The Right Coast

November 06, 2004
Final word on tradesports
By Tom Smith

Yes, tradesports and the other bookies had W down at around 30 percent for most of election day. Does this suggest markets aren't so efficient or rational after all, or at least betting markets?

Not really. The betting markets were reacting to the exit polls, just like everyone else was. The exit polls were way off, further off, I strongly suspect, than anyone had reason to think they would be. Why were they so far off?

Both Dems and the GOP made a huge effort to get out the vote, including getting out many new voters. The GOP effort in particular relied on an extraordinary, risky, and unprecedented effort to use informal networks of contacts to get out the rural vote. Republicans called friends, relatives and associates and reminded them to get out the vote. In my republican part of San Diego county, "Be sure to vote!" became the temporary replacement of "Have a nice day!"

Exit pollers have to make assumptions about turnout to allocate their boots on the ground. They apparently allocated their interviewers, rationally enough, in some way that did not anticipate the big increase in rural voters. Thus they oversampled suburban and urban voters, catching the big Dem turnout, but not the even bigger GOP turnout.

The market isn't magic. It only works with the information it has. It looked for a while like a Democratic rout because the information wasn't available that across Red Amuricah, pappy had fired up the truck and taken all the kin into vote. If exit pollers had more samplers in rural areas, or if we had a media that actually reported the news, we and the market would have been spared a panic. When did the market fix itself? As soon as real numbers started coming in in Florida. When the real numbers started coming in, tradesports showed Florida, Ohio, Iowa, Nevada and New Mexico going to W well before the networks called any of them. Good information drove out bad.

So, I think the betting markets did fine this election day. I think what the selling off of W contracts before the polls closed showed was simply that the turnout the GOP organized across the country was an utter surprize to all but a few GOP insiders, such as Herr Professor Rove.

I have been told that the Realclearpolitics moving average of state polls called every single state correctly, perhaps because aggregating poll numbers has the effect of increasing sample size. Also, if you select randomly from a database that includes rural voters, laws of probability should assure they get adequately represented in your sample. So the likely voter screens ended up working well enough so that exit pollers had no particular advantage just because they were talking to real voters. The fact that the polls did as well as they did does not mean anybody knew there would be such a big rural turn out. Moreover, polls must have been using some likely voter screen beyond, "did you vote in the last election" to do as well as they did.

BTW the only pollster I know who foresaw the Dem's problem was Pat Cadell, the rotund, very sad looking fellow seen rarely these days on TV. He said months ago he was getting numbers like 80 percent opposed to gay marriage, even in traditionally Democratic rural areas. He said the Dems were in deep trouble if they stuck to that issue, and it looks like he was right.