How to convert NBA win % to point spread??

**LT Profits** · 10-20-16, 02:47 PM

One way is to convert the win probability percentages into a money line and then enter those odds into Spread/ML Converter

http://www.sportsbookreview.com/picks/tools/spread-ml-converter/

**oilcountry99** · 10-20-16, 03:23 PM

I appreciate the response.

Do you know of a formula to input into an excel spreadsheet?

**LT Profits** · 10-20-16, 05:42 PM

Originally posted by oilcountry99

I appreciate the response.

Do you know of a formula to input into an excel spreadsheet?

You mean to convert probability to ML?

If greater than 50%, then -(probability/(1-probability)). So 55% would be -.55/.45 for roughly -122

If 50% or less, then +((1-probability)/probability). So 40% would .60/.40 for +150

**oilcountry99** · 10-21-16, 07:04 AM

thanks again for the post. What I need is a formula to convert win% to point spread

Appreciate your efforts thanks LT.

**LT Profits** · 10-21-16, 07:12 AM

Originally posted by oilcountry99

thanks again for the post. What I need is a formula to convert win% to point spread

Appreciate your efforts thanks LT.

There is no such thing because all point spreads are priced differently, and it varies from sport to sport. That is what makes the converter valuable, it is based on historical results.

**oilcountry99** · 10-21-16, 07:50 AM

LT,
What if I input different point spreads in the converter and created a table of WIN% = PT SPREAD and put that table into excel to reference from? so if team X has a win% of 56% excel will then look up the corresponding pt. spread in the table. Does that make sense? Would it work? I'm just trying to avoid having to input the numbers into the table and then input the results into excel...to time consuming.

**oilcountry99** · 10-21-16, 08:03 AM

LT, I found this...

W% = 1 / (1 + e^(-0.13959 * (Avg. Pt. Diff)))

Using a win% of 0.663 as an example:

0.663 = 1 / (1 + e^(-0.13959 * P))

Solving for P (i.e., the point spread) we get:

P = -ln(1/0.663 – 1) / 0.13959 = 4.85

**LT Profits** · 10-21-16, 09:46 AM

Originally posted by oilcountry99

LT, I found this...

W% = 1 / (1 + e^(-0.13959 * (Avg. Pt. Diff)))

Using a win% of 0.663 as an example:

0.663 = 1 / (1 + e^(-0.13959 * P))

Solving for P (i.e., the point spread) we get:

P = -ln(1/0.663 – 1) / 0.13959 = 4.85

I do not know if that formula is right for the specific sport you are looking at, but I DO know it is not good universally over all sports. You can test it for the sport you are looking at by inputting -197 (.663/.337) in converter and seeing how close it is to -4.5

**LT Profits** · 10-21-16, 09:53 AM

I was more exact and inputed -196.7 and got the following:

NFL: -4.5 (Voila!)
CFB: -6.5
NBA: -5.5
CBK: -4.5 (Voila!)

Doesn't really prove anything but you may be close for NFL. And it seems fluky that your formula gets 4.5 for both NFL and CBK given the mass difference in the scoring distributions for those two sports.

**oilcountry99** · 10-21-16, 10:12 AM

LT, the example was specific to the NBA, so the -5.5 is off.

**oilcountry99** · 10-21-16, 10:13 AM

here's a link to what I'm looking into...maybe it will help further

http://statitudes.com/blog/2013/11/12/home-win-probability-using-srs/

**LT Profits** · 10-21-16, 10:19 AM

Originally posted by oilcountry99

LT, the example was specific to the NBA, so the -5.5 is off.

Or the formula is off.

Not really sure which is true.

**LT Profits** · 10-21-16, 10:26 AM

Originally posted by oilcountry99

here's a link to what I'm looking into...maybe it will help further

http://statitudes.com/blog/2013/11/1...ity-using-srs/

This seems to make a lot of sense, so I would go for it for NBA if you want to. I wasn't too thrilled at first with usage of end-of-season SRS, but he addressed that later.

**oilcountry99** · 10-21-16, 10:43 AM

Originally posted by LT Profits

This seems to make a lot of sense, so I would go for it for NBA if you want to. I wasn't too thrilled at first with usage of end-of-season SRS, but he addressed that later.

I believe I would have to apply the formula's set out by TomTango in the comments below the article...

Justin’s equation can be rewritten as:

x = EXP(-dSRS * .167)
win% = 1.844 / (1.844 + x)

However, that dSRS in the model is the UNREGRESSED value.

If you used a regressed value (meaning Games / (Games + 12) x dSRS), then the “x” equation becomes:

x = EXP(-reg_dSRS * .191)

**The Hat** · 10-21-16, 11:50 AM

IMO, point spread and win probability do not correspond. The spread corresponds to public perception and does not reflect or equate to win probability.

**LT Profits** · 10-21-16, 12:22 PM

Originally posted by The Hat

IMO, point spread and win probability do not correspond. The spread corresponds to public perception and does not reflect or equate to win probability.

Precisely the reason you can gain an edge with a model that CAN generate an accurate unbiased spread.

**The Hat** · 10-21-16, 12:41 PM

Originally posted by LT Profits

Precisely the reason you can gain an edge with a model that CAN generate an accurate unbiased spread.

I see what you mean in regards to comparing the 'vegas spread' with the spread that that model generates and betting accordingly, but IMO, I don't think you can actually begin the process with an accurate win probability. What generates the win probability percentage? What comes before the win prob prcntg?

**The Hat** · 10-21-16, 12:52 PM

OP starts off with 'I have an estimated win probability'.

What margin for error are you dealing with in regards to the estimation? 2%? 5%? 7%?

Do you run the model plugging in the various 'win prob %'s' due to the margin for error? Once you get into doing that, not sure that you would see any significant variance between 'model line' and 'vegas line'. And again, just my opinion and I very well may be wrong.

**oilcountry99** · 10-21-16, 12:54 PM

Originally posted by The Hat

IMO, point spread and win probability do not correspond. The spread corresponds to public perception and does not reflect or equate to win probability.

The calculations however would give a 'base' as to what the spread should be, from there one can evaluate whether or not there is any value in the selection based on the given ML and PS on the game.

**oilcountry99** · 10-21-16, 12:56 PM

Originally posted by The Hat

OP starts off with 'I have an estimated win probability'.

What margin for error are you dealing with in regards to the estimation? 2%? 5%? 7%?

Do you run the model plugging in the various 'win prob %'s' due to the margin for error? Once you get into doing that, not sure that you would see any significant variance between 'model line' and 'vegas line'. And again, just my opinion and I very well may be wrong.

Refer to the link in post #12 for more insight.

I'm no expert here just looking at options, thanks for your thoughts

**The Hat** · 10-21-16, 01:06 PM

Originally posted by oilcountry99

Refer to the link in post #12 for more insight.

I'm no expert here just looking at options, thanks for your thoughts

Yeah I checked it out. IMO, the formula doesn't have enough criteria built into it. In basketball, you have to take into account teams playing back to back games vs a team that isn't. Also, coast to coast travel factors in. Injuries. etc etc etc.

**oilcountry99** · 10-21-16, 01:33 PM

Originally posted by The Hat

Yeah I checked it out. IMO, the formula doesn't have enough criteria built into it. In basketball, you have to take into account teams playing back to back games vs a team that isn't. Also, coast to coast travel factors in. Injuries. etc etc etc.

That's where the handicap comes in. The formula only gives a lean.

**oilcountry99** · 10-21-16, 01:40 PM

Originally posted by The Hat

I see what you mean in regards to comparing the 'vegas spread' with the spread that that model generates and betting accordingly, but IMO, I don't think you can actually begin the process with an accurate win probability. What generates the win probability percentage? What comes before the win prob prcntg?

The SRS is what generates the base. This stat is powerful in that it takes past games point differential and opponents strength and incorporates it into a number or rating based on the average NBA team.

Look at the tables in the article, there are some great correlations based on the formula.

**oilcountry99** · 10-24-16, 12:18 PM

How do you enter this formula into excel:

p = 1 / (1 + e-(0.613230 + 0.167546 × (-3.01)))

**Waterstpub87** · 10-24-16, 12:20 PM

=1/(1+exp(-(0.61323+0.167546*(-3.01))))

I assume you mean raising e to that power. Not subtracting it from e

The result is

0.527202

is that close to what you want?

**oilcountry99** · 10-24-16, 12:35 PM

Thanks Waterstpub87...appreciate the reply! worked like a charm

how about this one:
Where dSRS is a cell reference, the number in the cell could be positive or negative.

x = EXP(-dSRS * .167)

**Waterstpub87** · 10-24-16, 12:52 PM

Go to Formulas on the ribbon at the top.

Select Name manager
Select New
Type dSRS
Select the cell where you will have dSRS in the refers section
Close the name manager
you can use this formula

=EXP(-drRs*0.167)

is there a reason you want to lock it to a specific cell? You know that you can just select the cell and get the same result, right? If you type -A1, it will give you the negative of what is in cell A1

**oilcountry99** · 10-24-16, 01:13 PM

Originally posted by Waterstpub87

Go to Formulas on the ribbon at the top.

Select Name manager
Select New
Type dSRS
Select the cell where you will have dSRS in the refers section
Close the name manager
you can use this formula

=EXP(-drRs*0.167)

is there a reason you want to lock it to a specific cell? You know that you can just select the cell and get the same result, right? If you type -A1, it will give you the negative of what is in cell A1

Maybe the above forumula is the incorrect one to use. I'm getting a decimal number result which seems way off.

**oilcountry99** · 10-24-16, 01:16 PM

l also have this last one:

Where win% is a cell, for example N7

= -ln(1/N7 – 1) / 0.13959

Thanks again for your help

**Waterstpub87** · 10-24-16, 01:23 PM

You need to specify order of operations.

Do you want

The natural log of 1/(N7-1) or the natural log of ((1/n7)-1)

Option 1: =-ln(1/(N7-1))/0.13959
Option 2: =-ln((1/N7)-1)/0.13959

**oilcountry99** · 10-24-16, 02:29 PM

Option 2 worked....

I was inputting what you had in the above post and it wasn't working. I did a copy/paste of your formula and it worked fine, strange!

**Waterstpub87** · 10-24-16, 03:00 PM

Originally posted by oilcountry99

Option 2 worked....

I was inputting what you had in the above post and it wasn't working. I did a copy/paste of your formula and it worked fine, strange!

Excel can be a pain sometimes. You might have missed a ( somewhere and it caused it to fail.

**gojetsgomoxies** · 10-25-16, 11:16 PM

you could put moneylines and spreads into excel and do one of two things i can think of:

1: run a regression although will be curve-linear i'm sure... then you will have formula to convert.

2) vlookup it.

as mentioned no real formula as it's different for each sport and changes somewhat over time. lots of conversion tables out there though