From the transcript:

  Richard: They say Bin Laden is just going to put all his money in the market and then attack?

  Robin: Well, that was crazy because these were relativity thin markets, and they have a lot of money at stake. Basically a fact that people don’t know about the markets is that many people criticize by saying, “Well, somebody will try to manipulate the markets by betting on one side not because they know better, but because they’re willing to lose money in order to distort the market price.”

  That is true. There are people willing to manipulate markets, but that actually makes the prices more accurate. For example in the fire the CEO market you say, “Well, the CEO wants to keep his job, so he will bet in these markets in order to make himself look like the price will be higher if he stays, and lower if he leaves.”

  Yes he would have an incentive to do that, but when other traders know that somebody will be trying to manipulate in the market they know to increase their trading and their efforts and that compensates, and actually on net makes the prices more accurate. That’s something we see in theory and we’ve seen in the lab, and we’ve seen in the field. These markets are robust to attempts to manipulate. In fact people who want to manipulate them make the prices more accurate.

Oddly enough, in financial markets people do not rely on self-correction for manipulation, but rather there is a tendency for it to be illegal. Market manipulation is in general not believed to make markets more accurate.

Any kind of the suggested self correction here relies on the other traders having sufficient funds/capital/risk taking capacity to overcome the manipulator and to coordinate their views to some extent. In the CEO example, how would other traders even now that the CEO was making the bet? What if some other informed party made that bet? In small, open, and rather serene settings that might work out, but in a real market with fast movements, market makers, etc.?

That's actually a slightly different thing.

The quote from the transcript is talking about attempts to manipulate the market.

optimalsolver is talking about manipulating reality in order to win money in the market.

> manipulating reality in order to win money in the market.

Analogy: working extra hard to make your company succeed because it'll increase the value of your company stock.

I think the idea is that these prediction markets are set up in a way so that the outcomes are generally deemed desirable (increase GDP for example) and it's OK to manipulate reality (do work) to achieve them while making money.

How do you set up a market for "increase GDP" without counterparties taking the "decrease GDP" position?

1. Government body says "We are considering adding full Github-flavored markdown support to Hacker News. Please establish a market on whether or not this will raise GDP by 1% over the next year."

2. People place their bets, either in the "Yes please" pool, or the "No to markdown" pool.

3. "Yes please" gets the most betting money. Betting period ends. Government decides to add markdown. People in the "No to markdown" pool get their money back.

4. If GDP does in fact go up over that 1 year, people in the "Yes please" pool make money. If not, they lose their bets.

There isn't a mechanism to make money by sabotaging GDP.

> If GDP does in fact go up over that 1 year, people in the "Yes please" pool make money. If not, they lose their bets.

Who do they lose their bets to? If the answer is the money is just set on fire or some equivalent (which I think is actually a mechanism we should use more of when it comes to things like fines), where does the money come from if they win?

More details here: https://blog.ethereum.org/2014/08/21/introduction-futarchy/

That article doesn't deal with the issue of directly manipulating the thing measured by the success metric (and implicitly assumes that it is unmanipulatable).

To take the example given in the article for bailing out banks and GDP, as soon as the "yes" trades are reverted and the "no" trades are confirmed, we collapse exactly into the scenario many other commentators on this thread have talked about. Now, all holders of negative amounts of "no" tokens are incentivized to decrease GDP in 10 years because this increases their profits at the expense of holders of positive amounts of "no" tokens. The argument is presumably that there is an equal incentive on the other side to increase GDP, but that's a fragile assumption (since in the real world betting market we still see fraud in the direction of those with power, even though in theory you could have fraud "pulling in both directions") and still leaves open the more general fragility-of-value problem (as the article refers to it later on), namely that someone can manipulate GDP but doesn't expose this when betting occurs (which again happens in real-world betting markets), which I'll talk about in a bit.

(Although if someone could explain "after ten years everyone holding the asset on the “no” market gets $26.20 apiece." that would be great, because I think that's a typo and the $26.20 just exchanges hands immediately, or alternatively everyone who's sold the asset gets $26.20 apiece rather than those who hold it? That's however irrelevant for the larger point.)

The problem we're talking about is basically the same one as a problem the article itself points out later.

> A futarchy-as-government, especially if unrestrained, has the potential to run into serious unexpected issues when combined with the fragility-of-value problem... Of course, in reality, futarchies would patch the value function and make a new bill to reverse the original bill before implementing any such obvious egregious cases, but if such reversions become too commonplace then the futarchy essentially degrades into being a traditional democracy.

The problem here can be recast as a version of the fragility-of-value: the value function is no longer accurate because it has/can be manipulated. But the half-solution that the article hand-waves, namely "futarchies would patch the value function and make a new bill to reverse the original bill before implementing any such obvious egregious cases" is doing a lot of work here and should be viewed with a great deal of suspicion.

It's not as relevant for the article, which explicitly points out it's not advocating for futarchy as government (or at least not for all governance rather than e.g. just party selection), which is probably why it's hand-waved away, but if you care about futarchy as government this is extremely important and it is not at all apparent that this "patching" would occur, especially given that the financial incentives are magnified vs a traditional democracy and there is potentially no way of knowing the value function is being manipulated, up until the very moment it is manipulated (and even then it may not be apparent that that is happening!).

In a way, truly solving the fragility-of-value problem is basically solving the same problem as AI existential risk and I would assume most people in the futarchy community agree that the latter problem is a very difficult problem. A failure mode of futarchy can then be thought of as a "monetary AI" completely optimizing for the wrong thing in spirit, even if it's the right thing in letter, e.g. manipulation of the value under measurement.

How do the No people win in this scenario? What happens if Markdown is added and GDP goes down? How do you resolve that markdown actually caused the change in GDP? (Aren't you just asking people to conditionally bet on GDP going up?) Who are people betting against if No loses? (Who's money do they get?)

Wait, that can't work; it's comparing against a counterfactual. Let's say Hacker News gets markdown. Also COVID disappears at the same time. GDP goes up by 1%, but maybe that's because COVID is gone. How do we know what GDP would have done if Hacker News didn't get markdown?

>How do we know what GDP would have done if Hacker News didn't get markdown?

We do not, but that's not the purpose of a prediction market.

The purpose of the market is not to prove that the change caused the result, but to choose the best path forward in a world of uncertainty.

You don't have to have a perfect crystal ball to make useful predictions about the future. Yes, sometimes a black swan event will cause a "good" prediction to not pan out, but on average if you can make better-than-chance predictions about the future and policies based on those predictions will personally gain predictors value, then people are incentivized to make good predictions on average.

Just like someone who can count cards won't win every blackjack hand, if they play for long enough, they are a favorite.

Sure, but the point is that with futarchy as proposed you don't actually have a prediction market on whether the policy is good, you have the government looking at the delta between markets for and against a policy, and the one for the policy which isn't enacted gets voided so anyone manipulating its price loses nothing.

Big Markdown shorts the "No Markdown" policy, it doesn't get implemented and they get their short positions back again.

There's really no prediction market incentive to bet against them: so long as they keep pumping money in to move the rate, none of your bets on the correct rate of GDP without Markdown will ever get paid out on. They only need to bid it one basis point below the price of the With Markdown GDP futures contract to get their policy, so even if you have deep enough pockets to outbid them and good reason to believe Markdown has no effect on GDP, you'd be risking a lot (GDP is pretty volatile and expert forecasts are regularly more than a basis point out in either direction) to win very little if they didn't get their way.

Maybe I don't understand the bets because they weren't clearly spelled out. Even if HN markdown was actually what held GDP back from 5% growth, you wouldn't be able to tell.

The government asked for market "on whether or not markdown will raise GDP by 1%". Presumably, they want to know if it gets an additional 1% on top of other growth (which is what? Let's say 2%?).

But there are four possible future outcomes to consider:

1. HN gets markdown, GDP grows 3% or higher

2. HN gets markdown, GDP grows less than 3%

3. HN gets no markdown, GDP grows 3% or higher

4. HN gets no markdown, GDP grows less than 3%

Of these possible outcomes, two will be discarded based on the decision to implement markdown, presumably made based on the betting odds, and then one more will be discarded based on GDP's measured performance over the following year.

Let's structure this as a bet. If you believe markdown is good, then you'll believe P(1)+P(4) > P(2)+P(3).

So you say: "I bet that either HN will get markdown and GDP will grow 3+%, or HN won't get markdown and GDP will grow less than 3%". I don't like markdown so I consider taking the opposing side.

But, I also believe COVID vaccines are going to cause GDP to grow 10% next year, and markdown will have a trivial impact in either direction. So in my opinion, P(1)+P(3) >> P(2)+P(4).

Then in that case I'm mostly betting on whether or not I think markdown is going to get implemented, not whether I think it's going to be beneficial. And if the implementation decision is going to be based on which side bets more money, then that's mostly a popularity contest. I just want to bet alongside whichever side is winning if I think the GDP is going to go up anyway for other reasons.

I think that your example largely shows that it's not useful to make prediction markets about GDP for policies that have effectively no impact on GDP.

If HN wanted to run a prediction market for features, they should use a measure like user growth, or average score of posts that use the feature or something. Something directly related to the feature in question.

But presumably there are policies that have an impact on GDP, right? A prediction market there is going to function properly. If you want to know what tax rate or pandemic policy the government should implement, it will absolutely have an impact on GDP.

Wouldn't this also open up to foreign influence dumping money and misinformation to promote negative policy decisions.

Or not even foreign influence. It's just rephrasing money = speech. Instead of spending millions to influence voters, you spend millions on positions, and you get your money back if you don't win your side? So it lets the megacorps/trade groups get richer by removing the uncertainty of lobbying in place of buying policy once you hit a certain critical threshold of money to throw around.

Yes, plausibly. Maybe you could limit this by requiring bets to be made by individual citizens (in the case of government policy markets) or by shareholders (in the case of a corporate market) or something.

There are still avenues for corruption, but it's not clear to me that they are worse than the present state. There is of course always the chance that those attempting to manipulate the market will lose their shirts.

You could set up a market for increase GDP by a range (-inf, +X] and the counterparty takes the trade (x, +inf). X does not need to be zero.

Doesn't that still incentivize counterparties to do whatever they can to undermine GDP growth? Just by X% less.

"An engine, not a camera" is a great book about this effect in financial markets.

https://mitpress.mit.edu/books/engine-not-camera

Review: http://quixoticfinance.com/the-perennial-criticism-a-review-...

He has written a new book about HFT which covers some of the same ground.

That's where Hanson's idea of futarchy (as opposed to prediction markets) comes into play. You don't bet on an event A, you bet on A given B or on A given not B. So the prediction market wouldn't be for an asset paying $1 if there's a streaker, but for two assets: one which pays $1 if there's a streaker given there's 100 or more security guards and $1 if there's a streaker given there's fewer.

I don't see how this helps any with the underlying problem. All it does is introduce an additional incentive for another party to manipulate the outcome in order to win money by betting: whoever runs the security guards.

In this scenario, the person running the security guards is the one organizing the market.

Which just means the person organizing the market now has an incentive to manipulate it. Doesn't seem like a good idea.

So the potential streaker has one extra step, discovering the amount of guards, before placing the bet? How does this solve the problem instead of creating even more of a plutocracy? Gaming the system will always be possible, except for a smaller amount of people.

The market is closed before the number of guards is decided. Indeed the whole point of the market is to decide the number of guards.

The potential streaker would have to bet in both markets, so won't affect the difference in prices between the markets.

Well, but the reason we want to decide the number of guards in the first place is that we want to prevent a streaker. If, in order to decide the number of guards, we use a process that increases the odds of a streaker, isn't that counterproductive?

This is why, for example, decentralized life insurance policies have been compared to hit contracts.

>Prediction markets would influence the processes they're supposed to be impartially observing.

Who said anything about impartiality (/cordially a strawman)?

Changing behaviors is a feature not a bug. Your health insurance writer (who's bought "No optimalsolver will not get sick") loses money if you do in fact get sick. Your fire insurance writer has an incentive to provide you free fire inspections because it reduces their payouts. A farmer plants a lucrative but fragile crop because a meteorologist can better price weather risks than they can.

Swapping exposure across space and time is a productive act.

Swapping exposure is only productive if you have many guardrails outside of the system that absolutely constrain what actions players inside the system are allowed to do. Otherwise, and this has been borne out it in reality, you end up with very distorted incentives.

This is covered today by what we call fraud (whether that be insurance fraud or market manipulation fraud), which tries to set bounds on what acceptable behavior is so that you can try to eliminate pathological edge cases. I don't see how this would be handled if everything at a top-level is handled through prediction markets.

More important than guard rails (which makes a value judgement on what actions are "allowed") is a way to determine how to settle the thing that bets are being made on ("did these actions that have stakes on them, actually happen"), auger[0] does this by:

"Once a Market’s underlying event occurs, the Outcome must be determined in order for the Market to Finalize and begin Settlement. Outcomes are determined by Augur’s Decentralized Oracle, which consists of profit-motivated Reporters, who simply report the actual, real-world Outcome of the event. Anyone who owns REP may participate in the Reporting and Disputing of Outcomes. Reporters whose Reports are consistent with consensus are financially rewarded, while those whose Reports are not consistent with consensus are financially penalized."

Where the market here can take these states/phases:

    > Pre-Reporting

    > Designated Reporting
    
    > Open Reporting
    
    > Waiting for the Next Fee Window to Begin
    
    > Dispute Round
    
    > Fork
    
    > Finalized

[0] https://v1-docs.augur.net/

The alternative to putting your money into the prediction market to influence the outcome is to try to influence the outcome in some other way (e.g. a disinformation campaign).

At when if bad actors put their money into the prediction markets, they draw further attention to their false claim--something that can't be said for the alternatives.

This is true, but in practice a very small problem.

People who can control or majorly influence these big outcomes - elections, sports matches, acquisitions, etc. - generally have a much, much larger stake in the outcome itself than any side bets.

A lot of dynamic Econ models have some condition where my choice today depends on an expectation of the future. Like how much I choose to eat today depends in part on my best prediction of how much I’ll eat tomorrow. For aggregate decisions, getting better predictions should improve welfare.

I hadn't heard that before. Can you give an example? Or do you mean in the "fixed game" way?

Imagine a prediction market had an entry for whether or not someone would streak across the field during a Super Bowl. Someone might see this, buy into the market, and then go streak across the field to force the outcome to favor their position.

It’s not even theoretical. This actually happened (or rather was attempted with traditional betting markets, not prediction markets): https://www.insider.com/super-bowl-streaker-bet-on-himself-p...

Prediction markets give financial incentive to force specific outcomes. They aren’t just observations: They become incentives to influence the outcome. The bigger the market, the bigger the incentive.

This is also the fundamental idea behind an assassination market.

https://en.wikipedia.org/wiki/Assassination_market

I feel like this is left as an exercise to the reader and no proof is needed. Bettors in a prediction market are not divine speculators causally divorced from the real world. They are embedded, and when there are a lot of them and their financial incentives are towards a particular outcome, they might act in ways that aggregate to a greater likelihood of the event transpiring than in the counterfactual setting where they are pure observers of the simulations waging in a vacuum. It seems to me there are multiple ways to formalize and prove this and this contributes to my perception the original comment seems self-evident. If that is not the case then something is wrong in my intuition.

Numerous scandals in european soccer. an endemic problem in any big popular sport, really.

https://en.wikipedia.org/wiki/French_football_bribery_scanda...

https://www.dw.com/en/police-expose-european-soccer-bribery-...

etc etc etc

You can find many simple but observed real world examples of unintended consequences from indexing on observation by searching for examples of Goodhart's Law (which, as commonly generalized, should actually be called Strathern's Law):

"When a measure becomes a target, it ceases to be a good measure."

See Wikipedia to get started, then google search for examples: https://en.m.wikipedia.org/wiki/Goodhart%27s_law

There was a prediction market on the number of times celebrities tweet a week. Someone found one of the celebrities live streaming and kept paying them to delete their tweets.

Years ago there was a libertarian proposing this as "assassination politics": simply create a system that allows people to bet that "politician X will be assassinated before time T". People who want X assassinated take the other side of that bet. Eventually there may be enough money in the pot that someone considers it worth making the hit and collecting the bet, and the people on one side have paid for it without directly paying for it.

Yes. The market facilitates payment for crime, e.g., you can't buy a company's secrets from employees directly but in a free predictions market leakers would be paid. Looks like a cynical way to expand the scope of what you can buy with gobs of money, beyond the current boundaries of law, decency and fair play.

>you can't buy a company's secrets from employees directly but in a free predictions market leakers would be paid.

Conversely though it creates an incentive for information to flow to everyone in the market instead of just to insiders. Despite shitty motives, it seems like it forces more transparency which seems like a net positive.

Markets are not based on fixed principles but are basically a game of psychology mixed with a few principles. So expectations play a big role in market outcomes.

If I had 500M to manage, I may just buy a lot of real estate in a specific place to drive up the prices in that area. I want prices to increase and with money I can create circumstances that would favor my preferred outcome.

It also creates adverse incentives for bad behavior. If I have a stake in a prediction, I am incentivized to make that happen by hook or by crook.

Yet we still use financial markets.

Isn't the idea that we put up regulations to minimize this, just like we have restrictions on insider trading?

The solution seems simple: predictions should be blind until the deadline. This also exploits the "wisdom of the crowd", where crowds are smarter when each individual's decision is independent of the others.

What am I missing?

Prediction markets are literally markets.

You can’t have a market if the price is secret.

The goal isn’t to have people place bets and see who is right later. The goal is to expose the predictions to market forces and make people put their money on the line, thereby (theoretically) improving the quality of predictions.

Yes you can have market with secret price. Check out dark pools.

Dark pool prices aren't available to the general public, but they are visible to those participating in the market.

If you reduce transparency you also lose out on a lot of utility that centers around the wagered amount being an indicator of certainty.

Suppose you're using prediction markets to encourage thorough code reviews, specifically with an eye towards catching malicious commits. Run of the mill non-malicious PR's get lots of little yes wagers, and are merged without exceptional scrutiny from the package maintainers. Then a malicious commit comes along and a reviewer wagers $100 on "not merged". This captures the attention of the maintainers and they give the PR extra scrutiny. Turns out it has a malicious commit, so it doesn't get merged and the reviewer who found the flaw is rewarded with the money wagered by those who didn't find the flaw (plus some from the stakeholders, who seed the market with some "no" money to encourage participation in the game even in boring non-adversarial times).

If you hide the predictions and the amounts, you can't use the unsettled bets as inputs for decision making.

What do you mean by predictions are blind? The issue the parent is describing is (as an extreme case): I go on a prediction market for when someone will die, put all of my money on tomorrow, and then kill the person tomorrow.

You could imagine a lighter version where I ruh for some public office, bet a billion against me winning, and then drop out of the race.

The latter example is easily taken care of though, because there has to be a market to take your bet if the market has no reason to believe you would win your odds are really low thus your winnings over your billion are really low. On top of that your risk is not symmetric as you've exposed your self for some to now spend what millions to campaign for you and thus bet against that pool.

My point is your example is contrived and not really useful.

> What am I missing?

That by so preserving the independence of predictions, you’ve also removed their utility, which is predicated on their availability.

The price of a prediction depends on what people are currently predicting. So you can't keep it blind. Well I guess one way would be to put in limit orders on odds you're willing to take, and not revealing the market price. However I don't know if such a market condition / feature would have people using the prediction market.

If you know the price and payout of a bet, it's no longer blind. And if you don't, you wouldn't buy a ticket.

Sounds clever, until you realize you want to predict stuff before it happens

True, but speculators and those utilizing predictions to guide policy should be separate.

Why?

If you're not willing to speculate on your own predictions, how can anyone know how confident you are about them?

If you're not interested in guiding policy, why not go bet on horse races instead?