Another restart for the blog

Once again medical issues have interrupted my blog. A couple more hospital stays, and an overall deteriorating condition haven't helped. Also, most of my available time was spent on the actual football season (not overly successful). And more recently a massive amount of time spent developing a college basketball system (CBB). It is the biggest and most complete system I’ve developed. This is the system I’ll be blogging about. (As I write this, I don’t yet know if it will be successful)

We’ll start where we left off, dealing with the question of what to set the minimum number of like games to be used. This is a parameter to the Like Game Routine. The purpose of the system is to identify a set of previous games most like today’s game, and hence help predict likely final scores. One technique is to look at the correlation of the average scores of the games in the set compared to the actual final scores. 

I wrote a routine to set the minimum number of games at 50, then process each of the games in the past 11 years of data and calling the find like games routine. I saved the average scores from the like games, the expected scores from the spreads and totals, and the final scores. I computed the correlation between these numbers. I repeated this for 100 games, then 150 games, and so on.

The sport being addressed is CBB. The number of previous games in the lookup table is enormous, 37,562 past games. Following are the results of the correlation run.

# Min Games	Corr1	Corr2	Corr3	Corr4
50	0.554	0.569	0.548	0.563
100	0.553	0.566	0.548	0.563
150	0.552	0.565	0.548	0.563
200	0.551	0.564	0.548	0.563
250	0.55	0.564	0.548	0.563
300	0.55	0.564	0.548	0.563
350	0.549	0.563	0.548	0.563
400	0.549	0.563	0.548	0.563
450	0.548	0.562	0.548	0.563
500	0.548	0.562	0.548	0.563

Corre1 and Corr2 are the correlations for the like games versus the actual scores. Corr3 and Corr4 are the correlations of the expected scores from the lines and the actual scores. An interesting note is that early on, the like game scores are a better predictor than the lines. It’s not until about 400 games that the lines catch up.  I’ve decided to use 400 games because each score in the selected set will only have a 0.25 % impact on the probabilities, while at 50 games each would have a 2% chance.

In my next post, I’ll put the like game system to work in looking for CBB money line wagers. I’ll tweet on Twitter, @ole44bill, when I post. If you have comments or questions add them to the blog or Email them to me at ole44bill@gmail.com. The forum at sportsbookreview.com is no longer an option. The thread I started on this topic has been deleted. Apparently, I violated a rule by listing a link to my blog.