Euro is reaching its climax tomorrow and I guess a lot of bets are put on which team may win.
The most important thing for placing a bet is good estimation on how likely an outcome is. If you can find a better estimation than a betting company, you could win significant amounts of money. I would guess though that this happens very rarely.
Betting companies provide us with some good estimations of how likely is for a team to win. We can “unveil” these probabilities by inverting the coefficients for a match and then normalizing so they sum up to 1. The normalization phase is necessary because their sum is never 1 and therefore they don’t correspond to a probability function. If you find a company where the sum of the previous process is larger than 1 (even by ε), don’t forget to share with me. 😛
By combining these two ideas, we could use the data from a “reliable” betting company to have a positive gain in expectation or even a positive gain in worst case by betting to another less “reputable” company. As a proof of concept, here is a draft implementation:
From now on, I think I will post my code snippets on github for easy retrieval.