On the EA Corner discord server, various participants, most notably @wolframhead, tossed around the following idea: Some people of an EA bent may want to donate vast amounts of money, huge percentages of high salaries, but be held back in case they turn out to need that money as savings later. A health crisis, their job market failing them, or some other unforeseen but expensive eventuality could make them regret prior profligate giving once they need a runway for a few months with no paychecks, run into a pay cut, or have to contend with huge surprise bills. But of course not all of these people will actually experience such a crisis even though all of them run some risk of doing so. If there was a way for them to get their money back from the charity they donated to conditional on an emergency cropping up, the charity would be able to keep most such conditional funds even not taking into account the possibility that they could make interest on the capital in the time between donation and refund. Overall charity revenue would probably go up if this "fractional reserve" model were feasible. Here's a writeup of my understanding of the topic; the EA Corner has open membership for anyone who wants to go view the conversation in its original form.
1. Possible Structures
A - Charities could directly provide this service. This would require them to deal with the overhead of assessing emergency needs in their donors, which is probably not their core competency. The charity would operate under uncertainty about not only its future budget but also its current budget, as the rug could be pulled out from under them on sizable donations at any time. It'd also probably affect how EA money flowed to charities that managed this versus ones that did not, as a charity that has a refund possibility is "cheaper" to donate to than one that does not. The concept of "room for more funding" could possibly limit the strangling effect on charities that didn't participate to a point, but only to a point, and making lots of charities spend on overhead and uncertainty-tolerant budget forecasts just to remain competitive donation targets seems like a deadweight loss.
B - A middleman organization could form (or it could grow out of an existing meta-charity like Givewell). Such an organization would mostly just pass on donations to the intended recipients - you'd send them a thousand dollars marked for AMF, and AMF would get most of those dollars to spend on AMF things with no worries about paying it back. Some of what passed through the middleman organization would float as liquid assets and be kicked back to donors who needed it, the way a bank doesn't keep enough cash on hand for everyone to withdraw their savings all at once but can handle a normal rate of withdrawals without a blip. There would be some loss in the form of overhead for this organization, but plausibly less than the gain in effective donations generated by the insurance scheme.
C - A tontine-like structure with less formal scaffolding could handle it. Groups of EAs who know and trust each other - and ideally have comparable budgets - could all pool donations of the same size, keep back that amount or perhaps twice that amount held by someone responsible in the group, and kick that amount back to whoever needs it first as assessed by other group members. A person with more money to throw around could belong to several of these. Relying on interpersonal rather than institutional honesty would reduce access to the model for people who are less socially in with other EAs and might make it harder to solidly verify people's emergency conditions by strict rubrics.
2. Probable Obstacles
A - Assorted regulatory complexity. Are any of the above structures legal? How will they interact with taxes, with nonprofit status, with whatever laws govern insurance even though (since in the vision I describe above payouts never exceed pay-ins) it's technically not insurance? Does it look like a tax dodge (is it a tax dodge)? Is it sort of like a donor-advised fund? What are the rules about those and which apply? How would you explain any of this to an accountant or an auditor without sounding insane?
B - Adverse selection. While anyone can worry about a trip to the hospital or a sudden firing, people who are more worried about these things might have more real reason to be worried, carrying elevated risk; expectations about payback rates based on baseline statistics about how common these things are may not be reliable, and a middleman that didn't account enough for that could wind up without enough cash on hand to return money to all its distressed donors, or a charity which provided the payback service in-house could try to budget its operations based on a guess that turns out too optimistic. Insurance companies have some institutional knowhow about how to handle this but it would increase overhead a lot.
C - Theft and fraud. While people presumably trust the charities they donate to, a donation the return of which could be debated based on assessed need would present an incentive for whoever held the bag to assess the need unsympathetically. A middleman organization or a private individual holding money for a tontine-style structure would face temptation to run off with the cash, too, and it might not be obvious this was happening right away if everyone who donated through them happened to be lucky and not expect the money back for an extended period of time. In the opposite direction, people might fake serious catastrophes to get their money back even if their use case for the money was not intended to be one that got a payback - this wouldn't be as appealing as normal insurance fraud, since you wouldn't get back more money than you put in in the first place, but it could get to be a problem if (say) a deceased donor's beneficiaries wanted to puff up their inheritances, or a third party believed they could intercept the money by posing as a donor in distress.
D - Bikeshedding and competition. If people have a lot of small differences of opinion about how this should work, that could support a large number of tontine-style setups, and different charities handling it in-house could "cheapen" their donations to various audiences of assorted inclinations, leaving each donor to decide if they prefer to deworm children under scheme 1 or net beds under scheme 2. But it'd sink a middleman organization to have only a handful of participants and be in competition with a lot of similar orgs, because it would run entirely on distribution of risk, and that requires a high population. Regular insurance companies have such a large pool of potential customers that they can manage splitting the pie; I don't know if the market for refundable donations is that big or could easily get that way (it's a weird idea that not even all preexisting EAs might want, for one thing).
A - Does something like this already exist? I would likely not know about it if it does (I don't handle my household finances or charity budget) but nobody else on the EA Corner mentioned "oh, you're talking about thus-and-such a program!", so perhaps it doesn't - or maybe there's a fledgeling one, or a poorly advertised sub-product belonging to some financial firm. Or maybe there's a way to squish something from its original shape into this shape without inventing new practices altogether.
B - Would this see enough use to benefit from risk pooling and to increase effective charity revenue? EAs move a fair amount of money already. While it would be nice if they endured less risk in so doing, the really high leverage of setting up fractional reserve charity options would be if it increased donations relative to the counterfactual. I don't have a sense of how large this space really is.
C - Are EA emergencies too correlated, e.g. are too many of us in a position to suffer from collapse in the tech sector for any EA techies to consider this employment insurance lite? We probably don't have correlated medical problems unless someone torches an EA Global, but if a bubble a lot of EAs work in popped, and every displaced software engineer or quant were each of them expecting five plus figures returned from their charities, that would capsize anything inadequately diversified in its path. Is there much use to a version that only refunds in case of medical emergency?
I learned the word "tontine" in a form different from the apparently central use case; the meaning I intend is a group savings/loan plan where (as an example of one way it can be structured) every one of N people in a group pays X per month into a shared pool, and every month a different member of the group gets N*X in a lump sum until N months have gone by.