NFL Week 6 | Coin vs. Machine
3-Model and 5-Model Unanimous Picks Have Performed Very Well So Far
OCT 13 2017
Week 1 - Week 2 - Week 3 - Week 4 - Week 5 - Week 6
Week 7 - Week 8 - Week 9 - Week 10 - Week 11 - Week 12
Week 13 - Week 14 - Week 15 - Week 16 - Week 17
Wild Card - Divisional - Conference - Super Bowl
3-Model and 5-Model Unanimous Picks Have Performed Very Well So Far
OCT 13 2017
Week 1 - Week 2 - Week 3 - Week 4 - Week 5 - Week 6
Week 7 - Week 8 - Week 9 - Week 10 - Week 11 - Week 12
Week 13 - Week 14 - Week 15 - Week 16 - Week 17
Wild Card - Divisional - Conference - Super Bowl
Welcome to Week 6 of Coin versus Machine. Last week’s NFL games might feel like a long time ago for sports fans – with the U.S. soccer debacle and two excellent MLB division series Game 5s sandwiched in between – but it earned itself a few words to open this week’s column.
First, the average betting line for Week 5’s 14 NFL games finished as the second-lowest of any week since at least 2002. At the time of my post late Thursday afternoon, the average line was 3.2 points on FootballLocks.com. The final average was 3.07 points, settling in three-hundredths of a point behind the final average in Week 8 of 2016 (3.04 points).
(Exact lines vary from sportsbook to sportsbook, but only a handful of weeks in the past decade have had an average line in the 3s. Week 5 of 2017 and Week 8 of 2016 are the only two weeks in the very low 3s.)
Second, last week’s games largely delivered on their promise of close outcomes. Ten of the 14 games were decided by seven points or fewer, good for the third-highest percentage (71%) in a single week since 2012. The average margin of victory for the week was 7.93 points – seventh-lowest since 2012.
Unfortunately for fans, half of those close games came in decidedly unsexy matchups, including 49ers-Colts, Jets-Browns, Bills-Bengals, Chargers-Giants, Titans-Dolphins, and Vikings-Bears. That list includes nine of the 12 worst-rated teams on FiveThirtyEight’s Elo standings through Week 5.
First, the average betting line for Week 5’s 14 NFL games finished as the second-lowest of any week since at least 2002. At the time of my post late Thursday afternoon, the average line was 3.2 points on FootballLocks.com. The final average was 3.07 points, settling in three-hundredths of a point behind the final average in Week 8 of 2016 (3.04 points).
(Exact lines vary from sportsbook to sportsbook, but only a handful of weeks in the past decade have had an average line in the 3s. Week 5 of 2017 and Week 8 of 2016 are the only two weeks in the very low 3s.)
Second, last week’s games largely delivered on their promise of close outcomes. Ten of the 14 games were decided by seven points or fewer, good for the third-highest percentage (71%) in a single week since 2012. The average margin of victory for the week was 7.93 points – seventh-lowest since 2012.
Unfortunately for fans, half of those close games came in decidedly unsexy matchups, including 49ers-Colts, Jets-Browns, Bills-Bengals, Chargers-Giants, Titans-Dolphins, and Vikings-Bears. That list includes nine of the 12 worst-rated teams on FiveThirtyEight’s Elo standings through Week 5.
|
Looking Back at Week 5
The machines produced their best week of the year against the spread (10-4) in Week 5. And after lagging the other tracked models by a few games over the first four weeks, Jeff Sagarin’s “Ratings” methodology delivered the best week of any model covered here this season (12-2). The coin went 6-8 and now trails the machine by six games on the year.
(To avoid a push in the Monday night game, I updated the line from Thursday afternoon’s Chicago +3.0 to the final line of Chicago +3.5. This change made the Bears the eighth unanimous pick among the three models for the week and moved them from a Tier V “toss-up” to a Tier IV pick overall.)
The machines produced their best week of the year against the spread (10-4) in Week 5. And after lagging the other tracked models by a few games over the first four weeks, Jeff Sagarin’s “Ratings” methodology delivered the best week of any model covered here this season (12-2). The coin went 6-8 and now trails the machine by six games on the year.
(To avoid a push in the Monday night game, I updated the line from Thursday afternoon’s Chicago +3.0 to the final line of Chicago +3.5. This change made the Bears the eighth unanimous pick among the three models for the week and moved them from a Tier V “toss-up” to a Tier IV pick overall.)
The three models agreed on eight games – most this season – and won six of them, improving their “three-model unanimous” mark to a very strong 19-10 on the year. The week’s only “five-model unanimous pick” was Kansas City (-1.0), which covered, bumping the machines up to a very impressive 9-3 in Tier I “best bets” through the season’s first five weeks. They went 5-2 in Tier II games (15-15 YTD), 1-1 in Tier III games (6-4 YTD), 1-0 in Tier IV games (7-2 YTD), and 2-1 in Tier V “toss-ups” (6-8 YTD).
|
Unanimous Picks and Tier I "Best Bets" Have Performed Very Well So Far
These are the complete standings through Week 5:
These are the complete standings through Week 5:
The first group of listed prediction methods includes the three-model unanimous picks and five-model-based tiers that I share each week. The second group shows the individual records for each of the five models that are covered here. (Note that the records I show are based on the lines used in this column. If you look at their records on ThePredictionTracker.com, they might vary by a couple of games.)
The third group starts with the “coin” (literally flipping a coin) but otherwise requires a little more explanation. The “Machine Majority (3 Model)” listing is the majority pick among the three main models (FiveThirtyEight’s “Elo,” Jeff Sagarin’s “Rating,” and Romeo Peričić’s “RP-Excel I”). When at least two of those three models pick the same team, that selection becomes the “Machine” pick each week.
Listed just below it, the “Machine Points (3 Models)” listing disregards how many of the models pick a team and instead looks at the aggregate difference between each model’s implied line and the game’s actual betting line. Only one of the three models may like a given team, but if they really like them, that point difference could outweigh the degree to which the other two models like the other side.
On Sunday, for example, Washington is favored by 10.5 points at home against San Francisco. (It's now listed at 11.0 points but let's stick with the 10.5-point example.) FiveThirtyEight’s “Elo” has Washington at -11.0 and Sagarin’s “Ratings” has them at -12.2. They both like Washington relative to that 10.5-point spread – FiveThirtyEight by 0.5 points and Sagarin by 1.7 points.
R-P Excel, meanwhile, has Washington at -8.0. So they like San Francisco against the spread and think the actual line is 2.5 points off the mark. The majority “machine” pick is Washington (-10.5), but R-P Excel likes the 49ers more (2.5 points off the line) than FiveThirtyEight and Sagarin together like Washington (a combined 2.2 points off the line). The 3-model points pick is therefore San Francisco.
Through five weeks, the 3-Model Machine Points method is 48-29 against the spread (62% winning percentage). It has performed better than any other direct or deduced method that has picked every game. The sample size is small, but we’ll continue to monitor it as an interesting solution to games in which a couple of the models barely like one team and the third model really likes the other side.
The “Machine Majority (5 Models)” and “Machine Points (5 Models)” follow the same logic but also include David Wilson’s “Nutshell Sports” ratings and Jeff Sagarin’s “Offense-Defense Method Predictions.” The 5-model majority pick reflects the team that at least three of the five models agree on. And the 5-model points pick runs the “how much does each model like them by” math across all five models.
The third group starts with the “coin” (literally flipping a coin) but otherwise requires a little more explanation. The “Machine Majority (3 Model)” listing is the majority pick among the three main models (FiveThirtyEight’s “Elo,” Jeff Sagarin’s “Rating,” and Romeo Peričić’s “RP-Excel I”). When at least two of those three models pick the same team, that selection becomes the “Machine” pick each week.
Listed just below it, the “Machine Points (3 Models)” listing disregards how many of the models pick a team and instead looks at the aggregate difference between each model’s implied line and the game’s actual betting line. Only one of the three models may like a given team, but if they really like them, that point difference could outweigh the degree to which the other two models like the other side.
On Sunday, for example, Washington is favored by 10.5 points at home against San Francisco. (It's now listed at 11.0 points but let's stick with the 10.5-point example.) FiveThirtyEight’s “Elo” has Washington at -11.0 and Sagarin’s “Ratings” has them at -12.2. They both like Washington relative to that 10.5-point spread – FiveThirtyEight by 0.5 points and Sagarin by 1.7 points.
R-P Excel, meanwhile, has Washington at -8.0. So they like San Francisco against the spread and think the actual line is 2.5 points off the mark. The majority “machine” pick is Washington (-10.5), but R-P Excel likes the 49ers more (2.5 points off the line) than FiveThirtyEight and Sagarin together like Washington (a combined 2.2 points off the line). The 3-model points pick is therefore San Francisco.
Through five weeks, the 3-Model Machine Points method is 48-29 against the spread (62% winning percentage). It has performed better than any other direct or deduced method that has picked every game. The sample size is small, but we’ll continue to monitor it as an interesting solution to games in which a couple of the models barely like one team and the third model really likes the other side.
The “Machine Majority (5 Models)” and “Machine Points (5 Models)” follow the same logic but also include David Wilson’s “Nutshell Sports” ratings and Jeff Sagarin’s “Offense-Defense Method Predictions.” The 5-model majority pick reflects the team that at least three of the five models agree on. And the 5-model points pick runs the “how much does each model like them by” math across all five models.
|
Looking Ahead to Week 6
The three main models agree on three picks this week – Baltimore (-6.5) hosting Chicago, Houston (-9.5) hosting Cleveland, and Detroit (+4.0) at New Orleans. Baltimore and Houston are unanimous picks across all five models, making them this week’s Tier I “best bets.” The Tier II picks are Jacksonville (-2.5) versus the L.A. Rams, Oakland (-2.0) versus the L.A. Chargers, and New England (-9.5) at the N.Y. Jets. (The fourth and fifth models like the Saints (-4.0) over the Lions, pushing Detroit down to Tier IV. Revisit the tiers here.)
The N.Y. Giants (+12.0) at Denver would be a Tier I pick and the Titans (-2.5) would be a Tier II pick, but I’m excluding them for now given the spate of Giants’ injuries last week (which the line factors in but the models do not) and the uncertainty regarding Marcus Mariota’s availability Monday night. [Update: Mariota is playing and the line has moved to Titans (-6.5).]
For what it’s worth, the three-model majority (“Machine”) pick is Tennessee up to -4.0, and then flips to Indianapolis at 4.5. And as we discussed last week, watch out for your lines. The three-model majority pick is San Francisco (+11.0) but flips to Washington at 10.5; New England (-9.5) but N.Y. Jets at 10.0; and Baltimore (-6.5) but Chicago at 7.5.
And how’s this for weird? I’ve seen Pick’em Pool lines for the Vikings-Packers game range from Vikings (+8.5) to Vikings (+3.5), the latter of which is closer to actual line of +3.0. (I’m only left to wonder if the +8.5 was a typo or if it was produced prior to the Vikings’ victory on Monday night, though I don’t think their three-point win over the Bears would be worth five points in this week's line at home against Green Bay.)
The three main models agree on three picks this week – Baltimore (-6.5) hosting Chicago, Houston (-9.5) hosting Cleveland, and Detroit (+4.0) at New Orleans. Baltimore and Houston are unanimous picks across all five models, making them this week’s Tier I “best bets.” The Tier II picks are Jacksonville (-2.5) versus the L.A. Rams, Oakland (-2.0) versus the L.A. Chargers, and New England (-9.5) at the N.Y. Jets. (The fourth and fifth models like the Saints (-4.0) over the Lions, pushing Detroit down to Tier IV. Revisit the tiers here.)
The N.Y. Giants (+12.0) at Denver would be a Tier I pick and the Titans (-2.5) would be a Tier II pick, but I’m excluding them for now given the spate of Giants’ injuries last week (which the line factors in but the models do not) and the uncertainty regarding Marcus Mariota’s availability Monday night. [Update: Mariota is playing and the line has moved to Titans (-6.5).]
For what it’s worth, the three-model majority (“Machine”) pick is Tennessee up to -4.0, and then flips to Indianapolis at 4.5. And as we discussed last week, watch out for your lines. The three-model majority pick is San Francisco (+11.0) but flips to Washington at 10.5; New England (-9.5) but N.Y. Jets at 10.0; and Baltimore (-6.5) but Chicago at 7.5.
And how’s this for weird? I’ve seen Pick’em Pool lines for the Vikings-Packers game range from Vikings (+8.5) to Vikings (+3.5), the latter of which is closer to actual line of +3.0. (I’m only left to wonder if the +8.5 was a typo or if it was produced prior to the Vikings’ victory on Monday night, though I don’t think their three-point win over the Bears would be worth five points in this week's line at home against Green Bay.)
In any case, good luck and enjoy the week!
|
The models referenced are created by and sourced from FiveThirtyEight, Jeff Sagarin via USA Today, RP-Excel.com, and NutshellSports.com. Lines are sourced from FootballLocks.com as of Oct 13 at 11am PT. Data was compiled and analyzed by ELDORADO. All charts and graphics herein were created by ELDORADO.
ELDORADO | Berkeley, CA | New York, NY
eldo.co | @eldo_co
ELDORADO | Berkeley, CA | New York, NY
eldo.co | @eldo_co