- This topic has 132 replies, 20 voices, and was last updated 16 years, 1 month ago by
.
-
Posts
-
Dave,
Nassim Taleb is certainly an original thinker and he cuts right through the garbage spouted by economists and traders. I read FBR some time ago and it impressed me as one of the few books worth reading in this area. Just got The Black Swan, his latest, for some autumn evening reading.
I read it and thoroughly enjoyed it.
It certainly helped me in reaching better strategies as well.
I believe a lot of Nassim Taleb’s observations(I haven’t read the book yet) are valid where subjective probabilities are concerned.
There are some patterns which emerge from looking at thousands of racing or football results, but the chaotic nature of each individual race or football game does make you wonder if there is any point in systems based on deductive approaches(i.e. nearly all systems).
For example, we know by observation that 28% of all football games ever played in the main English football leagues have been drawn, but that information has only limited use when looking at individual games played in the present or future. We need more information about the past games such as league positions of each team before the game started and perhaps several more facts. Even then, we cannot be sure that the pattern will continue.
Horse racing is even more problematic. Any systems based rigidly on what happened in the past to various categories such as favs, trainers or jockeys, have all been shown to be very shallow. Also, other systems that produce various ‘rules’ are invariably worthless when given a prolonged trial.
It seems to me that the best approach is that of form study using a problem solving model. This takes the form of thinking about how an event might turn out by using all available information, assessing it and then assigning your own probabilities to the possible outcomes of that event.
Happily, this is how most people here proceed. Those who are still searching for a repeatable formula using inputs based on past events are frankly barking up the wrong tree.
It took me a long time to see that basic truth.
What really interested me about it was how your mind can deceive you because it always looks for patterns and logic even though the events are completely random.
I think the betting exchange and the 100% book has caught out a lot of punters. It seems to work something like this. You are lucky and get a lot of positive results. Your mind adds two and two and you think you have an edge because you are in front, laying 100/1 shots would be a good example of this. It goes on for long enough and you develop the Messiah syndrome.
The inevitable happens and your results then fall onto the negative side, as things start evening themselves out a bit.
When you’ve blown your bank and a bit more, you then post up something like .. ‘Betfair has ruined my life’. In fact all that has happened is that you have been fooled by randomness and you never had an edge in the first place.
I’m quite interested in how random other people think that random events actually are, is it possible to measure it? Or maybe by its very nature it cannot be measured.
Obviously, things are more random at lower chances of probability, so you are more likely to be lucky/unlucky, than if you are at higher chances of probability. Its interesting that the on-line hi/lo games results are based on 300,000,000 games to generate their results. These games are audited and are judged to be ‘fair’. So, what implication does this have for the effect of luck in horse racing, can any conclusions be drawn ?
What really interested me about it was how your mind can deceive you because it always looks for patterns and logic even though the events are completely random.
I’m quite interested in how random other people think that random events actually are, is it possible to measure it? Or maybe by its very nature it cannot be measured.
Obviously, things are more random at lower chances of probability, so you are more likely to be lucky/unlucky, than if you are at higher chances of probability.
So, what implication does this have for the effect of luck in horse racing, can any conclusions be drawn ?
Horse racing is more of a closed universe in that it has specific rules and the event type is repeated daily, worldwide. So you have a chance of prediction if you can pin down, prove and predict causation, which Artemis was getting at. A system might have a provable cause (or Nick Mordin type theory) as to why it might work and when it won’t work. A system that as you say, just looks at past results and backfits some rules is as hopeless in predicting future events as the commentator saying that the jockey winning the last race is "bang in form"- whatever that might mean "and must be followed" – ’til another jockey wins the next race, ad nauseum.
Causation to me has to be, which horse on the day has the speed, ability and fitness to reach the winning line before the others. It might still not be completely predictable as form in the past is the past; fitness is not even fully known by the trainer; the going might be totally different from the advertised and the racing pace that dictates the result is yet another estimate. At least you are pointing your estimates in the same , consistent direction.
If you stick to one method then you can have a chance of seeing how well it behaves, when it behaves well and what random effects might apply. If you chop and change every time a result does not go the way predicted then no chance of learning anything. The method / system is as random as the result.
Being random means that it is as unpredictable at low odds as for high odds. So a hot favourite that breaks a blood vessel for the first time can fail and can fail for many other reasons that are ignored as unlikely in a single event even if they do occur quite regularly over a season. Is that random (unlucky) or something that is known but not included for properly?
I can agree with all of that, particularly .. Artemis ‘problem solving model’ and everything that you have said there Robert.
Uncertainty (luck), is really what I am getting at here. I know some people look to adjust their models to factor in how badly a horse might run and create a sort of rating band for each runner, which doesn’t work imo. Like horses that will either finish first or last.
Being random means that it is as unpredictable at low odds as for high odds
I’m going to have a think about that. I don’t think that you are right, but you might be .. you’ve got me thinking, well done!
Followed the links from the original and there is some interesting reading
It made me wonder if we overestimate/underestimate probability and think, are the outcome of events less predictable than we as estimators of probability assume.
Are we possibly putting a larger bias on certain factors than we should, are these events more competitive than we really want to believe?
Ie:is that top class horse giving weight to all his rivals up against today or will his class still see him take the race, but the probability, dominance he has is less than we, the market assumes and therefore one or two of the opposition has an higher probability of wining
Robert .. I think you may well be right about luck being the same regarless of the chances of something happening, but it’s effect would be different when measured over time.
Are we possibly putting a larger bias on certain factors than we should, are these events more competitive than we really want to believe?
Yes and no. If a horse has a 25% chance of winning, it also has a 75% chance of losing. If we run that over 100 races we would expect the horse to win 25 times and lose 75 times. But what if we don’t get exactly 25 winners? what if we get 23/77 split, could we expect the next 100 to give us 27/73, or might our split never reach parity in relation to its chance? When all is said and done about pricing races, all we are doing is glorified guesswork, based on what has happened in the past. I am coming around to think that we would be better thinking that odds should have a +/- built in to their chances to cater for luck. I dont know if anyone does this?
Someone once wrote, ‘systems dont work because they are overcome by massive waves of luck both good and bad.’ I’d agree with that .. I think the randomness that Taleb talks about relates more to being over-confident (and thereby over staking) in your own ability which has been inflated by good luck. When things level off over time your in a mess.
Robert .. I think you may well be right about luck being the same regardless of the chances of something happening, but it’s effect would be different when measured over time.
Are we possibly putting a larger bias on certain factors than we should, are these events more competitive than we really want to believe?
Yes and no. If a horse has a 25% chance of winning, it also has a 75% chance of losing. If we run that over 100 races we would expect the horse to win 25 times and lose 75 times. But what if we don’t get exactly 25 winners? what if we get 23/77 split, could we expect the next 100 to give us 27/73, or might our split never reach parity in relation to its chance? When all is said and done about pricing races, all we are doing is glorified guesswork, based on what has happened in the past. I am coming around to think that we would be better thinking that odds should have a +/- built in to their chances to cater for luck. I dont know if anyone does this?
Someone once wrote, ‘systems dont work because they are overcome by massive waves of luck both good and bad.’ I’d agree with that .. I think the randomness that Taleb talks about relates more to being over-confident (and thereby over staking) in your own ability which has been inflated by good luck. When things level off over time your in a mess.
Dave,
For your example of the 25% true chance of winning expecting to win 25 races in a 100:
The Bernoulli equation which is the basic formulae for such random events gives the following probabilities of wins in a 100 races:15 wins 0.57%
20 wins 4.93%
25 wins 9.18%
30 wins 4.57%
35 wins 0.70%, and so onYou have to add up all the possibilities from 1 win in a 100 to a 100 wins in a 100 to get 100%.
The chance of the "expected" 25 wins is only 9.18%, when most people would believe it to be nearer 100% – they would put money on it but again are fooled by randomness. This may be where "luck" in systems or pricing can be more rationally explained.
You can never really say a horse is 3/1 but you can say it is somewhere between 5/2 and 7/2 with some probability. You cannot believe a horse is rated at exactly 100 but at best is rated somewhere between 95 and 105, say. Your system which on average picks out 25 winners in a 100 system race trial may only score 20 for the next series of 100 races. Bad luck ?-. 30 wins – Good Luck? No – absolutely nothing has changed – it is just how probability works.
I would say yes, Dave as robert’s excellent illustration shows.
Good luck is when you are favoured quite randomly by chance; bad luck when you are not. You can make probability work for you by skewing the odds in your favour but you have no control over luck or random chance.
The longer you play with odds in your favour, the more likely you are to reduce the effects of luck.Using Bernouilli’s equation shows the expected probability distibution is less skewed in proportion to the number of trials. In other words, as you increase the number of trials, you will find more values nearer to the middle of the range.
But using Bernouilli’s equation, I could expect to have between 20 and 30 winners 79% of the time, the remainder I get 100 – 79 = 21% of the time I don’t, providing that the distribution is normal. Taking the distribution to be normal, 1/2 of the time I will get less than 20 winners and half of the time more than 30, so how can I possibly conclude that I have an edge and I’m not just getting random results?
I don’t believe you can ever prove conclusively that you have an edge in relation to betting on horses or most other sports.
If you were to bet on about 800 races and show a profit, I would accept that you had an edge but someone else might accept a lesser or greater number. Pedants might say that it can never be proved because your next series of 800 bets might wipe out those profits.
Some will say that bookmakers(obviously) enjoy an edge. While this is self-evident, it is possible for them to lose over a period. Nothing is certain where subjective probability and luck are concerned, although we can make much more confident predictions about things like cards and dice, where the parameters are objective and well defined.
Even with cards, dice and roulette wheels(in particular), computer simulations have shown that only in the very long run do these probabilty distributions hold true. People who have bet against the above proposition have often gone broke.
The next step in this discussion involves posing philosophical questions such as:
given the randomness of chance or luck(or its capricious nature), how can we deal with it in our lives? No, I’m not going to suggest joining the Jehova’s Witnesses. Successful punters will have had to tackle this thorny question to make any progress in their quest for the elusive edge.
I don’t have any pat answer to this, but thinkers from Socrates to Berlin have mused over similar questions and generally concluded that knowlege is the key while ignorance might be blissful for most people who don’t even want to think about it.
Know yourself, know the game, know statistics. With all that knowledge you might just have the ghost of a chance of getting that edge.
Or, don’t bother about the edge. Don’t even think about it. Just enjoy the day with all its surprises, then go to bed and forget about it.Good post Artemis ..
I don’t think you can know if you have an edge. But what I think you could do is make an estimate on what degree randomness actually plays in gambling models. I’ve not quite worked out how to do that yet .. I wont sleep until it’s done.
If we agree on the impossibility of calculating the true value of an edge it leads to several real issues for horse racing betting.
Successful Kelly betting relies on knowing a precise edge for each and every bet. We cannot find the true edge for even a single example of a 3/1 bet.
This leads to fudges such as taking half Kelly to cover up for that ignorance factor. Is half Kelly OK or should it be 1/4 Kelly – we won’t know until the end of a sequence of bets – too late.We looked at how even with a 100 bets at 3/1 the number of winners was far more random than common sense expected. If you look at 100 bets for each of the prices a system might be used for eg 1/1, 5/4, 13/8 etc etc, say 50 prices then that means at least 5000 races to test. To be more certain of any profitability it might mean 50,000 races, if you can find them. That means going back years, when the racing factors, market and SP calculation was quite different. So you never actually arrive – systemwise.
Artemis pointed out the benefit that value betting brings to solving the randomness ("luck") conundrum.
So sticking with getting 100 bets at a price of 3/1. That means 25 winners are needed to break even.
Say you could calculate absolutely true prices for each and every bet then if you could back the following true price chances but get paid 3/1 for each of them you have the following chances of making any profit ie get more than 25 winners:Your true price Chance of profit (paid at 3/1 each winner)
6/4 99.89%
2/1 94.71%
5/2 74.88%
3/1 44.65%
(7/2) 21.26%
(4/1) 8.75%So even if you can do that feat, you are still not certain of making a profit but you have tamed randomness significantly if you get paid 3/1 on about 5/2, or shorter, true chances.
Getting closer to reality if some times your 5/2 true chance is actually a 7/2 true chance, but you still get paid 3/1, then your average chance of a profit decreases to only 48.07%. You are slightly more probable to lose than to win.
Regularly accepting 3/1 bets about true 4/1 chances will ensure you lose about 91.25% of the time.
So the moral is that, apart from selecting a viable potential winning horse, an equal amount of effort should be put into estimating true prices and even then forgoing backing them unless you can get at least one half a point plus half a point for "luck" advantage.
My problem with trying to find an edge in terms of estimating the true probability of a horse winning/performing is the number of races that is required for that edge to shine through. Is it 1,000, 10,000, or even 100,000? How many times will a coin have to be tossed for the split to be 50:50, or even 55:45? Yet if you were to get 11/10 about the coin toss, you would consider that an edge as such, but even those odds aren’t enough to compensate for the randomness factor.
- You must be logged in to reply to this topic.