Line Convergence with Prediction

**FourLengthsClear** · 01-31-12, 11:05 AM

Originally posted by Waterstpub87

I built a mathematical prediction formula for NBA totals, it has hit around 61% over the last 80 or so bets. However, I have been running into a problem with line convergence. The line movements tend to move the total closer to my predictions. Does the closing line being closer to the prediction total then the opener mean that my system is more accurate then if it was the other way around? I have run a regression on it, and come up with a higher R2 then the line only on games that I have predicted were 5 points or more off the total. Am I on the right track if the closing line tends to converge with my prediction?

This is exactly what you would hope/expect to see. The usual caveat about 80 games being too small a sample applies, of course.

**Waterstpub87** · 01-31-12, 11:12 AM

Originally posted by FourLengthsClear

This is exactly what you would hope/expect to see. The usual caveat about 80 games being too small a sample applies, of course.

Absolutely, I know it is a small sample size. Promising results so far, but I did take a Binomial Distribution on it and came up with .97 or so. So that at least somewhat statistically significant. If lines continued to converge, could I possibly scalp games that are near the convergence point?

**FourLengthsClear** · 01-31-12, 11:17 AM

Originally posted by Waterstpub87

Absolutely, I know it is a small sample size. Promising results so far, but I did take a Binomial Distribution on it and came up with .97 or so. So that at least somewhat statistically significant. If lines continued to converge, could I possibly scalp games that are near the convergence point?

Broadly speaking you should only be looking to scalp/middle games that reach/go past your calculated fair value.

**mathdotcom** · 01-31-12, 01:30 PM

It baffles me how guys who are sharp enough to make these models can't figure out these questions.

If you put the total at 200 and the line is 195 then you hit the over. Great. Obviously you don't buy back the under if the closing line is 200. Maybe you take some Under if it overshoots to 201.

A strategy is to double your unit on the O195 with the expectation of buying back some U201 so you can risklessly bet more on +EV bets. But don't count on the line going that far.

**mathdotcom** · 01-31-12, 01:31 PM

On a side note I wish the word R-squared were banned from the think tank.

It is the source of so many problems/misconceptions.

Some of my most successful models have had R-squareds of 0.01-0.02.

**Waterstpub87** · 01-31-12, 01:32 PM

Originally posted by mathdotcom

It baffles me how guys who are sharp enough to make these models can't figure out these questions. If you put the total at 200 and the line is 195 then you hit the over. Great. Obviously you don't buy back the under if the closing line is 200. Maybe you take some Under if it overshoots to 201. A strategy is to double your unit on the O195 with the expectation of buying back some U201 so you can risklessly bet more on +EV bets. But don't count on the line going that far.

I wouldn't normally even do that. I meant that if I have it pegged at 198 and it is 200 at one book and 196 at another book should I scalp the middle, because I am hitting numbers that would give it a high % of hitting in the middle.

**mathdotcom** · 01-31-12, 02:32 PM

No shit

How are you smart enough to make a model but not know what to do with it??

**bztips** · 01-31-12, 03:06 PM

MDC, being a bit harsh, aren't you? Two different skill sets - creating a model requires some formal modeling and statistics knowledge, plus info about the sport. Knowing what to do with the results requires betting knowledge, not the same thing.

I know lots of economists/stats guys who could model-build all day (if they also had some minimum amt of knowledge about a particular sport), but who wouldn't have the first clue about how to bet.

**Spektre** · 01-31-12, 03:20 PM

I think somebody woke up on the wrong side of the textbook.

**Waterstpub87** · 01-31-12, 04:49 PM

Originally posted by mathdotcom

No shit How are you smart enough to make a model but not know what to do with it??

I appreciate your help. Thank you for your honest answers.

**mathdotcom** · 01-31-12, 05:29 PM

Originally posted by bztips

MDC, being a bit harsh, aren't you? Two different skill sets - creating a model requires some formal modeling and statistics knowledge, plus info about the sport. Knowing what to do with the results requires betting knowledge, not the same thing.

I know lots of economists/stats guys who could model-build all day (if they also had some minimum amt of knowledge about a particular sport), but who wouldn't have the first clue about how to bet.

OK I am taking at least knowledge of how a book works, i.e. what -110 means

But after that it is a piece of cake

**RicoJay13** · 02-05-12, 05:49 PM

Question: are basketball totals normal distribution?