as settled.
That's a genuinely different posture than most trading education
teaches. It's not "identify the trend and trade with it." It's
"hold your read of the current condition loosely, weight it by how strong
the evidence actually is, and update it the moment new evidence arrives" —
closer to how a detective builds a case than how a forecaster builds a
prediction.
What This Means for Us
- Treat
your read of "what the market is doing right now" as a
probability, not a fact — and be willing to revise it as new data comes
in, not once a day but continuously.
- Resist
the pull to explain why the market is in a given state before
acting on the evidence that it is. The state matters more than the story
behind it.
- Notice
how much of your own process assumes you can see the market's condition
directly, versus inferring it cautiously from incomplete signals.
Leonard Baum never worked in finance until Renaissance recruited him,
and he never really thought of himself as a trader. But the way of thinking he
brought with him — infer, don't assume; observe, don't explain — became one of
the quiet load-bearing walls of everything Simons built afterward.