Quote:
Originally Posted by MPRanger
To account for multiple dependent variables;
Is it better to run a single multi linear regression
or multiple single regressions? Or is either way OK?
|
That's a notch over my pay grade.
I'm doing simpler things.
I tend to test single handicapping ideas in my database to see if they have any betting value (for example showed speed against a closer bias). Then I start breaking them down into smaller and smaller categories (sprint, route, dirt, turf, stakes, claiming, favorites, longshots, turf routes, dirt sprints, projected to get loose today etc..)
If I find something negative or positive (in terms of ROI relative to the take) that doesn't look like it's due to a couple of fluky results, I create a permanent query that brings up all the horses that fit that category that are running today.
I have a bunch of queries like that to go along with my handicapping information.
I try to refine the queries over time and also try combining different factors. I also keep testing them to stay current and make sure they've been working well recently.
What I like best is when I find both a negative and positive inside the same race and the positive is a serious contender and the negative is the favorite. If the positive is probably not good enough to win, I'll see if I can construct something that makes sense where all he has to do is hit the board.
It's very hard to find sustainably positive stuff. I had one that was close to break even with no handicapping at all for a few years. Then it suddenly collapsed for over a year. I recently checked it again it's doing great in 2020 so far. So you can get whipsawed by some of this.
