Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:00
I always used to misspell Renaissance as I was
0:02
typing it out at R-E-N and then I would
0:04
sort of like not really know what came from
0:07
there, but I learned a mnemonic to make sure
0:09
I get it right. Oh, I
0:11
thought you were going to say you've typed it so
0:13
many times now over the past month. Well, there's that
0:15
too, but you ready for this? You
0:18
can't spell Renaissance without A-I. Oh,
0:21
oh, oh. Touche,
0:24
touche. All
0:27
right, let's do it. Welcome
0:45
to Season 14, Episode 3 of
0:47
Acquired, the podcast about great companies
0:49
and the stories and playbooks behind
0:51
them. I'm Ben Gilbert. I'm David
0:53
Rosenthal. And we are your hosts.
0:56
They say, David, that as an investor,
0:58
you can't beat the market or
1:01
time the market, that you're
1:03
better off indexing and dollar cost averaging
1:05
rather than trying to be an active
1:07
stock picker. They say there's
1:09
no persistence of returns for hedge funds,
1:11
that this year's big winner can be
1:14
next year's big loser, and
1:16
that nobody gets huge outperformance without
1:18
taking huge risk. When
1:20
I was in college, I actually took an
1:22
economics class with Burton Malkiel, who of course
1:24
was involved in starting Vanguard and is a
1:27
big proponent of all that. And that is
1:29
what I learned, Ben. Well, David, it
1:31
turns out they were wrong. Today
1:34
listeners, we tell the story
1:36
of the best performing investment
1:38
firm in history, Renaissance Technologies,
1:41
or RENTEC. Their
1:43
30-year track record managing billions of
1:45
dollars has better returns than anyone
1:47
you have ever heard of, including
1:50
Berkshire Hathaway, Bridgewater, George Soros, Peter
1:52
Lynch, or anyone else. So
1:54
why haven't you heard of them? Or if
1:56
you have, why don't you know much about them?
1:59
Well, their eye popped performance is matched
2:01
only by their extreme secrecy, and
2:03
they are unusual in almost every
2:05
way. Their founder, Jim Simons,
2:07
worked for the US government in the
2:10
Cold War as a codebreaker before starting
2:12
Renaissance. None of the founders
2:14
or early employees had any investing background,
2:16
and they built the entire thing by
2:19
hiring PhD physicists, astronomers, and speech recognition
2:21
researchers. They're located in the middle of
2:23
nowhere in a tiny town on Long
2:25
Island. They don't pay attention to revenues,
2:28
profits, or even who the CEOs are
2:30
of the companies that they invest in.
2:32
And at any given time, they probably
2:34
couldn't even tell you what actual stocks
2:37
they own. Now, you may
2:39
be thinking, okay, great, I just learned about
2:41
this insane fund with unbelievable performance. And
2:44
to be specific listeners, that's 66% annual
2:47
returns before fees. And,
2:49
you know, well, I want to invest. Well,
2:51
you can't. To add to
2:53
everything else that I just said, RENTEC's
2:56
flagship medallion fund doesn't take any outside
2:58
investors. The partners of the firm have
3:00
become so wealthy from the billions that
3:02
the fund has generated that the only
3:04
investors they allow in are themselves.
3:08
Oh, we are going to talk a
3:10
lot about that towards the end of
3:12
the episode, because I think it's kind of the key to
3:14
the whole thing. Ooh, cliffhanger,
3:17
David, I'm excited. So
3:19
what exactly does Renaissance do? Why does it
3:21
work? And how did it evolve to be
3:23
the way it is today? And
3:25
while the resources are out there are
3:28
scarce, because for one, employees sign a
3:30
lifetime non-disclosure agreement, David and
3:32
I are going to take you through everything
3:34
we've learned about the firm from our research
3:36
dating all the way back before Jim Simon
3:39
started as a math professor to understand it
3:41
all. This episode was selected
3:43
by our acquired limited partners. And to be
3:45
honest, I didn't think enough people knew what
3:47
RENTEC was to pick it. But when we
3:49
put it out for a vote, the people
3:51
have spoken. So if you want to become
3:53
a limited partner and pick one episode each
3:55
season and join the quarterly zoom calls with
3:58
us, you can join at acquired.f If
4:02
you want to know every time a new episode
4:04
drops, sign up at acquired.fm slash email. These
4:06
emails also contain hints at what the next
4:09
episode will be and follow up facts from
4:11
previous episodes. For example, we
4:13
had a listener, Nicholas Cullen, email
4:15
us this time who found the
4:17
actual document with the bylaws of
4:19
Hermes's controlling family shareholder H51, which
4:21
we linked to in this most
4:25
recent email. Tom talked about
4:27
this episode with us after listening at acquired.fm slash
4:29
Slack. If you want more from David and I,
4:31
check out ACQ2. Our most
4:33
recent episode was with Lata Bier-Newton, who
4:36
led the team that created the first
4:38
GLP ones at Novo Nordisk. So awesome
4:40
follow up to the Novo episode if
4:42
you liked that one. Before
4:44
we dive in, we want to briefly share
4:47
our presenting sponsor this season is JP Morgan,
4:49
specifically their incredible payments business. Yeah,
4:51
just like how we say every company has a
4:53
story, every company's story is powered by payments. And
4:56
JP Morgan payments is a part of so many
4:58
companies that we talk about on acquired. It's not
5:00
just the Fortune 500, too. They're
5:03
also helping companies grow from seed to
5:05
IPO and beyond. So
5:07
with that, the show is not investment advice. David and
5:09
I may have investments in the companies we discuss or
5:11
perhaps wish we did. And this
5:14
show is for informational entertainment purposes
5:16
only. David, where do we start
5:18
our story today? Well,
5:20
we start in 1938 in
5:24
Newton, Massachusetts, which is
5:26
a fairly wealthy suburb just outside of Boston,
5:29
where one James Simons is
5:31
born. Both of Jim's
5:33
parents were very, very smart, especially
5:35
his mother, Marsha. His
5:37
dad was a salesman for 20th Century
5:40
Fox, the movie company. His job was
5:42
he went around to theaters in the
5:44
Northeast and sold packages of
5:46
movies to them. Super cool. And
5:49
we know all this because we have to
5:51
thank Greg Zuckerman, author of The Man Who
5:53
Solved the Market, which is the only book
5:56
out there that is solely dedicated to RENTEC
5:58
and Jim Simons. got to
6:00
talk to Greg in our research. He helped us out
6:02
a bunch. Thank you, Greg. And helped fact check a
6:04
few of our assumptions of what happened after the book
6:06
came out. So that
6:08
was Jim's parents. But really
6:11
a major influence on him growing
6:13
up was his grandfather, Marsha's
6:15
dad. There's already echoes of the Bezos
6:17
story here with the grandfather, the mother's
6:19
father, and spending a bunch of time
6:22
with him, and rubbing off on young
6:24
Jeff or young Jim in this case.
6:27
And Bezos, of course, would get his
6:29
start in his career at D.E. Shaw.
6:32
A quant fund coming up at the same
6:34
time as RENTEC. But back
6:36
to Jim here in the 1940s, his
6:39
grandfather, Peter, owned
6:41
a shoe factory that made women's
6:43
dress shoes. Jim spends a ton
6:45
of time there growing up at the factory. So
6:49
Jim's grandfather, Peter, was
6:51
quite the character. He was
6:53
a Russian immigrant. And he's
6:55
kind of like still more Russia than Boston at
6:57
this point in time. As
6:59
Greg puts it in the book, Peter
7:02
reveled in telling Jim and his cousins
7:04
stories of the motherland involving wolves, women,
7:06
caviar, and vodka. And he teaches young
7:09
Jim when he's a child here in
7:11
the factory to say Russian phrases like,
7:13
give me a cigarette and kiss my
7:16
ass. Which I think he probably would
7:18
say that thousands of times the rest of his life.
7:20
I think so. If you watch
7:22
interviews with Jim, his hands are always
7:25
twitching because he has chain smoked
7:27
his entire life probably going back to
7:29
age 10 in the factory. Three packs
7:31
of merits a day. Unbelievable.
7:33
Although I think he quit later in life,
7:35
but he definitely chain smoked the better part
7:37
of the first, call it, 75 years or
7:40
something. I mean, there's these famous stories of
7:42
the conference rooms at RENTEC and the war
7:44
rooms when the market is going through a
7:46
crazy gyration and it's just filled with cigarette
7:48
smoke. And it's all Jim. Different
7:50
time. Different time. So
7:53
back to Jim's childhood though, here in
7:55
the Boston suburbs. He
7:57
grows up certainly not uber wealthy.
8:00
are uber rich, but very, very solidly
8:02
upper middle class, and especially he's an
8:04
only child. He has all the resources
8:06
of his parents, his family, his grandfather's
8:08
this sort of well to do entrepreneur.
8:11
And Jim, you know, he gets to rub
8:13
shoulders in the Boston area with people who
8:15
are really rich. And
8:18
he says later, I observed that it's very nice
8:20
to be rich. I had no interest in business,
8:22
which is not to say I had no interest
8:24
in money. Yes, important to
8:26
tease out the difference between those two
8:29
things. Yes, very, very important. And
8:31
what he means when he says he has no interest in business,
8:34
it's because from a pretty young
8:36
age, he gets really into
8:38
math. So the legend has
8:40
it when Jim is four years old,
8:43
he stumbles into one of Zeno's famous
8:45
paradoxes from ancient Greek times. Yep,
8:47
this is great. The basic gist of
8:50
Zeno's paradoxes, if you are always taking
8:52
a quantity and dividing it by two,
8:54
you will never hit zero,
8:56
you will asymptotically approach zero, but you will
8:58
never actually touch zero, you need to do
9:01
addition or subtraction to do that division won't
9:03
cut it. And so Jim as a four
9:05
year old, when he observes they need to go
9:08
to the gas station to fill up the tank,
9:10
he throws out the idea, well, let's just use
9:12
only half the gas in the tank, because then
9:14
we'll still be able to after that only use
9:16
half the gas in the tank. And you know,
9:19
the funny thing that doesn't occur to a four
9:21
year old is, well, then we're just not going
9:23
to get very far. So
9:25
Jim's dream is to go to MIT down
9:28
the street in Cambridge and study
9:30
math. He graduates high school in
9:32
three years. And during the second
9:34
semester of Jim's freshman year there,
9:36
he enrolls in a graduate math
9:38
seminar on abstract algebra. So
9:40
pretty, you know, heady stuff. Yeah,
9:43
and Jim would go on to finish his
9:45
undergrad at MIT in three years and get
9:47
a master's in one year. Yeah,
9:49
pretty, pretty smart. But it
9:52
turns out that that freshman year grad seminar
9:54
he took actually has a big
9:56
impact on him because he doesn't do well
9:58
in the class. He can't keep up. And
10:01
Jim's pretty self-aware here. There
10:04
are other people at MIT who
10:07
never run into problems. They never
10:09
hit a limit. They never struggle
10:11
understanding any concept. And
10:13
he realizes that, oh, I'm
10:16
smart. I'm very, very smart. I'm smarter than most
10:18
other people here. But I'm not
10:20
one of those people. Right,
10:23
which is, what do you do with that information?
10:25
You realize you have to add a few of
10:27
your skills together to become the best at something.
10:29
You have to be smart and something
10:31
else. Yes. So Jim's own words on this
10:34
are, I was a good mathematician. I
10:36
wasn't the greatest in the world, but I was pretty good.
10:39
But he recognizes, like you said, Ben, that
10:41
he has a different advantage that most of
10:43
the supergeniuses lacked. And that's that as he
10:45
put it, he had good taste. So
10:48
these are his words. Taste in science is
10:50
very important. To distinguish what's a
10:52
good problem and what's a problem that no
10:54
one's going to care about the answer to
10:57
anyway, that's taste. And I think I
10:59
have good taste. By the way,
11:01
this is exactly the same thing as Jeff
11:03
Bezos. In college, realizing he
11:05
wanted to be a theoretical physicist, he
11:07
met some of the extreme brainpower
11:09
people that would go on to become the
11:11
best theoretical physicist in the world. And he
11:14
said, I'm smart, but I'm not that smart.
11:16
And so switched to computer science. I
11:18
think the analogy here is
11:20
like sports. There
11:23
are all-star players. There
11:25
are hall of famers. And then there's
11:27
LeBron and MJ. And
11:29
Jim ends up being a hall of famer
11:31
mathematician. But he's not Tom
11:33
Brady. I mean, he's got a pretty important theorem
11:35
named after him. That goes on to become a
11:38
foundation of string theory and physics, which isn't even
11:40
Jim's field. Crazy. So
11:42
this realization that Jim has about
11:44
himself, though, both that he's not
11:47
the smartest person in the room at a place like
11:49
MIT, but he can hang with them, and
11:52
that he has this taste concept,
11:55
I think becomes one of the
11:57
most important keys to the secret
11:59
sauce. that ends up getting built at RENTEC, which
12:02
is that he can relate
12:04
to everybody. He understands what's going on.
12:06
Any person off the street probably couldn't
12:09
even really have a conversation with these
12:11
folks, but he can. And
12:13
yet he also has the perspective, maybe some
12:15
of this is from his grandfather, of what
12:18
is important out there in the real world.
12:20
And as a result, all
12:22
of his friends at MIT and these super smart people, they
12:24
look up to him because you
12:27
aren't like the kid in the
12:29
corner at the high school dance.
12:31
You're cool. He's the extroverted theoretical
12:33
mathematician. Yes. So
12:36
he was elected class president in high school.
12:39
He smokes cigarettes. He's popular with
12:41
the ladies. He kind of
12:43
looks like Humphrey Bogart. He's a popular
12:45
dude, especially at this point in time.
12:48
We're now in the late 50s when Jim's at MIT. This
12:51
is kind of James Dee and Rebel Without
12:53
a Cause era. So
12:55
after graduation, Jim leads his
12:58
buddies on a road
13:00
trip with motor scooters. You
13:02
can't make this stuff up from Boston down
13:05
to Bogota, where one of his classmates is
13:07
from. The idea is that they're going
13:09
to do something so epic that the newspapers are going
13:11
to have to write about it. So
13:13
they all load up on scooters and
13:16
drive down to Bogota. They get into all
13:18
sorts of adventures. There's knives, guns, and they
13:20
get thrown in jail. It's honestly crazy that
13:22
this group of people took this type of
13:25
risk. Totally crazy. So
13:28
after he's done at MIT and after the road
13:30
trip, Jim heads out to
13:32
Berkeley in California so that he
13:34
could do his PhD with the
13:36
professor Xing Shen Chern. And
13:38
much later in life, Jim would collaborate
13:40
with Chern for the Chern-Simons theory that
13:43
we talked about earlier that becomes one
13:45
of the foundational parts of string theory
13:47
in physics. But before Jim
13:49
leaves for the West Coast, he
13:51
meets a girl in Boston. And they
13:54
decide to get engaged in four
13:56
days. I mean, this is This
14:00
is him back there. These are the times. And
14:03
when they get to California and they
14:05
get married, Jim takes the
14:08
$5,000 wedding gift that I believe they
14:10
got from her parents, and he decides,
14:12
I want to multiply this. So
14:15
he starts driving from Berkeley into San
14:17
Francisco every morning to go hang out
14:19
at the Merrill Lynch brokerage office and
14:22
just be a rat hanging around the brokerage
14:24
and find ways to trade and turn this
14:26
money into something more. Which is
14:28
so interesting to think about because at that point in time,
14:31
there was such an advantage to just
14:33
being there. This wasn't even the trading
14:35
floor, but information is all so manual
14:37
and all so relationship-driven in the markets
14:39
that there was basically no way to
14:41
be in on the action unless you
14:43
were physically there in on the action.
14:46
Exactly. Yeah, you couldn't just
14:48
log into Yahoo Finance or something or open
14:50
the stocks app on your iPhone, which even
14:52
the information they were getting was God knows
14:54
how long delayed from New York or from
14:56
Chicago for the futures and commodities that are
14:58
being traded that Jim gets into. He's
15:01
as close to the action as he can possibly
15:03
be, but he's a long, long way from the
15:05
action. Yep. Nonetheless, when
15:08
he starts out doing this, Jim hits
15:11
a hot streak and he goes up 50% in a few
15:13
days. Trading
15:15
is easy. Trading is easy. He says,
15:17
I was hooked. It was kind of
15:19
a rush. I bet. Except
15:21
he ends up losing all of his profits
15:23
just as quickly. Important
15:26
to learn that lesson early. Yes. And
15:28
also right around this time, Barbara, his
15:30
wife, gets pregnant with their first child
15:32
and is like, you can't be
15:34
driving into San Francisco every morning at
15:37
gambling our future like this. Right,
15:39
effectively playing the ponies. Yeah, exactly.
15:42
So Jim's like, okay, okay, I'll
15:44
stop. I'll focus on academia for
15:46
now. So he finishes his
15:48
PhD in two years. They come back to
15:50
Boston and he joins MIT as
15:53
a junior professor at age 23. So
15:55
they stay one year in Boston. But
15:58
Jim, even though he's got a family... even
16:00
though he's super successful as a young academic
16:02
here, he's got kids. He's
16:04
restless. So one
16:06
of his buddies from the scooter trip to Bogota is
16:09
from Bogota and lives there, his family's there. He
16:11
has an idea to start a flooring
16:14
tile manufacturing company, because he's like, you
16:16
know, the flooring at MIT and in
16:18
Boston, it's so much nicer than at
16:20
Bogota. We should start a company and
16:22
make the same kind of flooring here.
16:24
When I read this, I couldn't believe
16:26
that this was Jim Simon's first business
16:29
venture. It's so random, but it really
16:31
is emblematic of just how much
16:33
he was thrill seeking and just looking for anything
16:36
that was unexpected, different,
16:38
exciting. He just gets bored
16:40
fast. Totally. Not
16:43
just is this the start of his
16:45
entrepreneurial career. The seeds of
16:47
this financially are what go on to
16:49
start Red Tech. It's wild. Totally
16:52
wild. So Jim takes a year
16:54
off and goes down to Bogota.
16:56
This is a guy with an
16:58
MIT undergrad and master's
17:00
and a Berkeley PhD in theoretical
17:03
math. Who's now a professor at
17:05
MIT. Who is taking a year off
17:07
to go work on a flooring company in Bogota. Yes,
17:11
accurate. So he does that for a year. They get it set
17:13
up. He gets bored again. He's like, all right, I don't want
17:15
to run this company. I've helped set it up. I have an
17:17
ownership stake in it now. He
17:19
bounces back to Boston,
17:21
this time to Harvard, as
17:23
a professor there for a year. He's really racking
17:25
him up. But he
17:28
spends a year there and he's like, ah, got
17:30
the itch again. And the junior professor's
17:33
salary isn't that much. And like
17:35
we said about him back from his childhood days,
17:37
he sees the appeal in being rich. He's like,
17:39
this is not a path to be rich. So
17:44
he's like, I'm going to go put my skills
17:46
out on the open market. He gets
17:48
a job in Princeton, New Jersey,
17:50
not at Princeton University, but
17:53
at the Institute for Defense Analyses,
17:57
which is a nonprofit organization.
18:00
that consults exclusively
18:03
for the U.S. government, specifically
18:06
the Defense Department, and
18:08
specifically the NSA. These
18:12
are the civilian codebreakers. Yes.
18:14
It was basically formed with
18:16
this idea that, one, across
18:18
various branches of our government,
18:20
we need better collaboration and
18:23
cross-funding of the same initiatives.
18:25
And two, there are going to be a
18:27
lot of people who don't work for the government that
18:30
we're going to want to hire to do some pretty
18:32
secret work. Yep. So
18:35
the IDA there in Princeton kind
18:37
of functions like the Institute
18:39
for Advanced Study, which is also
18:42
in Princeton. That's where Einstein went
18:44
when he came to America, kind
18:46
of an independent think tank research
18:48
group, except it's solely focused on
18:51
code-breaking and signal intelligence with the
18:53
Russians during the Cold War. Yeah.
18:56
And it's a pretty wild charter, and especially how
18:59
special of an organization it was. The way these
19:01
people would spend their time is part
19:04
code-breaking, but part kind of
19:06
goofing around because the creativity
19:09
of mathematicians working together on
19:11
passion projects is important to
19:13
discovering clever new algorithms. Yes.
19:16
This is so, so key. And
19:19
this culture ends up getting translated
19:21
whole cloth right into RENTEC. So the
19:23
way IDA worked, and I assume still
19:25
works to this day, is
19:28
they recruited top mathematicians
19:30
and academics to
19:32
come be code-breakers there. They would double
19:35
their salaries. And importantly, it couldn't have
19:37
been a government division if they were
19:39
going to be doing that because there's
19:41
very specific, congressionally approved budgets for payroll.
19:44
Exactly. They figured out that
19:46
they needed to attract the smartest people in the
19:49
world who weren't going to come just go work
19:51
for the Department of Defense. This was
19:53
the way to do it. So like
19:56
you said, Ben, the charter
19:58
of the group was that employees had to...
20:00
to spend 50% of their time doing code
20:02
breaking. But the other 50% of the time, they
20:06
were free to do whatever they
20:08
wanted, like research, pursue whatever they
20:10
were doing in academia, publish papers,
20:13
kind of the appeal of going there was, hey,
20:17
it's the same thing as being a professor at
20:19
MIT or Princeton or Harvard or whatever, except
20:21
you're doing code breaking instead of teaching.
20:24
And there's no bureaucracy to worry about,
20:26
there's no politics. It's just like, hey,
20:28
you do your code breaking work and
20:30
then you publish it, you can collaborate
20:32
with your colleagues there. Now,
20:36
this is pretty crazy. Very quickly
20:38
after Jim arrives at IDA,
20:41
remember he's in money-making mode at this point
20:43
in time, he recruits
20:46
a bunch of his very brilliant
20:48
colleagues to come work with him in
20:50
their 50% free time on
20:53
an idea to apply the
20:55
same work and technologies that
20:57
they're using in code breaking
20:59
and signal intelligence to
21:02
trading in the stock market. So
21:05
they come together and they publish
21:07
a paper called a probabilistic models
21:09
for and prediction of stock market
21:11
behavior. And everything
21:14
that they suggest in this paper really
21:17
is rent tech, just 20
21:20
years before rent tech. It's
21:22
crazy, 1964, this was published?
21:25
Yes, now at
21:27
this point in time, fundamental
21:29
analysis was then, as in most of
21:31
the world today still is, the
21:34
primary way of investing in things of,
21:36
hey, I know this company, I'm gonna
21:38
analyze their revenues, their price multiple, or
21:41
I'm gonna think about what's happening in
21:43
the currency markets or in the commodity
21:45
markets and why copper is
21:47
moving here or the British pound is
21:49
moving there and I'm gonna invest on
21:51
those insights. You're effectively looking at the
21:54
intrinsic value of an asset, trying to
21:56
assign it a value and make investments
21:58
based on that. Yes. fundamental investing.
22:01
There also existed in the
22:03
60s technical investing,
22:06
which kind of is
22:08
voodoo. This
22:11
is like I'm looking at
22:13
a stock chart and I've got a feeling
22:15
that it's going to go up. Like, I'm
22:17
tracing this pattern and like it's going up,
22:19
baby, or no, no, no, this pattern is
22:21
going down. Yeah, using the phrase technical might
22:23
be a little generous. But what
22:25
they're looking for, basically trying to mine
22:27
trading behavior for signal about the way
22:30
that it will trade in the future
22:32
rather than mining the intrinsic information about
22:34
an asset for what you think it
22:36
will do in the future. Right.
22:39
And what Jim and his colleagues here are
22:41
suggesting is that, but
22:44
just not really done by humans.
22:46
It's that with a lot more
22:48
data and a lot more sophisticated
22:51
signal processing. And importantly,
22:53
you might say, why is it this group
22:56
of people that came to that conclusion of
22:59
applying computational signal analysis
23:01
to investing? Well, it's effectively the
23:03
same thing as code breaking. You
23:05
are looking for signal in the
23:08
noise and trying to use computers
23:10
and algorithms to mine signal from
23:12
something that otherwise kind of looks
23:15
random. Totally. When Jim started
23:17
working on code breaking, I think
23:19
he just looked right back to his experience trading in
23:21
the markets and was like, whoa, this is the same
23:24
thing. Which is not an
23:26
insight other people had. That was the
23:28
amazing thing about his background, priming him
23:30
to realize that. Yes, there's all this
23:32
noise in this data. And it is
23:35
impossible for a human to sit here
23:37
and look at this data and say,
23:39
oh, I know what the Soviets are
23:41
saying. No, no, you have to use
23:43
mathematical models and statistical analysis to extract
23:45
the patterns. So mathematical
23:48
models, statistical analysis. We actually hear
23:50
a lot of that in the
23:52
world today because machine
23:54
learning is a thing. Yes,
23:57
what they are really doing here at IDA
24:00
and then soon in
24:02
RENTEC is early machine learning. And
24:05
Jim just had this incredibly
24:07
brilliant insight that you can
24:09
use these techniques and this
24:11
technology for making investments, which
24:14
makes this the perfect time to talk about
24:17
our presenting sponsor for this season, JP Morgan
24:19
Payments. Yes. The
24:21
finance industry has a rich history
24:23
of innovating dating all the way
24:25
back to the literal Renaissance where
24:27
double entry bookkeeping and letters of
24:30
credit revolutionize global trade and economic
24:32
development. And JP Morgan Payments really
24:34
continues that tradition in their technology
24:36
investments today. They move $10
24:38
trillion a day securely. That is
24:40
a quarter of all US dollar
24:43
flows globally. Just think about the
24:45
sheer volume of data at 5,000
24:47
transactions per second and how
24:49
important that is to the global economy. Unsurprisingly,
24:52
JP Morgan Payments has been in
24:54
the AI game for years now.
24:56
Similar to RENTEC, they were also
24:58
early to recognize the value of
25:00
AI to gather, process and analyze
25:02
those massive troves of data to
25:04
provide solutions for their customers and
25:06
mitigate risk. Like when they incorporated
25:08
AI into their cashflow forecasting tool,
25:10
which helps businesses manage liquidity and
25:12
that proved especially valuable during the
25:14
pandemic. Yep. So also unsurprisingly, JP
25:16
Morgan was ranked number one in
25:18
a recent global banking index of
25:20
AI capabilities with Fortune saying they
25:22
were quote, head and shoulders above
25:24
the others. Their customers get AI
25:27
powered payment solutions for fraud prevention,
25:29
customer insights and treasury insights, all
25:31
of which grows the bottom line.
25:33
They can even analyze transaction data
25:35
to predict and mitigate fraud patterns
25:37
in real time with their validation
25:40
services, helping stop millions of dollars
25:42
for customers in attempted fraud. Yep.
25:45
We were doing some research to prep
25:47
for this segment and we came across
25:49
something pretty wild. The United States Treasury
25:51
Department has started using AI to detect
25:53
suspected check fraud and recovered over $375
25:56
million in 2023. utilizing
26:00
the new tools. The U.S. Treasury
26:03
Department disperses trillions of dollars annually. So
26:05
if they continue to employ new technologies
26:07
like this, it could really add up
26:09
to the tune of billions. So
26:11
how does this fit in? Well, the
26:13
Treasury Department recently selected J.P.
26:15
Morgan to provide account validation
26:18
services for federal agencies. Obviously,
26:20
payment integrity and this issue of improper
26:22
payments is top of mind for them
26:25
and at enormous scale. So whether you
26:27
are one of the largest institutions in
26:29
the world or a small business like
26:31
us here at Acquired, J.P. Morgan offers
26:33
you peace of mind and protection. Yeah.
26:37
One more playbook theme in common between
26:39
Rentec and J.P. Morgan Payments. They both
26:41
analyze data to uncover patterns and insights
26:43
you may never think to look for.
26:45
One of their clients, a furniture store,
26:48
discovered a correlation with customers who also
26:50
shop at pet stores where shoppers spent
26:52
76 percent more than the average customer
26:54
when this was the case. So
26:57
the furniture store launched a line of
26:59
pet friendly furnishings for that audience. These
27:02
are the sorts of insights that drive growth with J.P.
27:04
Morgan Payments as your partner. To
27:20
learn more, check out jpmorgan.com/acquired and
27:22
fun fact listeners, it is fraud
27:24
prevention month. So listeners can learn
27:26
even more by following JP Morgan
27:28
on LinkedIn. Okay,
27:30
David, so this paper is
27:32
published. They're gonna trade and make
27:35
a whole bunch of money in the stock
27:37
market by applying this code breaking signal
27:40
processing data analysis approach
27:42
to investing. Yep, so
27:45
then the natural question is, okay, what
27:47
is the model here? How are they gonna do
27:49
this? And it turns out that
27:51
one of the employees of IDA at this time and one of
27:53
the members of this sort of rebel group,
27:55
shall we say, within the organization is a
27:58
guy named Lenny Baum. And Lenny
28:00
just happens to be the world
28:03
expert in a mathematical
28:05
concept called a Markov model. Specifically,
28:08
a version of the Markov
28:10
model called a hidden Markov model. Now
28:14
a Markov model is a
28:16
statistical concept that's used to
28:18
model pseudo-random or
28:20
chaotic situations. Basically, it
28:23
says, let's abandon any
28:25
attempt to actually understand what is going
28:27
on in all of this data that
28:29
we have, and instead
28:31
just focus on what are the
28:33
observable states that we can see
28:36
of the situation. Can we identify
28:38
different states that the situation is
28:40
in? And if we just
28:42
do that, can we predict
28:45
future states based on
28:47
what we've observed about the patterns of
28:49
past states? And the answer to that
28:51
is usually yes, even
28:53
if you don't know anything about fundamentally how
28:55
the system operates. So the great
28:58
example that Greg Zuckerman gives in the book
29:00
is... Yes, a baseball game. There's
29:02
three balls and two strikes. That
29:05
state has a narrow
29:07
set of states after it. It's
29:09
going to be a strikeout, they're going to get
29:11
on base, it's going to be a walk, or
29:13
maybe they foul it off and it keeps going.
29:15
There's only really a narrow set of things that
29:17
could happen after that. Whereas when it's zero balls
29:19
and zero strikes, there's a lot that
29:22
could happen. They could just keep pitching. And
29:24
if you don't know the rules, you're like,
29:26
why do they just keep pitching? And so
29:28
it's this sort of great way to explain
29:30
this idea of the black box that if
29:33
nobody tells you the rules to the game
29:35
by observing the outputs enough and observing, okay,
29:37
in this state, these outputs are possible, you
29:40
actually can kind of get pretty good
29:43
at least, if
29:45
not predicting, understanding the probability distribution
29:47
of the outcomes for any given
29:49
state in the game. So
29:52
we brought up machine learning and AI a
29:54
minute ago. This is
29:56
a foundational concept to modern
29:58
day AI. If you think
30:00
about large language models and predicting what
30:03
comes next, it's not like these large
30:05
language models necessarily understand
30:07
English. They're just
30:09
really, really good at predicting states and
30:12
the next state, i.e., characters and the
30:14
next character, or pixels and the next
30:16
set of pixels or frame, etc. And
30:19
obviously, they're much fancier than that, but that is
30:22
kind of the underpinning of it all. I
30:24
mean, I remember in my sophomore year of
30:26
college computer science class, I had a Markov
30:29
chain assignment, and it was basically write a
30:31
Java program to ingest this public domain book,
30:33
and then I would give it a seed
30:35
word, the first word of each sentence, and
30:37
press return, return, return, return, return. And it
30:39
would scan through the probability tree and give
30:41
me the most probable word based on the
30:43
corpus of the book that it just read
30:45
to create some sentence. And it feels like
30:47
magic. And of course, in these early rudimentary
30:50
Markov chain things like the one I did
30:52
in college, it kind of spits out nonsense.
30:54
But that would evolve to be the LLMs that
30:57
we know of today. Yes,
30:59
totally. And that is what they
31:01
were using at IDA to do code breaking. And
31:03
that's what they propose in this paper that
31:06
they could use in the stock market. Exactly.
31:09
And the way that this applies to investing
31:12
is just like you might
31:14
not know the rules of baseball. But if
31:17
you've watched enough baseball, you can kind of
31:19
guess at what the probabilities of the next
31:21
thing to happen are based on the state.
31:24
That's kind of the same thing, or at
31:27
least the stock market movements are where you
31:29
don't know the future, you don't know what's
31:31
going to happen. You don't know if stock
31:33
X affects stock Y in some way, because
31:36
you don't know in what way those companies
31:38
do business together or who holds both stocks.
31:40
Are they overlapping investors? You don't know the
31:42
relationship between those companies. So you can't
31:45
forecast with 100% certainty what is
31:47
going to happen. However, if you
31:49
suck in enough data about what has
31:51
happened in the past and the probability
31:53
distribution from every given state in the
31:55
past, you probably could make some educated
31:58
guesses or at least understand. and
32:00
the probability of any individual outcome based
32:02
on a state today of what could
32:05
happen next. Yes, exactly.
32:08
So Jim and Lenny and
32:10
this whole little crew, they're
32:13
pretty fired up. They're like, oh,
32:15
great. Let's go
32:17
raise a fund and invest
32:20
in the markets using this strategy. Certainly
32:22
we're gonna be successful at raising that
32:24
fund and certainly we're gonna be very
32:26
profitable because we've got this great idea. Totally,
32:28
what could go wrong? Well,
32:31
in the mid sixties, the
32:34
idea that some wonky academics at
32:36
some random secretive agency in
32:38
Princeton, New Jersey could
32:41
go raise money was
32:43
non-viable. I mean, it was hard
32:45
enough for Warren Buffett to raise money at this
32:47
point in time for his fund. And
32:50
he was Benjamin Graham's anointed,
32:52
appointed disciple. And here
32:54
are these academics who are working
32:56
at some random unknown nonprofit saying,
32:59
give us money. We don't
33:01
know anything about these companies that we're gonna
33:03
invest in. We don't know anything about fundamentals,
33:05
but we've got a really good algorithm. People
33:07
are probably like, what is an algorithm?
33:10
So they just have no access to capital. Right,
33:12
this was decades before it became high
33:15
pedigree to come from a technical computer
33:17
science background in the world of investing.
33:20
Yes. So a bunch
33:22
of kind of Keystone Cops style fundraising
33:24
happens here. They're going around in secret.
33:27
They're trying to keep the IDA bosses
33:29
from knowing what they're doing. One
33:32
of the group ends up leaving a
33:34
copy of the investment prospectus on the
33:36
copy machine one
33:39
night and the boss discovers it and calls them
33:41
all into his office and is like, guys, what
33:43
are you doing here? Right, it's a little bit
33:45
of a clown show on the operational side, even
33:47
if the idea is good. Yes.
33:50
So they end up abandoning the
33:52
effort both because they can't raise money
33:54
and because IDA has found out about this and they're
33:57
not too pleased. Shortly after all
33:59
of this though, Jim. ends up moving on
34:01
anyway, because the Vietnam
34:03
War starts, and he, as
34:06
you can imagine from his background, is not
34:08
a supporter of the Vietnam War at this point in time.
34:11
Jim writes an op-ed in
34:13
the New York Times denouncing the
34:15
Vietnam War and saying, like, yeah, he's, you
34:17
know, sort of part of the Defense Department,
34:19
but like, not everybody in the Defense Department
34:21
is for the war. Which is
34:23
so naive, thinking you can write an op-ed
34:26
in the New York freaking Times, and
34:29
it's not going to create issues for you in
34:31
your job. Even more than that, amazingly,
34:33
nobody really paid attention to it except
34:35
a reporter at Newsweek who then comes
34:38
to interview Jim and ask him some
34:40
more questions, and he just doubles down
34:42
on this. And when the Newsweek piece
34:44
comes out, that's when the Department of Defense
34:46
is like, all right, you got to fire this guy. So
34:51
Jim gets fired in
34:53
1967, even though he's a
34:56
star code breaker, he made supposedly huge
34:58
contributions to the group, which are still
35:00
classified. But at age 30,
35:02
with a wife and three kids, he's
35:05
out on the street. And
35:07
even though he's super smart, his
35:09
colleagues love him clearly, he's now
35:11
bounced out of MIT, he's bounced out
35:13
of Harvard, he's gone
35:15
to this seemingly final
35:17
home for him, a great place at IDA, he
35:20
gets bounced out of there too. His
35:23
job prospects are not great. Yeah.
35:26
So he takes pretty much
35:28
the only halfway decent paying job that
35:31
he could get, which is
35:33
to be the chair of the newly
35:35
established or maybe reestablished math
35:37
department at the State University
35:40
of New York, Stony Brook,
35:42
which is the Long Island campus
35:45
of the State University of New
35:47
York. This is
35:49
not Harvard. This is not MIT.
35:52
No, it is not. But
35:54
it did have one very important thing going
35:57
for it, which is why Jim ended up
35:59
there. And that is that Nelson
36:01
Rockefeller, who was then the governor of New York,
36:04
had launched a campaign, a
36:06
hundred million dollar campaign to
36:08
try and turn this Long
36:11
Island campus of the State University of New
36:13
York into a mathematical
36:16
powerhouse to become the Berkeley
36:18
of the East. I
36:20
sort of thought MIT was the Berkeley of
36:22
the East already, but Rockefeller is waging a
36:24
campaign that he wants Stony
36:27
Brook to become a math
36:30
and sciences powerhouse.
36:32
And Jim is the key. He
36:35
wouldn't be able to recruit somebody like Jim otherwise,
36:37
but because he's now kind of tarnished
36:39
his career, here's a
36:42
like very talented mathematician that they can convince
36:44
to come be chair of the department. Yep.
36:47
So they basically give Jim an
36:49
unlimited budget and leeway to
36:51
go try and poach math professors
36:54
from departments all over the country and the
36:56
world and bring them there to Long Island.
36:59
And part of how Jim goes and
37:01
recruits folks is money, like the old, hey, I'll
37:03
double your salary line. But
37:05
the other part of it too is
37:07
he's given such leeway and
37:10
Stony Brook is so different from the politics
37:13
of an MIT or a Harvard or a
37:15
Princeton. He says, hey, come here, I'll pay
37:17
you more. But even more
37:19
importantly, you can just focus on
37:21
your research. You're not gonna have
37:23
to deal with committees. You're not gonna have to
37:26
do all this stuff. There is none of this
37:28
stuff here. You might have to teach a little
37:30
bit, but that's not even the point. Rockefeller doesn't
37:32
want this necessarily to become a great
37:34
teaching institution. He just wants to assemble talent
37:36
there. Yep. And
37:39
amazingly, it works. Jim starts
37:41
getting a bunch of great talent, including James
37:43
Axe, who is a superstar in algebra and
37:45
number theory from Cornell. And
37:47
he ends up at Stony
37:49
Brook recruiting and building one
37:52
of the best math departments in the world. Amazing.
37:56
Totally amazing. But in true
37:58
Jim fashion after a couple of years of this. this, and
38:01
also his marriage with Barbara falling apart,
38:04
he starts getting restless again. He decides
38:06
that he wants to go on a sabbatical
38:08
and go back to Berkeley and reunite with
38:10
his old advisor there and go spend some
38:12
time out on the coast in California. And
38:15
this is where Chern and Simons end
38:17
up collaborating and developing the Chern-Simons theory
38:19
that ends up winning the highest award
38:21
in geometry from the American Mathematical Society,
38:24
and really kind of is Jim's
38:26
personal mark on mathematics.
38:30
Now also, right around
38:32
the same time, remember the
38:34
Colombian flooring company, it
38:37
gets acquired. And Jim
38:40
and his buddies who are partners in it come
38:42
into a good amount of money. And
38:44
Jim is newly divorced,
38:46
he's restless in academia, he
38:49
has these ideas back from
38:51
when he was an IDA about what
38:53
you could do in the markets if
38:56
you had capital, he
38:58
starts trading again, and he gets
39:00
more and more into it. Meanwhile,
39:02
like we said, he's becoming disillusioned
39:04
again and restless at academia. And
39:07
in 1978, he leaves
39:09
to focus full time on trading, which
39:12
is a huge shock to the academic community.
39:14
Remember he's assembled this superstar team there at
39:16
Stony Brook. There's a quote in
39:18
Greg's book from another mathematician at Cornell.
39:21
We looked down on him when he did this,
39:23
like he had been corrupted and had sold his
39:25
soul to the devil. Yeah, I
39:27
mean, it was really viewed in the math
39:29
community as anyone who's going to do investing
39:31
is throwing away their talent. And it wasn't
39:33
even that it was common the way that
39:35
it sort of is today. Right, Jim was
39:37
the first one, but the idea that you
39:39
would leave to do anything commercial, you're
39:42
doing a disservice to humanity.
39:44
Yes, exactly. And leaving to
39:46
do anything, sure, but leaving to
39:48
do investing was almost just seen as dirty.
39:50
Like it's this rich person's game that
39:53
provides no value to society. Right.
39:55
Yeah, I don't think it was that the
39:57
rest of the math world was skeptical that
39:59
it could work, they probably were like, oh yeah, this
40:01
could work. But they were like, ew.
40:05
Academics tend to be much more motivated by prestige
40:07
than money. So I could totally see this other
40:09
people being like, oh, I could do that if
40:11
I wanted, but I have this higher calling and
40:13
everyone respects me for this higher calling and my
40:15
currency is the papers I publish and the awards
40:17
that I win and that's what I want. Yep.
40:20
Now, Stony Brook, we should say too, like it's
40:22
a very nice place. Yes. But it's
40:24
in the middle of Long Island on the North
40:26
Shore. This is not the Hamptons. It's
40:29
like the Long Island suburbs. Yep.
40:32
The wooded Long Island suburbs. Yes, the
40:34
wooded Long Island suburbs. Here's
40:36
Jim in a strip mall next to a
40:38
pizza joint, setting up his trading operation that
40:40
he decides very cleverly to
40:43
call monometrics, a
40:45
combination of money and
40:48
metrics or econometrics. And
40:51
he recruits his old
40:53
IDA buddy, original
40:55
partnering crime on the
40:58
trading idea, Lenny Baum to
41:00
come and join him. And
41:02
this time though, they have
41:04
some capital from the sale of the flooring company. And
41:07
how much did he make on that flooring sale? I
41:10
think together with Jim,
41:12
his partners and whatever money Lenny
41:14
put in, they had a little
41:16
less than $4 million in
41:18
this initial capital. In 1978. Yep.
41:22
Now, Jim also has another advantage at this point
41:25
in time, which is he's right
41:27
down the street from Stony Brook and
41:29
he's just recruited all of these
41:32
superstar mathematicians. The
41:34
table has been set. Yes. And those
41:36
folks are more loyal to Jim than they are
41:38
to Stony Brook. But they're more
41:40
loyal right now to academia than they
41:43
are to finance. This is not a
41:45
paved pathway until Jim paves this pathway.
41:48
Yes, in general, but some of them
41:50
and in particular, the superstar James
41:52
Axe, Jim convinces
41:55
to come join him in his
41:57
trading operations. So having Baum
41:59
and Axe and
42:01
Simon's, it's like suddenly this
42:03
extremely credible team in the
42:05
math world. Yes. Beyond
42:08
credible. Right. All the theorems that
42:11
a lot of mathematicians are using every day are
42:13
all named after these three guys who are now at
42:15
the same firm trading. Yes. And it's
42:17
led by Jim, who's somebody
42:19
that they respect as an academic, but even
42:22
more important is somebody they want to work
42:24
for and they look up to and they
42:26
think is cool. And he's out
42:28
there being like, hey, I think we can make
42:30
money. Right. Now,
42:33
at this point, they're primarily
42:35
trading currencies, not
42:38
stocks. And currencies are
42:40
obviously large markets, but they
42:42
aren't impacted by as many signals
42:44
and as many factors as
42:47
stocks are, or really even slightly more
42:49
complex commodities like, I don't know, soybeans
42:51
or whatever. And it seemed
42:53
to me like a lot of the trading
42:56
of currencies they were doing was basically based
42:58
on feelings that they had around how a
43:00
central bank was acting. Like if the head
43:02
of state of a certain country was going
43:04
to do something or not, it's basically like
43:06
betting on how one single
43:09
actor who was in control of
43:11
currencies at governments would act. So
43:14
to your point about very few signals
43:16
impacting price, it's knowing what one
43:18
person is going to do. Yes. And
43:20
this is super important. At
43:22
the end of the day, they built some models
43:24
there, they're getting the early
43:27
versions and infrastructure and scaffolding
43:30
of this quantitative approach set
43:32
up. But in terms
43:34
of the actual trades they're putting on, they're still doing
43:36
all of it by hand. And
43:38
they're still all really going
43:40
on a fundamental type
43:43
analysis. They'll take some signals from
43:45
the model, they'll see what's interesting, what they spit
43:47
out, but they're not going to act on anything
43:49
unless they can be like, oh yeah,
43:51
I see what is going
43:53
on here. I have a hypothesis. Right. The
43:56
computers are by no means running loose at this
43:58
point. By no means. at all. Yeah,
44:01
they're just suggesting patterns and ideas.
44:03
And Jim and Lenny and James, they have to then decide,
44:05
hey, are we going to do this or not?
44:07
Or are we going to do something just totally different than we
44:09
think is what's going to happen? Yeah.
44:12
And this actually does make
44:14
sense, really for two reasons.
44:17
One, computers and computing
44:19
power just wasn't
44:21
sophisticated enough yet
44:24
to really build
44:26
AI in a way that's powerful enough
44:28
that it could work well enough, you could
44:30
really trust it. That's one part. The
44:33
other part is these folks
44:35
are mathematicians. They're not
44:37
computer scientists. Right. And
44:40
they're really, really good at
44:42
building models, decoding signals, obviously,
44:45
but they're much more from this
44:48
realm of theory. And I actually
44:50
spoke with Howard Morgan, who's going to come up here in
44:52
a second. And he made this point to me. He's like,
44:55
in math, there's this concept of
44:57
traceability that's a really, really important
44:59
cultural tenet. It's like proving a
45:01
proof or proving a theorem or
45:03
something like that. You
45:06
really need to understand why to
45:08
get ahead in the field. It's not like you
45:10
can just say, oh, hey, the data suggests this.
45:12
It's like, no, no, no, you need proof. And
45:15
that's the world that these guys are coming
45:17
from. They're like, oh, we can use data
45:19
to sort of help us here. But ultimately,
45:21
we want to have a rock solid theory
45:23
of what is fundamentally happening here. Fascinating,
45:26
which is very different than we'll cram a huge
45:28
amount of data in and then whatever the data
45:30
suggests, we know it's true because the data suggests
45:32
it, which is sort of where
45:34
they would end up many years later
45:36
once they had both the hardware you're
45:38
referring to sophisticated computers, the clean data
45:41
that would be required to make all
45:43
of those incredibly numerous and
45:45
fast calculations, and also the
45:47
real computer engineering architecture to build
45:50
these scale systems to actually
45:52
act on large amounts of signals and understand
45:54
them all to come up with results. They
45:56
just didn't have any of that at the
45:58
time. So it was is hunches and
46:01
chalkboards. Yes. And so
46:03
much so that even Jim is
46:05
ringleader here. He's far from
46:07
convinced that he should put all of his wealth
46:09
into this thing. He's like, oh yeah, this is
46:11
interesting. We're building, we're experimenting, like great. But
46:14
I also wanna put my money
46:16
somewhere else too for some diversification. So
46:19
this is where Howard Morgan comes in. You
46:22
know, we used to talk about this on old acquired
46:24
episodes that in the early days of Silicon Valley, there
46:26
were only 10 people out here and they all knew
46:28
each other and they were all doing the same thing.
46:32
This was also the case in East
46:34
Coast finance and technology and early VC
46:36
in these days. Howard Morgan
46:38
would go on to be one of the co-founders of
46:41
first round capital. Which was essentially
46:44
spun out of Renaissance. Like it was kind
46:46
of the venture capital work that they were
46:48
doing at Renaissance that didn't fit with the
46:50
rest of Renaissance. Yes. So
46:52
here's how it all went down. And this is
46:55
so poorly understood out there. Yes. Howard
46:58
was a computer science and business
47:00
school professor at the University of
47:02
Pennsylvania. So he taught CS at
47:04
Penn and business at
47:06
Wharton. And he
47:08
had been involved in bringing ARPANET to
47:10
Penn and was kind
47:13
of like early, early internet pioneer.
47:16
And so as a result, he was super
47:18
plugged into tech and
47:21
early startups and really early, early
47:23
proto internet stuff. And
47:26
Jim gets excited about investing
47:28
together with Howard. So they say
47:30
like, hey, maybe we should partner
47:33
together. And in 1982,
47:36
Jim actually winds down monometrics
47:39
and he and Howard co-found a
47:41
new firm together that's gonna
47:43
reflect both of their backgrounds and be
47:46
a great diversification. Jim and
47:48
his group are gonna bring in
47:50
the quantitative trading thing. And
47:53
again, trading on currencies
47:55
and commodities at this point. And
47:57
Howard's gonna bring in private
47:59
companies. company technology investing,
48:02
and they pick a name for
48:04
a firm that is going to reflect this
48:07
Renaissance Technologies. It's crazy. And that
48:10
is why RENTEC is called RENTEC.
48:13
I could not, when we figured this out
48:15
in the research, I could not believe that
48:17
this is not a more widely understood story,
48:19
that this is the origins of what
48:22
is today a fantastic venture capital
48:24
firm, first round capital, but you
48:26
could not name two more different
48:29
strategies in investing. I mean,
48:31
a long-term illiquid thing
48:33
like venture capital, highly
48:35
speculative versus, you know, we're
48:37
going to trade whether we think the French
48:40
franc is going to go up or down
48:42
tomorrow based on the whim of some government
48:44
leader. It's unbelievable these were under the same
48:46
roof. Totally. But
48:48
when you know the whole background in history, it kind
48:50
of makes sense because this is their personal money.
48:53
This is Jim and his buddies, and
48:55
Lenny and James and Howard. There's
48:58
not institutional capital here. They're not
49:00
out pitching LPs of like, oh,
49:02
you should invest in my diversified
49:04
strategy of currency trading and private
49:06
technology startups. Yeah, when they say
49:08
multi-strategy, this is really multi-strategy. Yeah.
49:12
We'll get into what multi-strategy today means later.
49:15
But in these early days of
49:17
RENTEC, 50% of the portfolio was venture
49:19
capital and 50% was
49:21
currency trading. And in fact,
49:24
a couple of years after they get started, the currency
49:26
trading side of the firm almost
49:29
blows up when Lenny goes
49:32
super long on government bonds
49:35
and the market goes against him and the whole portfolio
49:37
drops 40%, which is wild. That
49:42
ends up triggering a clause in Lenny's
49:44
agreement with Jim and they
49:46
sell off Lenny's entire portfolio and he
49:48
leaves the firm. This
49:50
is crazy. Blow-up risk
49:52
is always an issue in the markets,
49:54
but this happened to RENTEC. And
49:57
because we quickly got to this point in the story, it would
49:59
be easy to say. Well, that's a clause that has a lot
50:01
of teeth. There were many sort of
50:03
rumbles of something like this potentially happening. Simon's
50:05
going to Lenny and saying, hey, maybe we
50:07
should cut some of our losses and it's
50:10
okay to trade out of these positions. And
50:12
Lenny was just very dug in on I'm
50:14
a true believer. And that's how you can
50:16
get into a situation where you'd trigger a
50:18
covenant like this. Totally. And again,
50:21
also shows they weren't doing
50:23
model based quantitative trading really at this
50:25
point in time. Now, so much gut.
50:29
So as a result of that, for
50:31
a while, RenTech is truly almost entirely
50:34
a venture capital firm. At
50:36
one point on the
50:38
venture side, just one investment, Franklin dictionaries.
50:41
Do you remember Ben Franklin electronic dictionaries?
50:43
Yeah, that was one of their biggest
50:45
investments. That one investment
50:47
is half of Jim's net worth.
50:50
What? At this low point for the trading side.
50:53
Yes. I had no idea.
50:55
That's crazy. Yeah. In
50:57
the book, Greg talks about, oh, Jim was
50:59
focused on venture capital. And that's kind of
51:01
the story out there. It's like, well,
51:03
he was focused on venture capital because that was the
51:05
only thing working and making money. Well, I
51:07
mean, it's the only thing where they actually had
51:10
an edge from Howard's access to deal flow because
51:12
they certainly didn't have an edge in the global
51:14
currency markets. So I think perhaps
51:17
in part because of the trading losses, James
51:20
Axe starts to get a little dissolution
51:22
too. And he tells Jim
51:24
that he wants to move out to California
51:26
with Sandor Strauss, who started working with them
51:28
at this point. Sandor was another Stony Brook
51:31
alum that joined them. And
51:33
the two of them want to move out to California and
51:35
do trading out there. Jim
51:38
says, sir, fine. I'm here with
51:40
Howard. I'm doing venture capital stuff.
51:43
Why don't you go move out
51:45
to California? You can start your
51:47
own firm, which they do. It's
51:49
called Axecom, A-X-C-O-M. And
51:52
we'll contract with Axecom to run
51:54
what's left of the trading operations
51:56
here for RENTEC. arm's
52:00
length thing where Jim
52:02
strikes a deal where he's gonna own a part
52:04
of Axecom in exchange for
52:06
this very favorable contractual relationship where
52:08
they're gonna hire them to be
52:10
the manager for this pot of
52:13
money that Renaissance has raised. But
52:15
you know, it's technically not Renaissance, it's
52:17
Axecom. Right, it's another company that
52:19
is now doing the quantitative trading. Yep,
52:22
and I think Jim owned a quarter of it,
52:24
is that right? Yes, that's right. And
52:27
importantly, I don't think anyone had
52:29
any idea what Axecom would become
52:31
or how unbelievably profitable
52:34
it would be. No,
52:38
nobody would have done what they did had
52:40
they known what was coming. Yes, wouldn't have
52:42
spun it out. No, so
52:45
once Axe and Strauss get
52:47
out to California, Strauss,
52:50
he's kind of on the computing data
52:52
infrastructure side, that's what he was doing
52:54
at Stony Brook, and that's what he
52:56
came into Renaissance to build. He
52:59
starts getting really into data, and
53:02
he starts collecting intraday
53:04
pricing movements on securities.
53:06
At this point in time, I think
53:09
really the best data you could
53:11
get from providers out there was
53:14
maybe open and close data on
53:16
securities pricing. Strauss finds
53:18
a way to get tick
53:20
data, like every 20 minute
53:23
data on the securities
53:26
throughout the day. Not only that,
53:28
he's getting historical data that predates what
53:30
your traditional data providers would give you,
53:32
and then ingesting it into computers and
53:34
cleaning the data to get it into
53:36
the same format as the tick data.
53:38
So he's getting early 1900s, even 1800s
53:40
stuff to
53:43
try to just say, at some
53:45
point, hopefully we'll be able to make use
53:47
of this, and I wanna have this just
53:49
really, really clean data set about the way
53:51
that these markets interact. Yeah, I
53:53
mean, he's doing ETL on the data.
53:55
Yes. Before anybody knew what ETL
53:58
was. Again, no one told him to do. That
54:00
was just a self-motivated, almost like obsession of like,
54:02
well, if we're going to have data, it should
54:04
be well formatted and well understood and labeled and
54:07
all that. So that's one thing that
54:09
happens. The other thing is
54:11
Jim says, oh, you're going out to California. Let
54:13
me hook you up with my buddy
54:16
who's a Berkeley professor out there, Elwyn
54:18
Burlacamp. And Burlacamp
54:22
had studied with folks like
54:24
John Nash and Claude Shannon
54:26
at MIT. I love that
54:28
Claude Shannon is coming in again. I know. We
54:31
talked about it a lot on the
54:33
Qualcomm episode, Father of Information Theory, really
54:35
the center of gravity for attracting tons
54:37
of talent to MIT and kind of
54:39
paving the way for what would become
54:42
phone technology and telecommunications broadly in
54:44
the future. But the fact that
54:46
Burlacamp is crossing paths at MIT
54:49
with Claude Shannon, so cool. So
54:51
cool. And most importantly
54:54
for this specific case, Burlacamp
54:56
had worked with John Kelly,
54:58
who developed the Kelly criterion on
55:00
bet sizing, which poker players will
55:03
likely be well familiar with. So
55:06
with this combination now of much, much,
55:08
much better and deeper data from Strauss
55:11
and Burlacamp coming in and working with Axe
55:13
on the models and saying, hey,
55:16
we should be smart about the bet sizing that we're
55:18
doing in the trades that are coming out of these
55:20
models. Versus I don't know what they were doing
55:22
before. Maybe it was naive of like
55:25
every trade was the same or just
55:27
like we should actually be systematic about this. The
55:30
models start really working. Yep.
55:32
This is the turning point. Yeah. In
55:36
these kind of mid 80s years,
55:38
Axecom is generating IRRs of like
55:40
20 plus percent on the
55:42
trading side. You know, not necessarily
55:45
going to beat venture capital IRRs, but
55:47
liquid. Yes. Reliable.
55:50
Well, that's the thing. They don't know how reliable
55:52
yet. They know they've done it kind of a
55:54
few years in a row here. But the
55:56
question is how uncorrelated to the stock market over
55:58
a long period of time. and
56:00
how predictable are these returns? Or
56:02
is it just super high variance?
56:05
Yes, but the early results are
56:08
really good. And Jim and
56:10
Berlekamp especially are very
56:12
encouraged by this. So
56:14
in 1988, Jim and
56:17
Howard Morgan decided to spin out
56:19
the venture investments and Howard goes
56:21
to manage those with basically
56:23
their own money. Fun coda
56:25
on this, when Howard
56:27
starts first round a number of years later
56:30
with Josh Koppelman, Jim of course is
56:32
a large LP. And
56:35
Howard of course remains
56:37
an investor in RENTEC. The
56:41
first institutional fund that
56:44
first round ended up raising was
56:47
a 50X on $125 million fund. It
56:50
had Roblox, Uber and Square.
56:53
So I believe this is
56:56
right. I think Jim made as much
56:58
money from his investments in first round
57:00
as Howard did from his LP stake
57:03
in RENTEC. That's wild.
57:06
Isn't that amazing? Wow, that
57:08
is a untold story about Jim
57:10
Simons. I think I read basically
57:12
every primary source thing on Jim
57:15
or Renaissance on the whole internet. But
57:17
I assume you got that from Howard. Yeah, it
57:19
was super fun talking to Howard about this. And
57:21
just the history of how first round started early
57:24
Super Angel Investing and everything that
57:26
became. I also didn't realize that
57:28
first rounds fund one was a 50X on $125
57:31
million fund. First
57:34
institutional fund, which
57:36
I believe they called fund two. I mean,
57:39
wild, wild stuff. Totally
57:41
wild. So when Howard
57:44
spins out the venture activities, Jim
57:47
then decides to set up
57:49
a new fund as
57:51
a joint venture between RENTEC and Axcom. And
57:54
they decide to name it after
57:56
all of the collective mathematical
57:59
award. that Jim
58:01
and James and Burlacamp and all these
58:03
prestigious mathematicians have won in
58:06
their careers. They name it the
58:09
Medallion Fund. Bada-da! And
58:12
listeners, we've arrived. This is the part
58:15
of the story that matters. The Medallion
58:17
Fund is the crown jewel, or you
58:19
might even say actually the only interesting
58:21
thing about Renaissance. And it
58:24
is born out of this observation
58:26
that, oh my god, what they're doing over there at AXCOM
58:28
is really interesting. Maybe they
58:30
shouldn't be doing it all the way over there. Maybe
58:33
that should be a deeper part of the
58:35
fold here at RENTEC, and we shouldn't have
58:37
let that get away, or frankly given up
58:39
on the quantitative trading strategies too early. And
58:42
again, still just currencies, still just
58:44
commodities futures, not playing the stock
58:46
market at all, but
58:49
the seeds and the ideas, the
58:51
huge amount of clean data, the
58:53
robust engineering infrastructure to process all
58:56
that data, the mining
58:58
of signals from data to
59:00
figure out what trading strategies to execute.
59:02
That is really starting to form here
59:05
in this new joint venture, this Medallion
59:07
Fund. Those ideas had
59:10
all existed before. This is the first
59:12
time that it's all brought together and
59:15
actually working and operationalized. And
59:17
frankly, that computers got good enough to actually do it
59:20
too. That's another big piece of this. Yeah,
59:22
I don't know that Strauss could
59:24
have done his data engineering
59:27
too much earlier in time. But
59:30
before we get into the just
59:33
absolutely insane run that this Medallion
59:35
Fund is about to go on,
59:38
that continues right through to this day, now
59:40
is the perfect time for another story
59:42
about ServiceNow. ServiceNow is one
59:45
of our big partners here in Season
59:47
14 and is just an incredible company.
59:49
Yep, ServiceNow digitally transforms your enterprise, helping
59:51
automate processes, improve service delivery, and increase
59:54
operational efficiency all in one intelligent platform.
59:56
Over 85% of the users are in
59:58
the industry. of the Fortune 500 runs
1:00:01
on them, and they have quickly joined the
1:00:03
Microsofts and the NVIDIAs as one of the
1:00:05
most important enterprise software companies in the world
1:00:08
today. So we talked
1:00:10
on our Novo Nordisk episode about
1:00:12
how ServiceNow founder Fred Luddy discovered
1:00:14
this core insight that software can
1:00:16
transform and eliminate manual tasks. And
1:00:19
on Hermes, we told the story of
1:00:21
how current CEO Bill McDermott came in
1:00:23
and turbocharged that into an absolute monster
1:00:25
$150 billion market
1:00:27
cap global behemoth. The
1:00:30
key thread that connects those two eras
1:00:32
is that from day one, Fred knew
1:00:34
the ServiceNow platform could be used across
1:00:36
the whole enterprise. But at the same
1:00:38
time, he also knew from his decades
1:00:40
of prior software experience that launching a
1:00:42
broad horizontal offering right out of the
1:00:44
gate as a startup was a recipe
1:00:46
for failure. You need to
1:00:48
start with a specific vertical use case. And
1:00:50
in this case, he chose IT service management.
1:00:53
Yeah, and that's been true for us here
1:00:55
on Acquired too, David, if we didn't name
1:00:57
it Acquired and cover technology acquisitions that actually
1:00:59
went well, we never could have broadened and
1:01:02
become the podcast that tells the stories of great
1:01:04
companies. You can't just start as that. Totally.
1:01:07
Well, this is what's so cool and where I think the
1:01:09
playbook lesson really is for listeners. Because you
1:01:11
can't just pick any use case, you have
1:01:13
to be strategic about it. And IT was
1:01:15
the perfect vertical because every other department has
1:01:18
to interface with them from the CEO on
1:01:20
down. So they're going to notice when IT
1:01:23
service management rapidly improves, all
1:01:25
of those support tickets that used to
1:01:27
take forever are now just magically resolved.
1:01:29
And that greases the wheels for the
1:01:32
other departments to say, hey, maybe we
1:01:34
should adopt ServiceNow to turbo charge and
1:01:36
digitally transform our service levels too. Yep.
1:01:39
Once those other departments do pull the
1:01:41
trigger on joining the ServiceNow platform, who
1:01:44
is in charge of rolling it out for them? Of
1:01:46
course, it's IT who
1:01:48
are already true ServiceNow believers. I'm
1:01:50
honestly not sure that there's a
1:01:52
better enterprise software playbook in history
1:01:55
than ServiceNow's. So once
1:01:57
they established the beachhead in IT, they then took
1:01:59
the same. platform to HR with employee
1:02:01
experience. They took it to CSM with
1:02:04
customer service requests. They took it to
1:02:06
finance with regulatory reporting, audit and expense
1:02:08
approvals. And now they're adding AI, which
1:02:10
will take everything to the next level.
1:02:13
Yup. So if you want to
1:02:15
learn more about the ServiceNow platform and playbook,
1:02:17
and hear how it can transform your business,
1:02:19
head on over to servicenow.com/acquired. And when you
1:02:22
get in touch, just tell them that Ben
1:02:24
and David sent you. So
1:02:27
they've got this grand new plan and
1:02:29
vision with the Medallion Fund. Unfortunately,
1:02:32
right out
1:02:34
of the gate, the fund stumbles
1:02:36
a bit. And
1:02:38
Axe ends up getting burned out.
1:02:41
Berlacamp though is like, no, no, no,
1:02:43
no, no. This is an anomaly. Like we're going
1:02:45
to fix this. I really, really believe that
1:02:48
what we're doing with these models is going
1:02:50
to be extremely profitable. So
1:02:52
he buys out most of Axe's stake
1:02:55
in the summer of 1989. And he
1:02:57
moves the
1:03:00
offices up to Berkeley. And
1:03:02
there he comes up with the idea
1:03:05
that, hey, we should trade
1:03:08
more frequently, a lot
1:03:10
more frequently. Because if what we're
1:03:12
trying to do is understand the state of the market
1:03:14
from the data we have, and then predict the future
1:03:16
state of the market, and then
1:03:18
combine that with figuring out the right bet
1:03:20
sizing to make, we actually want
1:03:22
to make a lot more trades to get a
1:03:25
lot more data points and learn a lot more
1:03:27
about the bets we're making so that we can
1:03:29
then size them up or size them down. It's
1:03:31
that and it's two other things. One
1:03:34
is the further into the future you
1:03:36
look, the less certain you can be about
1:03:38
it. If you know something is worth $10
1:03:40
right now, what you know
1:03:43
five minutes from now is it's probably gonna be worth
1:03:45
about $10. The most likely
1:03:47
situation is it's within 5% of
1:03:49
that. If you ask me three years from
1:03:51
now, I have almost no intuition about that.
1:03:53
And a state machine is the same way.
1:03:55
If you flash forward a whole bunch of
1:03:57
states, you sort of lose predictability you
1:04:00
sort of continue down that chain. The
1:04:02
second thing is, if your models are showing that
1:04:04
you're going to be right, call it something like
1:04:06
50.25% of the time, then the amount of money
1:04:11
you can make is gated by the
1:04:13
number of bets you can make at
1:04:15
a quarter percent edge. If I walk
1:04:17
up to the casino, and I think
1:04:19
I'm right about this particular roulette wheel,
1:04:21
which of course you're not, 50.25% of
1:04:25
the time, and I decide to play once or
1:04:27
play twice or play five times, there's
1:04:29
a chance I could lose all my money, or if
1:04:31
I have tiny little bet sizes, then I'm just not
1:04:33
going to make that much money. But if I walk
1:04:35
up to said game with a little bit of edge,
1:04:37
and I use small bet sizes, and I played 10,000
1:04:39
times, I'm going to walk out with a
1:04:41
lot of money. There is
1:04:43
a great Bob Mercer quote about this
1:04:46
later. He says, we're right 50.75%
1:04:48
of the time. And I do think he's
1:04:51
making up that number. I think it's illustrative.
1:04:53
Right. But we're 100%
1:04:55
right. 50.75% of the time, you can make billions
1:05:01
that way. It's so true. When
1:05:04
you have that little edge, it's about making sure
1:05:06
that you're not betting so much that a few
1:05:08
bets that don't break your way can take you
1:05:10
down to zero. And to make
1:05:12
sure you can just play the game a
1:05:14
lot, a lot. Yes. And then
1:05:17
back to the Kelly criterion, adjust your bet
1:05:19
sizes over time as you're making those bets.
1:05:22
Now, of course, this is all great in
1:05:24
the abstract. If it's that you're literally sitting
1:05:26
at a casino when you're somehow perfectly making
1:05:28
these bets, and you're just sitting right there
1:05:30
at the table, and then you can walk
1:05:32
over to the cashier, it gets a little
1:05:34
bit different in the market. For example, there
1:05:37
are real transaction costs, especially at this point
1:05:39
in history before some of these more innovative
1:05:41
trading business models with pay for order flow
1:05:43
and zero transaction fees and all this stuff.
1:05:45
There's real transaction costs to putting on these
1:05:47
trades. And of course, you're going to move
1:05:49
the market when you put on these trades.
1:05:51
Yes, this is slippage. There's all sorts
1:05:53
of practical consideration. You could get
1:05:55
front run by other people. It's
1:05:58
not just a computer program that gets executed. executed,
1:06:00
you actually have to meet the constraints of the
1:06:02
real world when you're deciding instead of a few
1:06:04
big bets, we're going to have 100,000 tiny bets.
1:06:08
Yes. And as time goes on,
1:06:10
and the whole quant industry emerges and becomes
1:06:12
much more sophisticated, I think it's
1:06:14
particularly the slippage there that becomes the governor
1:06:16
on how high velocity you can actually be
1:06:19
on this. And the slippage is that
1:06:22
once you are at a certain scale, you are going to
1:06:24
move the market with your trades. So the
1:06:26
deeper you get into the order book, like, let's
1:06:28
say you want to buy $5 million of something,
1:06:30
maybe your first $100,000, you're pretty sure you can
1:06:33
get the quoted price. But
1:06:35
by your last $100,000 of that $5
1:06:37
million buy, the price might have gotten
1:06:39
pretty different already. Yeah,
1:06:41
we're going to come back to this in just a
1:06:43
minute. But this certainly for early rent heck, and
1:06:46
then even now still for all of
1:06:48
quantitative finance is a really, really, really
1:06:50
important thing. Yeah, and
1:06:53
David, in a very crude way, calls back
1:06:55
to last episode on Hermes, the
1:06:57
idea that the price would be highest for
1:06:59
the family member that is willing to sell
1:07:02
now and sort of goes down over time.
1:07:05
If the family was going to sell to Bernard
1:07:07
Arnaud, it would behoove you to be first in
1:07:09
the order book, not last in the order book.
1:07:11
Yes. I feel like there's this meta
1:07:14
lesson that I've been learning through acquired and my
1:07:17
own personal investing over the past couple of years.
1:07:20
Every market is dependent on supply and
1:07:22
demand. You can see
1:07:24
quoted valuations and quoted price streams,
1:07:27
but oftentimes that's like the mistake of just
1:07:29
looking at averages. Exactly. Yes, looking
1:07:31
at the quoted price of an asset
1:07:34
is wrong. You actually should be looking at what
1:07:36
is the volume that is willing to buy and
1:07:38
what is the volume that is willing to sell.
1:07:40
And for all of those buyers and all of
1:07:42
those sellers, what are the price at which they
1:07:44
are willing to transact? And the
1:07:46
way that tends to manifest on a stock chart
1:07:48
is here's the price of the share right now,
1:07:51
but that's not actually what's going
1:07:53
on under the surface. It's a whole bunch of
1:07:55
buyers and sellers who have different willingness to pay
1:07:57
and have different amounts that they're trying to buy.
1:08:00
yourself. Yes. Now,
1:08:02
at this point in time when the Medallion Fund is
1:08:04
first starting to work in, say, late 1989,
1:08:06
early 1990, it's small
1:08:09
enough that this isn't a big consideration
1:08:11
yet. Yeah, right. Medallion
1:08:13
was about $27 million under management
1:08:15
when Burley Camp bought out ACK.
1:08:19
In 1990, the first full year after that, the
1:08:22
fund gains 77.8% gross, which after fees and carry
1:08:29
was 55% net. Now
1:08:33
what were the fees and carry? I mean,
1:08:35
either one of those numbers is shooting
1:08:37
the freaking lights out. Assuming that
1:08:39
this is not a crazy
1:08:41
high risk strategy that they executed and it'll
1:08:43
completely fall apart under different market conditions, like
1:08:46
if this is an actual
1:08:48
repeatable strategy that produces the numbers you
1:08:50
just said, unbelievable,
1:08:52
world changing. Hell
1:08:54
yeah, let's go. And
1:08:57
indeed, it was a hell yeah,
1:08:59
let's go situation. So the
1:09:02
numbers you quoted me, the gross and the net sounded quite
1:09:04
different. Talk to me about the fees and carry. So
1:09:06
carry, I've seen different sources of whether it was 20% or
1:09:08
25% in the early days, but the management
1:09:12
fee on the fund was 5%, which is crazy. The
1:09:16
top venture capital firms in the world charge a
1:09:18
3% management fee, and
1:09:20
even that is like everybody holds their nose and
1:09:22
is like, this is ridiculous. How
1:09:24
on earth were these nobodies charging
1:09:27
a 5% management
1:09:29
fee out the gate to
1:09:31
their investors? Well, a
1:09:34
couple things. One their
1:09:36
investors were not sophisticated. It was mostly their
1:09:38
own money and their buddy's money. So
1:09:40
they set that precedent. They set that precedent. But
1:09:43
two though, they actually needed
1:09:45
the money because Strauss's
1:09:48
infrastructure costs were about $800,000 a year.
1:09:52
So they just backed into the management fee based on like, hey,
1:09:54
we need $800,000 a year to
1:09:57
run the infrastructure. Plus, we need some money to... pay
1:10:00
folks and whatnot like great 5% management fee
1:10:02
and so the pitch they're making the investor
1:10:04
base is like if you believe that we
1:10:07
should be able to massively outperform the market
1:10:09
doing quantitative trading. What we're gonna need
1:10:11
a lot of fees to do that and so the
1:10:13
investors basically took the deal if they thought about it
1:10:16
enough. Okay so that's the
1:10:18
fees on the performance that twenty or
1:10:20
twenty five percent it's just not actually
1:10:22
that far above market if it's above
1:10:24
market at all what you're seeing is
1:10:26
a high fee normal ish performance fee
1:10:28
fund at this point in time. Yes
1:10:30
high management fee normal ish carrier
1:10:33
performance element. Yeah so
1:10:36
at the end of nineteen ninety simons
1:10:38
is so jazzed about what's
1:10:40
going on that he
1:10:42
tells brilacamp hey you
1:10:44
should move here to long island
1:10:47
let's re-centralize everything here. I
1:10:49
want to go all in on this I
1:10:52
think with some tweaks we can be up
1:10:54
eighty percent after fees next year. Really
1:10:57
camp is a little more
1:10:59
circumspect a he wants to stay
1:11:01
in berkeley he doesn't have any desire to move to long
1:11:03
island. And be I couldn't tell
1:11:05
how much of this is just he's a
1:11:07
little more conservative than jim or
1:11:09
how much of this actually might be his whole
1:11:12
poker bet sizing thing. He
1:11:14
turns to jim and he says well if
1:11:16
you're so optimistic why don't you
1:11:19
buy me out so jim
1:11:21
does at six
1:11:23
x the basis that berlacamp
1:11:25
had paid a year earlier.
1:11:28
On the one hand making a six x in
1:11:31
one year sounds great on the other hand this
1:11:33
is the equivalent of when
1:11:36
don valentine sold sequoias apple
1:11:38
steak before the IPO to
1:11:40
lock in a great game.
1:11:43
But miss out on all the upside
1:11:45
to come. David I think
1:11:47
we should throw this out so people understand the
1:11:49
volume of this they've generated on
1:11:52
the order of sixty billion dollars
1:11:54
of performance fees for
1:11:57
the owners of the fund over their
1:11:59
entire life. time. So on
1:12:01
the one hand, 6x in a year
1:12:03
ain't bad. On the other hand, you
1:12:05
owned a giant part of something that
1:12:07
has dividended $60 billion in
1:12:09
cash out to its owners. Yeah,
1:12:13
that's just on the carry side. I mean, the
1:12:15
owners are the principles. So
1:12:17
just like dollars out of the firm,
1:12:19
it's probably twice that. I would
1:12:22
estimate probably $150 to $200 billion
1:12:25
that have come out of Medallion over
1:12:27
the last 35 years. So
1:12:31
Jim buys out Brillicamp.
1:12:34
He rolls everything
1:12:36
in the Medallion fund back into
1:12:38
RENTEC itself, moves everything back
1:12:40
to Stony Brook. Strauss moves to Stony
1:12:42
Brook. So it's now the
1:12:44
Jim Simon show in New York with Strauss
1:12:46
building the engineering systems and Axe, I think,
1:12:49
still had a small stake. Yes,
1:12:51
that's right. And Strauss had a stake as well. So
1:12:54
once Jim takes control and moves
1:12:56
everything back, he
1:12:58
basically decides that
1:13:01
he's going to turn RENTEC
1:13:04
into an even
1:13:07
better, even more
1:13:09
idealized version of
1:13:12
IDA and the math department
1:13:14
at Stony Brook. He's going to make
1:13:16
this an academic's paradise,
1:13:20
where if you are one
1:13:23
of the absolute smartest
1:13:25
mathematicians or systems engineers
1:13:28
in the world, this is
1:13:30
where you want to be. So
1:13:33
of course, he starts reading the
1:13:36
Stony Brook department itself again. And
1:13:38
this is when Henry Laufer joins
1:13:41
full time. Laufer had
1:13:43
been consulting with Medallion in the
1:13:45
early days and working with Brillicamp
1:13:47
as they're doing bet sizing, as
1:13:49
they're making more frequent trades. But
1:13:51
now, once the whole operation has
1:13:54
moved back to Long Island, Laufer is like,
1:13:56
Oh, okay, great. I'll come full time. I'm here at
1:13:58
Stony Brook anyway. This is way more fun. And
1:14:01
listeners, I imagine this is probably the point where you're
1:14:03
starting to get confused and saying there are so many
1:14:06
people in this story. I think we're on eight or
1:14:08
nine. We just keep introducing more people. And
1:14:10
that is the story of
1:14:12
Renaissance. It is not this singular,
1:14:15
clean narrative. It is a very
1:14:19
complex reality of
1:14:21
a whole bunch of different
1:14:23
people that came in and out
1:14:26
at different eras where the firm
1:14:28
was trying different things and eventually
1:14:30
became phenomenally successful with a very
1:14:32
particular approach. But while they were
1:14:34
figuring it out along the way, it took a lot of people.
1:14:36
A lot of people. And just a lot of
1:14:38
time, too. This is 25 years. This
1:14:41
is a quarter century from the
1:14:44
time that Bauman-Simons
1:14:47
write the paper at
1:14:49
IDA until Medallion really
1:14:51
starts to work. It
1:14:53
takes a long time. We haven't
1:14:55
even introduced the two people who
1:14:57
would become the co-CEOs of this
1:14:59
company for 20 years. Yes.
1:15:03
Well, let's get to that. So
1:15:06
Jim Moves Everything Back to Long Island sets it
1:15:09
up as this academic paradise who's recruiting the
1:15:11
smartest people in the world. In
1:15:13
1991, the next year, the firm does 54.3% gross returns and 39.4% net returns
1:15:15
after fees. So
1:15:25
not Jim's bogey of 80%, but
1:15:27
still pretty freaking great. And
1:15:29
we should say the years of modest
1:15:31
performance are behind them. From every single year
1:15:34
forward, they shoot the lights out. From 1990
1:15:36
onward, they never lose money. And
1:15:41
on a gross basis, they never even do less
1:15:43
than 30%. It's
1:15:46
working. It's going. The whole rest
1:15:48
of the story is about hold
1:15:50
on, keep the machine working, and
1:15:52
we're on the train. The
1:15:55
historic run has begun, let's just
1:15:57
say. Yep. So... Nineteen
1:16:00
Eighty Two, Grace Returns Or Forty
1:16:02
Seven percent. Ninety. Three Their
1:16:04
fifty four percent. At.
1:16:07
The End of Ninety Ninety Three. Simon's.
1:16:09
Decides to close the find and not
1:16:11
allowed new Lps. and so if you're
1:16:14
an existing Lp, you can stay and
1:16:16
but they're no longer open for new
1:16:18
inflows the Us. So much confidence in
1:16:21
what they're doing. That. He thinks
1:16:23
they're all going to make more money. Without.
1:16:25
Accepting new capital by just keeping it to
1:16:27
the existing investor base. Maybe. Ninety
1:16:29
Four Gross returns are
1:16:32
Ninety Three Freaking percent.
1:16:35
Medallion. At this
1:16:37
point is stacking up cast it
1:16:39
is a. Meeting for
1:16:41
fun. It's about two hundred and
1:16:43
fifty million dollars total at this
1:16:45
point in time, which is small,
1:16:47
but we're talking about Nineteen Ninety
1:16:50
Four with a bunch of outsiders
1:16:52
and academics that have managed to
1:16:54
amass a quarter billion dollars here.
1:16:56
People. States pay attention. And.
1:16:58
The performances on this are. Seven.
1:17:01
Million Dollars. Thirteen Million Dollars. Fifty
1:17:03
Two Million Dollars. The free cash
1:17:06
flow flowing to partners here is
1:17:08
certainly becoming real to. Just.
1:17:11
But. As they get into that. Call.
1:17:13
It on the order of magnitude of a
1:17:15
billion dollars scale. They
1:17:17
start bumping into the Moving Markets
1:17:20
problem and the Slip. It's. That.
1:17:22
we were talking about earlier yup, and that sort of
1:17:25
in the mid nineties. Yup, As they're
1:17:27
hitting this two hundred fifty million half a
1:17:29
billion dollar scale right, the computer model spits
1:17:31
out we should go by this huge amount
1:17:33
of something at this price they go to
1:17:35
do with they can only buy ten twenty
1:17:37
thirty percent of the amount they want at
1:17:39
that price And and suddenly the price is
1:17:41
very different. Yeah up. To
1:17:43
this point. The. Vast majority of
1:17:45
what. Medallion. Is doing. Is.
1:17:48
Trading currencies and commodities.
1:17:51
Not. Equities. Because.
1:17:53
He might be thinking. Okay, yeah, I
1:17:55
hear you. The nineties was a different
1:17:57
era, but. Half. a billion dollars fund
1:18:00
doesn't sound that big. How are they moving
1:18:02
markets with half a billion dollars? It's
1:18:04
not the equity markets. It's because
1:18:06
they're in these thinner markets. It's
1:18:08
not that commodities and futures are
1:18:10
small markets. They're large, but they're
1:18:12
thin compared to equities. There's just not
1:18:15
that much volume and you just can't trade
1:18:17
that much without slipage becoming a huge issue.
1:18:19
And Medallion is now hitting that limit. So,
1:18:23
Simon's decides the
1:18:25
only thing we can do here to expand,
1:18:27
which I'm such a believer in what we're
1:18:29
doing, we need to expand, is
1:18:32
we need to move into equities. Equities
1:18:34
are the holy grail. If we can make
1:18:36
this work there, the depth
1:18:38
in those markets will let us scale
1:18:41
way, way, way bigger than we are now. And
1:18:44
there's so much more data about
1:18:47
equities pricing that we can feed into
1:18:49
our models and the signal processing that
1:18:51
we can do and the signals that
1:18:53
we can find are going to be even better. Right.
1:18:56
There's so many buyers and sellers every
1:18:58
day showing up to trade so many
1:19:00
different companies at such high velocity. It's
1:19:03
almost this honeypot for
1:19:05
Renaissance's systems. This is sort
1:19:07
of their moment. This is what they were built
1:19:09
for. And it's kind of funny that they've just
1:19:11
been in kid glove land the whole time with
1:19:13
these thinly traded markets with minimal data. Yes.
1:19:16
And this brings us to Peter Brown
1:19:19
and Bob Mercer. And in
1:19:21
1993, one of the mathematicians that
1:19:25
Jim had recruited to RENTEC, a guy
1:19:27
named Nick Patterson, gets
1:19:29
especially passionate about going out and recruiting new
1:19:32
talent along with Jim. And this is, I
1:19:34
think, one of the keys to RENTEC and
1:19:36
the culture there. People want
1:19:38
other smart people to come be
1:19:40
there too. Nick's sitting there like,
1:19:43
this is a joy. I want to go
1:19:45
find other best people in the world to
1:19:47
hang out with. And he had
1:19:49
read in the newspaper that IBM was
1:19:51
going through cost cutting and was about
1:19:53
to do layoffs. And
1:19:55
he also knew that the speech recognition
1:19:58
group at IBM had some them absolutely
1:20:01
fantastic mathematical talent.
1:20:05
Really, what they were doing was
1:20:07
again another vector
1:20:09
in the early AI machine
1:20:11
learning research. Specifically,
1:20:14
IBM's Deep Blue Chess
1:20:16
Project of the time had come
1:20:18
out of this group. Peter
1:20:21
Brown there was the one that actually
1:20:23
spearheaded the project. Yeah, and
1:20:26
it's interesting that you talk about
1:20:28
speech recognition as the perfect
1:20:31
fit for what they were doing. And you might say,
1:20:33
why is that? Well, the actual
1:20:35
work that goes into speech
1:20:38
recognition, natural language processing is
1:20:40
kind of the same signal processing
1:20:42
that Renaissance is doing to analyze
1:20:44
the market. It's not just kind
1:20:46
of, it's exactly the same signal
1:20:48
processing. Right, speech recognition is a
1:20:50
hidden Markov process, where the computer
1:20:53
that's listening to the sounds to
1:20:55
try to turn it into language
1:20:57
doesn't actually know English, right, obviously.
1:20:59
But what it does know is
1:21:01
when I hear this set of
1:21:03
frequencies and tonalities and sounds, there's
1:21:05
a limited set of likely things
1:21:07
that could come after it. And
1:21:09
in Greg's book, he greatly points out
1:21:11
this perfect example. When I say Apple,
1:21:13
you might say Pi. The
1:21:16
probability that Pi is going to be
1:21:18
the next word following Apple is significantly
1:21:20
higher. And so these people who have
1:21:22
spent their careers not only doing the
1:21:24
math and the theoretical computer science behind
1:21:26
speech recognition to help figure out and
1:21:28
predict the next words that you have
1:21:30
a narrow set of likely words to
1:21:32
choose from. So when you're listening to
1:21:34
those frequencies, you can say, it's
1:21:37
probably going to be one of these three rather than
1:21:39
search the entire dictionary for any word that it could
1:21:41
be to narrow the processing power. It's
1:21:44
not only the theoretical side, but
1:21:46
it's also people who have built
1:21:48
those systems at IBM, like a
1:21:50
real operational computer company. Yes, at
1:21:53
operational scale. And this is what's so
1:21:55
important and why the two of them
1:21:57
become Probably the most critical.
1:22:00
The hires. In. Rent
1:22:02
Accessory even including all the great academics
1:22:04
the game before them. Because.
1:22:06
They're good on the mass sides. But. They
1:22:09
have this large systems experience.
1:22:12
And. Gym and Nick know that
1:22:14
if they're going to move into equities. Because.
1:22:16
Of the volume of data and because of how
1:22:18
much more complex that market is. They. Need
1:22:21
more complex systems?
1:22:24
And. The current talent at rents are coming from
1:22:26
academia is just never experienced that are built anything
1:22:28
like it. And the world that
1:22:30
they're entering, his just exploding
1:22:33
in complexity and dimensionality. And.
1:22:35
When I say that, here's what I mean. The. Data
1:22:37
that they are mining that there
1:22:39
are for is this intraday tic
1:22:42
data between every stock trading. So.
1:22:45
They're in this sort of trying to map the
1:22:47
relationship between one stock and every other stock. Not
1:22:49
just about Mon and time, but every time before
1:22:51
it, in, every time after it. They.
1:22:53
Are also once they do identify patterns
1:22:56
which this is Keith the algorithms identify
1:22:58
the patterns is not a human with
1:23:00
a hunch saying I think when. Oil
1:23:03
prices go up to the airline. Prices are
1:23:05
going to get his. It's computers doing machine
1:23:08
learning to discover the patterns in the data.
1:23:10
Then. There's the second piece of wool. What
1:23:12
trades do you actually put on? To.
1:23:16
Be profitable from. The
1:23:18
probabilities that you just discovered all
1:23:20
these weights of relationships between all
1:23:22
of these different companies? Yeah, just
1:23:24
putting on one trade. You're putting
1:23:26
on ten. A hundred thousands of
1:23:28
simultaneous trades. Both two heads to
1:23:31
be able to isolate some particular
1:23:33
variable that you're looking for. Again,
1:23:35
not you been a computer is
1:23:37
looking for and you also need
1:23:39
to do it in such specific
1:23:41
bite sizes so that you don't
1:23:43
move the market. So you're looking
1:23:45
for a super multi, very yet
1:23:47
multi dimensional problem. both on the data
1:23:49
in just inside he and on the
1:23:51
how do i actually react to it
1:23:53
side and all of this computation can't
1:23:55
take a long time because you must
1:23:57
act you know not in milliseconds it's
1:23:59
not a high frequency trading that's front running
1:24:02
the market. That's not actually what they do. A
1:24:04
lot of people think it is, but we'll get
1:24:06
to that later. But they do need to act
1:24:08
with reasonable quickness, probably on the order of minutes.
1:24:11
So these need to be really efficient computer
1:24:13
systems too. Yeah. And
1:24:15
the universe of equities is so much
1:24:19
more multidimensional and interrelated. There
1:24:21
are only so many currencies in the
1:24:23
world, and there are especially only so
1:24:25
many currencies that are large enough trading
1:24:27
markets that you can operate. There's
1:24:30
not infinite, but thousands and thousands of
1:24:32
equities in the world that are deep
1:24:35
enough markets that you can operate in.
1:24:37
And to some degree, they're all correlated
1:24:39
with one another. And just
1:24:41
keep adding layers of complexity here. Keep
1:24:43
adding new things to
1:24:45
multiply by. Many of these are traded
1:24:48
on multiple exchanges. So you might also
1:24:50
be looking for pricing disparities on the
1:24:52
same equity on different markets at different
1:24:54
points in time. So there's just dimensions
1:24:57
upon dimensions of things to analyze, correlate,
1:24:59
and act upon. So
1:25:02
Patterson and Simons go raid
1:25:04
IBM. They're like Steve
1:25:06
Jobs, raiding Xerox, Burke. They
1:25:08
bring Peter and
1:25:10
Bob and one of their programming
1:25:13
colleagues, David Magerman, over from
1:25:15
IBM into RENTEC. And
1:25:17
they get started on building the equities
1:25:19
model. But it turns out, A,
1:25:23
they're obviously very successful at that,
1:25:26
but the impact that they have and what
1:25:28
they build is even bigger because
1:25:31
Bob and Peter realize
1:25:34
that pay actually, we
1:25:37
should just have one model for everything
1:25:39
here, for currencies,
1:25:42
for commodities, for equities.
1:25:45
Everything is correlated. Everything is a
1:25:47
signal. It's not like the equities
1:25:49
market is wholly independent
1:25:51
and separate from what's
1:25:53
happening in currencies or what's happening
1:25:55
in commodities. There are relationships everywhere.
1:25:59
We really. want just one model. This
1:26:02
is like a fantastical undertaking, especially
1:26:04
in the early to mid-90s. Right.
1:26:07
But if you can nail it, it means
1:26:09
that you can do interesting things like, hey,
1:26:12
we don't have a lot of data on this
1:26:15
particular market, but it looks
1:26:17
a lot like something we do have data on.
1:26:19
So if it's all part of the same model,
1:26:22
we can kind of just apply all the learnings
1:26:24
from this other thing onto this brand
1:26:26
new thing that we're looking at with little data
1:26:28
for the first time. And because we're putting it
1:26:30
all in one model and no one else in
1:26:32
the world is, we can discover patterns that no
1:26:34
one else knows about. It turns
1:26:37
out that this was actually the second
1:26:39
most important innovation that
1:26:41
Bob and Peter bring to RENTEC, the
1:26:43
actual product and performance of having one
1:26:46
model. The most important
1:26:48
thing is that if you
1:26:50
have only one model, one
1:26:52
infrastructure, everybody in
1:26:55
the firm is working on that
1:26:57
same model. You can
1:26:59
all collaborate all together,
1:27:02
which is especially important when you have the
1:27:05
smartest people in the entire world, all
1:27:07
in one building. Before this,
1:27:09
there were separate models within RENTEC.
1:27:12
So insights and innovations
1:27:14
and work that one team was doing
1:27:16
on one model wouldn't
1:27:18
get applied or translate over to
1:27:21
work that was happening by another team on another
1:27:23
model. They did have the cultural element where it
1:27:25
was encouraged that you share your learnings, but someone
1:27:28
would have to take the time during their lunch
1:27:30
break and go learn from you about those and
1:27:32
then implement it in their version. There's a lag
1:27:34
and it may actually not get implemented. Yeah,
1:27:37
this is wholly
1:27:40
unique and revolutionary. No
1:27:43
other at scale
1:27:46
investment firm, period,
1:27:48
and especially QuantFirm, operates this way
1:27:50
today with just one model. Their
1:27:53
portfolio managers and teams and
1:27:55
multi-strategy, people are culturally
1:27:57
competitive with one another, but even if they're not,
1:28:00
work that you're doing on this side of Citadel
1:28:02
is not impacting the work that you're doing on
1:28:04
that side of Citadel. What Bob
1:28:06
and Peter do is they unify everything
1:28:08
at RENTEC. So all the wood is
1:28:11
going behind one arrow. And
1:28:14
before we talk about the impact of
1:28:16
that, we want to thank
1:28:18
our longtime friend of the show,
1:28:21
Vanta, the leading trust management platform.
1:28:24
Vanta, of course, automates your security
1:28:26
reviews and compliance efforts. So frameworks like
1:28:28
SOC 2, ISO 27001, GDPR, and HIPAA
1:28:30
compliance and monitoring, which is quite topical
1:28:33
if you are in the heavily regulated
1:28:35
finance industry and you need a
1:28:37
lot of security and compliance. Vanta takes
1:28:39
care of these otherwise incredibly time and
1:28:42
resource draining efforts for your organization and
1:28:44
makes them fast and simple. Yep,
1:28:46
Vanta is the perfect example of the quote
1:28:49
that we talk about all the time here
1:28:51
on Acquired. Jeff Bezos, his idea that a
1:28:53
company should only focus on what actually makes
1:28:56
your beer taste better, i.e. spend your time
1:28:58
and resources only on what's actually going to
1:29:00
move the needle for your product and your
1:29:02
customers and outsource everything else that doesn't. In
1:29:05
RENTEC's case, this would be the model.
1:29:08
Every company needs compliance and trust with their
1:29:10
vendors and customers. It plays a major role
1:29:13
in enabling revenue because customers and partners demand
1:29:15
it, but yet it adds zero flavor to
1:29:17
your actual product. Vanta takes care
1:29:19
of all of it for you. No
1:29:21
more spreadsheets, no fragmented tools, no manual
1:29:23
reviews to cobble together your security and
1:29:26
compliance requirements. It is one single software
1:29:28
pane of glass, just like one model,
1:29:30
that connects to all of your services
1:29:32
via APIs and eliminates countless hours of
1:29:34
work for your organization. There are now
1:29:36
AI capabilities to make this even more
1:29:38
powerful and they even integrate with over
1:29:40
300 external tools, plus they
1:29:42
let customers build private integrations with
1:29:44
their internal systems. And perhaps
1:29:46
most importantly, your security reviews are now
1:29:49
real-time instead of static, so you can monitor
1:29:51
and share with your customers and partners to
1:29:53
give them added confidence. So whether you're a
1:29:55
startup or a large enterprise and your company
1:29:57
is ready to automate compliance and streamline security
1:30:00
reviews like Vanta's 7,000 customers around
1:30:02
the globe and go back
1:30:04
to making your beer taste better, head on
1:30:06
over to vanta.com/acquired and just tell them that
1:30:08
Ben and David sent you. And
1:30:10
thanks to friend of the show, Christina,
1:30:12
Vanta CEO, all acquired listeners get $1,000
1:30:15
of free credit,
1:30:18
vanta.com/acquired. So
1:30:20
David, the equities machine. Yes.
1:30:23
And indeed, a machine it is.
1:30:26
So Peter and Bob come in in 1993, and 1994, 1995, they're building
1:30:31
this, RenTech is getting into equities. And
1:30:34
yet just imagine the computers that you
1:30:36
were using during 1994 and 1995. It
1:30:38
is astonishing the
1:30:41
level of computational complexity and
1:30:43
coordination and results that they
1:30:45
are pulling off, again,
1:30:48
in real time analyzing these markets with
1:30:50
the technology that was available during those
1:30:52
years. Yes. And here's
1:30:54
what's amazing returns
1:30:56
go down maybe slightly, certainly a bit
1:30:58
from the blowout year that 1994 was,
1:31:00
but they're still above 30% every single
1:31:03
year, most years above 40%.
1:31:09
This is unbelievable that they're
1:31:11
maintaining this performance as
1:31:13
they're going into this hugely more
1:31:15
complex market, and they're scaling assets
1:31:18
under management. So by the
1:31:20
end of the 1990s, Medallion has almost
1:31:22
$2 billion in assets under management, while
1:31:25
maintaining roughly the
1:31:27
same performance by getting into equities.
1:31:30
This is huge. Yep.
1:31:33
And David, if you just kind of look at this and do the math,
1:31:35
okay, so 94, their AUM was 276 million,
1:31:41
and they grew 93%. And then their AUM the next year
1:31:43
was 462 million, and then they grew
1:31:48
52%. And their AUM the
1:31:50
next year was 637 million, you kind of
1:31:52
quickly get where I'm going here, which is, Oh,
1:31:54
they're scaling AUM Not by bringing in
1:31:56
new investors, right? It's close to new
1:31:58
investors. It's all just. Compounding. This.
1:32:01
Is the same capital that they had
1:32:03
in Nineteen Ninety Three that has gone
1:32:05
from a hundred and twenty two million
1:32:07
at the beginning of that year, to
1:32:09
Ninety Ninety Nine being one point five
1:32:11
billion. Just. And. Then.
1:32:14
In the year two thousand. They.
1:32:16
Just totally blow the doors off.
1:32:19
A hundred and twenty eight
1:32:21
percent. Gross returns.
1:32:24
Net. Returns after fees.
1:32:27
Of. Ninety Eight Point Five Percent
1:32:29
This. but man, they grow
1:32:31
the fund from one point
1:32:33
nine billion to three point
1:32:35
eight billion of assets under
1:32:37
management against purely by investing
1:32:40
gains, not by getting any
1:32:42
new investors. The. Year
1:32:44
the tech bubble burst. Just.
1:32:46
A while the whole rest of the market. Is.
1:32:49
Down big time. Medallion.
1:32:52
Is up a hundred and twenty eight
1:32:54
percent girls on the air and this
1:32:56
becomes a team. High volatility is when
1:32:58
medallion of really science and here you
1:33:00
go. On. Correlate, they have
1:33:03
their final stamp of approval right
1:33:05
here of not only are we
1:33:07
have money printing machine, we are
1:33:09
a money printing machine in all
1:33:11
environments regardless of the state of
1:33:13
the broad market. And save it.
1:33:15
As he said, volatility actually makes
1:33:17
their algorithms work even better because
1:33:19
what are they doing? There are
1:33:21
for scenarios where the market's gonna
1:33:23
act a radically and they can
1:33:25
take advantage of people making decisions
1:33:27
that they send and time any
1:33:29
investors are under pressure. There's. A little
1:33:32
bit of edge that can accrue to a medallion
1:33:34
that saying okay your fear selling right now? Well
1:33:36
I can determine if you should be for you're
1:33:38
selling or not and if I determine that you
1:33:40
shouldn't be dumping net asset and buy it from
1:33:43
you. So there's a really fun
1:33:45
story around this that. Really? Oh
1:33:47
streets. Gyms. Genius
1:33:49
in managing the from and
1:33:51
the people. And.
1:33:54
How. This. Year was when they
1:33:56
really figured this out. So. the
1:33:58
first couple days of
1:34:00
the tech bubble bursting, Medallion
1:34:03
actually takes a bunch
1:34:05
of large losses. And part of
1:34:07
it might be that the model wasn't tuned
1:34:09
right yet because nobody at RENTEC had seen
1:34:11
this type of behavior in the market before.
1:34:14
Part of it might also be too that it didn't
1:34:16
perform well for those couple of days. It's
1:34:18
a really stressful time for everybody.
1:34:21
You know, everybody's in Jim's office, Jim's
1:34:23
smoking his cigarettes, it's a cloud of
1:34:25
smoke, and they're debating what
1:34:27
to do. And Jim makes the call to take some
1:34:30
risk off. He's worried about blowing
1:34:32
up. We're not very far removed at this
1:34:34
point from long-term capital management. The
1:34:36
model may be saying we should stay long here,
1:34:38
but let's not blow up the firm. Yep.
1:34:42
After this goes down, Peter
1:34:44
Brown comes to Jim and offers to
1:34:46
resign, given the losses that they
1:34:49
incurred over these couple of days. And
1:34:51
Jim says, what are you talking about? Of
1:34:54
course you shouldn't resign. You are way more
1:34:56
valuable to the firm now that
1:34:58
you've lived through this, and you now know not
1:35:01
to 100% trust the model in
1:35:03
all situations. Fascinating. It's such
1:35:05
a good insight. That illustrates Jim
1:35:07
as a leader right there. It
1:35:10
totally does. There's a parallel
1:35:12
story when Jim ultimately does
1:35:14
retire in 2009, and
1:35:17
Peter and Bob take over as co-CEOs, where
1:35:20
a year or so before the quote unquote
1:35:22
quant quake had happened, where similar to
1:35:24
the tech bubble bursting, there was all
1:35:26
of a sudden very large
1:35:28
drawdowns about all quantitative firms in the
1:35:31
market and rent tech gets hit. And
1:35:34
during that period, Peter
1:35:37
argued very strenuously that we should trust
1:35:39
the model, stay risk on. This is
1:35:41
gonna be an incredibly profitable time for
1:35:43
us. And Jim pumped
1:35:46
the brakes and stepped in, intervened and took
1:35:48
risk off. And Peter
1:35:50
goes to Jim again around the
1:35:52
CEO transition and says, hey Jim,
1:35:55
Aren't you worried that with me running the place now,
1:35:57
I'm gonna be too aggressive and blow it up. one
1:35:59
of the. The Is. And. Gym says
1:36:01
no, I'm not worried at all. I.
1:36:04
Know you were only so aggressive in that
1:36:06
moment because I was there pushing back on
1:36:08
you. and when you're in the seat. You're.
1:36:11
Gonna be less aggressive resist such a master
1:36:13
it insight into human behavior. It is so
1:36:15
true though I don't find this about myself
1:36:17
that I was naturally take the position of
1:36:19
the foil for the person across from heat.
1:36:21
So if somebody is being pushy and some
1:36:24
way I'll find myself taking a position where
1:36:26
if I read, pause and reflect and like.
1:36:28
I don't think I expected to take this
1:36:31
position coming into this conversation, but in a
1:36:33
you naturally want to sort of play the
1:36:35
other side to balance out the person sitting
1:36:37
across from you. So
1:36:40
backs. Of year Two thousand and this
1:36:42
incredible performance. And to what you
1:36:44
were saying earlier about uncorrelated returns, Not only
1:36:46
do they suit the lights out that year,
1:36:48
they're doing it when the market is down.
1:36:51
We. Got to introduce this concept of a surprise
1:36:53
you have now with for all of you listeners.
1:36:56
Better in the finance world you'll notice but for
1:36:58
everybody else this is a really important concept. And.
1:37:00
I think people grasp at intuitively.
1:37:02
We've mentioned this concept a couple
1:37:04
times this episode where. Okay,
1:37:06
great. It's amazing to have a fund
1:37:09
that twenty five Axis or a year
1:37:11
where you have a hundred percent investment
1:37:13
returns. or I bought Bitcoin yesterday and
1:37:15
it doubled over night. Does that make
1:37:17
you one of the best investors in
1:37:20
the world? We all intuitively. No, No,
1:37:22
it doesn't because. Maybe. That
1:37:24
was a fluke. Maybe you're taking on an
1:37:26
extreme amount of risk and then the question
1:37:28
is always. Adjusting. For the
1:37:31
risk that you're taking, can you produce
1:37:33
a superior return? Taking. The
1:37:35
risk into that accounts and so you
1:37:37
basically can provide value to investors. As
1:37:39
a fund manager in two ways: you
1:37:41
can outperform the market, or you can
1:37:43
be entirely on correlated with the market
1:37:45
and get market returns. Or. What
1:37:47
you can do as Rent Tech
1:37:49
is both. You can be uncorrelated,
1:37:51
he and massively outperform, which is
1:37:53
effectively the holy grail of money
1:37:55
management. Just. And so the surprise
1:37:57
you is a measurement. combining
1:38:00
these two concepts. Exactly. So it's named after
1:38:02
the economist William F. Sharp, it was pioneered
1:38:04
in 1966. It is effectively
1:38:07
the measure of a fund's performance
1:38:09
relative to the risk free rate.
1:38:11
So if you performed at 15%
1:38:13
that year, and the
1:38:16
risk free rate was 3%, then you
1:38:18
know, your numerator is going to be
1:38:20
12%. And it is compared against the
1:38:23
volatility or the standard deviation is technically
1:38:25
what it is. But effectively, how
1:38:28
volatile have you been the last x
1:38:30
years and typically it's looked at as a three year sharp
1:38:32
or a five year sharp or a 10 year sharp. The
1:38:35
sharp ratio represents the additional amount
1:38:37
of return that an investor receives
1:38:39
per unit of an increase in
1:38:41
risk. And so David, you're starting
1:38:43
to throw out numbers, low
1:38:46
sharp ratios are bad, negative sharp ratios are
1:38:48
worth because that means you're underperforming the risk
1:38:50
free rate, high sharp ratios are good because
1:38:52
it means that you're producing lots of returns,
1:38:55
and your variance or your standard deviation or
1:38:57
your sort of risk is low. So
1:38:59
in 1990,
1:39:01
they had a sharp of 2.0, which
1:39:03
was twice that of the S&P 500
1:39:06
benchmark. Awesome. Yep. Good. 1995 to 2000 sharp ratio
1:39:10
of 2.5 really starting to
1:39:12
hum pretty unbelievable. Good. Where
1:39:15
do I sign up to invest? At some
1:39:17
point, they added foreign markets and achieved a
1:39:19
sharp ratio of 6.3, which is double the
1:39:23
best quant firms. This is
1:39:25
a firm that has almost no chance
1:39:27
of losing money, at least historically, and
1:39:30
massively outperforms the market on an
1:39:33
uncorrelated basis. And I believe
1:39:36
if I have my research right in
1:39:38
2004, they actually achieved
1:39:41
a sharp ratio of 7.5. Astonishing. You
1:39:43
know, again, back
1:39:45
to our sports analogy here, these aren't Hall
1:39:48
of Fame numbers. These are like, I
1:39:50
don't know, make Tom Brady look like a
1:39:52
third stringer. Yes, exactly. So
1:39:55
on the back of 2000, and this rise,
1:39:57
the Next year in 2001. They.
1:40:00
Raise the carried interest on the fund
1:40:03
to thirty six percent, up from either
1:40:05
twenty or twenty five percent Whatever it
1:40:07
was before. Now. Remember the
1:40:09
party, close the fund. To. New
1:40:11
investors so they're still outside investors in
1:40:13
the fund. But. No new investors
1:40:16
are coming in. And. Then
1:40:18
the next year in two thousand and two.
1:40:20
They. Raise The carried a
1:40:22
Forty four. Percent.
1:40:25
I. Mean. Great. Work if you can
1:40:27
get it. I'd for contacts the Sequoias the
1:40:29
benchmarks out there. They. Have obscene
1:40:31
carry of thirty percent. Forty Four is impressed,
1:40:34
and there's to interesting ways to look at
1:40:36
this one. They're just trying to jack it
1:40:38
up so high that they just purge their
1:40:40
existing investors out where they're saying we're not
1:40:42
going to kick anyone out. Yep, else. We've
1:40:44
been close to new business for a long
1:40:46
time now. You should see yourself out at
1:40:49
some point. The. Other way to look at this
1:40:51
which I think is probably the right way to look at it
1:40:53
is. Investors. Are
1:40:56
arbitragers. They. See a
1:40:58
mispricing. They come into the markets. The
1:41:00
six that mispricing. So anytime that there's
1:41:02
an opportunity to bring the way that
1:41:04
a currency is treating on two different
1:41:07
exchanges closer together, investors are serving their
1:41:09
purpose of coming in, arbitraging that difference,
1:41:11
taking a little bit of profit as
1:41:13
a thank you and then sort of
1:41:15
fixing the market to make the market
1:41:18
a true weighing machine, not a voting
1:41:20
machine, but making it so that all
1:41:22
prices reflect the value of what something
1:41:24
is actually worse. And in
1:41:26
some ways. That's. What Renaissance is doing
1:41:28
here. To themselves or to their investors. They're coming
1:41:31
and saying, look, This is obscene. We
1:41:33
so clearly outperform the markets the you're
1:41:35
still gonna take this deal even if
1:41:37
we take more of this because her
1:41:39
sister misplacing here. This product should not
1:41:41
be priced at twenty twenty five percent
1:41:43
Carry This products should be priced at
1:41:45
a much higher carried interest and you're
1:41:47
still gonna love it. You. should
1:41:49
pay twenty percent terry for
1:41:52
affirmed that delivers you fifteen
1:41:54
percent annual returns were delivering
1:41:56
you fifty percent annual returns
1:41:58
totally so i to imagine it
1:42:00
didn't go over well with the existing investors, but
1:42:02
they just have so much leverage that what's going
1:42:04
to happen. Okay, once again,
1:42:07
I'm sorry audience, I have to say hold
1:42:09
on one more minute for another perspective that
1:42:12
I have to offer on the
1:42:14
carry element, but I want to finish the
1:42:16
story first. Okay, so 2001, they raised
1:42:18
the carry to 36%. 2002, they raised
1:42:20
it to 44%. And
1:42:23
then in 2003, they actually say,
1:42:25
hey, we can't incentivize you out of
1:42:27
the fund outside investors. They can't
1:42:30
take you out. So starting in 2003, everybody
1:42:32
who's an outside investor who's not part
1:42:34
of the rent tech family, you know,
1:42:36
current employee or alumni of the firm,
1:42:40
kicked out. And not all alumni get
1:42:42
to stay there select alumni that get grandfathered
1:42:44
in. Yes. Now, why
1:42:46
did we do this? I'm going to talk about one reason in a
1:42:48
minute. But one reason is super obvious.
1:42:51
The medallion fund is now at $5 billion
1:42:53
in assets under management that they're trading. Even
1:42:57
in the equities market, they are now hitting
1:42:59
up against slippage. And
1:43:01
so if they want to maintain this
1:43:04
crazy, crazy performance, they just can't get
1:43:06
that much bigger. This is
1:43:09
the problem that Warren Buffett talks about all the
1:43:11
time and why he has to basically just increase
1:43:13
his position in Apple rather than going and buying
1:43:15
the next great family owned business. The
1:43:17
things that move the needle for them are so
1:43:20
big that that's really all they can do. And
1:43:22
when you are big, you're going to move any
1:43:24
market that you enter into. And
1:43:27
the strategy that RENTEC is employing right
1:43:29
now, they're just deeming doesn't work at
1:43:31
north of $5 billion. So
1:43:34
in 2003, they start kicking all the outside
1:43:37
investors out of medallion. But
1:43:39
clearly, there's still lots of
1:43:41
institutional demand to invest with
1:43:43
Renaissance. So what do they do? Well,
1:43:46
time to start another fund. So
1:43:49
they start the Renaissance Institutional
1:43:51
Equities Fund. And there's
1:43:54
a couple of things to add a little bit of
1:43:56
context to really why they decide to do this. Well,
1:43:58
the first one is... Sometimes there's just
1:44:00
more profitable strategies than they had the capital
1:44:03
to take advantage of in medallion, but they
1:44:05
weren't sure it would be on a durable
1:44:07
basis. If they were sure that they could
1:44:09
manage 10, 15, 20, 25 billion in medallion
1:44:11
all the time, then
1:44:15
they would grow to that. But if just sometimes
1:44:17
there's these strategies that appear, well, we don't want
1:44:19
to commit to a much higher fund size and
1:44:21
then not always have those strategies available. The
1:44:24
other thing is that a lot
1:44:26
of the times those strategies aren't really
1:44:28
what medallion is set up to do.
1:44:31
They require longer hold times. And
1:44:33
so there's a little bit of downside to
1:44:35
that because these new strategies, the predictive abilities
1:44:38
are less because they have to predict further
1:44:40
into the future to understand what the exit
1:44:42
prices will be on these longer term holds.
1:44:44
But they still figure, hey, even
1:44:47
though it's not quite our bread and butter with the
1:44:49
short term stuff, we should be able to make some
1:44:51
money doing it. Yeah, there's a
1:44:53
fun story around this that Peter Brown tells
1:44:56
of Jim came into his office
1:44:58
one day and said, Peter, I
1:45:00
got a thought exercise for you. If
1:45:02
you married a Rockefeller, would
1:45:04
you advise the family that they should
1:45:07
invest a large portion of their wealth
1:45:09
in the S&P 500? And
1:45:11
Peter says, no, of course not. That's not
1:45:14
a great risk adjusted return. And
1:45:16
these guys are very used to sharp ratios that
1:45:18
are far better than the S&P. Right.
1:45:21
And so Jim says, yes, exactly. Now
1:45:24
get to work on designing the product that they should
1:45:26
invest in. Right. And that's
1:45:28
basically what they come up with is, can
1:45:30
we create something that's like an S&P
1:45:32
500 with a higher sharp ratio? Can
1:45:35
we beat the market by a few percentage
1:45:37
points or frankly, even match the market each
1:45:39
year with lower volatility than if they were
1:45:41
buying an index fund? And you can see
1:45:43
who this would be very attractive to, pensions,
1:45:45
large institutions, firms that want
1:45:48
to compound at market or slightly
1:45:50
above market rate, but don't want
1:45:52
to risk these massive drawdowns or
1:45:54
frankly, just big volatility in general,
1:45:56
should they need to pull the
1:45:58
capital earlier. And the nice thing about being... invested
1:46:00
in a hedge fund versus a venture fund is you
1:46:02
can do redemptions. Like if you look at the 13
1:46:05
F's, the SEC documents that the Renaissance Institutional
1:46:07
Equities Fund files over time, it changes every
1:46:10
quarter because there's new people putting money in,
1:46:12
there's people doing redemptions. So it's a pretty
1:46:14
good product, or at least the theory behind
1:46:16
it is a pretty good product of a
1:46:19
lower risk, similar return thing
1:46:21
to the S&P 500. And
1:46:25
the marketing is built in. It's not
1:46:27
like there's any lack of demand of outside
1:46:29
capital that wants to invest with rent tech.
1:46:31
Right. It's really funny. There's only stories about
1:46:33
how the marketing documents literally say this is
1:46:35
not the medallion fund, we don't promise returns
1:46:37
like the medallion fund. In fact, we're not
1:46:39
charging for it like the medallion fund. You
1:46:42
know, David, you said that the fees and
1:46:44
carry on medallion went up to what five
1:46:46
and 44. Well, on the institutional fund, the
1:46:48
fees are one in 10. You're
1:46:50
only taking 1% annual fee and 10% of the performance. Clearly,
1:46:54
this is a very different product. But people
1:46:56
did not perceive that people were very excited.
1:46:58
It's a Renaissance product. It's the same analysts,
1:47:00
they're using all their fancy computers. I'm sure
1:47:02
we're going to get this crazy outperformance. And
1:47:05
at the end of the day, it is an extremely different vehicle.
1:47:07
Yeah, that has not performed
1:47:11
anywhere near how medallion has performed.
1:47:13
Correct. Has it served its purpose? Yeah.
1:47:16
But is it medallion? No, it's not
1:47:18
special in the way the medallion is
1:47:20
special. Yes. A
1:47:22
couple other funny things on the institutional
1:47:25
fund. So I spent a bunch of
1:47:27
time scrolling through 13 F's over the
1:47:29
last decade from the medallion filings. And
1:47:31
they're all from I think they have
1:47:33
two institutional funds. Yeah, there's
1:47:35
institutional equities and diversified alpha.
1:47:38
So the funniest thing is they filed these 13 F's.
1:47:40
And David and I are very used to looking at
1:47:42
13 F's of friends of the show
1:47:44
who run hedge funds we've had on his guests, or
1:47:46
perhaps really just any investor where you want to see
1:47:48
like, or what are they buying and selling this quarter?
1:47:51
And usually you see 15, 25, maybe 50 different
1:47:53
names on there. Well,
1:47:57
the 13 F's for Renaissance has 4300. stocks
1:48:00
in these tiny little chunks. And
1:48:03
there's a little bit of persistence quarter to
1:48:05
quarter. For example, weirdly, Novo Nordisk has been
1:48:07
one of their biggest holdings, biggest, I say
1:48:09
at like one to 2%. That's
1:48:12
their biggest position for several quarters in a
1:48:14
row. Hey, they've been listening to a choir.
1:48:16
That's right. That's one of the signals in
1:48:18
the model. You kind
1:48:20
of get the sense from looking
1:48:23
at these filings that these
1:48:25
things were flying all over the place. And this
1:48:27
was just the moment in time where they decided
1:48:29
to take a snapshot and put it on a
1:48:32
piece of paper. And even though this is the
1:48:34
end of quarter filing of what their ownership was,
1:48:36
if you had taken it a day or a
1:48:38
week earlier, it could look completely different. Yes.
1:48:41
The way that some folks we talked to
1:48:43
described the difference between the
1:48:46
institutional funds and medallion to us
1:48:48
is that medallions average hold time
1:48:51
for their trades and positions is
1:48:54
call it like a day, maybe
1:48:56
a day and a half. Whereas the
1:48:58
average hold time for the institutional
1:49:00
funds positions is like a
1:49:03
couple months. So across
1:49:05
4,300 stocks in the portfolio,
1:49:08
there's a lot of trading activity that happens on
1:49:10
any given day, but it's
1:49:12
a lot slower in any
1:49:14
given name than medallion would
1:49:17
be. Yeah, which makes sense. Again, it
1:49:19
gets back to this slippage concept. If
1:49:21
you have a bigger fund and you're
1:49:23
investing larger amounts, which the institutional funds
1:49:25
are, you can't be trading as
1:49:27
frequently or all of your gains are going to slip
1:49:29
away. Yeah. And
1:49:31
frankly, it just looks a lot like the S&P 500. When
1:49:34
you look at as of November 23, so
1:49:37
11 of the 12 months of the year had happened, they
1:49:39
were up 8.6%. Okay,
1:49:41
that sounds like an index type return. You
1:49:44
look at the first four months of 2020 right
1:49:46
after the crazy debt from the pandemic, they were
1:49:48
down 10.4%. Yes,
1:49:51
and the broader market, but they still
1:49:53
were sort of a mirror of the
1:49:55
broader market. So I think the RIEF,
1:49:57
their institutional fund, yes, it works as
1:49:59
expected. No, it's not Medallion.
1:50:01
And if it were standing
1:50:03
on its own, there's zero chance that we would
1:50:05
be covering the organization behind it unacquired. Zero
1:50:08
percent chance. Speaking of the
1:50:10
fund, that is the reason why we are covering
1:50:13
this company on this show. We
1:50:15
set up during the tech
1:50:17
bubble crash that volatility is when Medallion
1:50:19
really shines. Well, there's
1:50:22
no more volatile periods than 2007 and 2008. Yep.
1:50:27
2007, Medallion does 136% gross. 2008,
1:50:34
Medallion does 152% gross. Like
1:50:38
get out of here. Crazy. This
1:50:41
is 2008 while the rest
1:50:43
of the financial world is melting down.
1:50:46
And so this really does illustrate where do
1:50:48
they make their money from? Who is on
1:50:50
the other side of these trades? It's people
1:50:52
acting emotionally. They have effectively these really robust
1:50:54
models that are highly unemotional, that are making
1:50:56
these super intricate, multi-security
1:50:59
bets. And they are putting on exactly
1:51:01
the right set of trades to achieve
1:51:03
the risk and exposure that the system
1:51:05
wants them to have. And who is
1:51:07
on the other side of those trades?
1:51:09
It's panic sellers. It's dentists. It's hedge
1:51:11
funds who don't trust their computer systems
1:51:13
and are like, ah, crap, we gotta
1:51:15
just take risk off even though it's
1:51:18
a negative expected value move for us.
1:51:20
They're basically trading against human nature. And
1:51:22
importantly in this business versus every other
1:51:24
business that we cover here on Acquired
1:51:26
or most other businesses, this is truly
1:51:28
zero sum. It's not like they're
1:51:31
here in an industry that's a growth industry
1:51:33
and lots of competitors can take different approaches,
1:51:35
but the whole pie is growing so much
1:51:38
that I don't care if, no, you're fighting
1:51:40
over a fixed pie here. I'm
1:51:42
trading against someone else. I win, they lose.
1:51:45
Yes. Well, there's one slight
1:51:47
nuance to that, but I don't
1:51:49
know how much it holds water. And the
1:51:52
apologist nuance would be, well, Warren
1:51:55
Buffett could be on the other side of the
1:51:57
trade and Medallion could. could
1:52:00
make money on that trade with Warren over
1:52:02
its time horizon of a day and a
1:52:04
half. And Warren could make money
1:52:06
over his time horizon of, you know, 50 years.
1:52:09
Superfair. So I
1:52:11
think the argument against that,
1:52:13
though, is that Medallion
1:52:16
sold after a day and a half
1:52:18
to somebody else who bought at that
1:52:20
lower price. And so somewhere
1:52:23
along the chain, that loss
1:52:25
is getting offloaded to somebody. The
1:52:28
direct counterparty of Medallion and
1:52:30
the quant industry writ large
1:52:32
might not take the loss, but somebody is going to
1:52:34
take the loss along the way. It
1:52:37
is, as you say, a zero sum game. Yeah.
1:52:40
But I think the important thing is, can
1:52:42
you and your adversary both benefit? And I
1:52:44
think in this case, you and your counterparty,
1:52:46
the person you're trading against, yes, you have
1:52:48
two different objective outcomes. Like can I get
1:52:50
a penny over on Warren Buffett by managing
1:52:52
to take him on this one trade? Sure.
1:52:55
And the strategy is such that that is irrelevant. So
1:52:58
after the historic performance during
1:53:00
the financial crisis, as I
1:53:02
alluded to earlier, Jim
1:53:04
retires at the end of 2009 and Peter
1:53:07
and Bob become co-CEOs, co-heads of
1:53:09
the firm in 2010. They
1:53:13
take the portfolio size up to $10
1:53:15
billion when they take over. It had
1:53:18
been at five for the last few
1:53:20
years of Jim's tenure. They
1:53:22
take it up to 10. And really
1:53:25
with no impact, which I assume means
1:53:27
that RENTEC was getting better and the
1:53:29
models were getting better because otherwise they
1:53:31
would have gone to 10 before. Right.
1:53:34
They gained confidence that they
1:53:36
had enough profitable trades
1:53:38
they could make that they could raise
1:53:40
the capacity without dampening returns. Yes.
1:53:43
And perhaps they could have done it earlier and
1:53:46
they just didn't have the confidence that it would
1:53:48
work at larger size. But I bet they're
1:53:50
very good at knowing how large can our strategy
1:53:52
work up to before it starts having diminishing returns.
1:53:55
Yes. And importantly, during
1:53:58
periods of peak volatility. like,
1:54:00
say, 2020, Medallion
1:54:02
continues to shoot the lights out. So
1:54:05
from at least the data that we were
1:54:07
able to find on Medallion's performance over the
1:54:09
past few years, 2020, they
1:54:11
were up 149% gross and
1:54:14
76% net. So the magic is still there. And
1:54:16
one way to look at it, which may not
1:54:24
be the be all and end all, but
1:54:26
I think is a good way to compare
1:54:28
Jim's era at Medallion versus
1:54:31
Peter and Bob's era. During
1:54:33
Jim's tenure, Medallion's total
1:54:36
aggregate IRR from 1988, when
1:54:39
the fund was formed to 2009 when he retired,
1:54:41
was 63.5% gross annual returns and 40.1% net annual
1:54:51
returns, which of course did include many
1:54:53
periods of lower carry, 20% versus the
1:54:58
44%. During the post-Jim era, the Peter and
1:55:00
Bob era from 2010 to 2022 was when
1:55:02
we were able
1:55:05
to get the latest data. IRRs
1:55:07
are 77.3% gross and 40.3% net. So better on
1:55:09
both fronts, even
1:55:18
with much higher average fees. So
1:55:21
yeah, I think Medallion is doing fine. It's
1:55:24
amazing. And we weren't able to
1:55:26
tell, there's some sources that report that they've
1:55:28
grown from $10 billion in the last
1:55:30
few years to being comfortable at a $15 billion fund
1:55:33
size. And if so, that just
1:55:35
means that they continue to find
1:55:37
more profitable strategies within Medallion to
1:55:40
keep those same unbelievable returns at
1:55:42
larger sizes. And
1:55:45
at the end of the day, this is all just insane. So
1:55:47
as far as we can tell, Ben, you alluded
1:55:50
to this a bit at the beginning of the
1:55:52
episode. And as far as anybody else
1:55:54
can tell, Medallion
1:55:57
has by far the
1:55:59
best investing trends. record of any
1:56:01
single investment vehicle in history. So
1:56:04
give me those net numbers. So
1:56:06
during the entire lifetime so far of Medallion
1:56:08
from 1988 to 2022, that's 34 years. The
1:56:15
total net annual
1:56:17
return number is 40%
1:56:19
for zero, who over 34 years
1:56:21
after fees, it's 68%
1:56:26
for four fees, which
1:56:28
equates to total
1:56:30
lifetime carry dollars for the
1:56:32
whole firm of $60
1:56:34
billion just in carry by
1:56:36
our calculations. Astonishing. That is
1:56:39
a lot of money. Also,
1:56:41
David Rosenthal, good spreadsheet work on this. You have
1:56:43
not done a spreadsheet for an episode in a
1:56:45
while, so I admire your work on this one.
1:56:48
Yeah. I still know
1:56:50
how to use Excel. Barely.
1:56:54
It's going to be a
1:56:56
dying art now with co-pilot and TPTs.
1:56:58
That's right. Okay. So
1:57:00
$60 billion in total carry. So
1:57:02
$60 billion in total carry is a
1:57:05
lot of money. And
1:57:07
well, speaking of a lot of money, we
1:57:10
do need to mention before we finish the
1:57:12
story here that that
1:57:14
Ren tech money has bought a
1:57:17
lot of influence in society. So
1:57:20
Bob Mercer, that name may have sounded
1:57:22
familiar to many of you along the
1:57:24
way. Bob was
1:57:26
the primary funder of Breitbart
1:57:28
and Cambridge Analytica and
1:57:31
one of the major financial backers of both the
1:57:33
2016 Trump campaign and
1:57:35
the Brexit campaign in Great
1:57:38
Britain. No, lest you think
1:57:40
that Ren tech dollars are solely being
1:57:42
funneled into one side of the political
1:57:44
spectrum. Sam Simons is a
1:57:46
major democratic donor as are
1:57:49
many other folks at Ren tech. Yeah.
1:57:52
Henry Laufer and other folks are also huge
1:57:54
donors, approximately to the same tune as
1:57:56
what Bob Mercer is on the right. Yeah.
1:58:00
millions of dollars, many tens of millions of dollars on
1:58:03
all sides and through many campaign
1:58:05
cycles here from RENTEC employees and
1:58:07
alumni. This did become
1:58:09
a flashpoint for the firm in
1:58:11
the wake of the 2016 election. Mercer obviously
1:58:15
became a controversial figure both
1:58:18
externally and internally within the firm.
1:58:21
Especially once people realized he was
1:58:23
the through line through Breitbart Cambridge
1:58:25
Analytica, the Trump election, and Briggs
1:58:27
it. Yes. Ultimately
1:58:29
Jim asked Bob to step down as co-CEO in 2017,
1:58:31
which he did, but he did remain a scientist
1:58:37
at the firm and a contributor
1:58:39
to the models even though he wasn't leading
1:58:42
the organization with Peter from a leadership standpoint
1:58:44
any longer. Ultimately the thing
1:58:47
that surprised me the most is how
1:58:49
these people all still work
1:58:51
together despite having about the
1:58:53
most opposite political beliefs you
1:58:55
could possibly have. Yeah, understatement
1:58:57
of the century. And all
1:58:59
being extremely influential and active
1:59:02
in those political systems. Yes,
1:59:05
Bob Mercer is no longer the CEO
1:59:07
of Renaissance Technologies or the co-CEO. He
1:59:09
still works there. He's still
1:59:11
associated. They all still speak highly of
1:59:13
each other. It's unexpected. Yeah,
1:59:16
I think unexpected is the best way to put
1:59:18
it. Like everything with Renaissance, it works
1:59:20
a little bit different than the rest of the
1:59:22
world. Yes. Okay,
1:59:25
speaking of, let's transition to
1:59:27
analysis. And I have
1:59:30
a fun little monologue I want to go
1:59:32
on if you will bear with me. Ben,
1:59:35
I think this qualifies as the
1:59:37
RENTEC playbook, but I really kind
1:59:39
of think of it as the RENTEC
1:59:41
tapestry. And I was inspired by Costco
1:59:44
here because we were talking to folks in the research
1:59:46
and everybody said, you know, RENTEC,
1:59:50
it just has these puzzle pieces that fit
1:59:52
together on the surface. RENTEC
1:59:55
does the same things
1:59:57
that Citadel, D-Shaw, 2
2:00:00
Sigma, Jane Street, others, etc. do.
2:00:04
They hire the smartest people in the world
2:00:06
and they give them the best data and
2:00:08
infrastructure in the world to work
2:00:11
on. And they say, go
2:00:13
to town and make profitable trades. Those
2:00:17
are very expensive commodities, those two things, the smartest
2:00:20
people in the world and the best data and
2:00:22
infrastructure, but they are commodities.
2:00:24
Like Citadel can say the exact same
2:00:26
things, just the same as like Walmart
2:00:28
and Amazon can say we too have
2:00:30
large scale supplier relationships that we leverage
2:00:32
to provide low prices to customers just
2:00:34
like Costco. But
2:00:36
it's underneath that where I think the magic lies.
2:00:38
There are three very interrelated
2:00:41
things that make RENTEC unique.
2:00:44
So number one, they get the
2:00:47
smartest people in the world to
2:00:49
collaborate and not compete.
2:00:52
Pretty much every other financial
2:00:54
firm out there, employees
2:00:57
and teams within the firm
2:01:00
quasi compete with one another. Yeah,
2:01:03
I mean, typically in kind of a friendly way. But
2:01:05
yeah, let's take like in a
2:01:07
venture firm, you've got your lead
2:01:10
partner on a deal or a deal team,
2:01:12
they're working that deal. And
2:01:15
maybe some of the other partners help a little
2:01:17
bit, but mostly they're off prosecuting their own deals.
2:01:20
And I think that's the most collegial way
2:01:22
that this happens in finance. Then you've
2:01:24
got multi strategy hedge funds out there where
2:01:26
literally firms are being pitted against one another
2:01:29
to be weighted in the ultimate trading model
2:01:32
for the firm. At RENTEC
2:01:34
though, because of
2:01:36
the one model architecture, everyone
2:01:39
works together on the same investment
2:01:42
strategy and the same investment
2:01:44
infrastructure. That means everyone
2:01:47
sees everybody else's work, everybody who works
2:01:49
at RENTEC on the research team on
2:01:51
the infrastructure team, they have access to
2:01:53
the whole model. That's not
2:01:55
true anywhere else. Yeah,
2:01:58
that's a good point. The whole code base is completely
2:02:00
visible. And that also
2:02:02
means because it's just one model,
2:02:05
just one strategy, when
2:02:07
somebody else improves that
2:02:09
model's performance, that directly
2:02:12
impacts you as much
2:02:14
as it impacts them. This is really
2:02:16
different than any other hedge fund out
2:02:18
there. So why is that different than if
2:02:20
I roll some of my compensation into a multi strategy hedge
2:02:22
fund that I work at? Don't I
2:02:24
love other teams creating high performance also?
2:02:27
Sure, but you don't love it as much as
2:02:30
your team, because either compensation or
2:02:32
career wise, you are much
2:02:34
more dependent on your performance than
2:02:36
you are other people's performance. Oh,
2:02:39
yes. This is a big thing. You intend
2:02:42
to have a job after that job at most
2:02:44
places most of the time. So you care about
2:02:46
credit, and you care about smashing the pinata and
2:02:48
then going elsewhere or building reputation and then going
2:02:50
elsewhere. Most of the people at RENTEC are not
2:02:53
going to have another job. What
2:02:55
did you find on LinkedIn, at least
2:02:57
the median tenure of employees is like
2:02:59
16 years? Yeah, I just got
2:03:01
LinkedIn premium and you can see median tenure
2:03:03
and it's crazy. There's only like three, 400
2:03:05
employees at Renaissance and the
2:03:07
median tenure at least as reported by LinkedIn
2:03:10
is like 14 years. Yes. Okay,
2:03:13
this brings me to point number two, which
2:03:16
he said this is an absurdly
2:03:18
small team. There are less than
2:03:21
400 employees that work at RENTEC,
2:03:24
only half of which work in research
2:03:26
and engineering, and the other half are
2:03:28
either back office or institutional sales for
2:03:30
the open funds. So let's call
2:03:32
it, I don't know, 150, 200 people max
2:03:34
who are like hands on the wheel
2:03:37
here for Medallion. Yep.
2:03:39
Every other peer firm of
2:03:42
RENTEC, Citadel, D-Shaw, Two Sigma,
2:03:44
etc, all of them, you
2:03:46
lump Jane Street, jump the
2:03:48
high frequency guys in here.
2:03:51
Minimum 2,000 to 5,000 people work at
2:03:53
those places. Wow, I didn't realize it
2:03:55
was that big. It is an
2:03:58
order of magnitude more people. Who.
2:04:00
Are working at the other firms versus you're
2:04:02
working at Rent. And lest you think that
2:04:04
it's like a capital, These things know the
2:04:07
institutional fans have gotten big. They peaked at
2:04:09
over one hundred billion, but they're currently between
2:04:11
sixty and seventy billion that they manage on
2:04:13
top of the ten or fifteen that's in
2:04:16
the. Medallion. Fund yeah so
2:04:18
am. It's like the same job as
2:04:20
these big ones. This. Has. All
2:04:23
sorts of benefits, Number. One there's
2:04:25
like that are most at oh yea
2:04:27
Worksop Benefit. Everyone. Knows each
2:04:29
other by name. You know your colleagues
2:04:31
kids, You know your colleagues families. Yeah,
2:04:33
they put right on their website. There
2:04:35
are ninety Phds in Mathematics, Physics, computer
2:04:37
Science, and related fields. The. About Page
2:04:39
has these ten kind of random bullet points and
2:04:42
that's one of I'm just. Then. There's
2:04:44
the related asked I tell this. The
2:04:46
from is. In. The middle of nowhere
2:04:48
on Long Island. You. Actually know
2:04:50
your colleagues, families, and kids because you're
2:04:52
not going out and getting drinks with.
2:04:55
Someone. From to Sigma New York City of
2:04:57
to Not comparing notes are. Measuring.
2:04:59
Parts of your anatomy with someone else.
2:05:01
The only thing out of the swimming
2:05:04
pool? Totally. And since Renaissance doesn't recruit
2:05:06
from finance jobs, It's. Kind
2:05:08
of unlikely that you know someone else
2:05:10
and finance you came out of a
2:05:12
science related field you now work in
2:05:14
East to talk at. Long. Island
2:05:16
which has it's like ten thousand people or
2:05:18
something or less that live there. Seer: This
2:05:20
little town. You're not actually going into
2:05:22
the city that off and and if you are
2:05:24
it's to get not to grab drinks with other
2:05:27
finance people so. Even. If you
2:05:29
didn't have a. Many. Page
2:05:31
noncompete, He and a lifetime
2:05:33
and da. You're. Very
2:05:36
unlikely to be in the social
2:05:38
circles you decide getting expose exactly
2:05:40
and. Rent. Text firing
2:05:42
established scientists impedes these. They're
2:05:44
not hiring kids out undergrad
2:05:46
like Jane Street or Bridgewater
2:05:49
is. My. sense is that
2:05:51
the places like a college campus without
2:05:53
any students have you seen their pictures
2:05:55
online yeah if you look up renaissance
2:05:57
technologies a google and you go and
2:05:59
look at the photos on can It's
2:06:01
a little courtyard and winding, walking paths
2:06:03
and woods all around it, tennis
2:06:06
courts. Yep. So
2:06:08
then there's the last piece of
2:06:10
the small team element, which is just
2:06:12
the magnitude of the financial impact, which
2:06:14
I don't think is true. But let's
2:06:17
say that there were another quant fund
2:06:19
that made the same number of dollars
2:06:21
of performance returns that
2:06:23
RENTAC does. At RENTAC,
2:06:25
you're splitting that a couple hundred ways. At Citadel,
2:06:28
you're splitting that 5,000 ways. It
2:06:31
just doesn't make sense to go anywhere else. We
2:06:33
were chatting with someone to prep for this episode
2:06:35
and they told us, you can't ever compete with
2:06:37
them, but they'll pay you enough that you won't
2:06:39
want to. Yes. Okay.
2:06:42
So this brings me to what I've been kind of
2:06:44
teasing and I'm super excited about. I
2:06:47
think the third puzzle
2:06:49
piece of what makes
2:06:51
RENTAC so unique and defensible is
2:06:55
Medallion's structure itself.
2:06:58
That it is a LPGP
2:07:01
fund with 5%
2:07:03
management fee and 44% carry. So
2:07:07
it's not like a prop shop or
2:07:09
proprietary, it's just one pot of money.
2:07:11
It's literally a GPLP, even though the
2:07:13
GPs and the LPs are the same
2:07:15
people. So here's my thinking on this.
2:07:18
Now, I don't know how it is
2:07:20
actually structured, but there was something
2:07:22
about this whole crazy 44%
2:07:25
carry that just wasn't sitting with me
2:07:27
right throughout the research because I kept
2:07:29
asking myself, why? Right.
2:07:31
They've already kicked out most of the LPs, if
2:07:33
not all. So why are they raising the carry?
2:07:36
Right. It's all themselves. It's all insiders.
2:07:38
Why do they charge themselves 44% carry
2:07:40
and 5% management
2:07:43
fees? I think Jim talks about this though. Oh, I
2:07:45
pay the fees just like everybody else. Yes. It's always
2:07:47
a funny argument. It's like, who are you paying the
2:07:49
fees to? Right. So I was like,
2:07:51
what is happening here? So,
2:07:54
okay, here's my hypothesis. This
2:07:57
is not about having
2:07:59
crazy... performance fees. This
2:08:01
is not about having the highest carry in the industry. This
2:08:04
is a value
2:08:07
transfer mechanism within
2:08:09
the firm from the 10-year
2:08:11
base to the current people who
2:08:13
are working on Medallion in any given year. So
2:08:16
here's how I think it works. When
2:08:19
people come into RENTEC,
2:08:21
they obviously have way less wealth than the
2:08:23
people who've been there for a
2:08:25
long time, both from the
2:08:28
direct returns that you're getting every year
2:08:30
from working there and just your investment
2:08:32
percentage of the Medallion Fund. Which
2:08:34
by the way, I think
2:08:37
they took, it was either the state of
2:08:39
New York or the federal government to
2:08:41
court to be able to have
2:08:43
the 401k plan at
2:08:45
RENTEC be the Medallion Fund.
2:08:47
No way. Yeah, so like if you work
2:08:49
there, your 401k is the Medallion Fund. That's
2:08:51
crazy. So it really doesn't take more than
2:08:53
a few years before you're set for life.
2:08:56
Totally. I mean, depending on your definition of
2:08:58
set for life, I think it happens very,
2:09:00
very quickly. Yeah. Okay. So given
2:09:02
that though, how do you
2:09:05
avoid the incentive for a group of
2:09:07
talented younger folks to split off and
2:09:09
go start their own Medallion
2:09:11
Fund? Right. Especially when
2:09:13
they all have access to the whole
2:09:15
code base. The whole thing is meant
2:09:17
to function like a university math department
2:09:19
where everyone's constantly knowledge sharing because we're
2:09:21
going to create better peer reviewed research
2:09:24
when we all share all the knowledge
2:09:26
all the time. You would think that's
2:09:28
a super risky thing to give everyone
2:09:30
all the keys. Right. So
2:09:32
I think it's the 44% carry
2:09:35
structure that does it. Because basically what
2:09:37
you're saying is every year,
2:09:39
5% management
2:09:42
fee, so 5% off the top and then
2:09:44
44% of performance. So
2:09:46
let's say Medallion is on the order
2:09:48
of, call it doubling every year. Let's round
2:09:50
that up and just add up and say 49% of
2:09:54
the economic returns in any given year
2:09:57
go to the current team and.
2:10:01
51% of the economic returns go
2:10:03
to the tenure base. I
2:10:05
was like, what is the equivalent here? I think
2:10:07
it's kind of like academic tenure kind of thing.
2:10:09
The longer tenure you are at the firm, the
2:10:12
more your balance shifts to the
2:10:14
LP side of things. Interesting. The
2:10:16
younger you are at the firm,
2:10:19
the more your balance is on the GP
2:10:21
side of things. But at the end of the day, it's 51,
2:10:23
49. So
2:10:25
there's this very natural value
2:10:28
transfer mechanism to keep the people that are
2:10:30
working in any given year super
2:10:33
incentivized. And as you
2:10:35
stay there longer, you
2:10:37
are paying your younger colleagues
2:10:39
to work for you. Right.
2:10:42
Funny. I think it's a good insight
2:10:45
that it's structured like a university department tenure. Well,
2:10:47
I just kept asking myself, why?
2:10:49
Why? Why do they have this
2:10:52
if there's no outside LPs? And this was
2:10:54
the best thing I could come up with. And
2:10:56
I actually think it's kind of genius. Yeah,
2:10:58
it's more elegant than it's all one person's
2:11:00
money and they're deciding to bonus out the
2:11:02
current team every year and just give them
2:11:04
enough money to make sure you retain them.
2:11:07
Right. Which is how I think most
2:11:09
prop shops work. Like Jane Street is mostly a prop
2:11:11
shop. I think it is mostly the principal's money, but
2:11:15
that's a static situation. It's not
2:11:17
like, if that were true, then Jim would
2:11:19
just own this thing forever. And
2:11:22
I don't think that's true here at RENTEC. Yeah,
2:11:24
so essentially David, the real magic
2:11:26
is they've got one fund, it's
2:11:28
evergreen. And when
2:11:31
you start at the firm, you're only
2:11:33
getting sort of paid the carry amount,
2:11:35
but over time you become a
2:11:37
meaningful investor in the firm and you sort of shift
2:11:39
to that 51%, you're kind of the LP. And
2:11:42
then over time you eventually graduate out entirely and
2:11:45
you're only an LP. And so you're right, it's
2:11:47
a value transfer mechanism from the old
2:11:50
guard to the new guard in a
2:11:52
way that is clear, well understood, probably
2:11:54
tax advantaged versus just doing I'm
2:11:57
the owner and I'm giving everyone arbitrary bonuses.
2:11:59
Yeah. And at the end of
2:12:01
the day, I think these three pieces, to
2:12:04
me, are the core of this sort of
2:12:06
tapestry of RenTech. One model
2:12:09
that everybody collaborates on together. A
2:12:11
super small team where we all know each
2:12:14
other and the financial impact that
2:12:16
any of us make to that one model is great
2:12:19
to all of us. And
2:12:21
three, this LPGP model
2:12:23
with very high carry
2:12:25
performance fees that creates the right set of
2:12:27
incentives both for new talent on the way
2:12:29
in and old talent on the way out.
2:12:32
Yup, I think that's right. Okay, there's
2:12:34
a few other parts of the story that we skipped
2:12:36
along the way because there was no real good place
2:12:39
to put them in. But these are objectively
2:12:41
fascinating historical events that are totally
2:12:43
worth knowing about. And
2:12:46
the first one is called Basket Options. So
2:12:48
the year is 2002. RenTech
2:12:51
has 13 years of knowing
2:12:53
that they basically have a machine that
2:12:55
prints money. So what should you do
2:12:57
when you have a machine that prints
2:12:59
money? Leverage. Now,
2:13:02
there are all sorts of restrictions around firms
2:13:04
like this and how much leverage they can
2:13:06
take on. You can't just go and say,
2:13:08
I'm going to borrow $100 for every dollar
2:13:11
of equity capital that I have in here.
2:13:13
So you need to sort of get clever
2:13:15
to borrow a whole bunch of money from
2:13:17
banks or from any lender to basically juice
2:13:20
your returns. If, again, you have a money
2:13:22
printing machine that's reliable, most people don't. Most
2:13:24
people probably shouldn't take leverage because they're just
2:13:26
as likely to blow the whole thing up
2:13:28
as they are to be successful. So
2:13:32
Basket Options. I am going to
2:13:34
read directly from the man who solved the market because
2:13:36
Greg Zuckerman just put it perfectly. Basket
2:13:38
Options are financial instruments whose values are pegged
2:13:40
to the performance of a specific basket of
2:13:43
stocks. While most options are based
2:13:45
on an individual stock or
2:13:47
a financial instrument, basket options are linked
2:13:49
to a group of shares. If
2:13:51
these underlying stocks rise, the value of
2:13:54
the option goes up. It's like owning
2:13:56
the shares without actually doing so. Indeed,
2:13:58
the banks who. of course, loaned
2:14:01
the money, who put the money in the
2:14:03
basket option, were legal owners of the shares
2:14:05
in the basket. But for all intents and
2:14:07
purposes, they were medallions property. So this is
2:14:09
very clever medallion saying, well, the way we're
2:14:11
going to lever up is there's a basket,
2:14:14
we have an option to purchase that basket,
2:14:16
most of the capital in that basket is
2:14:18
actually the bank's capital, but the bank has
2:14:20
hired us to trade the options in the
2:14:22
basket. And then after a year, when long
2:14:25
term capital gains tax kicks in, we
2:14:27
have the option to buy that basket.
2:14:30
So anyway, all day medallions
2:14:32
computers sent automated instructions to the banks, sometimes
2:14:34
an order a minute or even a second,
2:14:37
the options gave medallion the ability to
2:14:39
borrow significantly more than it otherwise would
2:14:41
be allowed to competitors generally had about
2:14:43
$7 of financial instruments for every dollar
2:14:46
of cash. By contrast, medallions option strategy
2:14:48
allowed it to have $12 and 50
2:14:50
cents worth of financial instruments for every
2:14:53
dollar of cash, making it easier to
2:14:55
trounce rivals assuming they could keep finding
2:14:57
profitable trades. When medallion spied an especially
2:14:59
juicy opportunity, it could boost leverage holding
2:15:02
close to $20 of asset for every
2:15:04
dollar of cash. In 2002, medallion managed
2:15:07
over $5 billion, but it controlled over $60 billion
2:15:10
of investment positions. David,
2:15:12
this exposes something we haven't shared yet on
2:15:14
the episode, which is, it's not just that
2:15:16
they could find $5 billion worth of profitable
2:15:18
trades, it's that they wanted to lever the
2:15:20
crap out of $5 billion and
2:15:23
find $60 billion of profitable trades to
2:15:25
make and basket options gave them a
2:15:27
legal way to have an incredible amount
2:15:29
of leverage in a way that they
2:15:32
felt safe about. Yeah, the unlevered
2:15:35
returns, if you were
2:15:37
running this strategy would be much
2:15:40
lower. Yeah. So a big piece
2:15:42
of this playbook that we didn't talk about is leverage,
2:15:44
but every quant fund does leverage and so Renaissance was
2:15:46
just more clever than everyone else. Yeah, it's
2:15:49
an important point though. Nine out
2:15:51
of every 10 companies that we cover
2:15:53
on acquired leverage is zero part
2:15:55
of the story. Right. And for us coming from
2:15:57
the world we come from in tech capital
2:16:00
leverage is like a dirty word. Like I'm scared
2:16:02
of it. Right. I mean, you could
2:16:04
imagine, let's say it wasn't, they were right 50.25% of the
2:16:06
time, but they were right 50.00001%
2:16:08
of the time. They
2:16:11
would need to do a ton of trades in
2:16:13
order to generate enough profits. So that's
2:16:15
why you need, you know, $60 billion
2:16:18
of cash to actually execute the strategy
2:16:20
to produce the returns that they were
2:16:22
looking for, you know,
2:16:24
on $5 billion of equity. Anyway, there's a
2:16:26
second chapter to this, which is it's all well and
2:16:28
good that this is how they get a bunch of
2:16:30
leverage. That's one piece of it.
2:16:33
The other piece is they thought this was
2:16:35
a remarkably tax efficient vehicle. The
2:16:37
way that they were filing their taxes said, Oh,
2:16:40
sure, there's stuff in that basket. But the
2:16:42
thing that we actually own is an option
2:16:44
to buy that basket or sell that basket.
2:16:46
And we only exercise that once every 13
2:16:48
months or so. I don't know the exact
2:16:50
number, but something like that over a year.
2:16:52
And so therefore, we're buying something, we're
2:16:54
holding it for a year. We're selling it. Oh,
2:16:57
of course there's millions and millions of trades going
2:16:59
on inside the basket. But we don't own that
2:17:01
basket. The banks do. We're just advising them. You
2:17:03
can kind of see the logic here. Over
2:17:06
time, eventually in 2021, the IRS
2:17:08
said, No, you made all those
2:17:10
trades. That was not a completely
2:17:12
separate entity. And so you
2:17:15
guys owed $6.8 billion
2:17:17
in taxes that you didn't pay. You're
2:17:20
going to need to pay that with interest,
2:17:22
with penalties. And by the way,
2:17:24
Jim Simons, we're going to want you and the
2:17:26
other few partners to really bear the load of
2:17:28
that. And they did. So for Simons alone, he
2:17:30
paid $670 million to the
2:17:32
IRS and back taxes for this basket option
2:17:34
strategy that turned out not to be a
2:17:36
long term capital gain. All
2:17:39
right, so numbers on the business today, and
2:17:41
then we will dive into power and playbook.
2:17:44
So today, we've talked about medallion 10 or
2:17:46
15 billion, depending on who you ask. Historically,
2:17:49
it was more like five or 10 billion.
2:17:51
The institutional fund is about 60 to
2:17:53
70 billion. And that's one point
2:17:55
was 100 billion. The
2:17:57
total carry generated David, you said is $60 billion.
2:18:00
Forbes estimates that Jim Simons alone is worth
2:18:02
about $30 billion today, which kind of
2:18:06
pencils with a bunch of other stats over
2:18:08
the years that he owned about half
2:18:10
of Renaissance. The returns, obviously
2:18:13
the medallion fund generated approximately 66%
2:18:15
annualized from 1988
2:18:17
to 2020 after those fees was about 39% wild. So
2:18:24
an interesting thing to understand, I
2:18:26
ran a hypothetical scenario of how much money
2:18:29
do you think Renaissance the
2:18:31
business makes a year in revenue? And
2:18:34
so the institutional fund, let's call it
2:18:36
10% on 60 billion of assets. So
2:18:39
that's 600 million from fees and 600 million
2:18:41
from performance. So 1.2 billion
2:18:43
a year in revenue to the firm
2:18:46
from the institutional side of
2:18:48
the business. Because I always ask myself the question,
2:18:50
does that actually matter? They did all this work
2:18:52
to stand up the institutional side. Who cares? Well,
2:18:55
let's say medallion does their average 66% gross on 15
2:18:57
billion. That
2:19:01
is 750 million in fees and 4.3 billion on performance.
2:19:07
So a total of 5 billion
2:19:09
from medallion and 1.2 billion
2:19:12
from the institutional side of the
2:19:14
business. Now, of course, the employees are
2:19:16
the investors in medallion. So you could just argue it's
2:19:18
actually silly to cut them up, but I
2:19:20
don't know, it's a 789 billion
2:19:22
dollar revenue business. Right, because
2:19:24
that's not including the LP return on medallion.
2:19:27
A hundred percent, it's not. Which again, as
2:19:29
we spent a long time talking about, it's
2:19:31
all the same thing. Yes. But
2:19:33
it's kind of interesting just to compare it against other
2:19:35
companies to have this in the back of your head.
2:19:38
This is a 7, 8 billion dollar a year revenue
2:19:40
business. Now, I
2:19:42
think there are a lot of expenses on
2:19:44
the infrastructure side. Totally. That was another
2:19:46
thing I wanted to talk about, the fact that they do,
2:19:48
let's say medallion alone. So they have
2:19:50
750 million dollars in fees. I
2:19:52
don't think they come close to 750
2:19:54
million dollars a year in expenses, but
2:19:56
they are running, who knows
2:19:59
what infrastructure, some kind of. supercomputing cluster, what
2:20:01
does it cost to run one Amazon data
2:20:03
center? I mean, it's, I think, much smaller
2:20:05
scale. I don't know. I mean, you're talking
2:20:08
about a lot of data here. Yeah,
2:20:10
it says right on their website, they have 50,000 computer
2:20:13
cores with 150 gigabits per
2:20:16
second of global connectivity and a research
2:20:18
database that grows by more than 40
2:20:21
terabytes a day. That's
2:20:23
a lot of data. Right. Is
2:20:25
that 750 million a year? I don't know,
2:20:27
but it's not zero. I don't
2:20:29
think so. They're certainly
2:20:31
not losing money on the fees, but
2:20:34
there are actual hard costs to this
2:20:36
business. Right. I wonder too
2:20:38
if the fee element
2:20:40
of Medallion basically
2:20:43
pays the base salaries for the current team.
2:20:46
That feels like it's right. If
2:20:48
you're someone who has
2:20:50
done a data center build out before, or has
2:20:53
any way to sort of back into what the
2:20:55
costs of Medallion's operating expenses
2:20:57
are on the compute and data and
2:20:59
network side, we would love to hear
2:21:01
from you. Hello at Acquired.fm. Okay.
2:21:04
Power. Power. This
2:21:07
is a fun one. Yeah. So
2:21:10
listeners who are new to the show, this is Hamilton
2:21:12
Helmer's framework from the book Seven Powers. What
2:21:14
is it that enables a business to
2:21:16
achieve persistent differential returns to be more
2:21:19
profitable than their closest competitor on a
2:21:21
sustainable basis? And the
2:21:23
seven are counter positioning, scale
2:21:25
economies, switching costs, network economies,
2:21:29
process power, branding, and cornered
2:21:31
resource. And David, my
2:21:34
question to you to open this section is specifically
2:21:37
about RENTEC's lifelong non-competes.
2:21:41
That feels like a big reason that
2:21:43
they maintain their competitive advantage. And
2:21:45
I'm curious if you agree with that, what would you
2:21:47
put that under? Well, I think
2:21:49
it's lifelong NDAs and
2:21:51
non-competes as long as the state of
2:21:54
New York legally allows for, but
2:21:56
that is not lifetime. I've heard various figures,
2:21:58
six years, five years. five years, something
2:22:01
like that. I mean, at
2:22:03
the end of the day, non-competes are more like, what
2:22:05
is one side willing to get a court over? But
2:22:08
the reality is, people don't
2:22:11
leave. People don't leave, period. And people
2:22:13
especially don't leave and start their own firms. I
2:22:17
was thinking about this in the middle of the night.
2:22:20
And I think there's three layers to
2:22:23
the effective non-compete
2:22:25
that happens with
2:22:27
RENTEC. There's the legal layer,
2:22:30
the base layer that you're talking about, it's like the
2:22:32
agreements you sign. Then there's the
2:22:34
economic layer of what we spent
2:22:36
a long time talking about in tapestry of it
2:22:39
would just be dumb to leave. You're better
2:22:41
off staying there as part of that team
2:22:43
with a smaller number of people than going
2:22:45
to Sigma with a lot more people. I
2:22:49
think that's the next level up. And then I think
2:22:51
the highest level is just probably the social layer. You're
2:22:54
there with the smartest people in the world
2:22:56
in a collegial atmosphere where you're all working
2:22:58
hard on something that has direct impact on
2:23:00
you. Right, it's your community. It's
2:23:02
your community, totally. You're not in New York City,
2:23:04
you're not in the Hamptons, you're not
2:23:07
in Silicon Valley. You are
2:23:09
selecting into that. And
2:23:12
I think if that's what you want, what better
2:23:14
place in the world? All right,
2:23:16
so classify it. What power does that
2:23:18
fall under? Well, I
2:23:20
mean, I think the people specifically you
2:23:22
would put into cornered resource, but I'm
2:23:24
not actually sure that fully captures it
2:23:26
here. I was thinking more
2:23:29
process power because I think it is the combination
2:23:31
of the people and the
2:23:34
model and the incentive structures.
2:23:37
Yep. I think that's
2:23:39
right. I also had my biggest one
2:23:41
being process power. You actually can develop
2:23:43
intricate knowledge of how a system
2:23:45
works and then build processes around that that
2:23:47
are hard to replicate elsewhere. I
2:23:50
think these systems have been layered over time
2:23:52
also, where anyone who's come into the firm
2:23:54
in the last five years doesn't
2:23:56
know how it works start to finish.
2:23:58
I didn't ask anyone to... verify
2:24:00
that, but it's over 10 million
2:24:02
lines of code and the
2:24:04
level of complexity of the system
2:24:07
of when it's putting on
2:24:09
trades, what trades is putting on, why, the
2:24:12
speed at which they need to happen, I
2:24:14
actually don't think anyone holds the
2:24:16
whole model in their head. And so I
2:24:18
think there's process power just because
2:24:21
it's 30 plus years of complexity that's
2:24:23
been built up. Yeah. I
2:24:26
totally agree with that, particularly in the
2:24:28
model itself. I mean, maybe you could
2:24:30
argue the model is a cornered resource.
2:24:32
I am going to argue that the
2:24:34
data is a cornered resource. I
2:24:37
don't know for sure about the model, maybe.
2:24:39
I mean, I guess that's the same thing
2:24:41
as saying the knowledge of what the 10
2:24:43
million lines of code does, that's the model.
2:24:45
But I actually think the fact that they
2:24:47
have clean data and they've been creating
2:24:49
systems, like they have the best PhDs in
2:24:51
the world thinking about data cleaning.
2:24:54
That's not a sexy job. And
2:24:57
yet they have probably the
2:24:59
treasure trove of historical market
2:25:01
data in the best format
2:25:04
that nobody else has. That's an actual
2:25:06
cornered resource. I have a couple nuances
2:25:08
on this. So one, I
2:25:10
think it probably is true that they have
2:25:12
better data than any other firm, thanks to
2:25:14
Sandor Strauss and the work that he started
2:25:17
doing in the 80s before anybody else was
2:25:19
really doing this. Yeah. And
2:25:22
other firms don't. That
2:25:25
said, certainly all the
2:25:27
other quant firms are
2:25:29
throwing untold resources at all this
2:25:31
too. Right. They
2:25:34
want to do this and money is not the issue. So in
2:25:37
chatting with a few folks about
2:25:39
this episode, I had more than
2:25:41
one person say to me, there's
2:25:43
two ways that RENTEC
2:25:46
could work. And
2:25:49
one version of how it works is
2:25:51
they discovered something 20 plus
2:25:54
years ago that is a timeless secret.
2:25:56
And they've been trading on that for 20 plus
2:25:58
years. Right. relationship between
2:26:00
types of equities that they've just been exploiting
2:26:03
and no one can figure out except them.
2:26:05
Right. And that may entirely be possible.
2:26:07
Isn't that crazy? Right. Now, RENTEC will say, they
2:26:09
will all say that is a hundred percent not
2:26:11
the way that it works. It's not that at
2:26:13
all. If that were the way that
2:26:16
it works, they would of course still say that
2:26:18
because they don't want anybody to know. Right. Don't
2:26:20
look at the relationship between soybean futures and GM.
2:26:23
Just don't do it. Right. So let's accept
2:26:25
that there is a possibility that that might
2:26:27
be true. More likely
2:26:30
though is that what RENTEC does
2:26:32
say is true, which
2:26:34
is no, there is no holy
2:26:36
grail. What we do here
2:26:38
is we completely reinvent the whole system continuously
2:26:41
on a two-year cycle. Two years is kind of
2:26:43
what I heard. The model
2:26:46
is fully restructured every
2:26:48
two years. It's not like on a date
2:26:50
every two years. It's being restructured every day,
2:26:53
but collectively it's about a two-year cycle. So
2:26:55
that would be an argument then that the
2:26:57
people actually could, with five people left, they
2:26:59
probably could go recreate it and all they
2:27:01
would need is the data. It's also an
2:27:03
argument that there is no actual cornered resource
2:27:05
here in terms of either the model itself
2:27:08
and maybe not the data either. I bet
2:27:11
the data is though. Let's say you've
2:27:13
been working there for 10 years. You
2:27:15
don't know how the 1955 soybean futures
2:27:18
data ended up in the database. Even
2:27:20
if you're used to using that data and you're
2:27:22
able to go recreate the model elsewhere, you don't
2:27:24
know how it originally found its way in. I
2:27:28
think that's fair. I think there might
2:27:30
also be some argument to the data that that
2:27:33
older data is helpful but its value
2:27:35
decays over time as markets evolve. Definitely.
2:27:39
The broader point I want to make here is just that every
2:27:41
other major quamp firm out there is also spending
2:27:43
hundreds of millions if not billions on this stuff
2:27:45
too. And people are looking for
2:27:47
alt data everywhere. The Bridgewater's of the world
2:27:49
are paying gobs of money for things that
2:27:51
you would never dream could possibly have an
2:27:54
effect on the stock market and yet they're
2:27:56
paying millions or tens of millions or hundreds of millions of
2:27:58
dollars for it. Yeah. So
2:28:00
I think we can rule out scale
2:28:03
economies for sure if anything their anti-scale
2:28:05
economies here Oh, yes,
2:28:07
there's totally there's diseconomies of scale your strategy
2:28:09
stop working when you get too much a
2:28:12
UM Yeah, you get
2:28:14
snippage. I don't think there's any network economies
2:28:16
here. I mean they literally don't talk to
2:28:18
anybody Although
2:28:21
well they do
2:28:23
have some very well-established
2:28:25
relationships with Electronic
2:28:28
brokerages and different players in the
2:28:30
trade execution chain. I think they
2:28:32
have very good trade execution and
2:28:34
very fast Market data
2:28:36
their ability to pull data out of the
2:28:38
market is very high quality Do you think
2:28:40
it's actually better than their competitors though? I
2:28:42
don't know. That's probably not the secret sauce
2:28:44
Yeah, I don't think so. It's the table
2:28:46
stakes switching costs. I don't think apply Branding
2:28:49
maybe applies in their ability to raise money
2:28:52
for the institutional funds But that's not a
2:28:54
big part of the business the fee stream
2:28:56
on the institutional fund may entirely belong to
2:28:58
branding Yes But I think
2:29:00
there's a lot of public equity firms and a lot
2:29:02
of hedge funds that have a lot of branding power
2:29:04
that have on average market
2:29:06
returns with decent sharp ratios and Are
2:29:09
able to raise because they've built a brand. Yeah
2:29:12
venture firms the same way totally So
2:29:15
for me this kind of leaves counter positioning I
2:29:17
actually think there's some counter positioning here and I
2:29:19
think we're gonna have two episodes in a row
2:29:21
of Counter positioning at scale
2:29:24
tell me about your counter positioning who is being
2:29:26
counter positioned in what way? They're
2:29:28
direct competitors in the market the other
2:29:30
quant firms and when I say direct
2:29:32
competitors I obviously don't mean for LP
2:29:34
dollars I mean for like the same
2:29:36
type of trading activity like their counterparties
2:29:38
in trades. I don't think
2:29:41
they are counter parties I think they
2:29:43
are all seeking to exploit similar types
2:29:45
of trades I think the counterparties are
2:29:47
the people there the dentist that they're
2:29:49
taking advantage of well But quant funds
2:29:51
are often counterparties to each other. That's
2:29:53
true. But I think yes adversaries in
2:29:55
finding the similar types of Trains
2:29:58
and I think the counter position for
2:30:01
RENTEC or for Medallion
2:30:03
specifically is one, I
2:30:06
do think the single model approach versus
2:30:08
the multi model, multi strategy approach that
2:30:10
most others have does have benefits
2:30:12
like I was talking about in the tapestries. But
2:30:15
I think also and maybe bigger
2:30:17
is every incentive at
2:30:21
RENTEC is fully aligned to
2:30:23
optimize fund size for performance
2:30:26
in a way that is not true just
2:30:28
about everywhere else. I
2:30:31
think they have the most
2:30:33
incentive of anybody to truly
2:30:35
maximize performance we're able to
2:30:37
achieve. Right, even though the
2:30:39
dollars would continue to rise because they get
2:30:41
fee dollars from more money in the door,
2:30:44
they are incentivized in a unique
2:30:47
way that makes it so
2:30:49
they're not willing to trade the dampener
2:30:51
on performance to get those dollars.
2:30:54
Yes, particularly because
2:30:56
it's all the same people on the
2:30:59
CP and LP side. Oh, you keep going around
2:31:01
and around that axis. I
2:31:03
loosely buy the counter positioning thing. I
2:31:05
just think the answer is disgustingly simple and
2:31:07
kind of annoying here which is they're just
2:31:09
better than everyone else at this particular type
2:31:12
of math and machine learning and they've been
2:31:14
doing it for longer so they're just gonna
2:31:16
keep beating you. Oh, that's another argument
2:31:18
I heard from people in that RENTEC
2:31:21
basically is a math department in a
2:31:23
way that none of these other
2:31:25
firms are. It could be culture. Yeah, it could
2:31:27
be culture. I mean, honest to God, it could
2:31:29
just be that the culture is set up in
2:31:32
a way that continues to attract the right people
2:31:34
and incentivize them in a sort
2:31:36
of fake altruistic way. Like this
2:31:38
is just a fun place to do my work and
2:31:40
yeah, the outcome is getting really rich but I
2:31:43
wouldn't go work at Citadel. Yeah, I
2:31:46
think that could be. So maybe that feeds into process power.
2:31:48
Yeah. Okay, for me, it is
2:31:51
some combination of process power and counter-persisting and I
2:31:53
don't think it's any of the other powers. For
2:31:56
me, it is process power and cornered resource.
2:31:58
Yeah, okay, I buy that. And a
2:32:00
thing that's not captured in Seven Powers is
2:32:03
tactical, like execution. The
2:32:05
whole point of Seven Powers is
2:32:07
strategy is different than tactics. And
2:32:09
I think legitimately, RENTEC
2:32:12
may just have persistently been
2:32:14
able to out-execute their competitors. There's
2:32:16
part of it that's just like they're smarter than
2:32:18
you. Yeah. Well, if you
2:32:21
buy the whole thing gets reinvented continuously
2:32:23
every two years, then yes. And
2:32:26
there's remnant knowledge. Like, if you started building
2:32:28
a machine learning system in 19, whatever it
2:32:31
was, 64, you're going
2:32:37
to be really good at machine learning today. And
2:32:39
the people that you've been spending time with for
2:32:41
the last 15 years, learning all of your historical
2:32:43
knowledge and working in your systems, are also going
2:32:46
to be better at machine learning than probably
2:32:48
the other people who are out in the world learning
2:32:50
it from people that just got
2:32:53
inspired to start learning machine
2:32:55
learning based on the new hotness.
2:32:58
So learning is compound is my answer. Great.
2:33:01
Okay. Playbook. So
2:33:03
in addition to the three-part David Rosenthal tapestry
2:33:05
that you have woven. I have nothing more
2:33:08
to add. There are a
2:33:10
handful of things that I think are worth hitting.
2:33:13
So the first one is signal
2:33:15
processing is signal processing is signal processing.
2:33:19
They, by not caring about
2:33:21
the underlying assets, they literally don't trade on
2:33:23
fundamentals except in the institutional fund when they
2:33:25
trade on fundamentals a little bit. They use
2:33:28
price earnings ratios and stuff like that in the
2:33:30
institutional fund, which is kind of funny because that's
2:33:33
a completely different skill set. But
2:33:35
if you just look at the daily and it's
2:33:37
all just abstract
2:33:40
numbers. You don't actually have to
2:33:42
care about what underlies those
2:33:44
numbers. You just have to look for
2:33:47
whether it's linear regression or any of
2:33:49
the fancier stuff that they do just
2:33:51
relationships between data. And
2:33:54
once you reduce it to that, it
2:33:56
is so brilliant that they
2:33:58
can just recruit from. any
2:34:00
field. It's not relevant how someone
2:34:02
has done sophisticated signal processing in
2:34:04
the past, whether it's being an
2:34:06
astronomer and trying to denoise a
2:34:09
quote-unquote photo of a star super
2:34:11
far away, or whether they've tried
2:34:13
to do natural language processing, it's
2:34:15
just signal. There's this really
2:34:17
funny line that Jim and Peter and
2:34:19
others will say when asked about why
2:34:22
they only hire academics and not from
2:34:24
Wall Street and whatnot, and they're like,
2:34:26
well, we found it's easier to teach
2:34:29
smart people the investing business than teach
2:34:31
investing people how to be smart. Right.
2:34:34
That's ridiculous. They don't teach anybody
2:34:36
anything about investing. They're just doing
2:34:38
signal processing. I bet at
2:34:40
least half the people at Red Tech on the
2:34:43
research side could not read a balance sheet. It's
2:34:45
so funny. It's a whole bunch of people who
2:34:47
are in the investment business, none of which are
2:34:49
investors. Yes. Another one that
2:34:51
you can decide if this fits or not,
2:34:53
I was thinking a lot about complex
2:34:55
adaptive systems. It's always been on my mind since
2:34:57
we had the NDS Capital guys on a few
2:35:00
years ago and read their work and the Santa
2:35:02
Fe Institute's work on this. In
2:35:04
a complex adaptive system, it's really difficult
2:35:06
to actually understand how one thing affects
2:35:08
everything else because the idea is the
2:35:11
relationships are so combinatorially complex
2:35:13
that you can't deterministically nail
2:35:15
down this one thing is the
2:35:17
cause of that other thing. It's the butterfly flapping
2:35:20
its wings. But there are
2:35:22
relationships between entities that you
2:35:24
can't understand or see on the surface. Do
2:35:26
you remember way back when we did our
2:35:28
second Nvidia episode, I opened with the idea
2:35:30
that when I was a kid, I always
2:35:33
used to look at fire and think like,
2:35:35
if you actually knew the composition of the
2:35:37
atoms in the wood and you actually knew
2:35:39
the way the wind was blowing and you
2:35:41
actually knew that, could you actually
2:35:43
model the fire? When I was a kid, you always
2:35:46
just assumed no. But actually the answer
2:35:49
is yes. This is a known thing
2:35:51
of what will happen when you light
2:35:53
this log on fire for the next three
2:35:55
hours and can you see exactly the flames. I
2:35:58
think RENTEC has basically They haven't
2:36:00
figured that out for the market. They
2:36:02
can't predict the future. But if they have
2:36:04
a 50.01% chance
2:36:07
of being correct, then they
2:36:09
can sort of take a complex adaptive system
2:36:11
and say, we don't really care that it's
2:36:13
a complex adaptive system. Our models understand enough
2:36:15
about the relationships between all these entities that
2:36:17
we're just going to run the simulation a
2:36:20
bunch of times, and we're going to be
2:36:22
profitable enough from all the little pennies that
2:36:24
we're collecting on all the little coin flips
2:36:26
where we have a slight edge over and
2:36:28
over and over and over again that they're
2:36:31
sort of the closest in the world to
2:36:33
being able to actually predict
2:36:35
how the complex adaptive system of
2:36:37
the market will work. Now,
2:36:39
I don't think they can back out to it. No
2:36:41
person could explain it, but I think their computers can.
2:36:44
Yes. And I think when I've heard
2:36:46
people from RENTEC talk about this, they
2:36:49
will all say, the model does
2:36:52
not actually understand the market,
2:36:55
but it can predict
2:36:57
and we can be
2:36:59
so confident in its predictions about
2:37:03
what the market will do that we
2:37:05
rely on it, whether it understands or
2:37:07
doesn't understand, doesn't actually matter.
2:37:09
Like, it can't tell you why,
2:37:12
but that's okay. But it does know it has a
2:37:14
slight edge, and so it should trade on it even
2:37:16
though it can't explain why. Yes. Now,
2:37:18
speaking of models, I've been trying to nail
2:37:21
down an answer to this question. Do you think
2:37:23
RENTEC was the birthplace of machine learning? This
2:37:26
is such a tough answer to tell.
2:37:28
We actually emailed some friends who are
2:37:30
very prominent AI researchers and AI
2:37:33
historians and sort of asked this
2:37:35
question. And the answer
2:37:37
we got back is unsurprising. They said, we
2:37:39
don't know because they don't share anything. Right.
2:37:43
It's like the principle certainly came out
2:37:45
of the same math community
2:37:47
that spawned machine learning, but is
2:37:50
what RENTEC has figured out over
2:37:52
the last couple decades in
2:37:55
Google's Gemini model and in chat. No,
2:37:57
it's not because they don't contribute any
2:37:59
research back. It may be the case
2:38:01
that actually, RENTEC has beat everyone
2:38:03
else to the punch, and they have a
2:38:05
strong AI or something that is actually much
2:38:07
more sophisticated than all the AI we have
2:38:10
out in the world today. And they've just
2:38:12
chosen that they'd rather keep it locked up
2:38:14
and captive and make a bunch of money.
2:38:17
I mean, it could just be the case that Renaissance
2:38:19
is just taking in as much
2:38:21
unstructured data as it possibly can.
2:38:24
And they sort of were just a decade
2:38:27
or two ahead of everyone else in realizing
2:38:29
that you can have unstructured, unlabeled
2:38:31
data. And if you have enough of it,
2:38:33
you can make it, in the case of
2:38:35
an LLM, say things that sound right or
2:38:37
sound true, or in the case these trades
2:38:40
be right more than 50% of the time. Right.
2:38:42
Make trades that sound right. Right.
2:38:45
They figured out this big unsupervised learning
2:38:47
thing before anybody else all the
2:38:49
way up until last year when the AI moment happened. If
2:38:52
that were the case, we should have very different answers to
2:38:54
powers. To illustrate this point, it's
2:38:56
quite interesting. Peter Brown's academic
2:38:58
advisor was Jeffrey Hinton. Yes.
2:39:01
Oh, I'm so glad we brought this
2:39:03
up. Yeah. It was the exact same
2:39:05
stew and the exact same cohort of
2:39:08
people and social group and academic groups
2:39:10
that RENTEC came out of, that AI
2:39:12
came out of. The other person, just
2:39:14
for people who are like, why are
2:39:16
you saying that? To make it super
2:39:19
explicit, the other person whose academic advisor
2:39:21
was Jeffrey Hinton is Ilya Sutskever, who
2:39:23
is the co-founder of OpenAI. I mean,
2:39:26
many years later, but still. Yeah.
2:39:28
I mean, it's like we were talking about with Markov
2:39:30
models and hidden Markov models. That
2:39:33
is the foundation of RENTEC. That is one
2:39:35
of the foundations of AI and generative AI today. Okay.
2:39:39
Another big one is this concept
2:39:41
that you should trade on a
2:39:43
secret that others are not trading
2:39:45
on. So on the face
2:39:47
of it, it seems obvious. Of course, I should
2:39:49
come up with some strategy to trade on that other people aren't trading on.
2:39:52
But I said a couple of words there,
2:39:54
which is, of course, I should come up
2:39:56
with, and therein lies the fallacy.
2:40:00
I think most investment firms
2:40:02
try to get their ideas out of people
2:40:05
and then do an incredibly rigorous amount of
2:40:07
data analysis to figure out if
2:40:09
they should put those trades on or not. I
2:40:12
could be wrong, but I do not
2:40:15
think modern RENTEC does that. I
2:40:17
think all of their investment ideas come
2:40:20
from data and come from signal
2:40:22
processing. And so therefore, you are
2:40:25
going to put trades on that
2:40:27
make no intuitive sense. And
2:40:30
so when you're putting trades on that
2:40:32
are profitable and make no intuitive sense,
2:40:34
you aren't going to have competitors. If
2:40:37
you find a relationship between two things that a
2:40:39
human could never come up with or dream of
2:40:41
those relationships, and we're saying two, it ends
2:40:43
things, 10 things, 20 things, 100 things,
2:40:46
and in various different ways at various different
2:40:49
time scales, that is a
2:40:51
killer recipe to exploit a secret that no one
2:40:53
else knows and be able to beat other
2:40:55
people in the market. Such a good point.
2:40:58
And many, if not most of the other
2:41:00
quant firms are not doing that. Some
2:41:03
of them maybe, but I think
2:41:05
most of them are the model
2:41:07
is suggesting things. And there
2:41:09
is a person or persons who are the master
2:41:12
portfolio allocators that pull the trigger
2:41:14
or don't pull the trigger. Yes.
2:41:17
And to be super illustrative, because I think your
2:41:19
natural tendency is like, oh, I can understand
2:41:22
why these two things would be related. The
2:41:24
relationship may not be what you figure. For
2:41:27
example, there could be two things that
2:41:29
always move together, Tesla stock and wheat
2:41:31
futures. And you might try
2:41:34
to, because humans are storytellers, concoct some story in
2:41:36
your head of why those move together. And
2:41:39
if you believe it, then you might decide there's
2:41:41
some date where they should stop moving together. Well,
2:41:43
it could very well be that some
2:41:45
other big hedge fund just owns both
2:41:47
of those things. And when they rebalance,
2:41:50
it causes those assets to move together.
2:41:52
But you would never think of that.
2:41:54
You would think these things have a
2:41:56
direct relationship with each other, not just
2:41:58
that there's liquidity in the market. from
2:42:00
both of them at the same time because someone else
2:42:02
owns both of them. So I think what RENTEC sort
2:42:04
of admitted is, we have no idea
2:42:06
why anything is actually connected, but it doesn't matter.
2:42:09
Yep, totally. And that
2:42:11
was surprising for me in the research. Like I sort
2:42:13
of assumed that was the whole quant industry. And
2:42:16
it was very surprising to me to discover
2:42:18
that I believe, no, it
2:42:21
is pretty much only RENTEC and maybe
2:42:23
a couple other people. Okay, my
2:42:25
next one is brought to you by a
2:42:27
friend at the show, Brett Harrison, who has
2:42:29
worked in the quant trading industry for a
2:42:32
long time and shared an idea that he
2:42:34
has with us, which is that there's basically
2:42:36
this two by two matrix. You have on
2:42:38
the one axis fast and slow in terms
2:42:40
of trade execution. And on the
2:42:42
Y axis, you have smart versus
2:42:44
obvious. Yeah, the way he phrased it
2:42:46
to us was smart versus dumb, but
2:42:49
dumb doesn't mean dumb. Right, it's the
2:42:51
obvious trades. And the high level
2:42:53
point is all quant funds are not
2:42:55
high frequency trading firms and vice versa.
2:42:59
not coming from this industry and now makes total sense
2:43:01
to me. I think I thought they were the same
2:43:03
thing, but fast and
2:43:06
obvious is your classic high frequency trader.
2:43:08
They're front running trades. They're locating in
2:43:10
a data center that's really near the,
2:43:13
this is flash boys, or they've got a
2:43:15
microwave line between New Jersey and Chicago, and
2:43:18
they're trying to arb the difference between two
2:43:20
markets. You need to have the fastest connectivity
2:43:22
in the world to pull this off. Yep,
2:43:24
this is Jane Street. Yes, there's
2:43:27
fast and smart, which
2:43:29
you kind of don't need to be
2:43:31
both. You don't need the fastest connectivity
2:43:33
in the world and the most clever
2:43:35
trades to put on. So people kind
2:43:37
of tend to pick a lane that
2:43:39
they're either a high frequency trader or
2:43:41
they're trying to make the smartest, most
2:43:43
non-obvious trades possible. And that of course
2:43:45
leads us to Medallion, which is in
2:43:47
the slow and smart quadrant.
2:43:50
All the machine learning systems discovered the
2:43:52
relationships in the data. So there's a
2:43:55
huge amount of compute. The non-obvious trades.
2:43:57
Exactly, that goes into finding the non-obvious
2:43:59
trades. obvious trades, but then they're actually
2:44:01
made reasonably slowly. They still have to
2:44:03
happen within seconds or minutes, but the
2:44:06
advantage isn't that they're high frequency, the
2:44:08
way that all the Flash Boys stuff
2:44:10
is. My sense is, RENTEC
2:44:12
is not a high frequency trading shop.
2:44:14
They are not front running things.
2:44:16
They are not Flash Boys. Compared
2:44:19
to you and me, they still
2:44:21
operate incredibly fast, but
2:44:23
it's more about the smartness and
2:44:25
less about the fastness. Greg
2:44:28
has a quote in his book, they hold thousands of
2:44:30
long and short positions at any given time and
2:44:32
their holding period ranges from one to two days
2:44:34
or one to two weeks. They make
2:44:37
between 150,000 and 300,000 trades a day,
2:44:39
but much of that activity entails buying
2:44:41
or selling in small chunks to avoid
2:44:43
impacting market prices rather than profiting by
2:44:45
stepping in front of other investors. Oh,
2:44:48
this is another thing that we heard. RENTEC
2:44:51
is world class at disguising
2:44:53
their trades. They can
2:44:55
make it so that they don't move
2:44:57
the market and you don't know
2:44:59
who is acting or when. This
2:45:01
is because in the early days, they weren't good
2:45:03
at this. People basically intercepted the trades that they
2:45:05
were making and were front running them. They had
2:45:08
to adapt and develop these clever systems to make
2:45:10
it so you don't know who's buying and you
2:45:12
don't know in what quantities and you don't know
2:45:14
if they're going to keep buying. My
2:45:17
last one before we get into value creation, value
2:45:19
capture is that this is a terrifying business to
2:45:21
be in. The amount of controls
2:45:24
and risk models that you need and kill
2:45:26
switches are just so important. What if the
2:45:28
software has a bug? Is it possible to
2:45:30
make a ton of unprofitable trades in a
2:45:32
matter of minutes and lose it all? That
2:45:34
wasn't possible in the old world where you're
2:45:36
calling your broker. That totally is possible here.
2:45:39
It happened. Yeah. While it's never happened
2:45:41
to RENTEC, there was a company called
2:45:43
Night Capital in 2012
2:45:46
that lost $460 million in
2:45:48
a single day. There was a bug
2:45:50
in their process to deploy the new
2:45:53
code. Basically, what happened was a
2:45:55
simple flag error, a misinterpretation of
2:45:57
setting a bit from zero. zero
2:46:00
to one that caused this infinite loop to
2:46:02
run where once a certain trade happened, it
2:46:04
was supposed to flip the bit. It flipped
2:46:06
a different bit. The systems were not looking
2:46:08
at the same location and memory for the
2:46:10
same bit. And so it basically thought it
2:46:12
was never flipped. This infinite loop ran 4
2:46:15
million trade executions in 45 minutes and
2:46:17
there wasn't the appropriate kill switches built in
2:46:20
and they basically watched it all to just
2:46:22
drain out and there was nothing they could
2:46:24
do. Yeah. So like the
2:46:26
whole portfolio gone, right? Yes. Well,
2:46:28
I don't know if it's the whole portfolio, but it
2:46:30
was enough that they lost a huge
2:46:32
amount of the LP capital and then
2:46:34
they were a publicly traded firm overnight,
2:46:37
their equity traded down 75% and then
2:46:39
someone stepped in and bought them. Well,
2:46:41
they probably got marching called by all
2:46:43
their counterparties. So whoever
2:46:45
is in charge of the financial controls
2:46:47
and safety systems at RENTEC, that's a
2:46:49
huge job for someone in this industry.
2:46:52
Totally. All right. To
2:46:55
kick off value creation, value capture, I
2:46:58
have a provocative statement, which is, David,
2:47:01
Renaissance Technologies is actually not in
2:47:03
the investment business. They are in
2:47:05
the gambling business. And in
2:47:07
particular, they're the house. Well,
2:47:09
I thought where you thought you were going with this,
2:47:12
I was like, yes, I would totally agree they're not
2:47:14
in the investment business. They have no idea how to
2:47:16
invest. The model does. I'll say this, they're not investors
2:47:18
and they're not in the investment business. There
2:47:20
is investment going on all around them in
2:47:23
the markets that they trade in. But
2:47:25
the fact that they're in those markets, they're
2:47:27
not there as investors. They're
2:47:29
there setting up shop as
2:47:32
Caesar's Palace, letting everyone come in and
2:47:34
do business with them while they have
2:47:36
a slight edge. And they'll lose sometimes,
2:47:38
but most of the time they're going
2:47:40
to come out slightly ahead. And
2:47:43
I think, let's say they do have a 50.01%
2:47:45
chance of being right. They're
2:47:48
just there to collect their vig on
2:47:51
everyone who is willing to trade with
2:47:53
them over all these years. And
2:47:55
at scale, it really worked. Jim Simons
2:47:57
managed to drain $30 billion into his
2:48:00
own pocket out of everybody that he ever
2:48:02
treated with. Now, I think
2:48:05
where you're going with this
2:48:07
is perhaps similarly along the lines
2:48:09
to Caesar's Palace or a casino.
2:48:13
They are not in the investment business, but they are
2:48:15
providing a service. Sure. Is
2:48:17
this where you're going with this? Well, I
2:48:19
mean, the investment business, it sort of depends
2:48:21
how you define investor. If you want to
2:48:23
be like all hoity-toity about it, which I'm,
2:48:26
you know, in this illustrative example of kind
2:48:28
of being one and saying an investor is
2:48:30
someone who provides capital, you know, risk capital
2:48:33
to a business for that business to create
2:48:35
value in some way in the
2:48:37
future or you lend money to some intrinsic
2:48:40
underlying assets so that it can be productive
2:48:42
with that capital and produce a return for
2:48:44
you as an investor. And of
2:48:46
course, lots of things are called investing
2:48:48
that are not that. Is it investment if I
2:48:51
put money to work and then I get more
2:48:53
money back later and I don't actually care how
2:48:55
the money got made and it's actually zero sum.
2:48:57
I'm just vacuuming it out of right. Right. Yeah.
2:49:01
The money is not being invested in anything to
2:49:03
produce. Correct. But it's literally the same business model
2:49:05
as a casino. You have a slight edge and
2:49:07
you let a whole bunch of patrons come in
2:49:09
and lose money to you in your slight edge.
2:49:12
Well, where I was going with the service provider, I
2:49:15
think casinos are service providers. They are
2:49:17
providing entertainment to their customers. Everybody
2:49:20
knows that the games are stacked in the
2:49:22
casinos favor. Similarly, I think you
2:49:24
could make an argument and I think this is
2:49:26
probably quite accurate that RENTEC
2:49:29
and all other quant firms like
2:49:31
them are providing a service to
2:49:34
the market in that they are
2:49:36
allowing trades that people want
2:49:38
to make to happen faster and
2:49:40
at much lower spreads. Absolutely. That
2:49:43
is the undeniable yes,
2:49:45
quant funds create value in the world
2:49:47
thing, which I think it's very easy
2:49:49
to say quant funds provide no value
2:49:51
because it's like it's zero sum. They're
2:49:54
not actually providing the capital to businesses to
2:49:56
do something with they're purely looking to do
2:49:59
an arbitrage or. any of the
2:50:01
strategies we've talked about this episode. But
2:50:03
you're totally right that there is
2:50:05
a value to market liquidity. Creating
2:50:07
more depth to a market makes
2:50:09
it so that if we go
2:50:11
back to the era that Renaissance was started, there's
2:50:13
no chance that retail is able to
2:50:16
function like it does today with zero
2:50:18
transaction fees and people able to invest
2:50:20
in all these different companies at near
2:50:22
real time. And any single one of
2:50:25
us can go buy a
2:50:27
security in
2:50:29
just about any market, at just
2:50:31
about any time of day, pretty
2:50:34
much instantaneously and
2:50:36
get a very, very, very granular price
2:50:38
on it. Yup. None
2:50:40
of which used to be true. Nope. The
2:50:43
fact that there is a whole bunch of
2:50:45
quant funds, hedge funds out there that are
2:50:47
ready to be willing counter parties to anyone
2:50:49
who wants to trade, that is a service,
2:50:51
you're right. They're also not
2:50:53
all medallion. They actually don't all
2:50:56
have an edge, even though they might purport
2:50:58
to. Lots of them are gonna lose money
2:51:00
to you. Right. Lots
2:51:02
of them lose money. You too listeners could
2:51:04
beat the market. Not investment advice, please don't
2:51:06
try. Right. On average,
2:51:08
medallion will not lose money to you, but
2:51:10
there are plenty of other hedge funds out
2:51:12
there and high frequency shops and counter parties
2:51:14
for you where you could take them.
2:51:17
That's just not Jim Simons. Ha
2:51:19
ha ha ha. Ha ha ha ha. There's
2:51:23
this great, great vignette at the end of
2:51:25
Greg's book. It was during one
2:51:27
of the sell offs in the mid 20 teens in the
2:51:29
market where Jim calls the head
2:51:32
of his family office. He's
2:51:34
long retired from RENTEC at this point, calls the
2:51:36
head of his family office and says, what
2:51:38
should we do with all the sell offs in the market? It's
2:51:40
like, you're Jim Simons. Right. You're
2:51:43
Jim Simons, what should we do? What should we
2:51:45
do? Ha ha ha ha. All
2:51:49
humans are fallible. Totally. A
2:51:52
couple of other squintable, the value
2:51:54
creation exists. It's easy to
2:51:56
knock that all these smart people are going into
2:51:58
finance and you will. wish they were doing something
2:52:00
more productive for the world. At
2:52:03
the end of the day, humans are going to do
2:52:05
what they are incented to do. And
2:52:07
so absent a larger global concern that
2:52:09
is incredibly motivating to people. I mean,
2:52:11
you look at World War Two, people's
2:52:14
level of patriotism and wanting to go save
2:52:16
the world from evil was a huge,
2:52:18
unbelievable motivating factor to move mountains. When
2:52:22
that is absent, or when people feel that
2:52:24
there's some existential thing that is absent, they're
2:52:26
going to go do what's best for them
2:52:28
and their family and if they're an empire
2:52:30
builder, go build empires and if they're a
2:52:32
fierce capitalist, go make a bunch of money.
2:52:35
And so the system is set up
2:52:37
the way that it is. So like you can be mad about that.
2:52:40
Given that, okay, people
2:52:42
are going to go engage in quantitative finance
2:52:44
as a lucrative profession. Fortunately,
2:52:46
there's a bunch of valuable stuff that comes
2:52:48
out of that. And I think that is
2:52:51
often missed is that
2:52:53
these really lucrative professions
2:52:56
and businesses can often produce
2:52:58
R&D that becomes valuable elsewhere.
2:53:01
For example, we just did
2:53:03
this big Nvidia series. What
2:53:05
do you think Mellanox was used
2:53:07
for before large language models? Oh,
2:53:09
yes. This is such
2:53:12
a really mind blowing point
2:53:15
here in value creation, value capture. Go
2:53:17
for it. Take it away. Well, there's not much to it other
2:53:19
than a huge amount of
2:53:22
InfiniBand was used by high frequency trading
2:53:24
firms. And I don't know for sure.
2:53:26
But I kind of think Mellanox built
2:53:28
their business on quant finance. That's
2:53:31
one of many examples. But now, you
2:53:33
know, that has limits. But
2:53:35
I think it goes overlooked that there's
2:53:37
a lot of technology innovation here. Yeah,
2:53:40
these are all great points. They all came up
2:53:43
in the research. I totally
2:53:45
agree with all of them. It is,
2:53:47
in my opinion, false to
2:53:49
say that quantitative
2:53:52
finance does not create
2:53:55
value for the world. It definitely
2:53:57
does, in my opinion. But does it create anywhere in
2:53:59
the world? as much as it captures. That's
2:54:02
it. They're
2:54:04
really, really good at value capture. Yes.
2:54:07
This is not Wikipedia here. This is about
2:54:09
as far away on the spectrum as you
2:54:11
can get. There's a great Always Sunny in
2:54:13
Philadelphia where Frank, Danny DeVito, sort of goes
2:54:15
back to his whatever business he founded in
2:54:18
the 80s and he's like dressing in his
2:54:20
pinstripes and stuff again and he's taken back
2:54:22
over. He brings Charlie with him and Charlie,
2:54:24
you know, he's like, so Frank, what is
2:54:26
the business, what do we do here? What
2:54:28
does the business make? And Danny DeVito
2:54:30
looks at him and he goes, what do you
2:54:32
mean? We make money. He's like, no, no,
2:54:35
like what do you build? He goes, we build wealth.
2:54:38
I think that's a pretty good meme for kind
2:54:40
of what's going on here. Yeah, totally.
2:54:43
Very, very good at value capture too. Yes. Okay,
2:54:46
Bear Bull. So this was a section that we
2:54:48
had for a long time that we did not
2:54:51
put in the last episode and boy did we
2:54:53
hear about it. So listeners, thank you so much
2:54:55
for expressing your concern. Bear versus
2:54:57
Bull is unkilled and it is
2:55:00
back. Resurrected, like a Phoenix. Resurrected.
2:55:03
However, this is about the lamest episode to
2:55:05
resurrect it on. What's the bull case for
2:55:07
RENTEC? Past performance is an indicator of future
2:55:09
success. Right, like they're gonna keep attracting all
2:55:11
the smartest people in the world. They're gonna
2:55:13
have the ability to keep their
2:55:15
incredibly unique culture. They're
2:55:17
not gonna get tempted to let
2:55:20
the business of institutional funds become
2:55:22
the dominant business. Keep
2:55:25
on keeping on is basically the bull
2:55:27
case. Maybe that they're actually still ahead.
2:55:29
The bull case for the
2:55:32
GP and LP stakeholders in
2:55:34
Medallion, which is, I don't know, 500
2:55:37
people in the world. And none
2:55:39
of the rest of us can get any exposure to it. Yeah,
2:55:42
the bear case is things are changing.
2:55:44
And I think things are changing basically
2:55:46
on any axis is the bear case
2:55:48
for them. So things are changing where
2:55:50
competitors are catching up. Maybe
2:55:53
the fact that the tech industry
2:55:56
has figured out these large language
2:55:58
models, maybe that trickles in. to
2:56:00
making it easier to compete with RENTEC. It's
2:56:02
a blurry line, but it is plausible. Like
2:56:04
maybe RENTEC actually was here a decade before
2:56:06
everyone else, and now everyone else has arrived
2:56:08
to the party. There's things
2:56:11
that are changing maybe about their culture.
2:56:13
Like Jim Simes has been gone for
2:56:15
a long time. Bob Mercer is no
2:56:17
longer a co-CEO. Peter Brown is a
2:56:19
co-CEO, and they just announced that they're
2:56:22
making the guy who was in charge
2:56:24
of the institutional funds. David
2:56:26
Lippie, he is becoming a co-CEO as
2:56:28
well. So maybe there's a bear case
2:56:30
around that, that someone from the institutional
2:56:32
side of the house is
2:56:34
becoming the current co-CEO and maybe eventually
2:56:37
CEO if you believe the medallion is
2:56:39
the special thing and the institutional funds
2:56:41
are sort of a blemish on the
2:56:43
business. You know, they're the Hermes Apple
2:56:46
Watch strap in David's parlance.
2:56:49
Maybe that's a bear case. Maybe
2:56:51
there's a bear case that their talent is
2:56:54
becoming kind of the same as everyone else's
2:56:56
talent. When you look on LinkedIn, I
2:56:58
recognize a lot of the companies that
2:57:00
people worked at who are more junior
2:57:02
at RENTEC. And in the past,
2:57:04
I think it would have been all people
2:57:06
just out of university research shops. So I
2:57:08
think if it's true that
2:57:10
they're starting to see the same talent
2:57:12
flow as everyone else, that would be
2:57:14
concerning. These things are all sort of narratives
2:57:17
you can concoct and really no way to know if they're true
2:57:19
or not. There's no way for
2:57:21
us to know any of this because there's no way to know
2:57:23
any of this. Right. It's all
2:57:25
the secret. Yep. Okay.
2:57:28
Our new ending section, the
2:57:30
splinter in our minds, the takeaway... The
2:57:33
one thing you can't stop thinking about.
2:57:35
What is the one thing for each
2:57:38
of us personally from
2:57:41
doing this work over the past month on RENTEC
2:57:44
that sticks with us? For me,
2:57:46
perhaps this is obvious from my little
2:57:49
diatribe on the tapestry. I just
2:57:51
think this is such a
2:57:54
powerful example of the power of
2:57:56
incentives and getting them right and
2:57:58
setting them up right. And
2:58:00
culture too. I don't want to shortchange
2:58:02
that. I think the culture of managing
2:58:05
an academic environment in a
2:58:07
fashion like a lab, but without
2:58:10
letting it spin into the frivolity
2:58:12
of a lab that Jim Simon
2:58:14
set up. Right. In other words,
2:58:16
early Google. Yeah, this is like
2:58:19
early Google. Exactly. There
2:58:21
historically has not from our research, and as
2:58:23
best as we can tell currently is
2:58:26
not anything going on at RENTEC
2:58:29
that is frivolous. They
2:58:32
are all very focused, which again to me
2:58:34
then speaks back to the power of incentives.
2:58:37
When you're there with less
2:58:39
than 400 people and on the research and
2:58:41
engineering side, less than 200 people
2:58:44
and those colleagues who you work
2:58:46
with are the sole purveyors,
2:58:49
supervisors and beneficiaries of
2:58:53
all of this that you're doing, that is
2:58:55
so powerful. I can't think
2:58:57
of anywhere else like that in the world. I mean maybe some
2:59:00
venture funds or other investment firms, but
2:59:02
not on a day-to-day fully
2:59:04
liquid with returns like this.
2:59:07
There's nothing like it. Pure
2:59:10
gasoline right into the veins. Yeah,
2:59:12
which is not to say I would necessarily want to work there.
2:59:15
I think I would not, but it is
2:59:17
truly unique. The
2:59:19
one thing I can't stop thinking about is
2:59:22
the idea of the complex adaptive system that
2:59:24
I was talking about earlier. I
2:59:26
think from what everything we can
2:59:28
tell from the outside, Renaissance actually
2:59:30
has built a large scale computer
2:59:32
system that discovers relationships
2:59:35
between different entities
2:59:37
in the world. Stocks, commodities,
2:59:40
bond prices, and whether
2:59:42
it can explain them or not, it is
2:59:46
correct most of the time. It might be
2:59:48
a small most, but all you need is
2:59:50
most and then you can operate a casino
2:59:52
business. That is my takeaway, is that
2:59:54
they are the house and they have
2:59:56
an edge and that edge is predicated on
2:59:58
a graph.
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More