(crossposted to Substack)
Decisions about whether to donate now or to invest and donate later can be very impactful. Despite this, there’s little public EA writing about investing relative to writing on other important considerations, like object level forecasting or organizational analysis[1]. I think this gap partly reflects the sensitive and personal nature of investment decisions and is partly because people lack a shared vocabulary to express their thoughts about EA investing. This post aims to introduce vocabulary and thinking tools for investment decisions and includes several examples of how they might be used.
Assuming that we are choosing whether to donate now or at some specific point in t years into the future, the value of investing to donate later can be expressed as
or
The Impact Discount Factor is the dollar amount of marginal impact you’ll have in the future if you spend one real dollar, expressed as a fraction of the impact you would have if you spent $1 today. A of 0.5 means you expect real dollars to buy half as much impact in the future as they do today[2].
The Real Return Multiplier is the expected real return on your investment decision over the chosen period. For example, if you assume your investments will have a yearly return of 5%, then your real return multiplier is .
Example: I have $100k already invested and want to choose between donating it now or investing for one more year. I think the one-year impact discount factor is 0.9 and the expected real return multiplier on my investment is 1.07. The relative value of investing to donate one year from now is 0.9 * 1.07 = 0.963, so continuing to invest is only 96% as good as donating now.
Multiple Scenarios
Suppose you are planning to mission hedge and are considering whether to buy a financial instrument connected to the performance of an AI company over the next three years. This instrument will have a real return multiplier of 2.5 if the company does extremely well and a multiplier of 0 otherwise. You think the probability of the company doing extremely well is 50%. You think that in the world where the company does extremely well, is 1.2, and is 0.8 otherwise. Your relative value of investing to donate later is 0.5 * 2.5 * 1.2 + 0.5 * 0 * 0.8 = 1.5. Since investing is 50% more valuable than donating now, you should invest[3].
Investment Risk
Note that the relative value formula doesn’t explicitly consider the risk of the investment. For investments uncorrelated with broader EA funding, you should be risk neutral so long as your investment decisions and returns don’t impact . If the amounts involved are large enough to impact , your estimation of how changes when total funding shifts[4] drives your risk aversion.
However, most investments have at least some correlation with the broader EA portfolio, and common investment choices like US equities will have a strong correlation. If there is a correlation, a scenario-based model can show how less correlated options lead to higher relative value. It also can also show how levering up or otherwise increasing correlation could be unwise, since levered equities perform well in the worlds with more total funding, and poorly in the worlds with less total funding.
Some Notes
If you're spending down a large pool of money over time, your spending can have a substantial effect on the impact of the marginal donation. As you spend in the present, you lower the impact of the marginal donation today, making future donation opportunities seem more relatively more appealing and raising . Formally, you can be modeled as donating to raise until * = 1, or until = 1/. For example, using Coefficient’s 2023 5% expected real return per year, we can say that they should spend until their = = .
If you think the impact discount factor could become higher relative to the real return multiplier at some point in the future, you may need to calculate the relative value of investing for durations other than one year. This is true even if your investments are perfectly liquid, since the one-year relative value calculation will only use and .
Final Thoughts
While you may be able to estimate your expected returns with some accuracy, is driven by a number of difficult-to-predict factors, such as the investment allocations of other EA funders, the value of movement building, and the future capacity of EA orgs to use funding effectively. Still, since we are always making investment decisions, and these decisions can be consequential, I think there’s value in improving the public understanding of many of the above factors.
This post has been an attempt to make thinking about EA investing more practical and approachable, although I’m unsure if I’ve succeeded. If you have thoughts on the usefulness of the proposed framework to your actual investment decisions, I would love to hear them.
- ^
I am grateful for the public writing that does exist, particularly from Philip Trammel and Michael Dickens
- ^
The reciprocal of the impact discount factor is the Future Cost of Marginal Impact (FCMI). It is the amount of money in real terms that future-you would have to pay to have the same marginal impact as spending $1 today. An FCMI greater than 1 means that future marginal impact is more expensive, while less than 1 means it’s cheaper.
- ^
Beyond this simplified example, I think it’s currently easy to mission hedge AI risk by allocating more to US tech stocks. If you think capabilities will significantly improve over the next few years, you should probably shift your portfolio, unless you believe that the existing EA community portfolio is already sufficiently allocated into AI for the opportunities that will be available. In other words, you shouldn’t shift if is too low in strong AI worlds.
- ^
Or elasticity with respect to total funding.