“Staring at the world through my rearview” – Tupac
Year one in the books
It’s been a hell of a ride and I’m sad the 2018-19 NFL season is over. Only 193 days until next season kicks off on Thursday, September 5th 2019, but who’s counting…
This season was a pivotal season for me…both personally and professionally.
I welcomed my first child, a beautiful baby girl, into this world right after cashing out big on our Super Bowl investments! Never bet against Brady in the playoffs!
Also, our NFL model dominated this year.
Overall Record: 169 – 90 – 8
65.3% Win Rate
Not going to lie…even I was a little surprised at how well the models performed…
Official Launch
The start of this NFL season marked the official launch of A.I. Sports. I have been betting on sports for the last 8+ years, the last 4 years quite seriously. But this year I teamed up with like-minded colleagues and together laid the foundation of a high performing, sophisticated investment vehicle.
Our investments this season have been 100% transparent and public. I’ll admit, stepping out from behind my computer/phone was a pretty daunting task. There were some serious learning curves I had to navigate. I quickly realized how easy it is for people to pick apart your plays before you make them. And how it’s even easier picking them apart AFTER the final whistle has blown.
Everything is always so obvious afterwards… but trust me, it’s a completely different ball game when YOU are the one placing sizable bets with the uncertainty and randomness of sport staring you down like the barrel of a gun in a friendly round of Russian Roulette. No matter how finely tuned a predictive model is, it’s never going to be 100% accurate. We will lose bets. We own our losses, stay humble with our wins, and always trust the statistically backed process that is the backbone of our business.
NFL Process
We’ve studied the pros and researched the “sharps” or “smart” money. There are many things we agree on but one thing we kept hearing over and over again that we slightly disagree on when it comes to the NFL. We heard how “edge” is everything and how it should be used to determine your bet size. In other sports it is much more valuable following the Kelly Criterion or a variation of it, half or quarter Kelly but the NFL has too much variation for us to follow this ideology. There is also so much focus on slight movements in the line and a fear of pulling the trigger with a bet if the line moves a half point or a point away from when you first saw the line. We’ve crunched the numbers and feel comfortable enough to trust our models.
The standard deviation between the Vegas closing line and actual final scores is 2 touchdowns, which means that they are wrong or far off the majority of the time. 75% of all games played since 2000 have been greater than 4 points away from the Vegas Spread (50% of games land between 4-15 points off), with a heavy tail missing by 30+ points.
On average, the final score is 10.44 points away from the closing line.
A slight move in the line will not make or break your play. Only 3% of all games are affected by the half point line move. Which also tells us very clearly that buying the half point and taking on the added juice is just not statistically sound. Being afraid to bet a game because the line moved a half point is ridiculous. If you are confident with your bet, stick with it.
Don’t get it twisted though. A.I. Sports 100% believes that tracking your “edge” and understanding probabilities is crucial when investing in sports. More or less our issue comes from our belief that most people simply don’t understand how to accurately calculate this metric.
For Example –
(Hypothetical situation that happens all the time)
Rams +3 vs. Saints -3
3 betting syndicates:
- Penny and the Bets
- Adam Schefter’s Source
- Karen
Penny and the Bets’ model suggests there is a large edge betting on the Saints -3.
Adam Schefter’s Source’s model shows a large edge betting on the Rams +3.
The line moves a half point to +2.5 and Karen’s model says there is no edge anymore and decides to sit out the bet.
In the end the Rams win outright. Adam Schefter’s Source feels justified in their modeling and edge calculation. Penny and the Bets come up with an excuse why they lost the bet (The Refs clearly missed a pass interference AND helmet to helmet hit on the same play that would have sealed the game for the Saints in regulation…). Karen pipes up afterword saying she knew the Rams would win and that she was leaning toward them anyway… DAMMIT KAREN, NO YOU DIDN’T! IF YOU HAD THEN YOU WOULD HAVE BET IT!!
That’s exactly how it goes everyday in Vegas.
This perceived “edge” is arbitrary and measured differently by everyone. Different variables in a model can create different predictions ending up with varying degrees of edge. Every single game has bets on both sides with people feeling comfortable with their edge. That’s all fine and dandy but in the end there was a clear winner and the bet was paid out. Nothing more, nothing less. It had nothing to do with any type of “edge” measurement that everyone continues to rely when determining their bet size.
If you have a model that is able to inhale hundreds and hundreds of variables adapting to the subtle nuances and correlations that our eyes or basic prediction models miss then you will find yourself on the winning side of the bet more often than not. From all my years of experience in this game, no bet is a sure thing even if your “edge” is huge. It’s way too easy to go broke betting too big on a miscalculated edge.
Edge is a risky metric to adjust your wager size in the NFL
There is way too much randomness. A.I. Sports utilizes a flat rate money management system for every NFL game with our goal win rate between 56-60%. We play the odds, trust our model, and benefit by volume. There will never be a single game that could crush us. We rely on statistically sound methodologies and a safe investment strategy to generate continual profits over the seasons.
A.I. Sports has 22 unique algorithms helping us reach our final recommendation for each and every game. We use aggregated probabilities from the models, which tends to skew the numbers slightly lower. We’re still toying with the final output to help our numbers reflect our true confidence in the game but that will come next season…
What I mean by all that is: our bets on games our model predicted to win at a 55-60% probability, ended up winning 70% of the time.
“Every model is wrong, but some are useful”
George Box, a statistical godfather, stated that “Every model is wrong, but some are useful.” We spent so much time over the summer fine tuning our various NFL models and automating our workflows to gear up for everything ahead making them as useful as possible. There’s a euphoric feeling when you dedicate countless hours behind the scenes and it comes to fruition in such a positive way. Obviously, there is always more work to be done for our team, but it feels good to take a moment to look in the rearview to see how far we’ve come.
Our model is useful.
Trust the Model – Bet Every Game
There are 32 teams each playing 16 games over 17 weeks. That’s a total of 256 possible games. Add in the 11 playoff games and there are 267 opportunities to make money. A.I. Sports correctly bet 169 of those games. We lost 90 games and pushed on 8 games. This gave us a 65.3% win rate. Because of our sound money management strategy we were able to generate a 420% ROI over the course of the season.
Now the fun stuff:
If you were to bet $100 ($110 risk to win $100)* on every game the model predicted, you would be up $7,390. Bet $500 and you’d be up $36,950. Bet $1,000 and you’d be up $73,900… You get the point… That’s not too shabby for 17 weeks investing. Let’s see your mutual fund or even a hedge fund get 420% ROI in that time frame.
Breaking down the model
- Overall Investments: 169 – 90 – 8 = 65.3%
- $7,390 profit
- 24% ROI on money risked
- 420% ROI since week 1
- Model Probability 50-55% Investments: 44 – 32 – 4 = 57.9%
- $535 profit
- 6% ROI on risked
- 30% ROI since week 1
- Model Probability 55-60% Investments: 59 – 25 – 2 = 70.2%
- $3,585 profit
- 35% ROI on money risked
- 204% ROI since week 1
- Model Probability 60-65% Investments: 36 – 15 – 1 = 70.6%
- $2,100 profit
- 36% ROI on money risked
- 119% ROI since week 1
- Model Probability 65-70% Investments: 15 – 14 – 0 = 51.7%
- $110 profit
- 3% ROI on money risked
- 6% ROI since week 1
- Model Probability 70+% Investments: 14 – 4 – 1 = 77.8%
- $960 profit
- 46% ROI on money risked
- 55% ROI since week 1
As you can see across the board the model performed really well. You would expect to see lower win rates with lower probabilities and an increase as we move up. It appears that the model struggled within the 65-70% range. Let’s dive a little deeper into these games…
65-70% Investments
Games that gave us trouble in the 65-70% range appear to be games betting on home favorites with very large spreads. A home favorite is typically the public pick and covers at a much lower clip. This year there was a huge difference between the Have’s and the Have Not’s. Cardinals, Bills, Raiders for most of the year, 49ers, and Jets were god-awful. Chiefs, Rams, and Saints were in a league of their own for most of the season. Their spreads began to grow and grow. They still were able to cover big spreads but not consistently.
Large Home Favorites
- Lost – Vikings (-16.5) vs. Bills Week 3
- Lost – Chiefs (-16.5) vs. Cardinals Week 10
- Lost – Chiefs (-14) vs Raiders Week 13
- Lost – Saints (-9.5) vs Buccaneers Week 1
The Vikings (1-0-1) came into their game with the Bills (0-2) off a brutal tie with their rival Green Bay. The narrative around this game was that they were frustrated with the tie and wanted to blow off some steam beating up the Bills for an easy win. The Bills lost their first 2 games by a combined 8 touchdowns. Even though a 16.5 point spread was tied for the highest of the entire season, it was not too unreasonable with what both teams had shown so far. However, the Vikings laid an egg and lost outright by 21 points. That would have been a beautiful moneyline ticket to have on the Bills.
Chiefs are on here twice. They had a great year but these two games the spread was a little too high for them to cover.
Saints vs Buccaneers… How on earth are you supposed to model for FitzMagic?
Overriding these 4 large spread bets would have put us in the 66% win rate zone…
Week by Week Payout
We only had one losing week this year. Week 11 and 14 were 1 loss away from ending slightly down but the Monday night victories in both weeks saved us.
Bankroll Growth Over the Season
Win Rate
Payout Betting on Team
Betting on the Patriots each week of the playoffs is what really separated them from the pack.
Our model loved Kansas City and Washington this year. They were recommended to bet on 15 times each. The Bills and Browns were both 100% accurate but weren’t recommended much.
Payout Betting Against Team
It didn’t happen often but betting against the Chargers was golden. They were also the second ranked team to bet on this year for us. And as you will see in a minute they were the second best team overall for our model.
Overall Payout by team
There were only 4 teams in the NFL that cost us money. Eagles, Titans, Saints, and Dolphins. Every other team was in the black. Results like this bring a smile to my face.
The Chargers were the second best team in both categories, betting on and betting against. Our model was 15-3 picking their games.
Payout by Division
Our model performed the worst when betting on teams from the NFC South. Falcons, Panthers, Saints, and Buccaneers. I remember very well how these teams continued to evade us. We would bet with them one week and against them the next and lose both ways. We need to dive into these teams during the offseason and see what we can do different. Other than that dip we seemed pretty solid.
Splits by Day
Our model did just fine on Thursday nights. Many people will tell you that Thursday’s are the hardest to model. Pro Tip: Bet the home team. This year the home team went 12-3-2.
Splits by Type
The model did great predicting for Underdogs and Pick’em games. Favorites covered at a much lower rate.
Splits by Location
Betting on Home teams
Betting on the home team we were overall 65%. Home Dogs or Pick’ems crushed it coming in at 20-6 for a 77% win rate.
Betting on Road teams
There is very little difference overall between our home and road teams splits. However, road favorites were our lowest split by a solid amount, which makes sense knowing that home dogs cover more than road favorites.
Can’t wait til next year!
Footnote:
*$250 bet size for playoff games
“You ain’t knowing what we mean by staring through the rearview
So since you ain’t knowing what we mean let me break down understanding
The world, the world is behind us
Once a motherf***er get an understanding on the game
And what the levels and the rules of the game is
Then the world ain’t no trick no more, the world is a game to be played
So now we looking at the world, from like, behind us
We know what we gotta do
Just gotta put our mind to it and do it
It’s all about the papers, money rule the world” – Tupac
