clock menu more-arrow no yes

Filed under:

Predicting the Pac-12: a Battle of Vegas, Computers, Homers and Haters

New, 13 comments

In which we look at who is best at choosing winners and losers, and winning and losing money. As a warning, there are a lot of tables… But the outcomes are interesting. And turns out the wisdom of the crowds is sometimes wrong.

NCAA Football: California at Oregon – Cal coach Justin Wilcox, football coaching staff and several players on the sideline. Troy Wayrynen-USA TODAY Sports

An Evolving Blog – so What's it Say?

Well, time to give this little conference-stats blog another try… The last one had way too much info, and then I ran into some health stuff that kept me away from writing for a while. We’ll see how the blog can keep up going forward – fingers crossed.

This little piece is going to address things from a bit of a different angle: the wisdom of Vegas, compared to the wisdom of computers (namely SP+ and FPI). Finally, it'll explore the old adage that if you take some money to Vegas, "the house always wins" – and it turns out, if you go to Vegas with even the wisest stable strategy, that adage is almost always right.

The post will go in a few parts:

  1. Reviewing the games and margins-of-victory for all Pac 12 teams through week 7.
  2. Reviewing the Vegas projections right before kickoff (the projections are based on “VI Consensus” picks on Vegas Insider). This includes point spreads and “moneylines”.
  3. Reviewing the projections of a couple computer algorithms: Bill Connelly’s SP+ (which pops up on his twitter feed here) and ESPN’s Football Power Index (FPI). Both computers have win chances (in percent), while SP+ also adds a projected margin-of-victory (or loss). Both computers also provided preseason projections for each game (at least for percentages) and continue to update their projections week-by-week. This week, I'll only cover the computers' weekly projections.
  4. Asking: which has been most accurate so far, in terms of wins and losses?
  5. Asking: if you were a betting person, would any of a few proposed strategies actually make money by the end of the season, and which would make (or lose) the most?
  6. Asking: have any teams challenged conventional wisdom enough that being a homer or hater would be a good strategy?

Before you go too far down, I want to stress this is a little mental exercise and not an endorsement of gambling. One thing that really stands out to me, at the halfway point in the season, is that the only strategies that would even make money would do so at a pretty mediocre pace (i.e. someone would need to gamble a lot of money to get not much back) and the strongest of 4 winning strategies will probably move closer to even by seasons' end, now that we are in conference play. Meanwhile, using any strategy that would put someone in the red could lose a LOT of cash. Seems like a few popular phrases are correct: don’t gamble your life savings, the house always wins, and you need [a lot of] money to make money.

Game Results So Far

Let’s take a look: what have the overall records for each team been, and what were the per-game and season-long point differences?

A table of game outcomes for every team in the Pac-12 conference, with their overall records and records in conference play. There is also a figure at the bottom showing that in "cross-division" games, the North has won 4 contests while the South won 6.
The results for each Pac-12 game through Week 7

Through Week 7, three teams (UO, ASU & Utah) have records of 5-1, while one more 5-win team (UW) joins at 5-2; only Cal and UA have the other winning records at 4-2 apiece; four teams (Furd, WSU, Colorado & USC) sit at 0.50 with 3-3 records; OSU is 2-4 (better than some expected they’d be); and UCLA is miserable at 1-5. Conference records are pretty spread out: only Oregon is undefeated in-conference; half the conference (6 teams) have 2 wins, with Furd and UW being 2-2 and UA, ASU, USC and Utah being 2-1; four teams (Cal, OSU, Colorado & UCLA) are 1-2, and lowly WSU is Coug’n It at 0-3. Of the 10 cross-division games so far, the South actually has a winning record over the North, holding 6 wins to the North’s 4.

A table showing the margin-of-victory, or margin-of-loss, in points for each game of all Pac-12 teams, plus the total point margin all-season and for conference games.
Margin-of-Victory (or loss) for Pac-12 games through Week 7

As far as margins-of-victory and margins-of-loss, 2/3 of the conference (8 teams) are in the green for the whole season-to-date, while 4 teams are in the red. UCLA holds the conference honor of largest season-long deficit and largest deficit per game (69 and 11.5, respectively); of the other losers, OSU is least in the red (12 overall, 2 per game), as its large-ish margins over Cal Poly and UCLA, plus close losses to Hawaii and Furd, made up for a 2-score loss to Oklahoma State and 45-point demolish and by Utah this week. Cal is the OSU of the positive margins, closest to even at 12 points overall and 2 PPG, while every game has been within 2 scores; the top end is populated by the expected conference leaders of Oregon (164 overall/27.3 PPG), Utah (130 overall/21.7 PPG), and Washington (118 overall/16.9 PPG).

For the in-conference games so far, a majority (7 teams) are in the red while only 5 are in the green. Colorado is worst at negative 44 overall/14.7 PPG, followed by WSU (11 PPG), OSU (10.3 PPG) and Furd (9 PG); and then 3 teams are tied for least-bad (Cal/Arizona/UCLA) with negative 16 overall/5.3 PPG. Oregon leads the positive point-fest, being ahead 67 points in-conference/22.3 PPG, and is followed by South division leader Utah (63 in-conference/21 PPG); there’s a big drop to Washington, the next team in-the-green (27 in-conference/6.8 PPG), then USC (18 in-conference/6 PPG), and finally ASU (8 in-conference/2.7 PPG).

Projecting Outcomes

Plenty of folks have tried to come up with ways to project the winners and losers in sports. Some of this is out of pure entertainment, some is fodder for sports journalists and entertainers, and some gamblers are simply looking for ways to make some more moolah. The question remains, though: who’s best at projecting games’ outcomes? This little section shows that computers seem better at predicting winners/losers than Vegas does, but not by much. (Of course, taken with the grain of salt that the accurate-to-inaccurate ratio is a bit boosted considering we are just past halfway and the stats from early-season cupcakes were gimmes for everyone).

Vegas: Wisdom-of-the-Crowds

Vegas has interesting ways of demonstrating how much a degenerate better will earn or lose based on a game’s outcome. There are three main “betting lines”: the point spread, moneyline, and over/under. On any bet, the “payback” is denoted in either positive or negative numbers, in this way:

  • In general: if someone puts down money, and they win, they get the original bet back, and then some. If they put money down and lose, then “goodbye money.” The exact amount a winner gets back is based on whether something is listed as a positive number or negative number; numbers start at 100 (either +100 or -100) and go from there (positive to +1,000,000 onward, or negative to -1,000,000 onward).
  • For positive numbers, someone gets back more than they bet. So if Better A goes for something listed at +200 and put down $1, then wins, B.A. gets back their original $1 plus $2 (if they lose, goodbye $1).
  • For negative numbers, bettors get back less than they bet. So if someone goes for something listed at -200 and they put down $1, then win, they get back the original $1 plus $0.50 (or, another way to look at it, they’d need to put down $2 to make $1). (If they lose, goodbye bet)

The betting lines go this way:

  • “Point Spreads” are set by Vegas where they think half of bettors will bet that one team will win by more than a certain amount, or that team will either a) win by less than that amount or b) lose outright. These are usually set at -110 on either side of the spread – so if Vegas/casinos are right that half of people bet on one side while half bet on the other, Vegas makes about $0.09 per dollar bet, no matter what happens (“the house always wins”). If way more people are putting money down on one side of the spread, Vegas/casinos adjust the spread accordingly (that’s why the point spread on game day is often as much as 10 points more or less than the “opening spreads” set about 6 days before).
  • The “over/under” is a bet of how many points will be scored by both teams in the whole game – whether or not the favorite wins. It’s usually set around -110 on each side, similar to the point spread. (I didn’t track this, but perhaps in the future…)
  • The “Moneyline” is a simple bet that the team will win outright. In games that are anticipated to be blowouts, the favorite often has money lines as ridiculous as -20000 (like WSU had over New Mexico State in Week 1), while the underdog can have money like +10000. In conference play so far, the largest Moneyline discrepancy was actually Cal at Oregon (Cal +1100, Oregon -2500).

Here’s what Vegas has put down so far. I have 3 tables: the point spread in favor of one team, the Moneyline in favor of that team, and the Moneyline against the team. In the point spread, a positive number shows the team is a favorite while negative is an underdog; in Vegas, it’s flipped the other way… But details set aside, for now.

A few games had point spreads but didn’t even have Moneylines because they were predicted to be such huge blowouts: Oregon v. Montana, WSU v. Northern Colorado, ASU v. Sacramento State, and Utah v. Idaho State. (All the numbers I listed below are for the betting lines at kickoff, per “VI Consensus”)

A table showing the "betting lines" for each game for every Pac-12 team, from Week 1 through Week 7.
Betting point spread at kickoff, per "VI Consensus"

Let’s take a look at the point spreads, set by Vegas, for each game and through the whole season. 6 teams (50%) have assembled positive overall point spreads, while 6 are in the red. Washington and Utah are the only teams that have been favored in every game so far, while Oregon was favored in 5 of their 6 games (note: Vegas was right on Oregon) and had the largest per-game point margin (with an average of 19.4 PPG). The Bears were only favored in 3 games and are ever-so-slightly in the red (-0.7 PPG), and the Baby Bears (UCLA) hold the largest negative margin at -5.8 PPG.

A table showing the money line payback for each game for every Pac 12 conference team, from the first week of the season through Week 7.
Moneyline payback per “VI Consensus" at kickoff

Here’s the table for the Vegas Moneyline looking at each individual team – in terms of them winning. Remember, positive numbers mean the team is an underdog, but also that someone betting would make more money than they bet, should the team win. I’ve listed teams as red if they are underdogs – rather than green because it would be a positive payout.

Something interesting to me is that Cal actually wasn’t hugely favored over Davis, and looking back at it, the VI Consensus Moneyline seemed different than some of the official casinos’ numbers. Just how things go, sometimes. Also Bears-oriented, the major-underdog status at the Washington game would have been a smart bet. Other standout deals are Oregon’s huge negative Moneyline on pretty much every game (aside from the season kickoff against Auburn), and also the hilarious level of upset that was UCLA at WSU (a 7-to-1 payout in favor of UCLA).

A table of numbers for every conference team's game from the first week through Week 7 of the 2019 season.
Moneyline payback per VI Consensus at kickoff – for opponent

Here is the same table as last one, except the Moneyline in favor of the other team. Feel free to browse…

Computers’ Weekly Predictions

We’ve already seen Vegas’s projections of wins, losses and point spreads. What did the computer overlords think? Both FPI and SP+ publish weekly projections about the likelihood each team will win; SP+ goes one step further and includes numbers such as the expected point spread and over/under. The following few tables show both computers’ projections around likelihood of a given win (or loss) and also SP+’s projections for the point margin (but not the over/under or a few other figures). I won’t really explain anything in-depth, but feel free to browse…

A table showing the FPI's projected chance for each Pac 12 team to win each of their games, from the first week through Week 8 of the 2019 season.
FPI Projections – winning likelihood (week-of-game)
A table showing the projected winning likelihood for each team, over the course of the season so far (weeks 1 through 7).
SP+ Projections – winning likelihood (week-of-game)
Each Pac 12 team's projected point spread over the course of the 2019 season, either as a positive number (win) or a negative number (loss).
SP+ Projections – point spread (week-of-game)

Who Is Right More Often?

Okay, here is where I ask: which of the trio – Vegas, SP+ and FPI – is more accurate in projecting the winners and losers for each game? It’s a pretty simple question – so here goes:

A table showing whether the Las Vegas betting lines were correct or incorrect for each game in the Pac 12 season so far. Overall, the Vegas betting lines were right for 37 games and wrong for 17.
Accuracy of Vegas projections – Pac 12 winners and losers (2019 through Week 7)

Looks like Vegas correctly predicted 37 out of the season’s 54 games so far. Some teams were more accurate than others: Vegas was right for all of Oregon’s games and 5 games each for OSU (5-1), Washington (5-2) and Utah (5-1). 2 teams (WSU & USC) are 4-2; 3 teams (Cal, Arizona & UCLA) batted 0.5 with 3-3 records; and Vegas doesn’t understand the duo of ASU & Colorado, with both at 2-4.

A table showing the accuracy of SP+ in projecting the winner and loser of each Pac 12 game so far this season. The SP+ algorithm got 40 games right, and 14 games wrong.
Accuracy of SP+ projections – Pac 12 winners and losers (2019 through Week 7)

SP+ has a slightly better record than Vegas, at 40-14. Not a single team has an “incorrect” record, while only UCLA sits at 3-3; the others are all 4-2 or better, with Oregon the only perfect outcome.

A table showing the accuracy of FPI's projections in the first 7 weeks of 2019. So far, FPI has an overall record of 39-15.
Accuracy of FPI projections – Pac 12 winners and losers (2019 through Week 7)

FPI is still better than Vegas but just barely behind SP+, with a 39-15 record. UCLA has a failing grade at 2-4, while four teams are 3-3 and the rest are in the green. FPI was the only formula to guess all of Utah’s games correctly, as it gave USC the nod in Week 4.

Bill Connelly has been tweeting about the accuracy of his formula against-the-spread, compared to Vegas. Here's the table of those outcomes:

A table showing the accuracy of SP+ in going against-the-spread compared to Las Vegas wagers. 30 games were correct and 24 were incorrect.
Accuracy of SP+ against-the-spread – Pac 12 games (2019 through Week 7)

So far, in the Pac-12, SP+ has gotten 30 of the 54 games correctly when looking at which side of the point spread to wager (a 55.6% clip). They got the spread right for the Bears half the time, while OSU and UCLA (my bottom-of-the-conference Power Rankings choices) were 5-1 against the spread. Interestingly, the projections misread the conference leaders more often than not (accuracy was 2-4 for Oregon and Utah).

Where's the [Imaginary] Money at?

I threw together a few imaginary betting strategies, and then bet one imaginary dollar per game per team. This shows the outcome per team, per week, and over the course of the season. The strategies are:

  • “Homer” bets, both for the Moneyline and against the spread. Basically, you always bet in favor of your team – the sign of a true overly-optimistic fan.
  • “Hater” bets, both for Moneyline and against the spread. Always bet against your team – a sign of a true, cynical Cal fan.
  • Betting the “Moneyline” in 3 ways: for the teams that are favored by Vegas, for the teams favored by SP+, and the teams favored by FPI.
  • Betting for (or against) the spread using SP+’s formula.

Here are the tables – It shows the outcome by betting $1 per game for each of those strategies: any point spreads use a -110 payback, so a win on a $1 bet would yield about $0.91. I won’t go into detail for each one, but will wrap things up at the end.

Homers

A table showing the payback for betting in favor of one specific team, using Moneyline bets, in the 2019 season through Week 7. 7 teams would make money, and the strategy would have made money over the course of the season so far.
Moneyline bets as a "homer" – 2019 Pac 12 games through Week 7

Something important to point out with outright Moneyline: a team that’s massively favored in each game (such as Oregon) is actually harder to get a positive return by being a Homer, because any upset would probably wipe away the multiple small wagers. Meanwhile, betting as an underdog and praying for one or 2 upsets isn’t such an awful idea. Cal and ASU homers are actually the smartest folks so far, largely because of huge upsets over Washington and Michigan State. Go Bears!

A table showing the outcome of betting the point spread in favor of any given team in the conference, for the 2019 season through Week 7.
Point spread bets as a "homer" – 2019 Pac 12 games through Week 7

Being a Homer, i.e. betting that a favorite team beats the point spread, would have so far only worked for 2 teams: OSU and Washington. Those were the only 2 teams to beat the point spread more often than not, with OSU a 4-2 record and Washington 5-2. Six teams beat were 50-50, going 3-3 against the spread – which would have lost money in total, at 4.5 cents/game. UCLA and WSU – which are only 1-5 against the spread – would have lost $4.09 over 6 games, or $0.68/game.

Haters

A table showing the outcome of betting against a team for every game in the season. 3 teams – Washington, WSU and UCLA – would make money while the 9 others would lose money. Betting on every game would make money, but not much.
Money line bets as a "hater" – 2019 Pac 12 games through Week 7

Three-quarters of conference “hater” fans would have lost money betting the Moneyline, with a maximum drop of Oregon (because they keep on winning) at $3.39 total ($0.68 per game). Washington ($0.50/game), UCLA ($0.79/game), and WSU ($1.31/game) are the only schools where hating your alma mater could get you rich: UCLA because they keep losing games, UW because the Cal and Furd losses were such upsets, and WSU because the UCLA loss was an otherworldly upset (+700)

A table showing the amount of money made or lost by betting on each team's opponent to beat the point spread, for all 2019 Pac 12 games through Week 7. This strategy would make a small amount of money, but hardly anything.
Point spread bets as a "hater" – 2019 Pac 12 games through Week 7

One-third of conference teams would have made grumpy alumni going against the spread a little less grumpy, led by WSU and UCLA ($0.59/game) and the duo of Arizona and Furd ($0.27/game). On the other hand, while one-half (6 teams) of grumpy alumni would lose a few pennies ($0.05/game), and both UW and OSU’s haters would lose a good chunk of change ($0.45 and $0.36/game, respectively).

Following Vegas or Computers, for Moneyline and Point Spread

Moneylines: Vegas, SP+ and FPI Favorites

Here are a few little mental experiments. Vegas clearly has its favorites per game, and both FPI and SP+ publish their favorites by Wednesday/Thursday. So – how much money would someone make betting the favorite per Vegas (the “wisdom of the crowds”), FPI and SP+?

A table showing the payback for betting the Vegas favorite on "money line" bets. Only 4 teams are correct, and this strategy would lose $0.17 per game.
Payback from going Moneyline on Vegas favorites
A table showing the outcome of placing "money line" bets on the favorite for each game, per FPI's weekly projections. Half the teams would be in the green and half would be in the red – betting every game would lose $0.06 per game.
Payback for going Moneyline on FPI favorites
A table showing the payback for betting the Money line on the favorite team in each game, using the SP+ computer formula. 7 out of 5 teams would make money, but betting every game in the conference would lose one cent on the dollar.
Payback for going Money line on SP+ favorites

Long story short – all of these strategies, bet across the entire conference, would leave someone in the red. Imagining that someone put down one dollar per game: Betting the favorite per Vegas would lose $0.17 per game, betting FPI would lose $0.06/game, and betting based on SP+ would lose $0.01/game. It’s clear that computers are better than the wisdom-of-the-crowds – but either way, all 3 end up in the red, so not exactly a great strategy for investing toward a new house…

The assessment of individual teams also shows that Vegas is the least-accurate, followed by FPI and SP+. 8 teams would be in the red using Vegas (ranging from UW’s $0.11/game all the way to ASU’s $0.79/game), with the 4 money-makers being more modest, from Utah’s $0.02/game to OSU & Furd’s $0.19/game. FPI was dead-even with 6 teams in the green and 6 in the red, but ending up with a negative balance sheet. SP+ shows itself as the wisest of the formulas, getting 7 teams right (and 5 wrong), with the money gained or lost being closer to even, compared to other formulas.

Going Against-the-Wisdom (for Vegas and Computers)

Okay, so throwing money on the Moneyline for every team in the conference would fail going with the favorites of Vegas and both computers. What about using that trio of methods to "bet the underdog" – almost like the difference between homers and haters. Humorously enough, that would darn-near invert the losses into winnings, and in some cases give a little boost. So whereas betting SP+’s favorite would lose one cent per game, betting SP+’s underdog would make $0.03 per game; FPI would go from -$0.06 up to $0.08 in the green; and Vegas would go from $-0.17 up to positive $0.19. Seems like the strength of the upsets – e.g. Cal/UW, AZ/Hawaii, and the 3 UCLA upsets (losing v SDSU and OSU, and beating WSU) – is enough to cancel out the favorites' wins.

A table showing winnings and losses by putting down one dollar for SP+ underdogs, on the "money line" bet. It would get back an average of $0.03 per dollar, across every game in the conference.
Betting the money line for SP+’s underdog
A table showing the return or loss by putting down one dollar on FPI's underdog, for each game in the Pac 12 so far this season (2019 weeks 1-7). This would earn $0.08 per dollar bet.
Betting the money line for FPI's underdog
A table showing earnings or losses per game, betting the underdog per Vegas on the "money line" bet. This would make $0.19 per dollar bet.
Betting the money line for the Vegas underdog

The Final Strategy: SP+ Against the Spread

Bill Connelly is quite proud on twitter of the accuracy of SP+ against the spread, especially as he continues to revise the formula. As I mentioned above, SP+ against the spread got 30 of 54 games correctly (a 55% rate) – but because “the house always wins”, setting point spreads at -110 payback, that’s still pretty close in terms of actual moola coming back. The level is so much that a hypothetical “payback” would make a 0.67% return – or less than one cent on the dollar.

A table showing how much would be made, or lost, using SP+ to bet against the Vegas spread. It would only yield less than one cent per dollar bet.
Betting SP+ against-the-spread

Well, thanks for following along. I’m not encouraging or condoning any sort of betting, but it’s interesting to see how accurate each of these formulas are as they continue to compete for the crown of best prognosticator. As some of the others on the blog have noted, it’s looking like SP+ is most on-point in its predictions overall. (Which is sad to me, because post-week-7, SP+ has the Bears ranked lower than FPI does). Throwing down money using the "Moneyline" against the "wisdom of the crowds" that is the Vegas Casino, hilariously has the best ROI. We’ll see how accurate everything continues to be going forward…