RATINGS ANALYSIS: "overlapping" filters

Surely the one that is chucking out the 88.88% is intrinsically flawed - because if it loses the next two, you are at OVER 100% probability of winning for the next 7 games? Or have I misunderstood?

No, you haven't mis-understood.

A team that is rated at 80% loses three in row, yes, that one is now at 8/7.

So if
11 - loss
12- loss
13 - loss
on game 14, it would be 8/7.

This, in turn, would be rounded by the filters that were taking readings on:
2 to 11
3 to 12
4 to 13

My basis for this is that filters are created to rate how teams perform over "X" games.
A shorter read on form does not give a balance on performance (imo).
But then we create ratings that should be performance over a period that follows rather than the "next game".

Bear in mind, the decision on placing the bet is
a) first takes into account that the team is stronger than the opponent
b) then if it is value.

You also have to take into ac*****who they played in the previous 10.

For example if, say, Everton have played chelsea, man utd, arsenal, spurs, liverpool and man city, then the simple average calculated from those games will have little bearing on upcoming games against pompey, bolton, west ham and wolves.

*account!

Agreed, Zola. Was simplifying things. The number generated is based on the rated performance of each match. Thus, if an away team beats a top-rated team, the individual match/result is rated such.

But from there it is a matter of smoothing those results and I hadn't found anything similar in my on-line searches.

I've read it again.

I prefer your second option (b) because it puts less weight on the first rating which is now achieved over only 9 games rather than 10. Also the first rating now doesn't include the last game so its totally discounting last night's form.

This is a tough question to answer, it strikes me, since I remember reading about baseball players and their mentality over a season. Even the best in the world need to be used to taking beatings on a fairly regular basis. So very recent form doesn't have the effect that it does in say soccer.

My 2ps worth which may well be only worth about 1p in this situation!

Thanks for the input. Your reasoning for "b" and giving weight to recent play makes sense.

I also woke up wondering if a "moving average" approach would do the same thing (which is supported by what you wrote); in other words, take a reading over ten games and five games.

But I still think that the "overlap" will be one that will take form.

They are around eight games into the season so around two weeks to get it written and up!

What you're after us an exponential moving average.

Also your probabilities must be calculated for each possible combination of outcomes, from 0 out of 10 to 10 out of 10. They are all possible and the sum of them must be unity.

So that after the 10 games, one of them must have occurred. Yu can then back test to see if your predictions for the outcome sets match with what actually occurred.

Taking a prediction for 8 out of 10 and then inferring something from that after the next result seems error prone.

Taking a prediction for 8 out of 10 and then inferring something from that after the next result seems error prone.

I agree. We all create and read ratings based on the race/event at hand.

But spent last year reading papers by mathematicians, etc, who had taken on sports and ratings (in the States, U.S. college football is a constant debate of ratings).

I found several that created analysis and formulas that looked at team performances over a season to predict how they would do over the following season. From there, some would discuss how this could be incorporated against betting - in simple terms, if a team was underperforming, then the wins would be upcoming. Overperforming, the losses.

Of course this is all (over) simplified since they were simply taking math and fine-tuning it to try to get it closer to the real result (players are individually rated then if they leave the team, it is curved, etc).

But the main aspect they made me think about is the "long term" view rather than the "immediate result" (as Lola points out).

So taking form tends to, again, not to allow for the "20% result" when the team was "80%" and superior.

Taking what I have read over the past year, creating "mini-seasons" then betting not just on the next event but looking at the next "mini-season".

So on top of the over-lapping ratings, the step is to also look at which of the next series of games to bet on coupled with the value.

Another point that I have incorporated is to watch for which teams react to the ratings (whether predicted to perform or under perform).

Again I would be wary of how this is applied.

I can imagine a senario where say Man Utd got off to a poor start and based on previous seasons you might get value on their upcomming matches until they start producing the "expected" performances.

But I would expect you would be unlikely to find value on such a team in practice with the support bias.

Another prime example would have been Liverpool this season based on last season. You would be bankrupt.

My gut feeling would be this would only work on the sports where there was not so much of a dividing line between team performance in the first place (like the american sports) but consequently any value gains might be slight and extremely variable.

I am always wary of taking academic results at face value as they tend to test only specific cases and tune too much for my liking.

my advice is no1 can predict the future...

forget trying to predict sport outcomes and concentrate on predicting the markets (general public)

its a damn sight easier