It's not always about bring in 50-50 money. Not every game. It's a myth.
Point spread is a statistical-inferred value that makes the market efficient, like the stock market. That's the reason they are in business. Any one claiming that they can beat the bookie is like the proverbial example of one in a million who happens to flip a coin 20 times in a row and always getting head (or tail). [see e. g. "A Random Walk Down The Wall Street"]
Therefore, people who set the point spread, for their survival, actually set them to be truly reflective of the strength of the teams involved. They may not always be perfect in any one game, but aggregately, they are quite right over a large population samples and a long period time.
Think of it like the stock value of a company. Everyone has a different opinion. But together, the market set its fair value.
A number of us develop models to predict games (just for fun) based on the Game Theory. Our model is conceptually similar to Accuscore, but does not require Monte Carlo sim like Accuscore. The output of the model, for each game, is a probability density function (pdf) of two scores, for example, P[Pats, Eagles]. From this, we have the pdf of Pats-Eagles. The prob. for Pats win is the integration of this function from 0 to positive infinitive. Conversely, the prob. for Eagles win is from -infinity to 0. The line where it is 50-50 on each side (median) is the point spread.
So far, ever since we started several yrs ago, we have been very impressed with the bookies. I don't know which model the bookies use, but we found them impossible to beat, in other words, we bet only when our point-spreads differ from their point-spread (either toward the favorite or the underdog) by more than 35-65 split. Over the years, we beat them by just a small margin that is probably not better than monkeys throwing darts.
Clearly, these guys know their business. That said, this yr Pats is truly a market anomaly. They have been loosing big until the Colts game. The line 23.5 is dead on, amazingly.
BTW, this is just one web site that report the betting percentage:
http://www.thespread.com/nfl-football-public-betting-chart
(I can't vow for their accuracy though)
Bottom line: point spread (to me) is truly a correct measure of relative strength of two teams in a game. The distribution has a large spread of confidence interval (variance - with its origin from Gaussian distribution) is an INVALID CONCEPT in this theory), that's why anything can happen and why they play the game. There is enough of fluky things happen to make the game sufficiently unpredictable and interesting. The more predictable an event is (that is the narrower the confidence interval of an outcome is) the less interesting it becomes. The Pats can lead the Eagles, say 28 pts, then a last second garbage hail Mary can make it 21 and the house would win. Remember in SB 39, the Pats led by 10 until the garbage TD.
As a side note: relative strength is NOT a scalar with transitive property like A>B and B>C means A>C. It is like paper, scissor, rock. That's why linear power ranking has little value in Game Theory prediction.
Point spread is a statistical-inferred value that makes the market efficient, like the stock market. That's the reason they are in business. Any one claiming that they can beat the bookie is like the proverbial example of one in a million who happens to flip a coin 20 times in a row and always getting head (or tail). [see e. g. "A Random Walk Down The Wall Street"]
Therefore, people who set the point spread, for their survival, actually set them to be truly reflective of the strength of the teams involved. They may not always be perfect in any one game, but aggregately, they are quite right over a large population samples and a long period time.
Think of it like the stock value of a company. Everyone has a different opinion. But together, the market set its fair value.
A number of us develop models to predict games (just for fun) based on the Game Theory. Our model is conceptually similar to Accuscore, but does not require Monte Carlo sim like Accuscore. The output of the model, for each game, is a probability density function (pdf) of two scores, for example, P[Pats, Eagles]. From this, we have the pdf of Pats-Eagles. The prob. for Pats win is the integration of this function from 0 to positive infinitive. Conversely, the prob. for Eagles win is from -infinity to 0. The line where it is 50-50 on each side (median) is the point spread.
So far, ever since we started several yrs ago, we have been very impressed with the bookies. I don't know which model the bookies use, but we found them impossible to beat, in other words, we bet only when our point-spreads differ from their point-spread (either toward the favorite or the underdog) by more than 35-65 split. Over the years, we beat them by just a small margin that is probably not better than monkeys throwing darts.
Clearly, these guys know their business. That said, this yr Pats is truly a market anomaly. They have been loosing big until the Colts game. The line 23.5 is dead on, amazingly.
BTW, this is just one web site that report the betting percentage:
http://www.thespread.com/nfl-football-public-betting-chart
(I can't vow for their accuracy though)
Bottom line: point spread (to me) is truly a correct measure of relative strength of two teams in a game. The distribution has a large spread of confidence interval (variance - with its origin from Gaussian distribution) is an INVALID CONCEPT in this theory), that's why anything can happen and why they play the game. There is enough of fluky things happen to make the game sufficiently unpredictable and interesting. The more predictable an event is (that is the narrower the confidence interval of an outcome is) the less interesting it becomes. The Pats can lead the Eagles, say 28 pts, then a last second garbage hail Mary can make it 21 and the house would win. Remember in SB 39, the Pats led by 10 until the garbage TD.
As a side note: relative strength is NOT a scalar with transitive property like A>B and B>C means A>C. It is like paper, scissor, rock. That's why linear power ranking has little value in Game Theory prediction.
Last edited:
