Every picture tells a story

Who’s ROF?

“ROF” = Rest Of Field

FYI: ROF in this case is mostly driven by Edwards.

Some points about stock movements.

1. The market, from what I understand, has a heavy influence from college campuses, where Dean rides high. This bias, however, likely will be discounted from now on.
2. The market is defined as the winner at the convention, and each candidate will have his or her ups and downs–and, like a financial stock market, there will be corrections.
3. Information prior to recent days Iowa did not take into account of what one Iowa voter said: “Dean does not wear well.” It was hard for the market to discount how well Dean wears until the first major test.
3.

Daniel: From the moment I first saw that chart a couple of days ago, I knew it was only a matter of time before you posted it.

You’ve hit it right on the head. The myth of the accuracy of these markets’ predictions rests on the Psychics’ Defense. They holler every time they’re right. And when they’re wrong they forget about it.

Of course, everyone sees through this when psychics do it, because no one believes in psychics. But no one sees through it when markets do it because we believe in markets. It’s the defining faith of our age. If you don’t believe in their omniscience and omnipotence, you are by definition irrational.

I’m not sure you can undermine faith with arguments. I think eventually they make you drink poison hemlock.

Michael

“Can we no longer hear about the ‘predictive power’ of the Iowa Electronic Markets, please?”

Sorry, nope. The graph above is a graph showing stock prices for the nomination, not stock prices for the Iowa caucuses. Since the nominee hasn’t been determined yet, it’s a little early to start throwing stones.

It does point out the danger of mistaking a point in the process for the final decision, which I imagine is a mistake quite likely to be made while interpreting any market of this type.

You’re just sore because you put on a large leveraged position long Dean on January 2…

I think we should take seriously the possibility that the objective chance of Dean getting the nomination was 0.76 back in November and is now 0.34, and that in general this market has been tracking objective chances perfectly. I mean, it’s consistent with the data, and we all know how well markets are supposed to work.

One question is whether the numbers were gamed or distorted by frenzied Deanies. In which case, they lost their small amounts of money, as they deserved, but if they were trying to get a bandwagon effect before the election they attained that goal in a small way.

When the cost of buying in is much, much, smaller than the value of electing a president, as soon as the poll starts getting taken seriously (i. e. can be even a small factor in the election, at a cost which is also small) it seems to me that it will be gamed. Even if the people who are gaming it are fooling themselves.

Meaning that the poll was pretty good when it wasn’t being taken very seriously.

One thing I noticed, watching the numbers evolve, was that as Dean went up in the nomination market, the “Bush over Dean” price in the general election market went up even higher, and the gap seemed to widen disproportionately as if Dean’s perceived electability were actually decreasing as his stock rose.

I’m not sure whether that means the market was being gamed by Dean fans, or not. My guess at the time was that the traders tended to skew Republican; they were buying into the “stupid Dems are walking right into the Dean trap” story. That would also explain the persistent overvaluation of Hillary Clinton.

I think that the IEM was particularly likely to undergo a major correction after the Iowa caucuses – because of the unusual nature of those contests.

The Iowa caucuses are not normal elections, where there is one round of voting, held in secret, and each voter only gets to register their “first choice.” Rather, there are potentially multiple rounds of voting for any given voter (if a candidate fails to get 15% of the first round vote in a given precinct, then his voters have to either choose one of the candidates who did get over 15%, or leave). And the vote is not held in secret (which allows social forces to shape the outcome). Further, delegates are assigned to each precinct based not on population, but on turnout of registered (in this case Democrats) party members in the last general election. And of course the caucus process is so onerous that only a tiny handful of dedicated partisans bother to vote at all.

Put all these factors together and you get an outcome that is unusually hard to predict because of the underlying complexity of the process. Market boosters (like me) are not surprised by this. After all, the Beta on a biotech start-up is much higher than the Beta on an old guard industrial chemicals manufacturer. When the events underlying the performance of the “security” are more complex, the market will adjust to changes in information more dramatically.

In this year’s Iowa election we had an odd situation where the frontrunner (at least for most of the last few weeks – up until the last couple of days before the election), was the second choice of almost nobody. In a straight primary, Dean would have probably finished a few points behind Kerry, which is what the tracking polls indicated. But in the caucus room, in precincts where Gephardt didn’t get 15% of the vote, his people broke for Kerry and Edwards; where Sharpton didn’t get 15% (pauses to consider Iowa electorate and image of lim(x)=0 pops into head), his people broke for Kerry and Edwards. You get the drift.

Unfortunately for Dean (and fortunately for those of us who will probably vote for Bush but who consider a Kerry or Edwards administration mearly displeasing, rather than bone-chillingly scary), the winner in Iowa gets a rejuvenating bounce, and now the momentum has turned against him. If New Hampshire were the first election then I’d still be reasonably confident that Dean would win the nomination. Now I’m not so sure – I think “Other Than” has a little better than a 50% chance right now, and if Dean doesn’t win next Tuesday, I’d put his chances at below 25%, maybe below 15%.

Again, I don’t think this impacts the chances of Dean becoming President. Even if he had sewn up the nomination early it would only be a matter of time before he bit the head off of a small child at a campaign rally (PENNSYLVANIA!!!!!!!!!! MICHIGAN!!!!!!!! **Gobble Gobble** OKLAHOMA!!!!!!). But it does significantly change the chances of Some Democrat winning in November. I’d say Bush is still the favorite, followed by Edwards (materially lower chance than Kerry of winning the nomination but much higher chance of winning the general election if nominated), Kerry and then Clark (Who’s looking more and more batshit crazy every day).

The IEM will do a much better job at predicting the New Hampshire primary, and the general election for that matter.

SD: But aren’t markets meant to be adapted to exactly this kind of high-dimensional problem? Instead, they’ve slavishly tracked the opinion polls up until Iowa, then done the switcheroo on the basis of Iowa. That’s not consistent with semistrong form efficiency, as far as I can see. Furthermore, I reiterate my point above to James:

The IEM will do a much better job at predicting the New Hampshire primary

Kidding, right? They aren’t the New Hampshire Electronic Markets!

Couple o’ disagreements here:

you fail to note praktike’s point above that the market is for the nomination. Kerry’s performance in Iowa doesn’t wrap up his nomination.

A fair point, but not one that you can make as a believer in IEM. IEM thinks that the Iowa result makes Kerry much more likely to win the nomination.

I don’t really see that election outcomes are such good tests for markets. Only God has the inside information to make money out of this market.

I don’t see that this is true at all. Everyone has a vote, and everyone talks to people who have a vote. Far more people could be assumed to have a sensible opinion on who will win the election than the price of orange juice futures.

What futures markets do seem good for, by contrast, are events such as horse racing or terrorism characterised by asymmetric information: some people knowing a lot more or having a better insight than others.

I think you have to do a fair old bit of work to construct a category which includes “horse racing and terrorism”. And this is again not obviously consistent with what we know about markets; all the really well-functioning markets try to ensure (through fair disclosure and insider dealing laws) that there aren’t any such information asymmetries.

Who loses if markets or betting pools are found to be superior predictors than the ones we use now?

Well nobody. But we’ve got a lot to lose if they turn out to be worse, and we’ve in the interim turned over our intelligence systems to them on a point of ideology.

Lots of good posts here. Sorry I was asleep when they went up, and am responding so late. Anyway, a few things:

With regard to Michael Pollak’s comment, I don’t think that markets can only incorporate publically available information. Ideally, they provide an incentive for people with private information to reveal it as well. But I agree with him that markets — or any aggregative mechanism — have little chance of predicting an event that no one (or only a tiny number of people) knows is going to occur (like, say, a terrorist attack). There is one caveat to this: I still think that closed internal markets (or some other aggregative mechanism) would be useful within the intelligence community as a means of circumventing political and bureaucratic hurdles, and I think these might have some predictive power. To offer one concrete example, if there had been an internal market on the probability that WMD would be found in Iraq, I suspect the market’s verdict would have been very different from the one that the White House relied on (or determined).

I’m with dsquared. I think elections are an excellent testing ground for prediction markets, and I don’t think asymmetric information is a condition for markets working well, so when the IEM does badly we should pay attention. But I do think the nomination market, especially at this point in the process, is a very difficult one to use as evidence for either the positive or negative case, mostly because of the objective probability problem — how will we know, in the end, if Dean did have an objective probability of winning of .76 in November, given the fact that the primaries take place only once every four years (unlike, say, horse races) — and in part because of the volatility of voter preferences. I tend to think the vote-share markets (where the outcomes are unmistakably accurate or inaccurate) are better tests. Of course, this may point to the limits of markets’ predictive capabilities.

One of the things I am really interested in is Zizka’s question about gaming. Because the IEM is so small, it’s obviously more easily subject to influence, not by people with deep pockets (since the stakes are limited), but just by an influx of new investors. With more outside commentators paying attention to the IEM — though Daniel may put paid to that one of these days — the political benefits of gaming the market increase (assuming that, as it seems to, perceived popularity ends up influencing actual popularity). The result may be that the more attention people pay to the IEM, the less accurate it will be (although you’d think that in a two-person race, at least, both sides would be trying to game the market, and they would cancel themselves out).

dsquared writes:

“SD: But aren’t markets meant to be adapted to exactly this kind of high-dimensional problem? Instead, they’ve slavishly tracked the opinion polls up until Iowa, then done the switcheroo on the basis of Iowa. That’s not consistent with semistrong form efficiency, as far as I can see. Furthermore, I reiterate my point above to James:”

The ability of a market to accurately predict outcomes through a pricing system is determined by the complexity of the underlying events being priced and and the quality and quantity of research and analysis that are applied to that system. Nobody I know of (and I went to business school at the University of Chicago, where we routinely sacrifice goats to the Efficient Market) seriously believes that the financial markets would still be efficient if the number of people playing in them were much much lower, or if the players didn’t have VERY strong incentives to try to beat the market. The IEM certainly attracts a lot of attention, but somewhat less attention than Wall Street, I dare say. And nobody ever retired on a 50 foot yacht because they analyzed the IEM better than the crowd.

As the complexity of the event being priced goes up, we need more and more and better and better market participation to accurately price the market. The Iowa caucuses are just too complex to be accurately priced by the IEM. If there were $100 million riding on getting it “right,” then you can be sure that the IEM would do a better job. The New Hampshire primary (even the general election) is a much simpler system, and thus the IEM should perform better in predicting its outcome.

Conscious gaming aside, the Dean phenomena may also comparable to the fact that, for psychological reasons, horse-track betters tend to put too much money on long shots, and that occasionally a certain horse will pull a lot of bets for superstitious and irrational reasons which are widely distributed (a mania of crowds type of thing).

Dean of course was NOT a long shot in Iowa; my point is that there can be systematic, widespread types of irrational betting which don’t cancel one another out.

However, for insiders in intelligence services, this kind of market COULD send a message in the case when a large proportion of the staff disagree with the conclusions reached by the leadership. As far as that goes, it could also be deliberately used by a faction to send an anonymous message to the leadership. Which wouldn’t necessarily be a bad thing.

P.S. A lot of people lost money on Bob Hope over the years. I’ve wondered whether maybe he actually died years ago, but that his cadaver had come inder the control of gamblers. Of course, a good embalmer could have done a much better job than what we saw for the last several years, but obviously they didn’t want to tip their hand. Good crooks think of stuff like that.

In fact, on my point on asymmetric information, I become even more confused when I read James’ post again.

The whole point of markets like IEM he writes is “Ideally, they provide an incentive for people with private information to reveal it as well” but then adds in his very next paragraph “I don’t think asymmetric information is a condition for markets working well”.

Probably my problem is down to jargon. When I hear “asymmetric information” I also hear “private information”. Am I wrong to make that link?

An internal market roughly like the one I mentioned was part of the FutureMAP program that was killed in the uproar over PAM. PAM — the so-called “terrorist” market — wasn’t actually going to be set up to predict terrorist events. It was going to focus on changes in the economic condition, military preparedness, and civil stability of Middle Eastern countries (all subjects, I remain convinced, that prediction markets could give us good information about). The internal markets, from what little information I’ve been able to glean about them, might have dealt with more specific questions, although even there I imagine the questions would have been relatively broad (perhaps, and I’m just guessing, something like, “What are the chances of a major terrorist attack on U.S. soil in the next 12 months?”) In any case, the point is that there was an impulse in the government –or at least in DARPA and the intelligence community — to set something like this up. Sens. Wyden and Dorgan ensured that the funding for it would be cut off.

As to whether it could have worked, I guess I see the political and bureaucratic hurdles that need to be circumvented differently than you do, Michael. For the most part, I don’t think you get bad decisions and predictions within corporations or governments because of conscious malevolence or manipulation. I think most organizations — and this includes the CIA — want their view of the future to be as accurate as possible. But I think organizational biases influence the kinds of things people will say and the challenges they will offer to what those higher in the hierarchy believe. In a sense, I think the true “consensus” view (I’d say “collective” rather than “consensus,” but no matter) will often not emerge unless there’s a mechanism that allows pepole to express what they really think without fear of reprisal. And since, in many organizations, people are more likely to be rewarded for going along with the boss rather than for being right, the incentives for preferring accurate predictions to politically acceptable ones are small. An internal market is one possible way around that.

It’s absolutely true that there is no guarantee that an administration would act on information that they found uncongenial. But I guess I’m assuming that, over time, most policymakers would want to know what’s true, and not just what they hoped was true, and that if internal markets proved to be better predictors than other mechanisms, they would come to pay attention even when they didn’t like what the market was saying. But I may be naive.

James, how do you square the two statements:

On markets like IEM: “Ideally, they provide an incentive for people with private information to reveal it as well”

with

“I don’t think asymmetric information is a condition for markets working well”.

Thanks for that clarification James. Still, it takes me back to my original question about the IEM. Is it really a good test of the predictive power of markets? After all, what private information do people have that they can reveal to IEM about the nomination chances of each of the democratic candidates?

Also, when you write “Certainly in this case markets can work well (as long as the ignorant people are making choices that collectively average out to something close to what the people who know a lot know, because otherwise the ignorance of the many would swamp the intelligence of the few).” it puts me in mind of the Guys and Dolls discussion of betting. Why would someone who knows nothing bet much? That’s the point of futures markets surely? If you know a lot, then you’ll bet and that’ll show up. Why would the ignorance of the many swamp the intelligence of the few? These futures markets should give more weight to those who “know” most.

I would think people have better private information about the chances of political candidates than about most things. Their private information would be based on conversations with family and friends, their experience of the current economy, the way their business is going, their sense of how people are reacting to news from Iraq, etc.: all the myriad small things that collectively end up determining election outcomes.

James, you say:

_As to whether it could have worked, I guess I see the political and bureaucratic hurdles that need to be circumvented differently than you do, Michael. For the most part, I don’t think you get bad decisions and predictions within corporations or governments because of conscious malevolence or manipulation._

But on the contrary — that seems to be exactly what you’re presuming:

_In a sense, I think the true “consensus” view (I’d say “collective” rather than “consensus,” but no matter) will often not emerge unless there’s a mechanism that allows pepole to express what they really think without fear of reprisal. An internal market is one possible way around that._

Unless I’m misreading you, “fear of reprisal” is exactly the precondition that makes such a market useful. If there is no fear or reprisal (in the broadest sense of reprisal) or other systematic distortion, than there is nothing for the market to remedy — because in that case, the collective view will already be emerging more directly through people telling each other what they think and defending their opinions with reasons.

And that leads to my dilemma. If there is systematic distortion, I don’t see why it shouldn’t distort the results of the market (that is, lead to them being ignored or explained away). And if there’s no systematic distortion, then there’s no need to set such a market up because a clearer version of the collective view is already on tap.

With regard to the second point, I think our ideas of where the “systematic distortion” comes from are different, which leads to our differenct conclusions about the possible usefulness of an internal market.

I don’t think that mistakes in most organizations are induced by explicit, top-down policies that are meant to silence dissent. I think instead they’re the result of people’s relationships with their department heads, or their division heads, or their fellow VPs, etc., relationships that end up encouraging compromise and going along to get along rather than honest thinking. That’s not the result, in my mind, of malevolence, but rather of the nature of most bureaucracies. So the reprisal I’m talking about isn’t a system-wide reprisal — with, say, the White House punishing agents who speak out of turn — but rather the day-to-day punishments inflicted by immediate superiors worried about covering their asses. The virtue of an internal market is that none of that stuff matters. No individual is responsible for the market’s final verdict — both because the traders would be anonymous and because a market price is a collective judgment — so no one can get blamed for it. I think that would help.

I also think, more practically, that an internal market is very useful in cases where organizations are divided into silos, as the U.S. intelligence community has traditionally been. Again, the division is not the product of malevolence, but it’s clear that there was much less intelligence-sharing going on than there should have been because agencies saw other agencies as rivals. Cultural identity, in a sense, trumped the collective goal of getting good intelligence. In an internal market, people would be rewarded for the accuracy of their predictions. I assume, in that case, the desire to be right would trump cultural identity. And again, since no one agency would be given credit for the final judgment, no agency would have an interest in subverting it.

There’s no doubt, of course, that people higher up the food chain might shut down such a market as soon as it produced judgments they didn’t like. But as I said, I do think in the end most leaders want their picture of their world to reflect the reality of the world, and so if the market proved relatively accurate (that is, better than other mechanisms) I think they would keep using it.

The “marginal buyer” theory is the classic explanation for how the informed 10% set the price. But I think the idea of the marginal investor is just wrong. Prices are set not just by buying and selling, but also by not buying and not selling. If the price of a stock I — an uninformed investor — own rises on smart sharks’ buying and I don’t sell, my not selling is setting the new price as much as the shark’s buying did, because it’s only my decision (and that of all the other shareholders) to “not sell” that keeps the price at its new level.

Of course, someone is always setting the price on the margin in a purely functional sense, but unless they have enough capital to outweigh the buying (or selling) power of all other investors, they cannot, in any sense, determine the price of the asset. There are no intra-marginal investors in an asset (aside from those who are legally prohibited from selling), because at every moment (not literally, but effectively) an investor is deciding whether to sell or not-sell the assets he owns, and those decisions are as determinative of the price of the asset as any others. (I cribbed all that from something I wrote on Brad DeLong’s site, because even if it’s wrong I can’t be any clearer in my wrongness.)

The point about the investors who know everything about one stock and nothing about 499 is a good one. But that’s what’s interesting, that people don’t invest only in the stocks that they really have superior information about(otherwise more money managers would outperform the market). And that suggests to me that in most cases people don’t know what they know and, just as important, don’t know what they don’t know (acknowledging, of course, that putting all your money into one stock, no matter how certain you were of it, would not be a sound investment strategy).

James: I would say that if someone is not “literally” making a decision, then they’re not “effectively” making it, when the context is buying and selling. It’s not that you don’t have a point, it’s just that the point is dependent on a sense of “cause” which is just too rarified. As far as I can see, one could argue analogously to your above that the people of Iraq were responsible for keeping Saddam Hussein in power by “not rebelling”. Which in a sense they were, but not one which ought to gain any traction.

Furthermore, does this mean that people with no holdings of stock at all are helping to set the price by “not buying”? How about people with no wealth at all who exercise the choice not to take out a bank loan or sell their house to buy stock? It strikes me that as yet undiscovered Amazon Indians might be part of the price discovery process on this basis.

Have you read Fischer Black on “Noise”, by the way? Not that it’s relevant to this point, but it’s a good piece.