Question for anyone who beats closing money lines in MLB.

**tto827** · 03-19-13, 12:16 PM

Very strange. Although I have heard BTCL in baseball is not as important as in other sports. Definitely want to see the responses you get from some of the wiser guys here.

**matthew919** · 03-19-13, 12:37 PM

With the huge liquidity for MLB, I would assume BTCL to be incredibly important. Smaller markets and CBB maybe not as much. But I could be wrong.

I imagine that betting moneylines also introduces much more variance- hitting a few dogs will do wonders, while losing on a few heavy favorites will slaughter your ROI, even for a sample size in the hundreds. I'm wondering if that in itself is enough to necessitate much larger sample sizes for testing. My backup plan is to shrug it off and stick with what I do well- totals; but aberrations like this really bug me.

**HUY** · 03-19-13, 03:58 PM

Originally posted by matthew919

Ok, here's a question that's driving me bonkers. I'm modeling baseball moneylines and am BTCL 67.2% over 336 plays (an outlier subset from the 2700 game test set). The average line move (in implied probability) from open to close is over 1% in agreement with this model. Also, these data are 100% clean- that much I'm sure of.

But the ROI for these plays is underwater at a miserable -2.6%, when theory dictates it should be somewhere around 2-3%.

Why would theory dictate that? If the line moves only 1%, why do you expect to make a profit?

**matthew919** · 03-19-13, 04:19 PM

Common wisdom says that every 2 cents worth of movement, on average, should translate to about 1% ROI. So a movement of 1% in implied probability works out to about 2-3% ROI.

Are you saying that an average of 1% movement in implied probability is not very good? If so, I think you're wrong. I mean, at an average of 2% you can make free money by line shopping and arbing alone.

**u21c3f6** · 03-19-13, 07:12 PM

I think you are forgetting the vig built into the line. The line has to move somewhat just for you to break even and move even more than that in order to profit.

Joe.

**Pancho sanza** · 03-19-13, 07:29 PM

A 5 % swing from expected % over 336 plays can easily be attributed to variance.

**339955** · 03-20-13, 10:03 AM

matthe what is the subset of games you are working with? could you explain that?

sounds like you are losing because of variance. it is possible to lose over money over 1000 games while beating closing games.

**matthew919** · 03-20-13, 10:36 AM

No, Joe hit it on the head. I wrongly assumed that 2c ~ 1% ROI for the no vig line. Did some more looking into it and confirmed that you do need to account for the vig when estimating your ROI. After factoring this in, I should be a net loser on MLs, even though the vast majority of line moves are in my favor.

Which is very good to know. Now I'll be staying away from the ML until I can generate a better line. At least my totals are razor sharp.

Can't wait. 11 more days til bases. Get pumped.

**339955** · 03-20-13, 02:01 PM

no? variance can't explain your losses?

**matthew919** · 03-20-13, 02:18 PM

Extremely unlikely.

**339955** · 03-20-13, 02:29 PM

ok sorry matty didn't mean to give u wrong answer just wanted to help but i messed up

**mcuni** · 05-22-13, 09:15 AM

Matt, i've read your thread about MLB run totals and I have couple of things to say regarding both threads.

1. If you beat the closing line of pinnacle 67% of time and have a negative ROI then outlier subset you're betting on might be pretty specific. You might be betting on favorite most of time with average odd of less than 1.50 (sorry, i got used to european system). There is just no other possibility.

I'm trying to build MLB system myself. One of methods I tried is splitting games into several clusters and dealing with them separately. One of clusters then contains lot of games where home team wins with probability of about 70% and small home team odds (mostly 1.40-1.50). Doesn't your outlier subset consist of such games?

2. More data is definitely the answer as 300 cases if pretty small amount.

3. When you write about beating the closing line in thread about totals, you have a record like 13-5-1 for Unders which means that 13 times out of 19 you bought 'Under' for better price than you would have payed tight before the game, correct? It doesn't involve the actual result of those 13 games, does it? 10 of those 13 games could be 'Over', do you still count them like if you managed to BTCL? If so, is it because you believe in market efficiency?

I asked these questions just to ensure we are on the same field. But what if market is not efficient? I read what you think about this statement, but here are the arguements: a) line is calculated by models like ones you or me are trying to construct, these models give cold weighted view on O/U, who will win etc; b) wages are done by public, and crowd is told to be generally stupid, it tends to expect more goals, it has sympathies so this distorts the real picture, quite accurately simulated with algorithms; c) as known, 95%+ of players lose money; d) so how can we talk about market efficiency? I have tested opening and closing lines from covers.com once and I believe I found that opening lines were sharper than closing ones (I need to find calculations to confirm that I remember it right).

So in this logic beating line movements (not closing line!) doesn't contribute to your ROI. Line moves are quite big, if market was effiecient it would have eaten bookies' juice and they would have gone belly up.

**matthew919** · 05-25-13, 09:04 AM

Originally posted by mcuni

Matt, i've read your thread about MLB run totals and I have couple of things to say regarding both threads.

1. If you beat the closing line of pinnacle 67% of time and have a negative ROI then outlier subset you're betting on might be pretty specific. You might be betting on favorite most of time with average odd of less than 1.50 (sorry, i got used to european system). There is just no other possibility.

I'm trying to build MLB system myself. One of methods I tried is splitting games into several clusters and dealing with them separately. One of clusters then contains lot of games where home team wins with probability of about 70% and small home team odds (mostly 1.40-1.50). Doesn't your outlier subset consist of such games?

2. More data is definitely the answer as 300 cases if pretty small amount.

3. When you write about beating the closing line in thread about totals, you have a record like 13-5-1 for Unders which means that 13 times out of 19 you bought 'Under' for better price than you would have payed tight before the game, correct? It doesn't involve the actual result of those 13 games, does it? 10 of those 13 games could be 'Over', do you still count them like if you managed to BTCL? If so, is it because you believe in market efficiency?

I asked these questions just to ensure we are on the same field. But what if market is not efficient? I read what you think about this statement, but here are the arguements: a) line is calculated by models like ones you or me are trying to construct, these models give cold weighted view on O/U, who will win etc; b) wages are done by public, and crowd is told to be generally stupid, it tends to expect more goals, it has sympathies so this distorts the real picture, quite accurately simulated with algorithms; c) as known, 95%+ of players lose money; d) so how can we talk about market efficiency? I have tested opening and closing lines from covers.com once and I believe I found that opening lines were sharper than closing ones (I need to find calculations to confirm that I remember it right).

So in this logic beating line movements (not closing line!) doesn't contribute to your ROI. Line moves are quite big, if market was effiecient it would have eaten bookies' juice and they would have gone belly up.

Whatever odds data you are using to test is flawed. I can say with 100% certainty that closing lines are more accurate.

I've seen that the greater the discrepancy between my model and the opener, the greater the line movement between open and close. Additionally, the greater the discrepancy, the higher overall win rate. To me, that's not a coincidence.

**mcuni** · 05-27-13, 02:23 AM

Originally posted by matthew919

Whatever odds data you are using to test is flawed. I can say with 100% certainty that closing lines are more accurate.

I've seen that the greater the discrepancy between my model and the opener, the greater the line movement between open and close. Additionally, the greater the discrepancy, the higher overall win rate. To me, that's not a coincidence.

That's really interesting. It can be flawed, I don't remember the source of it.
I hope to check it this season, I'm tracking pinnacle line most of the days and hope to revert with some facts in autumn.