Do I have enough data to start using the Kelly Formula?

[Dylan]

After a combination of shadow betting and betting five percent of my bankroll on one collegiate basketball game per day, I'm 28-15. This means I have a win percentage of 65.1162790697674% and my average bet is -105.26 (1.95002892533637), which calculates to a Kelly Percentage of 28.3977%.

I know Kelly is sort of a controversial topic on here, but do you think I have enough bet data to start using Kelly or a fractional Kelly?


Thanks for any advice you can dispense!

[mathdotcom]

Nope. See how you do on your next 43

[Dylan]

Nope. See how you do on your next 43

Thanks for your answer!

[FreeFall]

Nope.

http://en.wikipedia.org/wiki/Confidence_interval

[Dylan]

Nope.

http://en.wikipedia.org/wiki/Confidence_interval

Thank-you so much for your response.

If my understanding is correct, and correct me if I'm wrong, if I add and subtract my confidence interval to current win percentage, I'd get the expected range of success?

Really basic example, if my confidence interval was ±10, and my current win percentage is 70%, I'd could expect anything from 60% (70-10) to 80% (70+10) for future success percentage.

Thanks again for the answer and any future answers.

[mathdotcom]

Nope.

http://en.wikipedia.org/wiki/Confidence_interval

Correct answer but this ignores the likely fact that he and others have silently tried 100 'systems' before and lost without posting on the forum about it.

[Dylan]

Correct answer but this ignores the likely fact that he and others have silently tried 100 'systems' before and lost without posting on the forum about it.

I don't use systems. I have a spreadsheet of data for each sport league with over 5000 game's worth of data. Each data set calculates the margin against the spread, and the game with the largest margin of victory against the spread is the game I wager on.

[mathdotcom]

Thats the right way to go about it, but forgive my skepticism about your model if you could not answer the question you posed in this thread.

- mathy

[Dylan]

Thats the right way to go about it, but forgive my skepticism about your model if you could not answer the question you posed in this thread. - mathy

No worries. I just like to bounce ideas off other people.

[Duff85]

Should be able to get a bit more volume than that too? One game a day seems like you are leaving a ton of money on the table.

[Dylan]

Should be able to get a bit more volume than that too? One game a day seems like you are leaving a ton of money on the table.

I somehow missed this post, sorry.

I actually moddeled this a while back, if I bet more than one bet per day, my win percentage takes a dive. For example, at two bets per day my win percentage goes from ~65% to ~54%. For me, it just makes more sense to bet on a game per day.

Good question.

[FourLengthsClear]

I somehow missed this post, sorry.

I actually moddeled this a while back, if I bet more than one bet per day, my win percentage takes a dive. For example, at two bets per day my win percentage goes from ~65% to ~54%. For me, it just makes more sense to bet on a game per day.

Good question.

Does it not seem odd/strange that there is exactly one game each day where you are able to identify a very substantial edge? Why would that be?

[Dylan]

Does it not seem odd/strange that there is exactly one game each day where you are able to identify a very substantial edge? Why would that be?

It's the game with the largest margin of victory against the spread.

Here's an example. Let's say there are three games:

Titans vs. Patriots (-9.5)

Lions vs. Packers (-10.5)

Panthers vs. Giants (+1.5)

I predict the Titans will be 10, the Packers by 4, and the Giants will win by 3.
The margins of victory ATS would be as follows:

Titans 19.5

Packers: 6.5

Giants: -4.5

The largest margin of victory ATS would be the Titans, and that's the game I'd chose to bet on.

To sum it up: As the predicted margin of victory against the spread increases, the probability of winning increases as well.

Let me know if that makes sense.

[FourLengthsClear]

It's the game with the largest margin of victory against the spread.

Here's an example. Let's say there are three games:

Titans vs. Patriots (-9.5)

Lions vs. Packers (-10.5)

Panthers vs. Giants (+1.5)

I predict the Titans will be 10, the Packers by 4, and the Giants will win by 3.
The margins of victory ATS would be as follows:

Titans 19.5

Packers: 6.5

Giants: -4.5

The largest margin of victory ATS would be the Titans, and that's the game I'd chose to bet on.

To sum it up: As the predicted margin of victory against the spread increases, the probability of winning increases as well.

Let me know if that makes sense.

I get that. But what happens if you have three games on one day that have a similar projected margin ATS?

Let's say you have three games all of which you predict the margin to be 11, 12 and 12.5 points respectively. Would you only play the game with the 12.5 point 'advantage'?

Obviously, I have no idea how you are deriving your predictions but as a rule, if you are way off the market's prediction (the spread), it would tend to indicate that something is wrong. Obviously your results to date refute that so I would suggest you just continue what you are doing but to put into context, in an NFL match where you calculate a 10 point advantage against the spread would equate to an edge of well over 60% against -110 odds.

Are you keeping a record of all the games that you are using this method to calculate or only the ones that you end up wagering on?

[Dylan]

I get that. But what happens if you have three games on one day that have a similar projected margin ATS? Let's say you have three games all of which you predict the margin to be 11, 12 and 12.5 points respectively. Would you only play the game with the 12.5 point 'advantage'? Obviously, I have no idea how you are deriving your predictions but as a rule, if you are way off the market's prediction (the spread), it would tend to indicate that something is wrong. Obviously your results to date refute that so I would suggest you just continue what you are doing but to put into context, in an NFL match where you calculate a 10 point advantage against the spread would equate to an edge of well over 60% against -110 odds. Are you keeping a record of all the games that you are using this method to calculate or only the ones that you end up wagering on?

To answer your first question, yes I would only place money on the game with a 12.5 point advantage.


I have one spreadsheet for each sport where I calculate the winner and margin of victory based off stats.


On "if you are way off the market's prediction (the spread), it would tend to indicate that something is wrong."...

There have been times I've been surprised at the calculation. A really good example of that occurred last year when the Miami Heat faced the Cavaliers in Cleveland. Miami were the favourites, 13.5, yet my model suggested a massive Cleveland win, 75 points. I took the moneyline on the Cav's. The Cav's didn't win by 75, but by 12.

There have been other times too, where situations like that burned me, but overall when my predicted score is way off the spread I'm 55-48. (Not necessarily the games I bet on.)


"Are you keeping a record of all the games that you are using this method to calculate or only the ones that you end up wagering on?"

Yes, I record every game and not just the games I bet on. NCAA Basketball for example, overall I'm 3150-2941.


Hopefully that makes sense. Let me know if you have further questions.

[FourLengthsClear]

To answer your first question, yes I would only place money on the game with a 12.5 point advantage.


I have one spreadsheet for each sport where I calculate the winner and margin of victory based off stats.


On "if you are way off the market's prediction (the spread), it would tend to indicate that something is wrong."...

There have been times I've been surprised at the calculation. A really good example of that occurred last year when the Miami Heat faced the Cavaliers in Cleveland. Miami were the favourites, 13.5, yet my model suggested a massive Cleveland win, 75 points. I took the moneyline on the Cav's. The Cav's didn't win by 75, but by 12.

There have been other times too, where situations like that burned me, but overall when my predicted score is way off the spread I'm 55-48. (Not necessarily the games I bet on.)


"Are you keeping a record of all the games that you are using this method to calculate or only the ones that you end up wagering on?"

Yes, I record every game and not just the games I bet on. NCAA Basketball for example, overall I'm 3150-2941.


Hopefully that makes sense. Let me know if you have further questions.

OK so in NCAAB you have a substantial sample. If you have a database, what patterns/trends do you see in the relationships between your predicted result, the closing line and the actual result?

[Dylan]

OK so in NCAAB you have a substantial sample. If you have a database, what patterns/trends do you see in the relationships between your predicted result, the closing line and the actual result?

Sorry for the late response.

Here's a few patterns:
The higher my predicted result, the greater chance of victory I have.
The higher the closing spread, the greater chance of victory I have.

My predicted result and the actual score have a correlation of R^2 = 0.131346639597348
The closing line and the actual score have a R^2 of 0.281201677023064

Predicted scores higher than 44 have the highest win % at 0.743589744., but that's only happened 39 times.

I can't think of any other obvious patterns, but if you ask for any (e.g., What's your win % when the predicted score is 10?) I'll be happy to provide.

[sixrolla]

Dylan, I think you should take Elway's advice for Tebow and stop being scared to pull the trigger. Unless you are concerned about your discipline, I don't see the value in limiting yourself to 1 bet per day. If your bets have good value, you are gonna be missing out in the long run by limiting yourself.

[Dylan]

Dylan, I think you should take Elway's advice for Tebow and stop being scared to pull the trigger. Unless you are concerned about your discipline, I don't see the value in limiting yourself to 1 bet per day. If your bets have good value, you are gonna be missing out in the long run by limiting yourself.

As much as I appreciate your confidence in me, I know my edge is reduced by enough that it's more profitable for me to bet on one game than to bet on more than one. For example, at two bets per day my win percentage goes from ~65% to ~54%.

Thanks though.

[mathdotcom]

As much as I appreciate your confidence in me, I know my edge is reduced by enough that it's more profitable for me to bet on one game than to bet on more than one. For example, at two bets per day my win percentage goes from ~65% to ~54%.

Thanks though.

This sounds really bizarre.

If you feel you need a huge edge to be comfortable betting, I suggest you give up on any major market and instead focus on props or an off-beat market.

[bztips]

I don't there's anything bizarre about it at all. There is always going to be uncertainty surrounding your point estimate; and the closer your point estimate is to the posted number, the more likely an uncertainty band of a given size (confidence interval) around it will include that number, which means it's less likely that you have a real edge.

[Dylan]

This sounds really bizarre. If you feel you need a huge edge to be comfortable betting, I suggest you give up on any major market and instead focus on props or an off-beat market.

I don't need a huge edge. I just know the most profitable way for me to make money is by only betting on the game with the highest margin of victory each day.

[Dylan]

I don't there's anything bizarre about it at all. There is always going to be uncertainty surrounding your point estimate; and the closer your point estimate is to the posted number, the more likely an uncertainty band of a given size (confidence interval) around it will include that number, which means it's less likely that you have a real edge.

Thanks for the reassurance. It makes sense to me.

[sixrolla]

If you are calculating an edge of more than 10% consistently your formulas are probably flawed. Like math said in major markets you just aren't going to see huge edges very often.

[sixrolla]

Dylan, it isn't about your winning percentage. Would you rather go 9-1 and hit 90%, or go 57-43 and hit 57%? Unless you are putting a disproportionate amount of funds into that 1 bet, you are likely missing out on some good opportunities. That said, if you are a losing bettor, limiting yourself to 1 really small bet is a great strategy.

[Dylan]

Dylan, it isn't about your winning percentage. Would you rather go 9-1 and hit 90%, or go 57-43 and hit 57%? Unless you are putting a disproportionate amount of funds into that 1 bet, you are likely missing out on some good opportunities. That said, if you are a losing bettor, limiting yourself to 1 really small bet is a great strategy.

It's not that big of sample difference. As well, I bet on more than one sport per day, so it's not as if I'm going nowhere.

Like I said, I've modeled every nearly every combination, (one game per sport, two games per sport, two-game parlay per sport, etc.) and the most profitable way was one bet per day per sport.

Again, though, I appreciate your confidence in me.

[sixrolla]

You're the man.

[Dylan]

You're the man.

Haha, thanks.

[FourLengthsClear]

Sorry for the late response.

Here's a few patterns:
The higher my predicted result, the greater chance of victory I have.
The higher the closing spread, the greater chance of victory I have.

My predicted result and the actual score have a correlation of R^2 = 0.131346639597348
The closing line and the actual score have a R^2 of 0.281201677023064

Predicted scores higher than 44 have the highest win % at 0.743589744., but that's only happened 39 times.

I can't think of any other obvious patterns, but if you ask for any (e.g., What's your win % when the predicted score is 10?) I'll be happy to provide.

OK, so based on the data you have - are you considering any adjustments to your method of calculating your 'predicted result'? Hitting at 75% on any subset is fantastic but needless to say this figure should be over 99% if you really had a 44 point edge against the spread. As mentioned above, huge differences between your predictions and the closing line are indicative of flaws. Seeking to iron out those flaws would make your methods more usable over a wider range of calculated edges.

Going back to your initial question, Kelly should not be utilised unless/until there is reasonable basis on which you are able to determine your edge on a wager to wager basis. If your current method results in a calculated edge on 5 or 6 points meaning no edge in reality (as is indicated by the big drop in win rate when considering/making more than one wager per day) then you really don't have that reasonable basis.

I don't there's anything bizarre about it at all. There is always going to be uncertainty surrounding your point estimate; and the closer your point estimate is to the posted number, the more likely an uncertainty band of a given size (confidence interval) around it will include that number, which means it's less likely that you have a real edge.

This is true at the margin but when we are talking about differences of 20, 30 or 40 points, I am with Mathy.

[Dylan]

OK, so based on the data you have - are you considering any adjustments to your method of calculating your 'predicted result'? Hitting at 75% on any subset is fantastic but needless to say this figure should be over 99% if you really had a 44 point edge against the spread. As mentioned above, huge differences between your predictions and the closing line are indicative of flaws. Seeking to iron out those flaws would make your methods more usable over a wider range of calculated edges. Going back to your initial question, Kelly should not be utilised unless/until there is reasonable basis on which you are able to determine your edge on a wager to wager basis. If your current method results in a calculated edge on 5 or 6 points meaning no edge in reality (as is indicated by the big drop in win rate when considering/making more than one wager per day) then you really don't have that reasonable basis. This is true at the margin but when we are talking about differences of 20, 30 or 40 points, I am with Mathy.

Thanks for the response.

I haven't really considered my method for calculating predicted result. I've had enough success with it in the past, just was curious if I was betting too small of an amount per bet.

The differences between closing losing lines and my predictions happen so infrequently, less than one-percent of the time. I doubt it's indicative of flaws in my method.

If I were to use Kelly, I'd adjust it the average odds and my win% with each bet, so if I were to go on a losing steak, my Kelly Percentage would drop.

However, if you guys seem to think I have what it takes for a Kelly percentage, than I'll heed your advice. Thanks!

[mathdotcom]

e way off the market's prediction (the spread), it would tend to indicate that something is wrong."...

There have been times I've been surprised at the calculation. A really good example of that occurred last year when the Miami Heat faced the Cavaliers in Cleveland. Miami were the favourites, 13.5, yet my model suggested a massive Cleveland win, 75 points. I took the moneyline on the Cav's. The Cav's didn't win by 75, but by 12.

How has anyone not commented on this??? Has this ever even happened in the history of the NBA?

This guy's model is garbage, and his money management is garbage too. If he had 3 games with pt edges of 10, 10.5, 11 he would only bet the game with the 11 point edge. He's trying to be conservative yet by doing this he's actually increasing variance, when instead he could split his unit 3 ways and have basically the same EV with less variance. Plus he took the moneyline which must have been like +800 and juiced to the hilt instead of taking the points.

Anybody who thinks like this regarding money management, and has a model that predicts 130-55 NBA scores is out to ******* lunch and you guys need to give him a dose of reality.

[mathdotcom]

It's not that big of sample difference. As well, I bet on more than one sport per day, so it's not as if I'm going nowhere.

Like I said, I've modeled every nearly every combination, (one game per sport, two games per sport, two-game parlay per sport, etc.) and the most profitable way was one bet per day per sport.

Again, though, I appreciate your confidence in me.

This is wrong for so many reasons

[Dylan]

How has anyone not commented on this??? Has this ever even happened in the history of the NBA? This guy's model is garbage, and his money management is garbage too. If he had 3 games with pt edges of 10, 10.5, 11 he would only bet the game with the 11 point edge. He's trying to be conservative yet by doing this he's actually increasing variance, when instead he could split his unit 3 ways and have basically the same EV with less variance. Plus he took the moneyline which must have been like +800 and juiced to the hilt instead of taking the points. Anybody who thinks like this regarding money management, and has a model that predicts 130-55 NBA scores is out to ******* lunch and you guys need to give him a dose of reality.

http://en.wikipedia.org/wiki/Outlier

[sharpcat]

Assuming you can accurately estimate your edge Kelly is going to help maximize the growth of your bankroll while minimizing your risk of ruin. One of the worst things you can do when using Kelly is overestimate your edge and over bet because it will greatly increase your risk of ruin and the destruction of your bankroll, the inability to accurately estimate your edge is one of the reasons why some choose to use fractional Kelly.

Considering that your example above suggests that your "model" gauges your "edge" depending on the point differential between the actual line and your prediction but does not take the value of each individual point into consideration, along with the huge differential in your prediction compared to an efficient market suggests that you have some serious flaws. Absolutely no way in hell your model is spotting inefficiencies in the market that are allowing you to have a 28% edge unless you are betting on the peewee football league with max wagers of $50.

My advice would be to chalk your 28-15 record up to luck and go back to the drawing board you are nowhere near being ready for Kelly.

[AlwaysDrawing]

I agree with Dylan. USC is -10 points today, but I have them at -86 in my model, which is the highest for today. The Kings are -68, but I'm not going to bet them because only one bet a day is +EV.