The Wisdom of Crowds

In his seminal work, The Wisdom of Crowds, financial journalist James Surowiecki notes that large numbers of people can often provide better or more reliable information in the aggregate than any member of the crowd can as an individual. The canonical example is that if you ask 50 people to guess the number of marbles in a jar, and provide them with an incentive to be accurate, the average of the guesses is likely to be closer to the correct answer than most of the individual guesses.

So, I am wondering, if we had enough people voting on where they think the S&P will close on any one day and provide them with an incentive to guess accurately, do you think we'll get an accurate collective guess? How do we incentivize the traders on this forum to give it their best guess? What could we give traders or do for them?

This portends as 'Social Choice Theory'. Some argue and doubt that mental states can be measured. At best have partial comparability. So, the sub-optimal social choices of accurate collective guesses could be malleable because of indivdual mental states.

gio

I've been mulling over this idea and think that I've had some very clever thoughts on how to put this into practice. Let me continue to mull over the weekend...

What a great potential idea. Great thought. Count this guy in for the vote!

I've just finished "reading" The Wisdom of Crowds. I say "reading" because I got the audio version and listened to it while exercising.

I recommended that you read or listen to it. The last disc (of 8) went into considerable detail of Long Term Capital Management (LTCM) and how their long/short strategy failed.

Now that I've finished it I can try and get this idea incorporated into the web and see if the theory has any merit in real time trading.

Here is the idea so far. A trader gets to vote on where they think a symbol is going to close for the next trading session. I've worked out a good (I think) incentive to get them to vote as accurately as possible. Once a trader has voted, he/she can see the average of all the votes and can return and see the new average at any time.

In order to make this simple let's just think in terms of one market (the Emini S&P 500) right now.

Restrictions that I'm currently contemplating:

• After the trader has voted, he/she can't change their vote.


• At a certain time voting is closed. I'm thinking the open of the market at 9:30am ET.

One of the problems that I envisaged was that a lot of traders will wait until they have the most information that they can get before voting - which is wise and logical. I thought that this delay in the voting would be a problem but now I see that it won't because the information that the voters will get will be usable throughout the day.

Thoughts...

What you are describing sounds similar to a binary contract (future or option).

Closing the voting at 9:30am might be problematic for days when economic numbers are released at 10:00am, or on FOMC announcement days. To obtain more pure predictions (before the fact) you may actually need to close voting at 8:20am to get ahead of the 8:30am economic reports.

For example, If I can wait until after the market reaction to a key 8:30am economic report, knowing there are no other reports due out that day, then I can base my 9:30am prediction on a great deal more market information than if I had to make the prediction at 8:20am (before the economic report).

So what you're saying is that the voting should close early enough to exclude the knowledge of economic reports?

I'm hoping that the predicted closing figure is going to be a usable figure by all voters that will be consistent enough to trade profitably. It's not really a competition among traders except for the incentive that I have devised.

By 9:30am (say) the voting is closed and all voters can see the average prediction. According to the research that James Surowiecki did, that average prediction should beat any single expert trader's prediction more consistently. The advantage for us, the voters, is that it should give us another trade that will, over time, be profitable.

What I'm envisaging is something like a 60% win/loss ratio using the same size stop as target.

What I was trying to say in that last post (which I don't think came across) is that it doesn't matter if traders have extra information. What we are looking at is an above average prediction at what the close for the market will be at 4:00pm given our collective knowledge and different approaches to the market. If we average our (incentivized) opinions, then the Wisdom of Crowds theory states that we should do better than using a random guess.

this is a lot like the folks that use math models, cycles, patterns, ect... to try to predict the trading day or the general shape of the trading day, like trying to predict the time or price of the day's high or low...ect

the difference is we are discussing using the collection of traders predictions as our model rather than a collection of math equations.

modelers have found it helps to make an initial prediction then re-run the model as new data becomes available and see if you can detect a trend or "lock" where the model keeps predicting a similar outcome, or just random noise (where each new run is completely different from all the others). This concept fits into my point about the econ. reports affecting the prediction. If you poll folks before and after an econ report, their prediction for the days outcome may be very different based on how they perceive the market's reaction to that new data.

Anyways, my guess is what you will find is folks will be accurate, but in a contrary way... saying it another way, the crowd will tend to be consistently wrong...

quote:

---

Originally posted by pt_emini
Anyways, my guess is what you will find is folks will be accurate, but in a contrary way... saying it another way, the crowd will tend to be consistently wrong...

---

- that's a distinct possibility.

What I'm thinking is comparing the crowd's guess against a randomly generated number in the allowable range.

Today's close for the ES was 1371.50. So I'm thinking of allowing guesses in a 30 point range from the previous day's close. There will be the odd day when there's a big market move and this range isn't enough but that will be the exception.

What would actually be a better plan (thinking aloud now) is to allow a 25 point range on the current price of the ES (at the time the vote is taken).

Anyway, over time I would then measure the crowd's result against the actual result and against a randomly guessed number in that range. That should, at the very least let me us know if there's any value in the crowd's guess.

For those of you following this thread/topic we have now implemented this feature. The help page describes the feature best here:
http://www.mypivots.com/forecast/help

Here is the page that you can submit your forecast on:
http://www.mypivots.com/forecast

And here is a topic dedicated to the forecasting feature:
http://www.mypivots.com/board/topic/6288/1/crowd-forecasting-discussion