I want to create as many prediction markets about the extinction risks of specific species as possible. The idea is to build a more precise and up-to-date alternative to the IUCN Red List of threatened species.
Since mid-February, I have created 41 markets. Will I have enough mana and find the time to reach 100 markets by the end of March?
If you support my mission, you can help by trading the markets I have created so far or simply by sending me mana donations. I will invest all the mana I earn in new markets.
You can find my markets here:
https://manifold.markets/ExtinctionRisk/portfolio?tab=questions
π Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | αΉ173 | |
| 2 | αΉ103 | |
| 3 | αΉ30 | |
| 4 | αΉ14 | |
| 5 | αΉ11 |
Longer the market span - less accurate the market is. If the profit from an answer can be only recieved in a century, it is not a good investment of mana.
If you believe the animal is gonna die soon, you bet yes.
If you believe it is not going to, you won't bet at all.
So all your markets are skewed and less accurate than IUCN Red List of threatened species.
@KongoLandwalker This is an experiment. And I think the results are surprisingly positive. The IUCN only has 7 risk categories and they often need 10 years or more to renew an assessment. I have the impression that prediction markets can produce more precise and more up-to-date results.
@ExtinctionRisk loans are still given out over a long time. And much less is given than was bet, just couple percent per day. And no more than original bet. That is still just frozen mana.
@ExtinctionRisk prediction markets CAN produce accurate results, WHEN there is an incentive to trade. Your markets do not have such incentive (for at least one side of the weights). And for another side the payout date would still be in a decade or more.
Another thing that is needed for accuracy of markets is the amount of traders. Do you hope to get 100 traders on each question? I predict the appropriate amount of traders needed for accurate prediction will usually accumulate right before the extinction lf the species - when the event is basically determined.
Trading markets REACT to information. You questions would be perfect on Metaculus (where interested people lose only a bit of their time), not on Manifold (where people in addition lose currency, which is proportional to their opinion weight on markets).
10 years to renew an assessment for a long term prediction is a rational thing, which filters out seasonal fluctuations of unknown impact.
@KongoLandwalker If you take the loan every day you have 97.5% of your mana back in 90 days, assuming no price movement. If no one trades, you will show a profit and actually get a loan a little larger than your investment. Trading a long term market like this can be rational if you think others will move the odds significantly in the course of some months. You do not have to wait years for the actual resolution.
@HarrisonNathan or i could do the same with a market which is guaranteed to resolve or guaranteed to be moving.
Trading on extinction is an opportunity loss. It is worse than all alternatives.
@KongoLandwalker If you know that a market is moving fast in one direction, it's better to bet on that. However, if you don't, putting some mana into sound long term bets is also a sensible strategy, given that you will get it back fairly quickly in loans.
@HarrisonNathan manifold does not take into account how long before the resolution you made the bet. So betting on extinction with high uncertainty would generally hurt calibration.
@KongoLandwalker What does calibration do anyway? They don't pay you for it.
Anyway if you would take high confidence long term bets and strictly control the size of lower confidence short term ones, you wouldn't have such heavy losses on your account as you do.
@HarrisonNathan I have loss not from a prediction, but from a mathematical debate. I was a champion for SnakeEyes debate and wanted to show my confidence in the answer. I was not predicting whether the judge will pick my side as the winner (whether i will be successful in convincing him), and I made it clear in market that i placed mana just to point and the mathematical answer I standed up for.
If I prove that extinction markets are a bad investment, that does not mean I use the website as an investor myself. I use the website as a series of games and do not care about mana much.
you wouldn't have such heavy losses on your account as you do.
Ad hominem, perfect example.
@KongoLandwalker Not really ad hominem. You were talking about your preferred trading strategy and I looked at your account to observe it. I didn't know why you greatly oversized a short term bet - which seems to be what you are advocating doing - and you have explained that now. The principle stands that keeping mana available for quick-resolving bets is preferable only to the extent that you are actually winning them, and mispriced long-term markets are a reasonable place to trade if you have a little patience.
@KongoLandwalker After reading about what they did to you there, I am really sorry, though. I'm surprised you are still here. I wouldn't be.
@KongoLandwalker The silly resolution relied on an extremely creative interpretation of the grammar. It was a straightforward question, though, and you were right.
@HarrisonNathan if you look at the plot of the market, And i was not the mega NO shares holder, it would have stayed at 99% for most of the duration of the market. Because dreev and shamba had 10 times more mana on balance on their accounts.
I lost 11 times more mana than i actually had, but at least I was fighting for the representability of the graphic :-) . Most of the market duration there was 60/40 proportion of YES/NO holders, but the plot did not reflect the opinion distribution before I came.
@KongoLandwalker You placed a correct bet on a simple math problem. The fact that it was controversial is astounding, and the longwinded "arguments" people were writing are just embarrassing. And then they get a mathematician to adjudicate and he comes up with a nonsensical linguistic reason to misinterpret the question as asking about the probability conditional on being chosen in a given round instead of conditional on being chosen at all, which was clearly what the question asked. This would have turned me off this place pretty hard.
@HarrisonNathan such events do not turn me away. Make me frustrated a bit sometimes, but that's it, i learned how to lose long before that. I think it is important to argue even when I will lose and my opponents will not change their mind. At least I am increasing the chance that some neutral mind crosses the debate and reads both sides to make his opinion. If I were quitting debates, all those places in the internet I took part in over the years would look like a total unanimity. And unanimity tends to be a strong brainwasher.
There are things I value more than mana.
@HarrisonNathan have you heard about Sleeping beauty? That's another black hole. What is the answer?
@KongoLandwalker Never heard of it, but I think the Wikipedia article gives a clear answer. If you interpret the question as asking about the probability of heads on a given waking, it's 1/3. It sounds like David Lewis was answering a different question, which is: after it's all over, what are the odds the coin was heads? That's 1/2. The discrepancy is due to the fact that half the time when it's tails, you fall asleep again, so that's not the waking event you remember when it's all over.
@HarrisonNathan my position is that the answer is strictly 1/2, and that 1/3 is a mathematical mistake. Wanna hear?
@KongoLandwalker So you have it that
P(Mon & heads) = 1/2
P(Mon & tails) =1/4
P(Tues & tails) = 1/4
And based on this, when you wake up, you think it's 1/2 likely to have been heads and 3/4 likely to be Monday.
That's a reasonable interpretation, actually.
@KongoLandwalker In this problem, I took it to be a premise that the wakings were equiprobable. That wasn't clearly stated, though. There was no such ambiguity in the other problem.
@HarrisonNathan no, you did not understand me.
P(Mon & heads) = 1/2
P(Mon & tails) =1/2
P(Tues & tails) = 1/2
Because the second and the third string are the same apple.
Your idea to have 1/4 would mean that monday and tuesday are excluding, that only one of them can happen. But we know that if one of them happens, the other is guaranteed too.
Probability theory is not defined on comparing consequtive events. You cannot teleport in time into only the second situation avoiding the first. So they are the same outcome, that brings 2 consequtive situations.
Let's say you are making a poll, what percentage of your city are Andrews. When you meet a person you write their name. When you meet Andrew, you write his name and... write his name again a day later. (Consequtive events of the same outcome). Your final stats with this corrupted method will show you that Andrews are twice more likely, but they are not.
@KongoLandwalker I'm talking about the probability, when you wake up, that it's heads and Monday, tails and Monday, or tails and Tuesday. Those are the only three possibilities and they are mutually exclusive.
@HarrisonNathan btw, that model that you mentioned 1/2,1/4,1/4 would correspond to the following:
If heads, wake the girl on monday and finish the experiment.
If tails, throw the coin again:
__ if heads, wake on monday and return
__ if tails, wake on tuesday and return.
In that statement waking on monday and waking on tuesday if the first coin was tails are exclusive. In the original statement they are not, you cannot wake up on tuesday without existing throough monday.
@KongoLandwalker There's no second coin toss. You're not being clear.
Of course if it's Tuesday then it was Monday yesterday and the coin came up tails.
@KongoLandwalker You seem to be misinterpreting the question, which asks your opinion on P(heads) after a given waking.
@HarrisonNathan you took three "situations", not "outcomes". It is impossible to draw a branching universe tree (and probability theory is only about those trees, not about our perception), such that you can end up in third of the situations without going through the second.
On each branch of the tree can only be one outcome, that is the axiom of probability theory, otherwise the outcomes are not exclusive.
"After a given waking"
Do you not see a parallel between your phrase and " being in a specific round of a snake eyes game"?
You appear to be confusing "it's Monday now" with "Monday happened."
Probability theory never answers what is now. It can only answer what can happen by definition of probability theory. It does not matter when or how many time you wake up - the universe goes linearly after the recent divergence point (coin toss).
You seem to be misinterpreting the question, which asks your opinion on P(heads) after a given waking.
As you said earlier, their is no opinions in math. I would rephrase it "there is no subjectiveness in math". No matter when or how you wake up, the answer should be the same. If the coin was tails, answer on monday has to be the same as on tuesday.
The probability that it is Monday today is a valid concept.
No, it is only valid if all the outcomes are exclusive. Waking on Monday and on Tuesday are not exclusive.
(But they are in the variant wirh 2 coin tosses that i named before).
@KongoLandwalker I'm not sure because you are not really being clear, but you seem to be taking the objective perspective that was advocated by David Lewis as described in the Wikipedia article.
I'm of course using "opinion" in a different sense here than earlier.
The question asks what you rationally estimate the probability of heads to be when you wake up. (This estimate is what I meant by "opinion" if that's not clear.) This isn't an incoherent question or one that probability theory is not allowed to answer and if you believe that, I don't know what to tell you. We can speak of the probability that things are happening now.
You wake up. Certain things are true about the world at the time you wake up. Specifically, it is either Monday or Tuesday and the coin was either heads or tails. The probabilities of these facts exist.
@KongoLandwalker I'm talking about the probability that this particular waking is on a Monday or Tuesday. I am not talking about the probability that any waking happens on Monday or Tuesday.
@HarrisonNathan If you accept that the question is asking you, conditional on a randomly (equiprobably) chosen waking, what is the chance that the coin was heads, the answer is 1/3, and there is a 2/3 chance that it's Monday.
Or you can say the question didn't offer you that premise, and have it as I wrote above, 1/2 likely to be heads and 3/4 likely to be Monday. But having it your way involves some fuzzy epistemology. It's certainly not a mathematical argument.
In the sleeping beauty problem you are asked about the probability of a coin being Heads.
Probability theory can't answer questions based on non-excluding sets of events, so it is impossible to answer the question "what day is it today" using the theory. Whatever fractions you invent as an answer for this question is not in the field of math anymore.
conditional on a randomly (equiprobably) chosen waking
This assumption is wrong. If one outcome comes after the second one, there is no randomness, there is no equiprobability, there is nothing. Probability theory just not applies to your formulating of the problem.
@HarrisonNathan it is strange you still don't see the problems are similar.
"Based on a specific waking you experience" as the "based on the specific round you end up in" are both the introduction of "where am I specifically" to both of the problems.
1/2 likely to be heads and 3/4 likely to be Monday. But having it your way involves some fuzzy epistemology. It's certainly not a mathematical argument.
My statement says nothing about probability of the days! You introduced this new question. Probability theory can only tell you about coin, because coin result is a branching event, but waking up is not.
@HarrisonNathan here is another problem with your attempt to take the amount of waking up into account.
You are sleeping beauty. And you know how the experiment will go, what is the procedure.
But instead of following the procedure, I continue to wake you up million more times. Am I changing the probability of a coin in the past by doing something from the future? Or the probability of the coin is independent of the amount of your wakeups?
@KongoLandwalker The difference is that in the snake problem, the wording was "conditional on being chosen," which does not mean "conditional on being chosen in a given round." The latter is also a perfectly valid concept but it is not what the question asked about.
I'm sorry I'm taking long to respond but I am really struggling to think of how to convey to you what your confusion is. I never had a student bring up these kinds of misconceptions, and I don't really understand what you are thinking.
@HarrisonNathan I am really thankful you really take part in this conversation. Especially if you are a teacher. May I ask what do you teach?
"Conditional on being woken" is equivalent to "conditional on being chosen".
"Conditional on being chosen into specific round" is equivalent to "conditional on being woken up on Monday". Here is the part of wording which links two problems together.
@KongoLandwalker I used to teach an introductory finite probability course at a university. I did two semesters of it.
In your example with the million wakings, yes, that lowers the probability that the coin was heads (in the equiprobable assumption) or the probability that it is Monday (in the other assumption).
really struggling to think of how to convey to you what your confusion is.
Just to be fair it could also be a sign that You are wrong in your understanding.
From a good-debate-point-of-view we do not know who is right and we a searching the answer.
In your example with the million wakings, yes, that lowers the probability that the coin was heads (in the equiprobable assumption) or the probability that it is Monday (in the other assumption).
So, you say the future changes the past? Whatever defined the probability of a coin (wind/weight distribution) is completely in the past.
@HarrisonNathan out of the question scope I want to ask another one, we are on manifold after all. How sure you are that, if preiodically exchanging messages with me for, let's say a month, you will be convinced you were wrong? Is it less than 1%?
@KongoLandwalker If you are told that each waking is equally likely to be the one you are experiencing now*, then the likelihood that the coinflip was heads is influenced by the number of wakings. This is not a causality violation.
*Maybe you have a problem with perspective here, so let's instead imagine we are doing this experiment a zillion times on as many sleeping beauties. For each, we flip a coin. If heads, (A) wake and send home Monday. If tails, (B) wake Monday, sedate, then (C) wake and send home Tuesday. You are shown a series of videos of wakings that have equal numbers of A, B, and C in them. In approximately how many of the videos was the coinflip heads? One third of them.
@KongoLandwalker Now you're asking about Bayesian probabilities and I'm less comfortable with those.
If you are told that each waking is equally likely to be the one you are experiencing now*
You can never be told that. You can only know facts. One fact is that the coin is fair. Another fact (but easily broken by the experimentator) is that you will wake up twice at most.
You are shown a series of videos of wakings that have equal numbers of A, B, and C in them. In approximately how many of the videos was the coinflip heads? One third of them.
You are shown a paper, where the name Andrew is written twice more times than the name Boris. Does that mean the amount of Boris'es in your city is approximately a third of the set of people with a name Boris or Andrew?
@KongoLandwalker "You can never be told that." You can debate whether that's a realistic assumption and in fact the version above with
P(Mon & heads) = 1/2
P(Mon & tails) =1/4
P(Tues & tails) = 1/4
would be the rational way to approach this if it actually happened to you.
You're rejecting the reformulation with many trials and an objective perspective. I don't think you have a good reason for that. In that case, if you are just shown all zillion videos, you will find the probabilities are as above.
If you are saying that this zillion trial version is not equivalent to the original problem, you have some kind of philosophical objection, not a mathematical one.
You're rejecting the reformulation with many trials and an objective perspective.
No, I asked you a question in response to your new model.
You see, I predicted that you will use the model with repetitions, and described the poll-methodology-mistake even before you use it.
The sleeping beauty is a problem about a coin. You forgot that.
P(Mon & heads) = 1/2
P(Mon & tails) =1/4
P(Tues & tails) = 1/4
This model has a slice of time (midnight) when all the branches do not sum up to one.
The correct model is
P(Mon & heads) = 1/2
P(Mon & tails) = P(Tues & tails) = 1/2
It sums up to 1 at any slice of time and shows, that those 2 situations (which are actually consequtive events of the same outcomes) are the same universe. Single entity is named in two different ways, it is like saying "dice shows 6 3 seconds after the landing" and "dice shows 6 10 seconds after the landing" are two different outcomes. No, they are not, they are consequtive, to they spare the same 1/6 probability no matter how many seconds you add to your waiting.
@KongoLandwalker This is going to be my last response because I do have to sleep.
What you are talking about, it seems, is the chance of heads conditional on what day it is, regardless of any data about the waking. You're using "&" as if mon/tues denote other events (mathematical sense) when really you are just talking about the probability that it's heads on a given day (always 1/2 of course) which I think is what misled me about your meaning before.
Unlike the "poll methodology" this doesn't tell you how to bet.
Suppose you are in the experiment. You wake up and are required to place a bet on what day it is or what the coin flip was. You're going to say it's equally likely to be heads as tails? Okay. You're going to say it's equally likely to be Monday or Tuesday? You will lose money, because it's 3/4 likely to be Monday. How do we know? With the "poll methodology."
If you have a problem with that, it's, again, a philosophical problem. This isn't mathematics anymore.
You treat monday and tuesday as permutation problem about boxes. Like there is some bag with your soul, your soul taken out is randomly placed into one of boxes and then you look at the result of "in which box does the soul exist" or with which probabilities.
With your model after the experiment has ended I can ask you which situations did happen? If within probability theory, you could name only one outcome. In your 1/2, 1/4, 1/4 model you sometimes would have to answer, that 2 happened, so your model does not satisfy the axioms of probability.
"Poll-methodology-MISTAKE".
Corrupted poll does not represent real distribution. Repeated questions do no affect the probability of something that happened before.
Forget about waking up.
<<If Heads, you get 1 dollar.
If Tails, you get 1 dollar, and one second later you get 1 more dollar, so 2.>>
You will bet on Tails, but not the probability distribution was the reason, the reward function skew was the reason for that!
Same with poll, and same with you zillion-repetitions example. You just multiply the output situations of one of the outcomes.
Even in zillion repetitions example if you record the video starting WITH the coin being tossed, you will see, that the coin was tails in half of outcomes.
The sleeping beauty is about the coin, not about "what time it is". You forgot that.
You also ignored my questions: about Boris and Andrew, about whether the future can change the past.
(To the second one you incorrectly used likelihood instead of saying "conditional probability", which does change with every new bit of information, but that was still the avoidance of the direct yes/no question).
Even if we talk about conditional probability, no information is gained by the beauty during the experiment, and there is no way of distinguishind days and repeats, so her answer has to be objective, the same as before the experiment.
I honestly think that if you participated in Snake eyes long enough, they would convinced you, and you would have voted YES, because your whole position on Beauty is the same as theirs: taking subjective instances as the set to be counted, not the amount of parallel universes.
In what ratio of all experiments, if repeated to infinity, (or parallel universes) will the situation happen "it is tuesday and coin was tails"?
A small advice: when feel confused, ask questions instead of making statements that your opponent is confused.
@KongoLandwalker I said that you were confused because I thought you were talking about mathematics; you said that it was a "mathematical mistake." I see now that you are talking about epistemology, so in this sense I was confused.
I actually learned something from the Sleeping Beauty problem, and I'm glad you drew my attention to it. I immediately made the - in retrospect - dubious choice of treating all wakings as equiprobable, which I just took to be a conceit of the problem even though it wasn't explicitly stated. Suppose I frame a related problem this way: You are put under anesthesia. If some pre-defined extremely unlikely event occurs while you are asleep, e.g. lightning strikes the hospital, they wake you up on Tuesday. Otherwise, they wake you up on Monday. You wake up and have to make a bet on what day it is. What's the bet? I think this version makes obvious the absurdity of treating saying it's equally likely to be Tuesday as Monday. (This is an epistemological, not mathematical statement.)
I don't mean this as an insult but rather just literally: I don't understand what you are talking about with all the other stuff about Boris and Andrew and the future changing the past. That's why I can't address that. I cannot figure out what you are trying to say.
It's the case though, that if we run the sleeping beauty experiment, and ask people to bet on what the coin flip was and what day it is, the people who think the coin flip was 1/2 likely to be heads and that it is 3/4 likely to be Monday will be the financially successful ones on average.
@HarrisonNathan financial success is not a valid method. When you calculate an expected value of an outcome, you MULTIPLY probability by profit from the outcome, and get a decision based on that. But it does not work the other way around: if you know you are going to be financially profitable in one outcome and work backwards, the best thing you can get is the PRODUCT of probability and profits function, bot thw probability itself.
I throw a coin. If heads, you get 0 dollars. If tails, you get 10 dollars. Of course you would "bet" on tails, but the reason for that is not an increased probability, but an artificially increased Expected number of dollars in that case.
In sleeping beauty any additional wakings are not independent events, they are just the increase of the return value. In Andrew and Boris example Andrew is not actually more probable to meet, his name was artificially multiplied in the list.
The question is about the probabilities associated with a coin, not with "when you find your consciousness at", which is not a mathematical question.
@KongoLandwalker I just want to mention that in respect to the snake eyes problem, I didn't read this as a problem about epistemology. As stated, it can be read as: People sorted into groups roll dice until snake eyes comes up. Each time it doesn't, the size of the group is doubled. What are the odds a random participant was in the last group? That's mathematically well-formulated and the answer is near 1/2, not 1/36.
The phrasing used was conditional on you being chosen, what is the probability you are in the last group? I read this as being the same as the above, and I'm skeptical that any other reading makes sense.
I am going to step out of this Sleeping Beauty discussion. It is not a mathematical problem and the assertion "any additional wakings are not independent events, they are just the increase of the return value" has nothing to do with mathematics.
the assertion "any additional wakings are not independent events, they are just the increase of the return value" has nothing to do with mathematics.
That is literally me describing the expected value formula with words, and that you cannot state that by default that expected values are proportional to probabilities of the options.
E=Ξ£pv . When you know e for each of the outcomes, you know the product of pv corresponding to the outcome. But v values, the returned profits, are different for different cases, and that is what people forget when converting betting strategies back to probabilities.