Tuesday 27 August 2013

Soccer In Play

Danny Murphy is back, but with a considered and accurate comment on a subject I don't think has been discussed too widely in the blog world:

Cassini I must congratulate you for the excellent reading of the Sunderland Fulham pre match odds in your recent Betting Expert article of August 13th.
This is in stark contrast to a similar attempt by Mr Mark Iverson on the Norwich Everton match in his blog.
To me his pre match model appears to be very crude, grouping teams into very broad bands as it does. How you could rationally argue Norwich to be 2.41 shots against Everton is beyond my comprehension and it should be noted the market had Norwich at 3.81 at kick off.
The funny thing is Mark's conclusion of either backing Norwich or laying Everton was correct, albeit for the wrong reasons. If you know your odds construction there is no way Norwich could be much less than 3.00 for this game unless Everton had a lot of injuries but in fact it was Norwich who had injury doubts themselves. Norwich at 2.41 is not a credible percentage against such a strong opponent.
If Mark had seen Norwich at 3.81 and viewed them as 3.00 shots that would be rational but the 2.41 was a rogue figure.

Perhaps it is a case of 'garbage in garbage out' with this model?
In my opinion, an app of this kind would be an excellent idea, and something that would well be worth a lot of money if it worked. The idea of being able to input two teams and the current game-state and see the probability of various outcomes based on historical data is as good as it gets, especially if it's for free, but unfortunately, the "Soccer In Play" app is not this model, and falls short in a number of ways.

For a start, it does not allow you to input two teams, but only two, often very subjective, categories of teams.

The input options are limited to A Top Team, Top Half Team, Bottom Half Team and Relegation Candidate. While the category for some teams in some leagues might be clear-cut, the definitions are vague, and many teams are hard to categorise, especially early in a season.

There is also no opportunity to select the league nor indeed any indication of what league or leagues the historical data is from. This is important, because the statistics from the Bundesliga are quite different from Ligue 1 for example, and each league and division has its own personality.

The app does say that it uses “data from across the top leagues in the world” which implies more than just Europe, but where these leagues are is unknown. South America perhaps, but presumably not MLS where there is no relegation and thus no team would ever be a "Relegation Candidate".

There is no consideration for form, but even more seriously, the numbers are simply wrong, or 'rogue' as Danny phrases it.

The image below gives the projected pre-game percentages for the 16 possible combinations of matches between the four classifications of teams.
Look closely.

Now does it really make sense that “A Top Team”, would have the same probability (0.6453) of beating a “Top Half” team as they would of beating “A Relegation Candidate” or a "Bottom Half Team"?

And those exact same numbers are coincidentally the same for a “Top Half” team beating a “Relegation Candidate”.

In fact, of the 16 possible combinations, (allowing for rounding) no less than five are 41 / 29 / 29, four are 65 / 22 / 13, another four are 28 / 27 / 44 and the other three are 49 / 28 / 23. 
Is a “Bottom Half” team really as likely to beat another “Bottom Half” team as it is a “Top Half team”? Mark uses Norwich City v Everton as an example of this latter match up, but his model would generate exactly the same odds were Norwich City at home to say Cardiff City, Stoke City or Sunderland.

Is “A Top Team” really as likely to lose at a “Top Half Team” as it is at a “Relegation Candidate”?

For a “Relegation Candidate”, they apparently have the same probability of beating “A Top Team” as they do a “Top Half” team, and identical probabilities when playing a “Bottom Half” team or another “Relegation Candidate”.

But only at home. Away, they have the same chance of winning at “A Top Team” as they do at a “Top Half” team. I would love to see the database for these statistics.

There are some other anomalies which you can see in the numbers above, but the bottom line is that while playing with the app may be a curiosity, it is certainly not an app that should be used for making investing decisions. The ULTIMATE in-play soccer app it is not.
The numbers discussed are those pre-game, but there is no evidence to suggest that the quality of the in-play data would be any better.

The presence of a bet365 advertisement on the app is revealing though. I've seen them advertise somewhere before. At first sight, one might wonder why on Earth a bookmaker would sponsor such a valuable weapon in the punter’s arsenal, but on reflection, it’s all rather obvious - along with why the "Soccer In Play" app is completely free. 
Out of interest, I looked at tonight's Freiburg v Bayern Munich match, and my model has Bayern Munich at 1.29. Soccer In Play has them at 2.26 (if Freiburg are considered relegation candidates, a bottom half, or a top half team). Betfair has them at 1.275.

Full Disclosure: I have no affiliation with Danny Murphy, the Soccer In Play app or bet365 and screenshots are for illustrative purposes only, and should not be considered recommendations. 

2 comments:

Random Man said...

I don't know about Danny Murphy, you seem more like what Fergie might have called the Dennis Wise of the betting blogosphere.

Anonymous said...

You really do seem to have some grudge against Mark, don't you? Dunno whether that's just jealousy of people doing this for a living or just your usual way of slagging off other bloggers to generate traffic.

I haven't any Apple pads etc so haven't actually used the app but can see the reasoning behind it. I'm sure you'd be aware any odds it gives out aren't meant to be 'true' odds of that event occurring but some indicator that the current odds are likely to be wrong. I've had quite a few automated ratings systems in the past and the majority of odds from any initial run is pretty much in line with the firms it's the ones that are out of line that merit further investigation as whether to bet or not.