Value Is Not An Option

A couple of intelligent comments on the debate of subjective versus objective analysis as it relates to football. Bayes is well known on the Betfair Forum and his thoughts are always well worth reading, and he had this to say:

I'm afraid I'm with The Investor on this one.

You are correct that match odds are determined by goal expectancy, but pre off goal expectation for both sides can be very different from what's observed. The market tends to be pretty slow at picking this up (although it's getting quicker) and good match readers have a decent edge.

In the glory days of Betfair treating their big customers like royalty, they took me and a few others to Roland Garros. It was 2008 and one of the punters there was a young guy who had a huge edge in pre off football. He was a mathematician but the crux of his edge was that he watched lots of live lower league football and recordings of all UK televised football. I'm sure he used some maths as well, but his edge was largely based on his subjective judgement.

No need to be afraid - the Investor is as much respected as Bayes is, and I am just facilitating a debate on this topic.

A match could very well begin in a style which the market wasn't expecting, but while I don't look at games in play these days, certainly in the past it was extremely rare to see the Unders price drift from the opening kick-off without a goal actually being scored, or conversely drop like a stone. It would sometimes stall, but never for too long due to the downward pressure from the time component, but if the goal expectancy was significantly different to pre-game, shouldn't either of these movements have been easily observable in these markets?

Perhaps this is what Bayes means by the market being slow at picking these up, but markets usually adjust pretty quickly, especially liquid ones with thousands of eyes watching and trading the games, and I'm just not sure how much opportunity there is for 'good match readers' to make decent profits in football. Other sports where 'genuine' trading happens (prices move more gently depending on the events being observed) I agree.

To the second part of Bayes' comment, I absolutely agree that in lower leagues this is very possible, but the debate was specifically related to 'big markets'. If you specialise in an area in which there is less interest, then it is clearly much easier to be an expert. I recall reading a few years ago about someone who specialised in the lower Scottish Leagues, watching games in person, building relationships with clubs and players and essentially becoming all-knowing about these leagues. The trick is finding a balance between being an expert but where there is little or no money and being knowledgeable, but less of an expert due to the number of rivals you have, in leagues where there is more money - 'big markets'.

Anecdotal piece here: I noticed yesterday that my son's former team of Chipstead were listed in Betfair's FA Cup matches. Not too many people watch Chipstead play, so the number of 'experts' is few, but was there any money to be made? When I looked pre-off, 1.02 was available on all three options (although given the result, that might have been value for the away option!). The point though is that there is a minimum level where you expertise can be viably profitable, and probably that is around the Football Conference level.

Emp is alive and well, and chimed in with a comment on the same topic:

"Football prices are derived from goal expectancies, the current score and the time remaining, and with occasional blips where a goal is more probable, the trend is pre-determined."

Whether they should be is debatable. My system doesn't care at all about "goal expectancies" for instance. I think this argument is coming perilously close to begging the question.

"The sporting markets are different, but one similarity would be that it seems unlikely that someone making little to no use of data would be able to consistently identify 'incorrect odds' and have a real and long-term edge over a more sophisticated (in technical / data terms) and professional (in every sense of the word) rival trader in a big market"

According to this logic, there is a 1:1 (or close) correspondence between the complexity of the data used and accuracy of your predictions. That's a very debatable conclusion. Even if professional firms use complex data, they can still be analysing it in the wrong way and asking the wrong questions. In fact, institutional hierarchies and group think make it very difficult to reverse such behaviour if it's there.

Finally if all your liquidity and prices come from goal expectancy models, the edge of people using them should theoretically constantly diminish (to the point that deviations from your expectancy are more likely to be due to valid circumstantial differences than due to mispricing).

While there is always more than one way to skin a cat, the only way you can be profitable long-term is by identifying value, and to do this you need to be able to price up matches, i.e. you need to be able to accurately determine the probabilities of various outcomes and compare these with the market. If professional firms are analysing it in the wrong way, their results will show this and the model will need to be improved.

"Whether they should be is debatable"? How is this debatable? I am not saying that the goal expectancy is static. As Bayes noted, this number may well change in-play, (and pre-game as team line-up news, injuries, tactical news filters out) but as option prices can be determined using the Black–Scholes–Merton formula, so prices in football markets can be determined using the parameters I listed. The Black-Scholes-Merton formula has been proven to be 'fairly close' to the actual prices, and the even the observed 'volatility smile' identified might be a parallel to market movements seen in football when there is a breakaway, a 'free-kick just outside the box', a corner, possible red-card, injury etc.  Incidentally, unlike many sports, the time remaining is also a variable in football, and I suspect the likely amount of time added on is an opportunity, but the formula holds regardless.

The issue of how complex the data used should be is another thing. Again a balance needs to be struck. Most of us do not have the time or ability to look at data beyond a certain point, and the usefulness of some data points is debatable anyway. Possession is a good example. In my opinion, the raw possession percentage is meaningless. The type of possession is what is relevant, and this information is generally not freely available.

As for liquidity coming from goal expectancy models, I'm not sure where this comes from. Money comes in from individuals or entities who either believe they have identified value, or want to back their favourite team or have an interest in a match. I doubt that too many casual punters have priced the games up using goal expectancies, but anyone serious about winning will have done so.

Successful long-term betting isn't about finding winners - it's about identifying value and doing this subjectively is a lot harder than doing it objectively - in my opinion.