Strongly upvoted for such a comprehensive answer, thank you Max! You've given me a lot to chew on.

You mention that

I see the most relevant application areas of this methodology in:

• Biorisk
• Economics
• Improving Institutional Decision-Making
• Climate Change
• Nuclear War
• AI Governance
• Longtermism

Are there any papers, writeups, blog posts etc I can check out to see how these techniques are applied "in the wild"? I'd be especially interested in writeups that show how simpler methods yield results deficient in some way, and how this methodology's results don't have those deficiencies. 

Hi Hugo, this is the metric: https://docs.google.com/document/d/1I0Ed8sYFLEKMlbPJG3SViqeE-POkDRASL0eE7fcjCn4/edit?usp=sharing

Likely to do with this; a little more discussion on that point here.

You might be interested in Gregory Lewis' person-affecting value of existential risk reduction CE estimate (Guesstimate model), which arrives at a mean 'cost per life year' of $1,500-$26,000 (mean $9,200) via this chain of reasoning. My sense is that it's a lot lower than even your pessimistic estimate mainly due to the person-affecting view constraint, but the takeaways still favor continued work & funding on reducing x-risk. Quoting Lewis:

[There] is a common pattern of thought along the lines of, "X-risk reduction only matters if the total view is true, and if one holds a different view one should basically discount it". Although rough, this cost-effectiveness Guesstimate suggests this is mistaken. ...

The comments section in that post surfaces a number of other x-risk CE estimates too. 80K Hours also has a (simpler) CE estimate. All of them seem pretty conservative.

Another remark is on discount rate, which you didn't seem to include in your post (maybe I missed it?). The discount rate effectively determines whether long- or near-termism is the best use of philanthropic resources is a post by professional cost-effectiveness modeller Froolow that explores this in more detail, using threshold analysis and assuming exponential discounting (although I suspect many people's actual discount rates including mine look more like Will MacAskill's, which is more lay-intuitive if not very mathematically nice).

Donation opportunities, yes. I'm not sure if donation opportunities in particular are something development economists look for; I'm not familiar with the literature. 

I broadly agree with the substance of your comment, I just admittedly find the tone off-puttingly abrasive ("delusional and arrogant" doesn't seem charitable), so I'll respectfully bow out of this exchange. 

Halstead and Hillebrandt didn't claim that a 4-person year research effort could discover the key to economic growth. Their claim is simply about finding good donation opportunities:

A ~4 person-year research effort will find donation opportunities working on economic growth in LMICs which are substantially better than GiveWell’s top charities from a current generation human welfare-focused point of view. 

My sense is such an effort might start from Hillebrandt's appendices, in particular appendix 4. The output of such an effort might look like one of Founders Pledge's reports (example; Halstead is a coauthor). 

In practice, "the charity that has the highest EV on the current margin" is more complicated than you may realize; see e.g. Section 1.3 of froolow's post on incorporating uncertainty analysis in cost-effectiveness modeling, showing how any of GiveWell's (older list of) top charities could be highest EV given reasonable assumptions:

and when:

I think it's also worth noting that GiveWell's CEAs don't actually calculate CE on the margin, and not the marginal impact of the donor's dollar but of all dollars generated by the donated dollar via leverage/funging (2nd bullet point in the Errors section); I was confused by this when I first tried to replicate (one column of) GW's AMF CEA. 

Another reason is just bets-hedging. It's probably unwise to put all one's eggs in one basket, not just due to theoretical considerations like moral uncertainty but operational considerations as well, like whether the organization can deliver on the forecasted impact next year (as someone directly impacted by the recent fiasco this is front-of-mind).

I'm also personally torn between EV-maxing and risk aversion. The former suggests donating to longtermist charities and the latter to GiveDirectly; I care that I have some impact as much as(?) I care about the opportunity cost of missing out on more impact. This is a little like how I think about personal investing, although in the latter case risk aversion is greater. 

Frankly this may just be me failing to find the right numbers or something, but I'd be curious to know if you yourself have identified any single charity you consider highest-EV on the margin (not historical EV), and what that EV number is (and preferably a link to how it's calculated).

I'd be curious to know as well, speaking as an FTX regranting program grantee.

HLI's research overview page mentions that they're planning to look into the following interventions and policies via the WELLBY lens; there is some overlap with what you mentioned:

Our search for outstanding funding opportunities continues at three levels of scale. These are set out below with examples of the interventions and policies we plan to investigate next.

Micro-interventions (helping one person at a time)

• Deworming programs
• Cataract surgery
• Cement flooring
• Friendship Bench
• Mental health apps

Meso-interventions (systemic change through specific policies) 

• Lead regulation
• Access to pain relief
• Immigration policy
• Psychedelic-assisted therapy

Macro-interventions (systemic change through the adoption of a wellbeing approach) 

• Advocacy for, and funding of, subjective wellbeing research
• Developing policy blueprints for governments to increase wellbeing