Overall, I think it's an excellent decision to bring on board a QB for future development who looks to hold good potential. What becomes of him after that is up to him and the coaching staff In depth, I'm not much of a draft guy. It's mostly a crap shoot as far as I'm concerned. You really don't know what you have until you face the professionals.
Cbrad, Since we previously have established that QB is the biggest individual player that makes a difference to a team’s performance, I suggest a modification to your model. Option 1 (Conservative) Do not bet on teams where the Starting QB is different to the QB responsible for [threshold level] of the previous 10 starts. My gut is that the threshold level should be 6 or 7 of the previous 10. Option 2 (Aggressive) If we want to be bet on those games you would add a bonus for returning a starting QB and a penalty for having a backup start the game.
Yeah.. let's gather data first and then we can look at how to improve on this simpleton approach. I deliberately recorded the moneylines for all those games so I can go back and test different modifications of this strategy.
OK, time to continue this experiment. So the tally for week 1 was 3 wins, 4 losses and 1 push (ties are pushes in 2-way moneylines) for a total loss this week of $69. So right now I'm at -$69. Any patterns? Well, it looks like Vegas did better by 3:2 when there was a major change at QB: stats correctly predicted Eagles and Bengals would win, but failed to predict the Dolphins, Broncos and Redskins winning. And the 4th loss came with the Raiders who of course went into rebuilding mode which was unknown to the stats. Stats did well in games where there wasn't a big change (Steelers and Jaguars games) by either making money or not losing money. Will that continue? It might. For week 2 there are MANY more games where both the favorite and underdog are extremely close to historical averages so the rule I'm using (+200 to -200 moneylines AND at least 5% difference in implied win% to historical win%) says we can only bet on 5 games this week, which is 3 fewer than week 1. And if that trend of more Vegas odds closer to historical averages continues, then it really shows why it's hard to beat Vegas! Either way, here are the picks for week 2 based on current moneylines (once again every bet to win $100): https://www.sportsbook.ag/sbk/sportsbook4/nfl-betting/week-1-lines-nfl-game-lines.sbk Bet $51 on Panthers (+197) over Falcons Bet $51 on Lions (+197) over 49ers Bet $51 on Raiders (+197) over Broncos Bet $155 on Cowboys (-155) over Giants Bet $65 on Seahawks (+155) over Bears Personally.. I'd NEVER bet on the Lions or Raiders to win after week 1 lol. But there it is. All this is still way too small sample size, but this experiment might end up showing that to beat Vegas using historical odds you want to bet on games where Vegas odds differ only by a small percent from historical odds, meaning that the time it takes to make a profit would be VERY long. We'll see!
Cbrad, Would it be possible to do the results as Case (1) pure stats Case (2) no bets when there are significant QB changes - let’s say same QB as for at least 6 out of the last 10.
Unless you're thinking of something else, the most obvious "pure stats" approach would be success rate as a function of difference between implied odds (by moneyline) and historical odds. I mean that kind of data would be gold, but it would require a TON of data. Theoretically I can do that if there is some database of archived NFL odds for every game, but man.. that would have to go back several decades at least, and I've never found a source like that. Case 2 is easy to keep track of so we can do that. For week 1 that would include only 3 games: Steelers, Jaguars and Raiders. The result is being up by $43 after week 1.
It makes sense to throw out the teams with new QB's for the predicted winning team since that makes the past wins/losses unreliable. That means we'd never bet on the Dolphins, Jets, etc. since they had a change at the helm...or maybe you use each QB's last 10 games as the indicator instead of team wins (so for RT, you'd look at game one then 9 games in 2016). Or maybe you do both. (Last 10 team games + last 10 starting QB results) /2 = X. That way you'd have better data for teams who switched QB's this season. Now for the big question. For the Steelers, Raiders and Jags, what were the formula's expected win percentage based on the last 10 games? In other words, who was the biggest favorite to win, second biggest favorite, etc? Did your "safest" bet win? We need a little more data to quantify this into a betting system. Shouldn't we have a control group as well? In other words, what we'd bet without the formula on the same games? If so, I want to be the control group!!!! =) I'd take- Panthers 49ers Broncos Cowboys Bears
Oh I see.. so just plot Vegas lines vs. last 10 games, but not the matchup. Yeah that should have enough data points. I'm guessing the means will be about the same but the 95% confidence intervals might differ and that would be interesting info. I'll do that at the end of the season though.. no point looking at it now with so little data. Yeah we'll keep track of both strategies, the one I started with as well as the strategy of throwing out games where a QB for one of the teams didn't start the majority of last 10 games. The passer rating idea is worth looking at but not with these stats. You'd have to redo posts #62 and #72 including passer rating first. Sadly I don't have that data yet so that idea will have to wait till another time. I did however redo post #62 with points scored rather than wins and the curve, which is obviously smooth since you don't have that "even" vs. "odd" distinction, peaks at 11 games back. So that's further evidence looking 10-11 games back is optimal. Steelers = 87.84%, Jaguars = 67.7%, Raiders = 43.1%. So the "safest" bet was a push (didn't lose money though), and the 2nd safest won. The degree to which a bet is safe is going to dovetail Vegas with few exceptions. I'm just looking at cases where they differed by at least 5%, and often it's just barely above 5% (e.g., implied Vegas odds for Jags was 62.3%!). LOL. Yeah I'd agree except I think the Seahawks will win. However.. you have to look at the moneylines here and then consider whether you'd risk the money (even if it's fake!). Using the moneylines at the time of post #84 this is what the "control" looks like: Panthers you'd bet $51 to win $100 49ers you'd bet $235 to win $100 Broncos you'd bet $235 to win $100 Cowboys you'd bet $155 to win $100 Bears you'd bet $175 to win $100 As you can see.. there's a hefty price for betting on favorites!
I think the goal here would be to evolve the base strategy through the season to get it as optimal as possible. As you said though, 16 weeks is a very small sample so that's pretty hard to do. I'm hoping we can make some adjustments over the year anyway- or at least get some insights on things to try next year. The Raiders are a tough one since they're going full rebuilding mode with the new coach- I think the safest bet would be to not bet them this year. =) I don't gamble at all, so of course I'd take the fake risk. I'm keeping up with you on this more for fun than for the actual strategy, but of course I want it to work as well. =)
lol.. Moral of this week is the KeyFin Control (KFC) is the way to go! The KFC method is up $349 in a single week LOL. Keep it up dude. The rule I'm using on the other hand has me down -$187. Pauly's suggestion of leaving out games where the starting QB for one of the teams didn't start at least 6 of the last 10 games is at -$24. I have no idea what random variation with this stats approach looks like, but if the next 2 weeks continue to show such losses, then it's pretty clear that the wrong strategy to bet is when Vegas and the stats differ by a lot.. apparently you'd have to win over time on tiny differences. Anyway, for this week's games there are so many heavy favorites (more than -200) or heavy underdogs (more than +200) so there are comparatively fewer games to bet on using the rule I'm using: https://www.sportsbook.ag/sbk/sportsbook4/nfl-betting/nfl-lines-nfl-game-lines.sbk But here are the picks: $69 on Jets over Browns $80 on Bengals over Panthers $80 on Redskins over Packers $80 Steelers over Bucs Notice a trend? In the last 2 weeks, 8 out of 9 picks are all underdogs. Not sure what that says.
The other thing I notice about those games- I wouldn't bet any of them in real life (not that I gamble anyway). The Jets/Browns is a disaster waiting to happen....I can make a strong argument for either team. Bengals/Panthers and Steelers/Bucs are also very strong teams clashing early. So I'll pick solely off team stats (total offense, team D, turnover differential). Tampa over Pitt (two of the top offenses BUT Tampa is +2 turnovers to Pitt's -4) Skins over Packers (Washington has been really stingy on D...but crazy tough to pick against Rodgers) Bengals over Panthers (the stats are pretty even....this is a pick-em game and shouldn't be bet on) Jets over Browns (my brain says to pick the Browns, but the Jets lead almost every meaningful category) What I would be inclined to bet in real life- Fins over Raiders Jax over Titans Pats over Lions Vikings over Bills I really think this formula has to include something with total defense to be more accurate. How about (QB rating last 10) minus (allowed QB rating last 10)? Calculate for both teams and the bigger number should win.
Yeah I'm not betting using this strategy in real life either. I will though if after tons of experimentation something is shown to work! As far as that QB rating suggestion, the database I have for game records doesn't include QB rating so that's an experiment we can't do atm.
Okay, what about- (total points scored last 10) - (total points allowed past 10)? That way, we're bypassing all the data and trying to figure out what matters/doesn't matter...we know points matter more than anything. For instance, the Fins are +15 the past two weeks where the Raiders are -21. That's a 36 point swing and surely a good indicator that the Fins should perform better. Of course, it's also biased since the Raiders got blown out week 1 and the sample size is too small. If we looked at that over 10 weeks, there should be some pretty good patterns there. With each team having an actual point value, we could bet the top 5 differentials each week. I'll try to figure out an easy way to do that- maybe I'll just make a spreadsheet that will give us the year's scoring totals for/against for each team and the final value. That shouldn't be too hard.
Yeah I've already done that here: https://www.thephins.com/threads/thoughts-on-the-dolphins-titans-game.93470/page-2#post-3087892 Similar predictive power at about 10 games back (technically using win% is a slightly better predictor at N=10). Now, one thing we could try is using both points AND wins. I can do that, but what specifically do you want to try? Those two things are so highly correlated you're not really improving overall predictive power.
That's right....I forgot about that one. I'm wondering if looking at overall averages of all 32 teams for "x years" blurs the results though. All we'd be interested in is the most lopsided match-ups, the low-hanging fruit, per se. Maybe we only bet the top 3 biggest differentials or something like that. I'll take it on as a pet project this season and update weekly as well. I'm kinda excited to prove/disprove the theory!
Fantastic! Love it when others do some stats work (keep in mind I might chime in with some critique here and there though lol). Anyway, as far as the database I'm using it's the "Schedule & Game Results" table at PFR: https://www.pro-football-reference.com/teams/mia/2017.htm I have that for every team since 1970. So I can look at anything in that table, except the last 3 columns dealing with Expected Points (I'd rather have raw stats). I can also filter across NFL history by anything like score differential etc.. Either way, whatever approach you use, we have to compare to Vegas and I'd suggest you use the same source I do: sportsbook.ag (best to wait till late Tuesday or Wednesday.. they don't have lines for many matchups if you look too early): https://www.sportsbook.ag/sbk/sportsbook4/nfl-betting/nfl-lines-nfl-game-lines.sbk If you need any help on translating odds to moneylines or so just ask.
I'll build a Google Docs spreadsheet and manually enter each week's scores, then create the formulas to give us the final "team strength" number for a few different variables (maybe past ten games, past 5 games, past 3 games, on the season). That way we can see lots of stats at once and look for micro trends that may be hidden in the data. I have a feeling that you're wrong about the past 10 games being the best indicator because we're looking at too much at once. For instance, the Fins in weeks 1-5 in 2016 were nothing like the team that played weeks 6-10. I think we need to be able to see those smaller trends to really get a handle on what to wager, when to wager, etc. And maybe I'm completely off here, but it's still not a waste of time since it will prove something regardless. So we're learning either way and it's simple enough for a non-stat guy like me to keep up with.
OK, time to continue this experiment (which isn't going that well so far LOL). However, for the first time in 3 weeks there was a profit. Nevertheless, overall this method is down -$136. Pauly's suggestion of leaving out cases where a QB didn't start at least 6 of last 10 meant only one game was bet on in week 3: the Bengals. That method is also down -$104. The KeyFin Control (KFC) lost some but is still making a hefty profit of +$193. Either way, I fear the worst this week because the bets look BAD (I'd probably bet on most of the opposing teams this week): $120 on Colts (-120) over Texans $56 on Bengals (+177) over Falcons $60 on Titans (+170) over Eagles $67 on Cardinals (+150) over Seahawks $53 on Broncos (+190) over Chiefs Like I said.. those bets look BAD lol, but so be it!
Just got back at 1am today.. catching up on sleep. Yeah, if you're referring to how this betting approach is going I stopped posting updates since I didn't think too many were that interested, but profit/loss-wise things leveled off after the initial losses, but this approach isn't beating Vegas. I think there's no choice but to try to incorporate more information. The database I have makes it easy to look at anything in the "Schedule & Results" table (except "Expected Points" columns) for every team from 1970 onwards.. all other information in the boxscores I don't yet have in easy to parse format: https://www.pro-football-reference.com/teams/mia/2017.htm So I'm playing around with ideas on what to try next. I've been thinking of two possibilities: 1) weighting past performance by game played to naturally incorporate a recency effect instead of just taking a block of "past N games", or 2) weighting past performance based on some measure of opponent strength. Not sure yet. Either way, I have to put some effort into getting a database of past Vegas odds so I can just test strategies up front instead of waiting for games to be played. Sites that have archived odds are either hard to parse or ask for $$ to put them in easy to parse format. We'll see what happens.. this isn't a short term project so I'm taking it one small step at a time.
Nice...I love Japan.... My wife, 5 kids, and I just got back from 8 days in Disney...I'm trying to catch up on sleep too. And I'm interested in this thread so please keep posting!
