Zoltar is my NFL prediction computer program. It uses a deep neural network and reinforcement learning. The results for the first half of the season were completely crazy. The covid-19 pandemic has had a massive effect on the NFL. I switched Zoltar into experimental mode starting in week #8. I fiddled with five of Zoltar’s parameters:
Base rating pts for a win: 16 to 14
Rout definition: win by 20+ to win by 22+
Rout rating pts: 4 to 2
Underdog away win bonus pts: 4 to 2
Home field advantage: 3 pts to 2 pts
These changes made Zoltar more conservative with regards to his betting strategy.
To me, the most interesting parameter is the home field advantage. For decades, both humans and Zoltar have given 3 points to the home team. But with covid-19, you can argue that the home field advantage is dramatically greater (say 5 or 6 points or more), or dramatically smaller (say 0 or 1 point), or even negative. I won’t know until after the 2020 season is completed and I have enough data to analyze.
Using the experimental parameters, here are Zoltar’s predictions for week #17 (last regular season games) of the 2020 NFL season:
Zoltar: steelers by 1 dog = browns Vegas: browns by 7 Zoltar: bills by 5 dog = dolphins Vegas: bills by 3 Zoltar: buccaneers by 6 dog = falcons Vegas: buccaneers by 6.5 Zoltar: vikings by 1 dog = lions Vegas: vikings by 6.5 Zoltar: ravens by 6 dog = bengals Vegas: ravens by 11.5 Zoltar: cowboys by 1 dog = giants Vegas: cowboys by 3 Zoltar: patriots by 5 dog = jets Vegas: patriots by 5 Zoltar: titans by 1 dog = texans Vegas: titans by 7.5 Zoltar: colts by 10 dog = jaguars Vegas: colts by 14 Zoltar: chiefs by 10 dog = chargers Vegas: chiefs by 3 Zoltar: saints by 5 dog = panthers Vegas: saints by 6.5 Zoltar: packers by 2 dog = bears Vegas: packers by 5.5 Zoltar: raiders by 1 dog = broncos Vegas: raiders by 2.5 Zoltar: seahawks by 1 dog = fortyniners Vegas: seahawks by 4.5 Zoltar: rams by 5 dog = cardinals Vegas: cardinals by 1 Zoltar: eagles by 3 dog = redskins Vegas: redskins by 1.5
Old Zoltar theoretically used to suggest betting when the Vegas line is more than 3.0 points different from Zoltar’s prediction. At this point, it’s beginning to look like an improved strategy will be something like bet on Vegas underdogs only if Zoltar and Vegas differ by more than 6 points, or bet on Vegas favorites if Zoltar and Vegas differ by more than 2 points. I’ll have to do an analysis after this season is over.
When you bet on an underdog, your bet pays off if the underdog wins by any score, or if the game is a tie, or if the favorite team wins but by more than the Vegas point spread. If the favorite team wins by exactly the point spread, the bet is a push. You lose your bet if the favorite wins by more than the Vegas point spread.
Theoretically, if you must bet $110 to win $100 (typical in Vegas) then you’ll make money if you predict at 53% accuracy or better. But realistically, you need to predict at 60% accuracy or better.
Because experimental Zoltar is very conservative, he mostly agrees with the Vegas point spreads in week #17. However, Zoltar tentatively likes five Vegas underdogs: Steelers against the Browns, Lions against the Vikings, Bengals against the Ravens, Texans against the Titans, and Eagles against the Redskins. Zoltar sort of likes the Chiefs against the Chargers, and the Rams against the Cardinals – but that’s based on very early point spread data that will certainly change over the next few days.
Against the Vegas point spread last week, experimental Zoltar was 1-2, correctly liking the underdog Jets (who upset the heavily favored Browns). But Zoltar missed badly by liking underdogs Lions (who got destroyed by the Buccaneers) and Patriots (who got crushed by the Bills).
For fun, I track how well Zoltar and the Vegas point spreads do when just predicting which team will win (not by how much). In week #16, just predicting the winning team, Zoltar was a mediocre 11-5. The Las Vegas point spread was the same at 11-5 just predicting winners.
My NFL prediction system is named after the Zoltar fortune teller machine you can find in arcades. Machine learning Zoltar uses a neural network; machine Zoltar uses a crystal ball. The use of movie scenes with a crystal ball was very common in the 1930s and 1940s, including four of the 38 Charlie Chan films featuring Warner Oland (16 films) and Sidney Toler (22 films). Left: In “The Black Camel” (1931) Chan (Oland) solves the murder of a movie actress who was filming in Honolulu and who often consulted with her fortune teller. Left Center: In “Charlie Chan’s Secret” (1936) Chan (Oland) is called to solve the case of a murdered heir to a fortune in a creepy house where frequent seances are held. Right Center: In “Charlie Chan at Treasure Island” (1939) Chan (Toler) visits the Golden Gate International Exposition (1939-1940) on Treasure Island in San Francisco bay and solves the case of a blackmailing, murdering magician. Right: In “Black Magic” (1944) Chan (Toler) becomes involved when a murder occurs at a seance attended by one of his daughters.