Stupidity, as defined by Prof Panos…
[...] It is largely stupid to call “successful†a continuous market that gets the average right but has a very high spread over the outcomes. Similarly, it is stupid to call a success a discrete market that is “correct†(i.e., the frontrunner being the actual outcome) when the contract price of the frontrunner is 15% or so.
Panos and I are in complete agreement. Regardless of whether the market is trying to predict a discrete outcome or a continuous variable, if the dispersion of the market “guesses” is large, this means there is great uncertainty. When such markets “get it right”, it is mostly by luck. We need to focus on getting prediction markets to have tighter dispersions (all markets). If it is not possible for certain types of questions, we shouldn’t be wasting any time on them. These are simply not useful and are likely to result in bad decision-making.
Earlier comments of mine have tried to make this same point. However, there are differences between discrete and continuous outcomes. With a continuous variable (e.g. quarterly sales), when the market predicts close to the actual outcome, this is often “good enough”. However, in discrete markets (e.g. horse race), being slightly wrong is completely inaccurate.
With frequent similar events like horse races, we approach ‘good enough’ for decision making. Of course interesting questions will tend to not be of that type.
Well Paul, you added a “nice” touch to that one. Are you suggesting that market volatility will lead to inaccurate predictions? I suppose you can account for all the micro-events, consciously, that influence the outcome?
>> We need to focus on getting prediction markets to have tighter dispersions (all markets).
No, you don’t.
Medemi, I think there is a misunderstanding here. I’m *not* saying that market volatility leads to inaccurate predictions. The market can be as volatile as it likes, until some sort of equilibrium (if there is one) is reached. Then, we take a snapshot of the predictions (weighted by dollars) to arrive at a distribution around the mean. I’m saying that if that distribution is too loose, the prediction will likely be very weak and even more likely to fail. If this is the case, it cannot be relied upon for decision-making purposes.
I’m also *not* saying at anyone can account for micro-events, consciously, either. The best that can be hoped is that the market participants (as a whole) will account for most of the micro events that are likely to occur, based on existing information in the market (completeness). Even if the market were *perfectly* informationally efficient, that would mean that the future changes in market prices would be perfectly random, and hence, unpredictable.
I don’t understand why you don’t agree that dispersions *must* be tighter. Can you explain how prediction markets might be useful for decision-making, where the dispersions are not tight?
Paul
Paul,
it is very common for highly liquid highly efficient markets to have high levels of dispersion. I’m talking about markets with a minimal spread and tons of money on them on both sides. There are a number of reasons for this, which I won’t get into.
Now, when a prediction market reflects an equilibrium for a sustained period of time this should worry you! It doesn’t always mean traders are in some sort of “agreement”, it probably means traders haven’t got a clue as to whether the market is representing fair value… Usually when volume (suddenly) increases the market will start moving again.
Sorry, but “your” obvious model doesn’t work, IMO. It simply reflects a human tendency to explain a complicated world in understandable and simple terms.
>> Can you explain how prediction markets might be useful for decision-making, where the dispersions are not tight?
It’s a simple matter of statistics, that’s all there is to it. Yet very powerful.
Let’s consider an example. Let’s say AC Milan is having a bad year and is in 10th position after 15 matches. The price before kick-off can in theory be all over the place, but when you study 200 of those cases you’ll have to admit that there is no chance in hell you will be able to make any money by betting for or against AC Milan. The market on average reflects the true probabilty of AC Milan winning. It takes very mature markets to be able to do that, and betfair’s markets will do that even if we dig a lot deeper.
Wait a minute, Medemi. Most of what you said is not attributable to me!
You said: “it is very common for highly liquid highly efficient markets to have high levels of dispersion. I’m talking about markets with a minimal spread and tons of money on them on both sides. There are a number of reasons for this, which I won’t get into.”
I disagree that is very common. More likely, highly liquid, efficient markets have less dispersion. It is possible to have a wider dispersion, but these markets would not be very useful for predictive purposes. I wish you would get into this.
Again, “Now, when a prediction market reflects an equilibrium for a sustained period of time this should worry you!”
One, I am doubtful that such markets do reach an equilibrium (and how would we tell?). Two, if they did reach a sustained equilibrium, I *would* be worried! I never said anything different.
“It’s a simple matter of statistics, that’s all there is to it. Yet very powerful.”
But of course! Statistics helps describe the mean and dispersion of the distribution of predictions. My point is that the dispersion *must* be reasonably tight to be useful for decision-making purposes. I will admit that I am primarily concerned with winner-take-all, continuous variable markets, not discrete, binary ones.
I agree with your comments on Betfair’s markets. You will probably never make any money betting on (or against) AC Milan on Betfair. However, if Betfair’s market provides a better estimate of AC Milan’s expected win percentage than any other method, that prediction may be used on *other*, less informed, betting markets, to your advantage.
>> My point is that the dispersion *must* be reasonably tight to be useful for decision-making purposes.
Why then??? Is it a confidence issue on your part or do you have scientific proof? The only “problem” I see with high levels of dispersion is how you’re going to interpret the data but that is always a problem with statistics. It depends on the type of market to some extent and while I can see a correlation (possibly) I just don’t see a necessity.
>> However, if Betfair’s market provides a better estimate of AC Milan’s expected win percentage than any other method…
Of course it does, everyone knows this.
>> .., that prediction may be used on *other*, less informed, betting markets, to your advantage.
Sorry, not sure what you’re saying here. You mean like the bookmakers? They adjust their quotes using information from the betting exchanges, well, betfair. They’re not stupid.
>> I disagree that is very common. More likely, highly liquid, efficient markets have less dispersion….
Not true. I’ve collected data from 300 matches, real time, using my own software which would then present that data graphical. I looked at this because I wanted to see if I could detect certain patterns to gain an edge. The AC Milan quote (a very liquid market) could easily go from 1.14 to 1.20 thirty minutes before the off, which is a huge difference. What is causing this is debatable but usually it is the big money coming in. The price could fall back to 1.14 or stay at 1.20, you just don’t know in advance.
Likewise, a not so liquid market could easily show a flat pattern.
Medemi…
It sounds like you’re trying to find a trend in real time bets over time. The fact that you weren’t able to turn this into a profitable betting strategy helps support the EMH. Not that I’m convinced of the EMH, as I will discuss in a forthcoming blog.
Perhaps we’re confusing “dispersion” in these markets. Not having seen your data (would love to see it), but I have observed similar trends in betting on Hubdub markets, it appears that you are describing dispersion as the volatile price *changes* that occur the closer the markets get to the close. I think this reflects a market reduction in uncertainty (in some PMs), and in betting markets (like horse races, football matches, etc…), it reflects the insiders taking advantage of those that are uninformed, making the markets more accurate. However, this isn’t *the* dispersion of the market. Dispersion is not a trend in prices. It is the distribution of prices around the mean when the market closes.
If you are trying to predict future odds based on current trends, you are implicitly assuming that the market is *not* efficient. Perhaps you need a prediction market to predict the prediction market. But, then again, you would need a … I’m kidding, but I’m sure you get the point.
Medemi…
Regarding dispersion in market prices, for decision-making, they must be relatively tight to be useful. I understand you are discussing betting markets, but I am more concerned with EPMs. If the dispersion is relatively flat, there is too great a chance that the outcome will not be reasonably close to the mean (prediction value) of the market trades. Even if such markets are perfectly calibrated, they will be useless for decision-making. I don’t believe there is any other way of spinning this fact to make these types of markets useful.
I would like to see your complete data on AC Milan (if you’re willing to share). Alternatively, can you calculate the *final* distribution statistics for these markets?
Paul,
I still don’t know what the hell you’re talking about, but it seems to me you’re looking at very illiquid markets. I will let it rest.
My data, I still have it on my hard disk but I don’t think I can start my applications with Vista. It’s been a while and I don’t feel like going back to it. There is a service out there where you can buy data for $1 per match or so and look closely at the data. There’s a link from the betfair site somewhere, at least there was. I have no in-running data myself and was only able to sample the markets 60 times/minute.
This was always a hobby to me although the prospect of being able to make some money was certainly a key driver. I picked the soccer markets as a positional player because I heard they were so tough to beat. We all need a challenge once in a while.
Unfortunately the fun resided as time passed because
1. I didn’t agree with betfair’s integrity policy, if there is such a thing.
2. Commission is a killer, and I did not want to work for them
3. I could see it coming that one day I wouldn’t be able to place bets from holland. Or that betfair would screw me in some other way.
There are many people like me who turned into researchers, and this is a key benefit of betting exchanges. I’ve waited a long time for them to come on here and tell their story so I wouldn’t have to. Alas, no luck. Anyway, I don’t think my ramblings will hurt anyone as people can choose what to do with it.
Always nice talking to you.
Have a great Sunday.