I continue to think Hantavirus becoming a pandemic this year is much less than 5%. There’s a chance the opportunity to make 10x gains on the prediction market leads a WHO insider to falsely characterize it as a pandemic in what technically counts as an “official communication.”
Andes virus, the strain of hantavirus that provoked the prediction market, is endemic in two countries. It causes hundreds of cases per year. The strain that infected the Hondius appears to be wild type. In other words, unlike COVID-19, it doe not have new, recent mutations that make it unusually efficient at infecting humans compared to its historic baseline.
If this virus had pandemic potential, I believe we’d have seen it causing the same kinds of mass outbreaks in local villages and towns that we saw in Wuhan at the start of COVID. Argentinian and Chilean villages have the kinds of rodents that spread it efficiently. If we aren’t getting mass outbreaks there, why think we’ll get mass outbreaks in regions where that rodent reservoir is absent?
None of this means we shouldn’t take it seriously. A 40% CFR hantavirus pandemic would be one of the worst catastrophes in human history. I just think that outcome is highly unlikely.
I’m not sure people on LessWrong entirely understand how Polymarket works on a mechanical level, and suspect it leads them to grossly misinterpret these markets.
Currently, the “hantavirus pandemic 2026?” market is at 8.9 cents. To participate, you and counterparty jointly collateralize a new $1 share—in this case, ”no” pays 91.1 cents, “yes” pays 8.9 cents. Your capital is then locked up for the next 7 months waiting for the market to resolve. Whoever’s correct gets $1.
If you believe the likelihood of a pandemic is 0.1% and want to sweep the order book to reflect that probability, it’ll currently cost you about $7M to collateralize all those shares. If correct, you stand to gain about $160k in 7 months—a measly 2% gain, far worse than you’d be likely to do buying and holding an index fund or even T-bills. It is below the risk-free rate. It is a worse return than the safest investment you can make.
So there’s no financial incentive to more than marginally participate for small change, and that’s discounting the possibility of insider corruption at WHO being used to generate a “yes” resolution despite no pandemic. Likely? Probably not, hard for me to quantify. Possible? It only takes one person under acute financial pressure at WHO. But this tail risk matters a lot in parsing the resolution criteria. Not that this really matters, since we’ve established there’s no meaningful financial incentive to bet on “no.”
Remember the strong cultural aspect to Polymarket. It’s regarded as a bad thing in the progressive academic circles. Particularly markets on bad outcomes like a pandemic. Whether or not you agree, it means that the experts who know most about it this topic are not only financially disincentivized from buying shares of “no,” they’re also culturally disincentivized from participating on either side of the trade. Heavy forces of adverse selection influence who participates in this market.
If you didn’t already understand this about prediction markets yet believed they are benefitting global epistemics, then does this lead you to revise your opinion? If you didn’t know it requires deep financial sacrifice to correct the prediction market probability downward, does learning that this is the case make you take it less seriously? Why or why not?
Doing a bit of trading on Manifold (I’ve been active for around 3 months) has drilled this into me well enough that the general principle of “low enough/high enough markets generally don’t go to their true probability” seems obvious, and that’s after I gained the theoretical knowledge that these things happen from the Jesus Christ returns market. If people on LessWrong are taking nearly all prediction markets purely at their face value, I think that wouldn’t be good, but I don’t think they are.
I will note that Polymarket does have a 4% annualized holding reward (basically 4% interest on your market positions, the rate is variable), so the potential gain isn’t quite as bad as you state. With this in mind, people not betting down a 9% probability does seem meaningfully different to me than if it were five or less percent.
Yes, this. Markets seem to basically never go below 3% or so, but 9% is very meaningfully different from that, and I don’t think is adequately explained by just the cost of holding capital for this.
I will note that Polymarket does have a 4% annualized holding reward (basically 4% interest on your market positions, the rate is variable), so the potential gain isn’t quite as bad as you state. With this in mind, people not betting down a 9% probability does seem meaningfully different to me than if it were five or less percent.
This is helpful information, thank you. That does meaningfully change how we should interpret the probabilities, conditional on having little to no concern about the risk of fraud/reflexivity.
If you believe the likelihood of a pandemic is 0.1% and want to sweep the order book to reflect that probability, it’ll currently cost you about $7M to collateralize all those shares. If correct, you stand to gain about $160k in 7 months—a measly 2% gain, far worse than you’d be likely to do buying and holding an index fund or even T-bills. It is below the risk-free rate. It is a worse return than the safest investment you can make.
I believe that Kalshi and sometimes Polymarket gives about 3% annualized on investments, roughly the risk free rate, to encourage long term investments. So the actual returns would be somewhat larger.
Incentives-wise, you could correct for this with shorter resolution times, right? Like instead of resolving a pandemic question in 7 months, ask if there are more entirely new cases outside of South America 2 months from now compared to 1 month from now. (2 months is still kind of long, so with clever question engineering one could probably find something even better that is relevant to what we’re interested in.)
The strain making the news was picked up by German tourists watching birds in a garbage dump in Ushuaia, Argentina.
A pandemic-grade strain of Andes virus has just emerged from Ushuaia. The German tourists were potentially the first victims, but the Hondius passengers and Ushuaian residents are all about to become the first victims of a coming Andes virus pandemic, which will start in Ushuaia and the immediate contacts of the Hondius passengers, radiating out from there.
The German tourists picked up a strain of Andes virus that already had pandemic potential before they were infected by it. But Ushuaians, unlike German tourists, don’t frequent their own garbage dumps and have no meaningful exposure to the rodents that are its reservoire, so they aren’t exposed to it. For this reason, the rest of the world gets pandemic Andes virus before Ushuaia itself.
As (2), except Ushuaia is about to have a mass outbreak of Andes virus and get quarantined. The quarantine works on Ushuaia, protecting the rest of South America, but because the tourists had already spread it globally, it can’t be contained. Ushuaia and the rest of the world get rapid growth in Andes virus cases, but South America is temporarily isolated from the pandemic.
The German tourists picked up a strain that mutated within their bodies on on board the Hondius to be much more human transmissible than the virus the tourists picked up in Ushuaia. In this case, the rest of the world would be getting exposed to pandemic-potential hantavirus before Ushuaia and the rest of South America, resulting in the rest of the world having more cases over the next few months than Ushuaia.
If we were about to have an Andes virus pandemic, scenario 1 seems massively morely likely to me than 2, 3, or 4. But in that case, we’d see cases mounting in South America in tandem with or at a faster pace than the rest of the world.
I do think that seeing enough de novo cases of Andes virus outside existing quarantine centers that infections transmitted at close quarters on the Hondius can’t account for it would be a massive, terrifying update for me about the potential for an Andes virus pandemic. It’s just that growth of Andes virus cases in South America probably shouldn’t update you away from the risk of a global Andes virus pandemic.
I don’t think that’d work. Fundamentally, you are locking up capital for some returns; and so your opportunity cost is what you’d have gotten in the stock market. Unless you change the payout structure, or do something clever, I think you’ll be stuck with not being able to trust probabilities below (or above the complement of) market returns.
What I meant was shorter lockup/resolution period makes the market still accurate for lower probability estimates. You’re right that below market returns it becomes unable to distinguish anything anymore.
With pandemic risks, it’s a bit stupid if the market fails to differentiate between 0% and 8%, but if you can get that down towards it being able to differentiate between “anywhere below 3%” and “likely 3+%”, then that’s starting to become useful.
Well, for a real money prediction market, market returns should prevent you from distinguishing anything below ~7%. Short resolution criteria help, but would still be compared to how liquid an investments in the regular stock market can be.
For a play money market like manifold, I’d expect shorter resolutions to have more impact.
It’s 7% per year (or 9% per year or whatever the figure is, the important part being PER YEAR). So if we resolve in 2 months, surely the opportunity costs are lower and that will let us get more precision on lower probabilities?
Important point towards low pandemic chance to me is that we saw it occur on a cruise ship. Those are tightly packed sure, but the chances that it originally started on the ship itself seems rather low. Assuming it is a new mutation, it either
Mutated in rodents that were already on the ship
Mutated in rodents on land that later slipped into the ship before sailing off.
Infected someone before they boarded.
Exceedingly unlikely but maybe it was planted there as a bioterrorism or something.
Both 2 and 3 pose the question of why the infection didn’t spread anywhere else. The infectee didn’t stay at a hotel the night before or spend time with anyone on land right before? The rodent didn’t happen to be near anyone on land and spread to them before it boarded? It’s not impossible, and maybe the incubation time was just long enough that it only started while on board the cruise, but that takes away from the likelyhood. Likewise it’s possible maybe some mouse or rat or whatever onboard just happened to be one that developed a new mutation but again, really unlikely.
That being said, we do have to face one issue. If Y is a possible (even if unlikely) cause of X, then observing X to be true means Y is way more possible than a non X world. We live in a world where hantavirus seems to have infected a significant amount on a cruise ship, thus while unlikely to be a major mutation that spreads human to human easily, it is something to be a bit concerned about.
We know it’s not a new mutation, that’s been confirmed. We also know where it started: two of the cruise ship passengers got infected while bird watching in a garbage dump in Ushuaia, Argentina. Argentina is one of the two countries that harbors the specific rodent species that are the main reservoires for Andes virus.
The clear sequence of events is that the tourists went to a rodent-infested location neither tourists nor locals frequent. This unusual decision exposed them to a much higher risk of infection from virus contained in aerosolized rat droppings in that setting. They were infected. Then they boarded the cruise ship, which sailed on. After an incubation period (which can last up to 6 weeks for hantavirus), the close quarters of the cruise ship enabled it to be transmitted to a limited number of other passengers, as we’ve seen in the limited number of reported cases of human-to-human transmission in the past, which took place at events like parties or among nuclear family members at home.
This incident has confirmed that hantavirus can undergo human-to-human transmission if the host is very sick and other people are in close quarters with them. It doesn’t appear to require physical contact in that circumstance. As I understand it, the human-to-human transmission cases have typically occurred when the patient is symptomatic, though we don’t know for sure about the risk of asymptomatic spread. Overall, however, it looks to me like it takes a rare, though not impossible, combination of events to result in human to human transmission that amount to it being too inefficient to pose even a high single digit risk of a pandemic.
I’d also add that, unlike COVID, the extreme danger of hantavirus means that if it were to show global outbreaks, the social reaction would be immediate and extreme. People would not be flying around and going out in public without extreme precautions if there was a 40% CFR virus with pandemic-level transmissibility. They’d be demanding lockdown. COVID-19 was in the perfect sweet spot for triggering a pandemic, being too low-risk on a case by case basis for the general public to instantly and uniformly lock down. Hantavirus is a different beast—to inefficient to be likely to transmit effectively, and too deadly to provoke the same cavalier behavior.
I continue to think Hantavirus becoming a pandemic this year is much less than 5%. There’s a chance the opportunity to make 10x gains on the prediction market leads a WHO insider to falsely characterize it as a pandemic in what technically counts as an “official communication.”
Andes virus, the strain of hantavirus that provoked the prediction market, is endemic in two countries. It causes hundreds of cases per year. The strain that infected the Hondius appears to be wild type. In other words, unlike COVID-19, it doe not have new, recent mutations that make it unusually efficient at infecting humans compared to its historic baseline.
If this virus had pandemic potential, I believe we’d have seen it causing the same kinds of mass outbreaks in local villages and towns that we saw in Wuhan at the start of COVID. Argentinian and Chilean villages have the kinds of rodents that spread it efficiently. If we aren’t getting mass outbreaks there, why think we’ll get mass outbreaks in regions where that rodent reservoir is absent?
None of this means we shouldn’t take it seriously. A 40% CFR hantavirus pandemic would be one of the worst catastrophes in human history. I just think that outcome is highly unlikely.
I’m not sure people on LessWrong entirely understand how Polymarket works on a mechanical level, and suspect it leads them to grossly misinterpret these markets.
Currently, the “hantavirus pandemic 2026?” market is at 8.9 cents. To participate, you and counterparty jointly collateralize a new $1 share—in this case, ”no” pays 91.1 cents, “yes” pays 8.9 cents. Your capital is then locked up for the next 7 months waiting for the market to resolve. Whoever’s correct gets $1.
If you believe the likelihood of a pandemic is 0.1% and want to sweep the order book to reflect that probability, it’ll currently cost you about $7M to collateralize all those shares. If correct, you stand to gain about $160k in 7 months—a measly 2% gain, far worse than you’d be likely to do buying and holding an index fund or even T-bills. It is below the risk-free rate. It is a worse return than the safest investment you can make.
So there’s no financial incentive to more than marginally participate for small change, and that’s discounting the possibility of insider corruption at WHO being used to generate a “yes” resolution despite no pandemic. Likely? Probably not, hard for me to quantify. Possible? It only takes one person under acute financial pressure at WHO. But this tail risk matters a lot in parsing the resolution criteria. Not that this really matters, since we’ve established there’s no meaningful financial incentive to bet on “no.”
Remember the strong cultural aspect to Polymarket. It’s regarded as a bad thing in the progressive academic circles. Particularly markets on bad outcomes like a pandemic. Whether or not you agree, it means that the experts who know most about it this topic are not only financially disincentivized from buying shares of “no,” they’re also culturally disincentivized from participating on either side of the trade. Heavy forces of adverse selection influence who participates in this market.
If you didn’t already understand this about prediction markets yet believed they are benefitting global epistemics, then does this lead you to revise your opinion? If you didn’t know it requires deep financial sacrifice to correct the prediction market probability downward, does learning that this is the case make you take it less seriously? Why or why not?
Doing a bit of trading on Manifold (I’ve been active for around 3 months) has drilled this into me well enough that the general principle of “low enough/high enough markets generally don’t go to their true probability” seems obvious, and that’s after I gained the theoretical knowledge that these things happen from the Jesus Christ returns market. If people on LessWrong are taking nearly all prediction markets purely at their face value, I think that wouldn’t be good, but I don’t think they are.
I will note that Polymarket does have a 4% annualized holding reward (basically 4% interest on your market positions, the rate is variable), so the potential gain isn’t quite as bad as you state. With this in mind, people not betting down a 9% probability does seem meaningfully different to me than if it were five or less percent.
Yes, this. Markets seem to basically never go below 3% or so, but 9% is very meaningfully different from that, and I don’t think is adequately explained by just the cost of holding capital for this.
This is helpful information, thank you. That does meaningfully change how we should interpret the probabilities, conditional on having little to no concern about the risk of fraud/reflexivity.
I believe that Kalshi and sometimes Polymarket gives about 3% annualized on investments, roughly the risk free rate, to encourage long term investments. So the actual returns would be somewhat larger.
Incentives-wise, you could correct for this with shorter resolution times, right? Like instead of resolving a pandemic question in 7 months, ask if there are more entirely new cases outside of South America 2 months from now compared to 1 month from now. (2 months is still kind of long, so with clever question engineering one could probably find something even better that is relevant to what we’re interested in.)
The strain making the news was picked up by German tourists watching birds in a garbage dump in Ushuaia, Argentina.
A pandemic-grade strain of Andes virus has just emerged from Ushuaia. The German tourists were potentially the first victims, but the Hondius passengers and Ushuaian residents are all about to become the first victims of a coming Andes virus pandemic, which will start in Ushuaia and the immediate contacts of the Hondius passengers, radiating out from there.
The German tourists picked up a strain of Andes virus that already had pandemic potential before they were infected by it. But Ushuaians, unlike German tourists, don’t frequent their own garbage dumps and have no meaningful exposure to the rodents that are its reservoire, so they aren’t exposed to it. For this reason, the rest of the world gets pandemic Andes virus before Ushuaia itself.
As (2), except Ushuaia is about to have a mass outbreak of Andes virus and get quarantined. The quarantine works on Ushuaia, protecting the rest of South America, but because the tourists had already spread it globally, it can’t be contained. Ushuaia and the rest of the world get rapid growth in Andes virus cases, but South America is temporarily isolated from the pandemic.
The German tourists picked up a strain that mutated within their bodies on on board the Hondius to be much more human transmissible than the virus the tourists picked up in Ushuaia. In this case, the rest of the world would be getting exposed to pandemic-potential hantavirus before Ushuaia and the rest of South America, resulting in the rest of the world having more cases over the next few months than Ushuaia.
If we were about to have an Andes virus pandemic, scenario 1 seems massively morely likely to me than 2, 3, or 4. But in that case, we’d see cases mounting in South America in tandem with or at a faster pace than the rest of the world.
I do think that seeing enough de novo cases of Andes virus outside existing quarantine centers that infections transmitted at close quarters on the Hondius can’t account for it would be a massive, terrifying update for me about the potential for an Andes virus pandemic. It’s just that growth of Andes virus cases in South America probably shouldn’t update you away from the risk of a global Andes virus pandemic.
I don’t think that’d work. Fundamentally, you are locking up capital for some returns; and so your opportunity cost is what you’d have gotten in the stock market. Unless you change the payout structure, or do something clever, I think you’ll be stuck with not being able to trust probabilities below (or above the complement of) market returns.
What I meant was shorter lockup/resolution period makes the market still accurate for lower probability estimates. You’re right that below market returns it becomes unable to distinguish anything anymore.
With pandemic risks, it’s a bit stupid if the market fails to differentiate between 0% and 8%, but if you can get that down towards it being able to differentiate between “anywhere below 3%” and “likely 3+%”, then that’s starting to become useful.
Well, for a real money prediction market, market returns should prevent you from distinguishing anything below ~7%. Short resolution criteria help, but would still be compared to how liquid an investments in the regular stock market can be.
For a play money market like manifold, I’d expect shorter resolutions to have more impact.
It’s 7% per year (or 9% per year or whatever the figure is, the important part being PER YEAR). So if we resolve in 2 months, surely the opportunity costs are lower and that will let us get more precision on lower probabilities?
Or am I missing something?
Ah, duh! Thanks, I think you are right.
Do you have examples of prominent people on LessWrong believing it’s greater than 5%?
Important point towards low pandemic chance to me is that we saw it occur on a cruise ship. Those are tightly packed sure, but the chances that it originally started on the ship itself seems rather low. Assuming it is a new mutation, it either
Mutated in rodents that were already on the ship
Mutated in rodents on land that later slipped into the ship before sailing off.
Infected someone before they boarded.
Exceedingly unlikely but maybe it was planted there as a bioterrorism or something.
Both 2 and 3 pose the question of why the infection didn’t spread anywhere else. The infectee didn’t stay at a hotel the night before or spend time with anyone on land right before? The rodent didn’t happen to be near anyone on land and spread to them before it boarded? It’s not impossible, and maybe the incubation time was just long enough that it only started while on board the cruise, but that takes away from the likelyhood. Likewise it’s possible maybe some mouse or rat or whatever onboard just happened to be one that developed a new mutation but again, really unlikely.
That being said, we do have to face one issue. If Y is a possible (even if unlikely) cause of X, then observing X to be true means Y is way more possible than a non X world. We live in a world where hantavirus seems to have infected a significant amount on a cruise ship, thus while unlikely to be a major mutation that spreads human to human easily, it is something to be a bit concerned about.
We know it’s not a new mutation, that’s been confirmed. We also know where it started: two of the cruise ship passengers got infected while bird watching in a garbage dump in Ushuaia, Argentina. Argentina is one of the two countries that harbors the specific rodent species that are the main reservoires for Andes virus.
The clear sequence of events is that the tourists went to a rodent-infested location neither tourists nor locals frequent. This unusual decision exposed them to a much higher risk of infection from virus contained in aerosolized rat droppings in that setting. They were infected. Then they boarded the cruise ship, which sailed on. After an incubation period (which can last up to 6 weeks for hantavirus), the close quarters of the cruise ship enabled it to be transmitted to a limited number of other passengers, as we’ve seen in the limited number of reported cases of human-to-human transmission in the past, which took place at events like parties or among nuclear family members at home.
This incident has confirmed that hantavirus can undergo human-to-human transmission if the host is very sick and other people are in close quarters with them. It doesn’t appear to require physical contact in that circumstance. As I understand it, the human-to-human transmission cases have typically occurred when the patient is symptomatic, though we don’t know for sure about the risk of asymptomatic spread. Overall, however, it looks to me like it takes a rare, though not impossible, combination of events to result in human to human transmission that amount to it being too inefficient to pose even a high single digit risk of a pandemic.
I’d also add that, unlike COVID, the extreme danger of hantavirus means that if it were to show global outbreaks, the social reaction would be immediate and extreme. People would not be flying around and going out in public without extreme precautions if there was a 40% CFR virus with pandemic-level transmissibility. They’d be demanding lockdown. COVID-19 was in the perfect sweet spot for triggering a pandemic, being too low-risk on a case by case basis for the general public to instantly and uniformly lock down. Hantavirus is a different beast—to inefficient to be likely to transmit effectively, and too deadly to provoke the same cavalier behavior.