Reasons for Low Final Score Predictions

**CHUBNUT** · 04-30-11, 05:09 PM

Its good your enjoying your betting, the vast majority of people dont. however dont mistake that with some bigger expectation as winning takes more than getting models from books, you only have to read the author of conquering risk asking for input on wind variance to see how desperate his models have got.
believe me, theres a huge industry out there trying to convince you there's an easy way to win at sports betting. well there isnt an easy way, get that in your head and realise winning takes a monumental effort that have negative consequences in other parts of your life.

your choice.

**Kaplan** · 04-30-11, 09:32 PM

Go to covers.com and look at my CFB bowl thread if you want to see what winning looks like.

**LLXC** · 04-30-11, 09:42 PM

Linear/Logistic Regression is pretty primitive...

**mjespoz** · 04-30-11, 10:58 PM

Originally posted by LLXC

Linear/Logistic Regression is pretty primitive...

LLXC,
Would you use logistic regression to predict a continuous dependent variable?? Not a very intelligent statement to make given the topic of discussion. Linear regression can be useful if used correctly, but yes, there are other statistical methods that are more appropriate.

Cheers,
mjespoz.

**BediMindtricks** · 04-30-11, 11:48 PM

I wouldn't suggest using logistic regression as a solid form of prediction.. I believe there are better forumals to use in this situation.

**LLXC** · 05-01-11, 12:04 AM

Originally posted by mjespoz

LLXC,
Would you use logistic regression to predict a continuous dependent variable?? Not a very intelligent statement to make given the topic of discussion. Linear regression can be useful if used correctly, but yes, there are other statistical methods that are more appropriate.

Well, I can't give it all away...but yea, you're right: it does work for some things, including primitive capping methods. It's great for stuff like...predicting housing prices, or predicting spam.

**mjespoz** · 05-01-11, 12:30 AM

Originally posted by LLXC

Well, I can't give it all away...but yea, you're right: it does work for some things, including primitive capping methods. It's great for stuff like...predicting housing prices, or predicting spam.

Yeah....good luck. From your responses you're going to need all the luck that you can get!

**mjespoz** · 05-01-11, 12:34 AM

Kaplan,

Are you making any transformation of your dependent variable before fitting your regression? The reason that I ask is because it's easy to incorrectly back transform predictions to the original scale.
Happy to help if you could provide some more detail in your procedure.

Cheers,
Mjespoz

**mebaran** · 05-01-11, 12:38 AM

Originally posted by Kaplan

My data sets are giving me a good "line", but the actual predicted final scores for both teams are coming in low. I may get a prediction of Home Team 29....Away Team 21 instead of the actual score of HT 35 AT 27. This seems to be prevalent often, but not always. My predictions are strong against the spread, but usually low on totals.

You may try his effective, yet primitive, way of compensating for too high or too low of score predictions. Just add in a constant to you equation. If your lines are consistently low, use 1.10 (only an example). You can probably use regression to come up with your constant fairly easily.

Luck with your modeling.

**ManBearPig** · 05-02-11, 03:43 PM

Based on your questions it sounds like you're trying to get to the total...is this the number you're going for or are you trying to predict the spread? There are a number of reasons for this and it depends entirely on what statistics you're incorporating. For example, league averages actually have more use in trying to come up with your own lines than I ever thought.

Like you I've been real interested in reading anything I can get my hands on just because it's interesting and read Conquering Risk which is a real good starting point. I'm not naive to actually think I'll create a rockstar system and get rich, but I think it's helped me become a better capper because I think numbers can be really telling if you know what to look for.

If you're in to reading some books there are two books I came across that I don't own but would like to. (Sports Investing: NFL Betting System and Sports Investing: College Football Betting Systems) that I've think may be interesting reads, would be curious if anyone has an opinion on them.

If you really want to get dirty look for books on machine learning and probabilistic modeling. There is a book called The Elements of Statistical Learning: ... (ISBN: 0387848576) that looks interesting. This type of stuff is out of people's realm so ignore if you don't feel like digging that much.

I'm always interested in some good reading material and although I don't need to be told how to fish, sometimes it helps to be shown where at least.

**Pushkin** · 05-10-11, 06:21 PM

Originally posted by Kaplan

After reading "Conquering Risk" in December, I have become obsessed with incorporating this regression analysis in my handicapping. I find it enjoyable, changing things up, after handicapping the same way for years. I've bought and read several books on Regression, model building, Probability, and even basic Algebra, as it has been a few years since I had a reason to polish up my algebra skills.

My data sets are giving me a good "line", but the actual predicted final scores for both teams are coming in low. I may get a prediction of Home Team 29....Away Team 21 instead of the actual score of HT 35 AT 27. This seems to be prevalent often, but not always. My predictions are strong against the spread, but usually low on totals.

I have not entered all of my games yet, so am currently working with a sample of just 506 games involving BCS conference teams. Also, I haven't really spent alot of time deciding exactly what the model will be. This is just an early observation derived from throwing all of my variables into one regression. Perhaps to early in the process to be concerned with such things? Thanks.

Kaplan, what books have you read on Regression, model building and Probability?

**LLXC** · 05-10-11, 06:29 PM

Originally posted by mjespoz

Yeah....good luck. From your responses you're going to need all the luck that you can get!

Sure, thanks.

**LLXC** · 05-10-11, 06:33 PM

Originally posted by Pushkin

Kaplan, what books have you read on Regression, model building and Probability?

Check out the syllabuses (syllabi) of Graduate Level Machine Learning courses. The problem is they might have too much advanced math.

**LegitBet** · 05-12-11, 07:28 AM

I thought regression w/football still had to overcome the turnover issue who's variance tends to outweigh the predictive benefit gained?

**Kaplan** · 05-12-11, 10:55 AM

Originally posted by mjespoz

Kaplan,

Are you making any transformation of your dependent variable before fitting your regression? The reason that I ask is because it's easy to incorrectly back transform predictions to the original scale.
Happy to help if you could provide some more detail in your procedure.

Cheers,
Mjespoz

No, I'm not doing any transformations. I am starting to get better predictions the more seasons I put in. With what I have so far, I'm starting to predict totals at a very good clip.

I think I need to divide my stat formula's into sub sets to get more accurate results. My rushing yards stat only goes up to around 3.5. I've noticed that teams above 2 usually run about 60% of the snaps in a paticular game, and win by double digits. However, I'm only seeing a 1.5-2 point bounce for these teams.