We have often discussed the possible benefits of prediction markets for reducing bias. One key element that could be particularly useful is a conditional prediction market, where betting claims are based on outcomes conditional on some factor we want to evaluate. Robin recently mentioned the possible benefit from market claims on longevity or health conditional on various interventions.

Unfortunately, there are few prediction markets in the world, and fewer conditional ones, so it is hard to know how practical this institution may become. Thanks to an initiative by OB contributor Peter McCluskey, Intrade – the large, real-money prediction market – has added conditional claims based on the outcome of the U.S. Presidential elections. These may represent the first major case study of real-money conditional futures markets. So how are they doing? Here is some data and analysis.

Because of how the conditional claims are computed, we need to normalize the results by dividing the claim prices by the probability of a win for that party’s candidate. At the moment I looked (2008-07-28, 5:00 PM PDT), clicking on Politics and “2008 US Election”, and expanding “2008 Presidential Election Winner (Political Party)” I saw that the odds for a Democratic victory are 66.7-67.4. The claims are based on Democratic vs non-Democratic victories, so we will use the complementary odds of 32.6-33.3 for a non-Democratic win.

Then clicking on the left on “US Pres. Decisions” gives us the six conditional markets funded by Peter McCluskey.

Here are the claims, the prices, and the normalized values found by dividing by the corresponding Democratic and non-Democratic victory probabilities above. To compute the normalized values, I divided the lowest claim price by the highest party-victory price, and the highest claim price by the lowest party-victory price, giving the maximum range consistent with current trading prices. (Peter provides continually-updated information on current implied values as well.)

 Claim Dem price non-Dem price Dem norm non-Dem norm Increase in US government debt (over \$10 billion) 51.2-53.7 34.4-36.5 76.0-80.5 105.5-109.6 Number of US troops in Iraq on 30 June 2010 (over 2000) 41.3-43.8 32.1-34.3 61.3-65.7 96.4-105.2

These prices imply that the market expects that under a non-Democratic administration, we will see substantially higher government debt, as well as much higher numbers of soldiers in Iraq.

In addition, there are two claims based on the chance that futures prices will move in the same direction as the Democratic party victory claim price, on election day:

 Claim Odds Oil Futures & PRESIDENT.DEM2008 prices will move in same direction on Election Day 49.8-49.9 T-Bond interest rate futures & PRESIDENT.DEM2008 prices will move in same direction on Election Day 49.9-50.1

These values are so close to the 50% mark that it appears that the markets do not expect any significant movement in oil prices or interest rates on election day, that can be attributed to developing information about which party will win. As critics have noted, this could be because they don’t see much effect of political parties on these values, or else because they expect that the election day results will be a foregone conclusion and there will be no surprises in that regard. (Or perhaps, that no significant information will leak out regarding election results during trading hours.)

Intrade has another set of conditional election claims under the heading of “Impact of Next Pres.”. Rather than party based, these claims are linked to specific candidates, meaning that at this point only the Obama and McCain claims have significant relevance. The claims measure economic growth, unemployment, crime, and control of the legislature, conditional on candidate victory – all interesting topics. But there is a huge difference in liquidity and trading volume for these claims versus the ones discussed above. These claims have bid-ask spreads that range from large to enormous; some have no bid or ask offers at all. And the trading volume is so far non-existent – apparently none of them has ever traded! These claims are essentially useless for shedding light on likely impacts of the presidential choice.

So what is the difference? The first set of claims is being funded and subsidized by Peter McCluskey, who has set up an automated market maker algorithm (based on an invention by Robin Hanson) that keeps the bid-ask spread within 2.5 points. When I first heard of the idea of such a bot, I thought it would have to be willing to lose infinite amounts of money, but actually this is not the case; losses are bounded. Nevertheless the potential costs to Peter are very significant, and he could be out almost \$50,000 to fund his six claims.

Still, we can see the very real and practical benefits of this subsidy approach. The claims sponsored by Peter have relatively good liquidity and produce meaningful price comparisons. While not intrinsically more interesting than the other batch, Peter’s claims trade relatively frequently since there is always an attractive offer, while the others have never traded at all.

The lesson, then, is that conditional prediction markets can be informative, but that there is great importance in the detailed structure of the institutions. Peter’s subsidies have made the difference between a failed market and a successful one. At the same time, we see that the claims based on price movements on election day are not working as well as the others, so that idea may not be as promising (perhaps circumstances will change as election day approaches).

Another issue I would see as a potential trader is the difficulty of protecting my position against shifts in prices of two different markets. The value I am betting on is the ratio of two sets of prices, so I may want to change my holdings if either of those prices moves. It would be nice if one could post bids which are based on the ratio rather than on specific prices in the individual markets. It would also be nice to see the imputed values based on price ratios live and next to the market prices, rather than on another web page.

Even with these problems, it appears that the overall concept is a rather spectacular success that suggests that the concept has great promise. We all owe a debt of thanks to Peter for making such a substantial contribution towards showing the practicality of conditional futures markets.

GD Star Rating
Tagged as:

I agree. Great explanatory post. Reputation points should accrue to Peter.

• http://yudkowsky.net/ Eliezer Yudkowsky

Much appreciation to Peter.

• http://hanson.gmu.edu Robin Hanson

These may represent the first major case study of real-money conditional futures markets.

Correction: The Iowa Electronic Markets had markets that could be interpreted as conditional forecasts of election given nomination, advising parties on who to nominate for President. But these new markets of Peter’s arguably offer more socially valuable info, helping voters see for whom to vote.

• http://goodmorningeconomics.wordpress.com jsalvati

I cannot see any way that Intrade lets you know the market is subsidized. It seems like this should be useful information for traders since, since it lets them know that their expected return is positive. I too am extremely happy to see experimentation with prediction market subsidization.

• http://www.philosophyetc.net Richard

Are these bets really ‘conditional’, or rather conjunctive? That is, do you get your money back if the condition doesn’t obtain in the end (say I bet on non-dem.DebtIncr, but then a Democrat is elected)? A cursory glance at the contract rules made it look like you instead lose — “contract expires at 0” — but maybe I’ve misunderstood something…

• Gwern

‘Increase in US government debt (over \$10 billion)’

Wait, what? Is this per year or increases over current rates or something I am misunderstanding? Did you mean ‘\$10 trillion’? (That makes more sense to me.)

• http://silasx.blogspot.com Silas

Just a pleasant reminder: Back on the original OB thread where we discussed these conditional futures, I proposed an alternate conditional futures market: instead of betting specifically on whether e.g. oil futures and the Democrat contract move the same direction on election day, what we should do instead, is bet on the overall correlation over all the days up through the election, between *changes* in oil futures each day, and the *changes* in the Democrat contract each day.

So e.g. you would bet on whether oil’s price always goes up as the market deems Obama more likely to win, and vice versa.

This is obviously a superior metric in that it is robust against noise and/or manipulation that can happen on that one vital day. It inputs variation on many days, exploiting much more relevant information.

I explain this all and argue the point with Robin Hanson and Peter McCluskey in the thread linked above, but unfortunately, this is kind of like talking to a wall. They seem to spend very little effort reading my posts, resulting in responses that stem from a misunderstanding, rather than being serious criticisms. And, Peter McCluskey makes arguments against my metric that apply several times over to his own.

(Of course you’re not obligated to read my posts, but if you’re going to respond, don’t do a half-job.)

Does anyone here agree with my point that my metric is superior and would be a far better prediction market contract? Would I get the same appreciation and reputation points if I gathered this data and funded such a contract? (Oh, and can someone turn off the italics in the linked thread?)

• http://riskmarkets.blogspot.com/ Jason Ruspini

What I find most pleasing about this design is that it avoids the awkward construction that Richard alludes to, and at the same time spurs trade by virtue of the changing denominator. Excellent.

I think the actual subsidy is much smaller although it did seem that over \$1000 was given away at inception by setting the initial prices at 50.

• http://profile.typekey.com/bayesian/ Peter McCluskey

I can’t agree that the claims I’m subsidizing have been a success yet. Note that NONDEM.PRES-GOVT.DEBT is priced as if it could be expired above 100, but the rules say the maximum expiry price is 100. That’s a clear sign that there hasn’t been enough trading to make the price of that claim reasonable.
A few of the “Impact of Next Pres.” claims have traded. I bought and sold 2 contracts of PRES.McCAIN+LESS.CRIME in June. I’m not sure what time period the Intrade volume figures cover.
Gwern, “over \$10 billion” and “over 2000” mean that you divide debt changes by \$10 billion and divide troop levels by 2000 to get the corresponding contract price.
Silas, it’s too late to change the existing contracts, and I’m not about to subsidise more contracts for this election. I will watch carefully for possible manipulation, and will consider better ways of minimizing manipulation risks if I subsidise similar contracts in the future. I think the rest of your arguments have been adequately answered. I certainly encourage you to subsidise additional contracts.
Jason, two of the contracts will expire at zero when the election is settled. I will lose nearly \$8000 on those. For the others, it’s unpredictable.

• http://silasx.blogspot.com Silas

Peter: Silas, it’s too late to change the existing contracts, and I’m not about to subsidise more contracts for this election.

Were you able to understand how I wasn’t asking you to change the existing contracts or subsidize others, but rather, proposing to subsidize my own superior ones that have better safeguards than “I will watch carefully possible manipulation”?

Silas, if you improve my ability to understand the present and predict the future, the yes, I will definitely award you proportionate appreciation and respect.

• Doug S.

When I told my father the current price for McCain on Intrade, he wanted to bet on him, but his bank blocked the credit card transfer.

(My father predicts that Obama will win the popular vote but lose the electoral vote, guessing that he’ll sweep the blue states and lose enough red states, but barely.)

• michael vassar

Silas: Sorry to hear that.

Robin: That claim definitely demands some serious acknowledgment. It sure does sound like Silas proposed a better system for extracting info from markets.

• http://hanson.gmu.edu Robin Hanson

Silas, as I explained before, we expect that on election day most of the causal direction of the correlation is from the vote to the consequences, while for most days before there are large causal forces in both directions. Also, we want a market estimate as useful as possible right up until the election, but just before the election the overall correlation price would be dominated by the previous correlation history. (And do you really imagine I had not considered your obvious variation when I chose my design?)

• Ben

Props to Peter for taking on the risk in order to get the market going, and more props if he ends up making a profit. Anyone interested in setting up a prediction market based on whether Peter McCluskey comes out ahead?

• Michael Sullivan

I doubt Peter makes a profit on this batch. That would require a lot of trading with that small a spread.

But in principle you are right. There’s got to be some sweet spot of volume and good guesstimation of the initial line where this is profitable even with the fairly small bid/ask.

If a somewhat larger spread would still invite interest in the market, it should be fairly easy to make it profitable as long as transaction costs to the market maker are low (much smaller than the spread).

• http://silasx.blogspot.com Silas

Robin: Silas, as I explained before,

Where? Would you mind pointing me to the first comment you made in that discussion, in which you understood what I was proposing, and correctly stated what you meant? Because that was kind of an issue back then.

we expect that on election day most of the causal direction of the correlation is from the vote to the consequences, while for most days before there are large causal forces in both directions.

Well, you certainly *hope* so, but then, the market has usually largely incorporated expectations about the winner before election, making the signal really really small, and allowing the random influences to dominate.

Let me take a gander at Peter McCluskey’s shock futures market. Is it consistent with my theory that bidders will assume that other random forces will determine the election day correlation?

“49.9-50.1”

Yep, that’s betting on a coin flip alright!

Also, we want a market estimate as useful as possible right up until the election, but just before the election the overall correlation price would be dominated by the previous correlation history.

Yes, so in the unlikely event that the correlation of “probability of democratic win” and “oil futures” is very far from the unknown noise-filtered, manipulation-filtered election day correlation, this metric will have inaccurate results.

If you really think the election day’s correlation is that much more important, then you can say:

“Silas, that’s a really good idea, but to account for election day’s more relevant information, the metric should treat election day as 30 days of price history for purposes of determining the correlation that resolves the bet.”

Then, you give stronger weight to election day, but still incorporate all the available information and dilute irrelevant influences!

(And do you really imagine I had not considered your obvious variation when I chose my design?)

I invite anyone to read the thread and decide if Robin’s responses indicate that he had considered the idea before.

P(Robin responds to misunderstandings of my idea | Robin has considered my idea before) = low
P(Robin responds to misunderstandings of my idea | Robin has not considered my idea before) = high

• Matt Huang

Being argumentative for the sake of argumentation isn’t productive.

• anonymous

Biases?

“Dem price non-Dem price Dem norm non-Dem norm”

Why not Democrat and Republican? Those are the two major parties. What about third
parties?

“Democratic” is the system of Democracy. Democrat is a political party in America. Although I’m sure the DNC loves how people conflate the two.

• http://www.allancrossman.com Allan Crossman

Why not Democrat and Republican? Those are the two major parties. What about third parties?

I suspect the intention was to make the probabilities (“Dem presidency” and “non-Dem presidency”) add up to be exactly the same as the probability of any kind of presidency.

“Democratic” is the system of Democracy. Democrat is a political party in America. Although I’m sure the DNC loves how people conflate the two.

I don’t think it’s a conflation; it’s simply a word with more than one meaning. The word Republican is the same: it also has a meaning distinct from the name of a party.

• Doug S.

Now, if only Intrade worked in the United States without requiring a bank transfer…

• http://silasx.blogspot.com Silas

Blogged and flogged.

I’ll miss my posting privileges here. I really will š

• http://www.midasoracle.org/2008/08/05/subsidizing-conditional-prediction-markets/ Midas Oracle .ORG

Subsidizing real-money prediction markets and real-money conditional prediction markets = BULLSHIT IDEA