Fantasy football analytics - Josh Lee, The Undroppables | Data & Analytics | Podcast

foreign

agreement that we would only talk to

guests that at least interests one of us

a lot

and today's guest Josh Lee I'm very

interested in because he is a fantasy

football analyst and just generally

obsessed with fantasy football and so am

I so I love that

and yesterday I was explaining fantasy

football to you something that you've

never really heard of or dabbled in and

I saw that you got a little interested

too so I'm just curious what piqued your

interest so far I think the most

important thing I found is like how

how real life player performance and

fantasy performance of the same players

kind of merged together and there is

just so much math goes on behind what a

player is expected to do there's just so

many different parameters that are at

play which I've found like very

fascinating

compared to let's say a scenario where

where you're not considering so many

parameters in in play

also like I found like so much

similarity between investing in stocks

and mutual funds which is something I

used to do at Goldman

and there are just so many similarities

between how you're trying to invest in

stocks that are hard to find that the

normal Market doesn't know and when you

are kind of explaining the first four

rounds of the draft to pick up the

obvious players but wear fantasy

football because like really interesting

is the remaining seven eight nine

players that you pick in the team where

you have to really know and you should

have done your research and yesterday

when I was explaining you all the my new

details about fantasy football it wasn't

making a lot of sense but then when both

of our light bulbs clicked we realized

oh it's like the stock market like you

just said and just like the stock market

you fail a lot I guess the big

difference is that when you lose in the

stock market it can really affect your

life and in fantasy football you know

hurts for a day but it's all all the

benefits of stock trading and none of

the bad stuff sounds like the perfect

thing to be doing exactly

and before Josh hops on so talking to a

fantasy football analyst for me is

perfect for the show just because it

marries my interest in data and

analytics and data science and fantasy

football

what type of guests in the future would

really interest you I think I think I

always find it interesting talking to

people who are doing non-intuitive

things uh in a way is the same thing

that we are doing here right now like

talking about a topic where we are

marrying data analytics and data science

with fantasy football there are hundreds

of others uh in the world who are doing

really smart things in Fairly

non-obvious places

for example like who is probably looking

at

weather patterns fine trading

commodities

you know a classic example is lentils

you're a lentil broker now

I mean it's so fun to talk about all of

these things and like or there was

somebody else who was having like a

drone Fleet almost like a fleet of micro

databases to cut down data ingestion

costs

and these are the really interesting

topics to talk about right the obvious

ones there are enough people talking

about it there is enough content around

it there is enough to read about it and

enough to be confused about it it's

really the hidden ones that always picks

my interest

so if there's a listener out there

that's doing either of those we want to

talk to you

and now let's hear the interview with

Josh Lee

all right welcome Josh Lee I think I

spoke to you like a year and a half ago

about a similar topic we were talking

about data and analytics and fantasy

football and A lot's changed since then

I think you were working at Capital One

and doing the undropables but now

seems like you've doubled down or you've

just gained a lot more popularity and

success in fantasy football so what

exactly are you up to now I have since

left Capital One I have moved on to kind

of a capital one-esque startup Mission

Lane it's a credit card startup focused

on the subprime population that's kind

of how Capital One got really big and

it's been a lot of fun I've been working

there for the past nine months now the

amount of attention and growth that I've

seen the fantasy football space has been

incredible just the amount of likes

retweets followers that I've gained over

the past year it's been a ton of fun the

goal is to make that my full-time job

yeah so you are the CEO of pharaoh

analytics you're also head of analytics

for sports and content Publishing

Company the undropables and you also

host the unending rebuild football

podcast is that all true

I don't know either I love fantasy

football so like it's it doesn't feel

like a job it doesn't feel like a chore

where like it's just it's just so much

fun for me to do all these different

things and I like dedicate like

different nights of the week to each of

the different jobs that I have to do

it's kind of one of those things where

like

if you do what you love all the time

like you don't even really think of it

as a job it's more so just like spending

time working on a project that you

really want to figure out the answer to

so that's kind of how I think about it

Married to the Game as some say

yes exactly Love of the Game love the

game and cow stuff is the CEO of our

company Paradigm he's brand new to

Fantasy Football I explained it to him

yesterday kind of I used the example of

of the stock market like you're trying

to find the best assets you know you

invest in some you buy low you sell High

you get really sad you're really happy

sometimes and I kind of count stuff like

you mentioned the light sparked and you

got a little interested absolutely I

mean like the parallels between trading

in the stock market and finding the

hidden gems outside the obvious ones

and that I've always find extremely

fascinating and when Parker started

explaining fantasy football TV I was

like okay I can understand that I can

get behind that so yeah very very

interesting topic and I'm used to

feeling like the dumb one with Cal stuff

he's always explaining stuff to me so

this is like my first experience where

where I'm teaching Cal stuff something

and I get to watch him feel like the

dumb one so this is an honor for me I am

a student here I

love it I love it so

I have a few big questions that I've

prepared and one of them is so in

fantasy football there's this division

between the Nerds and the football

purists right a football purist would be

you know that guy you grew up with that

played High School football and thinks

that he knows everything there is to

know about football and fantasy football

I guess so you as an analyst

where do you sit are you more analytics

driven or are you more of a football

purist or are you both

it's kind of funny that you bring that

kind of schism up between the two

because like it's definitely very

prevalent it's even prevalent in the NFL

but growing up I actually played I

played football for four years played

running back and wide receiver from when

I was sixth grade to my freshman year of

high school send me your huddle

highlights

I'll have to go digging around for for

some of those but I did suffer three

concussions while playing so I had to

hang up the cleats during High School I

played soccer all four years played a

lot of basketball now but still love to

play like flag football actually this

past week and I was at the the fantasy

football Expo

I saw that yeah it was it was a ton of

fun first game my team was up scored the

opening touchdown of the entire

tournament so it was a ton of fun

absolute blast highly recommend both you

and kostov and anybody listening to the

show you should definitely go out to be

a fantasy football Expo next year it

doesn't matter if you are a huge data

analytics person you just love fantasy

football or you're just looking for

Community it is just a ton of fun it's a

wonderful experience Dez Bryant was

there he's like the nicest guy ever like

oh that's all I was taking pictures like

having conversations with anybody that

approached them so he was super super

cool but kind of back to the original

question am I more of a analytic sort of

guy or more of like a football purist

I'd like to say I'm a little bit of both

we all build like General intuition just

from watching football right so kind of

the most beautiful part of it is when

you see something and the data agrees

exactly with what your intuition is

telling you

um and kind of that compounding factor

is usually where I like to put my chips

I like to put my money behind where I

can see something on TV on the football

field and then the data agrees with

exactly what I'm thinking and those are

the types of guides that I liked about

that's very that's very interesting

because when I used to work at Goldman

uh or even before that uh every time

you're looking at trading you have a

understanding of a company and then you

see the data to see that whether that

understanding gets bolstered by your

thinking about the company so one of the

questions I have is as an analyst you

probably spend like hours working

through models and working through repos

to discover and answer that you found

right data

so what or when making a big fantasy

football decision how do you decide what

should be analyzed versus what you

already know instinctively Ah that's a

that's a very good question it's a

little bit more Broad in general so I'm

gonna get a little bit more specific so

let's take I'm gonna get like very very

specific on from from the player

perspective let's take someone like

Derrick Henry for example Derek Henry is

a 29 year old running back wage for the

Tennessee Titans and he's been a beast

his entire career when you watch him on

TV he he looks like a man amongst boys

he's uh stiff farming he's giant NFL

linebackers into the ground like they're

like they're made of paper he's an

absolute Behemoth of a man when you're

looking at the data for running backs

though from an average perspective it

says as soon as a running back hits the

age of 26-27 their career starts to

decline one of the things that I love

about kind of blending what you see on

TV and blending the analytics is the

importance of of adding context to these

sorts of situations when you're watching

football like you can kind of identify

certain guys as outliers guys who are

going to break the model per se so

Derrick Henry is one of those guys that

not gonna follow your typical running

back production curve trajectory and

it's funny because the machine learning

models that I built were able to

identify that and it has Derrick Henry

continuing to score a lot of fantasy

points this season but I'm trying to

draft Derek Henry every single in every

single draft that I'm in as a first

round or are you looking at him more

second round that totally depends on the

size of your league and all that but is

he like do you want him number two on

your roster number one on your roster

what are you looking at if I can get him

number two on my roster that is

phenomenal if I'm in a room where it

feels like someone else might take him

before me I do not hate taking Derek

Henry in the first round typically the

back end of the first round so if I'm

drafting at like the 111 112 and it's

like a two running back two wide

receiver league and top guy drop the

board I like to click on Derek Henry

right there and I'm gonna I'm in a

couple leagues but one of my leads is

really competitive I've been in it for

15 years so I was like 14 with when we

started these things I'm curious what

type of analysis are you performing

before a draft and this can be one of

your analyzes or it can be someone

else's that you leverage yeah so you

mentioned that you're in a home League

that you've been in for a long time one

of the beautiful things about being in

these home leagues is like you have you

kind of have a good idea of like what

your opponents are going to do so you

have an idea of like who they like and

so you can play around with the draft

from a little bit that way but kind of

going back to the original question

there's no Silver Bullet when it comes

to one piece of analysis dang it there

is there's a lot of different things

that you want to blend together so that

you can build the most competitive

roster so let me ask you what is the

format of your league is it half PPR

full PPR how many wide receivers do you

start it's more popular now but it's

full PPR and we have two wide receivers

an offensive player and two quarterbacks

interesting okay yeah okay so from that

perspective you're going to want to

draft quarterbacks certainly right so

when you're let's dive into quarterbacks

a little bit while you're drafting

quarterbacks you want to look for

rushing upside like that is like the

number one thing for quarterbacks

because if a quarterback has rushing

upside

they're going to have a floor that is

much higher than any other quarterback

in the league because like let's say

guys who don't want to run a lot let's

say like in Aaron Rodgers he's not going

to get his his variance for his weekly

scoring output is going to be much

greater than that of like a Justin

Fields who does have a lot of Russian

production because Justin Fields has a

floor of like 15 points just purely

based off of the amount of uh Russian

production that he gets Aaron Rodgers on

the other hand if he has a really bad

game and he doesn't throw any touchdowns

his floor is probably like five points

and that's going to really screw you in

a given week so that's why whenever I'm

in superflex leagues or I guess in your

case two quarterback leagues I'm looking

for guys that can run so in the Scott

fishbowl

if you're not familiar with the Scott

fish boy it's a really large Tournament

of like 3 000 players when it's for

charity every single year it's an

absolute incredible sort of organization

and something that I definitely highly

recommend looking into I was actually at

a Scott Fishbowl live draft this year in

DC and it was a ton of fun you could

really see like the camaraderie the

community amongst everyone it was such a

joy my two quarterbacks are Justin

fields and Daniel Jones and I took

Daniel Jones in the third round of of

that draft because Daniel Jones has one

of the safest floors out of old quartos

my models love Daniel Johns because I

know they've been like yeah exactly

right it's like public perception kind

of use it where it's like okay Daniel

Jones is kind of a goofy guy reminds you

a little bit of like Eli Manning because

he wears the that Giants blue as well

he's the Giants quarterback so like he's

not like the most he's not he's not a

sexy pick but Daniel Jones will get you

points last year's qb9 his rushing prop

on all the biggest sports book has him

finishing fifth in the NFL uh for at for

quarterbacks for rushing yards and I'm

super excited for him in the second year

Brian doubles offense so he's he's a guy

that I've been drafting a pretty decent

amount of all right I'll take Danny

dimes I'm fine with that

yeah of course yeah I mean like it's

fascinating to hear you kind of having

that fluency of thought and you know

literally all the players by her and

what they do and it's just it's just

incredible so I I would like I'm just

wondering like

let's say and it's a it's a right it's a

question I have if I have to start

tomorrow in fantasy football and I'm

already getting excited right where

should I start what should I start from

the undroppables website and look at the

scores uh or how should I start the

andropos website is a phenomenal

resource so I go on the andropos website

and just start looking at uh some of our

our team's rankings so if if you're more

interested on like the model side I do

have a few tweets that I can send you

and they kind of walk through like my

process of like building the models and

what the model determines as the most

important variables in terms of

predicting the amount of fantasy points

that a player is going to score each

season so getting a little bit deeper

into the weeds into the model side of

house because that's something that I'm

very passionate about and I think gives

me a pretty distinct Edge in fantasy

football so I vote four different models

I built four different XD boost models

one for each position one for

quarterback running back writer super

tight end just like an understanding of

exactly what variables I should be

looking for in drafting each of those

positions so let's take quarterback for

example the one thing that popped

rushing rushing production like I like I

talked about when I was giving uh Parker

the advice to Jack Daniel Jones for

running back

the thing that popped the most there

were two things actually the first was

Russian EPA so EPA stands for expected

points added so expected points added

are calculated in terms of when the

running back gets the ball how many

expected points do they add to the team

score in terms of touchdowns so

touchdown is a big driving factor for

EPA and first downs because if you get

your team a first down the drive extends

and you have a higher likelihood of

scoring a touchdown or getting a field

goal thus expected points so Russian EPA

and then receiving yards per game was a

very predictive metric so when running

backs when they catch the ball as well

as run the ball they are some of the

most valuable players in fantasy

football because essentially they are

both a running back and a wide receiver

combined take Christian McCaffrey for

example Christian McCaffrey back in 2019

had 110 receptions if Christian

McCaffrey

had zero Russian production and only

white and only his receiving stats he

would have still finished in the top 10

for fantasy points

but you take his receiving production

and you add in all of his Russian

production as well he was the most cheat

code broken fantasy player in oh it was

awesome yeah it was incredible you

guarantee he guaranteed like you to win

the championship That season it was it

was Unreal what he what he what he could

do so that's what I'll say about running

backs for receivers it's pretty

self-explanatory actually my receiving

model was my best one by far it was able

to predict the expected fantasy points

per game within two points and the r

squared the r squared for that was 60 so

it's able to explain 60 of the variance

in the in the data just off of the the

inputs that I fed it which I thought was

pretty incredible because like there are

so many different things that go into

fantasy points that I was I was amazed

that it was able to predict it at a 60

rate and then for tight ends tight ends

is actually the in my opinion worst

model the R square was at a pretty low

32 percent and I think that speaks to

the tight end position itself because

the tight end position is so touchdown

dependent basically if your tight end

scores touchdowns then they are going to

score more points so that ties into the

strength of the offense as a whole and

that's why this offseason I actually

built another model to predict the

amount of touch challenge that each team

is going to score I haven't released

that model that's actually part of my

work with Pharaoh analytics and that

that's kind of the the driving Factor

behind us starting this company because

we've done all this work and it's it's

going to be pretty value I think and

we're trying to sell API licenses to a

lot of different sports books betting

companies content companies there's a

lot of opportunity in the fantasy sports

and gaming space so I'm excited for

what's to come yeah I'm still free man

like you're you're in the perfect

industry at the perfect time and you

have all this experience so I'm excited

to keep watching this is exciting I'm

curious in that tight end model have you

found anything interesting about Brian

keithy

who oh wait not kid oh who's the the

bills guy the the new bills tied in I

just forgot his name Dawn Kincaid Stone

Kincaid that's it what have you found

interesting about him Dalton Kincaid is

one of my favorite players this season

yeah I love Dawn Kincaid

he is he's essentially a wide receiver

in the tight end position he's the type

of profile that I that I like to chase

inside a position because if if you can

get receiving production out of someone

in your tight end spot that gives you

such an advantage over the field take

Travis Kelsey for example Travis Kelsey

earns the most targets out of anyone on

the Kansas City Chiefs roster Kansas

City Chiefs have a phenomenal offense

he's going to score a lot of points Dawn

Kincaid while he doesn't have the upside

to be the number one target earner on

the Bill's offense because Stefan Diggs

is there I think he is I haven't slotted

in for a hundred Targets this season

just based off of the amount of passing

volume that the bills typically have and

the strength of their quarterback and

Josh Allen I'm very excited for

Jefferson kid I think he's going to be

the third option in that Bill's offense

I've kind of been searching for that

third guy for a long time Isaiah

McKenzie couldn't do it for them last

season and there was a lot of Buzz

around Dalton Kincaid going into the

draft and the bills actually ended up

trading up to go ahead and draft Dalton

Kincaid in the first round so I'm very

excited to see what he can do this year

what I was like really getting

fascinated by is you have all of these

models running and constantly running

them what's like a data stack looks like

what do you use to keep the lights on to

make sure that all the predictions come

out at the right time yeah uh kind of

the beautiful thing about this is this

is all like preseason so I don't have to

worry about real-time information

real-time data anything like that I only

worry about getting all of my data from

the previous Seasons actually my

historical data set goes back to 2003

and I'm able to analyze all of that data

to predict this coming season's uh

fantasy points and because

fantasy football drafts they all

happened before the NFL season starts so

I just have to make sure that I have my

models ready by the time I I enter my

drafts and kind of the big thing behind

me spending so much I I probably spent

about two and a half months building uh

my models and it was all based off of me

trying to win this one fantasy football

tournament called best Ball Mania

Basketball Mania is a tournament hosted

by Underdog fantasy it is a tournament

of 680 000 different teams and first

place of the tournament wins three

million dollars and every month entry

yeah game on exactly every single entry

every draft cost 25 so my upside is I

can win three million dollars on just a

25 investment Arab to follow up on that

I I mean on the on droppables website

you're using xgboost as the underlying

technology uh on the models or um for

most of the predictions why XG boost do

you sort of combine random forest with

xgboost have you tried like catboost and

then that didn't work so what was some

of the things that you tried and it was

like though that's not really worth my

time so

um I'm not gonna lie I don't come from a

data science background I I'm a Statin

econ major and I had done a bunch of

research on what the best algorithms to

use for predictive analysis and a lot of

the things that that one like these

cowgirl competitions these data science

competitions was actually boost I had

done some very initial base model

building with just simple regression

analysis logistic regression too but the

results weren't nearly as strong as I

wanted them to be and then the second I

plugged in xgboost the results

skyrocketed and I was like oh yeah this

is the one gotcha I think your skin

still comes into how you build your

features right what you feed into these

models because that's where the garbagey

and garbage out problem happens right

that is right

and the last question before we let you

go I'm really interested in this one so

in that XG boost model that you well I

think it's actually the undroppables

risk score model that you created you

fed it all these data sources you had

historical fantasy football data you had

weather data you had a ton of other

stuff and it says that the real

variable that mattered was just historic

performance and if that seems so obvious

when you learn that like is that what

you expected or were you kind of

surprised I mean it was what I expected

but like I wasn't expecting it to be

that simple at the same time intuitively

like yes that it should be what we're

all looking for but because there's so

much data that's out there now it's like

oh maybe this is important maybe this is

important maybe this is important like

so many different analysts are talking

about so many different things now that

you kind of lose sight of what's truly

important and what's really important

it's how they performed in previous

Seasons so it was a nice like little

refresher kind of like take a step back

and you realize they're like all that

other stuff doesn't really matter the

only thing that you should really be

looking at is how they performed in

previous Seasons yeah Occam's razor

still stands true the simple answer is

the right one exactly that's the signal

in the noise yeah great that's all the

questions I have cow stuff is there

anything else you'd like to add before

we let Josh go I I think this was a very

fascinating conversation and I learned

so much from you Josh now I'm probably

checking out a few websites to get a

geek out and nerd out on on fantasy

football now so yeah thanks for making

the time for us Josh where's the best

place for people to reach you or which

website I'll include your Twitter where

else should people find you

yeah Twitter is definitely the best spot

to find me for all things fantasy

football everyone can find me at Joshua

jkun on Twitter jayguns about

j-a-e-h-y-u-n outside of that you guys

should start to follow along at

pharaohan analytics Fair analytics.com

more start we're building something

pretty cool over there so I'm excited to

see what the future holds for us

I will Josh thank you so much that was

awesome I'm excited for more Twitter

content more help on my teams

of course of course thank you so much

for having me on

thanks Josh see ya

so how awesome was that I had so much

fun hearing from Josh I mean like I was

totally Blown Away by the the depth of

his knowledge I really like being able

to call player by player and what

they've done in 2019 onwards and being

able to have that command over the over

so many players so many teams so many

variables is truly a truly impressive

yeah and the the best analysts or data

scientists are always the one that have

the technical expertise and the domain

expertise so like they might be experts

in marketing and they might be experts

in data science and that's exactly what

Josh is for Fantasy Football

100 100 I have seen so many times like a

data scientist or a data analyst come

into a job and then they are like

completely clueless because they're they

cannot differentiate the signal from the

noise they do not spend time in the

domain understanding the domain where

they are operating as a result they're

never able to

be successful because they just don't

know what they're operating if you are

if you're working in a revenue team

you've got to understand how the company

makes money right if you're on your

product team you got to understand what

product metrics the product teams care

about and what does the product do how

the users interact with the product if

you do not have that understanding you

will never be able to separate the

signal from the noise same here like so

many different parameters so many

different things and once you kind of

know that hey what matters is the

player's performance in the previous

season if separated so much noise that

you can focus on what really matters

yeah and my biggest takeaway was when he

was talking about this guy Daniel Jones

you may not have heard of him but he's

this very if you just saw if you watch

the game of his you'd think oh he's an

average fantasy scorer but according to

Josh in the analysis he ran he's a top

10 quarterback

and that's really interesting that he

found that Golden Nugget and I have a

draft coming up and I'm for sure gonna

be drafting Daniel Jones so I got some

value you got some value out of that I

think with a lot of this underrated

players or The Underdogs is like when

you add when you analyze their

performance across games what you see is

a very different metric compared to when

you just look at their performance on a

per game basis and all of a sudden

everything starts to look very different

it's like okay this guy is probably

really good in certain situations and

he's a consistent performer in other

situations which kind of makes him a

very valuable player to have in the team

because he's dependable and he performs

consistently absolutely and to wrap up

what was the main takeaway that you got

out of that whole conversation

so my main takeaway is that to be

effective in data analysis you have to

separate the signal from the noise and

to be able to do that you need domain

knowledge amen I can agree with that

that's all for this episode of somewhat

analytics and

thank you everybody for listening and we

will see you next week awesome