Modelling teaser winchance

[Ominous]

I have programmed a method I use to estimate the implied winchance from buying points in NFL games.

The Halfpoint calculator overestimates the importance of buying many points. This is because the favorite margin of victory, favMoV, is more likly to be close to the actual spread.

The halfpoint calc assumes that the push frequency for any favMoV is the same as the push frequency of the associated spread(which can be "measured"). (No news here)

The basic idea with my method is that I want to multiply the Push frequencies listed in the halfpoint calculator with a function(ranging from 1 to 0) that has lower value the further away from the actual spread of the game examined.

I currently use a normal distrubution function, divided with its value at the "expected value"=spread.

Newpushfreq=Oldpushfreq*normpdf(X,Spread ,Stdev)./normpdf(Spread,Spread,Stdev);
where Y = NORMPDF(X,MU,SIGMA) Returns the normal pdf with mean, MU, and standard deviation, SIGMA, at the values in X.

The standard deviation (Stdev) is fit numerically so that if you buy points from the current spread to "infinity" then you have a 100% winchance.

Does this approach make sense? Input is appriciated.

[Justin7]

I use a similar approach - the degradation of point values as you get further away from the actual spread. The tough question is what is the most accurate way to do it, and I haven't found a perfect solution.

[Ominous]

Ok, how do you degrade the push frequencies?

Im in on houston +4 tonite at the odds of 1,4. Has to be value pure point buy wise right? Pinny line is -3 houston -119/ +3 JAX +111.

My calc shows 7,6% value to the pinnacle line

[Ganchrow]

I have programmed a method I use to estimate the implied winchance from buying points in NFL games.

The Halfpoint calculator overestimates the importance of buying many points. This is because the favorite margin of victory, favMoV, is more likly to be close to the actual spread.

The halfpoint calc assumes that the push frequency for any favMoV is the same as the push frequency of the associated spread(which can be "measured"). (No news here)

The basic idea with my method is that I want to multiply the Push frequencies listed in the halfpoint calculator with a function(ranging from 1 to 0) that has lower value the further away from the actual spread of the game examined.

Yes, this is definitely the way you should be thinking about this.

I currently use a normal distrubution function, divided with its value at the "expected value"=spread.

Newpushfreq=Oldpushfreq*normpdf(X,Spread ,Stdev)./normpdf(Spread,Spread,Stdev);
where Y = NORMPDF(X,MU,SIGMA) Returns the normal pdf with mean, MU, and standard deviation, SIGMA, at the values in X.

The standard deviation (Stdev) is fit numerically so that if you buy points from the current spread to "infinity" then you have a 100% winchance.

Does this approach make sense? Input is appriciated.

Although you're on a reasonable track there are a few theoretical problems with the particular methodology you've outlined that I'll try to touch on in a later post. I will mention however, that one extremely sharp occasional contributor to this message board created a somewhat similar model.

[Ganchrow]

Ok, how do you degrade the push frequencies?

Im in on houston +4 tonite at the odds of 1,4. Has to be value pure point buy wise right? Pinny line is -3 houston -119/ +3 JAX +111.

My calc shows 7,6% value to the pinnacle line

For what it's worth, given the spread of 48 my own proprietary models would judge this teaser leg to be very slightly -EV.

[Ominous]

For what it's worth, given the spread of 48 my own proprietary models would judge this teaser leg to be very slightly -EV.

Hmm, lets see.

Pinnacle Match odds of -170 / +160 at the time would imply a winchance of 62,08% at pk for houston.

After I set my program to center the spread around "-3" it gives the following push frequencies:
JAX by

1 0.0237
 2   0.0182
  3  0.0866
  4  0.0253

For a total winchance of 0.62+0.12=0.74 and a push chance of 2,5%. Total return= 0.7485*1.4+0.0253 = 1.073.

Anyhow, its interesting that we get so diffrent results. What do you use to compute your "degraded" push freqencies?

[Ganchrow]

Hmm, lets see.

Pinnacle Match odds of -170 / +160 at the time would imply a winchance of 62,08% at pk for houston.

After I set my program to center the spread around "-3" it gives the following push frequencies:
JAX by

1 0.0237
 2   0.0182
  3  0.0866
  4  0.0253

For a total winchance of 0.62+0.12=0.74 and a push chance of 2,5%. Total return= 0.7485*1.4+0.0253 = 1.073.

I must have misunderstood your earlier post. My apologies.

Based on the Pinnacle spread at ±3 and the Pinnacle ML at -170/+160 I get a 74.17% probability of covering the +4 and a 2.1612% probability of pushing.

What do you use to compute your "degraded" push freqencies?

As I said, I use my own proprietary models.