Hedging Grants And Donations
Epistemic status: speculative ideas.
Core Idea
Large philanthropic institutions might be better off buying insurance for some of the projects they fund: they shouldn’t be risk neutral with respect to money because money has decreasing returns at the scale they are working at.
Moreover, this would have two major additional benefits:
It would create natural prediction markets about the success rate of various projects, giving the community valuable information about them;
It would make it easier for people at these organizations to face the perspective of making loosing bets.
A Simple Example
Let’s say you are a large philanthropic organization with $8B at your disposal. Two potential use of this money seem promising to you:
Use $6B to fund a large campaign to pass a piece of legislation that makes it harder for the US president to use nukes. You estimate that without your intervention, the piece of legislation has a 0% chance of passing, and with your intervention, it has a 75% chance of passing. If it is passed, you expect that it reduces the probability of an existential catastrophe by 0.2% in the long run.
Wait until crunch time for AGI, and then spend $6B to help the team which you think has the highest chance of developing safe AGI. You expect this investment to reduce the probability of an existential catastrophe by 0.1%.
You expect money spent beyond $6B on either projects to have drastic diminishing returns, and projects other than these two to be much worse. You also expect it to be infeasible to raise enough money to finance both projects.
What is the right course of action?
I think the right course of action is to buy insurance:
First, convince an insurance company that the probability of the piece of legislation being passed, if you do project A, is 75%. Then they should agree to the following deal: you give them $2B now, and they give you $6B if the legislation doesn’t pass despite your best effort (their expect value is $2B − 0.25 x $6B = $500M). Use the remaining $6B to execute plan A. If the piece of legislation doesn’t pass, use the money the insurance company gave you to execute plan B.
This raises your expected existential catastrophe reduction to 0.175%, up from 0.15% if you chose A, and up from 0.1% if you chose B.
Note: all incentives are aligned with you doing your best to make the legislation pass: the “best effort” clause of the contract makes it mandatory for you to spend $6B on campaign-related expenses. If you made a poor use of these $6B, you would end up with reducing existential catastrophe probability by only 0.1% (by carrying out project B), and would lose the $2B insurance premium in the process.
The General Setup
If a project requires spending that is obviously only related to this project, and if this project changes the probability of a short-term objective outcome, then philanthropic institutions should probably try to get insurance for this project.
Additional Benefit 1: Insurance Markets Are Predictions Markets and Will Give You Valuable Information
This is not legal advice. I don’t have the knowledge required to evaluate if the law would allow this.
It seems to me that it would be possible to express your interest in an insurance against failing to pass a piece of legislation, and then have people making you insurance offers. The price they are willing to give you already gives you an idea of what the success probability of project A is in the simple example above.
You can turn this into a proper market by giving people 2 billion contracts which say “if I carry out project A, I will give you $1, and you will owe me $3 if the project fails”. The expected value of this contract if you carry out project A should be 1 − 3 x 0.75 = 0.5, so they should each sell at $0.5. If they sell below this price, the project is likely to fail, if they sell above it, the project is likely to succeed. A marketplace could allow people to buy and sell these contracts, with actual money being transferred only if you decide to carry out A (just like in crowdfunding projects, where you only pay if the project happens).
The price of such a market gives you a crowd estimate of the probability of success of the piece of legislation, which is an extremely valuable piece of information.
Note: This exact financial contract is probably not feasible, but I expect some related financial contract to be feasible.
Note: Insurance contract prices will be impacted not only by the probability of the event, but also by its correlation with the overall market, and by the irrationality of the market. The former means that you shouldn’t use this system when the event has a strong correlation with the overall market. I’m unsure what to make of the latter.
Additional Benefit 2: It Lowers Irrational Risk Aversion
I would like to be responsible for leading a $6B project with a failure probability of 25%. I would probably prefer leading the project if the organization only lost $2B if the project failed. If this feeling is shard by many, it would mean that insurance would make it easier:
given a set of leaders at the organizations, to make them carry out risky projects
given a set of projects you want to carry out, to find talented individuals to lead them
Is Insurance Valuable for Small Donors?
The main argument doesn’t apply to small donors because there are no decreasing returns of money when you are responsible for a tiny fraction of the projects you are contributing too.
But the two additional benefits still apply:
Insurances for common charities might allow good forecast of the effectiveness of current actions, which will be measured in the future. For example, AMF would probably gain some valuable information if a portion of its donors insured themselves against future studies showing that AMF’s action wasn’t as effective as claimed (though the details of the insurance contract might be tough to work out, since AMF would probably have incentives to make the insurance price higher or lower).
Small donors can be risk averse. You might feel bad about giving money to GPI (Global Priorities Research) because you feel like the probability of them finding anything interesting in the next five years is too small. Then you would benefit from using insurance contracts: give money to GPI, and pay an insurance premium such that if GPI doesn’t find anything interesting in the next five years, the insurance gives to AMF (Against Malaria Foundation) the amount of money. Then your minimum impact will be the same as if you directly gave money to AMF. The expected value of your impact will be close but smaller than if you gave money to GPI without paying an insurance premium (if you believe the expected impact per dollar of GPI is higher than AMF’s). In any case, if there are enough actors on the market competing on insurance prices, your expected impact won’t drop bellow “giving your money to AMF directly”.
Interesting idea, but I forsee several challenges in implementation:
First, few organizational outcomes are truly binary—it is possible that the organization gets some, but not all, of its objectives achieved, in which case there is going to be litigation about whether the actual outcome is an insured loss.
Second, it is going to be expensive for an insurance company to develop an accurate sense of the odds of success, especially because many of the relevant pieces of information are under the control of the organization and may be very difficult to measure without organizational influence. If I were the insurer, I’d require a significant application fee just to provide a quote, and I would quote very conservatively.
Third, incentives can change. Even if the insurer believes your assertion about preferences, those preferences could change over time and you could then have an incentive to “throw” the first project. Detecting failure to provide “best efforts” is challenging and uncertain. I think the workaround for this is for the insurance company to require significant co-insurance—e.g., only half of the loss from the failed initiative is covered. That gives the organization a much more concrete sense of skin in the game than its mere assertions that it would prefer success to collecting the insurance payout.
Finally, the hypothetical scenario (in which you don’t seem to have any good alternative use for $2B) is fairly unlikely. That doesn’t mean that insurance would have no use cases, only that they may be limited.
One interesting possible application would be having different EA cause areas potentially “insure” each other. E.g., if animal-welfare people want to try a high-risk, mega-high-reward intervention but is having a hard time tolerating the idea of losing some high-value and fairly safe options if the intervention fails, groups from another cause area might be willing to “insure.” As opposed to an insurance company, other EAs are going to be better at developing an accurate sense of the odds of success and assessing whether the insured’s interests are likely to change.
Moreover, the insurance “payout” would likely still have good value for the “insuring” EAs—even if I would not have donated to animal-welfare causes in the first instance, the fulfillment of high value options in that area still brings me utilons. Likewise, if you’re a animal-welfare person, the payment of an insurance “premium” to global health/development still generates utilons in your book, even if not as many as applied to animal welfare.
nit: the per-mille symbol (‰) is easily confused for the percent symbol (%), and isn’t well-known. I think this would be clearer if you stuck to percent (‘0.1%’ instead of ‘1‰’ etc)
Agreed, I’ll edit the post.