Originally Posted by
Dark Horse
3/4 Kelly? That's very high.
Unless the bet is purely mathematical, as in a prop, sports bettors deal with sample sizes that are generally considered adequate in the 300-400 range (per model). A statistician will frown upon that number, and will set his required sample size closer to 10,000. But that is not practical in sports betting, and to protect themselves against fluctuations that the uncovered 9,600-9,700 sample may yet have in store, sports bettors reduce their risk by turning to fractional Kelly.
Even if a sample size is considered clean, it may not be. Let's say you perceive a winning play over 50 games. It just jumps off the page. Money in the bank. After those 50 games you start betting it. You didn't include the 50 games, so statistically your sample is clean. And yet in many cases it may not be. Because you caught a fluctuation. If it continues for the rest of the season, that fluctuation will make your 300-400 sample look very good. But the 10,000 sample will have additional fluctuations, and they may not all be in your favor.
It is certainly possible to develop models that are good each season. But that is a huge step beyond where most players are. In the big sports this may be easiest in baseball, which in my experience is the most mathematically clean; not to mention that the teams play twice the number of games as NBA and NHL teams, and ten times that of NFL teams. Hit a rough stretch in your NFL season and the ATS effect will be beyond repair for that season. Hit a rough stretch in a MLB season, and it will hardly matter in the end.
In any case, a Kelly discussion to me is about taking it out of the realm of theory and into the reality of sports betting. That reality includes the fact that the sports betting market has become much tighter after the anti-gambling legislation from a few years ago. Who do you think did not walk away? Do you think it affected the lines? If so, what does that tell you about long term models?