Comments on: Presidential Decision Markets

By: Robin Hanson’s conditional prediction markets | Midas Oracle .ORG

Robin Hanson’s conditional prediction markets | Midas Oracle .ORG — Tue, 11 Aug 2009 08:38:43 +0000

[...] prediction markets Written by Chris F. Masse on March 4, 2008 — Leave a Comment His first group of conditional prediction markets are traded quite well (although thinely), but his second group is a thermo-nuclear disaster. [...]

By: Silas

Silas — Mon, 07 Jan 2008 18:24:55 +0000

Peter_McCluskey: The trader doesn’t have to fundamentally shift the market. He just has to make one trade show up right before the cutoff point for your measure, thus making “the” price where he wants it to be.

I’m not suggesting that voters are sensitive to daily price changes, only that they’re sensitive to changes over months in prices they pay at the pump, and traders may change their predictions about that effect as fast as oil futures change.

Okay, point made. But then, the reverse correlation problem could show up on election day too. The oil prices change voters’ minds as going to polls, which changes odds, which affect futures on financial markets, which affect oil prices …

By: Peter McCluskey

Peter McCluskey — Mon, 07 Jan 2008 02:49:06 +0000

Silas, the December 2011 oil futures are currently traded heavily enough that a trader could not reasonably expect to manipulate the results of the shock response future contract without trading oil futures worth millions of dollars.
I’m not suggesting that voters are sensitive to daily price changes, only that they’re sensitive to changes over months in prices they pay at the pump, and traders may change their predictions about that effect as fast as oil futures change.

By: Silas

Silas — Sun, 06 Jan 2008 20:06:32 +0000

Peter_McCluskey: Can you describe a more plausible way in which shock response futures would reflect (a week before the election) a causality other than election results causing change in oil prices?

I can imagine on election day, an errant trader might eat a 2% trading loss ($2000 on a $100,000 trade) just to make it look like “Democrats cause higher oil prices”, or because he stands to win more than $2000 on shock futures that he’s bought. (Just like someone took a loss with $100 oil because of other goals.) People buying futures, then, have to incorporate the risk of this, as well as the risk of anything with a larger magnitude than the election results and opposite in sign.

I think it’s a good point that my metric might capture causality in the opposite direction, i.e. people change their votes (which changes the odds) because futures prices change. However, traders more closely monitor conditions that could affect their portfolios (because of the financial interest) than voters monitor conditions that could affect their votes, so traders are much more sensitive to small changes in election odds, than voters are sensitive to small changes in futures prices. Thus, a correlation is more likely to indicate the odds –> futures-prices causation direction.

By: Peter McCluskey

Peter McCluskey — Sun, 06 Jan 2008 17:00:53 +0000

I don’t want to establish these particular indicators as standard measures for voters to use. But mnuez’s suggestions that people use a larger number of more special purpose indicators would be a step in the wrong direction, as the more indicators that people attempt to use, the easier it will be to emphasize those that support one’s biases.
I want future decision markets to try to use a few broader measures of wellbeing that capture a larger portion of what it is that people want to maximize.
One such measure that I have some desire to use is life expectancy.
One reason I haven’t used that is that I suspect fewer people have useful knowledge of how politicians affect life expectancy, so it’s harder to attract informed traders with the kind of subsidies I’m currently willing to provide.
Another reason is that the effects of politicians on life expectancy will take longer to become measurable, and Intrade’s business model does not encourage them to design their system to make long-term forecasts work well.

Silas, I can imagine that the correlations you want are caused by, say, increased oil prices causing voters to prefer a Democrat.
I could imagine that we see something similar with the benefit of hindsight for the shock response future contract (i.e. a sudden surge in oil prices starting on Monday afternoon of election week influencing the election result). But in order for that to affect the shock response future price weeks before the election, the shock response future price would need to incorporate an expected last minute oil price surge better than oil futures prices do. I find that far-fetched. Can you describe a more plausible way in which shock response futures would reflect (a week before the election) a causality other than election results causing change in oil prices?

By: Silas

Silas — Sat, 05 Jan 2008 20:44:57 +0000

Robin_Hanson: Neither would an election-day-only correlation, as Peter_McCluskey’s is. Was your point that the months-long, day-to-day correlation is necessarily unrelated to the causation direction, while the election-day-only correlation is strongly related to the causation direction? Or was it a more general point about “correlation != causation”?

If the former, I disagree. Any countervailing event on election day (like the errant $100 oil trader) larger than the magnitude of the candidate’s effect, flips the sign. In contrast, on my metric, those events must also operate for n days, consistently in the same direction as changes in the candidate’s changes. That why I bring up that my metric is more statistically significant.

And any chance of taming the italic tag?

By: Robin Hanson

Robin Hanson — Sat, 05 Jan 2008 15:02:33 +0000

Silas, a consistent correlation would not show the direction of causation.

By: Silas

Silas — Sat, 05 Jan 2008 14:57:13 +0000

Robin_Hanson: That would be a valid point if I were proposing simply comparing the initial and final values. But my method involves comparing the correlation over a succession of days. To win on my metric, the economic variable would have consistently, over a longer time, move in the same direction as the candidate’s chances.

In any case, financial markets on election day close before the results are in, or close to certain.

By: Robin Hanson

Robin Hanson — Sat, 05 Jan 2008 13:16:17 +0000

Silas, I mispoke. The point is to disentangle the direction of causation. On election day most of the causation goes from election to other parameters. Over the preceding months it is much less clear.

By: Aaron Brown

Aaron Brown — Sat, 05 Jan 2008 02:05:16 +0000

“effect oil prices” –> “affect oil prices”