Three pointer at buzzer by team up by 25 to go over by 1. this is why there is NEVER a lock or any such thing in sports. Too much randomness. 60%, or maybe a smidge higher is all one can hope for, long term. This model is not 60% - but 57-58 (I believe, and was in real time last year). Hope to improve it to 60%+ without eliminating alot of games. 0-1 so far today...
You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an alternative browser.
You should upgrade or use an alternative browser.
Pattern Recognition on NBA Halftimes
- Thread starter TheNeuralNBANet
- Start date
Your statement(s) are incorrect, or at least naive. Let me take a few minutes to explain.a 2pt loser and 1pt loser is still losing. (signs of a rookie bettor) "but but i only lost by .5 point" your system is not very good.
This system predicts a 2nd half line and we compare it to the vegas posted line. If its 3 points different or more we will bet it. The standard deviation on NBA 2nd half totals vs the spread is 12 points. (The distribution is slightly skewed to the right (because of OTs, 2OTs and some 3OTs) but basically normal in its shape). If the standard deviation is 12 points we can calculate that the value of 1 point is 3.3% - so the value of 3 points is 9.9% or an EV of 59.9%. (This is consistent with around 9 cents per half point on offshore buying lines which means, bookmakers obviously understand the above - its just probabilities/math)
To your 'complaint' about my system 'not being very good' and '2 pt loser and 1 pt loser are signs of rookie bettor' etc. To someone betting games, a loss is a loss, but few people wandered into this thread and are betting full units on something they don't understand and don't have a track record - at best a $500 bettor is betting $50 or $100 on these picks to get a feel/confidence. To someone researching or designing a technical system, how big wins and losses are is an important datapoint both in small sample sizes to see how big of role variance is playing and in large sample sizes to see the robustness of the algorithm. If I claim my system has a 3 pt edge on all bets and say for example my record is 15 wins and 20 losses and you make a decision that 'this is a bad system' what if I additionally told you that the 20 losses lost by an average of 4 points and the 15 wins won by an average of 11 points. This additional data should lead you (or someone versed in this) to acknowledge the system was probably better than just the W/L record indicated. On the other hand, if the system was 20-10 but all the wins were narrow and the losses were all blowouts, that should take away a bit from the power of the system. Now, if the system had a W/L of 150-200 - in that large of sample the short run variance would be far less likely to be extreme, and therefore you could probably reasonably state the model was 'not very good'. I write this only because in historic testing (and last years real time) our 'points won by' average is significantly higher than our 'points lost by average' (normalized for overtimes).
Good luck.
agree w this...Neural....don't waste your time explaining....tell the guy to take probability and staitstics 101 and be done w itDon't even give that guy the time of day. Not worth your time.
If he considered that bitching of a rookie bettor, he ought to go scroll through some other posts to see what real bitching is.
I appreciate the plays.
Thanks.
