Pro Noggins Prognosticate

Reader "Bret" provides an extended and insightful reply to my post immediately below (which was in turn a reply to him). Since he's threatening to start his own blog, I'll make this my last direct front-page response. But his points are so well-stated and the comments threads are so obscure, that I think it merits a direct response.

First, he notes that a "bellwether" needs to be predictive, not reactive, and thus finds fault with my definition of Tradesports as a "bellwether." It is important to bear in mind when considering polls or bets or what-have-you, that such indicators represent the state of things at present, and are not explicit predictions of the future. As someone trying to make a forecast, polls or futures prices are useful as a basis of present information upon which to build a prediction.

I don't mean to impute this view to "Bret", but I have often heard the complaint that "such-and-such a poll indicated that Candidate X would win, but instead he lost." This can certainly be frustrating, but it need never be attributable to a flaw in the poll. The poll allegedly represents the outcome of an hypothetical election held at the time of the poll - NOT an election held at any specified future date. The measure of a poll's worth is not whether it accurately captures the final result of an election, but whether it accurately describes the present intentions of the electorate. It's always possible for 30% of persons to change their mind overnight. Should such a thing happen, you would expect a poll, even one conducted a day before the election to be wildly off-the-mark. If, on Nov. 4th, 30% of the electorate was leaning towards Bush, then on Nov. 5th, that 30% changed their minds and voted Kerry, a poll which anticipated the swing would misrepresent the nation polled before the election.

So, it would be unwise to take any given poll as an prediction that a candidate will win. It can only indicate that one candidate is leading, and the other trailing. "Bret" points out that Tradesports, after Iowa, saw a crash in Dean's "share-price." Since this plunge was a reaction to Dean's defeat in Iowa, rather than an anticipation of it, he finds "As a bellwether, it seems pretty ineffective." But bear in mind, there is not yet a nominee for the Democratic Presidential candidate! We're still months away from the actual nomination, but the odds on Kerry are effectively 100, and the odds on Dean effectively 0. So, as an indicator, it's actually doing well! Of course, an idiot could do so well. My point is NOT that the market is any more prescient than any given individual. My point is that the market cautiously but rapidly prices in ALL relevant criteria and thus anticipates the consensus prediction NOT the final outcome.

By contrast, any given poll only constitutes ONE criterion. If a series of polls alternate between a Bush lead and a Kerry lead, then a cautious trader would hold onto the Bush stock because the advantage of incumbency would give Bush a slight edge (hence, most of the swing states trading just above 50). If a drop appears to be temporary, then there's a speculative incentive to buy up the plummeting shares on the anticipation that they can be sold on their way back up. Thus, a stock price won't crater unless the complete odds are starting to move decisively.

"Bret" then writes:

What I am curious about is whether this market of limited (and I was speculating, non-uniform "gamblers") might be favoring Bush in an inefficient way that we could take advantage of, i.e. the market doesn't trust Zogby, because he's considered too liberal, but actually he's a leading indicator, and other polls will catch up to him and then tradesports will be incorporate this knowledge later.

Now, this is certainly possible. Zogby is considered liberal, and his polls tend to skew ludicrously leftwards (Bush down by five IN NEVADA?!?!?!) And of course it's possible that Zogby will be right. But Zogby is a pollster, and if you've read his forecast you find that he limits the universe of data he draws upon to his own poll results. What's more, he makes the prediction on May 10, MONTHS before the election. Presumably, the "bungee jump" isn't much of a dare if he can change his forecast in September. So, Zogby has arbitrarily limited the input of his forecast. By way of contrast, Tradesports remains free to incorporate Zogby's prediction, Zogby's polls, and the far more ambivalent polls of groups like the ARG, SurveyUSA, and Rasmussen, and to factor in a conservative "incumbency bias." This means that Tradesports is an "indicator" because it is reflective, not because it is predictive.

So, as a prognosticator myself, Tradesports makes a nice fact-checker. If a state drops below 50% on the market, that indicates a high likelihood that the state is presently strongly pro-Kerry. A present state of strong pro-Kerry-hood is the best indicator available of the future electoral outcome.

It pays to watch the Tradesports prices not because they anticipate the election to come, but because they reflect most conservatively the current state of affairs. Since the current state of affairs is the best indicator of the final outcome, Tradesports is indeed (possibly) the best indicator available.