Hey, merry Christmas and thank you so

much for listening to Odds on Open. I

hope you enjoy this episode.

>> Bill, thank you so much for doing this.

>> Oh, great. Glad to be here, Ethan.

Thanks.

>> You use both a discretionary and a

systematic approach to investing at 10

Dynamics. Can you walk me through that?

>> Sure. But it's it's um I guess the

discretionary component is brand new for

us really. Um know we spent the last

five years being very dedicated to and

automate everything as much as possible

and and really be as low touch as

possible.

Um and that's I think that approach is

still a very valid uh way to at least

start off with systematic.

Um and you have this I guess just a

debate in general about you know

technology. Are we are we building

technology to replace us or we building

it as a tool? And I think we have been

down the road pretty far down the road

about building technology to replace us.

Um but recently started to

run into a situation where we felt like

we were able to

using our own sort of human input uh

make improvements on the decision the

system was making. Um and so we kind of

came up with a few um particular uh

situations where we thought that was

true and we did a lot of testing to see

if can we implement our own like what

what's our decision-m process behind the

scenes like can we can we codify that

somehow um and haven't really figured

out how to do that and yet we still seem

to be able to identify these these times

where you know for instance I think the

big one is when when is should the

system not be trading at

Um I think you can you know we're doing

trend following right so that's our

that's our approach and you know I think

we've we we perform well on on uh trend

following in general and we like to

apply it to lots of different assets to

get the diversification which is great

but you can really I think just look at

certain markets and just tell that trend

following is not you know markets like

for inance a market that's in

significant over supply and is really

low priced right so it's it's unlikely

to go any lower uh but it's also overs

supplied so it's unlikely to go higher

and what you kind of get is a very you

know jelpy price path or in a very sort

of volatility can even be kind of high

in that situation but no real sustained

trend where you know you whether you

either just look at you know look at a

visually how the market has behaved

recently or even if you do a little bit

of you know investigating in the

fundamentals you can identify well this

is not really a market where trend

falling is going to be successful right

and so Um, so this is this is a bit of a

departure from us philosophically, I

guess. Um, we took, you know, I read an

article quite a while ago, and I don't

know if it's still true, that the best

chess players in the world are AI

augmented, right? That they're they're

humans who use um sort of the chess, you

know, uh, AI or or even the older not

AI, but actual just chess um chess

expert systems to compete and that

that's better than the expert system

alone, right? And so I don't know if

that's still true today, but it seems to

to have been true in the past. And I

like that idea of, you know, is there a

partnership between humans in technology

that is better than either one in

isolation? And and it's an interesting

test for us. I think it's a it's a

really big test for for humanity really

is what we're facing going forward is is

the combination of of artificial

intelligence and humanity going to be

better than or is AI just going to

replace us? I I you know certainly

remains to be seen on that front. But um

so yeah so this is this is a is a good

time to do this this podcast actually

because you know it is a bit of a shift

for us um as a firm and and uh so now we

run both we run our systematic strategy

and we have a new strategy which is the

application of a discretionary filter

effectively over the top of the um over

the system and our initial results are

are actually very promising. So so we're

happy with it. still would love to be

able to codify it somehow, but um you

know we haven't been able to to find the

magic that that does that yet.

>> How do you go about

judging the past performance of

uh you know discretionary filter

overlay rather on top of a systematic

strategy. So in this case, you know,

with with trend, um obviously you can

look to the past and say, "Oh, maybe I

would have probably would have done that

or that." Uh and and you can do that

with a lot of hiding sight bias. How do

you think about that?

>> Well, it's it's interesting. I think you

you know, you really can't at the end of

the day. You can't really there is no

quote unquote back testing of

discretionary um strategy. So you have

to have, you know, some sort of

philosophy, I guess, that you're basing

it under and some belief that it's going

to work. And um and then it just comes

down to experience. And what's what's

interesting in the quant space and the

systematic space is nobody except the

people doing back tests believe back

tests. So you can build all the back

tests you want and you know justify how

you know there's no look ahead bias.

there's this that and the other thing

and no investor believes it or very very

a few investors believe it. So all that

really matters from a systematic

perspective from an investor point of

view is your track your actual track

record with live money which really puts

you in the same ballpark as then a

discretionary investor. You know all

that matters for a discretionary person

is what their track record is with

managing live money and and the story

that they tell around that I guess this

is helpful from a sales point of view

but but in reality you know you can

really only judge by how their their

track record has been. So, so really the

the back testing element is how you

convince yourself more than it is about

how you convince the outside world

because the outside world is really

going to care about your your actual

performance. And um I was I guess we

were maybe surprised or a little bit

disappointed that that was the case on

the systematic side because you know

there's um you hope that people believe

your back test. But I I you know you can

see if your shoes on the other foot

you're investing. You have no idea what

what an investor's claiming is true and

that they they tested the system the way

they said or they might even have a

mistake in their code that they don't

even know about, right? They might have

some forward-looking bias that they

haven't seen or don't know or

understand.

Um so yeah so for for us I think the

confidence came from trading the system

for 5 years and and really feeling like

you know watching it trade certain

markets and just being frustrated oh

this is you know the system's just not

going to make any money in this market.

It's quite obvious and being right about

that time and time again or being right

about the markets that are really

performing. So you know if you look at

the last couple years

all of the excitement and trend has been

in sort of the eggs and you know those

sort of meats and a little bit soft. So

things like coffee, cocoa, uh life

cattle, uh orange juice that's pretty

liquid, but orange juice certainly

moved. Lumber lumber seems like it's

always moving. It's still a liquid but

but that's where the big trends have

been. You know, I think you look at

coffee went to all-time highs and um

cocoa similarly was was a big driver in

some of the performance and you could

you knew those were going to be good

markets. So, you know, not only did we

know you shouldn't be trading some of

the things like rates have been really

tough the last, I'd say two years is,

you know, people have been trying to

decide what's going to happen with the

with the rate cycle. But, um, you know,

being able to say, well, actually, we

probably really shouldn't be trading

rates or being have have much risk

allocated to rates at the moment, but

put more rate. you know, there's just

not a lot of of kind of the softs and a

there's not a there's not a huge you

look at the kind of broad buckets in

commodities. Energy is probably your

broadest space, particularly if you do

Europe like we do. So, you have lots of

commodities and you get a decent um you

know different products, you get some

diversification. So, that tends to

dominate any equal weighted portfolio is

going to have an energy tilt to it

naturally if you're just trading

everything that's that's liquid, you

know. So um so yeah so we started to

feel really like gosh we were seeing

things operating the system that maybe

weren't obvious when we built the system

so much but still you know as I said we

haven't really been able to figure out

what it is about what are we doing when

we look at a market and just see that

okay the volatility might be high but

the trend is the trend it's just not

going to trend right it's just doesn't

have the constructive behavior

um and it's you know I mean you probably

don't have to go I mean, you go back as

far as you want and everyone will tell

you the key for trend following is

identifying when you're in a range,

quote unquote, and when you're in a

trend. If you can identify those two

things, that's that's how you make all

your money. And we we the way we apply

our model systematically, we do attempt

to do that. So, that's our, you know, we

we've figured out a way to to doing

that. But even then, um it's like the

it's like we should exaggerate the model

positioning more than it is. So, you

know, if if we're relatively lightly

positioned in a market, we probably

shouldn't trade it at all. Maybe you

that's what that's effectively what

we're doing. We're saying that, you

know, Mario, you know, this this product

is is because we're we haven't really

talked about we scale in and out of our

position, right? So, we're constantly

adjusting our position um based on fixed

risk allocations across the different

assets and um which is nice from a trade

cost point of view and I think it's it's

really what the model is built to do. Um

but really what we're thinking is you

know that that markets that are trending

is where our model seems to be deployed

highly and we should actually have it

more highly deployed there. So that's

what our discretionary uh system is

doing is putting more risk on on the

markets where the model is positioned

anyway and and avoiding markets where

positioning is quite light.

>> And before that you were fully

systematic. Can you walk me through um

those those strategies you were running?

>> Yeah, it's so we run a single called a

single strategy. So our at least our

philosophy is that

you know price call price dynamics or

price movements or whatever that if it's

driven by human psychology which is kind

of what we think is going on um that it

should really operate across all markets

in the same way and it should also

operate across different time horizons

in the same way um within limits. you

know, if if we're too, you know, we

don't do high frequencies, so we don't

do anything shorter than like hourly

type type inter, you know, um, sampling.

Uh, and then, you know, out past weekly,

it gets to be pretty long time between,

you know, trade decisions. So, it's hard

to, you know, hard to assess whether

it's it's working or not. So um we apply

the same set of column signals uh to

every market and within each market we

have multiple time frames and we use the

same signals independently on each time

frame. So the idea being kind of a you

know like a lot law large numbers thing

we're trading lots of markets we're

trading lots of different time frames.

The idea being that if we have an edge,

we're sort of maximizing the number of

places we're trying to exploit that edge

across time and and across the the asset

space. And um and so we don't we don't

optimize parameters. I think that's the

kind of biggest thing that we don't do.

You know, I you can call we do technical

analysis, I guess, because it's it's

price-based past performance. Um but we

don't optimize parameters. We don't have

moving averages. We don't have any of

the other indicators that people talk

about or you see you know people

discussing from a from a technical point

of view. Um we have a very kind of

structured just way of trying to

identify whether there's a train a trend

operating on a on a particular time

frame. Um and that comes from several

different

ways of looking at that. So you know we

look at price which is what a lot of

people look but also look at pure time

measures. So we will look at you know we

will compare if you think about kind of

market movements back and forth you know

we compare the time that it takes a

market to you know if let's say let's

say the um you know equities have been

up for 6 months on some time frame and

then they go down for a couple days.

Even if that price movement down was

bigger than the movement up, we would

still say that the time element is still

bullish, right? Because you were

spending a lot more time going up than

you were going down. Um, so we have a

couple different time type of uh signals

we use. We have price and then we have I

guess loosely kind of identify as more

patterny based things. Um, and all

relatively recent. So everything we do

is look at the very recent past. We

don't use you know for for any given

time frame. we don't go back very far um

in terms of our our analysis. So um so

yeah so we we we apply those same

indicators to said to every every time

frame every market and uh you know I

think that on average there's an edge

there and and uh that edge is sort of

exploitable the more you diversify

across the the portfolio.

Bill, I'm really interested in your

background because you started off doing

mostly discretionary. Correct me if I'm

wrong.

>> So, what was that transition from

discretionary to systematic to both now?

>> Yeah. Well, it's you've missed the

beginning of the cycle. So, you know, I

did I did a PhD in finance and

>> part of my dissertation was looking at

trading strategies and and um I worked

for a bit at uh Barclay's Global

Investors, which was Barclay's old sort

of asset management, equity asset

management, and and there it was, you

know, standard equity type long short

signals, very quantitative, you know, um

no discretion whatsoever. Um and and so

I kind of been through a few iterations.

Um and yeah, you're right. When I was

when I started my commodities career, it

was really focused on fundamental

science. It was kind of bottom up

fundamental modeling. Commodities being

quite different from equities and that

you know, you really can use

microeconomics. You know, if you figure

out if you know what the supply is and

you know what the demand is, you can get

a get a pretty good idea of price should

be. Particularly for things like my

background was tend to be more power and

gas and power where you had no storage.

um you really was a microeconomic you

demand and supply curve bang that's the

price right so um so that seemed you

know back back then when you this was 25

years ago where there wasn't as much

information around you really could

model demand and supply um differently

maybe than other people do and get an

edge through how you were modeling or an

edge the data you could get access to

because there wasn't um launch stuff

wasn't available on the web or if it was

available on the web was kind kind of

buried in a government website somewhere

and so you would find the data and

somebody else wouldn't. So you could put

you know components for demand supply

together easier than than other people

and get an edge that way. And so but

what you know over that over my career

over those 25 years I saw the sharp

ratio for kind that fundamental bottomup

trading steadily decline and that was

just because the the competition for

information availability for information

the people you know would leak. you go

from one firm to the next and that firm

would learn what the other firm was

doing and and so and that's kind of the

state of play now where I think most

people have access to almost all the

same information, same modeling

techniques and so fundamentals is is a

lot tougher than it that it used to be.

Um and while that was happening I was I

had developed some of the tools that we

still use today back way back when I got

my PhD and and we was using them more

for timing alongside of the fundamentals

that that we were doing. And eventually

got to the point where I just thought,

well, I don't really see what the

fundamentals are doing here. It seems

like this is all timing. Um, and so

probably by about I guess this was maybe

10 years ago, I started to really

believe that the fundamental approach

because of its declining value was not

really adding any value over what you

get simply by, you know, trend following

on on most markets. Um, so that was the

the transition and yeah, I would say,

you know, I I had this discretionary

approach and I kept refining what I'm

doing now over over many years to the

point where I felt like the system was

better than my discretion.

And so on a you know it was in in terms

of identifying trades and identifying

places that that you know trends were

were performing um you know the system

was was quite good and that's that's

what we built the the firm on you know

starting in 2020 really was that that

concept and while that's true I guess I

guess the the second step of that

transition was

you know I felt like okay if you look at

a given market you know the idea was

capture the trend when they exist and

try not to lose very much when there

isn't a trend. That's what we built the

system to do, right? Is that you know

most trend if you if you have a trend

following system better catch the

trends. If you're not catching the

trends that then that's a terrible trend

falling system, right? But but mostly

what happens is with most trend falling

systems you burn a lot of cash. It's

almost like being long options, right?

That's that's how trend falling behaves

like being long options, right? you have

a couple big winners, lots of losers,

and in the option space, you know, the

bleed when you're long volatility, you

know, it tends to be quite high uh for

the few times you you get paid off. I'd

say trend following is the same has the

same shape, distribution shape. You have

positive skew and your returns just like

you do in a long volatility strategy. So

very there's a lot of analogies between

the two and and so really what we try to

do is really manage those those times

where the market's not really moving and

you know still try to capture the trends

and that that's great. That's what the

system does. But what what you realize

is you can actually even avoid those

times completely. So instead of not

losing money at all, it seems like we're

able to figure out actually the odds

that a big move originates from this

market at this time is really really

low. And so why waste the money getting

chopped around even though that we think

we're not going to we're not going to

lose as much. So so it became more of a

portfolio decision. So it's like we

built we built the system at a market

level and even at a time frame level and

thought that well we don't have any

reason to bias in terms of what markets

we trades. We apply to all markets that

are liquid. We don't we don't bias on

time frame because we don't know how to

choose a time frame. So we tried to have

this really agnostic approach. Um but

what we've seen now with the portfolio

is actually you can at least we feel

like we can be a bit smarter about that

and and um uh you know decide where we

want to apply the model and where we

want to just sit put it on the sideline

for a bit.

>> How big is your team at 10 dynamics?

Just curious.

>> Uh it's just four. voice.

>> What's it like [clears throat] making

those decisions like having to make

those fundamental discretionary

decisions um and having to build out um

a like continuously improve the

systematic approach at the same time?

How do you maintain focus? That makes

sense.

>> Yeah. you know what what we've actually

found I think the um

uh

really so you know one one of the things

that I used to tell people that you know

I didn't people ask well how's the you

know how's the trading going or whatever

and I would always say well I don't

really feel like a trader anymore I'm

I'm more like a mechanic

>> I just make sure the system runs I don't

really care if the system's long or

short or buying or selling I just trust

the system's going to go and I'm the

mechanic and make sure that it runs and

that's true of the whole team Right? We

were just we were mechanics. But the

advantage now of using kind of more

discretion is it's make it's gen it's an

idea generator, right? You start to get

more real time feedback with okay we

thought this at that time why didn't we

you know why did we do this or didn't do

we do this and then you can start say

well is that a signal are we missing a

signal there we can then we can go back

and test it and so so actually I think

the combination is really good for idea

generation. I think our our ideas were

not getting stale but if I looked at our

research queue um even you know well

definitely at the beginning of this year

it was pretty small because we weren't

doing the discretionary stuff at that

point and we kind of exhausted what we

thought we could do with the the system

a bit and we relatively happy with it

and then all of a sudden we started

trading discretionary thought oh well

what about this what about that you know

you started to and you you catch more

when you get closer to the market I

guess you you start to get more ideas

and things like we started, you know,

one of the ideas we've got now is well

maybe this discretionary stuff should be

informed by news or you know maybe we

should be doing one of these things

where you're constantly monitoring

social media or whatever it is you know

like ideas for how to embed other

signals potentially using AI to do that.

It's a lot easier than it used to be for

a small team like us you know it's

easier for us to build than you know it

was a few years ago. So um so yeah I

think it's been it's been quite good

actually. Um, and there's been a few

places where now because we're really

watching on a micro basis because, you

know, before the system's making lots of

trades, right? So, we're trading,

you know, across whatever 60 or

something commodities, you're probably

making 80 trades a day or something like

that. So, you're not looking at every

trade and saying, well, do this make

sense or that. And now, so now, because

we are seeing that a bit more, it's

like, oh, why the system do that? Well,

that's kind of a maybe we should think

about that as a as a filter to say,

well, don't take those trades that are

in that kind of situation because, you

know, that was clearly a an oddball type

type trade. So, um yeah, so that's and

and what we've seen um in our in our

testing is that we definitely are

affecting the statistics in terms of our

um win loss ratio. So we are choosing

more winners and excluding more losers

on average at least over the short

horizon that we've we've been doing this

than in than the system which is which

is good.

>> What you mentioned there about the idea

generation from being more active in the

markets um from a discretionary

perspective I find that fascinating. Um

my guess it does make sense uh that if

you are continuously monitoring the

market you will you'll get ideas and

I guess I'm curious now what cuz what

sorts of edges have come about as a

result of that and you mentioned the

news thing. Um, are there any other

stuff that you're thinking about

building into the pipeline maybe from a

systematic perspective or maybe just

improving uh your discretionary overlay

and adding more dimensions to that? I'd

love to hear it.

Yeah, I mean uh what first one is

definitely the the news or or whatever

you call like some sort of sentiment

trending thing coming from real time

coming from what's happening because you

know what's what what we've seen this

kind of goes to a second idea too but

but we've what we see in the model quite

clearly is that the I'll call it alpha

right we'll call it what just I mean you

know talk about technically what it

actually means but but the

outperformance of the of the system the

alpha um that it is it has a time series

structure to it and this is true across

um sectors. So if you look at say energy

you have you know you'll go through a

cyclical period where you'll have months

of very high alpha and then you'll have

months where the alpha declines and you

might have a couple months of even

negative alpha and then it'll go back up

right and you see this across all the

different sectors and you see it across

the whole market as a whole. So whatever

is driving the returns to time series to

to trend following it has a structure to

it a time series structure to it. Um and

so we've been thinking about what drives

that time series structure and if if we

can figure that out you know maybe

that'll improve them all right and and

so one thing is just do pure time series

forecasting. That's one of our ideas,

right? So look at can we can we do time

series forecasting on the alphas on

given sectors. Personally I that's a

that's was one of our first ideas but I

think it's hard. I I really I don't like

you know say it sounds follow funny

being trend falling say you don't like

time series forecasting but like the the

traditional time series forecasting I

really don't like because you know

usually you're doing something in a

monthly type granularity at least what

we'd be doing it's sort of slowmoving so

you don't have a lot of data you know if

you have monthly and then you have even

20 years of monthly data that's not a

lot of data to be testing your

parameters on stuff so so I I am less

confident that we'll get much out of it.

You know, maybe we will. I don't know.

We'll have a see see if that whether

that works or not. But the other is

that, you know, it either points to well

what what drives this this sort of

structure. And one of the things is that

we can see is there's a trade-off

between volatility and call it trend. So

you hear, you know, if you think of a of

a a grid really where on one one axis

you've got high, low and medium

volatility and the other axis you kind

of have like downtrend range uptrend.

Um, you know, everybody says, well,

hedge funds in general, but systematic

too, that that trend followers make

money in high volatility, right? That's

the I think you hear people say that all

the time. That's actually not exactly

right. What you want is the ideal for

trend following is low volatility but a

strong trend. So, if you have something

that's just moving up consistently and

doesn't really have a lot of chop in it,

that's your best market, right? So, so

it's actually low volatility

environments but that have a trend which

perform the best. High volatility can

work but it's very asymmetric. So high

volatility tends to work good at bull

markets but high volatility of bare

markets is really tough because um short

covering rallies tend to be really fast

relative to the downtrend that precedes

them. So you tend to lose you get back a

lot from what you made. So, so a high

volume environment in a downtrend is not

really ideal at least for our system

might work for for other people. Um so

it's trying to figure out how can if we

break if we create that grid you know

can we identify what quadrant we are in

the grid you know are we in a high

volume trending environment or a lowvall

trending uptrend or downtrend and you

know what are the drivers that well I

think potentially fundamentals are so

you know I mentioned before so what's to

me an overs supplied market that has a

low price relative to the history that

that's a bad market for trend falling

because probably there's no trend And

probably the volatility is high because

everybody gets caught out. You know,

when when a market's overs supplied, you

have a tilt towards the short side and

then a little something changes and all

of a sudden the market spikes. You see

this in like US net gas is it happens

quite a bit. So, um so that's not an

ideal environment to to be involved in.

And you know, do we think there are ways

to identify which one of those

environments we're in? It could be

fundamentals in in the equity side. It's

definitely like some sort of psychology

chatter, you know, like everybody's

talking about Tesla might not might be

going up, might be going down, but it's

definitely trending, right? And it might

have high volatility too, but it's

definitely moving where, you know, most

of the particularly like when you get

out of the large c when you get out of

the technology

space, you know, there's a couple other

areas where it's pretty good for for

kind of the chatter um that that seems

to drive I guess I guess it's attention.

Um then equities become really tough to

trade from because I think then it's

become it's just about you know what's

happening with sales and you know real

fundamental stuff that is great for the

long short equity guys but I think it's

pretty tough for for trend following to

generate any alpha over and above

whatever the the market's doing. So, so

there's a whole I mean we went from

having a very short list of of research

ideas to now a kind of massive list of

research ideas and then it becomes

picking which ones you think are more or

less likely. Um and and AI making things

a lot more easy in terms of speed to do

research. uh which we've seen even in

the last couple months I would say it

that what AI's contribution to our

research is much higher than it was um

you know even 6 months ago which is

pretty good

>> you talk a little bit about that um how

is how of these genai models enhanced

your research process and pipeline

>> yeah so look you know if we take the the

call it the alpha time series analysis

right So

yeah, I would say maybe people will

argue with me, but my my feeling was

with AI, you know, which are pick your

model or whatever that earlier this year

it was still kind of like a a Wikipedia

summarizing tool like yeah, you know,

it's good at communicating, but it

really in terms of facts and ideas, it's

kind of just summarizing, you know, does

a good good job, very good job of

summarizing Wikipedia. But this last

version that came out, you know, I was

working with it to kind of to hammer out

the the time series alpha forecasting

and it generated a whole research plan

for me that I would have said came from

a master's level of financial

engineering student. It was really

really um welldone sort of mathematical

equations. All the math was correct. I

didn't see any errors with it.

um quite interesting some some little

applications of stuff that I hadn't

thought of doing and and um um you know

ways of implementing the forecasting

that might you know maybe it maybe it

has some advantage to it. So it suddenly

was like instead of just having a a

summarizing tool, it was actually like

having a kind of M's level um colleague,

you know, who who managed to generate a

pretty complete plan in about 15

minutes.

>> So that was um that was the first time I

was really like, whoa, that was that was

good. That was really good. Um, so yeah,

so I don't I don't know where it goes

from here, how much further it can it

can continue to progress, but it's

really helpful now. So now I feel like,

you know, to build this, you know,

social media news monitoring thing for

all the assets that we trade, I think

it's going to be all done by. I don't

think, you know, we I think, you know,

my partners and I will probably do the

project management work, but I don't

think we're going to do any of the

coding. And and I think you know we'll

have most of it built built by AI which

is um you know very different than I

would have said 6 months or a year ago.

>> Oh yeah. I I think this the ideation and

to execution in terms of speed and

building with these models it's

ridiculous. You know even myself if I

have an idea I can easily and that's not

even for something I want like markets

related just yeah you know project on

the side. So I'm curious like how how I

could do that. I can go into I mean I

could go into even just chatbt but you

know you stack up cursor and these other

like tools and it's it's ridiculous. And

and the other funny thing you mentioned

is um the fact that an AI model like a

genai model can do the work of a

master's level financial engineering

student. And I just find that funny

because I'm currently a master's in

financial engineering. Sarah, uh oh

guys, you know, this is this is you. I'm

looking at my friend though. Maybe this

is this is just not the best position to

be in. Um

>> yeah,

>> but yeah.

>> Yeah, it's I think it's it's going to be

a challenge. But even then, you know,

like when I say it's mast's level, I

think it's like u it's just getting

really good at the call it almost like

boilerplate. Like you you know, when you

look at I always think of from like law,

right? You ask somebody to write a

contract and they charge you, you know,

your lawyer charges you a lot of money

to write a 20page contract. Of that 20

pages, 18 of it is a boiler plate. 18 of

it is in every other contract. It's all

the same. And there's a little bit of

math that's like that, right? There's a

little bit of engineering that's like

that. There's just the sort of basic

stuff, you know, that that um to have it

do that really, really quickly is great.

I still think a human then can take that

and say, "Yeah, what about this? What

about that? What about, you know,

tweaking it this way or that way?" And

um and I'll I'll give you a story. I

I've I would love I don't I mean, you

know, my background was finance in terms

of my PhD. So, I know a bit of math, but

I'm definitely not a mathematician. I'm

definitely not a programmer. Actually,

I'm getting embarrassed like when I give

when I give chat GBT some code, you

know, how much better it writes the code

than I do. It's like a bit bit

embarrassing but um when I was getting

my PhD I had a math student or a math

professor and this is you know when I I

was getting my PhD in the 90s and and in

the '9s neural nets were a big deal.

Everybody thinks like AI just came

around yesterday, but actually neural

nets were a massive deal even in finance

back in the 90s and people were trying

to use them for prediction, stock prices

and stuff. And uh the math professor

said, you know, AI will never replace a

human mind because of imaginary numbers.

And I said, like, what do you mean? He

said, well, how would an AI ever invent

an imaginary number system? like it took

for for a human to say hey here's the

you know we we're having to deal with

this square root of negative1 thing well

let's just represent it by I and if we

represent it it opens up an entire

branch of mathematics that actually

completely changed mathematics and it

it's like it's like the analogy of

giving a a computer that plays chess you

know is the best chess playing computer

for the computer to say one day hey you

know be really cool it'd be really cool

if turn the board over and play it on

both sides of the board at the same

time. Like how would a chess playing AI

ever come up with that idea, right? So,

you know, his his argument was how would

something that's trained on data ever

invent the imaginary number system? And

and I think there's something to that

and I would I would love somebody who's

really a an expert in the field to tell

me how that would happen because you

know people and again AI people will be

critical of me but my my view of all

these machine learning models is they're

just really fancy regressions. They're

just really really fancy regressions and

and we all know the limits of

regressions when you try to forecast

outside the data set, right? You how can

you forecast outside the data set? So to

me like imaginary numbers were outside

the data set somehow. Now maybe a

mathematician would tell me differently

but he was he was a good mathematician

when he thought it was that that was the

case. So um yeah so that's just an aside

of where you know where we go and where

the where the human element can still

come in and it's still the creativity

side um which I think you know as a as a

master's level engineering student it's

you know focusing on where the

creativity is um rather than the the

boil call it the boiler plate you know

>> I mean the ingenuity to come up with

imaginary numbers and all these

different tools within mathematics like

definitely It's I I don't think we're

definitely not there at the moment.

Maybe we'll never get there. I hope we

never get there. Actually, I don't know.

Um but uh when I think about because I

did my degree in in in math and physics

um for for my undergrad and I mean yeah

there's so many tools

um

coming up with that um physics like I

think I mean for me it's ridiculous even

that Albert Einstein didn't learn

quantum mechanics when he went to school

for physics because now it's just a part

of a a course Right. But I remember I

was, you know, before the like one of my

exams, I was talking to a friend and he

said, "Dude, don't you think it's crazy

that like Einstein never learned this?"

And you think about it, it's like,

"Yeah, like that. That's wild." Or that

Newton never learned the relativity

stuff, right?

And

I guess the beautiful thing about human

ingenuity is it just keeps building on

top of um like the the last thing that

was super novel at the time and then is

this new paradigm like whoa. And I mean

we don't know what it's going to be

next. Um but yeah, I mean I'm I'm

probably with you. I don't think that I

think that that the the ingenuity

something truly like like a new a truly

new novel idea. I think it's hard to get

there and I think it's easy to get to

some fake novelty um and um where you

know oh this is novel we've applied this

existing

solution to a nicheer problem um that

hasn't been solved before but oh we

found a way to map it and I think that

like that it can do I I can see it being

able to do that in the future. Yeah,

maybe like the majority I think the

majority of PhD the whole work tends to

be um applying

um you know uh you know solving these

these niche problems um and and and may

maybe it'll be able to do that. Uh but

fundamentally that something that's

truly novel I think that that comes from

human ingenuity. I think that's what

your math professor was getting at and I

think this leads us into

your broad view for how these

technologies are going to be utilized

within the markets because I think

the returns and performance of traders

is always going to it's always going to

be a bell curve always and you know you

add you give everyone access to these

generative AI tools maybe everyone can

model the use a bit better and so the

bell curve maybe

shifts but like or like some top

performers are able to use things better

capture some nonlinearities

uh but as soon as everyone starts using

it yeah it's it's back to a bell curve

fundamentally and I think edge comes

from from anticipating what other market

participants are doing at least that's

one component of edge right and so how

would you think about the how do you

think about these technologies being

applied in the future and where do you

see where do you see edge going in the

next couple years with with respect to

these new tech

>> yeah I mean I not I'm not 100% sure you

know there's um there's a part of me

that wants to believe that certainly

machine learning is going to struggle in

the markets and and and part of that is

at the end of the day the data set

people don't think it the data set is

pretty limited it's not People think,

oh, if there's tons of financial, not

really. There's a lot of really

correlated stuff out there and there's a

lot of high frequency stuff, but high

frequency stuff is only good for high

frequency. High frequency is not going

to help you for longerdated stuff. So,

if you talk if you think about like,

okay, let's let's talk about energy in

particular. So, energy has seasonality

in it. So, behavior in January is

completely different from the behavior

in March. What does that mean? It means

you literally have one data point a

year. you have January 1st and then

January 1st the next year and then

January 1st the next year and maybe the

kind of days around that but those days

don't have anything to do with how

energy behaves on March 1st and so

really your data set becomes if you're

going to use daily data it's not just

you know 250 days a year it's actually a

couple days a year each year so suddenly

you're you know if you want to go back

whatever 15 years you've got 30 data

points or 45 data points you're tra

can't train in AI on that like it's not

going to happen so so there is a there

is huge data limitation and the other

thing is you know like LLMs are amazing

but language is stable you know a dog is

a dog today and it's a dog tomorrow and

it might be a different definition a

thousand years but it's pretty much

likely to maintain its still general you

know meaning but the the relationships

in the markets over my career they've

they're constantly changing what people

look at what people care about changes

every year it seems like and and so how

is an AI going to pick up you know what

where's the underlying sta stable

structure you that the AI is actually

training on. I don't I don't think it

is. I think the the AI is just training

on what happened in the you know past

however long your training set is. And

and if there is a stable structure deep

down in there somewhere that the AI is

going to find out the minute one AI

finds it all the AIS are going to find

it'll that information will spread

spread like wildfire and I still believe

in market efficiency at the end of the

day. So if there's an underlying

structure and AI discovers it, that's

the end of trading in a way because how

you know everybody's going to know what

it is. But then you go back to, you

know, the whole theory of finance and

and economics, right? It's trading is

price discovery. So if the AI knows the

structure and can figure out exactly how

to, you know, translate all information

at price and then everybody trades out

the same price and then trading stops

and there's no price discovery and then

what happens, right? So I don't I don't

I don't think that's possible. So

somehow I think AI is going to struggle

to to I think you know financial like

pure price forecasting be it from

fundamentals or sentiment or whatever

you're going to do you know or even mean

reversion or whatever you know that it's

either going to get armed away or

there's something there that the machine

can't pick up one of the two and and I

think there's still at least you know

right now do we see the giants are they

like do they have one AI model that's

trading all the markets and gotten rid

of all their traders No, it's not

happening. And investors don't really

even like it. I think it's hard enough

with a systematic approach to talk to

convince an investor. Again, they don't

believe the track record. You can't give

away in detail everything you're doing.

You don't want to give away the the

whatever your you you you think your

secret sauce might be. So, you know,

they're they're basically trusting a

machine with their money. It's like, you

know, I've used and I've talked about

it, you know, I think on a podcast or

two about you, nobody wants to fly on a

plane that doesn't have any pilots, but

we could make we could make pilotless

planes today that'd be safer than planes

with pilots because pilots have been

the, you know, the behind the majority

of crashes over the last however many

years. So, so there is a little bit of

that in finance, too. You don't find

people have a little bit of discomfort

with saying particularly what we would

say is, well, we don't really ever

intervene. You just let the machine do

what it's supposed to do. that that

makes people a little bit a little bit

nervous. So yes, I think that you know

that role of the human somehow is is

embedded into into finance from both

from a kind of scientific point of view

maybe even from a investor point of

view. It's hard to hard again is even

worse. So I don't know how you sell an

AI system. Well, it worked. It worked in

the training set. It worked today.

>> We don't know what's going to happen

tomorrow. So yeah, I don't know. It's a

tough cell.

>> Yeah. I I mean Ken Grein I think that he

recently said it genai is not good for

alpha discovery like that's I think yeah

he recently made that comment and

yeah I mean this first thing that comes

to light it is

and I think this is similar to my

experience of it um I've found that when

I use these tools for

certain tasks. And I don't know if you'd

agree, um, but I I found that it can

make me intellectually basic where I

just I I just, you know, it's

>> it's

what

for me it would be similar to comparing

just scrolling on in Instagram versus um

like really going to the source of

things, right? It's just it it for me it

props intellectual laziness. I've run a

regression a million times. I run your

regression dose some like um some do

some um some add some panel L L2 or you

know just and and I I I kind of don't

really think that deeply about it.

Whereas when I just go you know if I

scrap the AI model and maybe I can make

it build a plan. Um but even then I

think the act of building plan can be

very good mentally speaking. Um, and I

mean I'm sure you're on Exobot and uh

there's there's this new term of being

oneshotted by AI uh where you know and

and and I see it like I can see people

not even just from a

from it from from using these tools to

solve their problems but just by over

reliance on them for their inner

monologue where they're there and they

the moment they have a thought that they

have a mild problem or a mild anxiety

they instantly go to the genai model and

that's you know and that and it it calms

their anxiety temporarily but the

exercise of doing that I think is just

the worst because you you build this

habit of relying on um not even just a

phone but but an AI model for mental

clarity and and for being able to to to

to not have anxiety with certain things.

I think that's that's insane. And I

guess it it leads me into a question

about how you would deal with using

these tools for young people um who are

honestly in my opinion over myself

included. You know, how would you use

how would you try to maintain that

intellectual rigor thinking deeply about

things? uh in this age where we're fed

not just bite-sized information on

social media, but just even our thinking

can be bite-sized where we're literally

prompting an AI to do our work. And how

do you think about that?

>> Yeah. I mean, this is the this is this

is one of my biggest concerns. I mean,

if you combine that with remote working

for young people, I I I mean, it's it,

you know, it really makes me, you know,

it's it's the one area I'm not concerned

about the big, you know, AI taking over,

murdering everybody, right? I think

that's that's it. It's it's way down the

field, you know, it's it's not it's not

the pressing concern. The pressing

concern is is exactly what you just you

just spoke about in the combination with

you young people don't get that

experience with older people in the

organization to teach them you know you

kind of do it like this or which which

is okay but it's not sitting you know I

used to say like you know why don't big

organizations it's tough for them to

just become giant monopolies because you

can see at work that you were more in

tune to what the person next to you was

doing than you were you know we used to

say open floor plan right the person on

the other side of the monitors who you

kind of saw once in a while, you kind of

knew what they were doing and you hear

their conversations but you didn't

really know compared to the person who

was next to you and the person who was

three rows over you didn't really know

what they were doing and if somebody was

in another office you had no idea what

they were doing right so so that that

loss of connection which what we have

now I think is is tough and and you know

I what you talked about I kind of went

through in a microcosm because again my

background is not coding it's not and I

as tool. You know, I my PhD well way

back when I did my undergrad, I was

using forran, but eventually I was kind

of a mat lab guy. I love mat lab. And

then when I as I was getting ready to

start this company, I thought, well,

that's not a modern tool. Let's start

doing Python. I knew nothing about

Python. So, if it had not been for Stack

Overflow, I would not have been able to

build this business with my colleague

because we didn't know Python really

didn't know coding to the level, you

know, it didn't know much about

object-oriented coding. It was all like

very scripty from kind of mat lab style

whatever and and stack overflow you know

that the idea that you could just ask a

bunch of experts a question you get 10

different answers you'd have to dig into

those answers and then you'd have to

code yourself debug it figure out why it

wasn't doing what you wanted it to do

and you learned and so you know I've

gotten to be a reasonably proficient

python coder only because of stack

overflow if it had been you know I think

about reading mat lab I used to have

five mat lab reference books like this

big each one of them and every time you

had a question about how to deal with

arrays you have to go in the book and

you look it up and but you learn mat lab

to a really deep level that way and you

would because you were always you

couldn't zero in on exactly the solution

you'd have to learn like 10 peripheral

things before you would actually find

the solution with Stack Overflow that

that 10 peripheral things got reduced to

four or five potential ideas and now

with AI you just get the one answer. And

then what I do, what you might do, I

don't even actually really read the

answer. I I test it. I test it. See if

the output gives me what I want. You

know, do a little bit of of um you know,

consistency testing and just bomb, you

know, use it. You know, I don't even

look at, hey, there's a really clever

way that it did something in Python that

I hadn't really seen before. You know, I

don't I don't do that. So, I don't I

don't learn that. So, you know, how do

people coming up, you know, they're

they're getting the they're getting the

the the exact answer every time or very

close, you know, how do you learn all

the stuff that goes around that and

really learn something? I, you know, I

don't know. I don't know. And nobody

maybe there's somebody out there, but I

can't imagine there's anybody that has

the discipline to say, "Actually, I'm

not going to use AI for this. I'm going

to go out. I'm going to write three

different versions of this this function

and if it doesn't work, you know, it'll

take two weeks to do it. And nobody will

do that. Literally, no one will do that.

So, so I don't I don't know. I don't

know. I feel like if I if I started

today, you know, just like I talked

about with this, you know, I'm already

thinking about with this, if we build

this um, you know, little sub

application to monitor all the stuff on

social media for what's going on in

different markets or whatever, I

probably won't really dig into what the

code's actually doing deep down. you

know, if it if if if our AI says, you

know, I shouldn't do this in Python. You

should do it in some, you know, some

language I'm not familiar with, there's

no chance I'm going to go learn the

language to figure out what it's

actually doing, you know. So already I'm

thinking I'm I wouldn't if I started

today, I wouldn't know what I know about

coding, you know, and I wouldn't learn

it. I just wouldn't. I'd be using this

thing to to to do my ideas and then I

don't even know if I would have the

ideas, you know. I don't know. It's

>> Yeah,

>> it's a really I wish I had a great

answer. I don't know that anybody has

any answers.

>> Yeah.

>> The one thing I do think though is, you

know, if you're, you know, if you're

young, kind of, you know, get a job

where you actually have to be around

people. I know remote working sounds

like a great idea. I just don't think

it's going to help you in the long run.

That's the one place where you can kind

of force yourself to bite the bullet a

bit and go into an office and

>> yeah,

>> spend some time with people. um you know

there's companies are still requiring it

but um you know I know a lot of people

who are coming out of school are like

actually looking specifically for remote

working opportunities some

it's it's tough so yeah I don't know I

guess yeah everybody should you know

seem like I can remember um this shows

how old I am that that you know the

people who are really engineer so I did

I did chemical engineering as my

undergrad and there were a few guys who

you know or girls who would commit to I

don't want to use a calculator. I'm

going to use a slide rule because when

you when you have a slide rule, you

really understand what's going on,

right? And it was a bit of a joke, you

know, with some of the really smart

engineering students, they do everything

on a slide rule just to prove they could

kind of thing. So, you know, I think I

think some of this stuff is becoming the

slide rule of today. You know, who would

ever use that tool anymore? It's even,

you know, even calculators. So, use I've

got this I've got I've got this bad boy

sitting on my desk. I bet there's very

few people that have a a proper HP

calculator and use it. I still use it,

but uh we should just be using something

else.

>> Yeah. I mean, path of least resistance,

everyone is trying to go path of least

resistance. And so I think I mean the

only solution I can think of and this is

something I'm trying to do myself. Um,

but it's putting myself, so I'm not

banning myself from using AI, uh, from

like because I think the ship sail, you

know, you might as well, but I think

what we can do is embrace intellectual

exercises that force you to go deeper.

Um, and so like one of the things I'm

doing, and this isn't, and there's

definitely not a onetoone correspondence

between this and doing a a quant job

better or being a better trader, but u

like I'm trying I'm I trying to make a

conscious effort to read more. And that

could be literally anything. But I think

this thing over here and doom scrolling

has is has fried my

attention span. Like it truly has. Uh,

and I don't I can still focus if I have

a exam or a project and a deadline.

Yeah, I can I can get it done. But

even then, like I'll do my 90 minutes of

focus with my AirPods in and my white

noise and fully fully fully focused only

focusing at the task at hand. And then

I'll leave the room and I'll doom scroll

for a couple minutes and I'll walk

around and then I'll go back and focus.

But even that is not ideal. Like

ideally, you should be able to sit

there, focus, leave the room, maybe take

it like a five, 10 minute break and just

let their mind be still and just and I

think that intellectual exercise I even

found that when I force myself to do

that the ideas just flow, right? even

about the t like I I was last working on

let's say very some sarcastic calculus

course right for example and and and

that 5 to 10 minute break afterwards I

feel like the ideas just get to simmer

and I get to really internalize those

and and and come up with insights about

that or about the the principles within

that that I can apply to other domains

and it's just great for your thinking

and when you pick this up and start to

scroll a bit you kill that even if you

focus during the session. And I mean, I

guess Bill, last question. If there's

one piece of advice, you're young today,

um, and you want to be in a, let's say,

let's say there's someone listening to

this and he wants to one day run a

systematic hedge fund the way you are.

What what advice would you give?

Oh well so

you know I've over my career I've seen a

lot of attempts the systematic

systematic trading you know to it's uh

it attracts quantitatively minded people

um you know so I've seen lots of very

very smart you know PhD level math and

physics people come into the field and

by and large they struggle because they

never really had any real market

experience. So, you know, talking about

that that trend, you know, like we're

getting again where it's like we're

getting reback into or back into having

actual trading experience with the model

and seeing how the two things work

together, you know, recognizing when the

model is doing something stupid or not

and and that kind of stuff is um I think

that real world that real experience is

helpful. Now, you can get that a lot of

different ways. you know, people can

talk. Yeah, you can go on whatever's

Robin Hood and punch around your own

stuff and and try to try to get that

experience. And I suppose that's that's

a little bit helpful, too. But um yeah,

it's it's um

you know, for I think of all of the

stuff I did on my PhD and all that I use

is the the way of thinking that you get

from a PhD program and from a master's

program and from engineering, you know,

I think I think engineering, you know, I

think it's just such a great degree.

Doesn't matter what you do. I think any

engineering the the process that you go

through to learn how to be an engineer

is so good um in terms of thought

process and breaking a problem down and

being methodical and how you solve it

and all that kind of stuff. I think it's

a brilliant skill set. Um but I don't

really think I use you know I don't

really use anything that I learned as

you know any actual techniques that I

learned as a PhD student you know in

finance or whatever. I um I took kind of

my own ideas and maybe I'm I I know

again it's like you know how not to fool

yourself a little bit. So knowing things

like knowing things like you know how

easy it is to

allow some forward-looking data sneaking

into your forecast and next thing you

know you think you're a genius and

actually you're cheating, right? It's

really easy to do that. It's really easy

to I mean finance is all about avoiding

pitfalls and and you know and how like

you know the just the whole thing of if

you're going to back test stuff that if

you give yourself full flexibility

you're 10,000% going to find a couple

amazing systems you know and they might

even work for a little bit but they're

not going to work long term and how do

you how do you separate that out? And I

think I think if you just rely on back

testing I don't know how you separate

that out. I think you have to have had

some experience or some philosophy like

we we built you know I I what what we do

is built on sort of our philosophy on

how prices move and you know not a

random walk you know we know kind of

have this kind of fractal concept in our

heads that that guides a little bit how

we think the market should unfold and

and things like that that you you got to

have you got to start somewhere with a

principle or or a belief you know and

and Um

um yeah my you know it's not this is not

an easy finance is not easy to break

into first of all it's it's in a lot of

a lot of

um there's a lot of finance that is kind

of turn the crank type stuff you know

that's you know whether it's you're

working at banks you know you have they

have products that they sell the product

has an edge built into it and there's no

you know it's not like you're not like

you're out forecasting somebody and

you're making money off the back of

that, right? It's there's there's a lot

of finance good finance jobs that are

really about just using your skills to

solve people's problems, right? which is

a different thing to a company's

problems, different thing than um than

actually trying to to beat all these

other super competitive places because

the question is, you know, either you

want to make yourself marketable to um

to Citadel or you know, somebody like

that, you know, which is which means you

better be top top of your class, right?

So, you know, that's the reality of it.

you know, you're gonna go that route or

well, you're gonna go on your own and

try to compete with S. Oh, that's pretty

tough. So, you better you better have

some you better you better have gotten

some some really good experience or have

some really good idea, you know, and

that's again, as you said, that's that's

a real challenge is what do you have

that's unique, you know, that you think

is you believe in or have experience

doing. And um yeah, it's not it's

definitely not an easy road. And I I got

super lucky. So I I I w I degreed myself

up, right? So I I learned I had my I did

my bachelor's degree. I had a master's

degree in finance and then I I realized

that because I went to you I went to a

good school. I went to the University of

Colorado, but that's not like a high

pedigree school. So I knew that to get

where I wanted to go, I had to go back

and get a degree. And so I went to

Cornell. I got a PhD. So that that

>> credentialized my background that got me

the job that led to where I was, right?

And then got me the experience. And had

I not done that, I couldn't have I

couldn't have made it. I wouldn't have

done it as a master student. I don't

think I could have gotten the the job

that I got got in the position I was in

to then had the opportunities that I

did, you know, had. And um so that was

one one way to way to approach it. But

um

yeah, it's it's um I guess you got to

you know, one thing. So, here's one

thing. One piece of advice I I will give

people. I think,

and this is based on managing people,

there's a there one mistake a lot of

people make is they want something

they're not good at. They want to be

something they're not good at. Like

there's a lot of people in my career who

wanted to be traders who just weren't

traders and they didn't, you know, I'm

talking about discretionary traders.

They just weren't they didn't have the

skill set. They might have been great

analysts, right? They might have been

great at fundamental analytics and and

analyzing you know whatever demand and

supply or had some specific knowledge on

a specific uh type of company right a

partic particular industry or you know

expertise in a country or something um

but they really wanted to take that

knowledge and translate it into being a

trader which is a completely different

skill set and completely different

challenge to being good at that and um I

think my my mantra I hope it's true is

that you know if if you do something

you're really good at, your life's going

to turn out well, right? Because people

make money if they do stuff they're good

at. The problem is aligning what you're

good at with what you want. And and I

think a lot of people make that mistake

and being being really um honest with

yourself about where's your competitive

advantage and stick to your competitive

advantage and do everything you can to

extend that competitive advantage and

see where that takes you in life

because, you know, you're never going to

be able to plan out. I've I never ended

I never imagined I'd end up where I did.

So So I don't think I don't think you

can plan your life out. But if you if

you focus on that, you know, what what

you have that other people don't um you

know, whatever it is, people who do what

they're really good at, they do well and

they're, you know, as long as they like

what they're good at and they're happy.

>> I love that. I think applying edge to

your personal life.

>> Yep.

>> Is great. And thanks so much for coming

on Odds Open.

>> Sure. Absolutely. Yeah, I appreciate it.

Thanks a lot, Ethan.