I'm beginning to gather ideas about various models to study, research, etc., but have some questions about how exactly to go about the process. I know from reading some books that 1,000 sample games seems enough to study about one's theory, but before I begin, just curious about finding an edge, and if some of the more advanced guys could give input, I would appreciate it.
Theoretically, let's say that I have found an edge in capping over 1,000 games. I have come to the consensus that I can win 62% of games with my model. For comical purposes, let's say my model (absolute fabrication) claims that NBA basketball players with a high amount of facial hair cover the spread at a higher rate than players without facial hair. I can even dictate points to value players. Clean shaven = 1 POINT. Stubble = 2 POINTS. Goatees = 3 POINTS. Full Beards = 4 POINTS. And lastly, Lumberjack, Mountain Men Beards = 5 POINTS.
So, based on my, again, theoretical model, I begin placing bets and win at 62%. If Lebron and the Heat (4 points) face off against Jason Kidd and the Mavs (2 points), I'm placing a large wager. I continue to win at an above average rate, and feel pretty good about my model.
However, the more I win, the more the odds makers adapt. Maybe they begin profiling my play, looking for trends, adapting, and set their lines accordingly. Suddenly, a clean shaven team is now favored over the Lumberjacks, and my model has not only been discovered, but now dissipates. Now, I am left to start over, beginning another model to determine my edge.
So, in a round about way, I guess my question is, without getting too far ahead of myself, before I begin my research, being that one is not playing against a computer that doesn't ever adapt at any point an time, won't my model be discovered in the long run? Once again, it may sound like a dumb question, but in my humble opinion, there is much man power looking at the spreads right now, and trying to determine why teams cover at a higher rate than others. So why wouldn't they look for trends (especially above the 60% rate) and dissolve one's model. And I know the more complex models would be much more difficult to dissect, but for the simplistic models, how much time is there before time runs out? There has to be some with first-hand experience of this out in the think tank. Thanks for your time. Much appreciated even if I wasted some.