Week 8 BTBS Picks!

That’s what they do in a lot of marketing modellling – for each observation you take the number of weeks it was ago, and combine it with some weight parameter to use as a reciprocal exponent (or something like that, I’d have to look it up, slightly hazy). You can jigger the weight to make a steeper or shallower gradients that smoothly weights more recent info more strongly, while discounting info slightly more as it gets older.

This could even solve your woes of what to do with preseason info. you just start its’ “weeks ago” at a relatively high number.

As the saying goes, “nothing is impossible for the man who doesn’t have to do it himself.” :)

Anyway, if you care about that you can email me and I’ll look up how to do it. It’s pretty simple as I recall.

So last night I decided to embark on some analysis of the BTBS predictions. My plan initially was compare the predicted BTBS scoring differentials with the actual results for each week, and then see if the BTBS predictions are improving. I ran the data, for week 7, realized how tedious and time-consuming the task was, and didn’t get any farther than that one week.

What I realized along the way, though, was that my approach was probably not the best. For this data to be meaningful I would have also had to run the numbers on the differential between the vegas spreads and the actual results for each week, and then see how much BTBS is a better predictor than Vegas and whether Vegas is catching up or BTBS is pulling away. But, again, then tells me very little….which is illustrated by the following example. Assume Vegas has set this line: Mizzou is -8 against Gotham Univ. BTBS predicts that Mizzou will win by 12 against Gotham. Thus, BTBS picks Gotham. If Mizzou were to beat Gotham 45-12, BTBS would have been right, but its prediction would have been way off base. If Mizzou were to beat Gotham 39-32, BTBS would have been wrong…..but yet the BTBS prediction would have been nearly spot on.

So, I am scrapping that idea in favor a better one. Namely, ascertaining the correlation between BTBS and Vegas prediction discrepancy (e.g., if Vegas had Mizzou 9 against Gotham, BTBS would tell us to bet on Gotham…but it seems like a very faint-hearted suggestion. On the other hand, if Vegas has Mizzou +1 against Gotham, this is a 9 point differential - which is significant.).

Bill is already employing this concept by putting predictions that vary from the vegas lines by (is it 13 points) into the “LOCK” category. But what I would like to do is take this a step further from the current binary approach, and see if/how the chances of prevailing against a given line increase with an increased discrepancy.

Problem is: I have no idea how to do it. I messed around with excel last night for a while, but couldn’t figure out what functions might do this. Basically, we would just have two columns. One column with the discrepancy. Another columing with a binary representation of a win or a loss.

Is this actually really easy to do? The end graph that I would like excel to spit out would be something the winning percentage as a function of the number discrepancy. Any tips?

So last night I decided to embark on some analysis of the BTBS predictions. My plan initially was compare the predicted BTBS scoring differentials with the actual results for each week, and then see if the BTBS predictions are improving. I ran the data, for week 7, realized how tedious and time-consuming the task was, and didn’t get any farther than that one week.

What I realized along the way, though, was that my approach was probably not the best. For this data to be meaningful I would have also had to run the numbers on the differential between the vegas spreads and the actual results for each week, and then see how much BTBS is a better predictor than Vegas and whether Vegas is catching up or BTBS is pulling away. But, again, then tells me very little….which is illustrated by the following example. Assume Vegas has set this line: Mizzou is -8 against Gotham Univ. BTBS predicts that Mizzou will win by 12 against Gotham. Thus, BTBS picks Gotham. If Mizzou were to beat Gotham 45-12, BTBS would have been right, but its prediction would have been way off base. If Mizzou were to beat Gotham 39-32, BTBS would have been wrong…..but yet the BTBS prediction would have been nearly spot on.

So, I am scrapping that idea in favor a better one. Namely, ascertaining the correlation between BTBS and Vegas prediction discrepancy (e.g., if Vegas had Mizzou 9 against Gotham, BTBS would tell us to bet on Gotham…but it seems like a very faint-hearted suggestion. On the other hand, if Vegas has Mizzou +1 against Gotham, this is a 9 point differential, which is significant.).

Bill is already employing this concept by putting predictions that vary from the vegas lines by (is it 13 points) into the "LOCK" category. But what I would like to do is take this a step further from the current binary approach, and see if/how the chances of prevailing against a given line increase with an increased discrepancy.

Problem is: I have no idea how to do it. I messed around with excel last night for a while, but couldn’t figure out what functions might do this. Basically, we would just have two columns. One column with the discrepancy. Another columing with a binary representation of a win or a loss.

Is this actually really easy to do? The end graph that I would like excel to spit out would be something the winning percentage as a function of the number discrepancy. Any tips?

Twenty-two, thirty-two, and two! Man oh freaking man…