Hey Everyone!
It is week three which means more data and our models get a little bit better. Most models I make use 3+ weeks of current season data so week four is when things get fun. This week I will be introducing a new old model! The Tail Events Model! I will talk a little bit about how its made later, but only if you go retweet my tweet with this Substack link!
Don’t forget, I’m just some dude doing statistics and running models for fun. I do not bet much on any individual bet and like to be SUPER aggressive with my SGPs.
Week 2 recap:
The Colts sucked last week. I’m sure I wont make another bet with them this week. Giants actually did half of what we wanted? but Daniel Jones somehow ONLY threw the ball to Nabers lol. Kamara was the best TD call ever but no one else in the NFL scored a touchdown so whatever. Not a terrible hit rate on random players I liked last week.
SGPs
Okay. This one comes straight from the new model (which doesn’t have 100% of the data it needs yet, but who cares).
Commanders VS Bengals
The Story: Total line comes in at 47 (4th highest this weekend) and Commanders … have to pass to at least try and keep up (+7.5)? They should be playing from behind a lot and the Bengals have one of the highest adjusted pass rates in the NFL + a positive predicted PROE while winning (meaning potentially more time for WAS to rack up yards). All this on top of the fact WAS secondary is… rough to say the least #1 worst by a margin in EPA per Pass given up.
The Bet: +4200 on DK
Daniels 275+
Terry 80+ (BOSS)
Chase 100+ (not a BOSS but good lord man, its the Commanders)
feel free to get weird with TDs. I added Ertz (60+) in one of my builds since he is a BOSS this week but honestly a TD might have been the smarter move. if you wanna get REAL WEIRD Trent Irwin TD could do that for you.
Saints vs Eagles
The Story: maybe its just confirmation bias after seeing it two weeks in a row but the Saints are legit and I do not know how the Eagles are going to stop them. (#3 worst EDP per drive, #2 worst overall EDP, #7 worst EPA per pass allowed, #4 worst EPA per rush allowed). Saints decimation in all offensive facets:
The Bet Core (+850 on DK)
Saints -9.5
Kamara 125+ Rushing & Receiving
Carr 225+
Any WRs you want - I lean Shaheed because he’s the #1 but Vegas still has Olave as the CLEAR #1, but Olave TD seems appealing (-.57 XTD in my model. -.3 in PFFs). Shaheed 60+ pushes the core to +2000
Hurts 250+ is interesting with the core (thinking Saints get an early lead and Eagles pass the whole time) pushing the bet to +2500
Steelers vs Chargers
The Story: Judah got me on this story through his podcast and then the data just backed it up. PIT passes the ball … decent against LAC after LAC runs the ball well against PIT. Now, I do like this a little less with all the Herbert injury news, but its still playable imo. Both teams can stop the run (#1 and #2 in Def EPA allowed per rush) but LAC is the #1 team in perfectly blocked plays which should give Dobbins a few chances at breakaway runs?
The Bet DK +6000:
Fields 225+
Pickens 100+
Dobbins 70+ (this leg i feel less good about, but one of my sub models has LAC with the #2 predicted yards before contact so the hope would be a breakaway)
New Model has Fields 225+ at 26.6% (DK implied prob is 12%… remember they are an NFL team and Fields can definitely get the ball to Pickens (BOSS and massive buy low right now). Dobbins 70+ has to do with LAC perfectly blocked run plays and his legit explosiveness. I like the idea of just Pickens and Fields too. Fields threw for 200+ seven times last year.
Colts vs Bears
Okay okay okay okay hear me out.
The Story: The Colts zoned out last week and forgot they were playing Malik Willis. Using Sportsfolio Kings EDP we see the Bears are the 2nd worst offense in the NFL on a per drive basis. When I see that, I hear “more ball for AR5” and you know what AR5 likes to do? yeet dat thang (per PFF: AR5 #2 deep throw rate in the NFL).
The Bet (depends on how aggressive you get):
Colts -1 (though I would be willing to push this)
Richardson 250+ (new model also likes this more than Vegas)
The hard part here is the WR. BOSS Models like AD Mitchell or Pittman, but Pittman is hurt. Judah likes Alec Pierce. I see a case for all 3 tbh.
New Model likes this more than Vegas as well. (DK 20% implied, Model: 29%)
Players I Like:
Brandon Aiyuk - this became a super obvious play after all the injuries
George Pickens - if you include penalties as plays that happened. Pickens is the overall WR2. I was going to link my source and I simply cannot find my source. He is in a great position to explode one of these days.
Chris Olave… Rant - was the first player I was like “I am betting on him” but now it seems like so many people are on him. Fwiw, Olave isn’t exactly “underproducing” his opportunity. PFF has him slightly positive for FP-XFP and I have him positive. My point is “intuitively” I like him but I cant find a reason to justify it in the data. If anything, the data screams Shaheed is legit and might actually be the WR1.
Deandre Swift. Okay so the bears offense sucks BUT the Colts knew GB would be running the ball last week and didn’t stop it AND Mixon went off the week before. oddly enough they are a top 10 team in EPA per rush allowed but #1 in rushing yards allowed. Teams just want to run the ball on the Colts lol
Tony Pollard. BOSS and #3 overall RB in my metric RBOPR (just a weighted opportunity metric …. its good) against GB (#4 in RYOE/ATT allowed)
Diontae Johnson. easy right? Andy steps in and just plays better than QB32 lol.
Tank Dell. hes been on the field a lot + getting decent opportunity with a good XTD. his time will come
Amari Cooper - BOSS, Giants 2nd worst EDP allowed, and well… i was going to say “and he has a good PFF grade!” well…. he doesn’t its the 6th worst… he might be as the kids say “cooked” that makes me a little hesitant, but he’s been getting great opportunity so I’m okay with it.
Jake Ferguson - back from injury BOSS
Trey McBride - ARI is #1 in EDP and hes a BOSS… don’t overthink it
Stupid TD Parlay
Parlay:
Brandon Aiyuk (The 49ers only offensive player good XTD)
Bijan Robinson (Good XTD an RBOPR)
Tank Dell (HES GOTTA SCORE EVENTUALLY)
Carson Steele (-.51 XTD but + Pacheco makes it look good)
Trey McBride (BOSS -.36 XTD)
+8227 on FD
NEW [old] MODEL
Okay so i revamped the tail events model from last year. with a few steps:
Use time series to predict upcoming week’s variables for all of our underlying variables. The underlying variables are stuff like: Time To Throw, Time To Pressure, EPA per Pass, EPA per Rush, defensive “Get off”, adots, series conversion rates, pace of play, etc etc ~100+ variables
put those predicted time series variables into a Quantile Regression Forest to predict the desired variable. In our case: Passing Yards (for a specific game). We can look at all the quantiles now: 5,10, 15…. 50…. 85… 95
bad news about that is it only gives us a point estimate for the specific quantile BUT we can use linear interpolation to synthesize data points in between our quantiles to get a solid estimated distribution!
boom now we have a distribution and we can look at % chances for tail events!
here is the variable importance tree for the QRF! The total and spread are super helpful! but so are things like “winning PROE”, pass EPA, and series conversion.
Table is sorted by 275+ probability
Important to note: The model gets better the more weeks we have and we only have 2 soooo…. Also, the model doesn’t know about injuries at all so the model doesn’t know Tua is hurt or Herbert is hurt or any defenders are hurt etc. The model does know how teams played every week this season though so take that into consideration.
Also, I am still refining it etc etc.
Sorry for the TERRILE looking table (i made it super fast just now) but this is another concept I’m working on “Potential” … just the 75th percentile prediction - the 50th. My theory is that this number is a good representation of how easy it is for your team’s game environment to push you into a tail event. (remember model doesn’t know about Tua lol)
BOSS Models
and thats a wrap!
I may have some VERY EXCITING NEWS NEXT WEEK that MAY INVOLVE MORE DATA WHO KNOWS.
If you read this far, dude, you are awesome! and you should definitely go retweet something on my twitter <3 I would love to grow this Substack if at all possible.
Anyways, remember, I’m just some dude with a model that likes setting money on fire.