Results filed under: “supercontest”
Let's just get this ugliness over, shall we?
Last week: 1-4 (20%), Overall: 23-20-2 (53.4%), Fake rank: 225 (out of 745)
To make matters worse, the Giants took a huge hit to their Super Bowl chances by surrendering a 10-point, 4th-quarter lead to the Steelers. They are basically off the board - in 100 simulated runs, they won zero of them. Part of this is because the Bears are basically sucking up all the oxygen in the room - they are winning the NFC in half of my simulated runs, leaving table scraps for Atlanta, San Fransisco, Green Bay, and the Giants to fight over.
On the flip side, a team that has managed to claw back into their division race with a win is San Diego, who spent he last 3 weeks in a freefall after collapsing late against Denver. Their odds of winning the division nearly doubled after their win last week, increasing from 17% to 31%. Still, Denver remains the favorite for the AFC West. That, along with the AFC North, which is basically a coin flip between Baltimore and Pittsburgh right now, are the only divisions that remain meaningfully competitive at this point.
I recovered from my 1-3-1 performance last week with another 3-2 week.
Last week: 3-2 (60%), Overall: 22-16-2 (57.9%), Fake rank: 99 (out of 745)
This week's picks:
Yes, I picked Arizona again, even after that abominable performance on Monday night. I also have a terrible track record picking against Houston, but here I am again with Buffalo +10.5. So...3-2 again? I'm happy with 3-2 again.
Am I alone here in not really believing in the Falcons? Nobody from last week's top 5 lost (Houston was on a bye week), but San Fransisco's path to the NFC West became much clearer with losses by both Seattle and Arizona. New England continues to benefit from a weak division. At this point, the odds are being driven by standings, since the following divisions show a winner with 90%+ confidence at this point:
NFC East (Giants win 98% of 100 simulations run)
NFC South (Atlanta, 96%)
NFC North (Chicago, 93%)
NFC West (San Fransisco, 100%)
AFC South (Houston, 100%)
I left New England (89%) and Denver (82%) off this list. The only truly competitive division at this point appears to be the AFC North, with PIttsburgh (37%) and Baltimore (61%) slugging it out.
I was due for a losing week, I guess. The lines this week were very sharp: 5 games ended within +/- 1 point of the number. I picked 3 of those games for the SuperContest, and went 0-2-1 in those games. And yes, I'm aware these are bitter whiney sour grapes and I don't care.
Last week: 1-3-1 (25%), Overall: 19-14-2 (57.6%), Fake rank: 154 (out of 745)
This week's picks:
A new challenger has emerged!
Making their first appearance on the big board are the Chicago Bears, who ROCKET to the #1 spot. This despite 3 of the teams they jumped ahead of winning, and another having a bye week (and still being 6-0). So what's going on here? Is my model whacked?
Well, that's always one possible answer. To consider other possible answers to the question, I ran two versions of my model this week. In addition to the final one, I ran one before the Monday night game, so I could separate out how the Bears performance in that game impacted their standings, vs. what was happening around them in the league. To facilitate this, here is a detailed chart of Chicago's progress.
Let's start by looking at their week-to-week win projections. Chicago's stock has been rising steadily ever since their week 2 loss to Green Bay. This building process is important. Much of sports analysis seems to take place in a vacuum. To use their week 2 opponent as a particularly illustrative example, one week, the Green Bay Packers are losing at Seattle on a bogus official's call, and the punditry is asking what has "happened" to Aaron Rodgers and the Packers. Then the reigning MVP throws 9 TDs and 0 INTs in his next two games, and the story is, do the Packers have their groove back? Both statements were dumb.
Instead, each piece of new information should build upon itself. Based on the Packers recent track record, a couple of early season wins should not be looked upon as the new normal, but as a bump in the road. Even the Saints 0-4 start may have less to do with Bountygate as it did with fluky luck.
That being said, we are now nearly halfway through the season, and your record at this point matters. The Saints are 4 games behind the Falcons with 9 left to play. Thus, their odds to win the division are only 3%. The Packers are trailing the division leading Bears AND the Vikings, leaving only a 7% slice of the division pie for them.
Getting back to my main point: each new data point should move the needle, but by how much? How much data does it take to convince us that what we are seeing is real?
In the case of the Bears, the answer was "alot". But once the model is convinced, each new datapoint will go a long ways towards convincing it that what it is seeing is real. Last week, the Bears were projected to win 10.6 games on average. After everybody else had played, but before the Monday night game, this number was 12.2. After the Monday night game, this increased to 13.2.
Why did they go from 10.6 to 12.2 before they even played? Let's look at their schedule moving forward: Carolina, Tennessee, Houston, San Fransisco, Minnesota, Seattle, Minnesota again, Green Bay, Arizona, Detroit. Collectively, these teams went 6-3 (excluding Detroit). But that doesn't tell the whole story. Only one team (Houston) looked truly impressive in victory. Everybody else either lost, or won by a touchdown or less. After putting a win in the books and Detroit looking pathetic, they gained another game in their expected outcome. Out of the 2.6 win increase modeled for Chicago, only 1.0 had anything to do with their performance Monday night. The other 1.6 had to do with what the rest of their future opponents did, which, other than the Texans, was be mostly mediocre.
This jump from 10.6 to 13.2 basically explains everything else. They went from being projected as the 4th best team in the league to the best. This gives them, on average, better playoff position, more bye weeks, and more home games, in addition to the model just thinking they are better to begin with.
On the one hand, the model's change in opinion regarding Chicago was very drastic this week. On the other, it took several wins in a row before the needle moved on Chicago at all, meaning it wasn't REALLY that drastic. The initial skepticism it had regarding Chicago meant that it took a few weeks to catch up to what somebody with a stronger starting opinion about Chicago might have already thought.
After 6 weeks of SuperContest picks, making 5 picks a week, I have picked 30 underdogs.This was not a conscious, purposeful decision. I just kept liking underdogs (this week is no different, by the way). Last week was an underdog pickers dream: dogs when 11-2 against the spread. 11-2!!! Unbelievable. So, my record last week must have been 5-0, right? At the very least 4-1?
Nope. 3-2. Only 2 underdogs failed to cover last week, and I picked both of them.
Last week: 3-2 (60%), Overall: 18-11-1 (62.1%), Fake rank: 58 (out of 745)
This week's picks:
This was a topsy-turvy weekend by any account, including mine. Of the 5 teams I forecast as the top 5 super bowl contenders, 4 of them lost. Only Baltimore avoided making it a clean sweep, and even then it was a gift from the Cowboys. So why, you may be wondering, did New England's super bowl odds shoot up, from 25% last week to nearly 35% this week?
Well... BECAUSE it was such a topsy-turvy weekend. Right now, the NFC appears to be the superior conference to the AFC. But it is for precisely this reason that New England is the Super Bowl favorite - they have an easier path there than any of the NFC contenders. There are currently only 2 teams in the AFC with a winning record: Houston (5-1) and Baltimore (5-1). In the NFC, we have New York (4-2), Atlanta (6-0), Chicago (4-1), Minnesota (4-2), Arizona (4-2), Seattle (4-2), and San Francisco (4-2) - that's 7 teams.
However, only one of them can make the Super Bowl. Collectively, their odds of winning it are 39%. But 3 of them are in the same division! Only 6 of them can even make the playoffs, and with 11 weeks to go, there is lots of time for things to change.
Now, let's consider New England. In their 3 wins, they have won by margins of 22, 24, and 10 points. In their 3 losses, they have lost by margins of 2, 1, and 1 point. Yes, a win is a win and a loss is a loss, but looking forward, the Patriots look significantly better than their 3-3 record. Baltimore counts among their 5 wins margins of 1 point (over the Patriots), 3 points, and 2 points. Who do you think is going to be the better team going forward?
In fact, it seems to be that New England and Houston are likely the class of the AFC (I think my forecast model is being too bearish on Houston at this point). Baltimore is a paper tiger with injuries that will start catching up to them, and soon. Probably this Sunday against Houston.
I think I'm doing pretty darn good at this fake Supercontest thing. In fact, after another solid week, I tried to explain to Suzi how good I was doing. In doing so, I made the mistake of noting that you have to finish in the top 20 before indicating my current standing.
Then I said: "And out of 745, I would currently be tied for 50th! That's in the top 7% of all entries!"
Blank stare, followed by: "so you AREN'T in the top 20."
Last week: 3-2 (60%), Overall: 15-9-1 (64.6%), Fake rank: 50 (out of 745)
This week's picks:
New England retains the top spot after taking care of business in Denver. Philly loses to Pittsburgh, but still leapfrogs San Diego after their loss to New Orleans. And Atlanta finally appears on the list after a 5-0 start.
After running this model for a month now, I'm noticing that it takes some pretty wild changes week to week on Super Bowl odds. This is because it is a highly dynamic model that is sensitive to changes in the initial conditions.
The way each model run starts is by simulating each game of the season, and predicting winners and losers. It then seeds everybody based on NFL playoff tiebreaker rules and plays out the playoff brackets to determine a Super Bowl winner. So not only does this model tell me Super Bowl winner odds, but season win totals, division winners, and conference winners as well. Let's look at how our front runner, New England, has changed over the course of the season in each of these categories.
Starting with our win predictions, we see that in the preseason the model thought New England would win an average of 13.7 games this year. After 5 games and a 3-2 start, this has dropped to 11.8. However, their division odds have actually INCREASED from the preseason. They started at 91%, and have gone up to 95%. Why have they gone up if their expected win total has dropped?
This is where the importance of a dynamic model comes in. Division odds are not just a function of New England's performance, but the performance of the three other teams in the division as well. Pre-season, I was very bullish on Miami: the model predicted 10.7 wins for them. Now, they still figure to finish in 2nd place in the division, but their expected win total is down to 7.6. This drop by their nearest competitor has allowed New England to stay strong in the division.
This dynamic plays out on a larger scale at the conference and Super Bowl level. Relatively small changes in the initial conditions week to week play out as significant swings at the Super Bowl level.
The second part of judging a forecast is to look back at how it has done. At the end of the season, I will have made 18 predictions for each team (1 per week plus 1 preseason) that can be tested. However, because there are many more results than predictions, it may take a few years before I can have any confidence that this method is accurate predicting outcomes - and more importantly, how that level of confidence changes as we get closer and closer to the end of the season.
First things first: I'm actually doing pretty good with these fake LVH contest picks so far.
Last week: 4-1, overall: 12-7-1, Fake rank: 67 (out of 744), top 9%
This week's card:
Here is the updated Super Bowl odds graph, including some new faces to the top 5. However, NE and SD have firmly planted themselves as the favorites. I join the rest of the world by welcoming the Houston Texans into my top 5. My computer finally thinks Arizona is for real (probably just in time for them to crash back to earth). And Philly is one of 5 teams that has at least a 75% chance of winning their division after skating by the Giants on Sunday night (the other 4: SD, HOU, NE, and ATL).
Now that a quarter of the season is done, I'm going to evaluate some of my preseason over-under win total bets. As a reminder, here are those bets.
Atlanta OVER 9 wins
KC UNDER 9 wins
SF OVER 10 wins
Indy UNDER 5.5 wins
NYG UNDER 9.5 wins
STL UNDER 6 wins
Miami OVER 7.5 wins
NO OVER 10 wins
TB UNDER 6 wins
Denver UNDER 9.5 wins
TEAR THAT TICKET UP
NO over 10 wins: for some reason (namely that it is retarded), my computer still has the Saints winning an average of 8.4 games this year. However, out of 100 simulations, they hit 11 or more wins only 8 times (they hit 10 for a push another 20 times). That means that my computer, which LOVED the saints, says I lose this 72% of the time. In other words: it is over.
SPEND THAT MONEY NOW
ATL over 9 wins: my computer has them at an average of 11.5 victories. They clear the 9 win hurdle 90% of the time, and push another 8%.
KC under 9 wins: this may be the easiest money I ever make gambling. KC fails to get to 9 wins 91% of the time, and pushes another 7%.
TB under 6 wins: or maybe this is the easiest money I ever make? TB comes short of 6 wins 95% of the time, and pushes another 5%. Or, in other words: out of 100 simulations, TB cleared the 6 win hurdle zero times.
TOO SOON TO TELL
DEN, at 2-2, projects for 6.4 wins, but the model may still be underestimating Manning's influence. IND has only played 3 games, so the 6.4 win projection is not as strong as some of the others. NYG, SF, and MIA are all currently project to be within a half-game of their totals (although MIA, frustratingly, is a couple of missed field goals away from 3-1 right now, and probably won't make it, if I am being honest with myself).
This week's fake SuperContest picks:
Last week: 3-2, Overall: 8-6-1, Fake rank: 120 (out of 745), top 17%
First, an updated Super Bowl odds graph.
For starters, Philly and New Orleans have fallen out of the top 5, to be replaced by Baltimore and San Diego. Second, this is a good opportunity to revisit some pre-season bets..
Before the season, I proposed some futures bets for super bowl, conference, and division winners. As the season progresses, books will update some of these lines. This gives us the opportunity to revisit these lines and perhaps take advantage of some additional value.
In our original strategy, it was not enough for a line by itself to have positive outcome value, because these bets are all mutually exclusive - since only one team can win the super bowl, a winning bet on Pittsburgh also implies losing bets on San Francisco, New Orleans, et al. (Also implying a losing bet on New Orleans? Betting on New Orleans, apparently.) That is still true. This means that any new bets we make must meet that same standard - a winning bet must have a positive expected value even after all other bets are losers. With that said, here's some additional super bowl plays after Week 3.
Original bets: PIT 14/1 (4.06 BU), MIA 75/1 (0.54 BU), SF 9/1 (1.64 BU), NO 18/1 (1.12 BU), 7.36 BU total.
San Diego Chargers (25/1): 2.7 BU
Pittsburgh Steelers (20/1): 2.8 BU
Seattle Seahawks (30/1): 1.8 BU
Before we made these bets, a total of 7.36 BU had been wagered on the Super Bowl. This brings the total to 14.66 BU. The consequences of these additions is that some of our pre-season bets have become less valuable. Specifically, if SF and NO were to win, they are still positive, but just barely. The threshold for adding additional teams from this point forward will be that much higher as a result, i.e. it will need to make up for the fact that some pre-season bets become losers even if they win. You will also notice that we are putting additional money on Pittsburgh at 20/1. This is because additional Pittsburgh money does not cancel out our original Pittsburgh bet, so its easier to have a positive outcome. The second reason is that the odds have gotten longer, so even with the Steelers at 1-2 right now this looks like a chance to grab some more value.
We can run the same exercise on our conference wagers (new division lines have not been posted).
Original conference bets: SF 9/2 (3.74 BU), NO 9/1 (2.04 BU), MIA 30/1 (1.32 BU), PIT 6/1 (6.56 BU)
Additional conference bets:
SEA (14/1): 4.1 BU
SD (10/1): 3.6 BU
PIT (9/1): 3.8 BU
CAR (40/1): 0.7 BU
This will be the last time I add any NFC Conference teams, because I have essentially soaked all the value from my original SF bet - any more money on the NFC and it becomes negative even if it wins. There is still some headroom in the AFC for additional plays if they become attractive.
Here is who I am on for this week's fake SuperContest entry:
In my last post, I mentioned Nate Silver, the statistical whiz behind FiveThirtyEight. In an election season, one of the features of FiveThirtyEight is its ongoing predictions. Nate not only calculates his predicted odds at any given moment, but keeps a historical record of where the race stood earlier in the year so that you can visually see how the race has progressed.
I plan to do something similar. I had hoped to unveil this after Week 1, but my ongoing predictive model had a bug that I've only now sorted out. In this space I will be keeping track of changes in projected Super Bowl odds, division odds, and win totals.
TOP 5 SUPER BOWL CONTENDERS:
#1: NEW ENGLAND (last week: 1, preaseason: 2)
#2: SAN DIEGO (last week: <5, preseason: <5)
#3: PIT (last week: 3, preseason: 1)
#4: NO (last week: 2, preseason: 3)
#5: PHI (last week: 5, preseason: <5)
falling off the list: SEA (was 4, <5 preseason), GB (was preseason 4), SF (was preseason 5)
Last week: 3-1-1 (7 pts)
Overall: 5-3-1 (11 pts)
Lacking the guts and necessary bankroll to actually participate in the $1000 buy-in LVH SuperContest, aka the World Series of Sports Betting, all I can do is play along at home. Here are this weeks picks.
AZ +13.5 over NE
CLE +7 over CIN
SEA +3 over DAL
MIA +2.5 over OAK
JAX +7 over HOU
Nate Silver, the statistical genius behind FiveThirtyEight, has a new book coming out called The Signal and the Noise. In an excerpt posted in the New York Times over the last weekend, he talks about how computer modeling has not necessarily made us better forecasters of the future, with one notable exception: weather forecasters. That is because weather forecasters have mastered the art of knowing when to listen to their computer models and understanding their shortcomings.
This is an art that I have yet to master with my sports prediction algorithms (or, as I call it, Gamblor). Gamblor is capable of producing some great insights, and I am confident that I will continue to find success with Gamblor moving forward. However, if I am to truly become great at this sports prediction business, I must begin intrinsically recognizing the shortcomings of my program.
All this is to say that I should have thrown myself in front of my MIA and PIT selections last week. Sometimes teams undergo too much change over an off-season for the computer to properly understand, and in those cases, sometimes discretion is the better part of valor.
And yet here is MIA back in my Top 5 picks. I am tempted to throw it out and replace it with my next selection, CHI +6 over GB. But unlike last week MIA isn't playing a good team. And they are at home. Let it ride, Gamblor.
Last week: 2-3
Overall: 2-3 (4 pts)
I have a dream. Specifically, a gambling dream. In this dream, like in many dreams, it starts with flying. I am flying through the sky, borne westward towards a gleaming city of lights. And hookers. Lights and hookers. I descend, the glittering lights (and hookers) stretching towards me and past me, and now I am among them, gliding across the landscape until I reach the palace at its heart. Once there, I am suddenly thrust into an arena, thrown into combat against the best of the best, until I am the last man standing, crowned champion of the Las Vegas Hotel Supercontest.
Put on every year by the Las Vegas Hotel, the Supercontest is becoming the World Series of sports betting. With an annual purse of nearly $1M, a $1500 entry fee gets you a chance to pit your NFL picking skills against the very best. The only problem? The $1500 entry fee. Until such time as my successful gambling renders me capable of tossing down $1500 on what is essentially a game, the best I can do is play along at home.
Each week, the hotel posts the card with the lines to be used for that game. Contest entrants must select 5 plays among those games to submit. A win gets you 2 points, a push 1 point, and a loss nothing. I will be picking along this year and keeping track of my virtual standing against the pros. Here are my Week 1 picks.
MIA +13 over HOU
By now, my foolish infatuation with this Miami team is well documented. I've picked them OVER wins, I picked them as part of my Super Bowl plays, my Conference plays, and my Division winner plays; and now I'm betting on them week 1.
IND +9.5 over CHI
Apparently I have a thing for picking rookie quarterbacks.
PIT +1.5 over DEN
This one at least makes sense. Jump on the betting against Peyton bandwagon now while there is still room. This is going to be a full ride by Week 4. That is, if Peyton is still playing in Week 4 and hasn't been sidelined with further neck trauma.
JAX +3.5 over MIN
Yeah, they are both bad, but Minnesota is worse than people realize, especially without a healthy AP.
CLE +8.5 over PHI
Did we learn nothing from overrating PHI last year? I sure hope not...
Also receiving consideration: SF +5 over GB, DAL +4 over NYG, CAR -2.5 over TB.