2013-01-14 06:27 am (UTC)
Yeah, but it uses an insane betting system.
2013-01-14 11:05 am (UTC)
It's similar in that it's a criticism of Mr. Silver's argument about modeling.
But, if anything, it's an argument AGAINST prediction markets. It describes how people with bad models (and intentionally bad models, at that) used those models to make money. The financial market, which is basically a prediction market on the subject of "how much money is this asset worth" not only failed to provide sufficient incentive for good models of reality to be created, it actually incentivized people to knowingly create bad models.
In the six hundred fifty years between the Norman Conquest and the neutering of the English monarchy, Wikipedia lists about twenty revolts and civil wars, all the way from the Barons' Wars to the War of the Roses to the English Civil War. In the three hundred years since the neutering of the English monarchy and the switch to a more Parliamentary system, there have been exactly zero.
You mean apart from the Jacobite Risings, the American Revolution, and various rebellions against British rule in Ireland and India?
Those were generally revolts by people who did not get a vote towards who ruled in Westminster.
The original Churchill quote is "Democracy is the worst form of government, except for all those other forms that have been tried from time to time." Technically, "futarchy" hasn't been tried yet...
2013-01-15 01:02 pm (UTC)
I am not sure how futarchy would actually work, since it requires hypotheses. (just unifformed.) In the US democracy, the President and/or individual representatives serve as hypothesis providers, and a big part of politics can be control of hypotheses. It seems like whoever creates the funds would be the target of intruigue.
What does resolving the gun control debate look like in the glorious prediction market future? Two bets open, one conditional on a proposed legislation passing and murder rate changing by >n% over the next 5 years, and another conditional on it failing and murder rate changing by >n% over the next 5 years, and you look at the difference they're trading at to figure out whether people bet that the legislation will help or hurt?
Something like that. I understand in practice it would be much harder than this, since lots of people care about things other than the murder rate, and since linking murder rate to gun control might be hard (like if we pass gun control, and at the same time we fall into a terrible economic collapse that drives crime through the room, does that count as gun control failing?)
I think futarchy is more of a sketch for a mechanism that could work in a lot of different systems (the same way "voting" works in democracies, constitutional monarchies, republics, and even some anarchies and communisms) and that the rest of it hasn't been fleshed out well enough yet.
I can't actually see how this will work. You open a pool on whether gun control will help - but you can't find out whether the betters were accurate without actually passing the legislation. There's no feedback loop, and thus you're not rewarding the market in a way that's likely to get the truth out of it.
Or am I missing something?
You have four contracts, one pays out iff legislation passes AND crime is above X in Y years, one pays out iff legislation passes AND crime is below X in Y years, one pays out iff legislation doesn't pass (etc.)
then you do math and derive conditional probabilities.
There's still a way to accuse a prediction market of bias: someone has to decide whether the bet pays off or not. One person says "Crime is low" and another person says "Crime is high" and the market judge has to decide who's telling the truth and who's a biased crackpot. [Disliked political figure] can simply say "Of course the prediction market will say official crime statistics are going to be lower/higher, because the people who come up with official crime statistics are manipulating the data so it comes out the way they want."
I figure the prediction market bet would be something like "The FBI's crime statistics for 2020, calculated using the exact same methods the FBI uses now, will report lower crime" rather than just being about crime directly.
Prediction markets also seem to have a self-fulfilling prophecy problem.
Suppose I, and lots of other people, wanted to give my support to the "best" candidate that I thought met a certain minimum level of support. So I look at the prediction markets, pick the "best" candidate with at least a 40% chance, and do things like volunteer for his organization and write it checks. And other people are doing similar things.
What I didn't know, before checking the market price, is that someone just bid up the prediction market price on that candidate from 30% to 40% for what seems like no good reason. Because of this, the candidate is now considered a "serious contender" and is therefore given much more media attention, finds it easier to attract donations and supporters, and so on. The intended causality worked in reverse - the probability of the event is tracking the prediction market price, rather than the other way around.
In general, any system that has aspects of a Keynesian beauty contest
is vulnerable to self-fulfilling prophecies of whatever kind.
(Another flaw that's possibly less relevant is that it's hard to collect on a bet that civilization will collapse.)Edited at 2013-01-14 09:02 am (UTC)
That's not much different from how it works with traditional polls, though, right?
"Another cool property of prediction markets is that they're impossible to corrupt."
Nothing is impossible to corrupt and someone will find a way to screw prediction markets if it involves making money. Look at the stock market as we have it now; it rewards companies that have high share prices. How do you get high share prices? Attract investors. How do you do that? Look profitable.
How do you look profitable? Well, one way is to shed a crapton of factory-floor level jobs and outsource to India; raid pension funds; take over an older but now struggling company, fold it into your own, strip all the assets out and let the shell collapse, etc. etc. etc.
Look at the banks - they jumped on the idea of loans funded by the construction and housing boom (I'm not pointing the finger solely at America, my own country did exactly this), packaged up debt, sold this debt on, this debt was sold on and on and nobody expected to be holding the bag when the game of 'pass the parcel' ended. But eventually somebody had to fork over the dough to pay for the debts incurred, and the assets weren't worth the liability, and here we are a few years' on still paying off on debts.
I think those show the system isn't very good at producing a happy society, but I don't think they show the system is wrong on its own terms.
Outsourcing jobs to India does raise profits. Raiding pension funds does raise profits. Buying sketchy financial industries does raise profits (in the short term, which is when the stock prices were going up).
These are all bad things for society, but if you think of the stock market as a prediction market for profits, it was correct to predict higher profits for companies that took morally questionable but profitable actions.
2013-01-14 04:38 pm (UTC)
Prediction Markets and the Singualrity
How could a prediction market help with predictions such as that unfriendly AI is more likely than friendly AI or similar futuristic ideas? There is no way to update on evidence. A person concerned with such a topic can only wait for the prediction itself to be either falsified or confirmed, in which case everyone either wins or loses. Which means that the usual incentive provided by prediction markets to guess correctly is missing.
Your "proofs" that prediction markets are "always be the most consistently accurate source of information available" and "impossible to corrupt" are fallacious as written; they depend on unstated assumptions that may or may not hold in practice.
Now, I'll give you credit for getting around to mentioning liquidity later in the post, but this point still really needs to be emphasized. For example, during the last election there were opportunities for arbitrage between InTrade and other betting sites that went largely unexploited:
I'm assuming that an actual government-supported prediction market that everyone in the country knows about and which has real financial industry people running it will be at least as liquid as NYSE or NASDAQ. At those levels I think it's safe to just model liquidity as infinite.
Comparing it to InTrade, which last I heard required you to send money in the mail and was technically illegal within the United States, is like saying elections couldn't work because when you tried a mock election with three voters one of them voted for his dog.
2013-01-14 05:08 pm (UTC)
One thing that supremely irritates me is the way that otherwise good rationalists seem to entirely forget the empirical method when markets are the topic of discussion. Your theory says that markets do all these cool things? That's nice. Now, open any newspaper printed in the last five years. Alas, your theory is not supported by the evidence.
I basically agree with this point.
2013-01-14 05:38 pm (UTC)
where does the money come from?
Have you done the math as to how much capital it would take to make prediction markets in _everything_ liquid beyond the point of profitable manipulation?
Note that obviosuly bets have to be in soem sense real capital, unless you allow everyone to say 'hey toss a coin: maybe I win 100 million, or maybe I go bankrupt'. Which I can't see ending well...
I suspect the answer is going to be something like '5 orders of magnitude greater than the planetary GDP'.
"First, it's supposed to place an upper bound on how terrible a leader can be."
Supposed to? Reign of Terror, anyone? Or, for that matter, the late Roman Republic? The transition to the empire was unquestionably an improvement, even if they had succession squabbles that made most monarchies look like pikers. All you have to do to be Hitler-level bad to most people is to do them in succession.
And who needs a leader? Anyone could be exiled from Athens at any time for any reason -- that's where we get the term ostracism from. If that's a minority, it still leads to Madison's observation:
democracies have ever been spectacles of turbulence and contention: have ever been found incompatible with personal security or the rights of property: and have in general been as short in their lives as they have been violent in their deaths.
"Reign of Terror"
They hardly got around to democracy, especially an orderly change of leadership. And pre-revolutionary France was full of its own terror too, which is part of why they had a revolution.
"the late Roman Republic"
Rome was never a good democracy, more like a formalized plutocracy in its constitution. That aside, only the city of Rome was the 'republic'; it's like having the US run by the city council of NYC. Plus slavery, women, foreign troops...
"democracies have ever been... in general"
Yeah, because Madison had *so* many data points to be extrapolating from.
Athenian democracy lasted as long as the United States, and got squelched by Macedonian conquest. The Roman Republic, such as it was, lasted longer before sliding into even worse plutocracy and civil war and dictatorship.
My reply to this ballooned out of control and has been moved to my own journal
Hitler was still pretty bad for the average voter.
Also, the idea of a mystical 'we' with free will that 'has a choice' when electing 'the leader's feebleminded son' seems to me to be at odds with the both simpler and more predictively valid models that actual candidates and their supporters take advantage of in order to win elections. From that perspective, ending up with a feeble-minded son of a previous ruler is what it is, and mythic choices as inputs to the result don't count as a point in the result's favor.
And the average voter didn't vote for Hitler; his party had 30+% of the seats, which was the most, so he got invited to form the government. And then taking over even further by force.
Of course, things were pretty bad before him too, with the late Weimar deflation.
2013-01-14 08:04 pm (UTC)
What prediction markets really do
I agree with parts of your argument, but I have a few issue with using prediction markets for actual policymaking:
First I look at what seems like the longest-existing prediction markets we have - sports betting; "who wins, A or B" is far less complicated than governmental policy and yet I see numerous examples where in retrospect betting lines aren't just a little off but were massively out of sync; I see no reason to expect that the occasional misvaluation we see in sportsbooks won't happen here as well and while those failures would be great for making money for whoever managed through luck or insight to bet 'correctly'... it's going to suck for the country as a whole if wrong policy is implemented, and unlike having the option of doing actual fight to see who wins policy decisions may be either irreversible (eg: bomb x country) and are particularly non-repeatable (no matter how much subsequent knowledge told us it was a bad idea, we can't go back in time to undo the housing crisis).
Next (and exacerbating the former) is corruption - you state that prediction markets are protected against this by market forces, which is true to a large point but once you use the market to determine policy (rather than letting policy or what have you chose the 'winner' of the market) you introduce an inflection point where short-term corruption can exist long enough such that the normal market forces will be unable to revalue things in time. Granted you could mitigate this somewhat by not deciding anything if the market is volatile but that means that a sufficiently monied actor could force a no-decision through constant investment - and no action is in many cases still an action. In this "noop due to perpetual volatility" case you also now have a negative where money is being tied up indefinitely rather than being funnelled into something useful.
Third, and more importantly I think that you're conflating prediction markets' abilities to predict what WILL happen versus what SHOULD happen, which makes policymaking using them problematic. Take last years debt ceiling debate, judging by market response and the end result it seems like it would have been long-term better for everyone if things had been decided early rather than fighting to the last minute, but smart money is definitely going to be on the debt ceiling in February resulting in the same partisan brinksmanship. If you wanted to edge into more heated topics take any moral issue that tries to go against human nature.
Finally, you're binarizing things in a way that isn't necessarily fair (or accurate). Take the presidential election - say Obama is up 70/30 on Nov 5th, does he win?, do we roll at d10 and 3 or less gives it to Romney?, if two weeks later the line switches do we get a re-roll? Extend this down to whatever the margin of error is on a prediction market (arguably smaller than that of an actual concrete vote count and how many times have we screwed that up in the last decade or so?).
|From: gwern branwen|
2013-01-15 11:44 pm (UTC)
Re: What prediction markets really do
> sports betting; "who wins, A or B" is far less complicated than governmental policy and yet I see numerous examples where *in retrospect* betting lines aren't just a little off but were massively out of sync
I think I see a problem here...
2013-01-14 09:56 pm (UTC)
I don't think your comments about Schelling points were very clear. As written, I disagree with them. But I think you really meant what you wrote in your "schelling point for rage" post, which you did link to.
Democracy does not provide any more certainty in identifying the leader than does monarchy. They are both Schelling points. What you seem to say in the other post is that the public always has the option of overthrowing the government. Democracy channels this into elections, making it easier, and much less violent. You could say that democracy provides a Schelling point for identifying the leader of popular uprisings, which often lead to chaos without this structure.
By your numbers, China failed a dozen times in 2 or 3 thousand years. That's almost as good as the 300 years you attribute to English democracy. There are two problems with this. One is that you are using wildly different standards to count rebellions in China, early England, and late England. The other is that England is cherry-picked as the best example of democracy. (You chose China in the same way, but with its population and history, it has to count a lot.) France has had 5 (or 6) republics in less time, 3 of them replaced from within. The English Commonwealth wasn't very stable, either.
I think my comparison of English rebellions was fair. It was pointed out that there were occasional colonial revolts like the US Revolution and several in Ireland, but those seem qualitatively different. I looked through the various English revolts I found and continue to think there were vastly more during the monarchy than the democracy.
I don't think England is cherry-picked. France did have several Republics, but AFAIK most of these were punctuated by periods of monarchy, empire, and foreign conquest, not by people in one democracy overthrowing it and starting something else. After they settled on a genuine republican form of government in the 1870s, I believe after that they only had constitutional crises that never made it to armed rebellion, plus foreign occupation.
If you think there was a monarchy more stable than England or China, I'd like to hear which one you think it is. As for democracies, they have a poor stability record in Third World countries but otherwise they're pretty impressive.
Yes, I meant about the same thing from my article above as I did in my Schelling point for rage article. I'm sorry if that didn't come through.
"In a democracy, the criterion for legitimacy is an objective and easily verifiable one - they got the most votes in an election. If there's any dispute, you can just hold another election. As a Schelling point, it's hard to beat."
The US does seem to be finding the failure mode of Democracy. Voter fraud has been getting more popular, and worse, more noticed by the public. Notice the uptick in "X isn't my president" and claims of illegitimacy in the last four elections. I expect those to continue in for whoever takes office in 2016 (can I bet on this somewhere?). At some point succession will be tried again, if not violent revolution. I just hope we have another Buchanan in office when that happens.
I recently heard a claim that voter fraud is less common than everyone thinks, and that most claims of "There were X thousand fraudulent votes cast in this election" usually just means "There were multiple votes cast by people with the same name and birth date on the voting rolls", which may just mean there were several John Smiths born January 1st.
More worrying to me are laws trying to gerrymander or decide who can or can't vote (the latter are acceptable in and of themselves, but scary when they're done with the aim of disenfranchising voters one expects to vote for the other party)
On the other hand, 2000 seemed like our nightmare scenario, and it didn't lead to even a hint of armed revolt. That makes me pretty optimistic.
I'm reposting a comment of someone else's that I had to delete because the user name was really long and screwing up the formatting:"I hereby correctly predict that if legislation X is enacted, my puppy will die (because I'll kill it myself).
What's preventing folks from covertly encouraging negative outcomes in order to win a bet? It doesn't have to be as dramatic as gunning down innocents yourself - for example, an NBA referee betting on his own games (http://en.wikipedia.org/wiki/2007_NBA_betting_scandal).
Any effort preventing or predicting such second-order effects seems costly.
What evidence is there that the benefits of prediction markets are robust enough to bear either this, or the difficulty in mitigation? Have any futarchy-dreamers addressed the problem? You can say "we embrace insider trading - thanks for giving us such accurate predictions!", but I'm not convinced.
(credit for this idea goes to my brother, Kyle Graehl)"
I'm curious how you think this would happen. Can you give a concrete example, using gun control and the crime rate? Would I have to say publicly to investors: "I will commit 20,000 gun crimes if gun control passes"?
Not only does that seem unlikely to work, but I think the police would be watching me pretty closely after that.
2013-01-15 06:03 pm (UTC)
Democracy and Monarchy
"America can do some stupid things sometimes, but we would never elect a Stalin, a Pol Pot, or a Kim Jong-Il"
Odd here you are implicitly arguing against Monarchy, by saying *Hitler* is the worst thing that can happen (you really should think about that for a second or two more). But at least we won't get a Stalin, Pol Pot or Kin Jong-Il! Oh wait those aren't monarchs. Can you say with a straight face that Louis XVI or Nicholas II where anywhere near as bad as those?
Oh and speaking of how democracy never picks leaders terrible for the majority... have you heard of a man named Robert Mughabe? I vaguely recall he won fair and square at least a few elections. A few other African and Latin Americanp states where the majority quite firmly picked one of the most terribel peopel available jump to mind too. But I guess those where no true democracy. And black people don't seem to *really* count anyway amirite.
As a side note I hope no reader will with a self-satisfied proclaim North Korea a Monarchy, formal rules of succession where the son's grip is automatically legitimate aren't implemented and indeed it has to pretend the chosen son of the previous ruler doesn't have a power base just because his dad was ruler but because he is this talented and awesome genious of socialism. Needless to say both of these dynamics fuck things up. I will admit a tradition might be forming now that two sucesfull transitions hvae occured. Should the Kim dynasty continuea and Juche evolve a bit more, I'm really not that sure North Korea will be a worse place to live in 2070 than South Korea. Hmmm ... maybe someone should convince Moldbug and Lee Kuan Yew to write a series of letters explaining to young Un explaining how he can become even richer and completely secure on his throne by reorganizing the country. Maybe he can set up a few prediction markets too.
"It never has to worry about being forced to accept the last leader's feeble-minded son as the successor."
A simple solution is for the royal family to pick queens on eugenic grounds and save the sperm of princes when they are young. Perhaps even a touch of genetic engineering. Even something as simple as the Roman Emperors did where the ruler adopts the man he sees as most fit for the job. http://en.wikipedia.org/wiki/Nerva%E2%80%93Antonine_dynasty
"Yes, reactionaries, I totally just went there. I just said democracy was better than the Mandate of Heaven, because it promotes stability."
This isn't the head spinning argument you may assume it is. If you check out my LessWrong account you will see that I made the same argument in favour of democracy several months ago. I'm well aware of it and have updated, I still don't see Democracy as obviously better than Monarchy.
Imperial China has behind it millenia of relative stability and civilization punctuated by periods of fragmentation. The US has what, 200 years of stability punctuated by a civil war as bloody as any dynastic struggle. I expect that before the 21st century is over another civil war or at least a break up is likely.
You make a good theoretical case. But looking at this empirically I don't see a clear trend especially considering confounding factors. Consider Democracies besides Britain. Like say 1930s Spain. 20th century Haiti.
2013-01-15 06:08 pm (UTC)
Re: Democracy and Monarchy