Thursday, 6 March 2014

Controlling The Process

On the topic of draws and my assertion that the value is in matches where the true goal expectancy is less than the market's goal expectancy, Danny Murphy comments:

I won't dispute a low goal expectation is a rich source for finding potential draws but it needn't be the only method.
Another method is finding a team near the bottom playing at home against a stronger team. You do get these teams that are decent coming forward but can't keep a clean sheet. Quite a few fall in to the 'too good to go down' category but at the end of the season you see they have drawn a stack of home games.
Could that be the case here?
Most seasons, there are more draws on the grounds of bottom-half teams than top-half, although not always. In the EPL 2007-08, the top ten sides drew 52 matches at home, the bottom ten drew 48, but what Danny says is usually the case, although until the end of the season, you may not be sure whether a team is top half or bottom half, which is why I prefer to use a more nimble Elo based rating system than use rigid and often misleading league tables when evaluating matches.

The thing is though that we are not just looking for draws. We are looking for draws that are value, and I suspect that the "lower / weaker side plus home advantage = away side" formula is well known and factored into the draw prices already. The 3:3 draw at Crystal Palace (v Southampton) or Norwich City (v Stoke City) this Saturday is available at 100-1 for a reason. Speaking of the 3:3 draw, something I remember from my Racing Post reading days was an article about the price on this outcome being 'wrong', and an individual, or individuals, were backing the outcome in multiples forcing the high-street bookmakers to reduce the odds from 100-1 to 80-1. Perhaps someone reading this can add some details.

The match that stated this discussion was Skeeve's draw selection in the AFC Chester v Nuneaton Town match.

Skeeve commented:
I've just replied to Steve over there, but you were right about the "that's more like it" part of the recap e-mail. 
I definitely didn't expect a goalless draw (I've now published the whole preview at Steve's blog) and of course it was lucky, but Chester have now shared points in five of their six home games against top-eight Skrill Premier teams. I guess this bet had it all - it was a good bet, a winning bet and a lucky bet. :)
We all need luck, but luck evens out over time. Skill (or talent) doesn't:
The key to identifying talent versus luck comes down to understanding the difference between process and outcome. As Barry Ritholtz put it last week: 
Outcome is simply the final score: Who won the game; what numbers came up in a roll of the dice; how high did a stock go. Outcome is the result, regardless of the method used to achieve it. It is not controllable. You can blow on the dice all you want, but whether they come up "seven" is still a function of random luck.
Process, on the other hand, is a specific methodology. It is a repeatable approach to any challenge or endeavour, be it construction or medicine or investing. And you can control a process.

1 comment:

Danny Murphy said...

Just to clarify about the lower team at home to a strong team situation, the method I read works on the principle of projected finish position and not current tables. In other words the crystal ball!

While that may be horribly unscientific it can be an aid to finding value. To give an example you can get a good side say Everton who have won the previous home game 4-0 and say West Brom have lost away. With a game like this Everton can end up getting backed heavily pushing the draw into value. You could regard it as an indicator rather than a method.

On the subject of value it could be argued a game with a low scoring profile like Norwich Stoke wouldn't produce value, yet you manage to find some. But then whatever approach we take we only need to be 2% better than the market and we're quids in, which is the inherent conundrum of value in the first place.

Cassini I have an idea for FTL. To spice things up between now and the end of the season why not allow 1pt correct score predictions? As luck would have it I have just finished calibrating a killer CS prediction model.

I could be a contender!