Writers of Pro Football Prospectus 2008

Most Recent FO Features


» Futures: Nick Chubb & Sony Michel

The Georgia Bullddogs' dynamic duo should be on NFL rosters at some point in the next 72 hours. Which will be the better pro? That depends on what kind of running back you're looking for.

14 Oct 2005

The Evolving Predictive Powers of DVOA

Guest column by Sean McCormick

Power rankings have become a ubiquitous feature at football websites both major and minor. By and large, they are a sloppy, subjective affair, and most of them simply add to the echo chamber rather than providing any real insight as to how teams are playing. Football Outsiders' DVOA rankings are different. They sort out which teams are playing well even when they are losing and which teams are winning in ways that are ultimately not sustainable.

(DVOA stands for Defense-adjusted Value Over Average, and is explained here.)

Football Outsiders has always stressed that the metrics on this website serve to provide a clearer understanding of how teams have played in the past, and what that means for them in the longer-term future. DVOA is not designed to pick a winner in each specific game, because it doesn't take into account things like injuries, home field, motivation, or specific matchups. But stands to reason that teams that are executing more efficiently are going to beat teams that are executing less efficiently more often than not. You don't need a terribly advanced metric to tell you that the Eagles are likely going to beat the 49ers. But most matchups are not nearly so imbalanced, and it's in those games that you would expect DVOA to help sort the wheat from the chaff.

The easiest way to determine the consensus opinion on a given game is to look at the weekly betting line. This is what the lines looked like for Week 3 of the 2004 season, along with the DVOA rankings after just two weeks of input:

Line vs. DVOA Ratings, Week 3 2004
Favorite DVOA Line Underdog DVOA Result
MIA* 29 1 PIT 14 PIT 14 MIA 3
TEN 8 6 JAC 26 JAC 15 TEN 12
NYG 9 3 CLE 24 NYG 27 CLE 10
BAL 19 2.5 CIN 22 BAL 23 CIN 9
KC 12 8.5 HOU 31 HOU 24 KC 21
STL* 30 7 NO 23 NO 28 STL 25
MIN* 27 9 CHI 18 MIN 27 CHI 22
PHI 1 4.5 DET 4 PHI 30 DET 13
ATL 3 10 ARI 32 ATL 6 ARI 3
DEN 10 10 SD 21 DEN 23 SD 13
IND 11 6 GB 20 IND 45 GB 31
SEA 6 11 SF 15 SEA 37 SF 0
OAK 17 3 TB 28 OAK 30 TB 20
WAS* 25 2.5 DAL 7 DAL 21 WAS 18

In ten of the fourteen games, DVOA agreed that the correct team is favored. In several of the games with big lines, it noted a significant discrepancy in the quality of the two teams. Atlanta had played like the third best team in the league during the first two weeks, while Arizona had been the very worst. Denver was ranked 10th overall while the Chargers, fresh off a beating from the Jets, were no better than 21st. Again, there was generally agreement between how the teams were playing and how most people thought they were playing.

But there were four games where the line and DVOA were at odds: Miami-Pittsburgh, Minnesota-Chicago, St. Louis-New Orleans and Washington-Dallas. In each of those games, the team with the higher DVOA rating was getting points, which is to say that in each of those four games public perception was at odds with how the respective teams had really been playing up to that point in the season. We call these games "DVOA Best Bets." And lo and behold, in all four instances, the team that DVOA preferred outperformed the line, and in three of the instances those teams won the game outright.

Was this a fluke or was DVOA picking up on hidden factors that were simply escaping public attention? Could DVOA consistently detect matchups where the favorites were ripe for an upset? As it turns out, the answer depends on which season you happen to be looking at.

Here is a table chronicling the DVOA forecasting record in 2003, the first season that FootballOutsiders.com was online. The predictions begin in Week 3, the first week for which DVOA information was available, and end in Week 16, thereby removing many of the games involving teams that were resting starters for the playoffs and causing a distortion in the results.

Success of DVOA Picking Games, 2003
Week Win-Loss Cover Best Bets
16 11-5 10-6 2-0
15 11-5 7-9 0-3
14 10-6 10-6 3-4
13 7-9 7-9 1-1
12 12-4 7-9 0-2
11 9-7 6-10 1-2
10 5-9 5-9 0-3
9 7-7 7-7 2-1
8 6-8 5-9 2-5
7 8-6 7-7 1-0
6 10-4 8-6 2-3
5 8-6 6-8 2-1
4 8-5 6-7 1-1
3 10-4 9-6 0-2
Total 122-85 100-107 17-28
% .589 .483 .377

In 2003, DVOA did not distinguish itself predicting games on a week in and week out basis. It managed to pick the winners of games at a roughly 59% clip. Once point spreads were included, DVOA's performance dropped to under 50 percent. The theory that DVOA would be able to detect faulty point spreads was obliterated, as the underdogs failed to even cover the spread 63% of the time.

So what happened? The biggest problem was DVOA's tendency to consistently overestimate a select number of teams and to underestimate several others. Here is a list of teams that were singled out as being a live underdog or a weak favorite more than once:

DVOA's Best Bets, 2003
Favorite Times Underdog Times
NYJ 5 SF 4
GB 4 OAK 4
NO 3 JAC 2
MIA 3 TB 2
CAR 3 SD 2
DET 2 KC 2
NE 2

Tampa Bay and San Francisco consistently underperformed every statistical measurement in 2003, thanks to their penchant for mixing dominating wins against good competition with lots of close losses. Both teams underperformed their Pythagorean Wins by two games, and they certainly played havoc with the weekly projections. Oakland started off playing reasonably well and losing before quitting on Bill Callahan and going into a tailspin, and those early performances lingered long enough to present a false impression regarding their week-to-week performance. DVOA picked up on the rise of the Cowboys early on, but overvalued them in late matchups against the likes of New England, Miami and Philadelphia.

More importantly, DVOA stuck Philadelphia and Tennessee at the bottom of its rankings thanks to big early defeats, with the result that both were misidentified as poor teams for much of the first half of the season. Indeed, it was this tendency of the DVOA ratings -- to not take sufficient notice of teams that had started off strong and were fading or teams that stumbled out of the gate but went on to play very well -- that necessitated the creation of the weighted DVOA metric. The Carolina Panthers were their own anomaly, as they looked like a lucky 8-8 team throughout the whole of the regular season before turning it on and playing three dominating games in the playoffs.

After the 2003 season, Football Outsiders tweaked the DVOA formula in an attempt to create rankings that more closely reflected the standings. The effect would prove to be dramatic.

Here is the list of weekly projections for the 2004 season:

Success of DVOA Picking Games, 2004
Week Win-Loss Cover Best Bets
16 11-5 8-8 1-2
15 11-5 10-6 5-1
14 12-4 9-7 3-0
13 11-5 10-6 1-2
12 13-3 11-5 4-1
11 13-3 11-5 4-1
10 9-5 9-5 5-2
9 6-8 6-8 1-2
8 7-7 7-7 2-2
7 7-7 7-7 2-2
6 11-3 11-3 3-0
5 9-5 9-5 3-1
4 9-5 9-5 2-1
3 11-3 11-3 4-0
Total 140-65 124-68 37-18
% .673 .596 .672

The improvement was dramatic and across the board. The standard win-loss selections rose to a robust 67% success rate. DVOA did a better job of picking games against the spread in 2004 than it did simply picking winners in 2003. Perhaps most tellingly of all, DVOA finally began to make good on its promise of identifying teams with records that were out of keeping with their performances, as the stronger underdog outperformed the spread 67% of the time, and in most cases won the game outright.

An examination of which teams were selected repeatedly as soft favorites or strong underdogs shows that DVOA accurately identified the rise and fall of multiple teams, and did so before the general public caught on. Here is a list of each team that was selected as either a favorite or an underdog more than once in 2004:

DVOA's Best Bets, 2004
Favorite Times Underdog Times
GB 6 CAR 5
SF 3 ARI 4
ATL 3 SD 3
TB 3 DAL 3

Early in the season, DVOA was quick to pick up on the fact that Tennessee, St. Louis, Carolina, and Green Bay were all radically worse than they had been the year before, and the lines were not yet reflecting that reality. It also picked up on the strong play of Atlanta, San Diego and Pittsburgh weeks before the lines caught up. Carolina and Houston both performed well in the second half of the season and were frequently picked as strong underdogs. The Bills' dramatic turnaround was identified quickly. The one team that gave DVOA a bit of trouble was the New York Giants, since it took many weeks for the insertion of Eli Manning into the starting lineup and the resulting offensive collapse to negate the team's strong early season performance.

But was the success of 2004 a result of an improvement in the prognosticating ability of DVOA, or were there other explanations that might suggest the season was a bit of an anomaly? The most striking thing about the 2004 season is that unlike 2003 the elite teams generally asserted themselves right away. The Eagles, Steelers and Patriots all came storming out of the gate early, meaning that in a good forty-plus games it was easy to discern the winner. That big a proportion of games could easily skew the overall statistics upward. So the question still remained, what would happen in 2005?

Well, the first thing that would happen is that FOX Sports would decide to replace their power rankings with the DVOA ratings, resulting in two distinct sets of ratings. The ratings posted here at FO continue to be strictly confined to what has happened in the 2005 season. But the second set was tweaked to factor in last season, and thereby to presumably mitigate the 2003 problem of slow-starting powerhouses taking a full season to work their way up the rankings by propping them higher than their early season performance might dictate. How has it all turned out? Not so well, at least not during the first three weeks. Here are the results so far for both the original and FOX DVOA rankings:

Best Bets
Best Bets
2 8-6 4-10 3-4 7-7 3-11 1-3
3 8-6 5-8 1-5 9-5 6-8 1-4
4 6-8 7-7 3-2 7-7 7-7 4-5

Uh-oh. The win-loss record for the straight DVOA rankings is .500, and both the regular DVOA and FOX ratings have been below par at detecting wonky lines. To a certain extent, this is to be expected in the early season when there isn't enough information to get a truly accurate fix on teams. (After all, those Week 2 DVOA ratings were based on just a single game.) An unusually good (or poor) performance in one game can still warp the overall ratings. Chicago and Buffalo still have high ratings because of a single dominating game (Chicago over Detroit, Buffalo over Houston) while New Orleans' rating is artificially inflated by their surprising opening day win over Carolina. The FOXSports.com power rankings have been giving the benefit of the doubt to Philadelphia, New England and the Jets, even though those teams haven't played with anything close to the same consistency as they did early last year.

There are other factors that go into early season games that simply cannot or will not be picked up on by DVOA with such a small sample size. Injury problems, particularly at quarterback, will take a while to settle out. There is no way for DVOA to account for the insertion of a Brooks Bollinger or an Alex Smith into a starting lineup. The Kansas City offense has been backsliding since Willie Roaf left the starting lineup, and the New England defense has fallen apart without Rodney Harrison. (The latter has made its way more quickly into the DVOA rankings, thanks to back-to-back games against teams with elite tight ends and strong inside running games.)

And then there are more subjective factors, which cannot be quantified at all but can nevertheless be readily identified by the human eye. It was obvious to most people that the Saints were going to play a good game in their home opener at the Alamodome, just as it was obvious that they were going to be dismantled by an angry and desperate Packers team a few weeks later. The Giants may have played well before their Sunday night tilt with the Chargers, but for most people, the question wasn't whether or not the Giants would lose the Eli Manning bowl, but by how much. As Bill Simmons noted last week, the Falcons looked like a good bet to continue the Patriots misery, right up until Tom Brady held a press conference and publicly told Marty Schottenheimer to mind his own business. At that point, you ignored the Patriots at your peril.

As sports fans, we all recognize these made-for-TV scenarios and the predictable results they generate, and over the course of a full season they tend to diminish in importance as depth and talent take over. By Week 10 or 11, DVOA may well have a better read on how two teams match up than your gut does. But in September and early October, you ignore your gut at your peril.

(Ed. note: I've been asked a few times whether the "FOX ratings" that take last year into account were "forced on me" by FOX. The answer is no, those were my idea, and I'll explain why I made that decision in the next mailbag, which will run sometime this weekend.)

Posted by: Guest on 14 Oct 2005

135 comments, Last at 10 Nov 2005, 10:01pm by TBW


by MIPackerBacker (not verified) :: Fri, 10/14/2005 - 5:57pm

New favorite FO phrase? "sort the wheat from the chaff."

by VarlosZ (not verified) :: Fri, 10/14/2005 - 6:26pm

Wow. The timing of this article is amazing to me, since I just yesterday finished work on a spreadsheet charting DVOA's success picking games against the spread last year. I may as well share the results, which are slightly different than the ones in the article due to varying lines and (probably) different methodology.

I only sampled Weeks 6-17, since DVOA fluctuates dramatically before then. The DVOA numbers through Week 8 of last year are not in the archives, so I couldn't include Week 9's games. I also did not count nine games in the last two weeks in which one team sat its starters (though DVOA wound up picking 5 of those 9 correctly).

I gave the home team a 15% DVOA boost (as per Aaron's suggestion in another thread), then worked out the difference between what DVOA thought the spread should be and what the actual spread was.

Overall: 87-70-2 (55.4%)
Best Bets (Top 4): 29-14-1 (67.4%)
Upsets (DVOA likes underdog to win outright): 14-14

DVOA picked the favorite 77 times and the underdog 82 times.

I'm currently working on spreadsheet using Offensive and Defensive DVOA to see if it can predict Over/Unders, but I'm not that confident that I know what I'm doing in that regard.

by VarlosZ (not verified) :: Fri, 10/14/2005 - 6:48pm

Looking a little closer, I'm now very curious as to how you determined DVOA's picks, Sean. I got the same results against the spread as you for Weeks 7, 8, and 16, but different results for every other week I processed. Most notably, you've got DVOA going 10-6 in Week 15, while I've got it 6-10. Also strange: in Week 13, your best bets went 1-2 -- mine went 4-0 (the overalls were closer: 10-6 you, 11-4-1 me). In fact, our best bets on a week-to-week basis seem to bear no relation to one another.

I'll post my methodology sometime today so everyone can point out its flaws and compare it to Sean's (if we get to see it, that is).

I'm really curious.

by Pat (not verified) :: Fri, 10/14/2005 - 6:49pm

Keep in mind that total DVOA is just the sum of offense, defense, and special teams. This is probably the best way to "generically" rate a team.

But if you've got two teams playing each other, those three components don't necessarily weight evenly.

Consider the ludicrous case where two teams never punt. Obviously their punt return/punting abilities completely don't matter. One team's punter could be dead and the other could be Todd Sauerbrun (on steroids) and it wouldn't matter.

It'd be interesting to toy around with a few things, like only using offense and defense and ignoring special teams completely, and seeing if that changes things.

Of course, if you wanted to be extremely advanced, you could always generate a toy simulation of two teams meeting each other, which would be pretty easy.

by Sean (not verified) :: Fri, 10/14/2005 - 7:02pm

Varlos- I consciously kept mine very simply by simply using the rankings rather than raw DVOA totals. You could get radically different results by matching up offensive DVOAs against defensive DVOAs, and I did some looking into that, but the article was going to get way too long and convoluted. So in the end I just took the rankings as were- if the second ranked team played the third ranked team, I assumed DVOA would take the second team, whether the differences in raw DVOA were negligible or very large.

by Pat Curley (not verified) :: Fri, 10/14/2005 - 7:11pm

I think the big difference between 2003 and 2004 was that Aaron did some tinkering with the ratings. I commented here in 2003 that the then-current version of DVOA was not doing well as a predictor; in 2004 I noted that it was doing significantly better (and better than my power ratings). Aaron mentioned in reply that he had changed some of the formulas.

by james (not verified) :: Fri, 10/14/2005 - 7:22pm

I've had a little success gambling using DVOA int the last two weeks.

I took the top 10 offensive DVOA teams and bet on all of the teams facing a defense with a positive dvoa.

Last week this resulted in
New England
St Louis(took Seattle bc of higher overall DVOA)

It ended up going 4-1 and getting me killed on the Philly bet.

This week spits out only a few teams so I need something new

by krugerindustrialsmoothing (not verified) :: Fri, 10/14/2005 - 7:31pm

very interesting, and I can't wait to read this again, however for now...

Car +1.5 over DET (Car has a 31% advantage in DVOA)

Cle +5.5 over BALT (Cle has a 31% advantage in DVOA, however lots of off field action in BALT)

nyg +3.5 over DAL (giants have a 29% advantage)

None of these picks have any adjustment for home field advantage. It would be interesting to see how the spread would factor in, you might also want to consider the following:

Mia +5.5 over TB (mia -6% DVOA vs TB)
was +6.5 over KC (was -5% DVOA vs KC)
ne +2.5 over DEN (ne -3%....)

a fascinating weekend shaping up to be sure.

by VarlosZ (not verified) :: Fri, 10/14/2005 - 7:36pm


-- A 0% DVOA has an expected value versus a league-average opponent of 21.5 points.

-- Home field advantage is worth roughly a 15% swing in DVOA to the home team.

I'm going to work through the first game on my sheet from last year, ATLANTA -4.5 vs. san diego, Week 6. Heading into the game, Atlanta's DVOA was 9.8%, San Diego's was 17.3%, and we add 15% to Atlanta's DVOA for being at home. The Adjusted DVOA Difference ("AD") is therefore +7.5% (that it's positive indicates that DVOA thinks the favorite will win straight-up).

To compare AD to the point spread, I convert that spread into Implied Adjusted DVOA Difference ("ID") -- i.e., what we would expect the difference in DVOA to be based on the spread. To get that number, I divide the spread by 21.5, then multiply by 100. In this case, its: (4.5/21.5)*100 = 20.93%.

Finally, to compare just subtract ID from AD, and look at the Discrepancy. In this case, its: 7.5%-20.93% = -13.43%. A positive Discrepancy means that DVOA likes the favorite to cover, a negative one means DVOA likes the underdog -- the size of the number say how strongly DVOA feels about the game. That -13.43 argues for San Diego with moderate strength.

I could also work the other way, and use DVOA to generate a "true" point spread, then compare it to the official spread. In this game, for example, DVOA said the true spread was Atlanta -1.6, instead of -4.5.

(Incidentally, Atlanta won the game 21-20, meaning that DVOA correctly picked San Diego to cover.)

So: How many mistakes have I made?

by VarlosZ (not verified) :: Fri, 10/14/2005 - 7:51pm

Re: #8.

Some of DVOA's thoughts on this weekend, according to my spreadsheet:

The numbers argue most strongly for SEATTLE -10 vs. houston, followed by CHICAGO -3 vs. minnesota. However, DVOA can't take into account injuries to Seattles top two WRs, and Chicago's rating is warped by their game against Detroit, which is looking like a crazy outlier.

However, DVOA likes cleveland +5.5 @ BALTIMORE almost as much as those two -- it likes Cleveland to win outright with a fair degree of confidence. It like the giants +3.5 @DALLAS to win outright almost as much (but beware the Giants coming off a bye). DVOA also very much likes Cincinatti and Buffalo to cover.

by Richie (not verified) :: Fri, 10/14/2005 - 8:00pm

I’m going to work through the first game on my sheet from last year, ATLANTA -4.5 vs. san diego, Week 6. Heading into the game, Atlanta’s DVOA was 9.8%, San Diego’s was 17.3%, and we add 15% to Atlanta’s DVOA for being at home. The Adjusted DVOA Difference (�AD“) is therefore +7.5% (that it’s positive indicates that DVOA thinks the favorite will win straight-up).


If Atlanta's DVOA is 9.8, and you adjust it 15%, doesn't that bring their new DVOA to 11.27%, and San Diego's at 17.3%, meaning that DVOA favors San Diego, who is the underdog?

by Aaron (not verified) :: Fri, 10/14/2005 - 8:05pm

Just to point something out, the whole "adjusted points" thing doesn't quite work as easily as that. (And no, I still haven't figured out if I want to start using it regularly or not... opinion was really split down the middle on the idea.) The idea of adjusted points also adjusts each team to the same number of plays per game -- remember DVOA is a per-play measurement. During the season, that means that a game between two teams that tend to run fewer plays (say, PIT and NYJ) would be likely to have a lower spread than the "21.5" thing would otherwise indicate, while a game between two teams that tend to run more plays would likely need a higher spread. If I'm remembering my own little equation correctly, that is.

by VarlosZ (not verified) :: Fri, 10/14/2005 - 8:44pm

#11 (Richie): If Atlanta’s DVOA is 9.8, and you adjust it 15%, doesn’t that bring their new DVOA to 11.27%, and San Diego’s at 17.3%, meaning that DVOA favors San Diego, who is the underdog?

Sorry, I should have said 15 percentage points, not merely 15%. You add the 15 to the 9.8, giving Atlanta an Adjusted DVOA of 24.8%.

#12 (Aaron): Just to point something out, the whole “adjusted pointsâ€? thing doesn’t quite work as easily as that. (And no, I still haven’t figured out if I want to start using it regularly or not… opinion was really split down the middle on the idea.) The idea of adjusted points also adjusts each team to the same number of plays per game – remember DVOA is a per-play measurement. During the season, that means that a game between two teams that tend to run fewer plays (say, PIT and NYJ) would be likely to have a lower spread than the “21.5″ thing would otherwise indicate, while a game between two teams that tend to run more plays would likely need a higher spread. If I’m remembering my own little equation correctly, that is.

That is a concern, but there are so many things that this system doesn't take into account, so something like that might be cancelled out by some other factor (or, possibly, excacerbated). In any event, since I'm measuring comparitive quality instead of gross output, I think that pace is only a small obstacle.

On the other hand, I'm discovering that game pace is a huge problem for DVOA Over/Under projections. For example, last year's DVOA says that the Jets were a very good offensive team and a below average defensive team, but that doesn't show up on the scoreboard at all. I suspect that any Over/Under system based on DVOA would have to be very complex to be worthwhile.

by Richie (not verified) :: Fri, 10/14/2005 - 9:19pm

Sorry, I should have said 15 percentage points, not merely 15%. You add the 15 to the 9.8, giving Atlanta an Adjusted DVOA of 24.8%.

That can't be accurate, can it?

Using the final 2004 DVOA, that means that Buffalo at home is expected to beat New England. I would think (and may very well be wrong) that you should add 15% for home field, not 15 percentage points.

by benjy (not verified) :: Fri, 10/14/2005 - 9:41pm

Just a friendly warning to limit the G word (that rhymes with Rambling) and related terms: our comment spam system tends to flag them.

We now return you to our regularly-scheduled discussion.

by Reno (not verified) :: Fri, 10/14/2005 - 9:58pm

As someone who's had enormous success betting with DVOA using a more complex incorporation of it, I don't really understand the point of Sean's article. A point spread is essentially a measurement of two teams' value relative each other in the gambling public's collective mind. The point spread doesn't just say "The Colts are better than the Rams," it says "collective wisdom is that the Colts are 13.5 points better than the Rams."

If you're just going to bet the team with the higher total DVOA, you're best off doing it against the win line. As Sean's study DOES show, you'll be succesful most of the time. But using that same methodology to bet against the spread just isn't utilizing DVOA to its fullest extent, or particularly intelligent. Just looking at total DVOA and seeing that Team A is better than Team B, does not necessarily mean that DVOA likes Team A against Vegas's spread. In fact, it often doesn't.

It's pretty easy to use just an even slightly more advanced look at DVOA like VarlosZ's to come up with a spread. Comparing DVOA's spread to Vegas', and then picking accordingly, is going to demonstrate a significantly less erratic (and higher) performance for DVOA, especially on "Best Bets." But using DVOA's win picks as its spread bets doesn't strike me as being particularly useful (or fair to DVOA) when it comes to writing an article about "how does DVOA do against the spread?"

I don't particularly like talking about how I make my picks - I go a bit further in-depth with DVOA than VarlosZ, but don't particularly have the time, energy, or interest to make it as accurate as possible. But using DVOA and the occasional sprinkling of common sense re:injuries and outlying past results that can skew DVOA, I've posted very, very good results the past couple of years. And I was a history major who went to a college that I chose largely because it DIDN'T have and a football team, and where I could graduate without ever taking a single math course.

by Sean (not verified) :: Fri, 10/14/2005 - 10:05pm

Reno- The DVOA against the spread stuff is just there for reader interest, as there is no reason why a team that was favored by DVOA should cover an arbitrary spread. What I was really focusing on were the instances where the higher rated team was the underdog, where you would expect an upset a good portion of the time.

(And I should point out that while using betting lines obviously tilts the focus of the column, that wasn't entirely my intention. I just wanted to see whether DVOA was picking up on rising and falling teams before the public perception that the betting line represents was.)

by VarlosZ (not verified) :: Fri, 10/14/2005 - 10:08pm

Heh, it's even worse than you think: what I forgot to mention was that I use Weighted DVOA when it becomes available instead of Total DVOA, so by the end of the year Buffalo would be expected to beat New England on a neutral field.

Two things about that. First, even if we look at that and think that the system is giving Buffalo a little too much credit, we can still assume that Buffalo is probably closer to New England than most people realize and pick accordingly.

Second, it's not a completely ridiculous outcome. Aside from their inexplicable stink bomb in the last game of the year, Buffalo was clearly one of the best teams in the league over the second half of the season. After losing to New England in Week 10, Buffalo had been crushing their opponents. They won their next six games by the following scores: 37-17, 38-9, 42-32, 37-7, 33-17, and 41-7. Also, look at these two common opponents Buffalo and New England faced over the last few weeks of 2004:

Buffalo @ Miami, wins 42-32
New England @ Miami, loses 29-28

Buffalo @ Cincinatti, wins 33-17
New England vs. Cincinatti, wins 35-28

(They both slaughtered Cleveland.)

Buffalo had been a better team from Weeks 11-16, and had fared considerably better against common opponents. So, predicting that Buffalo would probably beat New England at the end of the season would be wrong, but it would not be ridiculous.

As for the home field percentage adjustment, 15% isn't an arbitrary number. Aaron said in another thread last week that some tentative research showed about a 15 point DVOA swing to the home team. Incidentally, in terms of expected points that works out to 3.23 points for home field, which is pretty much exactly how much the home team gets in the point spreads.

by Aaron (not verified) :: Fri, 10/14/2005 - 11:30pm

17%, actually.

by Fnor (not verified) :: Sat, 10/15/2005 - 12:11am

You mean I can't talk about general Grambling? That's cruel, benjy :(.

by Jerry P. (not verified) :: Sat, 10/15/2005 - 12:29am

Varlos, the Bills were not a better "team" at the end of the season. Their defense went from merely good to butt crazy good. Over those 6 games you gave the scores for they were averaging like 4 possessions a game starting inside opponents territory due to having an amazing defense and probably the best punt/kickoff return unit in the league. The offense got barely competent enough at the end of the season to capitalize on insane field position. Check the difference between their offensive DVOA and Weighted DVOA last year and then do the same with the defense. You'll see the weighted DVOA was high because the defense got even better while the offense still generally stunk. Check the drive stats too. Thoroughly average except for the best field position in the league.

So I guess what I am saying is that Buffalo is kind of a whacky team to use for last years numbers since the quality of the offense and defense was so lopsided.

by EorrFU (not verified) :: Sat, 10/15/2005 - 12:42am

I started last week using the Offensive and Defensive DVOA matchups to pick my bets last week. It has been really good at picking up unexpected holes in passing and running games. So far it has been much more accurate than picking by total DVOA so far. It also has been really effecive at picking over's the past two weeks. The major games that the Off/Def comparison. I took a +/- 1.5% to the total offensive output. It whiffed on the Cincy and GB games completely. But it had the Skins in a tight game, SEA crushing STL in a shootout, IND covering, and Pit winning outright by a point. It is way too early for DVOA to be consistent but it seems to not be out there. Actually I was originally suprised at how often the results matched what the o/u was.

In reality it relly helps show bad matchups like a good passing game over a weak secondary.

by Brainster (not verified) :: Sat, 10/15/2005 - 6:23am

Seems to me that if you really want to convert DVOA into points, the easiest method is to convert to power ratings. If we assume that the best teams in the NFL are about 12.5-15 points above average (as they have ranged for many years), and the worst teams are about an equal amount below sea level (ditto), then it's relatively trivial to plot everybody else in on the line between them based on their DVOA ratings. After that it's just a matter of adding in the appropriate adjustment for home field advantage (usually 3 points, although it's averaging quite a bit above that this year), and you have a reasonable line to compare to the published lines.

The obvious question though is how applicable your derived line is given the injury situation of the teams involved. Is Pittsburgh as good as their DVOA indicates if Tommy Maddox is the starter instead of Ben Roethlisberger?

Still, DVOA worked tremendously last year in predicting future performance just on a won/lost basis, so I have no reason to doubt it would do very well if carefully applied and with adjustments for or refusing to bet on teams with significant injured players.

by VarlosZ (not verified) :: Sat, 10/15/2005 - 8:14am

17%? Hmmm. Making that change improves DVOA's overall performance last year by a few games, but brings its Best Bets record down from 29-14-1 to 25-18-1. It does do a better job with straight-up upsets, however, going 12-10 instead 14-14.

Say, am I just missing the Week 8 DVOA numbers from last year, or are they really not in the archive? Anyone?

by Jerry P. (not verified) :: Sat, 10/15/2005 - 1:37pm

Varlos, week 8 DVOA numbers for 2004 are in the archive. If you're talking about week 9, which I did not see, maybe the midseason report is the week 9 DVOA as it does contain the DVOA stats for every team. Just not in one huge table.

by John Gach (not verified) :: Sat, 10/15/2005 - 4:59pm

I believe that Reno's observations in #16 are exactly correct: use DVOA to bet the money line and for the most part ignore the point spreads. It's one thing to expect the superior team to win and quite another to predict by how much -- recall TMQ's citations of how rarely prognosticators ever get the actual score correct. As Mr. McCormick's essay points out, the revised DVOA is a better tool for selecting likely winners than it is for selecting teams that will beat the spread, though even there it's pretty good -- a nearly 60% success rate is itself extraordinary.

One issue Reno didn't mention is that using DVOA makes some parlay bets much stronger than would usually be the case -- especially when the Vegas odds favor the "wrong" team (as with DET this week). For example, betting this week on CIN, SEA, & CAR to win -- all strong DVOA-based wagers -- yields a significantly better return than betting the individual money lines.

Another good strategy is making very large bets on DVOA locks (like IND vs. SF or PHI vs. SF) whenever they are available. The caveat is that such moneyline wagers on heavy favorites often have to be made early in the week as they tend not to be available later (and even if they are, the likely return will be smaller). Unfortunately, these are not always available (e.g., [at least at SportsInteraction] the money line for IND vs. STL has been closed all week), so one has to seize such opportunities when they exist.

So, at least for now, DVOA seems to allow for significant arbitrage. How long that remains the case is a different matter, since there is no reason that oddsmakers can't factor it into their betting lines. So long, however, as the majority of wagerers make less rational decisions, the opportunity for arbitrage probably will continue to exist, since the odds are set to encourage equal opportunity wagering on both teams as well as to set an expectation for a probable outcome.

This year ha presented an unsual set of wagering opportunities: SF's & HOU's DVOAs are so minuscule that one needs a microscope to see them. Checking the past 5 years, I find, as one would expect, that the bottom team's DVOA is in the -30 to -40% range (the inverse of the top team). This year SF is at -70.8% and HOU at -56.9%. I suppose they could significantly improve; but just to achieve average bottomness [1] they both have to play at a nearly average level for a number of games. When either plays even an averagely good team on the road, it has little chance of winning.

[1] I'd write "mediocrity" but SF & HOU can only aspire to mediocrity in their fantasies.

by Reno (not verified) :: Sat, 10/15/2005 - 7:36pm

Yeah, even before FO I was a pretty big believer in betting the money line.

One interesting thing about this thread is the different ways in which people can interpret the same data. The Detroit-Carolina game, for example, doesn't seem to me like a "lock" for Carolina based on my DVOA interpretation at all. Despite the huge difference in total DVOA, my entirely DVOA reliant methodology actually likes Detroit (though with very little confidence) against the spread, and likes Carolina to win, but not overwhelmingly so.

There's the possibility that the Panthers' banged up backfield could effect them more than the Lions' missing receivers (who weren't doing a lot/being made much use of/whatever in the first place). However, now that Aaron has the DVOA vs. receiver type up (not something in my formula), we can see that though Detroit has a high DVOA against the passing game, they look ripe to be gashed by Smith. Yuck, it's just not the kind of game that inspires much confidence in me one way or the other.

It's not a game that I'm going to place a wager on. I wouldn't be suprised at all if they blow Detroit out of the water like Total DVOA says they should, but also won't be at all suprised if Detroit plays them close and wins it by a field goal.

by VarlosZ (not verified) :: Sat, 10/15/2005 - 7:46pm

"Varlos, week 8 DVOA numbers for 2004 are in the archive."

They are now, but I'm almost certain they weren't before.

Week 9 of last year turned out to be one of the strongest for DVOA. 10-4 overall, 3-1 best bets, and 3-0 in upsets.

"I believe that Reno’s observations in #16 are exactly correct: use DVOA to bet the money line and for the most part ignore the point spreads."

Actually, he said to use DVOA rankings only to bet the money line; the main point of his post was that the DVOA ratings could be used to bet against the spread very effectively. You're not trying to predict exact outcomes -- all you have to do is determine if the gap between two teams is significantly greater or narrower than the public and/or the odds-makers believe.

by Reno (not verified) :: Sat, 10/15/2005 - 8:42pm

Actually what I do is kind of a combination. I use DVOA to come up with an expected margin of victory, but then what I do is bet the money line based upon that. DVOA does well against the spread, but not with the same predictability it does against the money line. I keep track of my wins and losses, and start betting in week 5. The past 2 weeks, I've seen a 28% return.

I just did a run-down of how I would've done if I'd been making spread bets the same way I make win bets (not betting the team I'd bet to win to win against the spread, but placing my bets in a similarly weighted fashion based on expected performance against the spread), and the past two weeks I would've netted a -%34 loss. The actual record against the spread (12-14-1) on games I would've wagered on isn't nearly as atrocious, but due to weighting my bets, I would've gotten killed. I've never done this at the end of a season to see how things would've gone if I'd bet the spread the way I do the money line in terms of how I'd do in real money terms. Past experience tells me that the 12-14-1 would eventually edge up toward something like 55%-65%. But it wouldn't have helped me much if a bunch of those wins came on $50 bets and I lost too much on $500 bets.

The point spread is just not something I like betting on - the potential for coaching decisions on 2 point conversions and field goal attempts, the prevent defense, running out the clock, etc. isn't worth the headaches, in my opinion. The intention behind my original post was more about suggesting improved methods of evaluating DVOA performance vs. the point spread in terms of winning percentage. I like betting on wins rather than victory margins because it rewards me for identifying the teams that are most likely to do what they're ostensibly trying to do - win football games. I've never really understood the appeal of betting against the spread, especially since as most readers of this site likely know, points scored isn't necessarily the greatest indicator of a team's on-field performance. This is why we love DVOA and its high correlation efficients so much, it seems silly to forget one of its most basic lessons when it comes to playing with your money.

But maybe this is just my bias against/fear of the prevent defense and Easterbrook's "why are you kicking a field goal?!" situations. I'll chart how I would've done against the spread for the rest of the season, along with how I do betting money lines.

Lastly, since I've been reading this site for a couple of years now and my first posts have been about what a degenerate gambler I am, I feel like I should thank the FO crew for really and truly helping my understanding of and appreciation for the game as a FAN who still never bets against his beloved Raiders, no matter what the numbers tell me or how mediocre they are. You guys are doing great work and making a really wonderful contribution to sports journalism, heck, are actually engaging in sports journalism (something most so-called practitioners don't do). I appreciate that this isn't a gaming website, and am sorry if all this "DVOA in action" chatter feels cheapening in some way.

by Reno (not verified) :: Sat, 10/15/2005 - 8:47pm

Correction - I start betting in Week 4.

Addendum: The 28% return came on a w/l record of 13-9.

by Reno (not verified) :: Sat, 10/15/2005 - 9:22pm

Oh, and hate to be like Carl, but I had the wrong line for Lion-Panthers, I obviously don't think its going to be a tie.

by RowdyRoddyPiper (not verified) :: Sun, 10/16/2005 - 12:04am

After getting carted weeks 1-5 in my high dollar pick against the spread pool. Integrate DVOA into my process for making the picks. The process before was dropping my bloody mary, running to the computer and punching in whatever I could before the site locked me out.

At this point I just compare offensive vs. defensive DVOAs to come up with an adjusted offensive expectation. I just use a league average of 21 points to convert the DVOA to points and tack on some HFA numbers. After reading this thread I'm going to have to look for some adjustments to DVOA to account for ball control vs. balls to the wall offenses, but that's a later task. I'll try and keep people posted on how it works for me.

Benjy, no matter what you say, I'll keep talking about Eddie Robinson.

by bethehouse (not verified) :: Sun, 10/16/2005 - 12:35pm

Reno, I fully agree about betting the money line rather than betting the spread. I find it much easier to formulate the probability of each team winning, rather than the point spread. Although once you get that probability of a win, it is simple enough to match it up with a correlating spead. For instance, a 3 point favorite is expected to win 57-58% of the time. But I think that is just taking it one step more than necessary and dilutes the bet essentially. My next project is over/unders, which have been difficult to this point. Email me at bethehouse@gmail.com if you want to collaborate further. Maybe we can help each other out.

by Jason (not verified) :: Sun, 10/16/2005 - 2:40pm

I am not in love with any games this week but here are my 2 Locks of the Week.

Panthers +1
Steealers -3

by Falco Sparverius (not verified) :: Mon, 10/17/2005 - 4:20am

Re: 18, 19 and 24 on Home Field Advantage (as well as numerous discussions on previous mailbag posts).

If you are trying to incorporate and calculate HFA for predictive purposes, it is my belief that it is not a simple linear situation. HFA does measure to somewhere between 3.2 and 3.5 for all NFL games. However (and I am basing this on my research before I was married and working full time) it is more like a bell curve. HFA tends to be greatest when the home team is considered "slightly worse" than the road team (4-8 points on a neutral field). If you look at point spread performance of home underdogs of 4 or less, you should find that this group of home teams has "outperformed" other groups over the last 10 years. Back when I was measuring it to incorporate into handicapping, it was running at closer to +4.25 for teams that were slight home dogs. The bell curve then decreased as you moved away from that peak in either directions. The HFA for large favorites (10 or more) was closer to +2.5, as was the HFA for large home underdogs.

I first noticed this in NCAA games. I noticed HFA was closer to 5 in the slight home dog situation, decreasing to barely 1 point when the line was 30 or more.

Also, and DVOA would be an even better way of measuring this, home teams that rush the ball better than their visitor opponents (greater than 25 yds more per game) and stop the run better than the opponent (at least 25 yds fewer per game) tend to perform well against the spread at home, particularly when an underdog or slight favorite. Chicago met that criteria this week, and covered handily. There was an old site called www.sillysports.com where I first saw this angle.

I also think the home underdog HFA is greater when the home team is at least decent playoff caliber, and is good defensively.

Some of the largest blowouts/discrepancies result when a home underdog is actually better than or closer to the road favorite than is generally perceived by the general public. It's almost as if the perceived vs actual difference between the teams is magnified because the home team is playing with more of a chip on their shoulder and/or thinks the visitor is better than they are and is motivated accordingly. Chicago vs Detroit is one that I suspect at season's end will fit this bill, and also Dallas whipping Philadelphia at home last week.

I can think of at least some reasons to support these numbers:

1) despite what they may say, teams are aware of the lines, and coaches use this as a motivating factor. The perception that you are being "dissed" when playing a similar or competitive opponent at your place can be a powerful motivating factor.

2) emotions are more valuable to a defensive performance, which can be fed by a home crowd. A good defense will keep a crowd in the game, and be better able to feed off the crowd. Similarly, if home team can run the ball significantly better than the opponent, the opposing team will be more likely to face long yardage situation where crowds can have an impact on offensive line communication and audibles.

3) crowds know the quality of opponent and may be more fired up and responsive when they feel they matter, and less into it when facing a clearly inferior opponent, or say, when attending a game in Arizona.

In looking at the results this year, so far home dogs of 4 or less are 8-7 ATS and 7-8 straight up. Total Points scored in those games are 285 (H) to 286 (A). Breaking down the games, you can see that whether the home team could run/stop the run relative to the visitor does impact the likelihood of success of the home team. You could also look at the list and see that a few plays going the other way late in games would have made the home dogs show even better, and DVOA may reflect some of that.

The wins are:
ATL +1.0 vs PHI
TEN +3.5 vs BALT
CHI +1.5 vs DET
CAR +3.0 vs NE
NYJ +2.5 vs TB
MIA +3.0 vs CAR
DAL +3.0 vs PHI

The cover/loss was
GB +3.5 vs TB

The losses are:
BAL +3.0 vs INDY
OAK +1.0 vs KC
TEN +3.0 vs CIN (Ten led 23-20 with 5:00 remaining)
OAK +3.0 vs SD
CLE +3.5 vs CIN
CHI +3.0 vs CIN
ARI +2.5 vs CAR (Ari led and game could have gone either way late)

Cincinnati's top ranking on this site is justified, and is even more impressive considering they are 3-0 as a slight road favorite, where road disadvantage tends to be strongest. (As an aside, they also played Houston in a letdown, high favorite game where HFA is generally less) If you exclude Cincinnati, the other slight road favorites are 4-8 against the spread. I think not coincidentally, the other two comfortable wins (INDY over BALT, SD over OAK) by slight road favorites involved two teams ranking very high in DVOA.

by masocc (not verified) :: Mon, 10/17/2005 - 9:00am

I work for the bookies in Las Vegas.

So I am really getting a kick out of most of these replies.

Some of you guys are very good at making it sound like you know what you are talking about.

But trust me.... You don't.

I think you just want to make yourself sound smart, when in reality you dont know what you are talking about.

This is how bad info gets passed around.

If you dont know about the topic....Dont make yourself sound like you do.

Cuz some Football Outsiders belive anything they hear.

by david mazzotta (not verified) :: Mon, 10/17/2005 - 11:32am

masocc, instead of just condescending, why don't you join the conversation constructively and correct our errors?

by S (not verified) :: Mon, 10/17/2005 - 11:35am

Masocc, in your experience working for "the bookies in Las Vegas", do they take their coffees black or with cream?

by Astro Boy (not verified) :: Mon, 10/17/2005 - 3:15pm


If you have to stay mum on this issue, so be it, but I’m dying to know what you can say about the line in Hou/Sea last night. I set my own lines before checking out what the houses have to offer, and it’s occasionally very useful to capitalize on discrepancies between what I feel a line should be and what is actually offered. I thought the line of Sea -9.5 or -10 was a total joke; I had it at 14.5 and wouldn’t have been surprised to see it even steeper. As I noted in the thread on Dr. Z’s column on gambling, I like to bet teasers (Seattle laying only 3!!!) and to hedge in the night games and go for middles. I was completely confident that the line would move enough that I could middle 10 or even 11 by kickoff if I needed to hedge. Instead, what floored me even more than the opening line was that over the course of the week, the line dropped to 9 at the two houses I use; only one had it back up to 10 at kickoff. Yes, I know that lines aren’t supposed to predict outcomes; they’re supposed to split the money. Given the line movement, it looks like a lot of money went on Houston, and I’m just stunned about that. So maybe there was nothing more to this line than trying to split the money, but even with the line movement, I just can’t believe a line of -10 would do that. Houston did cover a similar line against Cincy, but I thought that line was a joke as well. Houston got a massive boost when Cincy had to go to their third-string center, but Houston was so ineffective on offense that my teased plays on Cincy were still able to survive that day. Sure, Seattle was out its top two receivers, but the backups did fine in St. Louis the week before, and if there were any questions about Seattle’s passing game, they still had Sean Alexander, who always seems to dominate weak teams on Sunday nights. The game unfolded exactly the way I expected to, and I did make money off it, but I found myself wondering the entire game why the line wasn’t steeper. Any info to pass on?

by another_farker (not verified) :: Mon, 10/17/2005 - 4:22pm


Astro Boy:
The 'I work for ___ ...' thing is the latest fark.com cliche.

by masocc (not verified) :: Mon, 10/17/2005 - 4:37pm

Thanks for that, farker! I was wondering how long I was going to have to sit here looking like an a**shole ;)

by 2 cents (not verified) :: Tue, 10/18/2005 - 2:32am

Do VarlosZ and Aaron run this site? Why does everyone try to calibrate their analysis with those 2?

I find it impossible to use DVOA to figure out what lines to bet on? VarlosZ does talk about it a little bit earlier but I dont his math as in #11, 13, 14. How can 15% become 15 points added to a DVOA??

Forget the spread, has anyone used the DVOA to predict the winners each week? Wonder what the results might be?

by Sean (not verified) :: Tue, 10/18/2005 - 2:36am

I suspect money went on Houston for the same reason that money will go against Indy in the next few weeks- because the odds tend to catch up with both winless and undefeated teams around now.

Good week for DVOA, btw. 11-3 straight up.

by Reno (not verified) :: Tue, 10/18/2005 - 2:43am

re: spread vs. money line

So, betting the money line this week, DVOA and I netted a 29.8% gain on an 11-1 record, taking my season totals to 52.9% and 24-10 over the past 3 weeks.

If I'd taken the same methodology and bet the spread accordingly, I would've posted a 28.5% gain on an 8-2 record. For the season, I'd be at -3.8% and 20-16-1. That I would've bet on two less games signifies their lines being very close to DVOA's expected margin of victory.

Percentage wise, this was a slightly better week than normal for DVOA on win bets, and as I haven't started looking at it until this season, I have no idea how aberrant the spread showing is. DVOA's straight up record against the spread for the season is about where I'd expect it to be. That there's a negative return so far means that DVOA's been pretty poor when it comes to "best bets" against the spread. This isn't terribly shocking to me, given my previously noted angst with betting point spreads. I also don't have any doubt that for the season, DVOA vs. point spread will result in a positive appreciation, and a pretty impressive one.

I also think that it'll continue to lag behind the money line. DVOA's so incredibly efficient at picking the winning team, most especially on its "locks," that I think its high winning percentage will more than make up for the disadvantage it gets in odds (it usually only likes one to three underdogs a week).

I know that everyone's going to cry "small sample size!" - and rightfully so. I don't really have the time or desire to go back through last year's DVOA ratings, spreads, and results to come up with results. I'll stick by my belief that betting on wins is better than betting on margin of victory, though.

If people want me to post weekly updates on this, I will, though I kind of doubt anyone's interested.

Oh, and throughout the above, when I say "DVOA" what I really mean is "my interpretation and application of DVOA." I should also say that I'm almost completely positive that no one on the FO staff would ever encourage gambling, that there's never any such thing as a "sure thing", and that I don't particularly endorse gambling very much, either, especially if you don't know what you're doing, and most especially if you aren't prepared to and able to lose every bet you make.

by Reno (not verified) :: Tue, 10/18/2005 - 2:55am

Oh, and I recognized Masocc's joke, and laughed all the harder because it's not entirely untrue. A Vegas oddsmaker would make me feel very, very stupid when it came to talking about betting on sports. I've got enormous respect for them, I'm almost positive that they're aware of and incorporate DVOA in more effective ways than I do. Hell, I wouldn't be suprised at all if they've had tools like it for years.

What gives me an advantage when betting is that:

A) The Vegas oddsmaker isn't trying to come up with statistically accurate measures of winning probability and margin of victory.


B) Most people aren't betting the way I do.

by Reno (not verified) :: Tue, 10/18/2005 - 3:00am

Argh sorry for the triple post. I should note that the gain/loss percentages are not taking into account teases. I bet each game seperately, and also tease on the side. These are just the individual game bets, because I didn't want to have to work out teases for spread bets I wasn't making.

by 2 cents (not verified) :: Tue, 10/18/2005 - 3:05am

(reno) I should also say that I’m almost completely positive that no one on the FO staff would ever encourage gambling, that there’s never any such thing as a “sure thing", and that I don’t particularly endorse gambling very much, either, especially if you don’t know what you’re doing, and most especially if you aren’t prepared to and able to lose every bet you make.

Agree 100%. For week 7 these are my only 2 locks:


cincy (if rothburger doesnt play)
saints (if bulger doesnt play)

everything else is too close to BET on.

by John Gach (not verified) :: Tue, 10/18/2005 - 3:05am

In case anybody is still reading this interesting thread: I had my best week ever, winning six bets with no losses. Since specifics are always useful, here are the wagers:
1) Seattle to win: 40 units, return of 8.6 units.
2) CIN & SEA to win: 10 units, return of 9.7 units.
3) CAR CIN SEA to win: 5 units, return of 19.28 units.
4) CIN SD SEA to win: 3 units, return of 10.48 units.
5) ATL CIN SEA to win: 3 units, return of 8.29 units.
6) CAR ATL CIN SD & SEA to win: 1 unit, return of 9.55 units.
Return here means "amount won."
The basic strategy entails firstly identifying one or two highly probably winners (SEA was this week's lock), assigning a degree of confidence in several other contests where DVOA strongly suggests that one team should win, then wheeling the one or two lock picks with probable but less certain predicted winners. Note that SEA is in all six bets and CIN in 5, while CAR is only in 2, since like Reno (and, really the oddsmakers, since DET was a 1 point favorite) I thought it was pretty much a pick-em. The large bet on the lock pick protects against anticipated losses with some less confident wagers (though that didn't happen this week). INDY was my second lock, but the money line wager wasn't available.

It's useful to be able to include road teams in parlays, since the return is higher. One needs, though, to be very careful doing it, since the risk is higher as well.

The 5 team parlay was an unusual, almost throw-away wager that I didn't expect to win. It came about because my analysis of the week's games provided no reason for believing that any of the 5 teams shouldn't win, though I certainly didn't feel overly confident about CAR & ATL. Actually it was originally a 7 team parlay, but that just seemed too crazy to go with (even though in fact both other teams also won). I have, however, already won two 5 team parlays this year.

I should add that I don't use spreadsheets or statistical apparatus other than DVOA, but make intuitive -- really aesthetic -- judgments after steeping myself in the statistical data. I should also add that being able rationally to wager on sports contests requires great discipline and, at least in my case, a tuition of spending about 5 years losing money by doing everything wrong. It's a skill like playing a musical instrument: one improves with practice and mistakes are ultimately the best form of learning. Anyone who thinks that they're just going to jump in and start making the big bucks wagering on sports contests is probably deluding himself. Even a sophisticated statistical armamentarium doesn't by itself suffice -- e.g., note how poorly nfl-edge's picks have done before this week (www.nfl-edge.com).

The statistical inferences and generally outstanding essays at FO provides the great benefit of an objective, abstract model that is divorced from the gritty reality of the games themselves. This makes it much easier to make inferences about probable winners & losers without the disturbance of "noise" like gut feelings and emotion (thus I rarely wager on my own favorite team, since emotion clouds one's judgment).

So far this year, beginning with week 2 (admittedly a tad early), I've had one losing week (week 2, of course). Since then, I've lost one wager (a 3 team parlay with PHI picked over DAL -- a bet that, looking back at it, shouldn't have been made since I knew that Philly in Dallas that week was a "trap game").

by Budman (not verified) :: Tue, 10/18/2005 - 6:17am

The only thing wrong with this is that the week when your "Sure winner" doesn't win, the whole deck of cards comes crashing down along with all the those ML Parlays you played with all those units.

If you been around sports gambling long enough you know there is NO "sure winner" no matter WHAT statistical measure says it will be. Trying to turn big ML favorites into better payouts by parlaying is a dangerous game to be playing.

I am glad for your success, but I don't think that Vegas enjoys being someones ATM. They have some pretty smart people working for them to make sure they don't for long.

PS - "Two for the Money" is a good movie for anyone who has ever placed a sports wager.

by Jay (not verified) :: Tue, 10/18/2005 - 8:01am

Re:9 Varlosz, Just curious but how that worked out on this years games?

by Jimmy Two Times (not verified) :: Tue, 10/18/2005 - 12:17pm


I'm not sure, but you might be me. It's not just that I go by Reno on a few other sites (Magnetic Fields reference), but I'm also a huge Raiders fan and degenerate gambler. I feel as though my identity has been stolen.

Anyway, I wanted to let you know that I'd be interested in weekly gambling updates. How about a best bets of the week post on Fridays? I know that the FO staff is sensitive about gambling, so I'd also be happy to take the discussion offline.

My DVOA-based system had a pretty good week against the spread, incidentally. The best bets were:

1. Chicago -3
2. Chicago/Minn UNDER 37.5
3. Seattle -9
4. Pitt/Jax OVER 34
5. Cleveland +6
6. Pittsburgh -3
7. Cincy -3
8. Cleveland/Bal OVER 34.5
9. KC/Was OVER 43
10. Sea/Hou OVER 46
11. Car/Det UNDER 41.5
12. TB/Mia UNDER 35

These bets went 8-4. Really, apart from getting the Cleveland-Baltimore game massively wrong, DVOA had a really strong week. A couple of things it got wrong were through no fault of its own; DVOA didn't know that Hines and Roethlisberger were out, for example, but if you'd exercised a bit of common sense you wouldn't have been steered wrong.

by VarlosZ (not verified) :: Tue, 10/18/2005 - 12:39pm

Re:9 Varlosz, Just curious but how that worked out on this years games?

I only just began tracking results this week, as up until now DVOA numbers fluctuate wildly. This week, we were 10-4 overall, 3-1 best bets. We were 1-2 picking upsets if you include Carolina, which opened as a 1-point underdog. Otherwise, 0-2 on upsets (Giants and Browns).

by Jason (not verified) :: Tue, 10/18/2005 - 2:08pm

Re52: Thank you VarlosZ. I'm looking forward to trying it this week.

by Jason (not verified) :: Tue, 10/18/2005 - 2:09pm

Re52: Thank you VarlosZ. I'm looking forward to trying it this week.

by COINFLIP (not verified) :: Tue, 10/18/2005 - 2:56pm

I've been comparing my own systems to DVOA for two years and have yet to beat it.

This year, I'm also tracking against Sagarin's ratings from USA Today.

So far, he's in the lead!

Click the link for more.

by John Gach (not verified) :: Tue, 10/18/2005 - 6:32pm

Of course Budman is correct in #49: there are no "sure" bets and over a long enough stretch several things are certain to happen: 1) a run of just plain bad luck (big favorites losing, for example); 2) just as Budman suggests, one or more weeks were the interlocked wagers built on one or two hypothesized victors result in large losses. So, one has to be prepared for both eventualities. This is partly a deep-pockets issue; partly an emotional issue (one has to be prepared for such occasional losses); and partly an issue of money management -- of not risking too much in any one week in relation to the pool of money available for wagering.

There is also the danger that one is having what economists call "a random walk" -- an above average performance that in fact has nothing to do with the reasons one thinks are responsible for the success (e.g., Darvas's stock picks in the 1960s, which on later analysis turned out just to be lucky guesses).

Of the three factors I mention, the deep pockets issue may be the most important -- assuming one's approach to the other two is reasonable. If I start with $100 and you start with $10,000 and we flip coins at $1 per call, you end up taking all my money in fairly short order, even though it's an exactly fair bet each time with no vigorish.

On the other hand, if one is wagering with the other guy's money and continues to exercise caution, restraint, and lack of greed, I don't see why in sportsbook wagering one can't continue to generate an above average performance of 55%-60%. After all, this is one of the areas of gambling where skill actually makes a difference and where it is known to be possible to generate above average results. My best guess as to why this is the case involves the distinction between frequentist and Bayesian probabilities (for a decent discussion of the latter see http://en.wikipedia.org/wiki/Bayesian_probability). Sportsbook wagers entail not just facts about the world but opinions about future facts that are themselves entangled with the opinions that others hold, which is a feasible arena for Bayesian analysis, which can assign probabilities to probabilities. I think that that is about what I am doing by using a combination of inductive reasoning and what C. S. Peirce called abduction, which more or less means using inductive inference to reach a correct conclusion, even though the inferential chain does not warrant the conclusion (i.e., educated guessing).

by Budman (not verified) :: Wed, 10/19/2005 - 2:41am


Very interesting approach. Anything statistical and innovative strikes my curiousity bone. This is very 'living on the edge', but it has worked so far for you for last three weeks. How many units are you up with this approach so far this year? I am also curious as to your dvoa plays this week as there seems to be more closer matchups.

by John Gach (not verified) :: Wed, 10/19/2005 - 3:59am

Re #56: Fair enough request from Budman (we two may be the last people bothering with this thread). I'm not sure I should post any choices publicly beforehand though -- one of my business consultants suggests that I'm being stupid writing publicly about the reasoning behind my approach, since I'm only increasing the likelihood that the arbitrage opportunities I'm taking advantage of will disappear. I checked my football wagers for the year yesterday and was astonished to find that I'm now something like 18/20 in calling winners -- a success rate that obviously cannot continue.

The two bad calls: 1) BAL over TEN in week two, before I knew that BAL was a deeply flawed team (even though Aaron Schatz predicted Baltimore's decline in both offensive & defensive performance before the season started -- would that I had been paying attention!). 2) PHI over DAL, because I made the mistake of believing the punditocracy's predictions: everyone (Sporting News, ESPN, NFL-Edge, etc. etc.) predicted a PHI victory, so I went with the opinion of the experts and included PHI in a 3-team parlay (both other teams won). Should have followed my own reasoning, which figured that it was a dangerous and risky game to bet on. The rule is simple: if I can't call a clear winner (meaning, of course, a team that should win and not necessarily the team that does win, given the vagaries of luck, official's calls [see BAL-DET], and so on), then I don't wager.

There are many factors involved in making such judgments. Relative DVOAs play an important role -- in fact without them as a basis for making inferences I'd probably be floating in the typical gambler's ocean of uncertainty. But one also has to take account of injuries (especially in positions that are important for the prior success of a particular team). One has to factor in home field advantage, which then has to be weighted -- so, for example, I expected STL to play fairly well at IND (for the Rams, that is) since they were playing in a dome. Without Bulger's injury the Rams may well have covered the spread. The Ravens, on the other hand, are terrible playing in a dome (now 1-9, I think) and always find some way to lose indoors. One has to account for what one might call "momentum:" teams that display signs that they are (or will be) better or worse than their current DVOA suggests.

Which leads to a comment about DVOA: it is a snapshot, a synchronic view of a dynamic system that, like most real-world systems, is constantly changing. If one views any week's DVOA rankings this way, then it's very much like describing a living language, which is changing into something else the very moment you write down your description. Just as any such linguistic description is, at the least, incomplete, and at the worst seriously wrong, so with DVOA. As with any statistical model, DVOA does not and cannot express the truth about the world of professional football, which is why Mr. Schatz is constantly refining it. Nope. It constructs an abstract model -- a kind of map -- that perforce must omit much (because by definition a model or map cannot have a one-to-one relationship to what is being modeled or mapped). DVOA helps one to comprehend fundamental relationships in a way similar (for me, anyway) to reading a drive-by-drive account of a football game, as opposed to watching the game.

There are many reasons for this (and for the similarity I'm suggesting): not least, the fact that any game is mostly full of redundancy and the human mind is evolutionarily built to factor out redundancy and to attend to the exceptional (e.g., movement instead of stasis). But often it is just the apparently redundant (e.g., the play of the left tackle in particular kinds of situations: see the two Ogden-Freeney analyses) that most of us don't attend to during a live contest that determines or helps to determine the final outcome. Instead, we attend to spectacular plays -- long touchdown runs and the like.

My preliminary sense after only a quick scan of the forthcoming games is that I may have a very light (possibly nonexistent) wagering week. Even before doing any detailed analysis, most of the games seem too close to call, from my point of view.

I'll certainly post any calls I make -- I just don't know yet whether I'll post them before Sunday.

Unrelated comment: One should look at certain kinds of grammatical errors in email messages a bit more charitably than did Mr. Schatz in a recent posting. Homonyms ("their" for "they're" or "our" for "are" or "to" for "too" for example) are commonly confused in email, much more so than in more formal writing. This is because writing email notes is a fusion of speaking and writing, which use different parts of the cortex. In speaking there is no difference between "to" and "too." What commonly happens in email notes is that the first visual image that pops into mind that corresponds to the sound of the word gets invoked and written down. One finds such errors even in the prose of educated writers, though a bit of editorial revision should eliminate most of them. I'd be surprised to learn that Mr. Schatz himself doesn't have to correct a few in his writing for FO.

by John Gach (not verified) :: Wed, 10/19/2005 - 4:10am

Apologies. Even with such a longish note I neglected to answer Budman's direct question. As of right now, I am up 92.9 units plus two long-term wagers of 10 units each calling division winners (IND looking good, NE likely, but PHI is worrying me). So, if you count those two wagers, I'm up 112.9 units in about 5 weeks. I'm not counting the lovely bonus given for depositing funds into my online gambling account, though perhaps I should. That would add 7.5 units to the total. Several of the parlay wagers (and one straight bet) involved the St. Louis/San Diego division playoff series, which I regarded as a lock for St. Louis. So a purist would have to subtract two or three wagers from my total, sinced they involved baseball.

by Jake S. (not verified) :: Wed, 10/19/2005 - 6:13am

Just a comment on why this might not be the best week to rate the DVOA system on. I might be violating some law of statistics, but I'm trying to help:

Underdogs went only 4-10 against the spread. Before this week, dawgs were 37-36-1 against the spread. Some variance is to be expected, but this seemed like an unusually easy week to pick games.

by Sean (not verified) :: Wed, 10/19/2005 - 6:13am

Just wanted to pop in and say the comments have been excellent in this thread, and have certainly gone well above and beyond the complexity of the article itself.

See, all I had to do was write an article with gambling implications and the lurkers come out of the weeds. :-)

by Jimmy Two Times (not verified) :: Wed, 10/19/2005 - 10:25am

I also wanted to re-appear and say that I'm keeping track of comments on this thread, with much interest. Play on.

My ATS picks of the week, using a modified DVOA system:

1. Was -12.5 vs. SF
2. Chi PK vs. Bal
3. Cin PK vs. Pit
4. Cle -2.5 vs. Det
5. Mia -2 vs. KC

Over/Under plays:
1. Dal/Sea OVER 45.5
2. SD/Phi OVER 47

Now, eliminating based on common sense: Cincinnati can't stop the run, and Pitt's DVOA is probably skewed down because of last week's game. If Roethlisberger plays, I don't bet that one. Chicago and Baltimore are both kind of flaky teams (high DVOA variance), so that one has a lower confidence interval. I'd take the remaining 5 plays, though:

1. Was -12.5 vs. SF
2. Cle -2.5 vs. Det
3. Mia -2 vs. KC
4. Dal/Sea OVER 45.5
5. SD/Phi OVER 47


by bethehouse (not verified) :: Wed, 10/19/2005 - 11:28am

I've always wondered at what point do you throw out the preseason DVOA rankings in the analysis? For instance, KC @ MIA, before the season started, the DVOA spread would be approximately KC giving 5. If you throw that out completely though, then I agree that MIA is a good play based on current DVOA rankings. But I think it must be factored into the analysis somewhere.

by John Gach (not verified) :: Wed, 10/19/2005 - 4:28pm

Springing from the weeds once again for a few preliminary comments. 1) Sean McCormick is too modest. His was an extraordinarily interesting and incisive account of the predictive powers of DVOA, from which I, and clearly many others, learned a great deal. No surprise that it has elicited a number of worthwhile responses. His essay and several early notes (especially Reno's) forced me to think about what I was actually doing when wagering on football games.

2)I think Jake S. is right that last week (I assume he meant last week) was easy picking. It's why I ended up with 6 wagers and really could & probably should have had several more. Few weeks have choices that are so clear-cut. It was like a license to print money. I mean, a 22% return for betting on SEA to win @ SEA against HOU? Roughly an annualized return of 11,000%.'Twas an early Xmas gift. My mistake was only to wager 40 units instead of 400 or 4,000. I'm curious whether another such opportunity will present itself this year. I'd also be curious for one of our statistical mavens to run a few hundred simulations of that game to see how many times HOU wins in, say, 500 tries. Not many.

3) 'Tis different this week. I don't like much of what's available. Impressionalistically (i.e., these are not picks but at this point mere inclinations sans any research) I tend to like CLE at home against DET straight up; ATL at home straight up against NYJ; maybe the SEA-DAL over at 46 (where I'm looking); IND and 16 at HOU is a siren call that tempts but will probably be resisted. For many reasons BAL @ CHI is a trap game that for me is completely unpredictable. Even the enticing under @ 30.5 I'll shun, since both defenses can score and create short fields. The real action should be on whether the two defenses will create enough exciting plays for more than 30 seconds of lowlights on NFL Prime Time.

4) Not a worthwhile parlay in sight. Not surprisingly the only two locks are such locks that you can't bet the money line on IND @ HOU or SF @ WAS. It looks like a quiet, possibly wagerless, week for me.

by Astro Boy (not verified) :: Fri, 10/21/2005 - 10:28pm

I was going to bore everyone with my week 7 picks, but a clever glitch seems to have thwarted me: the paragraph breaks got taken out, and my discrete game capsules got turned into a single, impenetrable stream-of-consciousness-style paragraph.

I just tried to start a new paragraph here, but it's still not working. Is this going to be a consistent problem with the posts after the system update? If so, Carl is going to have to go back to individual posts for each point in his weekly TMQ rebuttal.

by Astro Boy (not verified) :: Fri, 10/21/2005 - 10:31pm

Or maybe the paragraph breaks do show up after all. For now at least, the previews don't match the posts.

Is this a dead thread, and picks are posted elsewhere?

by Astro Boy (not verified) :: Sat, 10/22/2005 - 12:18pm

Triple post on a dead thread. Well, if my Week 7 picks suck, at least fewer people will see them. After all, these picks are coming straight from the Brandon Lang Tuesday Express to Ruin. What I mean by that is if you’ve seen “Two for the Money,� you’ll know that the main character’s picks went in the toilet when he started making them in the space of 5 minutes on a Tuesday. Well, I saw that movie on Tuesday, came home, set my own lines for Week 7 games in the space of about 5 minutes, looked up the actual numbers, and had most of my bets for the week in soon after. Tuesday is probably the worst time to make NFL bets because you’ve lost any early edge on the lines to the action that came in on Monday, and of course you’re betting too early for unreported injuries, sudden weather changes, etc. So, I totally understand if anyone wants to go contrarian on me and bet the opposite way in the games below. In fact, several lines have moved as of this posting, all to the benefit of the contrarian. If you go that way, please extend the courtesy of posting your intent beforehand rather than gloating after the fact.

In spite of Tuesday’s dangers, I like to get some early action on the games because it does set up middling opportunities in games that do have significant late movement in their lines. It’s all about positioning. I did my homework the rest of the week, and I still like these picks. The main reason for that is simple: there’s nothing exotic about these picks. 7 out of 8 are home teams, with the steepest line at -3.5. Sure, a lot of the home teams are weak, but their opponents aren’t so hot themselves, and many of those teams have stunk on the road. I like to bet teasers, but I didn’t see a lot of value getting created by teasing these lines, so there’s nothing exotic about my style of play this week, either. Like a lot of people from the sound of it, I had a good week last week, and I can afford to take a couple of risks this week. These picks could go 3-5, and even if only my weakest three plays come through, I’ll still be in decent shape for the season. As always, though, you’ve got to be physically and mentally prepared for a big loss or wipeout.

Early games:

Philly -3.5 over San Diego: I don’t love this pick, but I like it enough because I think the timing of this game is at least as important as the stats of these two teams. You’ve got an Andy Reid team coming off a bye playing a team facing a second consecutive road game and a third consecutive game against a rested team. I like going against teams playing a second consecutive road game, particularly if they’ve won the first game. The initial win tends to get the team overvalued in its second game, and here I think that overvaluing is just enough to counter the tougher line you’ll get on Philly with Reid’s proven success coming off a bye. On top of that, SD’s second game is a non-conference game after a big division road win, so there’s potential for a San Diego let down. DVOA totals give the nod to SD, but so much of that difference is due to special teams. Yes, I know Philly has deeper problems on special teams, but at least Philly now has a kicker with more than one hamstring. Tomlinson probably will get at least another TD, but no matter how fast and how often he scores, I think a rested McNabb against SD’s secondary can hang with them. If SD has to play catch-up, Philly’s secondary should perform better than they did in Dallas. The hook always scares me, so this is a 3 unit play.

Cincy (pick) over Pittsburgh. Basically, I’m taking Cincy every week. I predicted an 11-5 record for Cincy, but I didn’t have the guts to bet them over 8 wins on the season. One reason I don’t regret that cowardice too much is that as long as weekly lines undervalue Cincy, I can make more each week on Cincy than I could on one season-long bet. Seriously, this Cincy team is this year’s equivalent of the 98 Falcons or the 99 Rams, suddenly good teams that have been so bad in the past that it took most of the season for the lines to catch up to them. Already Cincy is 4-1-1 ATS this season. The one ATS loss (to Houston, at home, no less) didn’t hurt me too badly because my teased bets still covered that day, and in the case of the pushed +3 against Jax, I was able to take advantage of a temporary line move to a winning +4. Those close calls actually help by keeping the Cincy lines much cheaper than they would be if Cincy had the exact same stats and were a more established team. Given Pittsburgh’s injuries, I had this line at Cincy -3.5. 7 unit pick

Cleveland -2.5 over Detroit; Chicago (pick) over Baltimore (late game). The fun thing about gambling is that the garbage games can mean just as much as the must-see contests, and they can be much more lucrative. I like both home teams, and I’m not going to make things more complicated than that. TMQ’s generic “home team wins� prediction for all games has a 62.5% success rate this year, and the spread basically doesn’t come into play here. Both are 4 unit picks.

Other late games:

Seattle -3 -125 over Dallas. Last week’s pick of Seattle -9 over Houston was probably one of the strongest plays I’ll ever have. This week’s pick isn’t nearly as attractive, but it’s attractive enough. As a Bills fan, I think I can tell when the wheels will fall off for Bledsoe, and no, it’s not a good siren you’re hearing now that he doesn’t have a running game or a tackle who can pass protect. With the steep vig, this is basically a risk 3.5 to win 3 unit play.

Arizona -3 -120 over Tennessee. Simply put, this game is ugly, and if the line moves any more, I just might hedge off it. I wanted to take advantage of Arizona’s receivers against Tennessee’s wretched secondary, a matchup that makes Arizona the favorite. But by how much? I set my own lines before checking out the actual numbers, and for this one, I thought the absolute most Arizona could lay would be 3. Surely enough, the line was an expensive 3 when I checked it out, and now you have to lay the hook, or even as much as 4.5. I can’t imagine laying the hook with Arizona. They’ve already choked in winnable home games against St. Louis and Carolina, although in both games, I liked the visiting team going in. I thought Tennessee would be a bad, overachieving team this year, with just enough solid veteran play and coaching to steal them a few wins against more talented foes. But their secondary is just so awful, and it’s the veteran guys who are causing penalties and turnovers. As it stands, this is also a risk 3.5 to win 3 unit play, but that could change.

NY Giants -1.5 over Denver. After this week, there probably won’t be such a discrepancy between Denver’s perceived strength and DVOA strength. Denver’s dominating home wins tend to leave them overvalued on the road, and they often stumble on the East coast, as their meltdown in Miami will attest. Denver fans point to the Jacksonville game, but I don’t see that game as impressive a performance as some believe. Denver ran the ball well for two drives and dominated the clock. (For the record, I won that game as well, when I put up one of my favorite plays: teasing a +4 line to +10.5. Sure, the money line would have been a world more lucrative, but a line over +10 can bail you out of a lot of trouble). Bill Simmons did nail one thing in his losing pick of New England over Denver last week: Shanahan is absolutely terrified of putting the ball in Plummer’s hands, and he’s doing everything he can to win games on the ground. Unfortunately for that plan, the Giants’ weakness is their pass defense. Plummer made two big pass plays at home against New England, but has anyone noticed that NE’s opposing QB rating is 102.2? That’s a bit higher than the 56.2 they posted in 2003 or the 75.3 their depleted secondary managed last year. The Giants’ secondary gives up a ton of yards, but they do get turnovers. If Eli plays well enough to set a fast pace to this game, Plummer’s mistakes should catch up to him. Currently this is a 4 unit pick, but if I have success in the early games, I’ll up my stake here as long as the line stays south of 3.

Buffalo +3 over Oakland. This is my only road pick of the week. At the start of the season, I saw Oakland as a talented team with absolutely no leadership. They had the firepower on offense to win some games, but once things went badly for them, I couldn’t see them making any kind of significant rally. Looking at their schedule, you’d figure they’d get a full dose of adversity with a Week 1 road game in New England, and that’s basically how their season has gone down. Now, with Moss and Curry out, they’re much less imposing on offense. Yeah, it’s true that Jordan can probably move the ball against the Bills’ pathetic run defense, but Collins probably won’t contribute much more than one or two big plays to Porter. Oakland isn’t far from giving up on the season, and the Bills at least have some momentum. This is also an opportunity for Mularkey to atone for one of last year’s three pathetic losses (Jax, Oak, and Pit) that each kept the Bills out of the playoffs last season. Of course Mularkey atoned for last season’s 1-4 start with a 1-3 start this year. But like I said, the Bills have some momentum now and at least some competent play at the QB position. If the Bills are guilty of looking ahead to New England next week, then this game is still an excellent chance to go into next week a half game up. A half a game up, that is, before they either get blown out next week or give the game away at the end. Lousy Bills. 3 units ATS, 1 unit money line (+130).

by John Gach (not verified) :: Sat, 10/22/2005 - 2:11pm

Terrific, informative post from Astro Boy, which stirs me to report, as I sort of almost promised to do, what I've wagered for week 7. Firstly, after almost skipping the week, I decided to put some money into play, but not a lot. Here they are:

1) Seattle to win the division: 16 units with a payoff of 3.2 units. Couldn't resist, since who else is going to win the NFC West? St. Louis with their 27th ranked DVOA? 20% return in about 90 days, an annualized return of @ 80%. Not bad. [Speaking of returns, an extra zero somehow crept into an earlier note transforming an 1100% return into 11,000%. Sorry].

2) Parlay: Seattle to win the division + CLE and ATL straight to win. 5 units with a return of 8.3 units profit.

3) A crazy throwaway 6-parlay: SEA to win the division; IND to win by more than 16; WAS to win by more than 13; the over @ 47 for SD/PHI; CLE and ATL to win straight up. This, of course, is a loony wager with little expectation of realization. But it's a 1 unit wager with a possible return of 17.94 units. So, essentially no risk with a very high return should the Football gods really be smiling on me this week.

4) So far that's it. I am, however, tempted by SEA to win in SEA and may yet place a bet on it (probably in a small parlay to up the return). What scares me about it is Dallas's not very predictable performance record. That opening day win on the road against a very good SD team is a big red flag. On the other hand, as Astro Boy pointed out, no running game and the missing left tackle doesn't augur well for Bledsoe. As everyone on this list knows, Bledsoe sans pass protection ... is Bledsoe in Buffalo.

5) I'm avoiding all the sexy contests, since I can't call any of them. As Astro Boy wrote, there's often more opportunity in wagering on unsexy, low-interest games. I should add that I like CHI over BAL straight up but, as a Ravens fan, can't do it and probably wouldn't anyway, since I regard the Ravens right now as completely unpredictable. This is a team with a lot of talent (albeit with some of it sliding over the hill towards retirement). As I've analyzed the team's performance, its two major problems so far have been a completely ineffective offensive line and Jamal Lewis's inability to run the ball, suggesting that he really hasn't yet recovered from his ankle injury. As a consequence of offensive ineptitude, the defense has been on the field too often and too long. There aren't many options for improving offensive line performance, but already last week Chester Taylor had more touches running the ball. Those who have followed the Ravens know that Taylor is potentially a Priest Holmes in waiting. Smartest thing the Ravens front office did in off season was matching Cleveland's tender of $3 million to Taylor to get him back. On the other hand, the Ravens's passing game actually shows some promise. Heap and Mason are dangerous receivers. Whether the Ravens can, however, either pass or run against the superb Chicago defense ...

by Astro Boy (not verified) :: Sat, 10/22/2005 - 3:33pm


Thanks a lot for responding. I know all too well how those wagerless weekends go. Week 5 should have been a wagerless weekend with all the tossup contests, but I took the plunge and lived to regret it.

You seem to like parlays a lot, so I thought I’d offer some perspective on how you can use them. Forgive me if I’m preaching to the choir.

Parlays get rightfully bashed because the payouts don’t match the steeper odds. If you compound three successive 11-10 bets, your payout should be about 7-1, but in a three team parlay, you’ll only get paid 6-1 at most houses. That’s money that goes out of your pocket and back on the house’s pile. Parlays do have advantages, as you pointed out: an almost inconsequential risk can get translated into a significant payout. The other advantage is that you can spread the bets out over time and create hedging opportunities.

Take your throwaway wager that pays about 18-1. Suppose all the plays on Sunday pull through for you. The only thing between you and a payout is a Seattle division win. As you noted, this is an expected result. But it would be a shame if you lost your parlay here having made it through the hard part. What you can do is hedge, guaranteeing yourself at least some kind of payout, and set up the possibility for a middle. Suppose Seattle’s Jan 1 game at Green Bay is the last thing between them and the division title; if they lose, the division goes to another team. But Seattle is favored to win; Green Bay has been eliminated from the playoffs, and Favre is unable to play. So Seattle is favored by 7. What you can do is bet 10 units on Green Bay against the spread. That way, if Green Bay wins outright, you’re up 9 units, having lost the one unit on the parlay and won 10 with GB. If Seattle covers, you lose the 10 unit ATS bet, but the parlay gives you a net gain of 8 units. Granted, it’s tough to do that, but you’ve got a nice best case scenario: Seattle wins but fails to cover. That way you win the 18 unit parlay and the 10 unit ATS bet. You get a 28 unit return, and your risk never exceeded the single unit. Pretty good, no? Most gamblers live for moments like that, so again, I’m sorry if I’m preaching to the choir.

I agree with you on the Seattle over Houston money line bet. As I noted, the spread was way out of whack in that game, and that meant the money line was off as well. As much as I loved that game, the way I played things last week left me with a higher risk total on Seattle than I was comfortable with for a single game, and I had to hedge off some of that risk. The cheap way to do that is with the money line on the underdog, and the fact that Houston was only +330 on the money line seemed like a cruel joke.

by budman (not verified) :: Mon, 10/24/2005 - 3:53am

Sharp bettors took a bath this week in the NFL and CFB. Meaning the analytical types like us. Public sentiment ruled the day. Two big public favorites ended up pushing (Phil and SEA) if you got in on them before the crowd (sharps). But those pushes hurt when they are needed to be wins. Alot of big NFL bettors I know got pummeled today. Including most in the Handicapping contests who are considered ELITE.

The more I try to convince myself that I can outsmart the books if I keep trying, or if I can find that one thing they overlooked, the more I think I realize this really is a losing proposition in the long run except for the few ELITE who are ecstatic with a 55% winning percentage equating to small returns for the MANY hours put in. I am starting to believe Random Walkness is prevalent in sports gambling. I have just been on the wrong side of luck lately. I can't list all the 'beats' I have seemed to have taken lately.

ALL those Random events that occur that effect the outcome Against the Spread ventually even out in long run, but then you have paid vig. We are fools blinded by good luck when winning and when it regresses (usually hard) it blinds us some more. Maybe it isn't luck. Maybe the oddsmakers really do have knowledge way beyond what the public can gather. I am sure they can get gameplans and such that make huge difference. Maybe the oddsmakers DO take sides and lure the public to their choosing by phycological manipulation at every chance they get, Then just split the action on the rest? I read a study that proved they took sides. I also read some that the sports betting markets are efficient. Both done by major colleges. So I don't know who to believe. But my wallet continually tells me more in favor of an inneficient world we try to reason otherwise. And the fact that I know of NO other winning bettors in MANY years here in Vegas, meeting some pretty smart people, tells me even louder.

DVOA is just another ranking system that is most likely already factored into the line. So whatever past success it MIGHT have had (we don't know unless it was done in real time, not backtestes) it most likely was again randomness and/or small sample size. Not to mention if someone DID find an advantage it wouldn't be freely available on the internet.

Just my thoughts after many years of being around this "game".

(I will post those links if anyone interested or still reading this thread. Some very interesting stuff. I just have to dig them up again.)

by bethehouse (not verified) :: Mon, 10/24/2005 - 12:24pm

This was a tough week if you were using DVOA as a major tool in your picks. In a pick'em league I am in, I got 2 out of 13 right. Thats tough to do.

by Astro Boy (not verified) :: Mon, 10/24/2005 - 1:30pm

In the bottom of the ninth, with Lidge facing Podsednik, Tim McCarver said something to the effect that that Lidge had gotten over the “taste� of the home run to Pujols. One pitch later, Podsenik homered. I think I know how Tim McCarver felt.

Or maybe it’s just another delusion of grandeur for me to aspire to McCarver’s level of mediocrity.

Yeah, 2-5-1 ATS totally sucks, and no one goes far with results like that. It’s a comfort to see that I wasn’t the only one who got hosed. But the fact remains that proper money management can mitigate a lot of the damage in a crappy week and lock in success in a good one. Early bets can put you in a position to capitalize on line movement, and one thing I did right was to hedge off most of my risk on Seattle by taking Dallas +5 when that line moved. That turned a push into a win. Full disclosure is that I failed to hedge with two other lines that moved in my favor: I passed on Den +3 and SD +4.5, mistakes you can just throw on my pile. If you can find room. The point is that those extra hedging opportunities were there. Your ability to capitalize on those opportunities matters just as much as your picks when it comes to your ultimate success in playing the lines.

The money management aspect of gambling is an area in which I’m not a total idiot, but it appears to be beyond the scope of this discussion. It’s really cool that you’ve taken a look at the relationship between DVOA rankings and the lines. It’s certainly a tool that many people use when analyzing a week’s slate of games and coming up with opinions on them. As for those opinions, you probably don’t need many more ignorant ones coming in. On that note, I’ll sign off and get back to work.

by urge (not verified) :: Mon, 10/24/2005 - 2:50pm

"I read a study that proved they took sides. I also read some that the sports betting markets are efficient. Both done by major colleges...(I will post those links if anyone interested or still reading this thread. Some very interesting stuff. I just have to dig them up again.)"
I would be interested in reading them. Please, if you don't mind, look them up.

by John Gach (not verified) :: Mon, 10/24/2005 - 4:08pm

Posts #69-72 draw me in once again, despite not really having time for this. The comments are just too interesting not to throw a few more cents atop this week's loser's pile. Firstly, not unexpectedly, I had my first losing week since week 2. I knew (and wrote) that I should stay away from this week's games. I went with CLE & ATL in a 2-team parlay, even though I really didn't have sufficient confidence in CLE to warrant the wager. Like most of this week's games, I saw the probabilistic outcomes as muddy, anywhere within a range of losing or winning by a few points (or by some fortuitous, you-see-it-once-every-5-years occurrence [See PHI vs SD]). The only contest in the outcome of which I had/have any confidence is ATL at home against NYJ, upon which I've just plopped a 20 unit wager. Really, that's the only wager I should have made. Recall the mantra about what the road to hell is paved with.

Asto Boy comprehends my attitude about parlays exactly. But with an increased vig, one must be very very careful with them. Hyperambition spells guaranteed failure. Specific example, albeit mostly not football-related from the Patriots's miracle first Super Bowl of this decade (which also partly addresses budman's random walk commentary): Beginning with a parlay of the Patriots to beat the +14 point spread against the Rams combined with the UMD to beat the spread in a road game against UVA (basketball), I started a tremendous winning streak -- even though I'd now say I knew damn little about what I was doing. For six weeks I almost couldn't lose in 2, 3, and 4 game parlays on basketball games, both pro & college. To use numbers instead of units, starting with $50, all I had left in the online gambling account, and using only money in that account, I went in six weeks from $50 to $3500. I started fantasying, since this was so easy, about doing this full-time. Then came the decline. For the next month I couldn't win a wager. Not one, literally. Down went the account, just about to zero. I was betting on wild 6-team parlays, several of which nearly won (lucky they didn't, even though the payouts would have been huge, since I would have [see below] drawn exactly the wrong inference).

All along I had known that it was possible that, during the winning streak, I was having a fabulously lucky random walk (equivalent to correctly calling, say, 15 or 20 successive coin flips). Problem is, when you're having such a just-lucky walk, you can't _know_ that that's what's happening. Until, that is, I had the exactly corollary negative walk (that one right off the plank into the ocean with my hands tied behind my back). So, believe me, I know whereof budman writes.

Anent budman's remarks in #70, I do not regard the oddsmakers as enemies, but more as potential allies. As I see it, one has to identify opportunities. When they're just not there, then one really has to stay out of the action (the past week shoulda, coulda, didna). When they are there (e.g., the definitely too-low line on SEA vs. HOU in Seattle, with the consequent too-large payout for the money line wager), one has to pounce. In general I now favor few but large wagers over many small ones.

Astro Boy's brief comment in #72 about money management is right on target. In a way, it's the key to everything. This year I'm trying to make wagers only when I have a high degree of confidence in the outcome (a program not followed in week 2 & last week). There may be one game that qualifies, or none, or several. I'm also trying to wager amounts that reflect my degree of confidence in the outcome. Thus, I only wagered 5 units on CLE & ATL to win straight up. Not a disaster, but still not really a very good wager by my standards. Retrospectively, here's the multi-game parlay that I should have gone for: ATL to win, SEA to win division, WAS & IND both to beat the spread -- and no over/unders and no CLE to win. Why? ATL to beat the Jets & SEA to win division both strike me as very solid wagers with a high probability for success. SF & HOU are so awful that I'm inclined to bet against them a lot from here on, especially on the road. They are probably the only teams for which I'll break my rule about betting on the spread. Unless, of course, the spreads actually catch up with the two teams's futility. In fact, the most rational parlay wager would have excluded IND but left in WAS at home against SF. Any middling good team playing at home against SF is probably going to be a very good wager, even if you can't play the money line.

Regarding budman's comments in #71 (to return to them): Remember what initiated my participation in this discussion: Reno's comments (in #16) about wagering on the money line rather than the spread. Very very tough consistently to beat the spread. In order to do that you need to spend a massive amount of time (more than I can afford to take) and, as Astro Boy suggested, constantly keep your eye on moving lines. Figuring out who you think is going to win is hard enough. Consistently calling the winning margin is much tougher. So why take the harder and more likely losing route? Especially in contests that you expect to be close, too many random events can impinge and produce the "wrong" result. That's why I usually prefer lopsided contests, where the better team will probably win even if things go wrong.

Which leads to my next comment. Wagering on individual games is _much_ harder than wagering on multi-game series, which alas! don't exist in football. Over an extended 3 or 4 game series the better team almost always wins, though it may lose any one game. And so I did very very well in last year's NBA playoffs, especially in the early rounds, with 2- and 3-team parlays. My one, completely stupid error was wagering on Detroit to win its second home game against Indiana. I won every parlay on the teams but lost back 30 units on the one game -- still finishing up about 700 units during the stretch of the second NBA season.

That loss, which reduced my winnings substantially, illustrates how important one really dumb wager can be in its effect on one's final return. Here's where the money management really comes into play. Under no circumstances should I have made a bet that large on so uncertain an event as the outcome of a single game in the series -- shouldn't have wagered at all, but in any case it should at most have been a wager of 5 units, the loss of which would not have substantially reduced the total winnings.

So ... one has to weight probable outcomes -- for which in football, all praise and as many hosannas as one can muster for Aaron Schatz and FO, without which football would be seeing very little, if any, of my wagering. No doubt, as budman writes, FO's statistics or their equivalent are already factored into betting lines (though I think it's more likely that Sagarin's ratings are what get most used). If betting lines were only about assessing probable outcomes, then arbitrage opportunities probably wouldn't exist, since one must assume that this market is reasonably efficient, in an economic sense [1]. But they're not just that; more importantly they are an attempt to match wagers roughly evenly on possible outcomes, collecting a commission along the way. As in many forms of gambling, you pay the commission when you win in the form of a reduced payout (higher for parlays, as Astro Boy wrote). As I see it, one is _not_ wagering against the bookies, but against the expectations of other wagerers -- exactly as in the stock market (where, I hasten to add, I have no ability whatsoever to pick winners, partly because it bores me, whereas sports wagering intrigues and I'm actually interested in the outcomes). That must be why occasionally betting lines are wacky, as in SEA/HOU or BAL at only -3 in the Super Bowl against an unexceptional Giants team. I have no idea how & why the line on Seattle was so low but have a pretty good idea why the Ravens weren't favored by signifantly more against the Giants: small market vs. large market with lots more Giants fans wagering with their hearts rather than heads. All that money coming in on the Giants kept the line artificially low. Unfortunately, that was before I was wagering on sports contests, so I made no money from it. These days I would have placed a bundle on the Ravens. Anyway, that's why I suggested in an earlier note that a Bayesian analysis may be a better fit than a frequentist probabilistic approach. Think game theory, where you have to take account not just of the logic of your move, but the logic of your opponent's move as well.

[1] but note that in the short term markets can be quite inefficient.

Sorry to have gone on at such length, but it does seem that there are others interested in this discussion of the whys & hows of wagering on sports.

by John Gach (not verified) :: Mon, 10/24/2005 - 4:19pm

Sorry, I forgot to ask that I should very much like budman to post the links he mentioned in #70.

by John Gach (not verified) :: Mon, 10/24/2005 - 4:48pm

Forgot to note that the wagers I didn't make (one because it went against the team I root for) were SEA to win vs. DAL and CHI to win vs. BAL, both at home -- both came out.

Just in case anyone else might still be interested in the Ravens, let me add that 1) as I forget who wrote last week (MDS?), Chicago is the defense the Ravens think they are -- yep, we now got insurmountable evidence for it; 2a) Chester Taylor ran the ball twice for signficant gains; 2b) Jamal Lewis ran the ball 16 times for about 32 yards; 3) why?

It is clear that, at least right now, Jamal Lewis has zero acceleration. The offensive line does have problems, but I tried to watch what the line was doing fairly carefully and concluded that there are holes, they just aren't there by the time Lewis gets to them and he can't change direction and accelerate as he used to do. With no running game at all, Baltimore's passing game, not exactly Indy's or Seattle's to begin with, had little chance to be effective, even though once again it showed some promise against a very good pass defense. So the question is, why did Billick (or Jim Fassel) not give Taylor more carries? Right now this team has no power running game (except for 3rd or 4th and 1), so why keep pretending that it does?

And this is not the Super Bowl Ravens defense. DVOA looks to have it just about right at high second tier or bottom of the top tier. The defense is good enough to win games with some help from the offense & special teams, not good enough to dominate and win games by itself à la the SB Ravens. There is still a lot of talent on the defense (though Ray Lewis has slipped a lot as he heads for football old age), but it's now concentrated more in the defensive backfield. This team cannot stuff the run as the great Ravens defense did, against everybody, thus making every team they played one-dimensional. Most people probably don't remember that at the beginning of the season in which the Ravens won the SB, the Ravens's defensive backfield was highly suspect; in fact they didn't start performing really well until the second half of the season (and even then had a horrible last game against Vinny & the Jets -- probably the best thing that could have happened to the Ravens, since it cost them the #1 defensive spot according to the NFL's official statistics and they went into the playoffs really motivated to prove they were the best defense.

by Reno (not verified) :: Mon, 10/24/2005 - 10:29pm

I didn't get hosed at all this week. It wasn't sterling, but (so far), it's been a winner. If Atlanta loses tonight, I'll take a significant (but acceptable) loss for the week.

I've followed the conversation but haven't had time to sit down and write a post touching on all the fascinating things said. It's been very illuminating for me to read other bettors' wagering strategies (most especially John Gach's).

I also wasn't crazy about the games this week. I was, however, able to get a money line on Washington at -800 on Tuesday. I normally wouldn't bet on such small odds, but I had Washington as an even bigger favorite than that. So I placed a large bet on Washington to win, and then parlays of WAS-ATL; Washington to win and cover; and finally of Washington to win, Washington to cover, and Atlanta to win.

That was the parlay action for the week for me. Then, as I do every week, I bet in a weighted manner my interpretation of DVOA on games I felt I had a strong read on. So far those teams have gone 7-2 for a modest gain, with everything essentially hinging on Atlanta tonight. I considered expanding that to a WAS-ATL-CLE parlay, but didn't at that point in the week due to uncertainty about Detroit's quarterback. Interestingly, every one of the nine teams I bet (including, ultimately, Cleveland) was a favorite.

My reading of DVOA vs. the spread is 6-3-1 (not counting tonight's game), and would be ahead of the money line bets, percentage wise.

I'm not writing this to gloat (I hate guys who do that), but merely to again point out that there myriad ways for different individuals to interpret the same data set, and that everyone obviously has different wagering strategies. Mine is both more and less conservative than John Gach's, which is really fascinating to me, and (as has those of others in this thread) has given me a lot to think about.

I'll continue posting money line vs. spread comparisons at the end of each week if people are interested. I'm simultaneously surprised and not that this thread has had such staying power.


by Masse (not verified) :: Tue, 10/25/2005 - 1:08am

Can someone please explain how DVOA can be used for predicting over/under?

by Reno (not verified) :: Tue, 10/25/2005 - 2:53am

I'd also love it if budman would post the links. I'm in complete and full agreement with John's general philosophy of not placing wagers that you aren't extremely confident in. That might sound funny coming from someone who bets ten or more games a week, but it should be understood that in doing so what I'm essentially doing is betting my confidence in DVOA's ability to pick winners over time, the same way some people invest in the S&P 500 as a way of demonstrating confidence in the stock market as a whole. It's also very possible that one day the bottom will fall out, but I like to think that I've won enough already that I'll be able to walk away in the black. Lastly I'll agree completely with John about the difficulty of placing sports wagers (or any other kind) dispassionately. It's hard to do, and should only be done if you can afford it (and by afford it, I mean afford to lose every bet).

I'm really interested to read budman's articles, the question of market efficiency in this case seems to be more one of perspective. I have no doubt that the market's extremely efficient for the casinos - it practically has to be, given its nature.

I also believe that the potential for arbitrage occasionally arises. DVOA provides such an opportunity (an opportunity that diminishes, by the way, each time someone from FOX clicks a link, even moreso each time someone reads this thread). The key will be a willingness to get off the train when it stops running (the market adjusts).

(Un)fortunately, sports betting is increasingly "hot" right now, especially among college-age men, and the vast majority of football bettors are not yet looking at this website and/or making good use of it. The percentage that do will undoubtedly rise over time, and I'm impressed enough with the broader FO project that it wouldn't suprise me at all if at some point in the near future GMs and talking heads are talking about DVOA the way their baseball counterparts now talk about OPS. This strikes me as being, on balance, a very good thing for football, and when it comes down to it, I'm a big football fan, and betting on football is more of a hobby and means I use of trying to understand the game better.

If you want to have a try at using DVOA when betting the O/U, there are plenty of suggestions that sound pretty good in comments up above. Personally, I've never paid any attention to it until recently, but this thread has gotten me thinking about it.

re: spread vs. money line

So, betting the money line this week, DVOA and I netted a 12.5% gain on an 8-2 record, taking my season totals to 60.9% and 32-12 over the past 4 weeks on just my straight win bets. The low return on the good record is due largely to a significant bet on Cleveland.

If I’d taken the same methodology and bet the spread accordingly, I would’ve posted a 31.4% gain on a 7-3-1 record. For the season, I’d be at -9.5% and 27-19-1. The season percentage doesn't correspond with my earlier post in this case due to an error I'd made calculating these bets I didn't make.

This was a down week for DVOA percentage-wise against the money line, though to call a positive return "down" is a testament to its strength. Again, I don't have a base of experience from which to comment on its line performance. It's worth noting that if I'd followed VarlosZ's lead and not begun "betting" on the spread until week 6, DVOA would've posted two very nice positive returns and be substantially in the black with a gorgeous 15-5-1 win percentage. My early and completely groundbreaking hypothesis is that DVOA can pick out which teams are good and which teams are bad at a relatively early point in the season, but takes a while to figure out just how good and bad they are relative one another.

Again, throughout the above, when I say “DVOA� what I really mean is “my interpretation and application of DVOA.�

I'll also say again that I’m almost completely positive that no one on the FO staff would ever encourage gambling, that there’s never any such thing as a “sure thing�, and that I don’t particularly endorse gambling very much either, especially if you don’t know what you’re doing, and most especially if you aren’t prepared to and able to lose every bet you make.

Please! Click my name if your wagering (on sports, or otherwise) is causing you problems.

Oh, and speaking of markets, football, better football statistics, etc., does anyone else play at PROTRADE? It would be fun to do an FO challenge.

by Reno (not verified) :: Tue, 10/25/2005 - 3:10am

Ah, when I post it comes in bunches.

Re: getting a Protrade thing together (it's free!), email me at icebergsoup@gmail.com. If you're interested but not joined yet, let me refer you! I was thinking a second half of the season thing, weeks 9-17.

by Reno (not verified) :: Tue, 10/25/2005 - 3:11am

Stupid periods getting included in the links. icebergsoup@gmail.com that is.

by budman (not verified) :: Tue, 10/25/2005 - 7:15am


Have you ever posted your methodology before in the past where I might have missed it? Or have you sort of kept it to yourself?

The problems I see with wagering on ML are the fact that often you are wagering on the favorites and in which you will be LAYING corresponding amounts of $$ to how strong the favorite is. Like your WASH wager where you risked 8:1! With as much parity as there seems to be in todays NFL (where anyone seems to be beating anyone) that would scare the crap out of me. A loss would take you eight wins to get back! Sort of like trading on margin. Sure the lieklyhood of the loss is high, it exists nonetheless. ESPECIALLY in a sport where injuries are the norm and can change the outcome on one play.

I would have been sick to my stomach yesterday had I had a ML wager on Indy and saw that halftime score! Now imagine that AND Manning out for 2nd half. I can see ML wagers being much less volitle in a sport with low injury occurance, like NBA.

It is also hard to find ML's available on favorites of 10+.

by John Gach (not verified) :: Tue, 10/25/2005 - 5:21pm

This has been an extraordinarily interesting thread. What's really intrigued me are the various discussions of underlying rationale (including my own, I hope), of the _reasons_ behind what and how we are wagering. I'm delighted that Reno jumped back in -- had he not, I was going to ask him to report on how he did this week. Clearly Reno and I share a highly similar, though not identical, attitude about how best to go about wagering on sporting events -- even to the extent that I, too, think that I am coming to understand football much better by wagering. Thing is, it is clear that with Reno and Astro Boy there are minds at work trying to figure things out. Just fascinating to read their descriptions.

On to my final report for the week -- a copy of Reno's. I lost my not too intelligent 3-parlay on SEA to win division + CLE + ATL @ 5 units. Lost my wild 6-parlay, but learned from the cheap 1 unit loss that I have no business wagering on over/unders, because I can't "feel" them the way I can feel the probable outcome of certain games. That may change, but for now I need to stay away from what I don't understand (though I can sometimes call over/unders in basketball). Yesterday I wagered 20 units on ATL to win -- close to a lock, in my opinion, as I'll explain presently, and won back enough to cover the two small losses and go slightly into the black for the week. Remaining is a 16 unit wager on SEA to win the division, which should return 20% profit in several months. All in all, not awful for a week that has a lot of people complaining. Had I seen the opportunity last Tuesday to wager on WAS to win @ -800 -- and I'll address the reason for that as well in a moment -- I would have placed a large bet of at least 50 units on it.

I suppose I should first explain my approach, for which in an earlier post I gave one of several possible abstract models (the Bayesian stuff). That's not what I actually do, but an attempt at an explanation of why what I do seems to work. Being a statistical nincompoop who at the applied level can't distinguish a chi square from a triple Salkow [?], I don't manipulate numbers. Instead I directly use the concepts that correspond to the mathematics. Thus, I mean it pretty literally when I write that I feel a probable outcome. It's definitely not a deductive process, nor is it, really, quite inductive. Rather, I steep myself for a few hours in the data (relative DVOAa, pundit opinions, Sporting News War Room's predictions, and a few similar sites), after which I check the available wagers, usually discarding most right away as ambiguous and not worth betting on. At that point I'm usually left with a small number of possible games to wager on. If I look at it and pretty much know who should win, then it's a possible.

Let's apply this to specific wagers, always useful.

1) I missed that Garcia might start for DET. Had I known that, I wouldn't have touched the game. Sure enough, Garcia ran for the winning touchdown. Harrington would more likely have thrown an interception run back for a touchdown.

2) SF @ WAS. A lock and even a rational wager on the spread @ -13. Yes, in theory there's parity (and often in practice, too). But, as I wrote in an earlier note, this is a year where there's money to be made betting against SF & HOU, especially on the road against middling decent teams like WAS. In the real world SF had so little chance to win that game that virtually _any_ money line wager was an extremely good bet. Of course you have to wager a lot to get any return. If it scares one to lay such large odds, then don't do it. But viewed strictly as a rational proposition, it makes a great deal of sense to me. This year presents unusual opportunities. I checked back to 2000 and both SF & HOU appear to be signficantly worse teams than any of the bottom teams of the prior 5 years. Playing Smith at quarterback, though in the long run possibly benign for the team, in the short run guarantees more of what happened in the WAS game. HOU is just totally hopeless with no offensive line at all and no ability to protect Carr, who may even be a decent quarterback if he's ever given any protection and a semblance of a running game (see Bledsoe; see Brunell). Speaking of WAS, I'm starting to become a believer. Gibbs has turned last year's sorry outfit into a good (not great) team that doesn't beat itself. You have to play consistently well to beat WAS. With the return of Arrington to the lineup, they may even start creating turnovers, which could make them a very dangerous team.

3) ATL at home against the Jets. A near lock. I only wagered 20 units because of money management principles -- I'm up a lot and didn't want to risk overly much, mostly I wanted to get back about even for the week (plus the long-term SEA wager, which I fully expect to win). Here's the reasoning on lockhood: A) ATL rarely loses at home in the dome. B) ATL's weakness is the defense -- but the Jets have no offense to speak of. C) There was little chance, even with one of their two main runners out, that the Jets were going to stifle ATL's running game. D) 41 year old Turnover Vinny was the QB. Testaverde can make great plays but throughout his career has had a penchant for brain-dead turnovers at the worst time (which is why the Ravens let him go to the Jets). Did this script play out or not? The only way I could see the Jets winning this game was if ATL handed it to them with turnovers. Sure enough, just to throw a scare into Reno & me, ATL decided late in the game to do their best impersonation of the Jets. It still wasn't nearly enough for the Jets to get very close. But the Jets could have scored a TD on their last drive, which would have killed bets on the spread. And so bettors like Reno and I generally avoid the spread -- too many factors go into the final score, factors that one has no way at all to figure into one's assessment of the outcome in advance. But right now, playing at home, ATL is a far superior team that should have and did win easily.

To address Budman's concern in #82: You're looking at it emotionally and saying, "what if I lose $800 trying to win $100" (or whatever the actual numbers would have been). Wrong approach. What you do -- and [chuckle] I'd bet that this is about what Reno does -- is rationally assess outcomes. I feel them, Reno may figure them statistically -- it's the conclusion that matters, not the particular means used to reach it. If you have high confidence in the conclusion, then it's worth a serious wager. Of course, you may be wrong -- and every now and then you _will_ lose one of these. Here's how I'd look at SF @ WAS at, say 8-1: Would SF likely win once in eight games. Nope. Would they likely win once in 10. Not likely. They might somehow figure a way to win once in 20 tries, though I wouldn't want to be the person wagering that they would do so. Thus, at 8-1 this is an extremely good wager. If you think, as Reno and I did, that a WAS win was about as certain as things can be in pro football, then you should make the wager. And yes, it is not easy to find money line wagers on big favorites. Even when they are available, they tend to disapper quickly. So on those one often has to act quickly, early in the week. Re: IND @ HOU with the big spread (16 last I looked, which is about as wide as it gets for NFL games). That spread would have worried me. Yes, IND is totally superior in every aspect of the game -- except that their special teams DVOA isn't very good, which suggests an occasional opportunity for even bad teams to get good field position or get a good return on a punt or kickoff (of which you'd expect there to be many). Budman must have been gasping for breath at half-time with a 14-14 score. Of course, in the second half Indy played like Indy and Houston like Houston and that was that.

Reno wrote: "My early and completely groundbreaking hypothesis is that DVOA can pick out which teams are good and which teams are bad at a relatively early point in the season, but takes a while to figure out just how good and bad they are relative one another." Intuitively, I think that Reno is correct. Assume it is and take it as an axiom. I think, then, that several corollary theorems naturally follow. 1) Narrow DVOA spreads, especially early in the year, don't mean much and are dangerous to use as predictive wagering tools. 2) As the season proceeds, DVOA becomes increasingly accurate in its relative rankings (which we pretty much already know to be true). 3a) When early DVOA identifies teams like SF & HOU as profoundly bad, serious wagering opportunities exist. 3b) When early DVOA identifies teams as profoundly good, ditto with 3a. 4) When early-identified profoundly good teams play each other, stay away. 4b) Ditto with early-identified profoundly bad teams. I would add that two years in a row now the preseason projected DVOAs have correctly identified a formerly bad team that was going to play superior football: SD last year and TB this year. Advance DVOA also predicted the Ravens's decline, though even the folks at FO didn't seem really to believe it. I sense opportunity here and next August will be keenly paying attention.

From the first game (IND vs. BAL) I figured IND was going to be very very good. After several more games, with their defense winning before the offense really even started to play well, I _knew_ they were a terrific team. Most impressive (after the defensive prowess) was Peyton Manning's ability to manage a completely different, almost Ravens-like kind of game, just taking what the opposing defense was giving. At that point I started thinking of the recent great Patriots teams that could win any kind of game, from offensive explosions to titanic low-scoring defensive struggles. DVOA doesn't have them quite at the top yet, but I suspect that it will. True, for such a good team they have a pitifully easy schedule -- but they still do have to play well enough to win the games. As budman pointed out, there is much parity in the league.

One last thought (for now, anyway): the issue of arbitrage potential, especially in relation to the given odds. If everyone wagered rationally, then there would probably be few opportunities, since the Sportsbook market would then be completely efficient and the odds (spread or money line) for any game or combined wager would correctly reflect the likeliest outcome. What I believe Reno and I are doing is identifying the cases where things are out of kilter -- in some cases (e.g., SF @ WAS) even being able to lay a money line bet at all is an arbitrage opportunity. I would claim, for the reasons detailed above, that ATL at home against the Jets was almost as solid a wager -- not quite as rock-solid because the Jets, bad as they are right now, aren't HOU or SF. Like Reno, I was a bit leery of publicly stating what it is that I do. But at this point probably about five people are following the thread and, besides, knowing what I do isn't the same as being able to do it. It took me a number of years (of losing, I should add) to acquire the discipline, to be able to view these contests dispassionately and objectively, to develop somewhat of a money management scheme (both to protect against egregious losses and to preserve gains), not to be greedy, and to lose budman's anxiety about risking and possibly losing relatively large amounts for fairly small anticipated returns. None of the above came easily. I still learn something every week that, as it were, refines my working model. It does seem to be working reasonably well for me. I have no idea if it would for anyone else (except, I suppose, Reno). Assuming that I continue doing well using the same method for the rest of the year (that is, that I'm not once again just taking a random walk), my strategy next year will be to escalate the size of the bets considerably (though this is not always possible to do, at least at the one venue I'm currently using, SportsInteraction). I also need to start up multiple accounts, for a number of reasons ranging from the possibility of arbitraging against myself at different venues to increasing a highly favorable wager to what I want to wager, when confronted with a limit at a particular venue (e.g., I actually wanted last week to wager 50 units on SEA to win its division but could only wager 16). Here's how I look at a wager such as SEA to win the division with a 20% return if successful. Firstly, the result is, in my opinion, highly likely, else I wouldn't make the wager. Secondly, that's an effective annual return of about 80%. Pretty good for a very low risk bet.

by budman (not verified) :: Tue, 10/25/2005 - 7:31pm

I would add that two years in a row now the preseason projected DVOAs have correctly identified a formerly bad team that was going to play superior football: SD last year and TB this year. Advance DVOA also predicted the Ravens’s decline, though even the folks at FO didn’t seem really to believe it. I sense opportunity here and next August will be keenly paying attention.

But it farted badly on ATL (which Aaron is about to write about) this year. Not to mention CIN, CAR, WASH and NE (unfair to count NE?).

by Reno (not verified) :: Wed, 10/26/2005 - 3:58am

I haven't posted my methodology, and won't post the specifics. Here are the basics, though:

On Tuesday, I use the new DVOA rankings to come up with likely winners for each game. My method is more complex than VarlosZ's suggested one at the beginning of the thread, but it's hardly "complex." The last math I had was eight years ago in AP Calculus, where I paid little attention and understood little of what I paid attention to. It takes me about an hour to run through the slate of games. I use a pen, paper, and calculator to do this. I translate the percentages into points, because points make more intuitive sense to me. This isn't an attempt to predict a final score, it's just easier for me to understand that Team A is about 7 points better than Team B than that it has a 32.5% advantage or whatever. I suppose a spreadsheet would save time in the long room, but I've never even thought of taking the time to set one up.

Next, I look at the early lines. If there's a situation like WAS-SF last week, I grab it.

Throughout the week, I watch the news items on ESPN's NFL site (mostly the daily roundups), the injury reports on the NFL site, and this site (most especially Under the Knife). This usually averages out to about another hour a day.

On Saturday, I sit down with the information I worked out on Tuesday and, considering the week's news, decide which games to bet on. I then bet in a weighted manner based on how confident DVOA and I are on the outcome. For example, last week I bet on Cincinnati to win, but I did so with a smaller wager than DVOA suggested. This was because of uncertainty about how much better Big Ben at less than 100% would be than Maddox, whether Ward would play, how the Bengals' interior offensive line was holding up, just how much the Jacksonville game affected DVOA's opinion of Pittsburgh, etc., etc. This actually wasn't enough uncertainty to keep me from betting (as was the case with, say, KC-MIA), but enough to reduce my confidence.

In the WAS-SF game, I pounced early in the week because WAS didn't have any major injuries and my numbers had them as being massive favorites, even more than the 88.9% the line I found gave them. Facing Smith also meant that they might actually get a turnover or two to boot.

When I first started doing this, the thought of placing a very large bet on 1:8 odds was completely nerve-wracking to me. Over time, I've come to trust what I believe to be the superiority of my method. Now, I trust the information I have, and the Washington bet wasn't something I thought twice about. Yes, there was the potential for a very significant loss - there's always the potential for a loss. But over the past year and change, I've found that the bets really do pay off. Yes, there's parity in the NFL, but it's more on a season to season basis. In what other major sport is it considered entirely normal for the best team to win 87.5% of its games, and the worst team often wins 6% or 15%? Even the very best and worst basketball teams don't come all that close on a yearly basis. Going 14-2 translates to winning almost 72 basketball games (I think the Bulls' record for most ever?), to win as infrequently as a 1-15 team would translate into 5 wins in an 82 game season (record low is 9).

Of course, that's not where DVOA comes in handy - even the most casual fan could've told you that Indianapolis should've beat the pants off Houston. The gambling advantage it provides is often when it comes to teams like San Diego or Carolina, who maybe aren't as bad or good as their records suggest.

Still - if you'd just bet on whoever was playing San Francisco last year to beat them, you would've gone 14-2, and as John points out, the opportunity is there again with both the 49ers and Texans. Betting any team at -800 is never fun, but when your information tells you its the right bet to make, it's the right bet to make. Provided you trust your information. And if you don't, why bet according to it?

This is also why I completely ignore any and all "picks" columns or sites (excepting Bill Simmons', which I read for entertainment value and where I often don't even pay attention to who he actually picks). Those guys aren't offering a damn thing I want, I think they're completely full of it and offering Theismann/Maguire level insight, often citing bizarre and meaningless "trends" they can get to say whatever they want them to (witness the recently linked Dr.Z article on the subject). I'll sometimes read Clayton's First and Ten and Pasquarelli's Tip Sheet (what that the whole thing were free again!) and Morning After because I enjoy them, but don't really factor them into my wagering. If I lived in Vegas, I would avoid other sports bettors like they had the plague.

If this were Moneyball and I were Billy Beane, the columnists and other bettors would be the scouts in the draft room, only held in more contempt. Because the columnists and bettors have proven themselves to be more spectacularly and consistently wrong than the scouts ever were. I want hard information - injuries, DVOA, adjusted line yards, roster moves, sack rates, DPAR, changes in coaching strategy - anything. But I try to keep my head largely clear of the buzz of conventional wisdom. Conventional wisdom as embodied in the lines is, after all, exactly what I'm trying to beat.

by John Gach (not verified) :: Wed, 10/26/2005 - 2:47pm

Re: Reno's #85. A resounding yes! You can probably also put on the trash pile along with the picks of the punditocracy most you-gotta-pay-for-'em picks. I've occasionally bought some of these short-term just to see what you get. Not impressed. My simple method, which is unsurprisingly very similar to Reno's except that I don't even bother to express relationships as points but just as team x will [maybe / probably / almost certainly] win (or, as in many cases, "too close to call"). Since I'm going to bet on the money line, that's really all I care about.

DVOAs are just indispensable for the analysis that I do. This week there are lots of good pickings, compared to last week's mostly dreadful choices. Already yesterday I placed a number of wagers, which I'll post sometime later in the week. I usually wait until later (and so missed the strong WAS wager), but this week's early pickings just seemed too good to pass on. GB @ CIN. TB @ SF. Even KC @ SD. ARI @ DAL. BAL @ PIT -- just for starters. What a wonderful menu from which to select delicacies. What I really have to decide is how seriously to take my "always bet against SF & HOU" mantra. It gets a bit of a test this week with CLE @ HOU. Since sooner or later HOU will probably win a game, why shouldn't it be this one? Luckily SF has already won a game.

And to think that next year's iteration of DVOA will probably significantly improve its accuracy. Is that a lovely thought or what?

Re: Budmans #84: What I have in mind is to pay attention both to the preseason DVOA predictions _and_ to the first several games. Where performance looks like it's matching the prediction, some early arbitrage possibilities could come into play.

Quick note about why I really like SD at home this week: they have a losing record again (yes, it took a miracle, but the miracle happened). Even with their brutal schedule this team is too good to keep a losing record. Only worry is KC's strong DVOA against the rush, so they may restrain LT's antics, though not likely to nearly the extent that Philly did. But to compensate there's KC's matador pass defense. Not certain but a pretty solid looking play to me.

by John Gach (not verified) :: Wed, 10/26/2005 - 3:00pm

Just a quick thought after rereading Reno's last note: just why are the "experts" generally so spectacularly wrong? Partly it may be that they have to pick winners (straight up and against the spread) for every team every week. If I had to do that, my record would probably be awful too, since like them I'd mostly be guessing. Last week, for example, I thought that almost all the games fell into my "too close to call" category, meaning they weren't (for me) bettable. Bettors such as Reno and I and Astro Boy can _choose_ which games to wager on and ignore the rest. So we select only favorable wagers and pay no attention at all to the other, mostly guaranteed money-losing opportunities.

by TBW (not verified) :: Wed, 10/26/2005 - 8:40pm

This has been a fascinating discussion. It prompted me to use DVOA to make my own predicitions last week which went 11-3 and 3-0 for my "locks". We'll see what happens this week. I want to see what happens for a while before making any actual wagers.

The thing that is exciting about this week is that the games with the biggest differentials between actual "Vegas" spread and my calculated spread are also the games with the biggest calculated spreads. In other words the teams with the best overall chances of winning according to DVOA are also the best bets in terms of overlays. People aren't giving Dallas, Cincinnti and Pittsburgh their props. I'm not as crazy about the TB-SF game. Obviously TB should win, but the Vegas spread seems about right, which means it is unlikely that the money line for that game would represent an overlay.

by TBW (not verified) :: Wed, 10/26/2005 - 9:45pm

As for why the experts are wrong so much, I think it is because they essentially are incented to go out on limbs. Think about it, they're not wagering real money, so when they loses, who cares ? The worst thing they can be is boring, because boring won't sell papers or magazines, or attract viewers and listeners. So, they walk a fine line between being sensible and titilatiing or controversial. They're always chasing that one big upset pick because that's something they can crow about, and if they're wrong no one is really paying attention anyway. So, my theory is that chasing upset picks(and having to pick too many games, not being able to say I don't have any picks this weekend) drives down their rate of success.

by John Gach (not verified) :: Thu, 10/27/2005 - 1:02am

Re: TBW's #88: yes, exactly. There are lots of terrific picks this week, unusually many, I'd say.

#89: Sounds right. I like boring when wagering, whereas them gettin' paid to pronounce picks probably do need to stir up some excitement and controversy.

CIN @ home against GB is my lock of the week, around which I've built several parlays. You beat Cincy by running ... but GB, now having no running backs, can't run against St. Maria Goretti. With one of the league's best pass defenses, Favre is likely in for a long day. Short of Cincy handing them the game with, say, six turnovers and just about no take-aways, just how can GB win this game on the road? I understand that technically it is possible for any team to win "on any given Sunday" ... but as both a practical and theoretical matter it's a matter of probabilities and specific mattchups of strengths and weaknesses. Looks to me as though GB has actually been granted some power credit for dominating woeful MIN most of the game, while CIN has been given multiple demerits for losing at home to a powerful team that matches up perfectly against their weaknesses -- sort of like SD at NE earlier in the year.

Regarding TB @ SF: the issue for me is simply, will TB almost certainly win? I think so. Wouldn't surprise me if it's closer than the spread (but that never surprises me); also wouldn't surprise for TB to win by 24. The key for me is the impossible matchup of SF's mostly nonexistent offense against one of the league's best DVOA defenses. Again, other than TB mistakes or an exceptional special teams play, just how is SF going to score? Hard to win without scoring points, unless you're Chicago. The TB defense alone may score enough to win this game (or create enough easy scoring opportunities, which is about the same thing). Of course, there's always the dread possibility that TB might figure that they just have to show up to win, in which case I'm dead in the water. But I don't think so for several reasons: good motivator in Gruden; and the team needs to keep winning in order to stay in the hunt to win the division, which should be sufficient motivation to play at least well enough to win. After all they don't have to bring their A-game to win.

Speaking of Pittsburgh, I notice that the line has moved since I placed my wagers yesterday, when PIT was favored by 9 or 9.5. TB has moved, too. Somewhat surprisingly CIN has so far stayed the same.

Since I'm going to be away Fri & Sat, perhaps I should reveal my wagers for this week. Here goes (all wagers are money line):

1) CIN & TB 15 units, return of 21.3.

2) CIN DAL SD TB PIT, 5 units, 14.8

3) PIT TB CIN 20, 34.4

4) CIN PIT 10, 14.9

6) CIN 30, 36.8

If necessary, I could give the reasoning behind them, though, to be honest, the picks seem pretty obvious to me. The only rule I violated was wagering against the Ravens, which I don't like to do and usually do not do (does make rooting for them problematic). But just because their season is over, why should mine be? Seems to me I should get some reward for their woefulness this year. I certainly wouldn't wager on them in any game that they actually have a chance to win (will there be another one this year?).

There are other temptations, which I still may act on: NE at home against BUF and JAX @ STL (but for some reason as of right now no money line available where I'm looking). At -3 JAX doesn't look crazy and STL is not so invincible at home this year. But I haven't checked DVOA matchups for either.

Out of curiosity, TBW, which three games were your locks for last week?

by TBW (not verified) :: Thu, 10/27/2005 - 3:23am

I think TB will too, but I'm just saying I think the odds are probably fair. What I want to find is a situation where I believe a team has 95% chance of winning, but the money line says the public thinks they only have a 75% chance of winning. In that case the reward outweighs the risk. Because I believe the spread on the TB-SF is right, I'm assuming the money line is probably priced right too.

My locks last week were Washington, Indy, and Atlanta. I had the Washington spread at over 19 points by my calc, the others were in the neighborhood of 11 points. I call it a lock if I think the spread should be 10+ points. I wouldn't necessarily bet every lock though. Last week I probably would have shied away from Indy as they were heading into a bye week. I would have stuck with Atlanta though since the Jets were also heading into a bye.

by John Gach (not verified) :: Thu, 10/27/2005 - 1:42pm

Re: #91. I actually don't agree that the early TB/SF line was quite right, evidence for which is that the spread has now increased. This raises the different issue of early vs. later in the week wagering, since the lines and returns sometimes change quite a bit in the course of a few days, adjusting to the actual betting patterns. Currently I think that it may often be more advantageous to wager early -- if and only if one has high confidence in the outcomes. On the other hand, if a lot of money is being stupidly wagered, then the opposite can also be true as lines change in a favorable direction for rational bettors. Recall my earlier remarks that one is not playing against the oddsmakers but against the opinions of others. What we're all looking for are the cases where those opinions in the aggregate are wrong (such as SEA/HOU), since such occasions present huge opportunities. In between fair bets are fine, so long as one chooses carefully.

I also pay attention to patterns, about which there's a lot to say so I'll just gently broach the subject here. Think of a team's season-long performance as similar to a person's development across a lifetime. At the beginning there is all potential -- albeit potential constrained by the existential facts of biological inheritance and social setting into which the person has been born. Just so with teams, where the constraints are the actual players, coaches, cap money available, and schemes of play. With each game played potential diminishes with some possibilities being forever precluded (can't go undefeated as soon as you lose your first game). Sometimes I conceive of this potential/realization dyad as a kind of wave function collapsing into a single actualized state, which selects one of indefinitely many possible states. Other times I conceive of it as an emerging form, quite inchoate and indistinct in the beginning, that gets clearer and more obvious each week. A number of corollaries follow from this way of looking at it, perhaps the most
interesting being that individual games are not quite independent events but are entangled probabilistically with the team's past and future games of the same year as well as with the performance of other teams. From this matrix emerge pop notions such as "momentum."

For all teams there is a high degree of uncertainty at the beginning of the year -- thus the wildly fluctuating early DVOAs. Our beliefs about the likely success of teams is conditioned by their performance in the prior year (as is early DVOA). These beliefs are likely to be wildly wrong, since there just isn't enough information available yet for making rational judgments. For most teams there is considerable uncertainty about particular games. But for some teams there is vastly less uncertainty both about individual games and about their future (although random and completely unexpected events, like a season-ending injury to Peyton Manning or Tom Brady or [we now know for sure] Ben Roethlisberger, drastically alters things and immediately changes all the probabilities about future performance, eliminating vast numbers of possible futures as no longer realizable). I believe that DVOA describes reasonably well these interlocked probabilities. Clearly it's not yet accounting for everything important that needs to be accounted for (as the persistent outlier problem suggests), but on the whole it's pretty good and will get even better.

To bring this stuff down to earth somewhat: the several paragraphs above illustrate why I think SD is a very strong favorite to win Sunday at home against KC. That conclusion is not just based on this week's DVOAs but on conceiving of SD's season as a single form. This team has a strong likelihood of achieving a winning record (as the Pythagorean projected wins keep telling us). So look at their schedule: future road games at NYJ, WAS, IND, KC -- three of which they definitely could lose. Call those losses and their record is 4-7 (counting NYJ as a definite win). Their future home schedule is KC, BUF, OAK, MIA, DEN. None of these home games looks like a real gimme, though one would expect them, as things currently stand, to win against BUF, OAK, & MIA. If they win out at home, they're minimally 9-7, possibly 10-6 or better, which certainly accords with their DVOA strength, even with their difficult schedule. If they lose to KC and win the rest, then their home record would be 5-3, possibly 4-4 with a loss to DEN. Win against KC and we're probably looking at 5-3 or 6-2, the latter making the most sense. So, to have much chance of achieving their potential, they really need to and should win the KC game. If one performs a similar analysis of KC's schedule, one will find that this game pops up as a highly probable road loss -- KC has played exactly one good team on the road, DEN, and was crushed. Most worryingly, from a KC point of view, DEN gained 221 yards rushing.

Now, one might reasonably ask, "if you feel so strongly that SD will probably win, how come you only have a SD win as part of your least likely to win wager (a 5-team parlay)?" The answer is that I believe that there are even stronger bets available, so I went with them. Every week presents different opportunities and the pool of money assigned for this silliness is finite. If this week were like last week, I might be wagering a fair amount on SD.

Regarding the locks, IND & WAS were obvious (though IND wasn't a lock against the big spread), I was curious whether the third choice accorded with mine, and it did (ATL).

by John Gach (not verified) :: Thu, 10/27/2005 - 1:54pm

A quick correcton to the note I just sent: DVOA doesn't describe interlocked probabilities, it represents them as statistically based relationships. Describing them, albeit far too simplistically, is what I've just done. A team's DVOA takes into account the performance of all the other teams and measures it using a yardstick of hypothesized average performance; ditto for any individual player or team subgroup (Oline, DBs, etc.). What it doesn't do, so far as I can see, is account for the kinds of probabilistic entanglements I alluded to in my earlier note.

by Wes M (not verified) :: Thu, 10/27/2005 - 3:20pm

While I don't have anything substantive to say, (I don't place wagers on sporting events,) I want to commend the four/five folk who continue to post in this thread.

Fantastic stuff - the only "archived" thread I keep up on.

by John Gach (not verified) :: Thu, 10/27/2005 - 4:51pm

Re: Wes M's comment in #94. Though centered on DVOA and wagering, I too think that many of the notes in this thread have implications for stuff far beyond mere gambling. And why not? The history of probability theory _began_ with attempts to understand gambling (Cardan, Pascal), i.e., with attempts to predict uncertain future events. Things in this area have come a long way in 400 years; nonetheless, there are still many fundamental unresolved issues (for one: exactly what randomness means; for another: are stochastic systems inherently unpredictable). Many of the same issues percolate through the sciences, while entire sciences (e.g., thermodynamics) are founded on them.

So the crazy longevity of this thread does begin to make some sense. There are reasons for being interested in in that transcend an interest in gambling. Who doesn't have some interest in strategies for predicting future events?

by TBW (not verified) :: Thu, 10/27/2005 - 5:13pm

Re: 92

I don't think a fair bet is good enough. By definition it's break even. If I think that TB has a 90% chance beating SF, but I have to wager 900 to win 100, then I'm just going to break even over time. Now, if I believe that TB will win 95% of the time then I can make some money. TB will win 19 times, for $1,900 in winnings, but the 20th game I'll lose my $900 "investment". I still net $1,000 though for 5.6% return on my money($18,000 bet in total). In the end, your return is driven by your ability to find the best bets, the bets with the largest discrepancy between what the public believes and what you believe. If you only bet games where you think the line is correct, you shouldn't make any money, you shouldn't lose any either(ignorning the vig).

The thing I am most interested in right now is in turning point spreads or money lines into win probabilities. Do 3 point favorites win their games 55%, 65% or 85% of the time ? Also, I'm interested in the relationship between the spread and the money line. It seems like it should be a simple formula, based on the aforementioned win probabilities, but is it ?

by John Gach (not verified) :: Thu, 10/27/2005 - 7:11pm

#96 is a really interesting note that's got me thinking again. Would that I could say so much in such small compass! Conclusions in the first paragraph have to be right, except that one must slowly lose with fair bets because of the vig (the death of a thousand cuts). Breaking even is losing because of the sales commission on each winning wager.

What, then, are bettors such as Reno and I doing: consistently identifying misvalued bets? Just being very lucky over an extended period of time? Both? Let me give an extended example that may shed some light on the issue.

I have a games disk that has a number of casino table games on it. For about a year-and-a-half (though not much recently) I've played craps on it on two different computers, each starting with a pretend $5,000. Usefully in this case, we can assume that over a long series of wagers (many thousands of rolls by now), the results will be very close to the theoretical expectation. On one computer I've turned the $5,000 into $330,000; on the other into $8,900 -- after losing the $5,000 and starting over again, so that I'm actually down about $1,100. What's the difference?

By and large on both computers I'm following Frank Scoblette's 5-count rule; simultaneously wagering pass and don't pass for the same amount; and playing only the odds (here 10x, which reduces the house advantage to about 1/5 of a percent), which is the only exactly fair bet in craps. On the first computer I got very lucky early; before I even knew quite what I was doing I had transformed the original $5,000 into $25,000 in several weeks, at which point I started playing for much higher stakes: $100 pass/don't pass and $1,000 on the odds. At times, with all six numbers working, I can have $6,000 at risk, which can be wiped out instantly with a 7. But even with just $25,000 I could withstand somewhat longish bad streaks and wait for things to turn around and put me in the black for a session, at which point I leave. And I have had to weather very long runs of bad luck: over several months I saw $90,000 reduced to less than $30,000, before things turned back around again and I had several tremendous good runs that moved up over $100,000.

In the second computer I didn't start out with much luck at all, so I never had sufficient funds to make large wagers. There I have never been able to ride the positive random fluctuations in probability, though I've finally slowly begun to build the kitty by leaving whenever I'm up at all, even if it's only a few hundred dollars.

So what's the difference between the two? After all, I'm mostly playing perfectly on both computers. In the former case, where I quickly built up a bankroll sufficient to play like a high-roller, I can wait out negative sequences of rolls -- even going down $20,000-$25,000 doesn't scare me much now -- and wait for a benign sequence of rolls that will put me in the black. Doesn't always happen -- sometimes I do end up down $10-$20,000; but often enough it does. When I go up, I leave (since, of course, the most likely sequence of events is to start losing money again). As I view it, I'm riding random fluctuations in probability -- and getting off the horse in time not to lose my ass.

But just how have I made $325,000 in pretend money playing a game where 1) every 36 rolls on average I have to pay the house the pass line bet (when a 12 is rolled) -- in this case the house vigorish --; 2) the odds bets are exactly fair bets with no advantage either to the house or to me -- the equivalent of calling coin tosses? And why is my performance in the other computer -- where everything is exactly the same except the amounts that I now have available to wager -- so relatively dismal?

I believe that the problem as I've posed it is _very similar_ to the "fair SportsBook wager" problem as you've posed it. And here are my suggested answers:

1) In both cases the bettor can choose whether and on what to bet. I don't have to take every "fair" SportsBook wager. If I did, the result would be exactly what you predict. Ditto at the electronic craps table I can leave, whether down or up. If I'm just having an extended lousy session, I can eat my losses rather than extending them.

2) In SportsBook wagering one can assign confidence weights to wagers, independently of those assigned by the oddsmakers. One can then vary the amounts wagered, depending on level of confidence, betting the most on outcomes about which one feels most certain and least (like my wild parlays) on those about which one is much less certain. One can do this at the craps table as well, and sometimes I do. If things are going consistently poorly and I still want to play, I'll reduce the pass/don't pass and odds wagers. If I'm consistently sevening out, I'll sometimes even start making what is usually a terrible bet -- on the 7 to show up on the next roll (for which the house exacts an awful vig, much worse than the extra vig on parlays).

3) One doesn't have to wager at all, or only minimally. So, last week I wagered little (and should have wagered less) because I couldn't call any of the money line-bettable results except ATL.

4) So let's focus for a moment on the ATL wager, which we both thought was a near-lock. As I recall the spread was -7. I bet 20 units on the money line and earned a 26% return (which just about balanced the two losses plus the @ 20% profit I expect on the wager on SEA to win the division). I'm not sure what that translates into in terms of the oddsmakers expectation of the probability of an outright ATL victory -- and I didn't care much, because I felt pretty certain that ATL was going to win the game, for reasons that I detailed earlier.

5) So what may be the case is that one needs a bit of luck early. In this sense: Let's say that one can over the long term call 60% of the winners correctly (I think one can do much better than that, choosing games very carefully but the exact percentage makes no difference for this argument). Decidedly, that does not mean that in any week one will, with, say, 10 wagers, win 6 and lose 4. One might lose all 10, or win 2 and lose 8. Or win all 10. With such a small sample, you can't actually determine what the win frequency is in the long run. So, if you start out with an extended losing streak, you'll probably quit before getting the compensating wins. On the other hand, if you start with a winning streak, you're much more likely to continue wagering. The danger in the latter case, of course, is that you've misinterpreted the results (as I did with the "random walk" I mournfully but accurately reported a few notes back), that is, that you think you're consistently going to win a much higher percentage of wagers than in fact you will.

Whatcha think? I'd be interested in any serious comments on the thesis presented above -- including smashing it to smithereens as completely wrong.

by John Gach (not verified) :: Thu, 10/27/2005 - 7:40pm

Sigh. Again, even in such a long note, I left out something signficant: the importance of size of bankroll, which ultimately may be the most important factor, assuming that one is making rational, favorable wagers. It is ultimately absolutely required to have sufficient funds to withstand long negative runs. Recall my earlier example of you and I calling coint tosses, you starting with $10,000 and me with $100. I am absolutely guaranteed to lose all my money, the only question being how long it takes.

But what if I start with $1,000 to your $10,000 and don't have to wager on every coin toss? Now, one might cavil, "what difference could it possibly make that you (that is, I) don't have to wager on every coin flip, since the theoretical expectation is exactly the same on every toss?"

Here's the difference it makes, which is similar to my earlier probabilistic entanglement argument: One doesn't have to count just individual events. Nope, one can construe probabilities for sequences of events just as reasonably: 10 flips, 20 flips, etc. Let's say that heads have shown up 10 times in a row. The probability that a head shows up for the 11th roll is, of course, still exactly 50%. But what if the "event" for which one wishes to find a probability isn't the 11th roll but the entire sequence of 20 rolls? Then, having the knowledge that 10 rolls in a row have already been heads in fact gives the bettor an advantage -- because, though it is, of course, still possible that the next 10 rolls will still be heads and though in a long enough sequence of rolls that will doubtless happen many times, it is quite unlikely to be the case. In other words, in this example, the 11th through 20th rolls are in fact not independent events but probabilistically linked. Without knowledge of the prior 10 rolls, there is no advantage, since it's having the knowledge about what's already happened that confers the advantage. Naturally, this is about the simplest possible case and things quickly get much more complex with almost any real-world kind of prediction of future events. But I think the basic principle holds.

by TBW (not verified) :: Thu, 10/27/2005 - 9:26pm

RE 97: and not making every fair bet

This just means that on some level you have concluded that the bet isn't fair. That you would not be compensated enough for the risk you would assume. If you had a time travel machine and could go to Sunday night and see that the Cowboys had in fact beaten Arizona, you would come back and literally bet the farm on the Cowboys, regardless of what the odds were. Even if it would take 10,000 units to win 100 units, who cares, it's free money ?

If you pass up a fair bet, it's because on some level you are doing the math and deciding that expected payout from the bet is negative. After all, if the expected payout is positive, not betting is turning your nose up at free money.

Regarding your craps simulation, the most likely scenario is that the underlying random number generator isn't really random, or there is some minor bug in the software. Seriously.

As for the coin flip example what you are saying is true, but it isn't information that you can act on. If I watch a coin flip 10 times and it comes up heads every time, I know there is a very good chance that out of 20 flips, the 10 I saw and the next 10, 11 will be heads. Unfortunately, I doubt I could find anyone willing to take that action after watching heads come up the first 10 times.

The advantage that you point out only exists if you can go back in time to the start to place your wager. As an example, the odds of a coin coming up heads 11 times in a row are 1 in 2,048. Now, we start flipping coins and the first 10 come up heads. The odds of 11 heads in a row is now 1 in 2. Now if you could go back in time and bet and get the 2,048 to 1 odds of 11 in a row you could make a lot of money. But you can't. All you can do is bet on the next coin flip, and that is a 50/50 proposition.

by TBW (not verified) :: Thu, 10/27/2005 - 9:44pm

I did a little research on point spreads, winners and money lines. It wasn't very encouraging.

I looked at data from 2000 thru 2004. The most interesting thing is that 1 and 2 point underdogs win their games 53% of the time. Things level off at about a 9 point spread, about 20% of 9+ point underdogs win their games ! In fact, 11 out of 38 teams that were getting between 12 and 14 points won their games outright. That's much higher than I would have thought.

Now compare that to the typical money lines that you see. If you have to bet 400 to win 100, you have to win 80% of the time just to break even, that seems hard to achieve as even the biggest favorites according to the Vegas line can't beat an 80% win rate.

I'm not saying it's impossible to make money betting the heavy favorites like we've been talking about, but it looks a lot harder than I thought. I'm really surprised that 20% of big dogs actually win outright.

I suspect the real money making situation is the following. Find a 1 or 2 point dog(they win 53% of the time according to my numbers), that DVOA recognizes as a solid favorite. In a circumstance like that you might win 70% of your bets and only need to win 55 to 60% to breakeven.

I should note that the assumption I am making is that a DVOA line is not any better than a Vegas line at predicting winners. I don't have any data at the moment to assess that. Perhaps if a DVOA calculated line says team A is a 10 point favorite that really means they have a 90% chance of winning, rather than the 80% chance that a Vegas 10 pt favorite has.

Anecdotally, can anyone think of a game that had a 10+ point favorite, but that DVOA seemed to indicate would be closer ?

by Reno (not verified) :: Thu, 10/27/2005 - 10:32pm

I was all ready to suggest the John and I just take our conversation to email, and then the thread manages to live longer...

Anyway, TBW's point regarding the comparison of the money line's odds to predicted expectation of victory is one that I've actually thought long and hard about. I also haven't come to much of a conclusion, I'm torn each way.

In addition to tracking my bets for the year, I keep another document of how I would've done if I'd only placed bets when my expected win percentage was better than that suggested by the money line (and I expected the team to win). This shows a positive appreciation, but not one as large in percentage terms as the manner in which I actually bet. It seems, generally, to be a lower risk/lower reward proposition. I have no idea if my actual betting is out-performing the more conservative method's performance due to a random walk or not.

My ultimate attitude is basically this: if I think a team's going to win, I'll bet according to how likely I think they are to win, without much regard to the odds. But often I wonder if this is a completely intelligent attitude to be taking, especially when the odds get so incredibly low. It's been my observation, though, that the odds suggested by money lines often tend to be very much in line with the odds I see of a team winning. The odds on Dallas & Pittsburgh to win this week are tenths of a percent away from what I have, for example. There are typically "only" four to six games a week where I feel the odds are within my favor, but even when I feel a favorite is getting too much respect, it's often not by more than 5%. I won't bet, however, when I feel a team's winning probability is ten percentage points or more worse than I feel it should be.

This is unsuprising, as money lines are set according to the win money coming in, and not according to some ratio with the point spread. It doesn't suprise me in the slightest that it's "smarter" money than that bet on the spread.

Why in the world, then, after all this talk about quantitative analysis and arbitrage, do I ultimately place bets in situations where I don't feel that I necessarily have an odds advantage?

The main justifications I use on myself are John's: that I get to pick how much I bet, and on whom. But ultimately it comes down to a belief that if I think a team's going to win, and especially if I think there's a very, very strong probability of them winning, that it makes sense to bet on them. So what if my numbers tell me that a team has 75% chance of winning, and the line suggests an 80.9% chance? I still have a 75% chance of winning that bet. And if I'm weighting my bets and only betting on the games I want to, over time being right 75% on the time on my "best bets" and 70% on others, and factoring in parlays on those best bets, I believe I'm still going to make money. Especially since the games on which I'm getting significantly better odds often counter-balance the games where I'm getting correct or slightly worse ones due to the vig.

As I write this, I think I've actually finally managed to convince myself that my method of placing wagers really is more risky than TBW's suggested one, which I've been charting on the side.

But hey, it's more fun, and that's why they call it gambling! I'm still confident that the way I actually do things in practice will continue to accrue gains. What I'm not sure of at all is that it would continue to outperform a more odds-rigorous approach over the course of, say, 1000 weeks rather than 22.

by Jimmy Two Times (not verified) :: Thu, 10/27/2005 - 10:33pm

TBW: Not quite in your parameters, but the spread on this week's Buf-NE game is currently NE -9.5; my DVOA system calls it for NE by a touchdown.

Someone asked upthread how DVOA can predict over/unders. The way I do it is by adding Offensive DVOA to opponent's Defensive DVOA, adjusting for homefield and multiplying one plus the result by 21.5. I know that method leaves out special teams entirely, but it has had about a 60% success rate ATS so far this season.

Paris du semaine:

1. Chi/Det UNDER 32.5
2. Cle/Hou OVER 37
3. Min/Car UNDER 44.5
4. TB-11 @ SF
5. Was/NYG OVER 42

That said: A nice 2x6 round robin on Cincy, Dallas, TB and Pitt (pays approximately 25:12 if everything hits, and protects your downside in case as many as two massive upsets come in) is probably where I'll be putting my money.

by Jimmy Two Times (not verified) :: Thu, 10/27/2005 - 10:36pm


In the event you two decide to take this discussion to e-mail, I'd really like to crash your party. ankurvdesai@yahoo.com

That said, I think there are a bunch of us lurking on this thread who really enjoy the discussion, so I don't see any reason to feel as though you have to take the discussion private.

by John Gach (not verified) :: Fri, 10/28/2005 - 2:34am

1) I need to study and seriously think about TBW's critique in #99 & #100, which I probably can't do for a few days. Short-term silence on the topic will _not_ mean I'm ignoring the notes.

2) Like Reno, I'd thought that maybe we should at some point stop pretending that anybody else cares about what we're thinking and go private. Every time I think that, along comes a note saying "continue in public, please" or thoughtful provocative contributions to the continuing discussion, like TBW's. So I guess our private vice is still a public virtue. Who'd a thunk it?

3) Re: Jimmy Two Times last note: I notice that JTT's got the same parlay I've got (see #90) but without SD, the adding of which certainly does increase the degree of difficulty. I gave, however, my explicit reasoning why I think SD will win. All the others seem like solid bets to me. I'll have to play around with JTT's method for assessing over/unders using DVOA. Would it make sense to add to Offensive & Defensive DVOA 1/3 of the difference in the Special Teams DVOA in order to account for its contribution? Or does that distort the total incorrectly or make too little difference to matter?

4) Re: Reno's #101 & (partly) TBW's #100: As usual, just about everything Reno wrote corresponds to my own approach (which is getting scary: are we engaging in a folie a deux sharing the same delusion?). There's one thing that TBW hasn't taken into consideration: what if punters (the British term, I think) such as Reno & I are actually significantly better at picking likely winners than the average lot? Remember, we're not attempting to pick all winners -- in fact sometimes very few. For example, I don't bet on heavy favorites because they're heavy favorites in the minds of oddsmakers or the general public. I wager, as does Reno, on teams that I believe will win, regardless of the line. Sometimes they're technically underdogs; more often, of course, they're favored, sometimes heavily. The stats TBW provided are scary, because they suggest that I can't / shouldn't continue successfully to call the high percentage of winners that I have this year so far. But isn't that conclusion based on an assumption that we're all equally inept (or equally skilled, depending on one's point of view) at predicting future events?

5) In a few recent notes I tried explicitly to tease out what one might call the cognitive background to my approach. This begins with DVOA, which provides a superior tool for assessing actual relative strength of teams, which often differs from the public perception of the teams (see this week's various power rankings, some of which are bizarre in relation to the DVOA rankings). But it's not just the current week's DVOA (one snapshot out of 16) but the relative movement of DVOAs over time as well that I count, which provides a more dynamic view.

6) Many notes ago I suggested that SportsBook wagering was a skill. It does seem to be the case that I'm much better at it now than, say, two or three years ago. The sample size is still so small, though, that it's still nearly impossible to know for sure whether I'm significantly better at it or just once again in one hell of a lucky streak.

7) Since I'm still hacking away at 1 a.m. let me at least begin to address TBW's #99. TBW wrote:
"If you pass up a fair bet, it’s because on some level you are doing the math and deciding that expected payout from the bet is negative. After all, if the expected payout is positive, not betting is turning your nose up at free money."

Yes, that has to be right. But doesn't that suggest that, like the animals in "Animal Farm," some "fair bets" are more equal than others? Some more worth putting money on than others? In fact we know that such has to be the case: bets that yield a positive return are just the ones you want to make, while skipping all others. So, perhaps the underlying question is, "Is there any means for identifying at least some such wagers in advance?" Perhaps more anon on that in a few days (or perhaps not, if I can't think of anything worthwhile to say about it).

8) I think your suggestion about the vast discrepancy in the two craps simulations cannot be correct, to wit that there's a bug in the software or that the RNG is defective and not really generating random numbers. I'm running exactly the same software from the same disk (it has to be run from the disk rather than the computer's hard drive). Now, one might attribute the difference to some difference in the two computers (which are not identical machines), except that all the other games I run from the same disk seem to perform identically on the two machines. When I play the slots emulator, the same virtual slot machines are loose or tight on both computers, for example. When I play Texas Hold'em I tend to do on average just as poorly on one machine as on the other. So far as I can tell, the craps emulator also works identically. It's not that I'm luckier on the one machine, but that I was very lucky early and built up a large bankroll, which then allowed me to withstand longer negative sequences. On the machine where I have relatively little money, several strongly negative sequences in a roll leave me too poor to make the correct wagers.

Which, to me anyway, suggests that the answer lies elsewhere. I've suggested that the combination of close to perfect play, initial good fortune leading to a much larger bankroll, and then (with the larger bankroll) being able to outwait long negative runs account for the difference in performance. Here's what happens in a typical session in the large-bankroll machine: Sometimes I go up a few thousand right away. I leave, even after 3 or 5 minutes. Sometimes I go down right away. Usually I hang around playing and eventually hit one or several good runs of positive rolls that leave me ahead. Again, I leave. Sometimes I end up going down, say $20,000 and I leave, having hit a loss limit past which I don't want to risk any more money. It can then take several months of relatively small gains to get back to the initial position from which I lost the $20,000 (or whatever). Expectably, I've had multiple sequential losing sessions, occasionally going down $60,000 or $70,000 over a month or two. But such bad runs are inevitably followed at some point by a balancing set of good runs that bring the account back up. Remember that the way I play, the only long-term cost is the once in every 36 rolls loss of a pass line or come bet (usually $100 on the high-roller machine). All wagers on the odds are exactly even with no house advantage and pay off at the statistically correct amounts (2 to 1 for 4 & 10, 3 to 2 for 5 and 9, 6 to 5 for 6 & 8).

by John Gach (not verified) :: Fri, 10/28/2005 - 2:41am

Somehow a smiley got put at the end of my last sentence, which correctly reads:"6 to 5 for 6 and 8."

by TBW (not verified) :: Fri, 10/28/2005 - 7:41pm

ok, I added some more data. I now have 3,003 games from 1993 thru 2004, including playoff games. Overall favorites win the game outright 66% of the time. This number of course varies with the spread although it is far from perfectly correlated with the spread. At the low end 1 and 1.5 point favorites win their game 48% of the time. 3 point favorites win about 60% of the time. 7 and 8 point favorites about 75% of the time.(one interesting anomaly is that 5 pt favorites are about as successful as 7 and 8 pt favorites winning 77% of their games(sample size over 100 games). Teams favored by 10+ points win 81% of their games(369 game sample). Looking at an even smaller subset 15+ point favorites won 97% or 38 of 39 games.

It's interesting that the spread picked the winner 66% of the time and DVOA 67% of the time. It implies that DVOA isn't much more accurate than the point spread at picking winners. Of course, I think the methodology used to pick winners using DVOA was much too simplistic. Assuming that there are ways to use DVOA and get even better results than Sean did then there is still hope.

The question I have again is where do we look for the advantages ? Is DVOA best at divining which strong favorites are for real, and which actually matchup poorly with their opponent and thus are suceptible to the big upset ? (when a 10+ point underdog pulls the big upset is always due to unpredictable things like injuries, fluke plays, etc. or is there sometimes an inkling of a matchup problem for the favorite ?) Or, is DVOA best at identifying underdogs that really should be favored ? Considering that according to Sean the best bets(which if I remember correctly were dogs that DVOA liked to win) win 67% of the time, and that 1 and 2 pt dogs only win 50% of the time, I believe the latter may be the real money making situation. However, the road to riches is probably a lot rockier this way.

by budman (not verified) :: Sat, 10/29/2005 - 9:05am

Wow, this thread has kept going and has had great discussion but it is time for me to chime in here again and keep playing Devil's Advocate. So for the few that still poke in here lets play... (this one will be long)

First, a little history. I have been researching this topic for quite some time now (over seven years) in regards to statistics, rankings, rating systems, point spread, etc. I have been entrenched in football sports betting for equal time. During this time I have been privy to a few studies done by universaties as well as several done by myself that remain private. I also have experience in being a part of the 'lottery' each Sunday night at the Stardust that test lines for some TRUE professional bettors before releasing a final line to all of Vegas (and most of the world) as well as met someone who is way smarter than all of us who worked at LVSC (the place where lines are MADE). So I can objectively say that I know a thing or two about this topic and probably could write a book (if I was more organized as well as motivated). With this, please don't confuse any apparent ego or negativity as close mindedness, or worse, lack of knowledge.

So with that, I can truly appreciate some of the efforts others in here have been putting forth in finding an advantage in NFL wagering as I have also been there before and know what it is like. To think that you really are onto something that is different and pioneering and that has an edge over ALL other past attempts and found a way to beat the house. BUT, YOU AREN'T and you DON'T.

As unique as DVOA appears and how different it is for judging teams, it STILL is based on the SAME info that has been incorporated into countless rating type systems before it and SURELY are incorporated into point spreads and Money Line figures. DVOA is just another way to read _past performance_ using _traditional_ stats.

Think of it this way: Knowing that it is their livelihood to make accurate point spreads and Money Line numbers to make sure sportsbooks across the world don't lose money, there is a very high probability that LVCS currently employees, at least, 5+ outrageously gifted staticians that would laugh at DVOA wondering what took someone so long, and probably another 5 that took an interest and incorporated it into their analysis. If not, they will, provided they think DVOA actually would be proven to have a higher correlation of success than standard deviation would suggest.

Betting and winning requires more than just thinking one can find a 'code' hidden within some cold static numbers. What's worse is using a retrodictive rating system (which DVOA clearly is) to try and determine future performance of human participants added in with weather, injuries, etc.

I really hate being so negative, but I have seen these same types of threads numerous times over the past seven years for each new ratings system discovered and published for public consumption. Nothing new under the sun. It is Deja Vu for me.What success others might have had so far in their small sample size is just that, small sample size, or random walk in accordance to SD. Sort of like the point John made with his craps analogy where one sim started off hot and the other lost its bankroll. Some of you just started "hot".

I cringe to think how much John exposes himself with all those units being bet on 'high confidence' picks since even though they may have high win rate, are -EV in the long run. One can't bet on ML favorites (esp large ones) because once again this is factored. Selective betting of only your high probability picks don't matter when you are laying more than you should get back. Like a 5-team parlay that true odds to hit are 32 to 1, but are getting paid 26 to 1 (if you're lucky enough to find that good of a payout as most pay 20 to 1!), you still lose even when you win.

Sports betting has become hard and lines SO tight as technology and communications get better that sports betting may be already or will become impossible to beat in the long run unless you can find weak bookies. But they dry up so fast as they eventually adjust or go broke (yes, bookies can actually lose).

I once had a bookie a few years ago who was too busy during the day with work that he couldn't monitor lines all day and actually told me to use the lines posted in USAToday! MY god that was, to this day, the closest I ever came to having a license to print money. I got a few friends together and we monitored offshore books for line moves and injuries on the NBA side/totals during the day and we detroyed him to the point of us getting cut off in a month. Last I heard, he got out of the business. This is what I mean by weak books and in all my experience is about the only way left to surely be a winner. Handicapping contestants have dropped in winning percentage each year and the ones who do win are won by the hot hand. Fifty-five percent over the long run ATS is attainable by very few in the world, and I suspect they use inside knowledge.

But none of this surely won't stop others from trying and sadly doesn't even stop me despite how much I know and my inner voice tells me otherwise. It is almost a curse of intelligent people to think they can eventually outsmart and overcome failures of others who have tried, as well as ones ownself. But for me, statistics and numbers have always been interesting and I will always enjoy football along with in-depth analysis (which is why I found FO). Plus I am now at the point where wagering won't land me into Gamblers Anonymous because I know it is more just putting my research into practice in controlled amounts that won't effect me at all if I lose and more recently I have even entertained thoughts of just quiting all together because it is insignificant amounts anyway and I have begun to hate the roller coaster for small profits.

I hope many of you heed this ridiculously large post mainly to mean "Take it easy" on your bets if you are in fact wagering. There is NO easy way to sure money each week betting sports.

Ohh, I linked the study (pdf) I mentioned where ratings systems were studied on 1800 college games. It is ncaa but that is actually better since the opportunity for more trials exists in same time frame. Conclusion?? No advantage, still 50/50, all factors known are already factored into the line...

Thanks for reading.

by budman (not verified) :: Sat, 10/29/2005 - 9:31am

Ohh, one last thing I wanted to also bring up is how large some online books have become. They take in probably 10x+ more sports action a day than Vegas does. They also all offer Bonus money and/or reduced vig.

Pinnacle Sports has large limits and the smallest vig on the web. WAY larger limits than anything Vegas would take....so if they didn't REALLY know what they were doing, they would have been in deep financial trouble in no time (while they've actually thrived and become the market leader) as surely any winning player or 'syndicate' would play here because of the limits. Pinnacle is ruthless in their lines and amounts accepted while at the same time growing at a large rate.

Want a more SURE thing than TB over SF this weekend? No DVOA needed. They are offering Money Line on LSU over North Texas this week... It is currently only at -40000. :0


by TBW (not verified) :: Sat, 10/29/2005 - 2:58pm

RE: 107

Excellent points Budman, but I think perhaps you are being too negative. The fact is there are people who make money consistently betting things like horse racing, and there is no reason why football should be any different. If sports books were in the business of taking positions on games, then I think your point would be more valid, but my understanding is tht most of the time they are trying to balance out the wagering as equally as possible so that they are sure to walk away with the vig. Now I am sure that it is rare when they can get the wagering to be a perfect 50/50 split on a game,so on most games the probably do end up with some sort of minor interest in the game. Still, the point is that the lines while set by pros get moved by people who aren't pros, and they are set by pros not to accurately predict the final result, but to accurate predict the final BETTING. There is a substantial difference. If you have better information than the millions of Joe Six-Packs who ultimately set the line, you should be able to find some sort of long-term advantage.

The biggest problem using DVOA for wagering right now is the inability to backtest it. So, we have no idea if a team DVOA favors by 10 pts wins 70%, 80% or 95% of the time. Until we have that information, we are flying blind and simply speculating. My gut feeling is that the advantage on big favorites is very small if it exists at all. And I doubt very much that it is so large that there really exists several opportunities a week where we can expect a long term positive result. My gut feeling is that DVOA probably can identify REAL wagering opporuntites that should generate a long term positive return, but that there are probably 6 to 10 such opportunities per year. And as such, you would need deep pockets and have to place big bets to really profit from the advantages in a meaningful way. I would be SHOCKED if DVOA really conferred such an advantage that there were on average 2 or 3 games a week that could be bet on with the expectation of a 20% return.

At the end of the day point spreads are driven by a market, and markets are pretty damn effective tools for predicting the future. They're not perfect, but they also are not wholesale wrong about things. In fact we already know that the point spread market is very effective at picking winners(66% accurate, 75%+ once the spread is over 6 points). That doesn't leave a lot of room for improvement when it comes to betting money lines.

Your points are good ones Budman, I don't think the task is impossible, but it's a lot harder than we have made it out to be in our discussion here.

by TBW (not verified) :: Sat, 10/29/2005 - 4:28pm

So, I guess my best bet for the week is Cleveland. They're at +113 at Pinnacle. There are a couple of things working here. First they are a 1 pt dog according to the spread, and 1pt dogs actually win a little more than 50% of the time. Second my DVOA method makes them the favorite in the game. So, if they win the game 50% of the time I would make money, and I have two factors pointing that they have a better than 50% chance of winning. Finally, there is an easy to identify reason for why this betting opportunity exists. People figure Houston has to win a game eventually and they probably think Cleveland is the easiest game left for them. This is akin to betting tails because heads has come up 10 times in a row. It's just wrong.

The problem of course is that this is my "best bet" and I maybe have a 55% chance of winning it. And there is only one game like this in week 8, and I don't think there were any in week 7. So, it would take literally years before there would be enough data to prove this theory right or wrong.

If we only had easy to use historical DVOA data for backtesting......

by Michael T (not verified) :: Sat, 10/29/2005 - 6:05pm

Fascinating debate, which I've been following since it's inception...it does strike me, however, that there's a basic truth about betting that needs to be said - the only way you'll win long term is by identifying value bets - i.e. bets where the odds quoted by the bookmaker are out of sync with the true odds of that team (or horse, or whatever) winning. So it's about taking the 3/1 to win bet when the true odds of winning are 2/1 and so forth. DVOA may give you additional insights (and it's certainly helped me), but the point shouldn't be to use DVOA to identify winners, but to use DVOA to identify where bookmakers have priced up teams wrongly, and taking advantage of those discrepancies...

(apologies for terminology in this post,
I'm from England and there's different words for many betting terms...)

by budman (not verified) :: Sat, 10/29/2005 - 8:42pm

TBW... I'm currently digging up another study for you.

Michael... I agree in that the only way left is to find value bets. This is what I had going to a large degree with my USAToday story and basically will never happen again rest of my life. So outside of having an obtuse bookmaker taking lines out of the morning paper, the opportunity for 'scalping' weak lines in todays information age is almost dwindled down to not impossible, but at least not worth the effort because of small expected returns.

Another problem with this is the contradiction of who's line actually is trueness? And can a line ever be truly "true" for it to be used to find weakness? Does Having the public pound a line for a weak at the largest books online represent the "true" line. We don't know if it was true because we never know how even or lopsided their take was and therefor if the line was badly off to begin with and the books were forced to take a side on game, or if it was 50/50 at kickoff which is ideal for trueness. Maybe it is the opening line? Your line? LVCS? Etc.

But even then finding value is the next task and that is just as difficult. Even if one could KNOW what the true line is or should be, finding it far off enough somewhere else is increasingly becoming more difficult and/or usually not off far enough be statistically relevant anyway.

Say you could know for example that the true line for NYG this week is nyg -2 as that is where it is at kickoff. You could have had wash +3 earlier in the week but back then you didn't know it was value. Now it would be impossible to get wash at +3 anywhere and maybe can find a +2.5 for yourself. The difference between you creatively finding a +2.5 and having a +2 is minimal as 3 is such a key number. So now you would have to buy a 1/2 point to get to a key number and then negating any value you found.

Another example would be say you see a TeamA is -175 on the Money Line at where you think represents a TRUE line (where that is I don't know). But you check all your outs and one of them gives you TeamA at -167. So now you blindly take TeamA because it represents "value" but in meantime you have no idea about the team and have no means to say it is value and will win other than YOU were able to find it lower somewhere else. What about the countless others who found good "value" on the + side? It is a matter of perspective on good value.

Yes, there are people who make a living betting sports. But whether they got there by skill and actually having an edge, or by probability dictacting that some will have hot streaks and walk right into them while others come out of the gate stumbling never to recover is debateable. If you and I were to flip coins for the next year or two 20 times a week for money, I could very easily be broke should I have a bad run and bad money management while you would win $$ by default without trying basically and then go on to open a website proclaiming your expertise.

Fun and interesting debate.

by TBW (not verified) :: Sun, 10/30/2005 - 1:46am

The true line is revealed over time. It took me less than a day and no money to come up with 12 years of data about NFL spreads, over 3,000 games. We know that on average favorites win 66% of the time. It may be that one week they'll win 90% of the time, or even one year maybe they'll win 70% of the time, but the point is we have a very good idea just how accurate the point spreads are.

Also don't forget that the spread are not trying to accurately predict the final score, they're trying to balance the amount of money bet on each team. Lots of ill-informed people bet on football games, some people pick the helmet they like best, etc. Then there's human nature. Why are things often price at $1.99 instead of $2.00 because $1.99 seems cheaper even though it's just a penny. People who bet on sports are not always acting rationally. They are human,they are driven by hunches, greed etc. This is what can create inefficiencies in the market that can be taken advantage of. Have you ever seen people race to a roulette table to bet on red because black has come up 7 or 8 times in a row ? Those people think they have an edge when of course they don't. All of these type of behaviour are what can create betting opportunities.

Let's put it another way. Let's say that I have discovered a method using DVOA to 100% accurately predict NFL games, including the final score. Think I can make some money with that ? Of course I could. Well, what if my method wasn't 100% perfect, but only 85% ? That's what we're trying to do. You can't be perfect, but you can do better than the people betting on the coolest helmet, or the team that is 0-9 because they HAVE to win one game.

As argued in post 111,it's all about value bets. And the problem right now is not having enough DVOA data to really know how accurate it is, and thus what constitues a value bet.

by TBW (not verified) :: Sun, 10/30/2005 - 1:49am

One more thing, if DVOA truly is a more accurate predictor of games, then yes over time this knowledge will become common and DVOA will begin to be priced into lines. But, as it is now I doubt it has any effect on lines since I bet 99.9% of NFL bettors would think that DVOA was that band from the 80's that wore flower pots on their heads and did that crappy cover of Satisfaction.

by John Gach (not verified) :: Sun, 10/30/2005 - 3:23am

[Advance warning: this is a very long and complex note that ranges quite far from what we seem to be talking about].

My God! Leave the "thread with legs" for two days and you come back only to learn convincingly that one can't possibly have a future making any money wagering on the results of sports contests. Budman and TBW have made so many profoundly interesting points that I'll just pursue one or two strands in the recent movement of this thread.

Increasingly I am convinced by their arguments that, merely on the basis of statistical inferences, one cannot in the long run get a positive return. I believe -- even without the direct experience that budman has -- that all relevant information, statistical and otherwise, tends to be factored into betting lines. I already figured that some equivalent of DVOA was being used by oddsmakers. As I hinted at earlier, the history of probability theory began with and has been driven by the need to understand gambling uncertainties.

But that does not automatically mean that particular individuals cannot profit from market inefficiencies or from a different kind of knowledge that is not statistical. We know that in another highly similar market (stocks) some individuals consistently outperform the market itself (Warren Buffet, for example). This ability of a few to do better than the average lot at picking winners led first to the invention and then to the vast recent growth of hedge funds, now far past their heyday in my opinion since it makes little sense to think that a rarefied minority skill can be generalized to earn profits for everybody. And we have the salutary example of Long Term Capital Management, which went broke -- and which could in a worst case scenario have taken down the entire Western banking system because of interlocked hedged bets. Keep in mind that LTCM was founded by several of the most brilliant economic statisticians of the 20th century, one of whom invented the concept of hedging, which led directly to the creation of derivatives.

So, we are presented with a contradiction: it looks as though you can't "beat the market" (whether stocks or SportsBook) -- yet demonstrably a small number of people consistently do it.

Actually, I strongly suspect that there are a number of ways to do it (e.g., Buffet's approach of buying undervalued stocks and holding them long-term is only one of many successful strategies) but that such methods cannot both be public information and continue to work. Thus, if DVOA confers any advantage, it will be short-lived as the information it reveals gets absorbed into the statistical rules used in setting betting lines (see budman's note). What I mean here by "public information" is not the what but the why & how of choices for wagers -- knowing what Buffet does is not the same as being able to do it yourself.

As I think my two craps simulations strongly suggest, bankroll size is not an extra-added optional feature, but a necessity. You either need to be lucky early [1] or begin with a bankroll sufficiently large to withstand a number of early large losses. [2] To have such losses early and still maintain faith that you actually know what you are doing (not to mention still being able to convince others that you know what you're up to) is probably in itself uncommon. So, in that sense it is likely that most succesfful big-time gamblers in stocks or SportsBook had to have the Goddess Fortuna smiling at them in earlyish days -- else they would have quit.

[1] LTCM's approach was theoretically correct -- it got hit by a perfect storm of unlikely economic events that caused certain things, which usually could be counted on to behave in opposite fashion, to move in concert. Thus, its hedges turned out not to be hedges at all, but, like many of my weekly wagers, interlocked and contingent on each other.

[2] George Soros, who made billions wagering on the devaluation of the British pound (among other correct calls), lost back a billion or so in the late '90s by guessing wrong about currency movements, after which I think he gave up that particular kind of wagering.

All the above is merely propaedeutic background info for what I actually want to say. In an earlier note I made the claim that I actually make aesthetic ratber than statistical judgments, albeit the former require and take into account the latter. Nobody commented on that, which is probably not too surprising given the statistical interests of many folks who read the stuff at FO. However, I'm increasingly coming to think, after reading the recent notes of budman & WTB, that if I'm in fact making any judgments about contests that are out of the ordinary, that is that express a greater likelihood than the betting lines suggest -- something admittedly not proven with such a tiny sample size --, then it may very well be the case that I am making judgments about probabilities by, as I wrote earlier, feeling them. My ideas about this subject derive from a highly complex ontology and phenomenology -- that is, a theory about how the world actually works and about how it appears to us humans --, one that is in many ways quite similar to John Searle's ideas, though developed quite independently and using different concepts and terminology (and, quite unlike Searle, completely unpublished).

Though, unfortunately, it is impossible to describe in the brief compass available here (and would certainly take us far far away from the topics we're discussing), perhaps I can impressionistically sketch how it might be relevant. Living creatures evolved by learning to assess probabilities in their environment. At the primitive level of one-celled creatures (which are actually extremely complicated systems in their own right, being the result of a billion years or more of evolutionary adaptation) this amounts to making three basic kinds of judgments: can I eat it; can it eat me; can it hurt me (in which case I should run like hell). Where the probabilistic judgments come in is in identifying what constitute objects in the environment on the basis of sensory information (chemical in this case -- these critters got no nervous systems), that is sorting James's buzzing booming confusion into classes (in this case three). Then all the individual organism has to do is identify an object as belonging to one of the classes. Highly efficient and not requiring any of what we call thinking. But, without any ability to think, how do such creatures "know" that X is a baddie that will eat them? They feel it, by which I mean that the entire organism is disposed to react to X by moving towards it (to eat, and evolutionarily much later to merge to create new creatures), or by moving away from it. I call this unitary kind of feeling state an affect. It is, I believe, the origin of what became feelings in more complex organisms (like us). Though intimately related and using much of the same physiological apparatus, feelings are not identical with emotions. Both can express judgments, but feelings are always global and have no direct external object (that is, are intransitive): an object _affects_ the experiencer by inducing some particular state of the organism, which state is not, at the level of experience, itself divisible into parts (think of one's reaction to music or painting, as useful examples).

Now, though clearly the origins of affect lie in direct experience of sensed, physical objects, it is, in humans at least, not restricted to the sensed environment -- because we mediate our experience of the world through ideas and language (a triadic relationship that is far too complex even to begin to try to explicate here). Probably unlike most creatures, we can have feelings about abstract objects (literary texts, for example). As I see it, there is no intrinsic reason why we can't directly feel highly complex probabilistic judgments, something that would be similar to an expert pianist who, after having done all the grunt work and having learned perfectly the mechanics of playing the instrument, can relegate all of it to unconscious brain programs and concentrate entirely on aesthetic effects in his playing.

Finally, for the one or two of you still reading instead of snoring, we get back to gambling. My thesis is actually quite simple (if one is cognizant of the background stuff I've presented above): With sufficient training one can acquire the ability to feel the likelihood of the outcome of certain sports contests (or, for that matter, stock movements). This is an entirely different kind of knowledge from cognitive judgments _about_ the contests -- in fact the more I have to think about who will win a game, the less likely I am to wager anything on the contest, since I have no clear feeling about the likely outcome. Those are what I call "too close to call" games, a category that is not derived from the oddsmakers assessment. For example, the PHI @ DAL game, which I stupidly wagered on, following (that time only, I hope) the unanimous opinion of the pundits and oddsmakers, was actually a "too close to call" game for me, even with a wide spread in Philly's favor. In an earlier note I called it a trap game. Now, how did I know that -- and what is it that I actually knew, even though I dumbly didn't act on it by not betting? Firstly, I definitely did not "know" that Dallas would win, otherwise I would have made the highly profitable contrarian wager. I just felt the result as cloudy and indistinct, as messily clustered around a number of possible outcomes, as a kind of probabilistic wave form that I had no ability to decohere into some particular highly likely outcome. Along with the pundits, I consciously "thought" the most likely outcome was a Philly victory -- but my feelings argued otherwise and I just didn't listen.

If my thesis is correct (yes, probably dubious, and probably not provable, even if I wager successfully for years, since there's no limit to how long a random walk you can take), then such a minority ability to call certain games would be nought but noise for oddsmakers -- and thus their statisticians can't take it into account. It could only matter if individual wagers were so large as to affect the lines -- and the correct strategy, I think, for placing such large wagers would be to spread them around in a number of venues. As budman pointed out, some of the online gambling venues now handle such a large volume of wagers that individual bets would have to be very large indeed to matter much. Some (most?) sites also impose limits on certain kinds of highly favorable bets, precisely, I suppose, to limit possible losses on near sure things (like, when it was last availabe, wagering on IND to win the division, with a 3% return -- a license to print money).

On the other hand, for all I know from here on out I'll lose every football wager and have my gambling account back down to zero by the end of the year. As one reads in mutual fund prospectuses, past performance is no guarantee of future success. Every week until the games are played I have no clue how things will actually turn out.

Enough. Far too long and I apologize. But I couldn't see how to present my ideas about feeling probabilities in a way that made any sense without introducing the cognitive background.

by John Gach (not verified) :: Sun, 10/30/2005 - 3:40am

Quick comment about budman's #112. I think sometimes one can either know or make a reasonable inference that an opening line is probably incorrect. This week TB opened at -9 or 9.5, then quickly moved up. Since I don't wager on spreads, I didn't take the spread bet but did include TB in a parlay, getting a higher potential return on TB's win than I would have later in the week. In fact this week I placed all my wagers early (Tuesday or Wednesday), because I thought I saw a lot of value on the opening menu. The useful thing about these weekly experiments, is that one finds out quickly how one has performed and can learn from mistakes.

by John Gach (not verified) :: Sun, 10/30/2005 - 4:23am

Re: TBW's #109 & #110. Actrually, I like the Cleveland bet, too, though I didn't make it -- precisely for the somewhat silly reason you gave: If Houston is to win a game before SF, this looks like a likely candidate.

Point well taken in #109. I've been wondering if I'm making too many wagers and had already concluded that I might be better off placing much larger bets on a small number of outcomes that I feel [literally, see the earlier note] highly confident about. This would seem to be true, even if I'm just randomly walking (since it reduces the number of occasions when I could be wrong).

Regarding budman's pick-your-superlative mini-essay in #107:

1) Point well taken about parlays. I know they're regarded as sucker bets, like most of the props.

2) It seems to me, too, that DVOA by itself cannot in the long run confer a selective advantage in predicting outcomes. But what it does do is give predictors (degenerate bettors in most of our cases) a valuable fact base and analytical tool that they did not have before. What I'm suggesting is that, even though the degree of difficulty of SportsBook wagering has increased because of the statistical armamentarium available to the oddsmakers, DVOA actually levels the playing field somewhat by giving bettors information previously only available to oddsmakers.

3a) I can guarantee that arbitrage opportunities do exist, though they're certainly not common. For example, in last year's NCAA playoffs Duke was playing in the first round some high-scoring small school (from Arkansas, I think). When I saw the over/under I couldn't believe it: it was a gift. I can't remember the exact number, but I do remember that Duke alone figured to score nearly enough points to get close to the over/under. If its opponent scored 50 points, it was a lock for the over - and in a game that in the second half was going to have Duke reserves on the floor with nobody worrying about playing defense, this seemed like (and was) an easy bet to win.

3b) Then in the NBA playoffs, the early calls for series victors were ridiculously easy. For example, there was absolutely no way, short of both Shaq & Wade getting injured, that Washington had any chance at all of winning the series. I think Washington had lost ten straight games to Miami. Expectably they were swept. Almost any payoff for this kind of wager is a gift. That such opportunities exist at all is what puzzles. There must have been people betting on Washington, mustn't there? True a Washington victory in the series would have generated a huge return -- but hardly one commensurate with the actual risk.

3c) So, don't 3a) & 3b) illustrate that your thesis that all relevant information is factored into bettiing lines can't quite be correct, even if in most cases it is?

3d) Not that I know enough to do it, but I remember reading somewhere that wagering on baseball is for some bettors highly profitable, because, since it's a much lower volume deal for the bookies, less effort is put into getting the lines correct. Any thoughts about that?

by John Gach (not verified) :: Sun, 10/30/2005 - 5:57am

Re: my #98 & TBW's #99, which I had promised to think about. TBW's right, I completely misfigured the correct way to conceive of the probability of the next event in a series of coin-flips, for exactly the reason he gave: without a time machine one doesn't have in advance the useful knowledge that the first 10 coin tosses will all be heads. This is also exactly why one can't use quantum indeterminacy and spooky action at a distance (such as state-entanglement of particles created together and then separated) to predict the future: one doesn't have the knowledge in advance about the states of the two particles and can only get it by performing a measurement on one of the particles, at which point you know with absolute certainty the state of its Siamese twin, no matter where it is in the universe. That said, I should add that

1) there are some intriguing theories around these days in consciousness studies that do make use of quantum entanglement.

2) There have been in the past few years some ingenious physical experiments that demonstrate that it is in fact in certain highly specific situations possible to know what you shouldn't be able to know -- for example, exactly which path a light wave (or ray, depending on how you want to look at it) has taken to reach an object, even though this is completely undetermined and entirely a stochastic process.

* * *
If I understand correctly the various discussions in this thread, a point spread expresses the following judgments:

1) a judgment about the probability that a team will win (none for a pick-em wager)

2) a judgment about the probability that a team will actually win by more than a certain quantity of points

3) a judgment about the probability that the spread will encourage roughly equal wagering on both sides. Amounts actually wagered can and often do move the line up or down.

There are two distinct sets of opinions being expressed here: one about the predicted outcome of a future event and another about the behavior of bettors. These are closely linked in that the second judgement directly derives from the first. Furthermore, the actual behavior of the betting public secondarily influences the first judgment, changing it to accord not with the oddsmakers conception of reality but with that of the bettors.

Now, here's the question. Just limiting the event horizon to Anglo-North America, we're talking about a population, the majority of which believes in the reality of UFOS and of psychic phenomena like telepathy, precognition, and psychokinesis; a significant percentage of whom believe in Creationism and think that evolution is poppycock; and a significant percentage of whom believe, like the president, that God holds personal conversations with them. Exactly why should we take very seriously the cumulative predictions of such people about future events? Are they really very likely to have an objective grasp of and collective wisdom about the likelihood of future events? Yet, is it not likely that it is many people like this whose opinions are shaping many betting lines?

by budman (not verified) :: Sun, 10/30/2005 - 8:11am

#116... Very good point in the last paragraph. But, even though we snicker a the other types of mystical hobbies such as the many examples you give, one must also consider ourselves in that category as well since we believe that we can also predict future outcomes based on human performance with enough certainty to overcome house edge.

...budman’s pick-your-superlative mini-essay

Sorry bout that.

Anyway, great debate guys. But this thread is dead. Look forward to interacting with all of you more on other FO topics. Good luck on all your wagering and please remember to share with us all how it is going every so often. Us early birds should stick together before the swarm comes later begging.


by Michael T (not verified) :: Sun, 10/30/2005 - 3:09pm

#116 and #117

Important points - what you're actually betting against is the opinion of bettors at large - odds get adjusted due to weight of money wagered. You've got to outsmart the masses to consistently win.

FWIW, my DVOA system, which I'll write up when I've more time, has the following as best value bets (in order) this weekend (all to win, or as seems to be the US term, the money line):


Tampa Bay


(Buffalo and Cincinatti are next)

by John Gach (not verified) :: Sun, 10/30/2005 - 3:21pm

Re: budman's #118: "pick-your-superlative" was intended as a high compliment, not an implied criticism. And, yes, a belief that one has some method or tool for overcoming the house edge may well rank as its own brand of superstition. It certainly enriches the house, whether it's casinos or online books. The thing is, holding a belief, whatever its ultimate source, entails some degree of conviction about its truth (otherwise it wouldn't be a belief but some form of provisional hypothesis). One can only know that a belief is superstitious when one no longer holds the belief.

by John Gach (not verified) :: Sun, 10/30/2005 - 3:39pm

Re: Michael T's #119: Exactly. You stated the case a wee bit more succinctly. That's one reason why I believe that arbitrage opportunities do exist, though not the only one. For example, that opening -9 or 9.5 line on TB this week was, presumably, not yet influenced by the betting behavior of the masses. Yet it was, I thought and still think, definitely too low. Why? I pose it as a serious question, because I don't understand the opening line. It doesn't seem to make sense that the oddsmakers wanted or needed with the opening line to encourage betting _on_ TB. Does anyone still reading this thread know what goes into such lines?

By the way, when this thread does eventually expire from natural causes, anyone who wishes to continue this kind of discussion (I have no idea in what form) can contact me at john@gach.com. Despite budman's suggestion that it's already dead, I continue to find the contributions stimulating and thought-provoking.

by Michael T (not verified) :: Sun, 10/30/2005 - 4:03pm

John (#119)

Are you not focussing on a single example where the line is too low? Presumably there's an equal chance of the line being too high? Or do you think the opening line is consistently too low where one team is highly favoured? (HOU - SEA and SF - TB being two recent examples, but has this been the case consistently?)

by admin :: Sun, 10/30/2005 - 4:10pm

Just to make sure everyone understands, nearly every comment in this thread is getting kicked into the moderation queue because of the subject matter. So don't worry when your comments do not immediately appear, and you may want to consider not referring to comments by number because "new" comments may suddenly appear whenever we are able to approve comments waiting in the queue.

by TBW (not verified) :: Sun, 10/30/2005 - 6:57pm

New Rule:

No bets on teams that are underdogs to winless teams after week 3.

Even if it was a good value, you just feel so stupid when you lose....

by Michael T (not verified) :: Mon, 10/31/2005 - 6:59am


by John Gach (not verified) :: Mon, 10/31/2005 - 8:08pm

Regarding that too low opening line on TB ... I was a wee bit wrong, as it turned out. May I add to TBW's excellent new rule another one: stay away from games where Team A travels cross-country; Team B just lost by @ 50 points; Team B is a big dog (and earned the distinction). In retrospect I should have seen it as a potential trap; but in retrospect all mistakes would be erased. At least I avoided the CLE/HOU trap game, correctly smelling a rat. But, really, HOU & SF _both_ win the same week?! Needles to say, because of TB and my high risk interlocked wagers, this was my first big losing week. I'll report in full the all the gory details Tuesday, since I still have the PIT/BAL game active.

by TBW (not verified) :: Tue, 11/01/2005 - 3:16pm


my DVOA based picks went 8-6 overall and 4-1 in games where I favored a team by 10+ points. For the year, 19-9 overall and 7-1 in 10+ point games.

by John Gach (not verified) :: Tue, 11/01/2005 - 3:36pm

Postmortem too: Not a complete bloodbath, even after being bitten by the SF dog.
CIN & TB -15 units
CIN DAL SD TB PIT -5 units
CIN TB PIT - 20 units
CIN TB NE PIT -3 units
CIN PIT 10 units, won 4.9 units
CIN 30 units, won 6.8 units
Total: - 31.3 units, all because of TB, since every other contest was correctly called (though all the wins were closer than the spread and NE & PIT real squeakers). Still up @ 52 units on the year. I do need, though, to figure out a more rational, even algorithmic method for figuring actual wagers based on confidence level instead of just winging it.

by John Gach (not verified) :: Tue, 11/01/2005 - 5:14pm

In what is probably my last note with any theoretical pretensions, I'd like to return to TBW's #99 and the issues of predicting future events & probabilistic entanglement. My example of calling coin tosses was exactly the wrong one to use to illustrate what I had in mind. So, let's set the stage:

Instead of coin tosses, we have wins & losses by NFL teams. We begin with the following information: Every year we shall have a series of exactly 16 team-tosses. Almost all of these will result in a win or loss result. A small number -- in any year possibly none -- will result in a tie, a potential result that for the most part can safely be ignored. In any sequence of 16 team-tosses it is extremely unlikely that any team with end up with 16 wins and equally unlikely that any team with end up with 16 losses. Thus, one can predict with near certainty that even the best team(s) will lose one or more games and that even the worst will win one or more.

Going into the first of the 16 contests we know close to nothing about what the final record for 16 game-tosses will be for any of the teams. We have the not-too-useful preseason record and the prior year's record to go by. We have the occasional DVOA projection, often based on odd-seeming correlations, that Team A will perform much better and Team B much worse than in the prior year. How much such projections count as useful advance information still isn't clear (at least to me).

After several weeks of play, we are beginning, based on actual performance, to form ideas about which teams are for this 16-game series relatively better or worse than average. So this year after two games we already had strong reasons to suspect that SF & HOU were stunningly bad teams (despite SF's opening home win), that the injuries had finally brought NE back to the pack, that PIT was still going to be very solid, that CIN had indeed much improved, that the Ravens had, per DVOA's projection, really slipped, and so on.

After several more games we're starting to have enough information from actual play to make some reasonable projections about final records for all 16 game-tosses for particular teams. We have lots of other information as well: some teams have greater home field advantage than others; some teams demonstrate that for the most part they can't win on the road; some teams play much more consistently (whether well or poorly) than average; and so on.

By this point projected wins, variance, and strength of future schedule together begin to have some predictive value -- not, of course, for predicting actual outcomes but for predicting the likelihood of outcomes. I think of this as the prediction of a range of outcomes, including probable final records.

This is why I felt pretty strongly that SD would win at home against KC. Their DVOA consistently showed that they were, despite their record, one of the league's best teams, they had low variance, and, despite the difficult remaining schedule they still project to be a team with a winning record (9-7 or better). So there they are at 3-4, at home playing a team than which they are better. If they go to 3-5, then they have to go 6-2 the rest of the season just to reach 9-7. With several very difficult road games on their future schedule (ranked #1 in degree of difficulty this week), this is a game they really need to win in order to have a reasonable chance to fulfill their potential for the entire 16 game season. In other words this is a case where the future (all potential) impinges on the present (soon to be all fact) and allows one to assess an outcome partly by taking into account future possibility.

For similar reasons I should have been leary of TB at SF. TB has low variance but SF has the highest variance in the league. Furthermore if one examines its prior record, one thing jumps out immediately:
28-25 STL 3-42 PHI
31-34 DAL 14-31 ARI
3-28 IND 17-52 WAS
15-10 TB
This team on the road is completely hopeless, but at home they play teams much tougher. This is a case where DVOA and projected wins may be misleading. On the road SF may be setting some never-to-be-surpassed negative DVOA; at home they may be playing much closer to an average bad team or even mediocre team. I would be very curious to see the home & road DVOAs. Had I actually looked at the record presented above (omitting TB, of course), I would have demoted the game from the status of a near-certain win for TB. Bad as they are, SF could easily be 3-1 at home, with its only loss a convincing one to a superpower. A great deal of SF's variance must derive from the schizoid split between home & road performance.

The point is that in some cases future expectation affects the likelihood of the next game-toss's outcome. Note that I am not claiming that it necessarily affects the actual outcome, but rather the probability of the outcome. Those 16 game-tosses are, in pro football, not entirely discrete events as they are in coin-tossing. They are, as I wrote earlier, entangled, nonrandom events.

by TBW (not verified) :: Tue, 11/01/2005 - 10:07pm

I think your argument is basically a psychological one. A coin is an inanimate object. If it comes up heads 10 times in a row, it can't buckle down and try harder to come up tails the next time.

A football team can and does look at the schedule and think we can win this game, or we have no shot in that game, or we HAVE to win this game. As such, a football game is not a completely independent event, as it is affected by the level of effort, mood, etc. of the teams involved.

I agree with this completely and I think most people would.

by TBW (not verified) :: Tue, 11/01/2005 - 10:38pm

This is an interesting game:

Atlanta(-2) at Miami. I have Miami as an 8 pt favorite. At +110 at Pinnacle that's looking pretty tasty.

by John Gach (not verified) :: Wed, 11/02/2005 - 2:46pm

RE: TBW's #131. It's not exactly a psychological argument, though I see how one might view it that way, but more a Bayesian kind of argument. The 16 games of a season are linked, as they are to the results of other contests. So much is more or less obvious. The specific point that I was making, though, was that expectations about future events (such as how many games SD minimally should win, given the relative DVOA strength of the team coupled with its consistency of play) impinge on expectations about the next event in the series and _can_ sometimes provide information that is useful in predicting the result of the next game. In other words a bit of a wagering edge.

Not that I've done an actual analysis of all the games from this point of view, but impressionistically I'd venture that such opportunities aren't all that common during the course of the year (SEA/HOU was probably another one), so one needs to jump on them. Not that I did -- I only had SD in one cheap multi-team parlay, perhaps because I was treating my reasoning more as an abstract venture. Looking backwards, SD was a better bet than TB; instead I piled up on TB.

Miami does look enticing. Even just scanning down the list of games and odds today it kinda jumped out at me as a possible. So far, though, my only wager is 20 units on a parlay of SD & IND both to win, with a possible return of 41.8. The IND/NE game strikes me as a real DVOA opportunity, since the mere -3 spread doesn't seem to derive much from games actually played this year, but from the mystique of prior years and from the misperception that Manning hasn't played well against NE, which barely scraped out a win at home last week against BUF (and really shoulda lost). Of course, Edgerrin James could again fumble twice at the goal line with both fumbles lost ...

by urge (not verified) :: Thu, 11/03/2005 - 1:25pm

Allison's Precept:
The best simple-minded test of expertise in a particular area is the ability to win money in a series of bets on future occurrences in that area.

by TBW (not verified) :: Thu, 11/10/2005 - 10:01pm

FWIW, my system went 11-3 last week in including 2-0 in games where I felt there was a 10+ pt difference between teams. Overall, the record is 30-12, and 9-1 in 10+ pt games. This will be an interesting week with 5 10+ pt games. I'm beginning to think the way to go is parlays on these "sure thing" games.