I promised some analysis on the late odds changes a while back, but I haven't followed through. Some projects came up and I've also found some things that may have value, so giving them away isn't best for me.
A couple thoughts from what I've found though:
- If you can, focus on the last few minutes. As most are aware, the money that flows in later is smarter, and the later it flows in the smarter it is.
- I looked at prices 90 seconds to close, 30 seconds to close, at close (when betting is closed), and the final price. The largest jump in efficieny ("smartest") is going from close to final price. This means even if you could wait until the very last possible second, you won't know the final price, and there will still likely be a big change in odds compared to the last price you're able to see
- A simple linear regression of the market win estimate, the market exacta win estimate, and the market daily double win estimate from previous race will get you about 1/3 of the way to predicting the final estimate very accurately. Start adding other factors like morning line and your own handicapping info and you can do better. Change the model specification (i.e. try things other than linear regression) and you can do better still.
- Give consideration to the numeric space you're working in. Do you want to predict final payoff, final probability, etc? Should that include takeout or not? If final probability, do the probabilities need to be proper i.e. do they need to sum to 1 and be betwen 0 and 1? These may sound like silly things but once I started down the path they actually inform a lot of modeling decisions.
I know that doesn't help if you're not building a database of price changes and have the ability to work w/ that data either in excel or some other language. AmWager offers what they call TRU Odds, which I believe is sort of what you're after and what I describe in the 3rd bullet. Not sure exactly how they do it, but I know they combine prices from the different pools to come up with a "better" price.
|