98 Comments

regarding the VA workplace activity question, the prediction probability density graph seems to agree with your conclusion as there is a high spike at the end of the graph (img: https://imgur.com/a/ZduqJaK, question: https://old.metaculus.com/questions/7137/va-workplace-activity-reach-baseline-lvl/)

We should be able to see if people really shift all the way to the end if we could see the changes to that graph over time, is that possible in metaculus?

Expand full comment
Dec 27, 2021·edited Dec 28, 2021

Re. Futuur California fires market – the resolution source says there were 9,639 fires last year, but their description says that "In 2020 there were *8112 fires*, in comparison to there were 5687 cases in 2019." So if you go off the linked statistics it's an obvious 'No', but if you go off their description it's an obvious 'Yes'. I think I'll sit this one out...

Expand full comment

It's quite possible I don't fully understand the Futuur betting market, but it seems to me that as regards the people-on-Mars bet, there are plenty of people correcting this mispricing. Out of 6000 transactions on the market, most of them are saying "Nope, no chance", which is correcting the mispricing as much as they feel they can afford (as Scott says, three years is a long time to wait for 17%)

But there are also, predictably, people uncorrecting the corrections - I looked through the bets and there's one Musk fanboy who has wagered 4000 ooms on humans walking on Mars in 2024.

And the weirder thing about this is that it wouldn't change much if it was real money rather than ooms. The 17% would drift down to some Muskman constant and the serious money would stay away because the returns - though as good as guaranteed - are too small and too distant.

Expand full comment
founding

if the "Mantic" model became legal, then anyone with a big enough following and the requisite trust factor could be a William Hill style casino. A 1% "vig" might be really bad for many gambling businesses.

If you think about it, the current 10% vig charged by gambling books are gigantic in this era of communication.

Expand full comment

Tsk tsk, Scott. I had to look up CFTC to find out it means Commodity Futures Trading Commission.

Expand full comment

I am intrigued by all of Balaji's ideas about how to make speech more free and meaningful. I just wish I understood the terminology better. I too would like some help understanding this Oxford debate thing better. And if anyone can, please also explain his ideas on pseudonymous accounts simply, as well :).

Expand full comment

I like futuur. I already won my 2 first bets and made some bets that will resolve in about 3 years.

Using virtual ooms only at the moment. I am not touching crypto money out of principle. It is a Ponzi scheme that we should not encourage. If they had option to use better money (US$, euro, pound), I would participate.

Expand full comment

A permanent work-from-home situation doesn't seem like much of an actual improvement on the previous, just shifting the misery. Some people will like it. Others, like me, find it miserable and lonely. I'm trapped in my apartment all day with nowhere to go. Sometimes I don't leave my apartment for over a week. It sucks. And I don't even have to worry about small children. I'd be thrilled to go back to the office...except that my team moved to open space, which I hate even more than permanent work from home.

It might be coming, though I think a few companies are starting to regret trumpeting the glorious end of the office. If it is, I don't think we're going to see a vast improvement in happiness or productivity. We'll mostly become more lonely and alienated, which will cause a whole host of lousy effects across society. Unfortunately, that's hard to make into a prediction markets question.

Expand full comment
Dec 27, 2021·edited Dec 27, 2021

> But second, if you go all in on correcting this mispricing, you’ll lock all 10,000 of your Ooms for three years to earn a 17% return.

(Usual disclaimers hedging that I might be misreading Scott's words and/or wrong about the object-level question.)

As far as I can tell, Futuur works like most(?) other prediction markets, where 17% means that you can buy "Yes" shares for (slightly more than) $0.17, and win $1.00 if you're right. That's about a 450% return minus fees if Mars happens, or 150% annualized, plus whatever value you assign to the question resolving earlier than Dec 2024.

Expand full comment
Dec 28, 2021·edited Dec 28, 2021

The video mentions conditional prediction markets (having markets for P(tourism increase|park) and P(tourism increase|no park) ), which seem like a great idea. One problem, and the solution that comes to mind for it, that I'm curious if other people have thought about:

Problem: Having two prediction markets significantly hurts (halves?) your expected returns, because one of the markets will just give back your investment when the condition doesn't happen, so you have to invest ~twice as much capital for the same return. For example, say I think there's a 10% chance of the park being built, a 90% chance of it increasing tourism, the market says 70% chance it increases tourism, and it'll resolve in a year. In a normal market, great, I can get a 30% return ($0.70 investment with a $0.90 expected return, or a $0.20 = 30% profit) in a year, I happily take it. In a conditional market though, there's a 90% chance I break even and a 10% chance I make 30%, so my return is only 3% in a year, which isn't worth betting at all. This will get worse with more than 2 conditional options, or deeper levels of conditioning.

Solution that comes to mind, I'm curious if/where people have discussed this: Have a parent market, "will the park be built", and then instead of that market paying out in dollars, it immediately gives you credits to bet on the relevant conditional market. So betting on "the park will be built" immediately gives me a $1 credit to spend on the "given the park is built" market, which will pay in full if the park is built and gives me nothing back if it isn't.

So in the example above, I can bet $0.10 on "the park will be built" and then put a full dollar of that toward "this will increase tourism". Now I have a 90% chance of losing $0.10, and a 10% chance of [ending up with ~1.43 shares in P(tourism increase|park is built), which have a 90% chance of becoming $1.43 and a 10% chance of becoming 0 so have an expected value of $1.29 or] a profit of $1.19, so my total expected return is $-0.10*0.9 + $1.19*0.1 = $0.03, and we're back to a 30% return on my $0.10 investment.

Question, in addition to just being interested in people's thoughts in general: Is this just recreating some complicated option system or hedge or margin purchase that you can do without the need for the "parent market pays instantly in credits for child markets" system?

[Edited to clarify math and language a bit]

Expand full comment

Prediction markets demonstrate worrying dynamics on the Russian issue. The change perfectly reflects my subjective perception shift on this risk in the last two months (i.e. the risk slowly increases).

It’s important to say that a 40% probability within a year still tells us little. As Putin demonstrated in the last 20 years, he never attacks precisely when everyone expects him, but eventually he does.

The last serious escalation between Ukraine and Russia happened 3 years ago (The Kerch Strait incident). It’s a long period, almost a half of the entire war. So it’s reasonable to assume that Putin is willing to fuel the fire soon, as he never received any dividends from this conflict yet. Neither we should underestimate his sacred love for Olympics.

So if we expect an escalation within the next 1-2 years, 40% is a reasonable baseline.

Still we don’t have any idea about the supposed nature of this escalation: whether it will be a full-scale war or something like the aforementioned Kerch Strait incident.

It doesn’t depend on the concentration of Russian troops near Ukrainian border: we see such moves EVERY year since 2014. Whether Russia will start a full-scale invasion depends only on three things: Ukraine’s internal stability, NATO’s willingness to respond, and Putin’s mental health.

But this time the shitshow at the border triggered a few political conflicts inside Ukraine. I’m worried that our politicians and their followers are dumb enough to be blindsided by internal agenda. That’s what made my perception shift recently and that’s where we should pay the most attention now if we want to reason about the Russian threat clearly.

Expand full comment
Dec 28, 2021·edited Dec 28, 2021

Thanks, Scott, for the coverage of what we’re working on at Futuur! We’re fans of ACX and its precursors so it’s great to see Futuur mentioned here :)

As indicated, we’ve recently launched our real-money markets, with trading available in various cryptocurrencies (including the USDC stablecoin for those concerned about crypto volatility). We offer hundreds of markets across all major topics, facilitated by an automated market maker. We are just getting started, so there are certainly opportunities for savvy traders as we build liquidity!

Regarding the point about our play-money markets, it is true that they can generate some odd forecasts, and we are exploring ways to address this (for instance, phasing out bets that came from inactive accounts on long-term forecasts). Overall however our play-money markets have performed reasonably well; we’ve posted a couple analyses on Medium, here: https://medium.com/futuur

And of course, we expect our real money markets to perform even better, although it’s not guaranteed. We’ve created a market on precisely this question here:

https://futuur.com/q/141989/real-money-or-play-money-markets-which-will-be-more-accurate-through-the-end-of-2021-on-futuur

Very interested to hear everyone’s feedback — positive, negative or neutral — and happy to respond to questions or suggestions folks might have, including suggestions for new markets, feature requests, etc!

Expand full comment

> It seems like the general rule is that every month, the date when people expect normality to return moves about a month and a half further out.

Interestingly, there's a very simple Bayesian model which results in this behavior. Namely, if the waiting time for some event is exponentially distributed, and you have a gamma prior on the rate, then the additional expected waiting time after waiting for time T will scale linearly with the time waited so far. It's kinda neat, since you actually have less than one observation in some sense, but you can still compute a posterior (just using P(X > T), instead of P(X = t)).

I originally thought about this in the context of nuclear war (of course neither nuclear war or pandemics are memoryless processes, but all models are wrong etc.). If we haven't observed a single nuclear war so far, how can we estimate how likely they are? You can't, but if you have a prior it also tells you how much longer each year without a war means you should expect to wait.

(Original stackexchange question and answer I wrote: https://math.stackexchange.com/questions/4113942/growth-rate-of-expectation-of-exponential-random-variable-given-no-arrival-so-fa)

Expand full comment

One way to make a Mars mission profitable would be to make your preparations in total secret, then bet against everybody on prediction markets that someone would do it by a given date.

Of course it would be too hard to completely hide the existence of a giant, well-funded rocket company, so you'd just need to appear to be a giant well-funded rocket company that was making suspiciously slow progress, like Blue Origin. Hey, wait a minute....

Expand full comment

I'm not sure I know Balaji's idea but you could try to host a debate about a topic and see which way the market moves. You could make people bet on the resolution and hold their position.

Expand full comment

Someone has nefarious plans with that fires bet.

Expand full comment

Can Americans simply lie about their nationality and use Mantic anyway? If it uses crypto it sounds like you shouldn't need to give them bank account details or anything tying you to a location.

Expand full comment

Balaji seems to be using prediction markets as a buzzword the same way tech companies keep putting things on the blockchain that do not need to be on the blockchain. Oxford debate has people spend some time arguing in a particular structured format then has the audience vote on who they think had a better argument. There's no window for usefully predicting the outcome (are you gonna do it halfway through the debate and predict 30 minutes in the future, or before the debate has even started when all you know is the topic and who's involved?) and also debates are fundamentally entertainment, this is like proposing a prediction market on who will die in a TV show that barely anyone watches.

Expand full comment

"This is good because it maintains everyone’s faith in the objective process, but bad because what people actually care about is whether there will be something that feels like a crisis situation, which is hard to instrumentalize in hospital numbers. Mantic wants to lean into the subjective side of prediction markets and see what happens."

Yeah - I discussed this tradeoff more expansively here - https://www.lesswrong.com/posts/JnDEAmNhSpBRpjD8L/resolutions-to-the-challenge-of-resolving-forecasts and here - https://www.lesswrong.com/posts/KrvXNGLxhkifQCch9/systematizing-epistemics-principles-for-resolving-forecasts

Expand full comment

I had no idea that what I needed to explain prediction markets to me was an animated dog in a wizard hat, but there you go! 😀

However, it just looks more and more like betting on horse races to me ("I am really sure that Silver Shoes will win the 3:30 at Kempton because of the going, his past performances in these conditions, and my uncle's best friend's nephew is head lad in the stables and they swear Silver Shoes will run a blinder" - now just slap a probability estimate on how you think Silver Shoes will do, and it's the same principle as "Joe Doggen will legalise black magic").

Expand full comment

As I understand it, prediction markets are only profitable if there are enough people making bad predictions to lose money to the people who make good predictions, and prediction markets might tend to drive unskilled people out.

Presumably, prediction markets need to be enough fun to keep the fish playing. I have some evil ideas for that, probably combining the prediction market with a lottery.

Am I missing something?

Expand full comment

"seems like the general rule is that every month, the date when people expect normality to return moves about a month and a half further out."

The manana law:

https://books.google.com/books?id=XAc_AwAAQBAJ&pg=PA157&lpg=PA157&dq=power+laws+manana+law&source=bl&ots=pdvm6M8Vaj&sig=ACfU3U0wr7f0D_xP53jriB6mpHAa30tA1A&hl=en&sa=X&ved=2ahUKEwiF8Z6HkYf1AhUtn-AKHYT6AxsQ6AF6BAgiEAI#v=onepage&q=power%20laws%20manana%20law&f=false

Expand full comment

If I write a really good question, what stops a bigger fish from just copying it?

Expand full comment
founding

It's not Monday, but I'm feeling man(t)ic.

A taxonomy of prediction markets.

2. There are kinds that attempt to predict "if we took up a collection to buy a rare-book NFT of the Constitution, could we raise $40 million".

3. There are kinds that attempt to predict "if we took up a collection to buy a rare-book NFT of the Constitution, would $40 million be enough".

7. Attempting to predict “what percentage of the US population supports <insert hot-button political topic>”, something which would be best measured by a statistical survey rather than a financially motivated system.

Expand full comment

The Ukraine one makes sense to me. Biden isn't threatening to go to war if Russia invades Ukraine, he's threatening harsh economic sanctions. I don't think Russia invading Ukraine this year triggers an actual war between the US and Russia right away, but it does increase the long term risks of war.

Expand full comment

I think there is a reasonably chance that every intelligence agency of reasonable budget/power that decides having the general public (and their adversaries) having access to good prediction data is against their national interest, will immediately move to shut down/discredit/hack prediction markets until the markets are unworkable.

It seems to me that it would be very cheap and easy to do.

Expand full comment

A common problem with prediction markets is that it's inefficient to act to correct to prices far in advance of the resolution date, because of the opportunity cost of the returns that you could otherwise be earning on the funds locked up in the market. The obvious solution is to invest the market's funds to realize those returns.

It's less obvious how the details should work:

- Should the amount of investment returns payed to each participant be proportional to their purchase cost, their winnings, or the average value of their position over the period it was held?

- How should the funds be invested? Is there a way to accommodate market participants with different risk/return profiles?

- What happens when there's covariance between the prediction and the underlying investment? Doesn't that mess up the nice property of directly treating prices as probabilities? Can you correct for it?

- While we're investing the underlying funds, why not invest them in *other* prediction markets? To frame it a different way—a position in the prediction market (usually) has non-zero value and is fairly liquid, so you should be able to borrow against it to make other bets. Or, you could reinvest in the same bet and get leverage. Would available leverage improve or detract from the accuracy of a prediction? What special considerations are there for a lender against this sort of collateral?

I'd appreciate any pointers to prior work.

Expand full comment

I was wondering if someone could explain something to me.

How does a question on a prediction market generally get started? As in, how does the exchange itself sell the initial "futures" to predictors, like an IPO? They need to sell them for a high enough price to pay out, on average, to the holders of the futures if and when the event takes place. How do they ensure they achieve this with enough of a safety margin?

Is there some clever trick pairing off the "yes" and "no" markets together? Do they act as a normal market participant (would this not require the exchange themselves to partake in prediction and need to do research)? I've thought about it for ten minutes and haven't come up with a low-risk mechanism that the exchange can use to get a market started, but I'm sure I'm missing something.

Expand full comment

Assuming that: (A) large numbers of people will soon be displaced by automation, thus making something like UBI necessary to avoid pesky bread riots and whatnot, and (B) if large and liquid markets are generally the best available (albeit imperfect) predictors of future events, even taking "dumb" money into account, I am trying to think of a way to tie prediction markets to UBI.

Each participant gets a certain amount of money at the beginning of each month, which they are required to use to make prediction bets on one or more matters of social interest. The amount the participant receives at the end of the month, above a certain minimal threshold, is tied to their success in making such bets. For bets which don't or cannot resolve in the next thirty days, the participant can either let the bet ride for the next thirty days or until resolution, whichever happens first, or the participant can cash out at the current bid-ask price.

The participants get UBI and something potentially socially useful to do (and that has the potential of an additional payday), and the mechanism may be more effective than paying consultants, futurologists and analysts. Also, participants which exhibit consistently successful track records may be selected to be groomed for Bigger And Better Things.

I just thought this up because I don't feel like working ATM, so tell me what is wrong with this idea?

Expand full comment