10 Comments

What is the result if you run the test by betting only the point spread?

Expand full comment

Hey, thanks for the comment!

When I bet only the point spread, I found that the accuracy was too low (<50%). MLB games are just really slow and “boring”, so a lot of times while it would predict the correct winner, the winner would only win by 1 run, failing to cover the spread.

I think that the point spread approach has more value in higher-action sports like NBA. I’ll also be building a model for those sports eventually, so I’m definitely not ruling out the point spread approach.

Thanks

Expand full comment

Hi there, using the exact same methodology instead of backing this high rated teams to win could you lay them on a exchange as the odds are on an average -116 this could potentially result in a positive ROI ??

Expand full comment

Hey, thanks for the comment!

What do you mean by on an exchange? I’m relatively inexperienced with betting venues and am only familiar with sportsbooks like Bovada or Draftkings.

Hitting at 60% with -116 would be a HUGE improvement, I definitely agree, but I’m curious about what you’re saying.

I haven’t made the post yet, but I recently expanded the algorithm to bet on all 3 types of a bet once per day and I actually got my average odds to about -105, but the accuracy topped out at about 52% so it still wasn’t enough to come out profitable.

The next step is getting historical data for prop bets. 52% for those kind of bets would be what I need, but finding that data is really tricky.

Expand full comment

An exchange is peer 2 peer betting so theoretically punting against other punters. Betfair is excellent for this. On betfair you can not only back somthing to win but to not win known as a “lay” bet. If your method above is performing as it does goes massively into the negative ROI of $9000 because of the below even odds. If you placed a “lay” bet on these favs you would only need a much small percentage to profit.

YouTube betfair and lay bets. You can download a betfair api with historical results/odds to test if the lay bet on these favs would return a profit.

The other advantage to the exchange you can automate bots to place bets 24/7 , you can’t get banned if your a profitable punter and the odds a generally better. If you live somewhere betfair isn’t accepted. A vpn in a location works perfect

Expand full comment

Please remember that Betfair takes their commission. Couple of years ago it was something like 4% if you make profit out of the bet.

Expand full comment

They do yes which is another reason to test/ simulate the model first. They also offer discounted commission on high turnover accounts often as low 1%

Expand full comment

Also the higher odds available on betfair can sometimes negate some of the commission

Expand full comment

Hi! How about this: in Finnish ice-hockey my friend bets only visiting team when odds are over 3x, as European, so better than +200. He said that this is EV+ in the long run. Can you get data out of this?

How about the actual EV+ which occurred couple of years ago; when NHL-team come back from long away-trip (minimum 3 matches out-of-the-town) and then they lose the first home match. Can you get data out of this? Still occur as EV+, if thinking that 2.5x (+150) is possible NHL-match home odds.

How about NBA? They have also long away-trips.

One possibility would be add info: which team will have to fly back home to west. Example if Los Angeles Kings play versus NY,Devils and Tampa Bay and then fly back to LA do they lose the first home match?

Expand full comment

Thanks for the comment!

I think your friend might be confusing high-payoffs with +EV. Yes, if they're hitting bets and getting back multiples of their principle, then they'd do great in the long-run. But generally, the sportsbooks odds are pretty accurate, so when it offers a 3 to 1 payoff, that event typically doesn't happen. So that kind of approach would have a very low win rate, but would occasionally get a 3-to-1 payoff. In the long run, the system would go broke just from the repeated losses. This is just my initial assumption, I don't have the specific data for that, so take it with a grain of salt.

Those sound like pretty interesting experiments, I'll get back to you on them!

Expand full comment