Dr John Seo is co-founder and a managing director at Fermat Capital Management, LLC. He has over 30 years’ experience in fixed income bond and derivatives trading and has been active in the Insurance-Linked Securities (“ILS”) market for over 25 years. Prior to forming Fermat with his brother Nelson in 2001, Dr Seo was senior trader in the Insurance Products Group at Lehman Brothers, an officer of Lehman Re, and a state-appointed advisor to the Florida Hurricane Catastrophe Fund. Dr. Seo’s work in catastrophe funds was featured in a cover article for the New York Times Magazine (‘In Nature’s Casino’ by Michael Lewis, 26 August 2007), and he has also testified before US Congress as an expert on the catastrophe bond market (‘Hearings from the 110th Congress’, 6 September 2007). Dr Seo holds a PhD in Biophysics from Harvard University and a BS in Physics from MIT. He is based in Connecticut.
The part about the catestrophic risk/market and insurance/derivative markets converging on the same values was very interesting. This would be consistent with the efficient market hypothesis. But it seems he's carved out an opportunity from an inefficiency.
The point about climate change was also very interesting. It seems you could get a noisy estimate of the expected increase in temperature based off of that.
Of course, there are also prediction markets, but unfortunately there's not a huge amount of money in those, which seems like a huge waste. My vision for the future of academia is replication markets on studies, and then replication markets on hypotheses. We would expect with enough money involved, there is a huge incentive to get these right. Since hypotheses are conditional on studies, you would expect hypotheses markets to move depending on newly published research. This would provide a measure of a persons academic contribution. If you could drastically move a market with your research, you're making important contributions. If you publish an article and the replication market falls to 5% and the hypothesis market is unchanged, you're producing garbage. There's already been some work on using machine learning to predict replicability. Seems like replication markets could be run mostly by AI. Seems like a goldmine of human knowledge. Much better than "well he went to Princeton and it's a good journal, so its probably true!" We've learned that approach is egregiously bad.
This obviously has impact for longtermist EAs. An argument has been made that AI fast takeoff is not properly accounted for in the market. Eliezer has argued it can't be. Personally, I invest in leveraged semiconductor funds. The goal is partly to make money, but partly to balance myself psychologically. The anxiety from x-risk can be accounted for by making a bunch of money to the point where the anxiety and excitement from getting rich theoretically balance out.
The part about the catestrophic risk/market and insurance/derivative markets converging on the same values was very interesting. This would be consistent with the efficient market hypothesis. But it seems he's carved out an opportunity from an inefficiency.
The point about climate change was also very interesting. It seems you could get a noisy estimate of the expected increase in temperature based off of that.
Of course, there are also prediction markets, but unfortunately there's not a huge amount of money in those, which seems like a huge waste. My vision for the future of academia is replication markets on studies, and then replication markets on hypotheses. We would expect with enough money involved, there is a huge incentive to get these right. Since hypotheses are conditional on studies, you would expect hypotheses markets to move depending on newly published research. This would provide a measure of a persons academic contribution. If you could drastically move a market with your research, you're making important contributions. If you publish an article and the replication market falls to 5% and the hypothesis market is unchanged, you're producing garbage. There's already been some work on using machine learning to predict replicability. Seems like replication markets could be run mostly by AI. Seems like a goldmine of human knowledge. Much better than "well he went to Princeton and it's a good journal, so its probably true!" We've learned that approach is egregiously bad.
This obviously has impact for longtermist EAs. An argument has been made that AI fast takeoff is not properly accounted for in the market. Eliezer has argued it can't be. Personally, I invest in leveraged semiconductor funds. The goal is partly to make money, but partly to balance myself psychologically. The anxiety from x-risk can be accounted for by making a bunch of money to the point where the anxiety and excitement from getting rich theoretically balance out.
This was very interesting. Thank you!