More Ukraine warcasting, nuclear war risk, forecasters vs. experts
Regarding the degradation of Zelenskyy NO on Polymarket, a lot of these markets behave like options: they have a time value, and in absence of any new meaningful information, the time value falls until it reaches 0.
It's dangerous to immediately price in the updates you expect to see based on the trend! The forecasters are behaving as proper Bayesian agents and updating slowly, according to the trickle of evidence. Moreover, waiting makes existing evidence stronger - nobody debunked it, you've seen nothing that contradicts it despite time in which such information could surface.
"This is almost monotonically decreasing. Every day it’s lower than the day before. How suspicious should we be of this? If there were a stock that decreased every day for twenty days, we’d be surprised that investors were constantly overestimating it. At some point on day 10, someone should think “looks like this keeps declining, maybe I should short it”, and that would halt its decline. In efficient markets, there should never be predictable patterns!"
If you plotted a prediction for "will this iodine-131 nucleus have decayed by April 1" you'd also get a roughly linear decline (unless it decayed in which case it would jump up to 100%). Prediction markets are allowed to have "story arcs", so long as the *expected* change is zero.
Shouldn't the fall of Kiev by April decrease with each day closer to the April deadline that it hasn't fallen? It doesn't seem that strange to me
A prediction market for next week’s Powerball numbers makes no sense as that is an unknowable thing. Similarly how Putin, Zelinskyy, Biden, Sholtz, random software problems that accidentally send a Russian Cruise Missile to Warsaw all interact to influence the future is just as unknowable as the future state of the balls bouncing around at lottery HG.
What am I missing?
Is it possible that 'story arcs' in prediction markets come because most people have far heavier priors on other people's assessment and judgement, than they do on events in reality itself?
Maybe most people, even those using prediction markets, have a real low prior on something like 'prediction markets are totally wrong here' ? I have far more experience with 'prediction markets being general correct' than i do with forecasting military outcomes.
For 5., the Biodeterminist's Guide to Parenting, #5 reads "Supplement with nicotinamide mononucleotide (for fathers)"
Should we assume that #s 2, 3, 7, 4, 10, all of them really, apply to the mother prior to birth? Subsequent to birth? The baby?
"One possibility is that, by a crazy coincidence, every day some new independent event happened that thwarted Russia and made Ukraine’s chances look better. Twenty dice rolls in a row came up natural 20s for Ukraine. Seems unlikely."
That's more or less exactly what's happened. It's not unlikely at all, it's how we experience the world.
On day 1, Ukraine was still standing.
On day 2, Ukraine was still standing.
On day 3, Ukraine was still standing.
etc etc etc
These are all new information! On any given day N, different things might happen, we don't know. We have to wait and observe. Simply drawing a linear extrapolation through the first few days would be lunacy.
Also, labeling these as "natural 20s" is a huge overstatement of what we know about the underlying distribution. We have very little information on which to base an estimate of "Can Ukraine defend against an attack by the Russian army?", and I offer as evidence the fact that this post exists, we're having this conversation, etc etc. So it *may* be that Ukraine is on a hot streak and rolling 20s, or it may be that the outcomes we see are right in the middle of the distribution because it happens that Ukraine is a better match for Russia than some (many? most? idk!) people thought.
I'm thinking that one of the worst things about surviving a nuclear war would be finding yourself in a society organized and dominated by the kind of people who optimize their lives around surviving a nuclear war.
This past month has really brought back the '80s, when "is nuclear armageddon more or less likely this week?" was a perfectly normal topic for news discussion.
> If there were a stock that decreased every day for twenty days, we’d be surprised that investors were constantly overestimating it. At some point on day 10, someone should think “looks like this keeps declining, maybe I should short it”, and that would halt its decline. In efficient markets, there should never be predictable patterns!
Most high-growth tech stocks have been in secular decline since November (see eg NFLX, COIN, DASH, PTON, U, etc), so if you really believe this, you should go pick up those obvious $100 bills just laying on the ground :)
Another explanation for the slow decline in Ukraine predictions:
This might happen if people are thinking something along the lines of "Russia will not capture Kiev by April unless they do something to turn things around before then". If so, then every day that Russia fails to turn things around provides a bit more evidence that Russia will not capture Kiev. This doesn't violate conservation of expected evidence or the efficient market hypothesis, because every day there is some small chance that Russia WILL turn things around and the odds will swing substantially in the other direction.
I'm kind of echoing what's been said before in the comments, but I'm really not sure why the monotonically decreasing pattern is odd. A stock shouldn't decline over time, but a stock is also in theory an infinite prediction (there's no time expiry). A prediction with a specific deadline is quite different.
Suppose your mental model is something like "there is a 1% chance Kiev will fall on any given day". With 30 days until April, you'd predict about a 74% chance of Kiev not falling (.99 ^ 30). With 15 days left, conditional on it not falling yet, you're up to about an 86% chance. Your base rate probability of Kiev falling doesn't have to change (it's still 1% on any given day), but the decreasing time to expiry changes your odds of the actual contract resolving one way or the other.
I think the key here is that the evidence has mostly been of the following form:
"the Russians haven't moved very far, and they've already been invading for X days"
That's evidence that becomes monotonically stronger as X increases but there's no sharp cutoff point where it qualitatively changes in character. It's quite different from the updates you would get from crucial pivotal events (agreeing to ship a more advanced SAM system, the fall of a city, losing a key skirmish etc) but I also think it's extremely relevant evidence.
For instance, if we think one of the two situations is likely to be a correct description of the situation, we could reasonably believe that every day without solid Russian progress is evidence towards option 2 and update accordingly.
Scenario 1) Russian forces are fundamentally powerful enough to take Ukraine but take time to deploy effectively
Scenario 2) Russian forces are not powerful enough to take Ukraine
Every day X ticks one higher, and the P(evidence| scenario 2) becomes higher and P(evidence|scenario 1) becomes lower meaning a larger update but incremental update towards scenario 2.
Even now, it's still possible that the Russians are gradually getting into position and will still prevail, but the alternative story where their logistics, morale, tactical awareness etc are too constraining to achieve their objectives has become a lot more plausible.
(This was a comment made by the real Enopoletus Harding on Metaculus.)
I thought of leaving here due to my disagreements with the moderation here, but since Scott has mentioned me on his blog (I can't comment there since I'm banned for extremist posts), I will continue to make predictions here. Needless to say, the chances of Russia taking Kiev go down each day there is no advance. I personally don't think it makes sense to update very rapidly, since, for all I know, the Russians may have been using the past two weeks to prepare some lightning assault -not something that's likely, certainly, but the chances of that are too high to simply exclude off the bat. Maybe I should update more rapidly -I've been underperforming the community when evaluated at all times. But knowing when to update and having good starting positions are not easy skills to learn.
I've also created an account on Manifold Markets, I expect my predictions to be better there (I have a habit of putting the vast majority of my money on a few markets I can easily win in):
No, I'm not trolling. Wrong, maybe. It's happened before. I had 10% chance of a Russian invasion for way too long, but I obviously changed that as the evidence became more clear.
Ukraine is going to lose in the end. Maybe only after most cities are destroyed, but they cannot win. Then they will be ruled by russian marionettes & dictat in the foreseeable future. The dictat will be harder and longer, in particular on the West-ukrainians, the more blood the Russians have to spill to conquer the country. The only alternative is for the ukrainians to cut their losses some way, but so far that seems unlikely. I'm writing this because the endgame is what matters. Making "predictions" about which week Kyiv and other cities fall is a teenager-way of thinking and acting about a real-life war. It is of very marginal importance which week it will happen, and such speculations are also rather distasteful. Grow up.
> One possibility is that, by a crazy coincidence, every day some new independent event happened that thwarted Russia and made Ukraine’s chances look better.
This doesn't require Thursday. The "event" is just the absence of any big Russian breakthrough.
If Ukraine is stronger than anyone could observe, then each day of non-defeat reveals this fact just a little bit to a rational observer
Regarding expert aggregation, the University of Chicago does this for economics, a bit. Link: https://www.igmchicago.org/igm-economic-experts-panel/
I'm not sure I follow the statement "At some point on day 10, someone should think “looks like this keeps declining, maybe I should short it”, and that would halt its decline."
It doesn't make sense to me that prediction markets would improve estimates without additional causal information (information tied to actual "facts on the ground") vs. this notion of the market just predicting based on itself, which seems antithetical to the goals of meaningfully accurate predictions. Beyond the market forces at play (which are just a tool to motivate effortful predictions out of forecasters), "technical analysis" (as opposed to fundamental analysis) adds little to the causal information included in prediction markets' final estimates, correct (unless someone believes that the best forecasters are always leading major shifts/trends rather than lagging them, which seems questionable; why couldn't poor forecasters move the market?)?
This feels like a fundamental limitation of the current approach/emphasis on forecasting that I've never been comfortable with: trying to predict something important by staring at a stock ticker rather than looking at the facts on the ground (even given those facts are fuzzy/noisy and their interpretation is also fuzzy/noisy). Moore's Law held for the past 50 years. Therefore, based on that, many futurists (as far as I can tell) just extrapolate out that growth in hardware performance and cost indefinitely into the future rather than ask whether there are any concrete reasons to believe semiconductors can keep being shoved into dramatically smaller form factors (Moore's Law has failed to keep pace for about a decade as far as I understand). It feels like this idea of "the trend is moving this way, therefore it will continue" is fundamentally antithetical to the whole point of forecasting and serves just to reinforce current market predictions. "a real dog" points out that delayed/lagging Bayesian updating strategies are useful, stabilizing heuristics, which I'd agree with although this a form of arguing that we should be biased to towards our previous priors more than we would otherwise be (if we didn't assume that new info is inherently less trustworthy than older info). That said, this feels different from trying to predict new things (rather than old things that were presumably based on causal info) based on a trend. The Kyiv trend looks kinda linear; in 2 weeks it could go down to -50%!
In this case it seems like the prediction market forecast is too biased toward prior predictions given how rapidly the situation is changing, as Scott pointed out. However, it's a different thing to say "given the most recent news [causal information] I've seen, the current market is off and I should short" from "given the trend is moving downward, I should short", which assumes all changes in the market should be accompanied by positive feedback effects (more formally, in linear control systems, this is equivalent to trying to always have the feedback time constants perfectly tuned, which is hard, particularly in unpredictable situations and I'd imagine would lead to instability; how long is a "long trend" that we should predict based on vs. a "short one" that is just spurious noise? How do you know in advance?). The trend only looks clearly linear in retrospect... at least for me, not being an expert of course, the situation didn't feel so linear at the time.
On the nuclear forecasting I do wonder if one month is too short a time horizon for us to expect increased risk. In my mental model of escalation we would see several weeks or months of increased tensions that gradually drag NATO in and make nuclear exchange conceivable. Perhaps Russia bombs a school next week and the west responds by putting the MIGs on trucks and shipping them to Ukraine a week later. Perhaps a missile aimed at the trucks hits a Polish border station and NATO increases shipments. A week later after some of these new systems are shifting the battlefield, Russia uses something that is borderline classified as a chemical weapon and NATO makes strong statements about no fly zones if there's further provocation etc etc.
Over 3-6 months I can see the tit-for-tat escalation getting to the point where NATO forces are massing in Poland and tactical nukes are being mentioned seriously. But 1 month just seems too fast to get to that point.
Or to put it another way, I would predict that the month before a nuclear strike actually takes place, forecasters will have put much much higher probability on it happening the next month than these numbers.
The weirdest part of the forecast to me is that "Kiev falls by April" declined from 69% to 14%, while "3/6 cities fall by June" effectively didn't change from 71% to 70%.
If you put the two of them together, it seems as those the market's prediction is "Russia wins slowly" instead of "Ukraine holds". Or maybe these markets are behaving inconsistently with each other?
A prediction from Bret Devereaux on Twitter: https://twitter.com/BretDevereaux/status/1503457631347879949
"I suspect, on the balance, the most likely outcome is probably the messy one: Russia loses in that Putin will brutally overpay for whatever fig-leaf concessions he gets, but Ukraine doesn't win in that they either lose territory, access to Europe, or some degree of both."
I’m going to echo what a lot of people have said and say there’s nothing weird about the monotonic decline in the “Fall of Kiev” market. The key new information that has emerged each day is “Russia has failed to make meaningful progress towards capturing Kiev”.
Early on, you could ascribe that to temporary setbacks that would get worked out. But as time goes on and the advance remains stalled, it becomes steadily more likely that no amount of regrouping/resupplying is going to be enough.
I cannot wrap my head around Metaculus putting 20% on WW3. Are you telling me everybody on Earth is playing Russian roulette right now?
EDIT: I was so worried when I wrote this that I didn't even spot the pun.
One possible explanation for the forecasts updating monotonically - I (top 60 Metaculus forecaster by standard rankings, top 30 by points per question) use deference (to what extent depends on the question) to the community forecast as an input to my forecasts, so my inside view forecast of Kiev falling was lower, but my actual all things considered forecast updated down as the community went down.
Your highlighting this here gives me something to consider in future, thanks!
I asked about Augur on a previous Mantic Monday, in particular the question of "what happened to Augur;" from the response and other research alongside, it seems like Augur was doing a lot of fancy engineering around zero-knowledge proofs to make their markets work as a "trustless" system, and that this lead to long development times, high transaction fees, a token that was subject to standard crypto pricing weirdness, and so on.
If this is a fair assessment, it makes sense to ask if trustless-ness is a required property for a crypto prediction market. Would it make sense, instead, to use (regular, old, analog, boring, passé) trust? For instance, have a market system where users can register themselves as "judges," post something to the blockchain to the effect of "Hi, I'm [moderately known person], I promise to make obvious assessments of market resolution in exchange for small fees," and then market-openers can assign judges they deem trustworthy to resolve their questions (in exchange for said fees.) One of the key virtues of blockchains is that everything is by-default open, so it should be easy to see when an arbiter does something dumb or malign, whereupon you simply stop trusting them (and they stop earning judge fees.)
Obviously there's a few (!) details that would have to be worked out in such a system, but would scale to at least Metaculus-like scales, and not require any super-arcane proof engineering. Is there a giant flaw in this approach that I'm missing? Is anyone working on anything like this?
I am extremely excited about getting an updated Biodeterminist's guide to pregnancy, I used he current un-updated version over the last year and it was a great starting point for deciding which interventions to pursue. But... while you are at it, I would be even more excited about a biodeterminist's guide to parenting (since that is, as of this instant, even more relevant to me)!
If every day Kyiv has an independent 1% chance of falling, the pattern of "will Kyiv have fallen by April 30th" is pretty much a straight line over time. Good luck arbitraging that. Another way to look at it is "given that Kyiv is still standing, how likely is it to fall in N days", and each day you subtract one to N and use the exact same model. It will constantly yield this monotonous curve. It does not hint at anything suspicious to me.
Robin Hanson actually confirmed a bet, though he's just betting $1 to someone else's $100. https://twitter.com/robinhanson/status/1502993000238133250
So far as I know an "efficient" market is merely one that prices accurately in all available information. But obviously if the available information changes, as it does when actual events actually happen, then the efficient market changes with it.
Id est, if actual events steadily reduced the probability of the Ukrainian invasion succeeding -- and there are all kinds of obvious mechanisms by which that could, and probably has, happened, e.g. a steady stream of Russian casualties, the regular loss of general officers, the declining logistical resources of the invaders -- then the efficient market will indeed have a visibly steady trend, but this trend will not obviously have been predictable in the beginning.
Indeed, there are scads of phenomenon which trend steadily until they don't, such as stock market rallies or slumps, or case loads during an epidemic, or fads and bubbles, meaning the underlying dynamics are chaotic, and any impression given by a short-term steady trend that the dynamics are smooth and predictable is illusory.
I think you also conflate "efficient" with "omniscient". A market that prices in all available information may still fail to predict the future, because the information necessary to do so doesn't exist yet. The only way you could have a market where that is *never* true is if you believe in a determinism so strong that a mere handful of variables could predict everything perfectly -- no free will, obviously, but also no large random events, and no chaotic dynamics. Even fairly simple natural systems don't behave this way, so it seems exceedingly unlikely something as complex as a planet of 7 billion conscious thinking creatures could ever be anywhere in the distant neighborhood of that deterministic.
I wish Metaculus had a stop-loss system that works like this: If the consensus moves more than X% from where it was when I made my prediction, I automatically withdraw my prediction and get an email notification and decide if I want to re-predict using whatever new information became available.
Right now there is unfortunately no way to opt-out of the obligation to continuously update predictions. Maybe at some point in the past I had some extra insight about a topic and wanted to make a prediction using it, but now circumstances have changed and I don't think I'm especially well informed and I want to defer to the community without logging in every day to mimic the community prediction. I just want the ability to withdraw my prediction.
The steady decline is consistent with perfectly updated priors, imperfectly updated priors, or constant, deterministic probabilities that need no updating.
It could be perfectly updated because every day probably doesn't see something like "the Russians didn't take over Kyiv today, they must have a terrible army and will lose", but rather, "the Russians were unable to take over the airport today, they will have to instead take over the radio station which is a little easier to defend and doesn't quite provide the level of supply of the airport, so let's adjust our probabilities down a little bit". If they fail to take objective #2, then the probabilities turn to Kyiv falling if/when the Russians take objective #3. The next day's predictions are changed only slightly because the update isn't going from total win to not win, but rather, from winning the best versus winning the next best (conditional on losing the best), and so on.
But note that even a non-changing probability curve could look like that curve. If the probability of taking over Kyiv is coin flip uncertainty, with a changing coin every day regardless of conditionality from the previous day, say, 10% on day 1, 9.5% on day 2, 9% on day 3, etc. etc, then one would expect forecasters to produce that kind of steadily declining graph as "high probability coins" get taken out of the forward looking product.
Would a subsidized prediction market work to extract secrets from Russian officials? You would create a market with the following properties:
1. All the questions ask about stuff that only a Russian official would know for sure until the very end
2. The questions are narrow enough that a Russian official with a sufficient level of clearance would win a ton of money, but also wide enough that others will be encouraged to make bets
3. The odds/payouts will be pre-determined in advance and the market movements won't be revealed until the end, to eliminate the problem of Russians seeing the market and choosing to do something else
4. The payouts will be subsidized by the Western governments so that people could make unlimited bets if they want to
Examples of questions:
1. (asked on Jan 1st 2022) On which exact day will Russia invade Ukraine? (1 to 50 payout if you get the exact day right)
2. On which week will Putin be removed from power? (1 to 100 payout if you get the exact week right)
3. Which city will the Russian army attack next? (1 to 20 payout)
4. On which week will Putin announce he's signing a peace treaty with Ukraine? (1 to 20 payout)
5. What will be the name of the Russian President on January 1st 2024, conditional on this not being Putin? (1 to 100 payout, predefined values + free form available)
If you don't place a limit on how big the market can get, would this be enough to entice well connected Russians to spill the beans? Thanks to the general public making bets too this might not even cost too much money for the West to operate.
A problem with prediction markets for long term bets is that they freeze up money that could be earning some larger return elsewhere. If you're buying a 50% chance of $1M in a year, that's worth far less than $500k, since a 100% chance of $1M in a year is worth less than $1M right now. If you want to have a prediction market that after a year some unlikely event will/won't have happened - you would want that bet to be as good as alternative uses of that money, otherwise no one will take the nearly certain side of that bet. However, that seems trivial to solve by combining the prediction bet with some other financial instruments.
In essence, the simple solution is that the bet should not be denominated in dollars, but in some reasonable investments - e.g. US treasury bonds, or a major stock index ETF - IMHO that should make long-term bets much more attractive. Is this something that has been offered by some prediction market, and if not, why?
Suppose England is playing France in soccer, with ten minutes left on the clock and England up 1-0. There's some kind of (idealised, high-liquidity) prediction market going on for "Will England win?"
If the participants in the prediction market can only see the scoreboard, then the market should move slowly and monotonically in one direction as the clock runs out, unless someone scores a goal. But if they can see the field, then the market will move up and down in response to every kick, moving down slightly every time France gets the ball and down slightly every time England gets it, suddenly shooting up every time there's a shot on goal and back down again when it's intercepted.
Monotonic movement is an artefact of the lack of reliable information coming out of the war zone. If metaculus predictors had access to hourly satellite photos then I'm sure the market would fluctuate a lot more.
"This is almost monotonically decreasing. Every day it’s lower than the day before. How suspicious should we be of this?" -- not suspicious at all? Every day the time left until April is shorter than the day before, so of course all other things being equal that probability should be roughly proportional to the time left until April. (Note that the full question text says they mean by April *1*, not by April 30.)
Also, I think there are good ways of using prediction markets and bad ways of using prediction markets.
And I think that "Hey, these guys did pretty well on a prediction market that one time, let's declare them to be Superforecasters and listen to everything they say from now on" is not a good way to use prediction markets.
My thoughts on v1 of the biodeterminist's guide to pregnancy:
1a. The benefit of fish is partially because it's a proxy for iodine status. Foods that come from saltwater are high in iodine and foods that don't come from saltwater are low in iodine [98% confidence] (edit: with some exceptions in modern times for chicken, cattle, and some crops that are artificially provided with iodine)
1b. Some other important minerals may also be easier to find in seafood than in landfood. [90% confidence]
1c. Throughout history coastal peoples tended to outperform landlocked peoples and iodine is probably a significant part of that. [80% confidence]
2a. Unlike saltwater fish, fish oil does not contain adequate amounts of iodine [99% confidence]
2b. The underperformance of fish oil pills is partly due to rancidity. Some of the nutrient decays to a defective form that is probably toxic. [80% confidence]
3. Losing 2-3 IQ points per decade of paternal age was surprisingly high. The rate of de novo mutations is only about 2 per year of paternal age. This implies a cost of about 0.125 IQ points per mutation. I would not have expected a number that high, considering there are 6.4 billion letters in the genome and almost all de novo mutations are SNPs. It suggests you can turn an Einstein (IQ 150) into an imbecile (IQ 50) by changing 800 bases at random, or about 1.25e-7 of the genome. Compare: 1% ethanol increases mutation rate by 10^-8 per base per generation in S Cerevisiae. ( https://www.nature.com/articles/s41467-020-17447-3 ) Let's make the crazy assumption that the effect on yeast is the same as the effect on humans and calculate how many mutations we'd get from being at the legal limit BAC for the entire human growth cycle (30 generations of cells). 6.4*10^9*30*0.08*10^-8 = 153. That's only 20 IQ points. But we're probably less resistant to alcohol than the yeast that made the alcohol. Would like to find some alcohol induced mutation rate data for humans even if it's just HeLa cells in a test tube.
4. Wikipedia has a handy chart of the dose-response effect of alcohol on various diseases: https://en.wikipedia.org/wiki/Alcohol_and_health#/media/File:Health_risks_of_alcohol_consumption.svg The tiny bit of CHD hormesis for very light drinkers is dwarfed by the very bad effects on cancer at all levels.
5. Studies on alleged benefits of alcohol are confounded not only by the sick-abstainer effect, but also probably by the has-enough-disposable-income-to-waste-on-alcohol effect, the has-enough-friends-to-go-out-drinking-with effect, and the alcohol-industry-pays-for-the-research effect.
6. Imagine tobacco companies in 1970 found some small real benefit of smoking and hyped the crap out of that. Then it all gets confounded by doctors recommending light smoking to reduce the risk of obesity, and then people who follow doctors' advice start doing that, and the tiny benefit starts looking bigger. They could have actually done this by emphasizing the fact that smoking protects against obesity. I am suspicious that something very similar is going on with alcohol.
The nuclear-war probability forecast seems unfortunately low. Looking at the source, it looks like they tried to estimate the probability of a nuclear war as a result of this crisis, took that as a "current probability per year", and then took the one-month probability as 1/12th of that. The current crisis will almost certainly be either over, or stalemated, in much less than a year.
If that understanding is correct, then the truncated-aggregate of 0.067% per month really means "0.8% chance of Russia-NATO nuclear war as a result of this crisis", which, OK, is more plausible but most of that is going to be concentrated in 1-2 months.
Roughly, take the Metaculus prediction of Russia taking <3 major cities by June as the probability of Russia not being able to win a conventional war against Ukraine at a price Putin can afford. There's a possibility of Russia winning in a prolonged war despite their crappy logistics, but there's also the probability that they take 3 cities by June because they nuked or gassed them in April, so handwave those as roughly cancelling.
That's a 30% chance of Russia not winning a conventional war against Ukraine, which would be catastrophic for Putin. Possibly existentially catastrophic; note the 20% forecast of him not running Russia in a year. So, if he can't win a conventional war, 50% for him to try an WMD war (against Ukraine, not NATO). But only 20% for "WMD" meaning "nuclear". So, 0.3 x 0.5 x 0.2 = 3% for Putin to use tactical nuclear weapons against Ukraine sometime in this crisis.
That's going to put huge pressure on NATO to do something drastic, and put Russia on hair-trigger alert for the possibility of NATO doing something drastic at the same time, and that gets us into Cuban Missile Crisis territory. I think we can manage that with maybe an 80% probability of avoiding an actual Russia-NATO nuclear exchange, but that's still a 0.6% probability of "World War Three" coming out of this - close enough to EA's 0.8%.
The 82% for "we'll at least stop short of mass city-killing" and 75% for "if it does come to city-killing, we'll see it in time to run" seem about right.
Good news is, I think the 78% fatalities in London (or Berkeley or any other major city) is on the high side, probably by a factor of two or so. This isn't the Cold War, with tens of thousands of strategic warheads ready to go. I don't think he can spare more than 6-10 warheads in the 100-500 kT range for London, with everything else he'll have to deal with; NukeMap seems to be broken right now but think ~40% fatalities rather than ~80%. Picking the low-hanging fruit in the densely populated urban areas rather than trying to flatten every suburb while simultaneously digging out every deep reinforced-concrete basement.
So, BOTE, six times the near-term monthly risk of World War Three and half the expected casualties if we do get that, or 72 micromorts.
Doubled if you do live in a densely-populated urban area, and quadrupled on top of that if you aren't prepared to evacuate on either very short notice or very modest levels of escalation. So do keep your NorCal friend's number on speed-dial.
> Along with reassuring me I made the right choice not to run and hide...
Sorry, I'm confused: how come ? The table above this quote shows that there's a 0.8% chance/year of global thermonuclear war -- an even which will likely wipe out our technological civilization, if not for good, then for a long time. This probability is way higher than most other X-risk estimates, including those from AI and asteroid strikes. I am kind of tempted to run and hide, if not immediately, then at least soon -- why aren't you ?
Depressing that Putin is now *more* likely to remain president. What's up with that?
Is there a way to sell on Scott's manifold biodeterminist thing? I bought a few of the options but it would be easier if I could directly sell one or two that are (in my opinion) overpriced.
Also, Scott, do you want me to predict what I think you will think as opposed to what I think is *true*? Clearly the market incentivizes me to do the former, but I don't mind doing the latter and losing a little bit of fake money for the sake of contributing to the knowledge pool.
That is to say, if I think intervention X is very important but I also think that you don't think so, then it would be irrational for me to buy X on manifold if I'm trying to maximize returns, but it would still be rational to buy X if I want to encourage you (or others) look into X properly. Is this type of reasoning encouraged or discouraged?
20 % chance on Polymarket that Zelensky will stop being president by April 22 seems too high. I'd short that to something like 10 % if I could could be bothered to figure out how Polymarket works.
Ad WW3 on Metaculus, it was 14 % before February 24 compared to 21 % now. I don't have a good framework on what is the correct probability of WW3, but 7 point jump because of the invasion of Ukraine seems clearly excessive.
EDIT: Also more than 50 000 dead civilians in one city(!) on 12 % seems way to high.
I was surprised to see some commenters questioning the value of markets that give probabilities. We've had these in sports for centuries. The parimutuel odds tell you a lot about a horse race, and make it much more interesting. In football they use a point spread instead so the odds are always 50% but the event is tilted. Again, that's very useful information for a spectator. There are a few bookmakers out there who have to digest the odds in order to make a living, but for most people knowing the odds is part of being an informed spectator.
A reasonable summary of the forecasters vs. experts seems to be 'They're about the same but the experts have making coherent up-to-date predictions as a low priority so they half-ass it while the forecasters take making predictions seriously' Which makes a lot of sense if the forecasters are listening to the experts and translating what they're saying into coherent predictions.
The prediction market I really want to see is where the Ukranian war will be at the end of the year.
I think the point about the "natural 20s" is backwards. My model is, on any day, Russia wins if they roll a natural 20. Each day that they fail gives them one less chance to do so by April 1. Ukraine keeps rolling a natural 1-19, which is exactly what we should think is most likely on any particular day, but it does give us a little bit more information.
> Will Zelinskyy no longer be President of Ukraine on 4/22?: 63% —→20%
It's painful to see his name written like that
> 5: Related: I’m testing Manifold as a knowledge-generation device.
I raised the question in an Open Thread, but are there any successful examples of them being used as a "future-generation" device? Given big enough volumes, prediction markets are open to "insider trading", where you can make use of hidden information (e.g. something that didn't happen yet, but what you're premeditating).
I'm glad to see the edit in the first section. This is a really common pattern in a prediction market with an end date and default outcome. Each day that the thing doesn't happen, the probability will shift closer to 100% chance of the default outcome barring a major event (which will often lead to it dropping to 0%). Things can change as events occur but if you wake me up on March 31st and say "Kyiv hasn't fallen. What are chances it holds until April 1st?" obviously I'm going to assign a higher probability than I would on March 16th, which is higher in turn than March 1st. But I still won't say 100% without new information.
Jumped straight down the comments when I read "London is a hub for the effective altruism community". Is it?! Can someone tell me more? I just moved to London, and have always been quite sad that due to certain factors tying me to the UK I wouldn't be able to get involved as much as I'd like. How do I find out about/get involved in EA stuff in London?
So I decided to try writing one of these markets ... https://manifold.markets/AlexPower/will-the-nytimes-wordle-for-march-2
One possible rational explanation for probabilities falling so consistently is that the distribution of expected evidence is very skewed. To see why, consider a prediction market for "will aliens appear by 2050?". What would the price of this event look like? Realistically, it would start off at some number X, and then slowly decrease in a linear fashion: in 2030 it would be X * 2/3, in 2040 X * 1/3 and finally converge to zero by 2050. The only possible alternative is if in some year, aliens actually do come (or strong evidence of aliens appears), and the number shoots up really high.
The Russian invasion isn't _exactly_ like that, but there are many aspects of it that lean in that direction. The military situation now is roughly "stable", and every day that it remains stable is good for Ukraine. But there's always the possibility of some kind of significant breakthrough for Russia, and that could happen suddenly (if they launch a major offensive, and that offensive succeeds). Of course, there is also the greater possibility of minor new info favorable to Russia coming in, as well as Ukrainian breakthroughs (like the recent breakthrough at Mykolaiv: https://www.dailykos.com/stories/2022/3/17/2086493/-Ukraine-update-Ukraine-retakes-75-miles-of-Russian-held-territory-pushes-toward-Kherson ), which would make a "rational" probability trajectory look more Gaussian (and so if the probabilities update linearly despite these facts, that's evidence of irrational updating-too-slowly).
It may well be a combination of both of these effects.
This should be obvious, but prediction markets have an obvious fail mode for things like nuclear war.
To state the obvious, there is no logical reason to "bet" on nuclear war because even if you win, nobodys going to pay up.
https://smoothiex12.blogspot.com/2022/03/well-well-well.html how about including some very interesting endnotes from William Engdahl's latest missive (he wrote Full Spectrum Dominance) and I consider him a foremost historian and truthteller all within one's own life prism and experience of course; check out the above blog on Ukraine and Russia with regard the biological weapons. FYI and IMO only. Also, I agree with some of the people on this "site" like in the absence of any meaningful information why not just GUESS? sort of Blinken.. just make it up and pray no one notices. :) have a great Easter week, maybe I will be back next week: I don't like hidden or international sites with codes in it: who knows what could REALLY BE GOING ON?.. :) cheers !