[tweek]

Last year, I decided to start logging my own MLB money lines from several sports books, as this seems to be the hardest piece of "historical" information to find. I started logging in mid June 2008 through the end of the season.

Right now I'm looking at the difference between opening and closing lines. I have 3 different metrics that I am using to evaluate the performance of any particular predictor (that predicts a win% for each team). You can read more about the metrics I used here.

I have 1309 games logged from last year. I used the winning percentages "predicted" by the lines as an input to each of these metrics. Depending on the metric, the opening lines show an improvement of 2.5 - 3.5% over simply assigning 50/50 to each team. This number seams reasonable to me.

However, what seems surprising is the closing lines. The closing lines show only 0.1% - 0.15% improvement over the opening lines in terms of their predictive power. I would have expected a much larger improvement in predictive power from the closing lines.

So, what do you think? Do these numbers make any sense? Is my data bad? Are my metrics bad? Or, are closing lines really not much more "powerful" as a predictor than opening lines?

**One thing to note, I only logged lines ~10 times per day, so my data won't show any motion that happens either right after the line comes out or right before game time.

[Dark Horse]

I like what your research could signify, but am not sure how meaningful your experiment is without the earliest line movement.

Are there patterns, within line ranges, that repeat day after day?

[Wrecktangle]

However, what seems surprising is the closing lines. The closing lines show only 0.1% - 0.15% improvement over the opening lines in terms of their predictive power. I would have expected a much larger improvement in predictive power from the closing lines.

So, what do you think? Do these numbers make any sense? Is my data bad? Are my metrics bad? Or, are closing lines really not much more "powerful" as a predictor than opening lines?

**One thing to note, I only logged lines ~10 times per day, so my data won't show any motion that happens either right after the line comes out or right before game time.

I'm not surprised at this, as baseball is probably the most analyzed sport in existence, a more mature line is built probably by the lines makers. In the second most analyzed sport: NFL sides, I've seen similar trends, but then I don't follow the absolute earliest line as I don't have enough information early enough to take advantage of it.

[tweek]

I like what your research could signify, but am not sure how meaningful your experiment is without the earliest line movement.

Are there patterns, within line ranges, that repeat day after day?

I haven't looked into something like this. One thing I'm wondering, and may look at, is the improvement in predictive power of closing lines that change more than x% (x = 10?) from the opening line.

My guess is this sample would show a much larger increase in predictive power.

[reno cool]

Last year, I decided to start logging my own MLB money lines from several sports books, as this seems to be the hardest piece of "historical" information to find. I started logging in mid June 2008 through the end of the season.

Right now I'm looking at the difference between opening and closing lines. I have 3 different metrics that I am using to evaluate the performance of any particular predictor (that predicts a win% for each team). You can read more about the metrics I used here.

I have 1309 games logged from last year. I used the winning percentages "predicted" by the lines as an input to each of these metrics. Depending on the metric, the opening lines show an improvement of 2.5 - 3.5% over simply assigning 50/50 to each team. This number seams reasonable to me.

However, what seems surprising is the closing lines. The closing lines show only 0.1% - 0.15% improvement over the opening lines in terms of their predictive power. I would have expected a much larger improvement in predictive power from the closing lines.

So, what do you think? Do these numbers make any sense? Is my data bad? Are my metrics bad? Or, are closing lines really not much more "powerful" as a predictor than opening lines?

**One thing to note, I only logged lines ~10 times per day, so my data won't show any motion that happens either right after the line comes out or right before game time.

that would be my guess.

[MonkeyF0cker]

I haven't looked into something like this. One thing I'm wondering, and may look at, is the improvement in predictive power of closing lines that change more than x% (x = 10?) from the opening line.

My guess is this sample would show a much larger increase in predictive power.

Two things. First, your sample size really isn't large enough. It only encompasses about half of a season. You should have at least four or five seasons of data. Line histories are not that difficult to find. There are plenty of sites that offer them for quite a few seasons.

Second, you are including lines that have relatively little to no movement from open to close in your dataset. You should attempt to find correlation between line movement from open to close and the results of those contests.

[tweek]

Two things. First, your sample size really isn't large enough. It only encompasses about half of a season. You should have at least four or five seasons of data. Line histories are not that difficult to find. There are plenty of sites that offer them for quite a few seasons.

Agreed. Is there a place you know where you can get opening and closing lines? Never come across this before (hence why I took it upon myself to log my own lines).

[reno cool]

big line moves would have to be more "predictive". Otherwise you can make $ just betting lines that moved+.
It actually might not be the worst idea in the world to check that.

[MonkeyF0cker]

"

[Ominous]

big line moves would have to be more "predictive". Otherwise you can make $ just betting lines that moved+.
It actually might not be the worst idea in the world to check that.

I have recently taken up an intrest to this and I am currently following MLB games but I have to few data. Also I am doing it manually so...

Anyhow, @OP. Could you possibly relate your data to the actual linemoves? I mean, you say that the closing line only predicts the outcome a few 0.1%s better than the opening line, BUT only about 25% of the MLB games have line-moves of more than 10 cents.

It would be useful if you could somehow relate the actual moves and try to determine if they were justified, however, it might prove a mathematical challenge

[wintermute]

Closing lines in MLB seem to be quite a bit more efficient than the openers.

I have a baseball simulator and I've run it against the 2007 and 2008 seasons using Pinnacle's opening and closing lines.

The results I get against the opening lines are about 1.5 percent better than against the closers for both money lines and run lines. There is a smaller improvement for totals. This is over a sample of about 4000 games.

[duanedibley]

However, what seems surprising is the closing lines. The closing lines show only 0.1% - 0.15% improvement over the opening lines in terms of their predictive power. I would have expected a much larger improvement in predictive power from the closing lines.

Do you have the RMSE or MAE for opening and closing run lines vs actual scores?