Thursday 2 April 2015

Books, Bias And Popularity

A couple of comments on recent posts, one which I meant to publish earlier but forgot about. It happens when you get old. Both are from FTL entrant Jamie A who made an excellent comment on my suggestion that:

"I would have expected the opposite [better results later in the season] with models absorbing promoted teams and improving as the season goes on, but that appears not to be the case. A reminder if anyone needs it, that profits can turn into losses, and vice versa, and even a season is a small sample size."
Jamie's comment was:
The models might well be absorbing promoted teams and improving as the season goes on but not as fast as the bookmakers and that is the key!
A very valid point, and somewhat related to his next comment on the "Midweek" v "Weekend" topic triggered by Graeme's TFA observations:
Are the bookmakers more or less accurate for weekend v midweek?
That is in the "something to look at in the Summer methinks" I mentioned.

As I made it clear (or so I thought), I'm not a statistician, and the post did not claim any findings of statistical significance, but given that midweek results have previously been identified as being less predictable (by at least Skeeve) it's an interesting topic, but it can wait for the season's end.

As I have mentioned recently, if we all wait for our possible / probable edges to be statistically confirmed we'd never place a bet. Successful betting is about much more than statistics, as the (relative) success of non-statisticians shows.

Personally, I find it useful to know that, for example, backing all home favourites in the implied 70%-79% probability range in certain leagues has an ROI of 7.2% (at Pinnacle's beatable prices, and thus a bottom line) over the past two seasons (347 selections).

It may not be statistically significant, (I'm  a non-statistician, and I'm generally not a fan of 'bands' - although Queen are pretty good), but the favourite-longshot bias is well established and studied and if we sit and watch it for another ten years before placing a bet, well, in the long run we are all dead.

As Joseph Buchdahl has commented previously:
Incredibly, had a punter backed every favourite at the best available price over this 2-year period, his return on investment would have been nearly 102% from almost 8,000 matches. At £100 per bet that translates into almost £13,000 profit in a little over 2 years, simply from using an odds comparison, considerably better than many who claim to be "sports tipsters". One could not, of course, rely on such profit being replicated time and time again.
Some might call that data-mining, but I don't agree when you are confirming an observation that was proposed before the period in question. As with my own Bundeslayga observations, Joseph is simply confirming that the bias has continued, and the key is the last sentence.
One could not, of course, rely on such profit being replicated time and time again.
Joseph's use of 'best available' is problematic for me though, since I don't think this paints a realistic picture for most of us. In theory yes, the point is made, but what is the reality for most people? I like to use Pinnacle Sports' prices, not only because their lines are sharp, but also because they don't limit winners and are thus a useful minimum.

Moving on, and there's been something of an up-tick (not a statistically significant one I hasten to add, but an up-tick nonetheless) in hits recently, with many coming from a country now in second place:
A little further drilling finds a huge number coming from one IP address, which is a little surprising given the owner's previous comments.
423 visits! I'm flattered. Perhaps I'm not that stupid after all? I should start charging.

2 comments:

Marty said...

I suspect Denmark has google now installed so it keeps polling your site to see if it has changed.

Marty said...

I suspect Denmark has google now installed so it keeps polling your site to see if it has changed.