Fezzik's 2011/2012 football record

In the case of the Rays, they had a slow start, so if you were only betting on the Rays, you would have had a smaller bankroll when they started winning. This is why you end up with less money at the end. For the Indians, their fast start meant you were betting more when they started losing, so again you end up with less money at the end. However, this is not always the case. Some losing teams lose even more often as the season progresses. Some winning teams win more often as the season progresses. Teams that "regress to the mean" as Tampa and Cleveland did only seem to do so because you find out the mean at the END of the season. The system I would like to try is one that shows expect W-L vs actual W-L, because perhaps the lines are overly weighted to the teams actual W-L. Surely this has been tried by now by many.

Keep in mind this had nothing to do with the actual lines for these games. It's simply a way to sample a pattern of wins and losses that end the season at 56% (a good year for a bettor) and at 49% (a bad year for a bettor). I'm simply using their games to simulate if a bettor won or lost a bet they placed as bets # 1, 2, 3, through #162, etc. It has nothing to do with the actual baseball games at all. The only thing that comes into play is how they won and lost to sample a betting streak that wins and losses.

Summary through Monday, Nov. 21: W/L: One-wgt: 50-53-5, Two-wgt: 31-37-3, Three-wgt: 12-12-0, Four-wgt: 1-6-0 Overall weighted W/L: 152-187-11, 44.8%, -57.91u Overall straight bets unweighted W/L: 94-108-8, 46.5% Overall teasers unweighted W/L: 4-4 NFL Regular season record: 113-125, 47.5%, (70-74-6 unweighted), -33.51u CFB record: 27-38, 41.5%, (18-24-1 unweighted), -15.35u CFL record: 6-7, 46.2%, (2-3-0 unweighted), -1.70u NFLX record: 4-7-1 unweighted, -7.35u

[QUOTE=Sixth Sense;48488]Keep in mind this had nothing to do with the actual lines for these games. It's simply a way to sample a pattern of wins and losses that end the season at 56% (a good year for a bettor) and at 49% (a bad year for a bettor).[/QUOTE] Keep in mind that you're just datamining two trends where teams had the pattern you wanted. So basically your datamining is well done, your math is ignorant. The problem of optimal bet sizing (as a percentage of bankroll) with a known win percentage involving independent events over the long term was solved by JL Kelly, Jr in a paper written in 1956. As Irish Tim correctly points out, theoretically the optimal solution is to resize your bets after every change in bankroll.

I didn't datamine a pattern that I wanted. I chose two winning percentages that are on one side or another of what we can expect from a bettor - a winning bettor hitting at 56% or a losing bettor hitting at 49%. The formula is pretty easy and I'm happy to run every baseball team through the formula, which would give us 30 different scenarios of 162 bets - good, bad and middle of the road. I believe (could be wrong) that the same scenarios would play out. Like I said above, maybe my math is ignorant, and I am looking for advise on that. But, when I take two random samples (again winner and loser) and get numbers that suggest keeping the levels the same is the best way to do this, I'm looking for reasons as to why my math is bad. Not saying it isn't, but looking for suggestions as to why it is bad. At this point, we have a couple of people saying you should vary the size. That's it - no proof just saying that is the best way. I've provided actual numbers that suggest otherwise. As I've said many times, I could be wrong with my assumptions but my math is correct, I believe, so please show me where I am wrong so I can learn if I am doing something wrong. This has nothing to do with bet sizing with a known percentage. That has more to do with the Kelly criterion, where this is just saying one bets 2% (or whatever number) on every bet. Obviously the 2% comes from something but I'm sure it's not coming from the Kelly study or the percentage would probably be higher.

[QUOTE=Sixth Sense;48530]I didn't datamine a pattern that I wanted. I chose two winning percentages that are on one side or another of what we can expect from a bettor - a winning bettor hitting at 56% or a losing bettor hitting at 49%. The formula is pretty easy and I'm happy to run every baseball team through the formula, which would give us 30 different scenarios of 162 bets - good, bad and middle of the road. I believe (could be wrong) that the same scenarios would play out. [/QUOTE] You just admitted to datamining the two by choosing one winning and one losing season. How could you know to do that without pre-determining (i.e.datamining) their records?. You also stop at two teams that suit your argument, anyone with any knowledge of probability theory would then test out-of-sample data involving many other teams/seasons in an attempt to validate the findings of the two teams. In any case, these two season records prove nothing. They are small samples and statistically insignificant. I ran several simulations with a 2% betting size. In fact, in a 162 game stretch, a 56% capper will only perform better by bet sizing roughly 70% of the time. If you had tested 500 games, you would have found bet sizing produces more profit around 94% of the time. And with 1000 games, bet resizing produces more profit 99.5% of the time. Of course, a 49% bettor should not be betting anything unless he bets mostly dogs. [QUOTE=Sixth Sense;48530] At this point, we have a couple of people saying you should vary the size. That's it - no proof just saying that is the best way. I've provided actual numbers that suggest otherwise. As I've said many times, I could be wrong with my assumptions but my math is correct, I believe, so please show me where I am wrong so I can learn if I am doing something wrong. This has nothing to do with bet sizing with a known percentage. That has more to do with the Kelly criterion, where this is just saying one bets 2% (or whatever number) on every bet. Obviously the 2% comes from something but I'm sure it's not coming from the Kelly study or the percentage would probably be higher.[/QUOTE] This has everything to do with bet sizing. The problem with the Kelly criterion and sports betting is that no one really knows their exact winning percentage because of small sample sizes and the dynamic and changing nature of lines making and handicapping. Thus the "known" percentage edge in sports is very different than say blackjack where the winning percentage can be determined very accurately with millions of simulated hands. So we can make the argument that bet sizing changes based solely on Kelly and recent results (and therefore bankroll size) are not optimal for sports betting. However, the longer the over or under performance (to an assumed long-term rate) trends last, the more necessary it is to adjust bet sizing to accomodate the trend, i.e. bankroll growth/reduction.

Summary through Monday, Nov. 28: W/L: One-wgt: 52-54-5, Two-wgt: 31-38-4, Three-wgt: 13-12-0, Four-wgt: 1-6-0 Overall weighted W/L: 157-190-13, 45.2%, -56.11u Overall straight bets unweighted W/L: 97-110-9, 46.9% Overall teasers unweighted W/L: 4-4 NFL Regular season record: 118-128, 48.0%, (73-76-7 unweighted), -31.71u CFB record: 27-38, 41.5%, (18-24-1 unweighted), -15.35u CFL record: 6-7, 46.2%, (2-3-0 unweighted), -1.70u NFLX record: 4-7-1 unweighted, -7.35u

Summary through Monday, Dec. 5: W/L: One-wgt: 52-56-5, Two-wgt: 34-39-4, Three-wgt: 14-15-0, Four-wgt: 1-6-0 Overall weighted W/L: 166-203-13, 45.0%, -61.46u Overall straight bets unweighted W/L: 101-116-9, 46.5% Overall teasers unweighted W/L: 4-5 NFL Regular season record: 127-141, 47.4%, (77-82-7 unweighted), -37.06u CFB record: 27-38, 41.5%, (18-24-1 unweighted), -15.35u CFL record: 6-7, 46.2%, (2-3-0 unweighted), -1.70u NFLX record: 4-7-1 unweighted, -7.35u

Summary through Monday, Dec. 12: W/L: One-wgt: 52-56-5, Two-wgt: 37-43-4, Three-wgt: 15-16-0, Four-wgt: 1-6-0 Overall weighted W/L: 175-214-13, 45.0%, -66.06u Overall straight bets unweighted W/L: 105-121-9, 46.5% Overall teasers unweighted W/L: 4-5 NFL Regular season record: 136-152, 47.2%, (81-87-7 unweighted), -41.66u CFB record: 27-38, 41.5%, (18-24-1 unweighted), -15.35u CFL record: 6-7, 46.2%, (2-3-0 unweighted), -1.70u NFLX record: 4-7-1 unweighted, -7.35u

[QUOTE=ComptrBob;48603]You just admitted to datamining the two by choosing one winning and one losing season. How could you know to do that without pre-determining (i.e.datamining) their records?. You also stop at two teams that suit your argument, anyone with any knowledge of probability theory would then test out-of-sample data involving many other teams/seasons in an attempt to validate the findings of the two teams. In any case, these two season records prove nothing. They are small samples and statistically insignificant. I ran several simulations with a 2% betting size. In fact, in a 162 game stretch, a 56% capper will only perform better by bet sizing roughly 70% of the time. If you had tested 500 games, you would have found bet sizing produces more profit around 94% of the time. And with 1000 games, bet resizing produces more profit 99.5% of the time. Of course, a 49% bettor should not be betting anything unless he bets mostly dogs. This has everything to do with bet sizing. The problem with the Kelly criterion and sports betting is that no one really knows their exact winning percentage because of small sample sizes and the dynamic and changing nature of lines making and handicapping. Thus the "known" percentage edge in sports is very different than say blackjack where the winning percentage can be determined very accurately with millions of simulated hands. So we can make the argument that bet sizing changes based solely on Kelly and recent results (and therefore bankroll size) are not optimal for sports betting. However, the longer the over or under performance (to an assumed long-term rate) trends last, the more necessary it is to adjust bet sizing to accomodate the trend, i.e. bankroll growth/reduction.[/QUOTE] Good stuff Bob as usual. Cheers