Oct

31

Poll Numbers, from Andy Aiken

October 31, 2016 |

 "I have noticed that 538 are quite incompetent (and aggressively so)– they don't grasp something basic about an election probability as an estimator of a future binary outcome. The more uncertainty, the closer the estimator to 50%. But let us 'price' it as an arbitrage-free option."

I think that Taleb is correct here. The point estimate of an election probability is far more unstable than the 538 model portrays. Also, on the betting markets, the uncertainty should be reflected in wide bid-ask spreads, which is not the case in these markets either.

Stefan Jovanovich comments: 

The poll numbers are not trades that must be cleared; they are, at best, applied social science research.  There is no penalty for getting the estimate wrong; no one ever gets fired for having missed the spread.  When you all lay on a single trade, there is actual money at stake - far more serious money than anything these D List celebrities on the tube ever put at risk. (Reminder: "Politics is show business for ugly people.")  The stuff fascinates me because it is the random walk of actual history, but why do any of you serious punters pay attention to it as anything other than a minor sports bet - like wagering on Columbia basketball in Casino (the movie)?

For all the supposed money at stake in politics and the outcomes of political elections, the actual net expenditures - the money spent that goes outside the bubble of the campaign organization itself - are trivial.  There is more money spent by GEIGO and its rivals on pitching auto liability coverage than all the net payments to television for Presidential campaign ads for both parties. 

The poll numbers being reported, even this morning, are for samples taken before Friday's revelation about the Danger Man's laptop.  They "changed" because the pollsters decided to use a sample that was a statistical probability rather than one that was completely bent in favor of Mrs. Clinton.  ABC News had a headline that says it all "Shift in the Electorate's Makeup Tightens the Presidential Contest".  Yeah, right.  The makeup of the electorate over an entire campaign season does not change.  If it did, the Democrats would not be so passionate about enacting same day registration and voting. 

A week ago ABC News had a poll that showed Mrs. Clinton up by 12 points and at the magic number of 50.  The poll with the headline had her at 47 and Trump at 45.  What "changed" was the weighting of the sample.  The earlier poll divided the electorate as 36% Democrat, 27% Republican and 31% Independent.  The more recent one splits it 37, 29, 29. 

So, Stefan, how does this prove your thesis?  Easy.  The folks at ABC decided that 81% of the registered Republicans are now going to vote, as opposed to only 75% a week ago.  They also decided that 5% fewer registered Democrats were actually going to vote.

Who is in the electorate does not change.  Who is going to be foolish enough to waste their time to actually vote is always the question.

The enduring paradox of representative government is the fact that there is no statistically valid reason for any individual to bother with voting.  Your individual vote NEVER counts.  It is a pure act of faith.  That is the reason that the countries that are actual democracies are wise enough to make voting have a real cost; it is only in autocracies that voting is free, easy and compulsory.

As for who will win next week, I still favor the Harold Macmillan prediction.  When asked by a journalist (who else?) what will determine the coming Parliamentary election, he replied: "Events, dear boy, events."


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search