Bet Like a Bayesian & SP's DraftKings Beef | Ep 108 - The Risk Takers Podcast

Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
SPEAKER_00 (00:00):
It does you no good to be 100% accurate with things

(00:04):
that agree with the market.
All that matters is how accurateyou are when you disagree.
And so that's where, again,there's a premium on being right
on outlier people, just likethere's a premium being right on
things early.

SPEAKER_02 (00:35):
Hey, what's up, everybody?
GP and SP talking about ourfavorite old reverend this week.
The reverend, as we referred tohim and his theory, Thomas
Bayes.
And we're talking about,basically, this is going to be a

(00:56):
very, very light to not at allmathematical theory.
on Bayesian thinking in sportsbetting.
It sounds intimidating, but it'snot at all.
I would say it's incrediblyintuitive compared to even
frequentist statistics.
I know that's a big war.

(01:16):
This is more SP's world than myworld.
He's come from the stats wars.
I think you've started with theBayesians, but are you a closet
frequentist and what are wegoing to be learning today
that's going to differ fromeverybody's college stats class.

SPEAKER_00 (01:38):
Yeah, I'm glad you introduced it as a war between
the two camps.
I actually, in terms of like inthe actuarial world or
statistical world, I think Iwould actually be, I know
there's some, lots of actuariesare very passionate about
Bayesian thinking.

(01:59):
I think I'm generally less thanthat.
And maybe this is like a hottake, but in general, I think
either can be appropriate andthey will converge to similar
points depending on the problemyou're doing.
So we could talk about that alittle bit more.
But yeah, I think it's a veryhelpful Bayesian thinking.

(02:20):
I think it's just a helpfulconstruct to think about life
and embedding.
The idea here, if anybody's notsort of familiar, is is really,
when this is talked about, it'susually referred to as like
incorporating some type of priorbelief into like a probability

(02:41):
estimate of something goingforward.
So that's sort of like the cruxof it.
And this is really important inbetting or any sort of like
prediction type thing, becauseoftentimes you have relatively
minimal data.
And that is where Bayesianthinking will be at its

(03:04):
strongest, when data and datapoints are more limited.
I

SPEAKER_02 (03:10):
agree.
When we were briefly chattingbefore the show, I said this
reminds me of Kelly, vibes-basedKelly, as you coined it, a while
back.
And how...
If you were to use Bayes'theorem, and we can might as

(03:32):
well get to that right afterthis, but it's kind of like it
works really well in a vacuum,but conditions have to be really
nice and tidy for it to applyperfectly to the problem at
hand.
But the general concept of thetheorem is, And the ways that

(03:54):
you do things that you would sayare Bayesian and staying true to
the logic of Bayes' theorem is,like you said, very useful in
betting and in life.
And it's the same way inunderstanding what Kelly is, is
really helpful.
And it's less so applying Kellyreligiously, like the exact

(04:18):
formula to your bet sizing.
And it's more like understandingwhy what makes Kelly tick, why
it works and how to kind of usethat as a framework to think
about just like risk andposition sizing, right?

SPEAKER_00 (04:37):
Yeah, exactly.
So maybe just to give maybe likea context on like, A non-sports
betting example, and I didn'tnecessarily think of this ahead
of time, but just to give howthe thinking can work in real
life.
Let's say you had a friend stayover your house, and then a week
later, you notice he's wearing ashirt that you have the exact

(05:00):
same shirt of.
A non-Bayesian way of thinkingabout that would be sort of
like, Let's say you're trying toestimate what's the probability
this dude stole my shirt andtook my shirt and now has my
shirt.
I think the right way to thinkabout these types of problems

(05:21):
through a Bayesian lens is thisperson is my friend.
How trustworthy are they?
These things should be part ofthe decision-making calculus.
You should have some sort ofprior view of this person, and
that should sort of weigh intothe ultimate conclusion.
probability that you thoughtthis person stole your shirt or
whatever.

(05:42):
If it was a random person, youshould give them less, um,
benefit of the doubt.
It actually, you know, itreminds me of some of the stuff
going on on Twitter of, ofpeople complaining about like,
oh, you're defending yourfriends on these things.
Like in this whole pokerscandal, what there was, that
was a lot of

SPEAKER_02 (06:00):
this,

SPEAKER_00 (06:01):
you know, like you're, you're, you know, there
was a lot of the commentary of,of like your step, you know, um,
you wouldn't be saying this if,if it wasn't so or so person.
And to me, that's just like agap in Bayesian think, or, you
know, that type of thinking,like you should have a prior on
someone, you know, and if it's atotally random person that you

(06:21):
should view, you know, thelikelihood that someone, someone
like that is acting nefariouslydifferent than your friend or
someone you trust, even if theydo the exact same thing.
Right.
And so that's, that's like maybethe simplest real life example
is, you know, to incorporate oflike, What you view of the
person, the team, the event, thesituation should inform your

(06:47):
ultimate probabilistic estimateof the event.
Yeah,

SPEAKER_02 (06:53):
yeah.
I think that that's spot on.
And just to go over the coreparts of Bayesian thinking, You
have basically– what you'resaying, if we were to use that
example, your opinion of yourfriend is your prior, right?
So there's a couple key termsthat we'll probably use in this

(07:19):
episode.
So you basically have yourprior, your prior belief, and
we'll touch on priors morebecause I think they're the most
– controversial ornon-frequentist part of Bayesian
statistics because they are orcan be subjective and that's

(07:42):
fine.
But in this example, SP'sfriend, you have this opinion of
him.
You think, I've known him sincecollege, never seen him steal
anything.
You have all these all thisprior data.
So I haven't seen him stealanything.
He's been very honest on all ofthese examples.
Then you see him wearing thesame shirt that you have.

(08:06):
Now, what we didn't know is, didyou check to see if you still
had this shirt?
I don't know.
That

SPEAKER_00 (08:11):
would seem like an easier

SPEAKER_02 (08:12):
way

SPEAKER_00 (08:13):
to solve the problem.
In this example, let's pretendyou can't.
Okay, you can't

SPEAKER_02 (08:17):
check.
So seeing your friend wearingthe shirt is what in the theorem
is called the likelihood, butthat's just like the new data,
the data that you observe,right?
And so the new data is you seeyour friend wearing the exact
same shirt that you have, andmaybe it's even kind of a weird

(08:37):
shirt.
So like that, and that'simportant, you know, is it a,
just like a gap t-shirt or is itlike this, like a third
division, like Russian soccerteam shirt, you know, which that
would matter.
And then you combine those twoand you get what's called a
posterior.
which is your final belief,taking your prior.

(08:59):
And then people will say likeupdating it with the new
observed data and then have yourposterior belief, which is your
new belief.
So in this example, you'resaying like your prior on your
friend is very, very strong.
You think, you know, that's ahonest person.

(09:19):
He doesn't seem to have a reasonto have stolen a shirt, you
know, um, Whatever it is, andjust seeing your friend wearing
that shirt is not going to, whatyou'd say, adjust your prior so
much.
So maybe your prior was, I thinkthere's an only 5% chance that

(09:42):
this person or a 2% chance thisperson would ever steal from me.
And then you see him wearing theshirt.
And then you update it and maybeyou're like, okay, now I think
maybe there's like a 5% chance.
I still think it's very lowbecause I trust this guy, but
that's the new posteriorprobability in that example,

(10:02):
right?

SPEAKER_00 (10:03):
Yeah.
So maybe to get out of mycontrived shirt, that not well
thought out example at all,where you could just check your
closet or whatever.
Like in sports, I think to me,the easiest way to sort of
explain this is is a simple sortof football power or any sport,

(10:24):
but let's just use football,like a football power ranking or
rating either or sort of modelwhere you think team A is maybe
three points better thanaverage, but you don't know
exactly how much better thanaverage they are.
Let's say it's the first game ofthe season.

(10:45):
You think they're going to bebetter than average, but you're
not sure exactly.
But you think on average they'regoing to be about three points
better.
They could be four, they couldbe five, they could be two, they
could be one, but you thinkthey're better.
And they're playing like anaverage team.
So their sort of meanperformance is distributed
around zero.

(11:06):
And those two teams play weekone.
And let's say the team that issupposed to be average beats the
team that's supposed to be alittle bit better than average
by 30 points.
Well, you wouldn't justimmediately say, okay, this team
is 30 points better than thisteam.
That's the one data point wehave.

(11:26):
We need to adjust the rankings.
No, what you'd likely do isyou'd want to move the
distribution of where thesupposedly above average team a
little bit to the left.
So maybe they're nowdistributed.
Maybe you're guessing they'reone point better than average.
And the team that beat them,maybe you're guessing is two and
a half points better thanaverage.

(11:48):
Something like that.
The movement is the key and sortof the science of this.
But the idea is that you don'tsolely react to the data you
see, right?
You're reacting and moving someamount of weight.
And the other component of thatthat I just wanted to mention is

(12:09):
one of the big benefits ofBayesian thinking as opposed to
some of the frequentest way ofthinking, and we actually ran
into this problem a little bitwhen we were discussing the
modeling episode and thedistribution episode, is
Bayesian thinking is like...

(12:30):
it's always probabilistic innature.
There's much less you have to doto try and infer distributions
or probabilities.
It's just core to the methods.
And so that always gives you awhole distributional curve of it

(12:50):
rather than just pointestimates, which is generally
more how some other methodswork.

SPEAKER_02 (12:59):
Yeah, actually, I had a real-life example of this
where I was trying to judge,basically, if my handicap in
golf was low enough to enterthis one tournament.

(13:19):
It was the South CarolinaMid-Am, and they say you can
apply if your handicap's up to a10.4, but they'll only take...
a hundred and something people.
So then I was like, well, willI, if I apply, get in?
And then I had to be like, well,okay.
I think like a scratch woulddefinitely get in.

(13:41):
Like that is my like a hundredpercent.
And like a 10.4, a 10 probablywould not get in at all.
And then I have a curve, right?
Which like probably is like,starts to kind of coalesce
around like a six or a seven.
And I'm like, okay, that is Ithink the most likely

(14:03):
probability, but there's achance that it's actually a six
or a seven or an eight, youknow?
And you think about it in termsof like the curve of, instead of
like you said, a set point oflike, this is just the answer,
but you're thinking like, well,yeah, 95% of the curve is to the

(14:25):
left of 10.
A 10 handicap, I think, is onlygetting 5% of the time.
Then you start to build yourestimate that way.
That was just a really good lookinto how much of a nerd I am
when I just take up regularproblems and overthink them
drastically using some type ofprobabilistic thinking.

SPEAKER_00 (14:49):
You also probably get an aspect of People doing
what you're doing to some extentand saying– you actually might
get these spikes because there'sprobably people who don't– like
certain handicaps who don't evenapply because they think they
have like no shot and it couldactually be like a game theory
type of thing.
Because like if I'm a– I don'tknow what you said, 10.4 or

(15:11):
something with the

SPEAKER_02 (15:12):
– Yeah, I think that was the ceiling, 10.4.

SPEAKER_00 (15:14):
Yeah, so if I'm a 10.4, like there's no chance I'm
probably applying.
Right.
And similar– and you could makethat– that's sort of like
induction argument down a whiteways.
Like if I was a 10, um, Iwouldn't apply probably.
But, and so it's actually, solike saying

SPEAKER_02 (15:29):
like nine could be like, really?

SPEAKER_00 (15:31):
Yeah.
It's probably not even like cutoff.
Yeah.
Yeah.
Interesting.
The way you, you probably think,cause you have like a selection
thing.
Right.
I do want to get to, to some oflike the, um, betting stuff yeah
like how to actually use thisstuff because i think so far
we've been fairly theoreticaland just talking about it but
maybe before that i just want togive like maybe a couple more

(15:52):
really one more example thatthat hopefully people are
somewhat um familiar with thatum hopefully makes it less
intimidating all this stuffwe're talking about if it has
been uh confusing orintimidating up to this point
but um for anybody who's likefamiliar with chess and or lots
of video games like use a lot ofthe rankings and that uses like

(16:15):
some type of elo model um whichis like a way to basically you
know you have two people playeach other in something chess is
the the classic thing this isapplied to everybody has a chess
ranking um like a numericranking and you you if two chess
players play each other um youcompare the ranking and

(16:37):
basically update the rankingbased on who won in the relative
ranking.
So basically if I'm not anexpert in chess rankings, but if
like a thousand rated playerbeat like a 1200 rated player,
that thousand rated player isgoing to get a bigger bump than
if he won versus anotherthousand rated player, right?

(16:58):
He beat up a more competitiveplayer.
And so this isn't necessarily...
the direct application of Bayes.
And I think that's, but I wantto talk about this because I
think there are complicated, inmy opinion, like statistical
methods to apply Bayesian likeinference and modeling.

(17:20):
I think that's probably like notthe, what we want to be talking
about for the most part here.
I think, you know, from myperspective, what I want to then
move to is like, How can peopleactually be doing this in the
context of some of the otherepisodes we've done and more
basic start to thinking thisway?

SPEAKER_02 (17:38):
Yeah.
Yeah.
I think this will pair verynicely with both of the modeling
episodes.
But I think also the firstepisode, when you start to apply
that to low data problems, whichI think we both agree is a great

(18:00):
important caveat to kind ofsplit your techniques because
low data problems are, you know,a Bayesian way of thinking will
be a lot.
If there is an area for it, likethat's the area.

(18:20):
I think I wanted to maybe, Idon't know how you wanted to
kick this off, but there's likea very common low data problem
in golf.
that I could lead off with in

SPEAKER_00 (18:34):
terms of like how does- I think examples work
better for these types ofthings.

SPEAKER_02 (18:37):
Think about it.
Okay.
So in golf, you have likethere's different tours.
So before you get on the PGAtour- or the Euro Tour, although
the Euro Tour is now kind oflike a development tour for the
Euro Tour, you'll play on a tourcalled the Corn Fairy Tour, or

(19:00):
even below that, maybe like aLatin America tour.
And there's all these minorleague golf tours, right?
Or you play in college and yougo right to the PGA Tour.
So the PGA Tour has a veryrich...
database of shot data and allthis stuff.

(19:22):
But the lower tours don't reallyhave that data.
But every year, there's agraduate class of depending
on...
depending on a couple of things,but there'll probably be like 20
to 30 new PGA tour players ayear.
And that's a significant amountof players.

(19:43):
Now, some of them might'veplayed a bunch of tour events,
like a Luke Clinton while he wasin college or whatever, but most
of them, you kind of don't havea ton of high quality data on.
So you need to know what to dowith these guys when they first
come up and you'll have some,some understanding of like where
they rank, but you might notknow their specific, uh, skill

(20:05):
sets you might know theiroverall skill kind of so how do
you go about quickly umidentifying like what they're
good at and what they're bad atwell you could just right when
they play their first pga tourround like let's say they gain
um two strokes off the off the tor two strokes on approach now

(20:29):
you could be like well thatdoesn't matter i'm gonna wait to
10 rounds from this guy becauseI need some kind of big enough
sample.
And the problem with that isthat while you're waiting,
you're running a sim on the fulltournament.
So there's better things to dothan wait, but not necessarily

(20:53):
the Your intuition when you'resaying, I want to wait is not
wrong because what you're sayingis like, well, that's one round.
There's a lot of variance and itmight not actually mean they're
that good at approach.
So what you can do is you cansay, well, my prior belief of
how good a person that comes upfrom the Korn Ferry Tour onto

(21:16):
the PGA Tour is at approach islike a negative point.
negative 0.8.
Then you can update it.
How much weight you want to givethe prior and the likelihood
is...
How much weight you want to givethe prior and the observed data,
that's going to be somethingthat you can tweak a little bit.

(21:42):
It's a better place to startbecause it's not random.
Picking your prior a lot of thetime is going to be based on a
bigger kind of general samplethat you can apply to the single
player.
So like, what is all the KornFerry Tour graduates?
What's like their averageapproach scale?
Okay.
And then you can start updatingit.
And the difference between doingthat and just waiting to collect

(22:05):
your data is that like, by thetime you get to enough new data
that you would start to feelcomfortable, if you were using a
Bayesian method with a priorthat made it any amount of sense
you would have gotten to a pointquicker where you had useful
information

SPEAKER_00 (22:28):
right because because in sports betting so
much of the bettingopportunities that are good are
exactly like relatively earlylike it is when there's a
limited amount of data for forpeople generally it's not
something like in different

SPEAKER_02 (22:48):
oh sorry

SPEAKER_00 (22:53):
Shoot, I realized I was on mute.

SPEAKER_02 (22:55):
No, I muted you by accident.

SPEAKER_00 (22:59):
Oh, okay.
All right.
That makes you feel better.
I'll just keep going then.

SPEAKER_02 (23:05):
Yeah, go

SPEAKER_00 (23:07):
ahead.
Okay.
In different domains, I thinkit's totally fine to not use
Bayesian statistics.
This is where I'm more of in themiddle camp that each has a
role.
But in embedding, Becausethere's just a large percentage

(23:27):
of the betting board has to dowith relatively small data
problems, it's really important.
So I think using your golfexample and sort of jumping off
of that, if you think about howa lot of people model things
right now, and I'm going to usebaseball just because I know
more about that.
I don't know the statistics asmuch in golf.

(23:48):
But in baseball, let's say youwere trying to put together a
model that's going to predictstrikeouts for a pitcher.
A lot of how people would dothis would be they would get
some sort of strikeouts per gamemetrics or K per nine or all
different metrics into a dataset and then fit some regression

(24:10):
model.
And then they're going to usethat to predict strikeouts.
The problem with that, whatyou're going to end up happening
if you do that and you don'tmake any considerations for some
of this, like this prior beliefstuff is you are going to have a
disproportionate number of betson people who don't have a lot

(24:32):
of data because you're going tofit a relationship between say
like strikeout rate and how manystrikeouts a person has like
historical strikeout rate andhow many strikeouts they have in
each game.
The problem is like, if youthink of like someone who has 10
starts and, just by volatilityitself, like those people are

(24:52):
going to have strikeout ratesthat are outside the band of
like a quote unquote, normal,stable, um, strikeout rate.
So, so what you're going to endup having is you're going to
have some rookies who have threeor four really good games and
you're going to think they'relike the best pitcher of all
time, you know, and, andconstantly be betting their
overs.
And you're going to have some,some pitchers who get, um, beat

(25:14):
up pretty bad in the firstcouple of games.
And you're going to thinkthey're never going to strike
out a pitcher again, becauseyou're not, you're not
accounting for the fact thatlike, there's a, a credibility
in there, or maybe a better wayof saying a lack of credibility
in their underlying metrics thatyou're using, um, to predict
whatever you're predicting.
And so you need to account forthat some way or else you're

(25:38):
your bets are, again, becausewe're betting, like you're
betting where you're different.
And if you're not accounting forthis, this is where you're going
to be different.

SPEAKER_02 (25:48):
Yeah, I think actually the key point of the
whole episode is like a lot ofthe opportunities in betting are
a lot of the opportunities yourmodel might fly to you in
betting.
And basically a lot of theinteresting things in betting
are low data spots, right?
Like SP said.
So I think you're right.

(26:09):
And it's not a frequentist,first Bayesian episode or
whatever, but I think you'reright.
There's tons of frequentiststatistics that I use all the
time.
And it's not like one istechnically the be all end all.
I think they kind of are superrelated and whatever.

(26:31):
But the reason where...
I kind of had this belief thatsports betting in a way is like
sports betting and trading.
These are things that will makeyou into a Bayesian because it's
just such the clear use case ofit where you've used the term

(26:55):
inference a couple of times.
You're not...
trying to describe data, you'retrying to predict data, you're
trying to take small amounts ofdata, and that's where the prior
is going to actually matter.
One thing we can talk about isif you have a prior, and this
was an example that I think wetalked about before the show,

(27:18):
but if you go back to the golfexample, if I have Rory
McIlroy's approach prior,because you might be like, well,
What about a player that cameoff of the Korn Ferry Tour like
Justin Thomas?
And now he's a great approachplayer.
He's not a negative 0.8.
Wouldn't that affect himnegatively?

(27:39):
Not at all.
Because Justin Thomas now hasbeen on the PGA Tour for 10
years.
We can give him that negative0.8 approach prior.
And at this point, there's somuch data on him that won't
even...
So a lot of where Bays matters,like SPO is talking about, and

(28:02):
it syncs up with where there's alot of opportunity in sports
betting.
And I think that's why it's sucha match made in betting heaven,
because it just threads thisneedle of that's where the
opportunity is, and that justhappens to be the area where
Bays matters.

(28:22):
is clearly better than like afrequentist method or like
methodology.

SPEAKER_00 (28:29):
Yeah, exactly.
I mean, just this is, this is,this is sort of true in all
sports is like, if you just lookat it from like, how much do
lines move in anything, whetherit's the actual like game lines,
totals, player props, anything,you know, ignore injury news,
ignore all that part, but linesare just going to move further

(28:50):
forward.
from the open in most sportsearlier in the season.
And that's just because there'smore uncertainty for everybody.
And so if you are able to, whatthat means is if you're able to
have, because early in theseason is really all priors for
a lot of things.

(29:11):
It depends what we're doingagain, but it's much more
priors.
Like week one, college footballis is basically who modeled the
transfer stuff these days.
Who modeled the transfer stuffthe best would be my guess is
effectively what that game isand who's the most on the news
and stuff.
College football in general,there's 11, 12 games, maybe more

(29:33):
now with all the playoffs, butyou're going to need the priors
the whole season.
You're just not going to haveenough data points.
It goes to show that it isreally important in betting.
And what I will say is it's okayto think you can't do the prior

(29:55):
stuff well.
If you want to bet collegefootball, but you're not
grinding or have some process tounderstand transfers and
everything.
If you had some model from thatone six years ago, but it is not
equipped to handle all thatmovement and stuff, that could
be okay.
But you just can't be betting inspots where you need aggressive

(30:20):
priors, like week one, week two,those types of things.
So it's okay to havedifferent...
If you don't feel like you'recapturing priors well or...
In the baseball example, forexample, let's say I had a
strikeout model, but I don'tfeel like it's properly
accounting for actual pitcherskill coming from the minors.

(30:41):
If I just assigned...
everybody the same skill level,let's say I had no minor league
data, and I just wanted toassign everybody as the same
skill, like average minorleague, that could be okay.
But probably in that case, Iwouldn't want to be betting like
people's first starts orwhatever.

(31:01):
Because I have no, no, likethat's a pretty uncompetitive
prior in that case.
So I think you need to like sortof have a little bit of a self
reflection on like, is what I'mdoing for the prior actually
better?
And it's true on both ends,right?
It could be like, is what I'mdoing late season better than
these people who have reams ofdata and are like really good?

(31:23):
Because maybe you're just likereally into college football and
you follow the news and whatnot.
Then maybe that's a place whereit's okay to bet certain things
early.
But when people who are betterwith data have eight weeks of
data, like maybe you don't wantto be betting against those
people.
So, you know, it's like...
attacking where you think yourstrength is.

(31:45):
And

SPEAKER_02 (31:46):
this goes back to our comment, what is your edge?
Where does your edge come from?
Which, you know, the questionthat we should always be asking
ourselves.
But priors, and why I like thisis because it's like not, it
gives an avenue to the ballknowers to compete as long as

(32:09):
they know or where this is wheremy edge comes from.
Because to me, priorsdisproportionately rewards
actual sport knowledge comparedto collecting observed data and
making predictions based offthat.

(32:33):
Setting a prior, it's going toalways reward you for
understanding the actual sport.
And you could...
Your prior could be like, yeah,I think SP said it perfectly
with college football or tradeslike big off-season moves.
Like, okay, this team now hasthis quarterback.

(32:53):
Okay, so how do I adjust thisteam?
Because you know quarterbacksplay a massive role in team
rating, so that's important.
And then you have some belief onnot just the skill of that
quarterback, but how...
they're going to fit into theirnew team relative to their old
team and relative to the team'sold quarterback.

(33:14):
And that's a situation wheregoing into week one, if you were
an ex-NFL coach with some kindof basic knowledge of
statistics, I wouldn't thinkit's that insane for you to come
to me and be like, I reallythink...

(33:35):
this could be a spot.
If there was just some touch ofBayesian thinking involved in
it, but because you understandhow basically a prior could be
built better than somebody who'sreally a data analytic
specialist, week one, week two,I'm on board with with maybe

(33:59):
hearing you out and thinkinglike this is an actual edge
whereas if it's you know week 14i'm less inclined to unless it's
like a big trade or somethinglike i'm less inclined to think
like your approach is going tobeat the big you know the star
lizards of the world or whatever

SPEAKER_00 (34:19):
right right i mean if you're not competing with a
data advantage, you want to playgames where data is sparse.
You want to play in the poolswhere you have as many weapons
and tools as everybody else.
If you don't have a lot of data,tools, or ability, you don't

(34:41):
want to compete in data-heavyendeavors.
This is my opinion, nobody couldball knowledge their way to
projecting I don't know, someMax Scherzer's strikeouts price
better than someone who's usingvery data-intensive methods in
his 25th start this season.

(35:04):
There's no ball of knowledgethat's going to outdo that.
Now, rookie coming up, playingin, let's say, early in the
season, playing in Sacramentowhere the new team is and nobody
really knows how the fieldnecessarily is going to play
out.
I could buy that.
I could get more behind thattype of thing.

(35:24):
I think just competing where youactually want to be is
important.
The other thing I wanted to saywas, this is maybe a stereotype,
but in general, whenever I thinkabout this, And I talk to
anybody like that and it startsto go in this conversation.
I think in general, people whoare like new in their modeling,

(35:47):
Bayesian thinking, sportsbetting journey almost always
don't rely on priors enough.
Like they're almost always tooquick.
And you see this in like, media,like the way that like
non-betters talk about sports,right?
Like team wins a game, they'rethe best, they're going to win
the Super Bowl, right?
And I think that's like hownovices, you know, or people who

(36:08):
are new to sports betting,that's obviously like an extreme
version.
But I think the idea of likewhat I was saying where pitcher
has 10 good games, you're just,you're always betting the over
on them.
In general, I think people earlyon do not regress or rely on
priors enough.
Yeah.
And I think the opposite, Ithink in general, bettors who've

(36:30):
bet for a long time or moreadvanced bettors, obviously, if
they've had success, they'redoing well.
But if they're going to fall onthe wrong side, I think they're
more often to fall on the sideof relying too much on priors.
Or another way, I know this isslightly different, but it's
similar, at least in my head, isregressing too hard to means.

(36:52):
Because I think we've talkedabout this a little bit, a lot
of modeling...
techniques that that that peopleuse in sports like the most
common ones are not gonna pickup the the outlier when it's
actually there very well um andthat's sort of by design but um
you know it's just i alwaysthink about that because i try

(37:14):
and think of like where am iWhat outlier or what am I
possibly missing in thissituation?
Because for me, I think ingeneral, if I'm missing
something these days, it's muchmore I'm relying too much on a
prior belief or I'm regressingsomeone too hard or a team too

(37:34):
hard to quote unquote average.

SPEAKER_02 (37:38):
Yeah.
I saw that in the show doc thatyou sent and I was thinking
about it where I fall and I'm

SPEAKER_00 (37:46):
curious what you've

SPEAKER_02 (37:47):
done.
Yeah.
I've kind of tried to, I guess,probably fall on...
I think probably fall towardsmaybe regressing a little bit
because you feel like it's saferor whatever.

(38:08):
And also, if you're doingsomething...
at scale and you're trying tohave everything work really
smoothly without having a ton ofhuman input, a lot of times you
may sacrifice a little nuance atthe edge cases.

(38:30):
But I think what I try and do isI try and get off priors as
quickly as I can by gettingreally useful observed data you
know doing everything becauselike you could have what's a
good example of this okay let'ssay you had a a guy come up in

(38:51):
baseball and you want you thinklike oh this is like a power
hitter but you don't know ifhe's a power hitter so you're
just like watching how many homeruns he hits well that's gonna
be like really spiky and likenoisy and you might miss like
you might just like have acouple of balls caught at the
warning tracker or, or whateverthat were like hard hit balls.

(39:15):
I, again, I don't know thebaseball terminology, so I know
there's some stuff like this,but you could be like, okay,
well, what, what is like thevelocity of the ball leaving his
bat when he connects?
Like now I'm thinking like,okay, I don't need to, I don't
need to have like this priorthat like, he's everybody who

(39:37):
comes up from AAA is like a 10run hitter, 10 run, 10 home run
hitter.
When this guy has like a top 10velocity, even though he hasn't
hit a ton of home runs, like hisvelocity is just clearly at an
elite level.
Okay.
I'm off my prior or much faster.
Like I'm off my, my 10, my,whatever my made up prior is

(40:00):
from somebody coming up fromAAA.
It's because instead of usinglike, basically data that takes
a slower time to realizesomething useful, I found
something that does it faster.
And I think that's always mybattle, is to find that thing

(40:22):
that does it faster, basically.
Or to create some metric thatdoes it faster than just the raw
data.
basic data that's coming in.
So I'd say I am less of a priorguy, but I'm always cognizant of
it and I'm just fighting thefight of not having to be super

(40:44):
tied to my priors or tied tothem for as short of a time as
responsibly possible.

SPEAKER_00 (40:51):
Yeah, no, I think that's a really good way of
thinking about it, of trying toidentify In my example, and I
think we've talked about thisbefore, if certain things are
stickier and stabilize fasterthan others, like in the
pitching example, maybe a betterway to inform a prior rather

(41:12):
than strikeout rate where that'sgoing to take maybe a longer
time to get a credible number.
you can look at just how fast apitcher is throwing.
That's something that's going tostabilize in one game.
And that can help inform abetter, more accurate prior, if

(41:34):
nothing else.
Because I do think, just to beclear, I do think that's the, in
general, the side you want to beon probably is the
over-regression, the over...
reliance on like, quote unquote,like average and whatnot, and
not assuming everybody's anoutlier.
But the challenge with bettingagain is like, you're betting

(41:56):
when you disagree with themarket only, right?
So it's not like enough to, youknow, you have three guys,
they're all quote unquoteoutliers, maybe two actually
aren't, right?
but you're going to have themost betting opportunity on the
one who actually is an outlier,right?

(42:17):
And like someone else iscapturing that better and you're
going to have the most likemarket disagreement with that.
So it's like not enough to like,that's what I think is missed on
some of these like modeling,like these public models that I
see is like, they might be righton average or accurate on
average.
Like people will talk aboutlike, whatever, like mean

(42:41):
squared error or whatever, likein a model, that really doesn't
matter in betting because theonly mean squared errors or
differences to your, you know,that matter are the ones you're
actually betting, like to themarket.
You could be, it does you nogood to be 100% accurate with
things that agree with themarket.
All that matters is how accurateyou are when you disagree.

(43:04):
And so that's where, again, likethere's like a, premium on being
right on outlier people, justlike there's a premium being
right on things early.
Cause that's where theopportunity and disagreement is.

SPEAKER_02 (43:16):
I'm probably going to make that the intro.
That was good.

SPEAKER_00 (43:23):
Okay.
Good.
I'm glad.

SPEAKER_02 (43:25):
Do you want to talk about Markov chains quickly or
any of the, you, any of

SPEAKER_00 (43:32):
the, yeah, we can talk about something just to
introduce that for like maybepeople who are more advanced,
um, Maybe just before doingthat, I just wanted to maybe
give a couple warnings if you'regoing to go down that I wanted
to share with Bayesian thinking.
First being, and this is sort ofwhat I was talking about, is
depending how you're doing it,depending if you're using formal

(43:54):
techniques or whatnot, it can berelatively hard to overcome a
bad prior.
And Bayesian thinking isgenerally slower to react to
changes than other methods.
And so...
you should want to be sure youwant that in what you're doing.
There's some cases where that'sgood.
There's some cases where maybethat's not.

(44:14):
I think you were talking to goodways to maybe move your prior
faster, even if it's manual orlooking at things, which I think
is important because insomething like college football,
if you have a bad prior andyou're using a classic sort of
Bayesian framework, like if wewant to talk about which I'll

(44:37):
talk about in a second, you'rejust going to be betting on that
team for a while and it mightnot be good.
So the prior is very importanton certain things.

SPEAKER_02 (44:48):
Actually, wait, I wanted to put out a disclaimer
because, yes, my actual Bayesianmodeling approach is that I see
what it spits out and then Idecide if I want to move the
prior or not.

SPEAKER_00 (45:04):
That's, that's, that's the true vibes based
Bayesian.
That's the

SPEAKER_02 (45:08):
way to do it.
Yeah.
Anyway, so now

SPEAKER_00 (45:12):
continue.
Um, we, we covered the onlyother things I had, I, we
covered like, I would just askyourself, like, do you actually
want to be betting your prior?
Do you want, like, do you thinkyour prior is better than the
market?
We talked about this, like,where does your edge actually
come from?
It's fine to like have aBayesian model with a prior and
get a nice, pretty plot of yourdistributional like prices, but

(45:33):
if you didn't actually put a lotof effort into the prior, it may
not be worth betting when you'revery dependent on the prior,
which would be early on.
So that was, that was onewarning.
And then I think what comes withthat is, well, we'll talk about
this cause I'll talk about someof the more complex methods and
I'm not gonna get too into them,but like some of these more
complex, like Bayesian methodsgive you, I can, I feel like

(45:55):
they can give you a false senseof like certainty because they
give you again, like a nicesense, you know, distribution
and you're like, Oh yes,exactly.
72% of going over 16 and a halfpoints.
Um, but it is entirely dependenton some of the, you know, the
prior you choose and how you'reactually modeling it.

(46:16):
Um, so those were, those wouldjust be some warning signs.
Whereas if you're doing like,dirty math, as I would call it,
um, or like dirty Bayesian

SPEAKER_02 (46:25):
family

SPEAKER_00 (46:26):
podcast, like a mix of like, you know, just like
stuff like that, like more ofthe napkin back of the napkin
math we talked about or modelingwe talked about with, with
trying to incorporate some ofthese ideas.
You're, you usually have morehumility around that.
Um, so, um, Yeah, those are somewarnings.
I guess I didn't really havemuch to say because there's a

(46:49):
book I think that is fairly goodthat describes this if you want
to understand some of the actual– I know we've talked about
regression and whatnot.
There's other methods to usesort of Bayesian inference and
modeling that sort of stuffthrough a Bayesian framework.
I don't actually know the nameof the book.

(47:10):
It's by Andrew Mack.
I don't know if you know thename of the book.

SPEAKER_02 (47:12):
Yeah, it's– I think it's called Bayesian sports
models or something.
I don't know.
But anyway, it's the one book byAndrew Mack that has the word
Bayesian and sports in thetitle.
And it's really good.
Actually, I'll link it in thedescription.

(47:34):
We should probably have Andrewon.
We have no sponsorship dollarsfor this.
Exactly.
As good as the book is, I don'tthink Andrew's getting rich off
the BSN.
I can't imagine that's the bestseller.
That was a book for us, maybefor him.

(47:54):
We'll link it.
Obviously, no sponsorshipdollars, but we should maybe
reach out to Andrew.
He's written a lot of good Excelbooks.
I was actually going to bringthat up because it's really not
like...
for an audio, like a short audiopodcast.

(48:14):
So it's really hard to, it'sbasically go into the actual
implement, like the true in-codeimplementation.
You're going to want to mayberead that book.
And he gives good examples ofwhat we talked about in the
beginning that are more kind offleshed out.
And I think he has one aboutlike a hockey goalie in the

(48:35):
backup.
I forgot.
That was pretty good.
But yeah, read that book forsure.

UNKNOWN (48:39):
Yeah.

SPEAKER_00 (48:39):
Yeah, it's not necessarily a light read.
It's very technical just to bestraight with anybody.
If you're not into coding andwhatnot, it's probably not for
you.
But if you are into that, it'shelpful.
It provides some of theframework about how to think of
these problems.

(48:59):
And really, to me, it just addsanother tool into your toolkit
for certain problems of how toestimate some of this and you
know, there's different, there'sdifferent sort of ways to
actually do Bayesian inferencein, in coding or whatever.
And then he goes through acouple of them in there.

(49:20):
So,

SPEAKER_02 (49:21):
right.
And it ends with the Markovchain as like,

SPEAKER_00 (49:25):
yes, which I think is, you know, by far the most
common or popular way of using,right.
Like actual, if you want to notjust do sort of like the dirty
way of, right.
I do think it's worth callingout.
Like, I think you can, hackfrequentist methods, which
frankly are much simpler to mostpeople.

(49:46):
I think there's ways you canhack those to capture Bayesian
type thinking.
One of the things that I wantedto mention, I guess I forgot,
was let's say you were doingthat strikeout idea again.
You could, instead of justhaving strikeout rate in your

(50:06):
linear regression, you couldhave strikeout rate interacting
with number of starts that theplayer has actually had.
So what that's effectively goingto do is you're going to place
more weight in your regressionmodel, like on the strikeout
rate, you know, like for everystart.
So if it's someone's 10th start,it's going to get a weight of 10

(50:27):
times the strikeout rate in theregression.
And if it's someone's 100thstart, it's going to get 100
times the strikeout rate.
So that's like a dirty non-scientific way of capturing the
same sort of things.
And you're able to do otherthings with...
I mean, that's a very simplemethod in regression.
There's other things you can dowith some other non-Bayesian

(50:51):
methods of more non-linearmodeling or whatnot that can
capture some of the fact thatWhat is the prior?
What is a sort of regresseddistribution or number versus
how much can you rely on thedata as of the 10th game, the
20th game, et cetera, et cetera?

SPEAKER_02 (51:10):
Yeah.
Yeah, definitely.
News?
Yeah.
Okay.
Here's the news.
Okay.
Okay.
Some cool news this week.
So we'll kick it off with acouple news.
of our classics.

(51:30):
Uh, the first is there's hintsof that.
Fan duels kind of kick the can,kick the tires, sorry, kick the
tires on a call.
She partnership slashinvestment.
Um, this was reported by, uh,already, uh, front office

(51:54):
sports, uh, They actually hadsome good news on that website.
I think it's interesting.
I'm not shocked, but obviouslyit doesn't go into a ton of
detail as to how this would kindof play out.
I think it's kind of funny ifKalshi's like, we're not a

(52:15):
sports book, but FanDuel nowowns us.
It's just kind of like, we'rejust playing...
this charade that everyone knowsis just fake.
And like every time somethinglike this comes up, it's just so
funny to me because like, why isFanDuel interested?

(52:35):
I don't know.
Probably because they'reoffering a bunch of sports bets.

SPEAKER_00 (52:41):
I don't know if you've used any of like the, uh,
the sweepstakes books, but likeon those, you, you can click the
little flip or this switch theflip to go, um, Flip the switch
to go from like real dollars orwhatever they call it to like
fake dollars basically.
And I'm excited for this becauseit'll be like, you'll go look at
your bets and you'll click bettab and you'll have Lakers minus

(53:06):
four and then you'll go to thepredictions tab.
And you'll also have Lakersminus four, but it will be
totally different.
One will be bettering societyand one is just having a bet.
So yeah, I think it's a matchmade in heaven.
I mean, yeah.

SPEAKER_02 (53:22):
I think it's important to know FanDuel isn't
really FanDuel, it's Flutter andthey have Betfair and they
have...
which I guess is that predictionmarket.
I mean, I don't know, but thisis like, this will be
interesting.
I do think it's interesting.

(53:43):
Like what, because FanDuel andDraftKings, and we'll talk about
DraftKings and FanDuel in asecond, the next item, but
FanDuel and DraftKings like havethe decision to make as to if
they want to, because they have,there is support at the state
level of like not wanting to,and the prediction markets to
operate just federally with nostate licensing or none of the

(54:09):
burdens that come with operatingthe sportsbooks in the state and
none of the money flowing backto the state that comes with
those licenses.
So there is a group that'sfighting these and there is a
lot of resistance.
But it's like if the Andal andDraftKings then decide they want
to side with the predictionmarkets...

(54:30):
To me, that would just be kindof a telling sign of where
things will go.
Because you have to believe thatFanDuel is operating with some
of the, probably the mostcomplete set of data right now
on this of where they want tobe.

(54:52):
And if they're kicking thetires, I don't know.
The only thing that I'm takingaway from this is like, it's
good news for Kalshi just interms of like their future
outlook because FanDuel has theoption to decide to like really
put their flag down and fightthis.
So if they're even like intalks, that's pretty good news

(55:13):
for them.

SPEAKER_00 (55:15):
Yeah, I would agree because it seems like one of the
bigger oppositions has beenlike, I think it's called the
Sports Betting Alliance, theSBA.
And I think that's mostly fundedor entirely funded by like made
up of like, you know, the, the,uh, quote on the, the normal
sort of sports books, uh, forlack of a better word, uh, in

(55:36):
the U S.
And so if they're jumping sortof ship and saying like, we're
not going to be able to stopthis, let's just get like on, on
the ride.
Um, yeah, I agree that that is apretty good tell of what, uh,
where FanDuel thinks, thinksthis is going.

SPEAKER_02 (55:56):
Yeah.
Yeah.
I, I was surprised that, ofthis, but it did seem like
DraftKings was...
Wasn't DraftKings like...
They had put in some kind ofcopyright for their own
DraftKings.

SPEAKER_00 (56:09):
Yeah, right.
I had

SPEAKER_02 (56:10):
the

SPEAKER_00 (56:11):
same thought when we talked about this, using
Bayesian thinking.
DraftKings probably knows, likeI said, more than most people on
the actual temperature.
If they're getting into thismarket, I would assume they have
a good handle on that it's goingto be a good idea.

(56:33):
Now, ultimately, they withdrewthat, so maybe that wasn't
right.
But yeah, I just see it asFanDuel getting ahead of this
before it continues to grow.

SPEAKER_02 (56:47):
Well, that's a classic DraftKings move though,
right?
They do something, they withdrawit.
They do something, they withdrawit.
They do something, FanDuel saysthey're not doing it, they
withdraw it.
But in the next news item, it'sthe flip side.
FanDuel decided to do somethingand DraftKings copied them
exactly.
DraftKings is following suitwith the 50 cent Illinois tax.

(57:10):
I think like we talked aboutlast episode, this had to be
done eventually.
It had to be done eventually.
There needed to be a point wherethe sportsbook said, okay, well,
we're going to have to pass thison or whatever.
Because until there'sresistance, there's nothing to

(57:31):
stop the states from justmilking the cow as much as they
can.
So I kind of had the senseDraftKings was going to follow.
I think it's pretty clearFanDuel's the big dog in
DraftKings is the number two atthis point.
And certainly, you know, withFanDuel's kind of either

(57:56):
implicit or explicit blessing,like falling is probably a– it
can't hurt.
You know, I don't think theywere ever going to– like I think
if you remember when DraftKingshad imposed something and
FanDuel said no– I don't know ifDraftKings passing on this one
would have had the same effectto FanDuel.

(58:19):
So I think they're probably justdoing what's best for them in
this situation.
It'll be interesting to me.
I'm most interested to see ifthis has any effect on Illinois
law.
But even if it doesn't, it'llhave some effect on other
states.
So there will be...

(58:40):
I think a positive for, fromthis, for the books pockets at
least.

SPEAKER_00 (58:47):
Yeah.
There's, there's two aspects tothis that I think are
interesting.
Cause I, I also was notsurprised by DraftKings.
I think we've mentioned that onlast episode that we, that we
expect them to do that.
What I think is going to beinteresting and they're sort of
related is will any other sportsbook in Illinois be do this I
know the tax is like verytargeted towards DraftKings and

(59:08):
FanDuel so it'll be interestingbecause I think if any other
book does it and this is sort ofto my other point of will any
book do it in Illinois but alsois DraftKings and FanDuel going
to just do this in Illinois andrealize nobody cares and then
just do this in every otherstate because I think you know I

(59:31):
I If they implement it, it couldeither be very good, I think,
for betters, this whole thing,how this is unplaying, or it's
going to be very bad and there'sprobably not going to be an
in-between.
If it's very good, I thinkenough people will complain,
Illinois will repeal this orrescind this, and states will

(59:54):
get the message like, okay.
We've hit our line.
We can't be doing this.
That would be a good outcome forconsumers generally.
The very bad outcome, which Iactually think is probably more
likely, is they implement thistax.
Nobody cares.
They still place their$10parlay.

(01:00:14):
It never wins.
So it doesn't matter.
They don't care about the 50cent tax.
And then DraftKings and FanDuelsay, let's implement this in New
York and let's implement this inTennessee and let's implement
this everywhere.
And there's really no, you know,this is like, they basically got

(01:00:39):
like a free pass to increasethe, effectively like the rake,
vig, whatever, the takeout.
They got a free pass to do itwithout getting any sort of like
backlash, I think.
Cause now it can be pointed atthe state and it's giving them a
free pass to test this.
And if it works, like they'rejust going to do it

SPEAKER_02 (01:00:58):
everywhere.
So that was my, I have aquestion for you.
What do you think would be more,would consumers get more
frustrated by just the 50 centflat tax?
Or let's say like the juice, thelike the even juice going from

(01:01:20):
like minus one 10 to to minus115 if their bet size, like if
the effective, whatever theirbet size is, the effective like
increase is 50 cents.
So like let's say you normalizethem.
Like what would be more, what doyou think would be more
frustrating for the consumer orwould cause a bigger backlash?

SPEAKER_00 (01:01:42):
So that's a really good question.
And I honestly think, you know,FanDuel and DraftKings doing it
this way, I think their goal is,is to fight taxes and send a
message across the country thatthey don't want this and they
don't want to raise prices.
Because otherwise, they wouldhave done what you just said, in

(01:02:04):
my opinion.
And they wouldn't have even doneit in a transparent way at all,
like you said.
What they would have done isthey would have just buried,
they would have cranked up theSGP margin from 30% to 35% or
whatever.
And um, just collected enoughthere and nobody would be,
nobody would even have a clue.
Right.

(01:02:25):
So like that's, if they justwanted to like sweep it, like,
okay, we're just going to passit onto the consumer and we
don't want the consumer fightingon behalf of us.
I think that's what they wouldhave done.
But I do think they want, and Iunderstandably, so they want the
consumers in this country, um,the betters in this country to

(01:02:45):
fight on their behalf basically.
Um, and, uh, and send a messageto other states.
And so that's why they're tryingto make it as transparent as, as
possible.
In my opinion, I do think Dave,maybe answer your question.
I think if you buried it, nobodywould know.
I do think if you, if like, I dosee people angry about like in

(01:03:08):
certain monopoly States whereit's like run by the lottery,
it's like minus one 30 each way.
Like, yeah, I don't think peoplewould be happy about that, but I
do think you could bury it inways that, There's no real
uproar.

SPEAKER_02 (01:03:21):
Right.
Right.
Right.
Exactly.
Because their SGP product issuch a...
It also represents so much oftheir revenue.
That's turning the knobs there.
It's all beneath the surface.
So

SPEAKER_00 (01:03:38):
it's not hard.
Not only represents so much ofthe revenue, it represents so
much of the revenue that'simpacted by this because...
Small bets, that matters forsmall bets.
It doesn't really matter for$500bets.
That's quite a small percentageincrease.
It's like when people arebetting$10 or something, that's

(01:04:01):
a problem.
Just bury the extra 50 cents inthere.
Again, nobody would know.
It's funny to me because If thisplays out this way and nobody
actually cares about the 50 centtax, I don't know what all these
people who are trying toproclaim the financialization of

(01:04:22):
sports betting are going to do.
Are they just going to put theirhead in their hands and take
their ball and go home?
Because they've been fightingthis battle of consumers getting
ripped off.
It's ridiculous that they'regetting charged.
And if nobody cares...
you're in a tough business thenbecause you're selling.

SPEAKER_02 (01:04:43):
Well, isn't that what happened kind of like with
Sportrade at least?

SPEAKER_00 (01:04:48):
A little bit?
I mean, I think that that's thechallenge.
The exchanges like have or anylow margin like Sportsbook has
faced is it's just people don't–price is not enough.
Like people don't care enough.
So this could be like a realnail in the– it could also be
like a big– boon, right?

(01:05:08):
It could go either way.
But like, if people don't, ifyou add 50 cent tax to like,$10
better is effective hold, youknow, it goes.
Or

SPEAKER_02 (01:05:17):
it goes up because they bet more, like you said,
because they just want to reducethe percent of their bet.
That is in fact the tax.
So it could, it could just belike this weird, just absolute
boon for DraftKings and Vandal.

SPEAKER_00 (01:05:34):
Oh, for sure.
But, but I guess what I'm sayingis like, if, If you're a sport
trade and you see people likeFanDuel and DraftKings are the
only ones that enact this, andFanDuel's market share doesn't
go down, DraftKings' marketshare doesn't go down, you're in
a tough position.
That doesn't bode well for youbecause it effectively means the

(01:05:59):
hold can continue to be raisedand nobody cares.
They're still going to play onthe place that has the best
product, in their opinion.
If my idea was to compete onprice and the price raised on my
competitors and nobody cared, Iwould have to take a hard look
at the mirror and say, is thisactually what I want to compete

(01:06:21):
on?
I

SPEAKER_02 (01:06:23):
mean, I think you don't ever really want to
compete on price.
Just

SPEAKER_00 (01:06:27):
as a rule of thumb.
I think if you're sellingbutter, maybe.
Yeah, true.
But for what is an entertainmentproduct, I don't know any...
It's like saying there's thishuge movie coming out this
summer.
Everybody's excited about it.
But there's this homemade moviethat I made that's half off.

(01:06:48):
That's not how people consumeentertainment.

SPEAKER_02 (01:06:52):
Right, right, right.
And I mean, it seems likeNetflix realized and now they
just keep raising the prices andno one cares.
Exactly.
Okay, well, we talked about Weteased DraftKings in, but now
we're going to talk about aninternet shattering story that

(01:07:13):
you were right in the heart of.
DraftKings voided some of yourand other bettors pick six
contests.
I guess it was last week.
You want to let the audience inon this one?

SPEAKER_00 (01:07:29):
Yeah, we're not going to turn this podcast into
my personal grievances witheverything that I've been
wronged with.
But I just wanted to talk aboutthis really because I think it's
related to some of thispeer-to-peer or exchange voiding
we've talked about where someoneaccepts or takes a bad side,

(01:07:53):
basically, and do you void thator not, and some of the
considerations there.
Yeah.
I think it's sort of a quicksummary.
Basically, there were somecontests posted.
There were lines in the pick sixcontest that were the highest
win rate, I guess, if you wantto view it that way, based on

(01:08:15):
what I saw.
It was maybe minus 200 to minus230 price, so maybe around
there.
What's your

SPEAKER_02 (01:08:23):
effective price you're getting?
On pick

SPEAKER_00 (01:08:27):
six for that sport.
Yeah, right.
You're getting maybe like 66%.
Yeah.

SPEAKER_02 (01:08:32):
Oh, no, I mean, sorry.
What's the implied on thatsport?
On pick six.
Yeah, for that sport.

SPEAKER_00 (01:08:39):
It depends on pick level, so I'm not even really
sure.
I'm guessing it's like somewherearound like 140, minus 140,
somewhere like 135, 140.
At the lowest, it's probablyeven higher for other ones.
You need to win– like the lowestat like 57, 58, 59%, something

(01:08:59):
like that.
So we're talking like you needto place picks at like almost
60% to win and these were pricedat like 66% or something.
Yeah, God forbid.
To be fair, it is outside thenorm of like what's usually
available.
For anybody who doesn't know,what's usually available is just
like what's on DraftKings lines.
And so I'm sure there's likethat– that win margin if you're

(01:09:22):
modeling or something, but it'snot like that far.
It's not like they'reintentionally including stuff
that's like minus 200 in theirpick six set.
So long story short, they, theyvoided the contest because I
think they, they had this issue.
And I just think, um, you know,and I, I think I discussed this
when we talked about the, this,uh, exchange stuff, I totally

(01:09:44):
get voiding, um, you know, likewhen you're playing against a
sports book for the most part,like I wish there was clear
rules on what like a palp orwhatever was, but like that is
part of the game.
It's just like a bad look andbad decision.
I think when you're voiding likepeer-to-peer type stuff, cause

(01:10:09):
like that's just a bad user,like not even for me, but it's
just like a bad user experiencefor most people.
like casuals who may have putaction in and then they go to
watch the game or whatever andtheir stuff is canceled and they
don't have a sweater or anythingor stuff like that.

(01:10:29):
To me, something is eitherpeer-to-peer or it's not.
And I mean, this honestlydoesn't even work me up that bad
because of how DraftKings has...
un-peer-to-peerified the VIXthing.
So it's not even peer-to-peeranymore.
But it is funny how much thatgame has changed since it

(01:10:52):
started, where the guy, Iremember, who was leading the
charge when they were fightingthe DFS pick-em apps.
I literally remember himtweeting out verbatim, DFS is
meant to be peer-to-peer.
I honestly wanted to go back andfind the tweet.
Do you think you have ascreenshot of it?

(01:11:15):
Tweeting stuff like that out andthen to see what they've done to
the game.
This was just like, you have achild.
They've disappointed you so manytimes and just one more nail in
the coffin.
That's my only personalgrievance I'll share on the
show.
I

SPEAKER_02 (01:11:32):
assume we've aired our grievances with the un-PVP a
thawing of pick six?

SPEAKER_00 (01:11:40):
We actually haven't.
I could do a whole 45 minutes onthe saga of pick six and

SPEAKER_02 (01:11:46):
get

SPEAKER_00 (01:11:46):
a lot off my chest, but I don't think anybody wants
that.

SPEAKER_02 (01:11:50):
It could come in in a hot take or I don't know.
If

SPEAKER_00 (01:11:56):
you're ever unavailable, that's what I'll do
for a solo show.

SPEAKER_02 (01:11:59):
Yeah.
Well, I mean, look, pick six wasa big part of my life last year.
Last year, yeah.
And it was great.
And I loved it.
I really enjoyed a lot of theanalysis and building of tools I

(01:12:21):
did around PIX6.
And a lot of that had to do withthe peer-to-peer nature of it.
And now it's obviously anythingbut that.
And we won't do it.
But I do think what'sfrustrating with something like

(01:12:43):
this, and it goes back to thattweet that you referenced, and
any attempt I've had to talkwith my VIP or whatever and be
like, the issue is that it's nota peer-to-peer game.
It's like if you have minus 200lines, If it was truly

(01:13:04):
peer-to-peer and you're justtaking a rake, then DraftKings
would not care.
The problem with now whatthey've done with increasing the
floors, and that was really whatmade it a player versus house,
because all of a sudden therewas opportunities where you
would just win at the minimumpayout and everyone would jam in

(01:13:28):
the same place.
plays and you could win becausethere was no punishment for
being duped.
The problem now is you createall these offshoot problems of
now you can only go over oncertain props or now certain
props have a payout boost.
It makes literally no sensewithin the realm of

(01:13:49):
peer-to-peer.
Or this, you now void bets.
It's just become like...
where you tried to solve oneproblem, which was like recs
were probably a losing too much.
I think in my data, I think Ishared this with you in my
database, but like the rec ROIor like the not, if you weren't
like in the top 20 of players involume, at least for NFL or

(01:14:12):
something, you were losing like50%.
It was like crazy.
So one, the recs were losing toomuch, but two, when they won,
they won really small and theycomplained.
And like, I understand that is aproblem, but it just feels like
now we're here where you'reavoiding entries.
And like, it was a predictableroad.

(01:14:34):
I think you and I both set atime and you said to your rep,
like, basically this is exactlywhat will happen.
Like some form of this.
And for them to be disingenuouson if it's peer-to-peer or not,
or wanting to be peer-to-peer,it's just frustrating because
you, one, engage with them ingood faith around the discussion

(01:14:56):
on the product, but now theywill not basically engage in
that discussion because it'sbecome some kind of fake
peer-to-peer and it doesn't helpthem to admit that.

SPEAKER_00 (01:15:08):
Yeah.
I've had conversations withseveral people who work there in
some respect.
They've been open to hearing atleast me out but nothing has
changed, right?
So like they've decided whatcourse they want to go on with

(01:15:29):
this.
And I think you captured itperfectly.
It was one, you know, they had apivotal like fork in two roads
or fork in the road and theydecided to go into like the
price pick light product.
And yeah, so they're just facingall the problems that price
picks faces, limiting, you know,pulling stuff off the board,

(01:15:51):
voiding, one directional props,all this stuff.
You can only put so much, youknow, it's funny, like
DraftKings, it'd be crazy.
Like DraftKings DFS, salary capDFS, it'd be insane if you
could, you know, you can make150 lineups.
It'd be insane if you could onlymake, let's say like you enter
your 76th lineup and they'relike, no, no, no, you've had too

(01:16:11):
much Patrick Mahomes.
Like we need to put someone elseon lineup.
Like, Because it's apeer-to-peer game.
You should be able to dowhatever you want.
But, you know, pick six isn'tlike that now.
It's like, oh, you've hit yourlimit on this person.
So it's just, it's all because,you know, they're trying to
protect the soft liquidity.
And I'm not necessarily in aposition with the data I have.

(01:16:35):
I don't have all the data theyhave to make that decision.
What I do have, the data I dohave, and I do collect a decent
amount for this.
They're not.
Protecting the soft liquidity.
They are just as worse off asthey were before.
Maybe they're getting lesscomplaints.
That's the data I don't have.
But they're just as worse off asthey were before.

(01:16:56):
And it's just worse and lowerliquidity for everybody else.
So disappointing, but such islife.

SPEAKER_02 (01:17:03):
I loved the NFL picks.

SPEAKER_00 (01:17:07):
It was fun too.
That's the thing that getsmissed on it.
It's like...
a peer-to-peer game is justgenerally more fun than like
prize picks.
Like for, for someone who wantsto like think a little bit for
like a true casual, it probablydoesn't matter.
And so I get that perspectivefrom them.
Like, but like for someone likeyou or I, who wants to think, or

(01:17:30):
even someone like cat, likethere's plenty of people who
don't win a DFS, but they likethe game of it.
Like it's more fun than justlike making a, like betting on,
the Lakers tonight or whatever,like it's, it's more fun.
Um, so that's, that's another,that's just, you know, another
disappointing part of it is justless fun.
I

SPEAKER_02 (01:17:48):
agree.
It really was the most fun Ihad, um, in gambling or, or
betting in a long time and likea really long time because it,
it, part of it was causeobviously it was, it was
profitable or there was a lotof, there was a lot of square
liquidity, um, obviously, but I,The thinking part of it in the

(01:18:11):
building.
It was all a lot of fun.
Anyway, did we offer DraftKingsif they wanted to?
We'll give them lifetimesponsorship if they switch.
If they switch pick six back, Idon't know if I speak for you or
not, but if you switch pick sixback to normal...

SPEAKER_00 (01:18:31):
I'll sponsor.
Yeah, yeah.
You can just be the sponsor.
I'll give a shout out everypodcast.
Yeah, you put you on the

SPEAKER_02 (01:18:37):
cover.
You don't have to pass anything.
Just switch Pick 6 back to whatit used to be.
And you can be the sole sponsorof the show.
No money down.

SPEAKER_00 (01:18:46):
Correct.

SPEAKER_02 (01:18:47):
Okay.
That's out there for whoever.
Whoever from Pick 6 whodefinitely listens to this.
Okay.
Q&A?

SPEAKER_00 (01:18:58):
Yep, let's do it.
I think we got some good onesthis week.

SPEAKER_02 (01:19:00):
Okay.
So the first one was a buzzerbeater that just missed last
week, but it's a goodpeer-to-peer-ish question
transfer.
So Lady Lado Potato asks, whenbetting into exchanges, what
amount of available liquidityscares you enough into thinking

(01:19:22):
twice about placing the bet?
Of course, it depends on theleague and bet type.
mainline player prop, et cetera.
But can you give us someexamples of what would or would
not cause hesitation to placethe bet or get you instead to
look to bet the other side on anormal book entirely?
I can

SPEAKER_00 (01:19:44):
take a stab.
Okay, you go ahead.
So I think we've talked aboutthis a little bit in prior
podcasts.
And I just want to, because Ithink we've gotten this type of
question and I think it's atotally normal one.
But what I do want to warnpeople of is in these games,
it's never going to be thissimple.
It's not going to be enough tojust look at the liquidity and

(01:20:10):
determine, is it good or not?
Especially as Profit and Novigand these companies have more
and more API people using theAPIs.
And I've personally seen some ofthis.
It's much easier to...
effectively drip the liquidityin.

(01:20:32):
Early on, if you're doingeverything by hand, it may have
made sense to put, if you want afill of$2,000, to put the full
$2,000 out and just help peopletake it.
If you have API access, there'sno reason to do that.
What you should do is drip$200in, let it be taken, drip

(01:20:54):
another$200 because you, A, cansee if you're getting filled at
that price.
And maybe if you get filled atthat price, then maybe you can
offer a worse price, right?
So you can do it in chunksbetter.
And you can basically codify allof this with the API, whereas
it's not time intensive to bepunching this all in.

(01:21:17):
So I want to say that because Iworry that people are getting
the impression that it's like,As simple as look at the
liquidity.
If there's a lot of liquidity,then I don't bet it.
But if there isn't, then I canbet it.
I think anybody who's postingprices is smart enough to know
that people know this as well.
I think Alex Monahan and OddsJam is onto this and is tweeting

(01:21:42):
about this.
So if they know about it, thencertainly the counterparties on
these sites also know about it.
And they're always going to betrying to be one step ahead of
you.
Me, I just sort of think if youare not originating, for the

(01:22:04):
most part, you just have to beextremely careful on these
exchanges right now.
Just because you don't have atrue estimate of the price and
anything that's posted there isposted there intentionally by
some party.
It's not posted by...

(01:22:26):
You don't have a ton of bigentities at this point on those
sites from everything I can see.
You have bettors who are postingstuff intentionally to try and
get arbors, top-down bettors, etcetera, to take your action.
I don't necessarily just want totalk about it as a dollar amount

(01:22:49):
anymore.

SPEAKER_02 (01:22:50):
Yeah.
My opinion is this used to be...
More of a thing than it is nowfor the exact reasons SP
discussed.
And to be fair, I have data onthis, at least some.

(01:23:10):
And there definitely was apredictive effect of liquidity
on a prop, basically, with theP&L of trading into that.
prop so that historically wascertainly the case but just like

(01:23:32):
with any other game and I thinkthis is this is like you either
embrace the chaos or not andthis is kind of like you'll find
a wall with some peopleembedding where it's like okay
well then what exactly do I dowell you know it was like this
okay so that means I get thisdollar amount I should penny

(01:23:55):
jump them or I should, you know,go bet the other side of the
sports book.
Well, maybe like it's usefulinformation.
Okay.
So, but like, what is the exactdollar amount?
It's like, bro, I don't, I don'tknow.
You know, it's going to changepeople.
You're going to, you're going toadjust and then they're going to
adjust and then you're going toadjust and then they're going to
adjust.
And the thing is that that'slike everything in betting

(01:24:22):
follows that pattern.
kind of path so but I do thinkwhat is important it doesn't
mean like okay then nothingmatters the liquidity matters
certainly having higherliquidity just in a vacuum if
you were to be like should I betinto like let's say you have a
$100 bankroll so you're notconstrained by Kelly or Betsa

(01:24:49):
you know whatever like you'reagnostic to to the sizing
advantages of betting into abigger order.
That doesn't matter to you.
So let's say there's two propsthat show value, and one of them
has$10,000 in liquidity and theother has$10 in liquidity.

(01:25:10):
Since you have a$100 bankroll,certainly the$10 one in a
vacuum, I would say, is a betterchoice to bet into, as with
anything.
But does that mean if there's a$500 one and a$10,000 one, well,
now like SP said, you could bedealing with somebody who's

(01:25:33):
dripping orders in.
Now we're getting into thefinancialization of betting.
There's financial...
trading techniques that willstart to come into the market.
And a lot of execution tradersdo just that, is you never show
your full size on the bid or onthe offer.
You never execute all in onemarket order bet.

(01:25:54):
If you're taking liquidity,you're going to do it in some
type of VWAP-esque way.
If you're posting, it's going tobe dripped in and not posted all
your whole order right on theoffer, right on the bid.
That just never really happensin financial markets anymore.
And for just this case.

(01:26:14):
But that doesn't mean we cantake a few things away.
And one thing that Lady Potatomentioned was depends on the
league and the bet type.
I'd like to make a specific bet.
caveat for bet type and it goeswith what you're saying about
who are the people postingliquidity it's usually not i

(01:26:38):
don't think on these exchangesbig entities it's just a lot of
like sharp individual people sowhat in my having met a lot of
people who either originate ordo well betting their own stuff
most of them are doing well inplayer props Most of them are

(01:26:58):
doing well in player props.
So to me, it's a big differencebetween a mainline and a player
prop, specifically on Novig andProfit.
The player props, I think, arescarier to me.
This is like a reverse.
It's like the exact opposite.
I think if anything, themainlines are just going to be
connected to Pinnacle orwhatever, basically.

(01:27:20):
But the player props on theexchanges, based on who I think
the...
liquidity providers are, are themost interesting to me because I
think that's possibly the areawhere they're going to be
disproportionately sharp.
So that would be my onlyresponse to this question is
like in a vacuum liquidity, Vinoliquidity, like it has some

(01:27:45):
effect, but it's less than itused to be.
But the player prop thing, Istill think based on the
clientele is now the keydistinction of these exchanges.

SPEAKER_00 (01:27:56):
All of those points make total sense to me.
Like the liquidity, I guess theway I would think about
liquidity, like big liquidity isstill should be a red flag, but
I guess small liquidity shouldnot be a green flag is maybe the
way to think about it.
Yeah.
So, yeah.
Uh, to the next one.

(01:28:18):
Sure.
I can read this one.
I'm curious your, your thoughtson this one.
Um, it's from Wolf JB.
Can you talk about offshorebooks in Florida?
And he only uses hard rockexchanges in, in the pickums.
Um, he wants to make sure his,uh, like his money's good.
He's risk averse in general, butwants to make sure it has a,

(01:28:40):
maybe a little bit reluctant totrust a site.
He that's unregulated with hismoney.
Um, And basically just wantingto have you talk about offshore
and or how to evaluate offshore.

SPEAKER_02 (01:28:53):
Yeah.
I think there's a lot of good...
Honestly, there's a lot of goodoffshore.
We can name them.
One thing I like about doingthis show with no sponsors is we
just can name these books.
I'm not saying that your moneyis 100% good with any of these,

(01:29:13):
but these are the ones that Ithink...
are pretty good that you canjust sign up for without going
through an agent and gettingcredit or whatever.
BetOnline.
I love BetOnline.
Never had any problems.
You'll probably get limited fromPropBuilder on BetOnline if

(01:29:36):
you're hammering SGP angle, butoverall, for example, I can't
bet on PropBuilder, but all ofmy normal golf stuff and
whatever is just I can editclick it in never had a problem
cashing out cashing in Bovadathat's like a soft offshore book

(01:29:57):
I've I there I still think thatyou would call them trustworthy
but they're a little in myopinion they will sometimes not
have a fair ruling in thecustomer's favor.
And then you have Bookmaker,never had any problems with

(01:30:20):
them.
And then a newer one, Bet105,I've also had a very good
experience with them.
They actually, I think I'vemessaged you this, they hit us
up.
We'll say no sponsors, but we'llgive you the shout out here.
Bet105 used you.

(01:30:41):
felt like that's another verygood option.
And you want to be, in myexperience, there's no point in
transacting with these offshorebooks in any way besides crypto.
So if you're using crypto andyou're depositing, withdrawing
to these books using crypto,it's going to be really easy.

(01:31:03):
You can get your crypto fromthose, at least from BetOnline
Bookmaker, Bet105.
Although DraftKings has done abetter job of withdrawals for
sure, but you get it reallyquick, you deposit really quick,
and I've never had an issue withany of them.

(01:31:23):
And I think most people in thecommunity would recommend at
least those three.
And how do you make adetermination of if it's legit
or not?
I think it's just reputation.
for most people who arelistening to this show who've
done any type of like bettinglike not DFS 2.0 or whatever but

(01:31:48):
betting they'll have heard ofthose books and they've been
operating for dozens of yearsand them stiffing you is a bad
business decision for thembecause they do a lot of
business and they've been aroundfor a long time.

(01:32:10):
And they've had reputable peoplewho've been semi-public figures
in the gambling sphereassociated with them.
So again, any book can go under,but these aren't credit books
that are run by your localbookies.
These are multimillion,$100million companies probably in

(01:32:34):
certain cases.
I bet Bovada is worth a decentamount of money.
And yeah, I think to me, thoseare kind of the ones that jumped
to my head as been around.
I bet 105 hasn't been around aslong as the other ones.
But recently, I felt likethey've done a good job.

(01:32:58):
Yeah.
I don't know.
I love Offshores.
They're great.

SPEAKER_00 (01:33:02):
Yeah, I think that seems like a good list.
It is, I think, much easier nowwith Bitcoin than it once was.
I remember my first gamblingforays involved.
I think it was called SportsbookAG, if I remember

SPEAKER_02 (01:33:19):
correctly.
Yeah,

SPEAKER_00 (01:33:20):
they might still be around.
and some skeevy checks.
It's a different game.
I think the ones who have beenaround a long time, that's what
I was going to...
The only thing I was going toadd is the longer they've been
around, the probably lessincentive they have to screw.
The longer track record theyhave of building a reputation

(01:33:42):
that's not screwing theircustomers over.
I think your list was good.

SPEAKER_02 (01:33:47):
I'm sure I maybe missed one or two, but Yeah.
Better to have a shorter list.
Yeah, better to have a shorterlist.
Exactly.
Exactly.
All right.
Let's talk about– look, somepeople in some states don't have
a lot of options, and sometimesthey turn to some sketchy stuff.
But some of those sketchy thingsaren't the offshores.

(01:34:10):
Two Natural asks, thoughts onplayer profit, funded sports
betting, and quotes.
Obviously their challenge isextremely predatory looking at
the rules and sizing, but is iteven possible with a good run?
What the fuck are they doingwith the accounts that actually
make it through?
So did you look, look into thisat all?

SPEAKER_00 (01:34:31):
I have never heard of this.
And I was, I was looking at thiswhile I was, I was looking at
this today.
This is the most insane thingI've ever, I've ever heard of.
So I

SPEAKER_02 (01:34:42):
had, so I think it's, it's, it's definitely bad,
but my, My read on this is likeone of my friends who's just so
good at sniffing out ifsomething's an end or not went
looking into these and decidedit's not worth it.

(01:35:04):
And I think that some of it iscounterparty risk and then
obviously we can talk about it.
But it's like you...
I keep hearing funded.
I get funded trading ads.
Is that a thing?
Is this a thing that's not insports betting?

SPEAKER_00 (01:35:25):
Let me tee up my understanding because you may
know more about it.
I looked at a website forlegitimately three minutes
today, so maybe I don't have agood read.
My understanding is basicallyyou go into the site, you pay
the site some amount of money.
In the case I was looking at,you paid them$800.
Then you make It's not totallyclear to me.

(01:35:48):
It's either 15, like on the siteI'm looking at, you make like 15
to 60 bets.
And if you hold like 20%, thenthey give you like$50,000 to bet
and you keep 80% of what youmake on that.
And to be clear, I think whereyou went was actually where my

(01:36:10):
head went first too, of like,It's not that these companies
are predatory or anything.
That's not where my head went atall.
It was how stupid are thesecompanies that they're going to
get angle shot in some sort ofway.
Because my first thought wasyou're going to make someone,

(01:36:31):
let's say maximum 60 bets.
Again, it's hard for me todecipher what's going on in this
one I was looking at.
But you are going to have noidea about someone after 60
bets.
And my thought was, I'll get meand eight buddies to make 60
bets.
One of them is going to be up,hold 20% or whatever.

(01:36:53):
And then I have 200K of theirmoney that I get to hold 80% on
and just fire.
It's like a free roll.
It seemed like a disaster.
Now, I didn't actually do themath.
It sounds like you had a friendor partner who did it to see if
it can overcome the challengefee, I guess, to pay.

(01:37:14):
This is just one of the...

SPEAKER_02 (01:37:18):
Well, that's exactly what he...
He was like, I'm not fuckingwasting my time trying to hold
20%.
I would just be ripping.
I would just basically just bemaxing out variance and then
seeing how many accounts can getthrough.

SPEAKER_00 (01:37:38):
Yeah, right.
If you have enough people,you're going to get some
through, right?

SPEAKER_02 (01:37:44):
Yeah, and the thing is that this could never be
legit.
This could never be legitbecause you're either going to
get, like you said, it's eithergoing to be legit for two
minutes, whatever you're saying.
If the offer is able to be plusEV to the better, it's never

(01:38:07):
going to be end up being goodfor you.
And they probably will stiff youultimately.
So I think what this is, is likejust some really, really
sketchy, like scammy way ofbooking in my opinion.
But I don't, there, there isthese weird companies in trading

(01:38:34):
that like would, That you wouldbe like a trader, but then you
had to like pay money to tradethere or something.
Like I can't try and rememberthe names of some

SPEAKER_00 (01:38:45):
of them.
I didn't even catch that part atthe beginning that you had to
pay to, like when I read itoriginally, you had to pay.
And I'm like, this is, they'regoing to be out of business in
like 10 minutes.
What are they doing?
But still the, the, the sort ofmoat of making you pay.
I'm looking like$800 to$1,400depending on how much liquidity

(01:39:06):
you want to hope that you canhold 20% over 60 bets or
whatever.
You're right.
Effectively, what thesecompanies are is they're booking
that people can't do that andjust hoping, praying that they
don't have to give these peoplethe actual 200K or whatever.
Right.
I

SPEAKER_02 (01:39:27):
think basically you'd get There'd be some kind
of back off or void or whateverin terms of if you ever found a
way to make this profitable.
This is the other thing when Ihear people come to me with

(01:39:49):
angles.
I'm way more interested in aFanDuel angle than I am in this
angle because for something tobe actually an angle, the person
needs to pay.
And the people running thesecompanies don't have any money.
This is not the type of legitcompany started by somebody

(01:40:09):
who's really well thought out,understands sports betting, and
is doing something that has alot of longevity and
professionalism and everythinglike that.
So to me, regardless of ifthere's an angle here, there's
not actually an angle.

(01:40:30):
Because if there's an angle,they're just going to pull the
units.
Like

SPEAKER_00 (01:40:37):
I've been there.
You don't want to run up a 200Kbalance or whatever with someone
who has no money.
And there's just no way.
It's hilarious.
If you're listening to this, youshould go online.
I'm looking and they have liketestimonials and they are the
most like fake, like AIgenerated, like headshots of

(01:40:58):
people.
No.
Sarah L., who's a professionalsports trader, says, the
challenges are tough, but that'swhat makes them worth it.
Passing the challenge was a hugemilestone in my trading career.
Wow.
Sarah L., if you're listening,we'd love to have you on the
pod.
If you're out there somewhere,let us know.

SPEAKER_02 (01:41:16):
Yeah, we would really, yeah, we'd love to
celebrate your milestone withyou here.
Yeah, definitely stay away fromthese places, like, Don't go
near them with a 50-foot pole.
My friend who was poking aroundis very risk-on, willing to get
rolled, willing to get stiffed.

(01:41:40):
Just a legitimately good gamblerwho just has a really nice
overall understanding of riskand what it takes.
And he's deemed these to bestupid and not worth it.
So...
I would shout out to Nico theTico.
Let's all follow his path andstay away from this.

(01:42:05):
I appreciate that bring up.
That brought me a lot ofentertainment today, though.
It is pretty.
It just feels like it's anInstagram ad with some guy.
It feels like it's legitimatelythe GP Academy video I made, but
as a site.

UNKNOWN (01:42:23):
Yeah.

SPEAKER_02 (01:42:23):
Throwback to that.
This is the second part to thisquestion.
This is actually something Ithought, this is an interesting
one.
Basically, too natural.
He says he just turned orrecently turned 21.
He's gone to Vegas and he'sseverely underwhelmed.

(01:42:45):
And he thinks casual bettorsare...
He mentions that MGM, Caesars,and Circa have extremely poor
apps compared to FanDuel andDraftKings and extremely limited
options for whatever, all thisother stuff, which he mentions
like parlays and exotics,whatever props, which even his

(01:43:06):
non-sharp friends were reallydisappointed in, which that's
actually very important.
And then he says, personally,I'm from Cali.
and I would much rather betthere than Nevada, which is
wild.
Shout out to the sweeps, Iguess.
Add that to the physicalverification, no DFS, and I feel

(01:43:27):
like Vegas is just inconvenientand not what it's built up to
be, especially for a newgeneration of people used to
mobile betting.
Does this have weight?
I'd say it certainly has weight.
You nailed it.
You nailed it.
Las Vegas is not a sportsbetting town anymore, in my

(01:43:48):
opinion.
And it's just because of this.
I think it's because they hadsuch a legacy betting, gambling,
whatever business, that theytried to halfway mobile sports
betting.
They went half in the physicallocation.
That's something a lot of...

(01:44:13):
the Vegas apps have, I think, isyou have to be physically
present in a certain casino tobet on the app.
And it just turned into this,they halved it.
They didn't really do onlinesports betting, but there's no
real place for live sportsbetting, or at least they didn't

(01:44:37):
figure out how to make that todifferentiate that from mobile
sports betting, which I, I mean,I think it's really hard to do
because how do you, so they kindof just went kind of halfway and
made this like really weird,inconvenient sports betting
environment.
And Vegas is great for a lot ofstuff.

(01:44:58):
But there's a reason we nevertalked about it on our, when we
get asked the question of wherewould you live to be a full-time
sports better?
Like it definitely wouldn't beVegas.

SPEAKER_00 (01:45:09):
I mean, the way I think about it is Vegas, and I
have not been in quite sometime, but Vegas is more of like
a hospitality town.
And the companies you're talkingabout that are like big in
Vegas, like the MGM Caesars typeof like those companies, yes,
they have mobile betting apps,but really they are, again,
hospitality companies, whereasFanDuel and DraftKings, some

(01:45:33):
people might get mad at this,but they're tech companies.
Like some people might get madat that because they think their
tech is bad or whatever, andthey have to limit and
everything.
But at the end of the day,they're head and shoulders above
the other competitors in termsof their tech they're offering,
their user experience.
And so that's the difference.

(01:45:54):
It honestly reminds me of whatwe were talking about when we
were talking about that 50 centtax.
It's like the MGM, Caesars, etcetera, are like, still showing
Gone with the Wind and FanDuel,DraftKings are showing whatever,
Avatar in 3D or whatever.
It's just a better, more updateduser experience, especially for

(01:46:17):
young people.
So I will say the thing that Ithink Vegas has going for it is
it still probably has a lot ofnetwork effects as it pertains
to betting.
Yeah.
Bet Bash is there for a reason.
I think it's always going to beviewed as a hub for bettors in

(01:46:42):
some respect.
I don't know.
Maybe this isn't...
Tech companies are all over thecountry now.
There's different tech hubs, butI think Silicon Valley, San
Francisco still holds somenetwork effect benefits.
I think that will always sort ofbe true with Vegas.
Maybe that's wrong, but I thinkdefinitely that is still true

(01:47:03):
today.
So there's aspects of that thatare helpful.
You probably can just get intouch with more gamblers there
still probably than anywhereelse in the country for the most
part.
Is that necessary or not intoday's gambling world?
Maybe not, but I do think it hasthat part going for it.

SPEAKER_02 (01:47:26):
Yeah, that's a good point.
It's the spiritual home of, ofgambling at least.
And, uh, you need to have aplace, right?
Vegas is like St.
Andrews for golf.
You know, is it the absolutebest course in the world?
I mean, I don't know.
I haven't played it, but myopinion from watching, maybe
it's not the absolute best, butno doubt it holds a special, it

(01:47:50):
holds its place in the game and,you know, people come back to
it.
Um, Okay, PoutineSteam, coolname.
Want to really highlight thefirst part of this, which is
love the podcast.
Thank you.
Thank you.
Automatically, your question'sgoing to be read.

(01:48:11):
I'm a successful top-down livearb better who is having success
using mostly AusJam.
I've been testing automated botbetting scraping AusJam EV
software cheaper than API.
Uh-oh.
Don't let Alex hear you.
It's done over a thousand betsat positive supposed EV of 2.5%

(01:48:32):
or higher.
And I am at breakeven slashslight loss on these.
Okay.
So basically he's collectingdata on historic bets, which is
very smart.
This is something you should betracking.
And want to use AI to search forpatterns in the data to remove

(01:48:52):
patterns.
Bet types, betting times,whatever.
Basically use AI for filtering.
And is this a fool's errand oftrying to reverse engineer a
top-down edge instead of doingthe hard work you do and create
models?
No, I don't think so.
I don't think so at all.
I think there's like this...

(01:49:12):
I think that there's like...
There's...
this concept of like, you'reeither like an originator and
you make, you build bottom upmodels based on data from the
sports.
And then maybe, yes, you doregress it to the market at the
end, but the cornerstone of yourmodel is sports data.

(01:49:34):
And then there's like, I justuse odds jamming, but the thing
on the screen, but there's thiswhole middle kind of type, I
think of like being a smartcustom model.
top-down better.
And I think this is like, if youthink about someone who's really
successful with this, likeSpanky, as far as being

(01:49:57):
top-down, but doing it in a waythat is really thoughtful and
understanding the differenteffects, which you talk about,
betting times, bet size, maybe abook being sharp here, not
there, and you know, linemovement now using AI to search

(01:50:18):
for patterns.
Um, uh, there's, well, this isinteresting.
I don't, I don't want to get toolost in a rabbit hole, but like
maybe you'd be thinking of usinga classification model from AI
to basically tag bet or no bet.

(01:50:41):
Um, I don't know.
I, I, I don't think it's thatoutlandish to build.
If you have, I think a thousandbets is nowhere near enough to
train, uh, this model.
So I think you'd probably wantlike hundreds of thousands of
bets, but I don't think it'scrazy to think that if you have
the bets tagged properly withlike interesting, um, data like

(01:51:09):
time to post or, um, I don'tknow, distance from sharp
midpoint or size of move orwhatever, you do a really good
job of building a big database.
I actually don't think it'scrazy to think there's a black
box solution to this type ofproblem, but it's going to be

(01:51:32):
way higher than a thousand bets.
So that's why the most importantthing you can do is start
collecting your own bets.
data or your own alert data orwhatever, because this is
something that I think ispossible, but it's going to take
a lot of the right thoughtfulinputs.

(01:51:54):
I don't know.
Do you have any thoughts here?

SPEAKER_00 (01:51:57):
Yeah.
I don't think it's a fool'serrand either.
I think this is definitelysomething that could work.
What I would say I think wasgetting to some of what you were
just saying is, the way I wouldgo about it is not necessarily
just with the number of bets youhave and data you have.
And even if I had a lot, Iprobably wouldn't just throw it

(01:52:20):
blindly into either an AI orsome sort of more black box
machine learning algorithm.
What I would personally try anddo And this goes to like the
overfitting discussion we've hada couple of times of like,
because that's the biggest riskyou're going to run if you do
something like that without anysort of specification is you run

(01:52:40):
the risk of, you know, whetherit's the AI or the machine
learning, like saying like, oh,we don't want to bet on MGM at 2
p.m.
on baseball games that involvethe twins.
And that probably is not whatyou want to do.
Um, you probably want to, uh,sort of yourself use intuition

(01:53:04):
and, and, and, uh, common senseto construct variables or
features that are actually goingto be predictive in, in saying
like, this is something I wantto bet or how good, like if it's
where I'd sort of think about itas if, you know, you have it
flagged as like a 2% edge orwhatever you can model, like
what, like an adjustment to thatis based on some other features,

(01:53:26):
um, Whether that's actuallygoing to return 1%, negative 2%
or something like that usingsome other features.
But I personally would try andprobably be a little bit more
hands-on on the featureselection and just use my domain
knowledge of markets, whatmatters, what doesn't matter.
Where do I think Odds Jam isoverestimating things?

(01:53:49):
That's more of how I would doit.
But I think the general idea iscertainly a good one.
And I wouldn't...
you know, the way the questionends are doing the hard work you
do and create models.
To me, this is hard work andcreate a model.
So I wouldn't like sell yourselfshort and I wouldn't like over
glamorize like a bottom up ormodel, like a origination

(01:54:11):
approach.
Again, I think we've talkedabout it.
Like that gets, I feel likethose people sometimes get like
put on, on a pedestal orwhatever and on like gambling
Twitter, but I don't thinkthat's appropriate at all.
And in this case, like, I wouldstick with it.
It seems like you're 75% of theway there.

SPEAKER_02 (01:54:28):
Yeah.
I think top-down modeling islike a real thing.
And this is, this is somethinglike that.
I've recently gotten a lot moreexposure to with, with GP picks
plus and literally doing thistype of top-down modeling.
But like, this is a, a skilljust like bottom up and it's
hard work and like build yourbill.

(01:54:51):
You have a thousand bet databasethat just you have right now.
That's great.
You know, like keep buildingthat.

SPEAKER_00 (01:54:58):
Yeah.
Because if, if you get the, theother part I was going to say is
if you're building the featuresyourself, a thousand bets, well,
it's still not a lot is morethan like, basically if you are
using intuition, you need fewerrecords to fit something usable.
If you're solely just throwingit in and hoping for the best,

(01:55:19):
you need more records.
Right.
Right.

SPEAKER_02 (01:55:21):
And look, I'm the, I'm the, as I've said, dipping
my toes into the black boxworld.
There's some, I have some, I'mhave some positive feelings
towards it, but at a thousandbets.
No, you know, no, but you know,at a huge, this is something

(01:55:41):
that I'm going to look into atsome point.
So I don't know, like if I will,it would make sense for me to
even share my results or not,but this is at least if, if,
It's a fool's errand if youthink that me and SP are fools.

SPEAKER_00 (01:55:59):
It's an interesting point because I gave that sort
of insane example or contrivedexample about twins on Tuesday
or whatever.
But in reality, this is true ofanything.
If you have enough data, youwill pick up stuff that is
meaningful, which you wouldnever think to model.

(01:56:21):
So I

SPEAKER_02 (01:56:21):
have an idea of what it could pick up.
is it would have this nonlinearrepresentation of edge, where it
would be like 1% edge and 2%edge is okay, and then 3% and
then 4% model is good.

(01:56:42):
And then as it starts gettinghigher, it starts dipping.
And I think the black box modelwould

SPEAKER_00 (01:56:47):
be able to capture that.

SPEAKER_02 (01:56:49):
So if it shows a 15% edge, the black box model is
going to be like, Red alert, redalert, or at least you should
test it on that.
Because I think we can all inour own intuitive experience, at
least our experience of sportsbetting and know like that's
where it could get kind of wonkywhen you start showing these big
edges.

SPEAKER_00 (01:57:11):
That would make sense to me.
And I, yeah, I mean, you willcapture stuff.
I think you just, for every onething you capture that is
legitimate and In that stuff, ifyou're not careful, you also
will capture stuff that's notlegitimate.
So you just have to be careful.

SPEAKER_02 (01:57:27):
And it's all about the features, just like you were
saying, for doing even somethingthat's more descriptive with a
thousand bets.
You still want to make sure allyour features are as good as
they can be.
So just keep building thatdatabase is my...
you're doing the hard work rightnow.
Like keep doing that.
That, that to me sounds like areally worthwhile project.

(01:57:50):
And I wouldn't recommend you goto like leave this project right
now and start bottom uppingsome.
Yeah.
This sounds like a good project.
Good job.

SPEAKER_00 (01:58:03):
All right.
Last one.
JS23.
Maybe not a question, but couldyou talk about adding to a
position?
For example, he had a bet atrecorders of a unit at plus 116,
moved to minus 125, added more,basically added more to it,
moved again in his direction,added more to it.
He did mention that at loss, sowe appreciate that.

(01:58:26):
We all knew that was true.
He was humble bragging about allhis CLV, but he did say at loss,
so we appreciate that.
Basically, when adding to aposition, do you look at market,
limits, sport, time before theevent starts, or any other
thoughts about how you add to aposition?

SPEAKER_02 (01:58:44):
This is Bayesian.
This is a perfect last questionbecause you have your fair.
I assume this is a bottom-upsituation because I don't really
understand how this would be atop-down spot where you're still

(01:59:08):
betting it at minus 150 fromplus 116, but I don't know.
Maybe the steam just reallysteamed or something.
Assuming this is a bottom-upsituation, and I guess if you're
betting...
you know, at minus 150, let'ssay you have a fair of minus
200.
And so first of all, maybeyou're regressing to the market,

(01:59:30):
maybe you're not.
But let's just say for whateverreason you weren't doing any
regressing to the market, youbet, you know, whatever at plus
116.
And then that book moves tominus 125.
So anytime an action is taken, Igive that more weight.
And this is why I'm a firmbeliever that like, Line

(01:59:51):
movement is more predictive thanstale de-vigging.
But the sportsbook sees your betand they decide, okay, we have
this information.
So we now have the informationthat JS23 bet in at plus 116.
And we think that they're prettysmart and they're doing their

(02:00:12):
own Bayesian updating.
Now they've moved the line tominus 125.
But we got to give them creditfor knowing something.
So something that we don't know,right?
It's usually useful.
So, okay.
So now they've decided, okay, weknow this, we got that
information.
We're putting all of thisinformation together, new lines

(02:00:34):
minus 125, and we've moved it.
I'm now, after they've made amove and they've basically made
their recalc of their fare, I'mchanging my fare a little bit
because they've, made an actionin response to my bet or my

(02:00:54):
information.
Usually, I'm not re-betting onsuch big moves like this.
Let's just assume your fair islike minus 200.
I just think in practice, it'sfine, right?
Because it's minus 150 and yourfair is minus 200.

(02:01:15):
But in reality, oftentimes, theIf they're moving, their number
they move to is usually prettycompetitive with your fare and
there's not much value therebecause now you have to remember
they have your info, they havewhatever they use to set the

(02:01:36):
line, whatever other bets havekind of come in.
So once they move, depending onwho it is, I'm giving them some
respect.
I'm probably not rebetting.
at minus 125.
I'd probably re-bet at plus 110if it's just like an auto mover.
But something tells me this islike not an auto mover if
they're going to minus 125.
So, yeah, that's my thoughts.

SPEAKER_00 (02:01:59):
It's hard to give an exact example without knowing
like what you're betting, sizeof the market, all these things.
But to maybe also try and wrapit back to a Bayesian way of
thinking about it.
Like, If you say the price isminus 200 and it's already moved

(02:02:22):
considerably, you're effectivelysaying when you're betting after
it's moved 60 cents that theopen price was whatever, 110
cents off market.
And that is...
Possible, certainly in somemarkets.
It's possible in any market, butit's likelier in some markets

(02:02:42):
and less likely in othermarkets.
If it's bet ESPN, some obscureprop, there's a higher
likelihood that their open issubstantially far off of what
the true probability is.
If it's a sharper book in areasonably regularly priced type

(02:03:03):
of thing, They probably havesome...
I'm not saying their openers aregreat in any book or whatever,
but you have to ask yourself, doyou think they missed by 25%,
30% or whatever your price is?
Because you should use it assome sort of data point, like
what their open was.

(02:03:25):
Because likewise, they have someprocess to generate it.
It may not be as good as yours,but it is some process.
So...
Yeah, I think it really justdepends on– it also depends– I
think of this as a similardiscussion as how we've talked
about the sort of fractionalKelly in the past of you scale

(02:03:48):
that to how confident you are inyour edge and what you're
betting.
And to me, it's like if I knewwith 100% certainty that the
fair price was minus 200, Iwould bet till– minus$199 or
whatever.
But if I've only been bettingthis for two weeks, I might have

(02:04:10):
way more margin.
If I've been betting this forfive years and I'm quite
confident in my prices, that's adifferent story.
But that'd be how I would thinkabout it.

SPEAKER_02 (02:04:19):
Yeah, I think that's a good caveat because I was just
thinking there have beensituations where I've certainly
chased the line this far.
But because I know for sure whythey're wrong.
And it's just some big mistake.
So in that situation, I think myfair would probably have to be a

(02:04:45):
little further than minus 120.
But like you said, it's actuallyhow confident you are in your
fair.
And if it's like, oh, this is anobvious mistake because I know
exactly what they're doingwrong.
And it's not like, oh, theydon't know this player is good.
It's more like a structuralthing that they're doing wrong.

(02:05:07):
They don't realize that therules are different for this
event or something or whatever.
It's one of those where then, ofcourse, I'm chasing the line and
whatever if it's a huge mistake.
But it has to, like SB said, youhave to be confident in

SPEAKER_00 (02:05:26):
that.
Right, right.
It can't be saying like, oh, Iactually think this guy's
strikeout rate is 30% versus 25%like that.
I'm going to keep– no, it has tobe like this guy– like the coach
actually came out of pressconference and said he's
pitching one inning and nobodyknows about this.
Right.
Then I can keep doing this.

(02:05:47):
Right.
It has to be something like

SPEAKER_02 (02:05:48):
that.
Exactly, exactly.
That's our last question.
So I think that's a good way togo out on some– vibes-based
thinking around chasing a bet.
Any last thoughts on what peopleshould do next in their Bayesian

(02:06:13):
journey, maybe?
I think the Andrew Mack book isa good start.

SPEAKER_00 (02:06:19):
No, I don't really have too much.
I would just think about whattypes of things you're betting
and what potential holes or howyou can incorporate this type of
thinking into your process mostsimply.
I wouldn't immediately try andoverhaul what you're doing, but
there's probably opportunitiesor gaps in your thinking where

(02:06:40):
you can rely more on domainknowledge, priors, et cetera, to
fill holes.
And so that'd be my takeaway.
But I thought today was good.

SPEAKER_02 (02:06:50):
Yeah, this was fun.
If you have questions orcomments, join the Discord
community.
goldenpants.com.
Just click join and it's free.
And then you can go to thepodcast questions channel, drop
some questions in there.
If you compliment the show, wewill definitely answer your

(02:07:12):
question, but we actuallybasically answer every question
every week.
So if you have any questions forus or feedback, comments,
whatever, that's the spot to go.
Thanks everybody for listeningand we will see you all next
week.

All Episodes

Bet Like a Bayesian & SP's DraftKings Beef | Ep 108

Episode Transcript

Popular Podcasts

Stuff You Should Know

The Joe Rogan Experience

Dateline NBC

.css-15opob5{left:0;position:absolute;top:0.8rem;} All Episodes

.css-14f5ked{margin:0;word-break:break-word;display:-webkit-box;-webkit-box-orient:vertical;box-orient:vertical;-webkit-line-clamp:2;overflow:hidden;}Bet Like a Bayesian & SP's DraftKings Beef | Ep 108