NBA modeling

[specialronnie29]

now the interesting question is

given this approach, what is the line to feed into your kelly calculator!!!!??? funny how this comes up again

[Data]

you want to include the constant to best explain past data as best as possible to see if that explanation tells you if your model (variable model1line) provides any valuable input.

This is where you and the posters who agree with you are wrong. Our goal at this point is NOT to find the best fit but to make the best estimates of the slopes and this is NOT the same thing. If the theory suggests that the line goes through origin (and that is our case) then RTO will give better estimates of the slopes even though OLS provides a better fit. If you want to look closer into the math of this, I can refer you to "Introduction to Econometrics" by Henry Theil, 1978.

[Data]

if this is appropriate the regression will yield this by itself. it wont be 0 exactly but it will be very close to 0

Yes, that is true.

now youre assuming the model accounts for everything the closer misses so it must go through the origin so we might as well impose the restriction noconstant. i dont buy it.

You don't have to buy it. You can actually calculate intercept's T value and you will see that there is no reason to think that it may be non-zero.

[specialronnie29]

This is where you and the posters who agree with you are wrong. Our goal at this point is NOT to find the best fit but to make the best estimates of the slopes and this is NOT the same thing. If the theory suggests that the line goes through origin (and that is our case) then RTO will give better estimates of the slopes even though OLS provides a better fit. If you want to look closer into the math of this, I can refer you to "Introduction to Econometrics" by Henry Theil, 1978.

for the bolded part: you are correct if the theory suggests that the line of best fit should go through the origin RTO will be more precise. Precision refers to the standard errors!!

this is what Theil refers to. He uses the word precision, not 'better' which you seem to imply means non-RTO estimates will be biased or inconsistent

his point is that if you know the line is going through the origin then you can lower your standard errors by forcing it through the origin.

yet in this application we dont care about the standard errors too much (at least you didnt)
we care about getting the slope right
additional freedom is awarded to getting the slope right by allowing a constant. if theory is wrong then the the slopes from RTO will be awful. intuitively you're saying you want to make sure you get the slopes right yet youre restricting them by forcing the line of best fit through the origin, this is pretty bizarre. i am not obsessed with the line of best fit here.... i am interested in the slopes just like you are.

imagine the true DGP is y = 10 + error but your theory says its y=x. then your coefficient on x will be positive if you use RTO when in fact it should be 0 and it would be if you had a constant term.

anyways the pt is pretty irrelevant as like ive said the constant is going to be close to 0 anyways and the standard errors will probably be similar leading to all the same conclusions

[specialronnie29]

You don't have to buy it. You can actually calculate intercept's T value and you will see that there is no reason to think that it may be non-zero.

i have no problem with this as youre first implicitly running the regression with constant and finding it to be insignificant and then re-running without the constant. theres no problem with that as it follows from the data. i thought you were imposing it just because you want it to have that feature..

[Data]

i have no problem with this as youre first implicitly running the regression with constant and finding it to be insignificant and then re-running without the constant. theres no problem with that as it follows from the data. i thought you were imposing it just because you want it to have that feature..

I think it is true, the data suggests it is true. I am going to act on assumption it is true.

[Data]

for the bolded part: you are correct if the theory suggests that the line of best fit should go through the origin RTO will be more precise. Precision refers to the standard errors!!

this is what Theil refers to. He uses the word precision, not 'better' which you seem to imply means non-RTO estimates will be biased or inconsistent

his point is that if you know the line is going through the origin then you can lower your standard errors by forcing it through the origin.

He says "the slope coefficient may be estimated with far greater precision than with the intercept term left in" and this is precisely what I need.

[specialronnie29]

I think it is true, the data suggests it is true. I am going to act on assumption it is true.

absolutely no problem with this
thought you were just doing it because you think it is true

[Data]

Come to think of it, it may look like I have been jumping to some premature conclusions in this thread. While certain statistical properties of the OP model were not spelled out I already knew what they are because I am very much familiar with the models like OP's. So, yes, I have taken some shortcuts but you guys forced me to prove their validity. I hope I did.

[bztips]

I think it is true, the data suggests it is true. I am going to act on assumption it is true.

So why not PROVE that it's true by running the regression with the constant and showing that the constant is not statistically different from zero and does not significantly affect the other coefficients??? Then you can go back to the zero constant regression if warranted. Unfortunately the depth of your resistance to performing this simple test suggests that you're afraid it won't show what you think and hope it will.

Unfortunately, all of this appears to be an academic exercise anyway since we haven't heard anything from the OP.

[Data]

So why not PROVE that it's true by running the regression with the constant and showing that the constant is not statistically different from zero and does not significantly affect the other coefficients??? Unfortunately the depth of your resistance to performing this simple test suggests that you're afraid it won't show what you think and hope it will.

I do not resist this, I cannot run this on OP's model. On my models I do this habitually and the intercept is indeed very close to zero and insignificant. I think that the OP's model is similar and I assume his results will be similar too.

[bztips]

Fair enough

[mathdotcom]

good discussion gentlemen

[bztips]

Mathy long time since we heard from you!

[bolekblues]

Good discussion all. Thanks for the inpiut.

Here are the results that should make things clearer.

I'll post st.errors in parentheses next to coefficients in the equation. I know it is a standard practice to put them below slopes, but this way would take me some more time.

M - model1, C - closer

MOV = 0,191(0,445) + 0,558(0,157)*M + 0,343(0,116)*C Femp = 138,4
(so yes, the null of intercept being zero cannot be rejected, it is close to zero with relatively high st.error)

MOV = 0,524(0,136)*M + 0,361(0,108)*C; Femp = 162,2

MOV = 0,663(0,417) + 0,988(0,061)*C

MOV = -0,6(0,388) + 0,724(0,045)*M


new line (M*), M* = 2*M

ATS results ag.closers for new line, season 2009/10:

overlap. W-L-%
14.125-121-0,508
16.84-68-0,553
18.57-39-0,594
20.23-16-0,59
22.11-7-0,611
It seems that despite reduced number of plays, the model can still find some valuable lines.

[mathdotcom]

Mathy long time since we heard from you!

some health problems

[Data]

now the interesting question is

given this approach, what is the line to feed into your kelly calculator!!!!??? funny how this comes up again

Not funny, actually, seems very appropriate. I think that by now you have much better understanding of what the models do. So, my general answer is "the line we get after step 5 that I described in post #34".

[TU]

You are totally arguing with something you made up. Yes, I agree with you critics of the process if we took the following steps:
1) take the lines that the model produced
2) take the market lines
3) run the OLS regression
4) use the coefficients and the intercept to calculate our final line

However, here is what we actually did:
1) take the lines that the model produced
2) take the market lines
3) run the RTO regression
4) look how useful our model judging by the coefficients (which are between 0 and 1)
We stopped at this point, as you suggested, our model can "identify the weak lines before the market does"

Having said that, there is a next meaningful step:
5) use the coefficients (this time no intercept!) to calculate our final line
What we do at step 5 is just paying respect to the market and assuming that the true line is not our model's line but a number in between. And we do that intentionally, despite your objections of "diluting the results of your own model". Now, if the market line changes, our final line will change too and if this was not happening that would be just silly.

Is there anything that you guys would recommend that would teach me some of this stuff in detail? I understand some of it, but most of it is over my head at this point. Obviously there isn't a book that shows modeling step by step but i'm trying to read anything I can to piece it all together.