Models, Balancing and Unconvinced at 18-0

My last post attracted some interest, with a record number of comments and 'likes' on Twitter. The days of people commenting on blogs seem to have passed, but on Twitter they are alive and well. I don't always tweet my new posts, some are admittedly not hugely interesting and seem unworthy of a tweet, but for the ones that are touching on areas of high interest and controversy, I should probably do this as a matter of routine to keep the conversation going. Some highlights are reproduced below.

l0l0 started the ball rolling with some thoughts on the bookmaker's process:

the process I imagine:

1)calculating "true" probabilities

2)knowing the habits of your clients, distort the offer in order to have an equilibrate state of the book at the end to match the "true" probs

3)make noise(publish 'betshares') when the book isn't close to equilibrium

I think this is about spot on. I would add that they should be managing limits so that the earlier lines are open to smaller bets than the later lines as the consensus on the price builds, We actually see this on major sporting events with Pinnacle when they announce closer to the start time that limits are now at some huge number. Regarding the 'habits of their clients', if they have a client who for example bets only on Maidstone United matches, and has a high success rate, they might well be happy to have all their green on this outcome and risk a zero profit or even a small list on the other outcomes but a well run business can't afford to take big risks.

On this topic, Joseph suggests that:

@aluckyaday is probably right. Some of it will be weight of money, some of it will be bookmakers exploiting squares, and some of it will be bookmakers attracting new customers. The picture is nuanced.

We are in agreement. Doctor C ( @oddsvantage) writes, in language highly unsuitable for a family blog, but I'll let it go this one time:

So does this mean that people have such a hard-on for Pinnacle prices because they don't block/limit sharp punters (which allows for improvement on their odds) rather than because of their analytics and models?

Regarding the idea of 'loss leading' prices, Joseph clarifies that he is referring to recreational bookmakers. I really don't give these organisations much thought, because in my opinion and experience, betting with them is unsustainable if you are even slightly sharp. But sure, if you are clueless or a novice, you can get some generous prices. Just don't give up your day job for them:

My WOCs system cannot possibly work because squares are this dumb to force bookmakers to offer thousands of loss leading prices. Recreational bookmakers do this to attract new customers by selling the "we offer great prices" message. Then they ban for regular exploitation.

Indeed they do.  

Ian ( @maxpowers1981 ) and Joseph get into something of a spat, with the conversation started off by Ian suggesting that:

Balancing books is a load of nonsense. It's a sub optimal strategy. Doesn't exist in practice generally.

I'm not sure what evidence Ian has for this assertion, none was offered, but the idea that that balancing books is a load of nonsense seems to be itself a load of nonsense. No one is suggesting books are almost evenly balanced across all outcomes, but a book that starts gambling isn't going to be in business long, and there are certainly several books that have gone out of business in the past. A well run business like Pinnacle don't need to take risks. 

Anyway, the man from Del Monte, Jez, chimes in with a pithy comment:

You may be the only person in this thread up to now who actually has a bet, Ian.

An unsubtle dig at Joseph perhaps, who admits to not being a gambler, and who rises to the bait, responding:

You may be the only person in this thread up to now who has contributed nothing, Jez. My threads are to facilitate debate about things which are to a significant extent unknown and uncertain. I don't see your name in the Forbes rich list, so presumably you're not sure either.

Ouch. Jez:

Two tribes. Those who gamble - however moderately - and those who just like collecting numbers as a hobby and would otherwise be on a draughty platform at Crewe taking them off trains.

There are certainly some people on the thread who are known not to bet but who have no job and spend several hours a day tweeting profusely and writing absolute nonsense, but we're not here to name names. I've written in the past that for me, sports investing is my video game replacement:

I was thinking in the shower this morning, (yes, once a week, whether I need one or not), Betfair really is the ultimate video game. I've never been one for games, (friends at work spend hours playing Call of Duty - why? What's the point?), but in many ways the exchanges are one big on-line game. It's me versus an unknown opponent. My opinion versus yours, except in this game the points are real money.

The 'real money' part is important since it is absolutely pointless to study markets and look for edges if you're not going to make money. 

Regarding my clarification of the ROI and P-values calculated by Joseph:

I'm not sure Joseph has got his ROI calculation quite right, because an ROI** of 104.8% from 3,511 bets would certainly be impressive. I think from looking at his chart, theoretical profits are actually 168.53 points, which is a far more modest, but still decent, ROI of 4.8% and a P-value of 2.341 assuming average odds of 3.0, a far cry from the 0.0006 tweeted. Or am I missing something here?

Joseph replied:

Yes, you were missing something. I used my yield distribution calculator rather than my p-value calculator. The later is programmed to compare ROI to break even (ROI - 100%). In the former, I set it to the blind betting returns which from memory was around 97.5%

I'm still not clear on what the P-value should be or what I am missing. Is the ROI 104.8% or 4.8%? Is the P-value 0.0006 or 2.341? 

Anyway, @TechnoRondo commented:

Good article with some fair points. However, stuff like "this pattern was there in 17 of 18 of the seasons that pattern in 12 of 13 last seasons" etc is imo not really convincing.

The word 'pattern' was not used by me in those examples. I see a big difference between identifying an area of the market that is perennially inefficient, versus something that is a 'pattern'. 

A 'pattern' to me would be something like that in every third year, favourites priced between 1.63 and 1.87 are profitable in the AFC when the game is played during the day in a domed stadium, i.e. completely random and useless.

When 18 of 18 years sees a 50/50 chance tilt one way, (one year there was a small loss due to commission, even though there were more winners than losers) I'm happy to accept that the market isn't efficient. This is more than a 'pattern' because it's not random; the a priori is that bettors have a tendency to favour home favourites. We have a hypothesis, and the data appears to support that. Maybe not statistically proving it, but as I wrote in my last post, you can't wait that long and markets correct themselves. Usually. 

Do I think this market will continue to be inefficient? No, but I said that ten years ago when it was 8-0! 

At what point would TechnoRondo be convinced? 35 out of 35? 

I'd suggest that if you're unconvinced with a record of 18-0, then you'll never be convinced, and you really might as well be standing on a draughty platform at Crewe spotting trains.

I mean seriously, what record would be convincing? 

TechnoRondo continued with:

First you should get a limited set of patterns (draws between equal teams could be one). After this it should hold in several competitions / seasons (large samples are necessary). Otherwise it is still gambling but no longer on football games but on bookmaker patterns.

Again, these are not 'bookmaker patterns'. Prices are "in general" determined by the money coming into the market. As for looking at several seasons and different leagues, I looked at several leagues and divisions in June 2018. for example from the Championship review:

For games where the difference is 25% or less, the profit is 23.14 points, ROI 2.3%, which is a lot lower than the same parameter produces in the Premier League.

I may do the same again this summer if time permits, but I wouldn't expect the same results from lower leagues as from the EPL as the markets are different, both in terms of volume and the participants.