Executive summary: This exploratory post argues that while standard expected utility theory recommends fully concentrating charitable donations on the highest-expected-impact opportunity, a pragmatic Bayesian approach—averaging across uncertain models of the world—can justify some degree of diversification, particularly when model uncertainty or moral uncertainty is significant.
Key points:
Standard expected utility theory implies full concentration: Under a simple linear model, maximizing expected impact requires allocating all resources to the charity with the highest expected utility, leaving no room for diversification.
This approach is fragile under uncertainty: Small updates in beliefs can lead to complete switches in preferred charities, making the strategy non-robust to noise or near-ties in effectiveness estimates.
Diversification in finance relies on risk aversion, which is less defensible in charitable giving: Unlike financial investments, diversification in giving can’t be easily justified by volatility or utility concavity, as impact should be the sole goal.
Introducing model uncertainty enables a form of Bayesian diversification: By treating utility estimates as conditional on uncertain world models (θ), and averaging over these models, one can derive an allocation that reflects the probability of each charity being optimal across possible worldviews.
This yields intuitive and flexible allocation rules: Charities get funding proportional to their chance of being the best in some plausible world; clearly suboptimal options get nothing, while similarly promising ones are treated nearly equally.
The method is ad hoc but practical: Although the choice of which uncertainties to “pull out” is arbitrary and may resemble hidden risk aversion, the author believes it aligns better with real-world epistemic humility and actual donor behavior than strict maximization.
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, and contact us if you have feedback.
Executive summary: This exploratory post argues that while standard expected utility theory recommends fully concentrating charitable donations on the highest-expected-impact opportunity, a pragmatic Bayesian approach—averaging across uncertain models of the world—can justify some degree of diversification, particularly when model uncertainty or moral uncertainty is significant.
Key points:
Standard expected utility theory implies full concentration: Under a simple linear model, maximizing expected impact requires allocating all resources to the charity with the highest expected utility, leaving no room for diversification.
This approach is fragile under uncertainty: Small updates in beliefs can lead to complete switches in preferred charities, making the strategy non-robust to noise or near-ties in effectiveness estimates.
Diversification in finance relies on risk aversion, which is less defensible in charitable giving: Unlike financial investments, diversification in giving can’t be easily justified by volatility or utility concavity, as impact should be the sole goal.
Introducing model uncertainty enables a form of Bayesian diversification: By treating utility estimates as conditional on uncertain world models (θ), and averaging over these models, one can derive an allocation that reflects the probability of each charity being optimal across possible worldviews.
This yields intuitive and flexible allocation rules: Charities get funding proportional to their chance of being the best in some plausible world; clearly suboptimal options get nothing, while similarly promising ones are treated nearly equally.
The method is ad hoc but practical: Although the choice of which uncertainties to “pull out” is arbitrary and may resemble hidden risk aversion, the author believes it aligns better with real-world epistemic humility and actual donor behavior than strict maximization.
This comment was auto-generated by the EA Forum Team. Feel free to point out issues with this summary by replying to the comment, and contact us if you have feedback.