Half-point Calculator beta test

[Ganchrow]

All comments, questions, and criticisms welcome:
<script> function OpenCalc() { var winTool = window.open('http://bettingtools.sbrforum.com/newcalc/half-point.html', '_blank', 'width=215, height=275, resizable=yes, status=no'); winTool.focus(); } </script>
<input style="font-size: 11px;" type=button value="Open half-point calculator beta test" onClick="OpenCalc();">

Note that you can use the push probability estimates embedded in the calculator or can enter your own.

Currently supports:
NBA, NFL, NCAA BB, and NCAA FB Sides and Totals
MLB Totals

[zootiehead]

Looks nice. Why doesn't it work with MLB spreads?

[Ganchrow]

Looks nice. Why doesn't it work with MLB spreads?

Because that works a little a differently than spread/totals in other sports. Specifically, the marginal value of half-points in MLB spreads is considerably more sensitive to the level of total than in other sports and also depends on whether the favorite team is playing at home or away.

MLB spreads will certainly be included in an upcoming version, but in the meantime I'm really just trying to get users' opinions and feedback on the current state of the interface and features.

[Santo]

It looks very usable at the moment. Speaking personally I prefer to keep these things in excel spreadsheets than web tools because of the speed differential.

I haven't checked all your numbers with my spreadsheets, but before I do I'll ask whether you believe them all to be accurate, or are just "placeholders" so to speak.

[flyingillini]

Great and I like it. I will use it!

[Ganchrow]

I haven't checked all your numbers with my spreadsheets, but before I do I'll ask whether you believe them all to be accurate, or are just "placeholders" so to speak.

We'll say I believe them to be fairly accurate.

The MLB totals you can take to the bank.

The NFL spreads are very accurate except for the fact that they don't adjust for the relative magnitude of the totals.

The NBA and NCAA BB totals are currently the figures in which I have least confidence.

[vanzack]

Ganchrow - most of the time you talk way above my head (that is a compliment in case you are confused) but in simple terms what data are you using?

For example, how many years of NFL data go in to determining that a 6.5 point fav has a 5.7% chance of pushing a 7 point spread?

Thanks in advance.

[Ganchrow]

Ganchrow - most of the time you talk way above my head (that is a compliment in case you are confused) but in simple terms what data are you using?

For example, how many years of NFL data go in to determining that a 6.5 point fav has a 5.7% chance of pushing a 7 point spread?

Thanks in advance.

MLB goes back to 1999.
NBA goes back to 1990.
NFL goes back to 1990.
NCAA BB does back to 1997.
NCAA FB goes back (in theory) to 1990.

[vanzack]

MLB goes back to 1999.
NBA goes back to 1990.
NFL goes back to 1990.
NCAA BB does back to 1997.
NCAA FB goes back (in theory) to 1990.

So my first question / comment to you is can we assume that the NFL game has remained constant since 1990 and in 2008 is it prudent to use statistics from the 90's? (I dont know the answer to this, throwing it out as discussion...)

Are the years weighted in importance, and if so / not what is the best way to weight the statistics?

[Ganchrow]

So my first question / comment to you is can we assume that the NFL game has remained constant since 1990 and in 2008 is it prudent to use statistics from the 90's? (I dont know the answer to this, throwing it out as discussion...)

That's the assumption. Examining the data, it does indeed appear to hold.

Are the years weighted in importance, and if so / not what is the best way to weight the statistics?

No.

Here's how I footnoted the relevant chart on the NFL forum:

• <span style="font-size: 11px;">The push frequency for a given point spread was determined by the percentage of games with a closing spread within 2 points of the spread in question ending with a favorite margin of victory equal to that spread. For example, the push frequency of a spread of 3 was determined from all games with a closing spread between 1 and 5.</span>
• <span style="font-size: 11px;">This is based on Stanford Wong's methodology as described in Sharp Sports Betting.</span>

[vanzack]

That's the assumption. Examining the data, it does indeed appear to hold.

No.

Here's how I footnoted the relevant chart on the NFL forum:

• <span style="font-size: 11px;">The push frequency for a given point spread was determined by the percentage of games with a closing spread within 2 points of the spread in question ending with a favorite margin of victory equal to that spread. For example, the push frequency of a spread of 3 was determined from all games with a closing spread between 1 and 5.</span>
• <span style="font-size: 11px;">This is based on Stanford Wong's methodology as described in Sharp Sports Betting.</span>

Thanks, and thanks for the efforts on the calculator.

I would be curious to hear others thoughts on the validity of using non-weighted stats since 1990.

I dont have any evidence one way or the other but think it would be an interesting discussion.

[Ganchrow]

Thanks, and thanks for the efforts on the calculator.

I would be curious to hear others thoughts on the validity of using non-weighted stats since 1990.

I dont have any evidence one way or the other but think it would be an interesting discussion.

Testing a data set for stationarity is not difficult.

If you're interested on the underlying mathematics, see the Wolfram entry on the stationary of time series.

	1990s				2000s	

Spread	n	p		Spread	n	p
0	312	0.00%		0	195	0.00%
1	901	2.55%		1	780	2.44%
2	1,105	1.99%		2	917	1.96%
3	1,210	9.67%		3	1,006	9.94%
4	1,233	3.41%		4	1,038	2.50%
5	1,278	1.64%		5	1,159	1.73%
6	891	3.93%		6	726	2.75%
7	824	5.83%		7	628	5.57%
8	772	1.94%		8	627	2.39%
9	618	0.81%		9	490	1.02%
10	417	4.80%		10	276	5.07%
11	343	2.33%		11	243	2.06%
12	279	0.72%		12	179	0.00%
13	202	0.99%		13	105	1.90%
14	164	4.88%		14	81	4.94%
15	135	2.22%		15	70	0.00%
16	79	3.80%		16	36	2.78%
17	43	4.65%		17	23	4.35%

[ProlinePlayer]

MLB goes back to 1999.
NBA goes back to 1990.
NFL goes back to 1990.
NCAA BB does back to 1997.
NCAA FB goes back (in theory) to 1990.

I don't really know if it makes much of a difference, but for this kind of thing I prefer to use a cutoff of 1994 for the NFL.

The year the 2-point conversion was introduced. I would suspect that the probabilities around the key numbers may be somewhat different before/after that change.

[Ganchrow]

I don't really know if it makes much of a difference, but for this kind of thing I prefer to use a cutoff of 1994 for the NFL.

The year the 2-point conversion was introduced. I would suspect that the probabilities around the key numbers may be somewhat different before/after that change.

Certainly a good point.

In reality, however, it doesn't make much of a difference. Stanford Wong's apparent assessment is that the increase in sample size makes up for the slight regime change that may have coincided with the reimplementation of the NFL 2-point conversion in the 1994-95 season. Wong's data set, however, stretches all the way back to the mid 80s.

For what it's worth, with my particular (and to be blunt, somewhat arbitrary) choice of data set years, the figures barely hiccough if you exclude the four pre-1994 seasons. In reality, the sampling is pretty noisy to begin with, and it's probably even more disingenuous of me to take those percentages out of to 3 decimal places. (Of course the tiny sample sizes are probably a big part of why I personally don't much bet the NFL to begin with.)

[Bill Dozer]

This is a very cool tool that takes line shopping to the next lext level. I think this will be the most popular tool hands down.

[Pants]

Wouldn't it be more useful to present push probabilities as they are priced by the market?

Take the MLB default total figures for example:

8.5 has odds of -170.77 and +139.85, for a fair market price of +155.31 for the under. Over 9.5 has a fair price of 100. Therefore, the market says that a push on 9 happens 10.83% of the time. The same can be done for 10, yielding 7.10%.

[jjgold]

Ganch's gambling tools are the best I have ever seen anywhere, I am dying to meet this clown but hate NY City.

Thanks

[Ganchrow]

Wouldn't it be more useful to present push probabilities as they are priced by the market?

The idea is for a user to enter the current line, and based on the predicted push probabilities, the calculator then outputs the fair line when buying or selling point. The user could then optionally enter the actual line and the calculator would then output his expected edge at the price.

Take the MLB default total figures for example:

8.5 has odds of -170.77 and +139.85, for a fair market price of +155.31 for the under. Over 9.5 has a fair price of 100. Therefore, the market says that a push on 9 happens 10.83% of the time. The same can be done for 10, yielding 7.10%.

If the market for the 8½ is -170.77/+139.85, then the fair market price is actually &plusmn;151.27. (If you mouse over the edge in that row, you'll see the a 60.202% expected win probability.) This, when taken in conjunction with the 9 and 9½ markets, correspond to a push probability for the 9 of 10.202%, which is of course exactly the predicted value in the calculator. This is by design.

[Tchocky]

This is very nice. It would be very nice if I could download this as an application for my mobile.

[Ganchrow]

This is very nice. It would be very nice if I could download this as an application for my mobile.

I'll have to look into this for all the betting tools. So stay tuned ... this might be available soon.

[Tchocky]

I'll have to look into this for all the betting tools. So stay tuned ... this might be available soon.

Thanks for the quick response, Ganch. It would be awesome to have a parlay calculator and a teaser calculator on our phones.

[Dhuepaxuz]

.

Ganchrow,

Looks nice indeed. Thanks for sharing.

[Pants]

If the market for the 8½ is -170.77/+139.85, then the fair market price is actually &plusmn;155.31. (If you mouse over the edge in that row, you'll see the a 60.202% expected win probability.) This, when taken in conjunction with the 9 and 9½ markets, correspond to a push probability for the 9 of 10.202%, which is of course exactly the predicted value in the calculator. This is by design.

How are you calculating this? 155.31 gives 60.83%.
I was using U8.5 (39.17%) and O9.5 (50%), which give 89.17% together, or 10.83% for hitting 9. You could also do this with O8.5 (60.83%) and U9.5 (50%) and subtract 100 to get the same answer.

[Ganchrow]

How are you calculating this? 155.31 gives 60.83%.
I was using U8.5 (39.17%) and O9.5 (50%), which give 89.17% together, or 10.83% for hitting 9. You could also do this with O8.5 (60.83%) and U9.5 (50%) and subtract 100 to get the same answer.

Whoops, I had erroneously copied your figure of ±155.31 rather than correcting it.

The fair market price is actually ±151.27 (original post modified).

[Pants]

Whoops, I had erroneously copied your figure of ±155.31 rather than correcting it.

The fair market price is actually ±151.27 (original post modified).

You can probably guess my next question then. How do you get 151.27? 155.31 is the midpoint.

[ProlinePlayer]

The push frequency for a given point spread was determined by the percentage of games with a closing spread within 2 points of the spread in question ending with a favorite margin of victory equal to that spread. For example, the push frequency of a spread of 3 was determined from all games with a closing spread between 1 and 5

I've been working on compiling charts of this nature for years. As someone without a solid math background it's been a real trial and error affair.

To correct for the problems of sample sizes being so small using the games 'around' the actual spread is one of the techniques I use. But I've also feared that this may introduce a degree of error into the final numbers.

Take for example an NBA spread of 9. Use the numbers between 7 and 11.

The problem is that the quantity of each of the spread numbers used will in the end not be the same and there will be a bias toward the lower numbers. In fact most of the sample would end up coming from games which actually had a spread lower than 9. As a result would the push probabilities for the numbers below 9 be too high? And for the same reason those higher be too low?

Just wondered if you consider this a problem.

[Santo]

You can probably guess my next question then. How do you get 151.27? 155.31 is the midpoint.

It might be the mid point between the two, but you can't use that when calculating fair value, there is a bias.

[Ganchrow]

You can probably guess my next question then. How do you get 151.27? 155.31 is the midpoint.

Check out my introduction to expectations and theoretical hold, which explains the calculations.

[Ganchrow]

I've been working on compiling charts of this nature for years. As someone without a solid math background it's been a real trial and error affair.

To correct for the problems of sample sizes being so small using the games 'around' the actual spread is one of the techniques I use. But I've also feared that this may introduce a degree of error into the final numbers.

Take for example an NBA spread of 9. Use the numbers between 7 and 11.

The problem is that the quantity of each of the spread numbers used will in the end not be the same and there will be a bias toward the lower numbers. In fact most of the sample would end up coming from games which actually had a spread lower than 9. As a result would the push probabilities for the numbers below 9 be too high? And for the same reason those higher be too low?

Just wondered if you consider this a problem.

Right, so the trick is you'd either equal-weight it, or if you wanted to get fancier, provide an exponential-type smoothing such that more weight is given to numbers closer to the total or spread in question.

But again, we're talking about very small effect that's likely much smaller than the inherent sampling noise.

[Ganchrow]

Check out my introduction to expectations and theoretical hold, which explains the calculations.

Actually, a more relevant posting would be An introduction to betting lines and percentages.

But to summarize: a market of -170.77/+139.85, implies vigged probabilities of ^170.77/_(170.77+100) &asymp; 63.068% and ^139.85/_(139.85+100) &asymp; 41.693%, respectively.

This corresponds to an overround of 63.068% + 41.693% &asymp; 104.761%.

Hence, the fair market probability for the over is ^63.068%/_104.761% &asymp; 60.202%, implying a fair market price of -100 &times; ^60.202%/_(1-60.202%) &asymp; -151.27.

[hhsilver]

still beta?

Is there a later version available?

The one I'm using is : bettingtools.sbrforum.com/newcalc/half-point.html

Today, for CFB, I entered 1 as the spread with odds of +101 and -111 for Fav and dog.
I got results for 0.5 , 0, -0.5 with a push prob of 1.2 % for 0. Either I am missing something or this is a bug? Ties aren't possible (why the positive probability?) and should there be three different values for the odds at 0.5, 0, -0.5 when, to me, the 3 spreads are equivalent if ties aren't possible.

Also,
the default is MLB totals. I wonder if the user will be able to reset the default based on what he bets most. This is a minor point. No big deal to change the setting each time I begin.


Thanks, Ganch, for all the great work and help.

hh

[Ganchrow]

Is there a later version available?

The one I'm using is : bettingtools.sbrforum.com/newcalc/half-point.html

Today, for CFB, I entered 1 as the spread with odds of +101 and -111 for Fav and dog.
I got results for 0.5 , 0, -0.5 with a push prob of 1.2 % for 0. Either I am missing something or this is a bug? Ties aren't possible (why the positive probability?) and should there be three different values for the odds at 0.5, 0, -0.5 when, to me, the 3 spreads are equivalent if ties aren't possible.

I suspect you didn't reclick the "Calculate" button after changing the spread.

Also,
the default is MLB totals. I wonder if the user will be able to reset the default based on what he bets most. This is a minor point. No big deal to change the setting each time I begin.

Good suggestion. I'll look into using cookies to preselect defaults as soon as I get back from Vegas.

[hhsilver]

I suspect you didn't reclick the "Calculate" button after changing the spread..

No, that's not it. Tried again and made sure to "calculate". Also, I put spread of 0 in. Same results - push prob of 1.2% and different odds for 0.5 and -0.5.

[john pavlic]

very nice Ganchrow! Thanks

[Ganchrow]

No, that's not it. Tried again and made sure to "calculate". Also, I put spread of 0 in. Same results - push prob of 1.2% and different odds for 0.5 and -0.5.

Good call. It was actually using stale data.

It should be correct now.

Thanks for picking that out. Much appreciated.