I have a problem with "prediction markets" as news view. They just aren't informative enough.

If the price of oil goes down is that due to: A reduction in demand. an increase in supply, a large amount of investors finding a better investment or a large amount of investors wanting cash (due to having to pay creditors/taxes).

I want them to tell me enough information so that I can begin trading in an informed manner. When you see the market expects rain fall on montana is 2 cm in a day, what information is this based upon? If you read about a newly created huge man made lake in the are which you expect to change the micro climate, how do you know whether the simulations people and betting using are running take this into consideration?

If I don't get this information I can't trade with expectation of being able to make a profit and the market doesn't get any information that I may have that it doesn't. I could try and reverse engineer peoples climate models from the way they trade, but that is pretty hard to do. So I would like to try and lower the barrier of entry to the market by giving more information to potential players.

This lack of information is due to the signals each trader sends to the market, they are binary in nature buy or sell, with the traders strategy and information she used a black box that we can't get into. As potential traders we don't know if the market is taking certain information into account with the price we are shown.

The only way I have thought of being able to get at some of the information enclosed in the black box, is to only allow programs to bid in a market. The programs would be run on the markets server  and have acces to news sources, financial data and government reports. The programs could be stopped at any time by the trader but otherwise not communicated with or updated. So the would be trader would have to build models of the world and ways of processing the financial data.

How then would we get at the information? The source for each program that wanted to trade would be held in escrow and it would be released as open source code when it was withdrawn from the system. This would create a time lag, where people could profit from a better predicting system, but also allow everyone else to catch up after a certain time. There would also be some significant financial cost to starting a new bot so that people would not just start lots of one shot bots that did a single action based on their hidden knowledge.

It would have the side affect of have different biases to the human market. possibly more accurate as people would have to build the bias into the program and it would be visible and open to exploitation if it became prevalent and known.

I don't see this as a reality any time soon, but I would be interested if anyone else had any, more easily implementable, ideas about improving the informativeness of the markets.


19 comments, sorted by Click to highlight new comments since: Today at 6:52 PM
New Comment

As stated, it isn't clear why you can't write a "program" that just does one trade and never does another one. So you need a clearer proposal.

As I said there would be costs associated with getting a new program added that would make this prohibitive for large scale traders.

If you tax programs, where does that revenue go, so that traders can expect to gain on average?

What I am thinking is of is a scheme like the following.

Allow one free change of program per time period, say a month. Extra program changes would require a small administration fee. There wouldn't be any fee for the programs to trade.

The exact details of this would depend on the market.

This is an interesting idea, but I'm unsure that it solves more problems than it creates and I don't think its really fair to frame this as a problem with the prediction markets as news view.

If using a programmatic prediction market requires one to share private knowledge about the world which continues to be valuable, then people will refuse to use it, and will try to profit on the information some other way which does not force them to share their private knowledge.

Furthermore, its not as if this is the only way to answer the type of question you ask - "why did the price of oil futures fall?"

Intriguing idea, whpearson.

The biggest hurdle to the adoption of such a system is probably the fact that most current traders probably do not have enough programming skill to trade in such a system without incurring significant costs by starting lots of bots and probably do not have enough programming skill to extract more than a small fraction of the information revealed by such a system. One way to get over that hurdle is to target your new market at programmers and allied occupations (like project managers). Programmers and allied occupations could use an effective prediction market to help them, e.g., predict completion dates of their programming projects. I have written more on how prediction markets might serve programmers and allied occupations.

It would be much better of course if there was some math behind your assertion that a programmatic market is more informative than current market designs.

Whether the market is more informative depends on the type of information people put in the programs. The price may not be better, but the price on its own is not useful to me, if I want to be a trader and put in my information. See my updated intro.

I liked it better before.

Markets provide information, even if it isn't the explanatory information that you seem to be looking for. In the standard case, the information is about availability and cost of resources. You don't get the information behind the predictions about availability, but the predictions themselves have real value.

If you want to find a way to add value, it's important to understand what the existing value is so you don't break the underlying mechanism in your attempt to enhance it.

Perhaps you could expand upon this reply as a top level post, on how to use the information from markets in decision making.

The programs would be run on the markets server and have acces to news sources, financial data and government reports.

It seems as though encoding all of Earth's news is a task more suited to humanity. Perhaps I am being somewhat naive, but how is building a news database that is program readable more efficient than skipping the step of creating a program? As best as I can figure, the advantage is that the markets would all get the same info at the same time.

In terms of making money, I would expect such a model to begin competing on manipulating the news feeds. I suppose this already happens to a certain degree? The other trickiness I can see is selling a program to a million people and then creating your own that explicitly out maneuvers the masses.

It is an interesting concept, though.

The reason you can't profit on your possibly newly aquired knowledge of the local lake is that you don't know whether the other traders did or did not already include said lake in their calculations.

However, long term traders know which information they did factor in and would probably be interested in buying the information from you, iff it is new information to them. So, they should in theory be willing to pay you a price for your piece of information, even if you don't have the global view of the market.

And I think there is a way for you to profit, given that the transactions on the market are essentially cost-less. Namely, bet on whatever your information predicts, totally irrespective of what the market looks like, then publish your information to all traders. Wait for the market to settle at the new value (now guaranteed to include you info), zero out your positions.

If your information was already known to all traders, well, you have to zero your positions without any gain, and if transaction costs are zero, also without loss. (Random market fluctuations should even out).

Offtopic: How did you format the text as strike-through? (I tried everything I could find in the Markdown doc, but apparently LW doesn't support inline HTML - at least in the comments.)

Post markdown is different to comment markdown I'm afraid, I just used the UI controls.

HTML is allowed in posts but not comments, it would seem.

Your example is the current futures market for oil. That market is great if all you care about is the price of oil. The phrase "prediction markets" is supposed to evoke a much broader class of markets. Ideally, if you care (enough) about some question, you create (ie, subsidize) a market specifically to answer that question. For example, weather futures (which do exist, at least as insurance) tell farmers that the high future price of their crop is due not to increased demand, but to difficulty in growing it, and thus they should not bid it down.

Isn't that close to how today's stock and currency models work? The people in financial engineering who design advanced financial products seem close to what you're describing.

(Those aren't usually describes in terms of programs, but that's pretty much what they are)

[-][anonymous]13y -1

Yes, but the way it is now, traders are not required to reveal the source code of their trading programs.

I agree that this would generate better predictions than markets as currently envisioned, but the cost of allocating skilled programmers to this sort of task drastically exceeds the value of the predictions generated, even if those predictions are perfect.

This suggests an interesting area of research, at least: Make prediction-programming as accessible as possible.

New to LessWrong?