[danshan11]

@tsty thanks dude, I really appreciate the conversation, hope you keep one lining it. I now know why you one line it because if you start to say more you go off the reservation, LOL

[tsty]

LOL, you are drunk, that is like saying we are all the same because we drink water. we are all math teachers cause we know 1+1

you are tsty we all model the same cause we all use some form of numbers, so yes you are correct but even in reality it could be possible to model without numbers

Not even remotely close

Or r u suggesting one of ur variables is ur power ranking? Lol

[danshan11]

Not even remotely close

Or r u suggesting one of ur variables is ur power ranking? Lol

one lined it again and still did not answer a single question, LMAO at least you are consistent

[Waterstpub87]

waterspub
what do you think of my approach? pros and cons, just your thoughts of course!

I can't real say without looking at it with more detail, which I don't have time to do. It is different from my approach, but that doesn't really mean much. Rather than use power rankings or prediction formulas, I go more the path of simulating games, which in basketball means looking at individual possessions, and in other sports means simulating individual at bats/plays in football.

One thing to consider is rest scenarios for the NBA, which I didn't see you mention above. It tends to get pretty important, especially as the season drags on. I consider both if a team played the night before, and if the next game will be their 3rd game in 4 nights. So there are four situations, fully rested, BB not 3/4, 3/4, and BB 3/4, which are games which teams tend to play pretty badly/rest key players on late scratches. Prior to this year, I only considered the BB, but I have seen positive results thus far in the NBA in the 3/4 and BB situations, normally betting against the team with light rest.

I disagree that linear regression are the only way to solve the problem. Many people get in trouble with that. If you get 1000 statistics, and run 100 regressions, you are going to find random signals which pass significance levels. When you implement, you are going to lose, and your confidence will cause you to bet more. Linear regressions are a tool, which I use heavily, but I use it more to figure how things effect a game, rather than using it by itself to predict a game.

[bettingman6]

Basketball unfortunately is a lot harder to model play by play than football is. So you can't develop systems like FPI and S&P.

Basically all you can do in basketball is create systems based on previous final scores.

[bettingman6]

538.com has NBA ratings. Unfortunately they don't have college basketball ratings.

ESPN.com has a basketball power index. They use their BPI to determine the odds of each team in both NCAAM and the NBA winning the game, so you can certainly use this for the moneyline. And you can somewhat loosely use their game odds to determine what ESPN.com thinks the spread should be.

What's annoying is that ESPN doesn't post the BPI for the NBA anywhere. And although they post where every NCAAM team is ranked in the BPI, they don't say how many points separates one team from another. http://www.espn.com/mens-college-basketball/bpi . (Unlike the FPI, where for both NCAAF and the NFL they clearly post how many points a team is above or below an exactly average team.)

[danshan11]

I can't real say without looking at it with more detail, which I don't have time to do. It is different from my approach, but that doesn't really mean much. Rather than use power rankings or prediction formulas, I go more the path of simulating games, which in basketball means looking at individual possessions, and in other sports means simulating individual at bats/plays in football.

One thing to consider is rest scenarios for the NBA, which I didn't see you mention above. It tends to get pretty important, especially as the season drags on. I consider both if a team played the night before, and if the next game will be their 3rd game in 4 nights. So there are four situations, fully rested, BB not 3/4, 3/4, and BB 3/4, which are games which teams tend to play pretty badly/rest key players on late scratches. Prior to this year, I only considered the BB, but I have seen positive results thus far in the NBA in the 3/4 and BB situations, normally betting against the team with light rest.

I disagree that linear regression are the only way to solve the problem. Many people get in trouble with that. If you get 1000 statistics, and run 100 regressions, you are going to find random signals which pass significance levels. When you implement, you are going to lose, and your confidence will cause you to bet more. Linear regressions are a tool, which I use heavily, but I use it more to figure how things effect a game, rather than using it by itself to predict a game.

I do adjust for rest but only on if they played yesterday, I no longer adjust for 3 in 4 and other scenarios because there is far fewer now than there was in years past. but played yesterday games I adjust for

[Bsims]

I'd watch about the Kenpom predictions, especially when it comes to totals. Last year I got absolutely creamed at the end of the season and the tournment using a kenpom based system. Not that the same thing would happen to you, but it is something to keep in mind.

I focus more on how close I am to the line. If I within a point or so in NBA of the closing line on 80+% of games, I know I have a decent model.

As I work my way through my current project, which you may recall is an attempt to use 3 different predictive models to estimate the final scores of CBB games, I'm now looking at the 2017-18 season data and generating theoretical wagers based on a varying the factors for these 3 models. As expected, for point spreads, the like game score model is by far the most significant, followed by KenPom, and the power rating system the least significant. When I looked at totals wagers, again the like game model was the most significant, but the power rating system was second and KenPom fell to third. This reminded me of the post that you made (quoted above). This somewhat supports your observation that KenPom’s totals weren't very accurate. The question becomes are KenPom’s totals generally not accurate or was 2017-18 an anomaly. I'm not sure that I want to pursue it much. Looking back at 2016-17 reminds me of the old adage, “A man with one watch knows what time it is, while a man with two watches can’t be sure”.

You also mentioned that the way you measure accuracy of your model is how close it comes the closing line. There is an agreement that this is a good approach. I assume that most people are talking about using Pinnacle’s closing line data. The simple question that I have is where one gets this data.

[tsty]

Lol ask the countless people on this forum who say they are scraping it daily

Right?

[Waterstpub87]

You also mentioned that the way you measure accuracy of your model is how close it comes the closing line. There is an agreement that this is a good approach. I assume that most people are talking about using Pinnacle’s closing line data. The simple question that I have is where one gets this data.

The automatic scraping I did in excel broke a few years ago, it started to treat the cells like formulas, so -8 -110 became -118, which is rough for pinnacle because you cant know what the juice is because they tend to close things on uneven juice (+102 /-112) ect, instead of like 5 dimes or bookmaker where it -110 each side normally.

Currently, I copy/paste the lines then use VBA to paste the close to a separate sheet. I can post the process to do this if it will help you. I should try to fix the auto-download, but I haven't had a super need to do it so.

[oilcountry99]

Vegas insider, line history, use VBA and excel to pull it daily

[u21c3f6]

...
You also mentioned that the way you measure accuracy of your model is how close it comes the closing line. There is an agreement that this is a good approach. I assume that most people are talking about using Pinnacle’s closing line data. The simple question that I have is where one gets this data.

This approach is foreign to me. I do not nor have I ever created a line. However, if creating a line gives you positive results, then by all means continue to do so. But I see things differently.

I cut my gambling teeth on horse racing at a time where the only line you could get was the closing line. There was no such thing as BTCL in horse racing (at least not for me). After a year of trying to pick "winners", I realized that something was wrong with this approach. So my focus became finding criteria that led to a pool of selections that tended to be over bet and/or under bet. In other words, performed better or worse than their closing odds. This is very different than picking "winners" as oftentimes the picking "winners" horses were the ones that were largely over bet in the pools. By eliminating these "winner" horses and others with other criteria, I was left with a smaller group of horses that tended to perform better than their closing odds suggested and therefore produced profits.

I use this same focus for my sports event selections. Based on the criteria I use to make a selection, the only use of the closing line for me is to get an idea of the timing of my wagers to try to get the best price. Sometimes I wager early and oftentimes I wager later or close to gametime if I determine that is when I will most likely get the best price.

Joe.

PS. Merry Christmas!

[jacharron17]

Doing some backtesting, I can tell you that TeamRankings is arguably better than Sagarin in terms of MSE. TeamRankings has a bias towards the home team, so you would need to find a model inversely correlated (biased to road team - Massey, Dokter, DRatings), and combine the values to minimize bias. Also subtract your MSE from a benchmark of sharp book closing lines. The goal would then be to lower your Mean Absolute Error. Sorting data by your projected point spreads (smallest to largest), and assign a range where your MSE is even lower than the average of all matches (would yield a higher win percentage). Maybe your spread range is anywhere from -25 to 40, and your MSE is 120. But maybe a range from -25 to 30 would lower it down to 115 or so. TeamRankings has great stats on key numbers and how they perform against the spread. Another powerful tool to use in your capping! Live betting can be very profitable with these strategies as well. Lets say the live betting line is 15 points off from my projected margin, but the game is not even 50% completed. Maybe the projected favourite hasn’t been hitting their shots at all, while the other team is lights out from 3 superficially inflating the spread.

[jacharron17]

^ Pertains to college ball btw. Forgot to mention that part.