I don’t follow Iraq closely, but recent headlines seem to indicate a turn for the worse and that withdrawal of U.S. troops is now on the table.

It should not have been difficult to predict that invasion would turn out badly, but politicians make the same mistakes (less charitably–tell the same lies) repeatedly, in particular when it comes to war (one reason why).

Among all the tragedies of the Iraq war, a small one is that there was no set of conditional prediction markets to consensus check (an analogue of “fact check”?) likely outcomes. An arbitrary expert can always be countered with another arbitrary expert. The nice thing about prediction markets here is that they converge to a single consensus probability (or set of interlinked probabilities for a set of claims) given the possibility of arbitrage. Faced with a market that says what a politician wants to do will probably have ill effects, the politician can ignore the consensus, but can’t counter it will an equivalent, as can be done with any expert.

So should the U.S. withdraw its military from Iraq? Unfortunately I do not know of a conditional market set up to guess the impact. Iraq-related markets I found:

• U.S. military deaths in Iraq: last trade indicates the first month with no U.S. military deaths from hostile action in Iraq will be May 2008. This would seem to indicate an expectation of near complete withdrawal by then, or something magical.
• There will be fewer than 100,000 US troops in Iraq in April 2007 last traded at 24, or about a 1/4 chance that at least a serious troop level reduction will start soon (from 144,000 in October).
• Announcement of US Troop Withdrawal from Iraq, last trade indicates a 10 percent chance of such an announcement before the end of 2006.

Unfortunately all of these are play money markets and all only concern U.S. troops. What about Iraqi civil war or economic performance? Fortunately we can use one of these markets as an input for a conditional market that attempts to guess the impact of withdrawal on Iraq. I used the second, as it maps directly to a probability, unlike the first, and is not deemed to be an incredibly long shot, unlike the third.

The Iraqi Body Count currently says a lower bound of 47,781 Iraqi civilians have been killed since the invasion. I assume if that lower bound moves to 100,000 or greater by the end of 2007, a civil war has occurred or is in progress.

So I set up Iraq withdrawal and civilian casualties on Inkling, with four stocks:

• USLEAV07 true AND >= 100k IBC EOY 2007
• USLEAV07 true AND < 100k IBC EOY 2007
• USLEAV07 false AND >= 100k IBC EOY 2007
• USLEAV07 false AND < 100k IBC EOY 2007

I set the intial price of the first two at 12 each and the second two at 38 each, reflecting the 24 percent chance of substantial troop reduction given by Newsfutures traders and a 50/50 chance of civil war (I don’t know of a probability source for the latter). In theory prices should move to whatever traders think the probabilities actually are regardless of their initial settings.

There are two major problems with this experiment. First, a spike in violence may make troop reductions more (or less) likely, which makes it harder to divine the impact of troop reductions on violence.

Second, Inkling markets are sometimes at great variance with others or common sense, e.g., Hilary Clinton is given a 28 pecent chance of winning the 2008 Democratic nomination, others have her around 50 percent.

I surmise that there is something wrong with Inkling. That something could be just that it has no users. I set up this experiment on Inkling because it was trivial to do so, but I’d really like to see Tradesports/Intrade set up real money contracts along these lines.

Update: The first problem can be removed by ignoring deaths through April 2007. I will create a new market reflecting this…

Iraq withdrawal and civilian casualties (improved) is running with the following stocks:

• USLEAV07 true AND IBC >= 40k May-Dec07
• USLEAV07 true AND IBC < 40k May-Dec07
• USLEAV07 false AND IBC >= 40k May-Dec07
• USLEAV07 false AND IBC < 40k May-Dec07

Update 20061127: The improved market is now actually running, was previously held for admin approval.

Update 20061211: Followup posted at Midas Oracle.