A review of GiveWell’s discount rate

Editorial note

This report was commissioned by GiveWell and produced by Rethink Priorities from June to July 2023. We revised this report for publication. GiveWell does not necessarily endorse our conclusions, nor do the organizations represented by those who were interviewed.

The primary focus of the report is to review GiveWell’s current formulation of its discount rate by recommending improvements and reinforcing justifications for areas that do not require improvement. Our research involved reviewing the scientific and gray literature, and we spoke with 15 experts and stakeholders.

We don’t intend this report to be Rethink Priorities’ final word on discount rates, and we have tried to flag major sources of uncertainty in the report. We hope this report galvanizes a productive conversation within the global health and development community about discounting practices in cost-effectiveness analyses. We are open to revising our views as more information is uncovered.

Executive summary

Notes on the scope and process of this project

This project aims to serve the dual purposes of reviewing GiveWell’s current approach to calculating its discount rate(s) to:

Provide recommendations to GiveWell on how its approach to discount rates could be improved.
Strengthen the justifications for its approach in cases where we do not recommend changes.

The direction of this project was mainly guided by our priors^[1] that a prioritized investigation into three aspects could potentially make the biggest difference to GiveWell’s discount rate:

A review of how other major organizations in the global health and development space (within and outside effective altruism) choose and justify their discount rates.
A review of GiveWell’s overall approach to calculating discount rates to determine:
1. Whether GiveWell should use a different overall calculation approach.
2. Whether GiveWell should think differently about discounting consumption vs. health outcomes.
A review of the pure time preference component of GiveWell’s discount rate.

We also reviewed several other components of the discount rate (consumption growth rate, compounding non-monetary benefits, temporal uncertainty), but decided to spend less time on those as we deemed it less likely to make major recommendations or expected it would be harder to make meaningful progress. Table 1 summarizes our recommendations for GiveWell’s discounting practices.

The majority of this report focuses on the discount rate used for consumption benefits, as this appears to be the “main” discount rate used by GiveWell,^[2] but we also discuss discounting of health benefits. We do not discuss discounting of costs in this report as (1) GiveWell’s cost-effectiveness models rarely involve discounting costs, and (2) our general impression is that the typical approach across organizations is to discount monetary costs and benefits equally and we have seen very little discussion of alternative approaches.^[3] A review of the shape of the utility functions^[4] used is also out of scope for this review. Moreover, we focus exclusively on temporal discounting.^[5] If the time frame is not specified, all discount rates expressed as percentages are annual. Due to the variety of existing opinions and approaches with respect to discount rates and a relative lack of consensus, we opted to approach this project from a perspective of figuring out whether there are any compelling reasons to change GiveWell’s current approach, rather than starting from scratch and coming up with a discount rate independently of the current approach.

Summary of recommendations

Table 1: Summary of Rethink Priorities’ recommendations for GiveWell’s discounting

Consideration	Current GiveWell choice	Rethink Priorities’ recommendation	Comments
Overall annual discount rate	4.0%	4.3% (if current inconsistent choice of η is kept)	The 0.3 percentage point increase is a result of an increased consumption growth rate estimate and a small change in the formula used to calculate the wealth effect. We recommend no other immediate changes. However, as mentioned in the report, our recommended discount rate is also contingent on whether GiveWell decides to use a consistent utility curvature η across applications. For example, if η = 1, the wealth effect would be 0% and the resulting discount rate would decrease to 2.3%. In several instances, our main reason for recommending to keep the current approach is that we could not find strong reasons to justify major changes. Thus, our recommendations do not always reflect a strong endorsement of a current approach, but can also reflect high uncertainty.
Overall approach to calculating the discount rate	Social rate of time preference (SRTP) approach	SRTP approach [Confidence: Medium-high]	We argue that the following constitute compelling reasons to continue to use the SRTP approach: In line with GiveWell’s welfare-maximizing goals & focus on cross-intervention comparisons of cost- effectiveness; Limited applicability of critiques in GiveWell’s case; Transparency of key assumptions; Relative ease of use; Commonly and increasingly used in practice. However, we think the social opportunity cost of capital (SOC) approach could also be a reasonable choice.
GiveWell’s current SRTP approach vs. Ramsey equation	Determine individual components and sum them up Calculate wealth effect implicitly in spreadsheet	Use GiveWell equivalent of Ramsey equation, r = δ + (η − 1)g. This means that wealth effect is calculated explicitly via formula, (η − 1)g. We recommend using a consistent η across applications. Note that using η=1 in line with the implicit utility function would lead to a 0% wealth effect and thus a lower discount rate. [Confidence: High, though this is contingent on SRTP being the correct approach]	Current approach can already be considered a variant of the Ramsey equation that is more suited to GiveWell’s modeling; we just think this should be more explicit. Main difference to Ramsey: Ramsey is based on absolute increases in consumption, whereas GiveWell’s model is based on percent increases, which yields a different wealth effect. We found and fixed a small calculation error in GiveWell’s calculations of the wealth effect. We think that using a higher η for calculating the discount rate than for the underlying utility function to model income benefits (as is done currently) risks overdiscounting of benefits. However, we have not reviewed which assumption for η would be the best choice.
Discounting consumption vs. health benefits	Discount health benefits using only the temporal uncertainty component	Tentatively keep current approach Continue to discount health benefits at a lower rate than consumption [Confidence: Low-medium]	We find that discounting health and monetary outcomes at equal rates is still dominant but do not find the arguments in favor of the dominant practice convincing. Instead, we concur with the view, broadly and increasingly supported in the economic literature, that health outcomes should be discounted at a lower rate. However, given the lack of a consensus view in the literature and various uncertainties such as those with respect to the shape of the utility function, we could not devise a superior approach to GiveWell’s discount rate for health outcomes.
Consumption growth rate	3% consumption growth rate This results in a 1.7% discount rate from improving circumstances, or the wealth effect	Raise consumption growth rate to 3.3% This results in a 2.0% discount rate from the wealth effect (according to the correct wealth effect formula) We expect the growth rate to decline over time and recommend next revisiting this estimate in 2028 [Confidence: Medium]	We recommend using published projections of real GDP per capita growth rates instead of eyeballing past GDP figures. Our approach considers both sub-Saharan Africa and South Asia as part of a population-weighted composite. We define time windows during which growth is being projected; our approach anchors the time interval to the longest-effect duration programs. We define time periods during which the current wealth effect estimate applies; we suggest revisiting the estimate at the end of each period (5.2 years).
Pure time preference rate	0%	Keep current assumption of 0% [Confidence: Medium-high]	Our impression is that the predominant opinion in the philosophical literature is that δ = 0%. Several arguments in favor of δ > 0% exist, though these seem highly context-dependent and somewhat controversial. Empirical estimates of individuals’ time preferences are extremely noisy and seem implausibly high to us; thus, not very useful. In a large expert survey on discount rates, δ = 0% is the modal response. δ = 0% is rarely used in practice. In many (though not all) cases where researchers/organizations assume a positive δ, it actually reflects a positive temporal uncertainty component rather than what GiveWell considers pure time preferences. Some theoretical articles show that δ = 0% leads to absurd predictions, but we do not think those specific predictions would apply in GiveWell’s case.
Temporal uncertainty	1.4%	Tentatively keep current assumption of 1.4% [Confidence: Low]	Temporal uncertainty has traditionally been defined more narrowly, as the risk of death or human extinction, than GiveWell does. GiveWell also considers changes in economic structure, catastrophe, and political instability. The current GiveWell assumption is roughly in line with other estimates, e.g.: other organizations with links to effective altruism use a broad range of estimates for existential risk ranging from 0.1% to 2.3%.
Compounding non-monetary benefits	0.9%	Tentatively keep current assumption of 0.9% More reasoning transparency in public write-up [Confidence: Low]	Including a component for non-monetary benefits sounds intuitively plausible to us, though we have not investigated the magnitude and the way of modeling. Our main reason for keeping the current assumption is that we couldn’t find strong counterarguments quickly. The SPC approach to discounting is the only approach we know of that accounts for reinvestment of benefits, but it is highly impractical to use and covers only a part of what GiveWell tries to capture in this component. We think a higher level of reasoning transparency in GiveWell’s public write-up would facilitate more critical engagement with this component.

To read the full report, please click here.

Acknowledgments

Jenny Kudymowa and James Hu jointly researched and wrote this report. Jenny also served as the project lead. Melanie Basnak and Tom Hird supervised the report. Special thanks to Bob Fischer, Sagar Shah, and Ben Snodin for their generous assistance in specific sections of the report, and further thanks to Andrew Martin (GiveWell), Ruby Dickson, and Aisling Leow for helpful comments on drafts. Thanks also to Anthony Boardman (University of British Columbia), Vicky Cox (Charity Entrepreneurship), Sam Donald (Open Philanthropy), Spencer Ericson (SoGive), Sanjay Joshi (SoGive), James Snowden (Open Philanthropy), Joel McGuire (Happier Lives Institute), Caitlin McGugan (GiveWell), Andreas Mogensen (Global Priorities Institute), Mark Moore (Simon Fraser University), Chris Smith (Open Philanthropy), Katie Stanford (The Life You Can Save), Dan Stein (IDinsight), Aidan Vining (Simon Fraser University), and Damian Walker (Management Sciences for Health) for taking the time to speak with us.

GiveWell provided funding for this report, but it does not necessarily endorse our conclusions.

Notes

^
Our priors are based on a combination of what we understood to be of high priority for GiveWell from speaking with Andrew Martin, and arguments and recommendations made in the “Change Our Minds” contest entries by SoGive and Julian Jamison.
^
This is the discount rate that is discussed in GiveWell’s 2020 discount rate write-up.
^
The only case we’ve seen different discounting approaches being suggested for monetary benefits and costs is Dhaliwal et al. (2012, p. 38) (J-PAL): “The discounting of costs is representative of the choice a funder faces between incurring costs this year, or deferring expenditures to invest for a year and then incurring costs the next year. An organization or government’s discount rate is usually calculated as the social opportunity cost of capital (SOC). [...] The discounting of benefits, on the other hand, represents how an end user of the program would trade off between the uses of the services this year versus next year. The appropriate discount rate for such a calculation is the social rate of time preference (SRTP) [...].” However, J-PAL itself does not use differential discounting of costs and benefits, but SOC for both outcomes.
^
GiveWell uses three different utility functions for its cost-effectiveness analyses: an isoelastic utility function with η = 1 (also called log-utility) to model consumption benefits, an isoelastic utility function with η = 1.59 to calculate the “improving circumstances” or “wealth effect” component of the discount rate, and a linear utility function for health outcomes.
^
We are aware that GiveWell uses other types of discounting (e.g., generalizability/evidence discounting). We do not focus on those other types of discounting in this report.