Note: This is a cross-post from my blog. For people who are already quite familiar with hits-based giving, you can skim up to the second image, as I try to bring people up to speed before using the Venture Capital analogy.

 

Most of us want to do as much good as possible. In light of this, how should philanthropists or funders decide which projects to allocate money to? Let’s say there are three options:

  1. Donate all your money to one charity you’re very confident in (the “all your berries in one basket” approach)
  2. Donate to 10 charities that have excellent reputations (the “prudent investor” approach)
  3. Donate to a wide range of pretty risky ideas, including charities that others are unlikely to fund (the “venture capital” approach)

 

Which one should you pick?

It turns out that the third option — donating to a wide range of risky endeavours — performs the best, at least if research on investment applies here. [1] A well-performing venture capital (VC) fund taking risky bets will outperform a similar investment fund that follows a more safe and measured approach.

Why is this?

It’s because the best opportunities to invest in (or donate money to) are often hard to spot, yet the very best investment opportunities can return 70x your initial input. Therefore, spreading your money across a range of more risky, but extremely promising, ideas can be a way to find these home runs. This is the premise of hits-based giving - that we can maximise our altruistic impact by focusing on high-risk and high-reward projects.
 

Another helpful metaphor is prospecting for gold. Gold is unevenly spread around the world. In most places, there is no gold, but some small areas have a lot of it. In fact, just one mine – Witwatersrand Basin in South Africa – accounts for over 30% of all the gold ever mined. Owen Cotton-Barratt argues that most of the good we can do in the world might be concentrated in a few select opportunities, just like gold reserves vary dramatically in size. These small probability yet high-value outcomes form what is called a heavy-tailed distribution, where a small number of data points determine the average. Below is a visualisation of what a heavy-tailed distribution could look like; the opportunities to find gold are ordered from left to right, in order of decreasing amounts of gold.

Returning to our venture capital analogy, we see that this same heavy-tailed distribution applies to returns from investments, as seen in the graph below. On the left, it shows the number of deals (e.g. investments) made by different kinds of VC funds, namely ones that ranged from getting a less than 1x return on their investment, up to those that consistently return over 5x their initial investments. On the right, we see the proportion of the fund’s returns made up by different deals, ranging from less than 1x returns to over a 10x return on the initial VC investment.

As shown, despite deals that brought back over 10x the initial investment making up only a small proportion of total deals done, they provide most of the financial value to the VC fund. In the case of the highest performing >5x funds, approximately 20% of all deals made account for 90% of the total returns.

From the chart above, we see another interesting point. Despite >5x funds out-performing the rest of the field, 40% of VC deals fail and return less than their initial investment. This highlights another key principle of hits-based giving: Just as VCs get a lot of misses, we need to be okay with a lot of our donations being misses, if we’re going after high-risk, high-reward opportunities.

Coming back from the world of gold and venture capital - what do these examples imply for a philanthropic portfolio aimed at maximising good? A few principles emerge:

  1. Opportunities to do good vary wildly in value, so it’s crucial to aim for the most effective projects.
  2. Most of our potential impact comes from a small number of opportunities.
  3. We’ll likely have many donations or grants that “miss”, or otherwise return very little altruistic value. This is okay – provided we’ve invested in a range of possibly high-impact opportunities.

Whilst these principles might seem counterintuitive at first, a deeper dive into venture capital strategy can explain why they’re important.

 

What determines the success of a hits-based portfolio?

There are a few factors that might explain why the best performing VC funds do so much better than average ones:

  1. Miss rate
  2. Success rate of big wins
  3. Size of big wins

Miss rate

One might expect that the best VC funds are simply the ones with the lowest rate of failed investments. However, this is not necessarily true. In fact, the best-performing funds - the ones that return 5x initial investments overall - had a higher rate of money-losing investments than funds that return 2-3x or even 3-5x, as shown below.

So if not the miss rate, what are the main factors behind the outperformance of certain VC funds?

 

Success rate for big wins

It turns out the best-performing funds have a higher rate of “big wins”, where investments return over 10x, relative to lower-performing VC funds. Basically, if you want to follow a hits-based approach, you want to be pretty confident you can pick the best opportunities, otherwise, you might end up having less impact overall.

 

Size of big wins

In addition, the magnitude of the biggest wins is one of the key driving forces that determines the overall performance of a VC fund, as opposed to the number of failures or median return. Again, this is wonderfully illustrated by research from Horsley Bridge shown below. We can see that the best funds had big wins that returned roughly 70x on initial investments, whereas the big wins of average funds were just 20x. The key takeaway here is that if you care about maximising impact (or money), what matters most for your overall portfolio is the size of your most successful investments. To put this into a baseball context, Chris Dixon says “Great funds not only have more home runs, they have home runs of greater magnitude”. Again, this highlights the importance of good judgement in selecting the best opportunities, as this will determine your overall success.
 

 

Why should we believe that the most successful philanthropy follows a hits-based model?

One key assumption I’m making is that philanthropy follows a similar pattern to venture capital, in that a small number of projects can actually account for most of the value generated for a funder, such that failed high-risk investments aren’t important. This is not immediately obvious, but there are several reasons to think it might be true.

  • Cost-effectiveness estimates of charities also seem to follow a heavy-tailed distribution, where a few charities are much more cost-effective than the median. Analysis by Toby Ord on the cost-effectiveness estimates for global health interventions finds that the most effective intervention performs 60x better than the median, similar to the best VC investments returning 70x their original value.
  • Philanthropy is structurally similar to venture capital (or investing more generally), in several ways:
    • Both are based on allocating money to maximise value, which can either be altruistic value or financial value.
    • Both are based on subjective assessments of complex and often unpredictable situations (e.g. answering “Will this tech startup become a unicorn” seems as challenging to answer as “Will this nonprofit/social movement be the one to change the world”).
    • Both philanthropists and venture capital funds tend to invest in a wide range of different projects to create a portfolio approach, rather than focusing only on a narrow set of opportunities.
  • There are several examples of successful hits-based philanthropy throughout history, highlighted below, with further examples referenced in Open Philanthropy’s work on the History of Philanthropy.
  • Potentially quite importantly - most other funders aren’t risk-neutral. Most philanthropists don’t want 90% of their donations to go to failed projects, so there seems to be a gap in the space where more risk-tolerant philanthropists can have a particularly large impact. Whilst I don’t have concrete evidence for this, it seems like a topic that’s spoken about a fair bit within the philanthropic sector.[2]

 

What are some examples of hits-based giving?

Whilst we don’t have perfect data on the exact benefit-to-cost ratio for most grants, below are some examples of hits-based giving that I believe would qualify as successful hits-based examples:

Neither we nor they had any way of forecasting or quantifying the possible impact of the group and we were prepared to ‘at worst’ resource a couple of people who we felt could benefit from a little financial support even if the only thing this would achieve was to fuel their relentless personal quest for social change.”

Sadly I couldn’t find any information on the US government or Katharine McCormick ever saying they thought these grants were low-probability bets, or that they had very diversified funding portfolios, but I think these are reasonable assumptions (e.g. based on things such as the number of regime change efforts the US has been involved in).

 

How to (potentially) do hits-based giving well

Bearing all that in mind, here are some tentative recommendations for philanthropists and those trying to maximise the amount of good they do:

  • Accept that we’ll have many failed projects
    • As shown above, having a large number of “failed” projects is not a sign of bad grant-making. It’s challenging to pick projects well, and the most ambitious ones have a higher chance of failing, yet this makes them no less valuable in expectation.
    • We should be comfortable with low probabilities of success given the potential benefits are large enough.
  • Most donations and grants we make need to have the potential to be enormously successful. 
    • For our hits-based approach to work, we need to fund projects that have the potential to achieve huge things and deliver enormous amounts of value.
    • Just like some unicorn tech-starts can return over 100x investments for VC funds, our philanthropic investments should also be able to deliver value much larger than would be expected, based on our initial funding.
  • Take action despite uncertainty
    • Some of the best ways to change the world might be hard to quantify or prove, but this shouldn’t stop us from going for these opportunities, provided they are high expected value.
    • We should make exploratory grants and use these as a chance to learn - which could be easier (and more accurate) than trying to figure expected impacts out a priori using careful analysis.
  • Seek out counterfactual opportunities
    • To maximise your impact, it might be best to seek out only opportunities you have access to, which would lead to high counterfactual impact. A perfect example of this is the FTX Regranting program which had 100 regrantors recommending funding to the FTX foundation. In their recent update, they said that:

“A key hope underlying these outcomes was the idea that regrantors and grant recommenders could exploit local knowledge and diverse networks to make promising projects move forward that we might not have known about or had time to investigate ourselves. It seems like that is playing out [emphasis mine].” 

 

Reasons to be sceptical of hits-based giving

Whilst hits-based giving is one plausibly good way of effective philanthropic giving, it’s not obvious that it is definitely the best way of doing so. Some reasons to be sceptical of hits-based giving are:

The frequency or magnitude of “hits” might be less than we expect

If we expect the average “hit” to be 10x more effective than the current best evidence-based giving opportunity, we would want to experience a “hit” for at least 1 out of 10 of all our donations, meaning 9 could fail, to produce the same expected value.

In some great research in 2018, Rethink Priorities examined the cost-effectiveness of funding vaccine research and compared it to donating directly to Against Malaria Foundation (AMF), a GiveWell top charity that has an extremely high level of cost-effectiveness, as well as a strong evidence base. Rethink Priorities found that certain vaccine developments (namely smallpox, malaria and rotavirus) performed better than donations to AMF, however only by a factor of 2x when considering 60% vaccination rates and up to a factor of 30x when considering full eradication of smallpox. On the other hand, 4 of the 7 vaccine research programs looked to be considerably less effective than AMF, where the worst performing vaccine (widespread Ebola) was 40x less cost-effective under 60% eradication. 

The research concludes that AMF, an evidence-backed giving opportunity that provides an alternative to hits-based giving, seems to perform reasonably similarly to the best “hits” in the vaccine development field, to the point where model uncertainty and different assumptions could significantly change their results. Therefore, in this case, it’s not obvious that hits-based giving would have outperformed a more evidence-based approach when considering global health and development. Whilst this is only one data point, focusing on a very specific form of hits-based giving (vaccine development), it’s certainly worth considering when there are very strong evidence-based giving opportunities to do good.[5] 

 

Charities might not differ greatly in cost-effectiveness

In this blog post, Brian Tomasik argues the cost-effectiveness of most charities probably doesn’t differ by factors over 10-100 times. There are several reasons he states for this, namely:

  • There are many long-term flow-through or indirect effects of charities, which are extremely hard to model but might make up most of the value of a charity.
  • The intervention might have been started or funded by someone else anyway if it’s such an incredible opportunity to do good.
  • Rare opportunities are indeed rare.
     

As a key premise of hits-based giving is that some opportunities are significantly (e.g. more than 10-20x) more effective than average opportunities, further research into this topic would have bearing on the case for hits-based giving.

 

Acknowledgements

Thanks to Justis Mills, Leonie Falk, Steve Thompson , Joey Savoie and Lorenzo Buonanno for helpful comments and feedback.

  1. ^

    I think there’s good reasons to believe investment research does apply to philanthropy. Read the section “Why should we believe that the most successful philanthropy follows a hits-based model?” to find out why

  2. ^

    This principle only works if you’re considering yourself as a marginal philanthropist looking for a comparative advantage relative to the rest of the field. Obviously if all philanthropists decided to be risk-neutral, this comparative advantage would disappear.

  3. ^

    Thanks to Holden for this example in his article on hits-based giving. You can read more about Katharine McCormick in the book, The Birth of the Pill.

  4. ^

    For interested readers, I highly recommend this article, this book and this documentary on Otpor!

  5. ^

     I’ll be writing up a fuller critique of hits-based giving soon based on some of these findings - watch this space!

15

New Comment
2 comments, sorted by Click to highlight new comments since: Today at 11:16 PM

EA long tail investment is harder given the quantitive evidence & history of effectiveness you want for each incremental dollar. It would be great to expand this using a lookback at what factors made small new charitable efforts in critical areas successful (careful and quantitive), and push forward with  not just funding existing small charities but creating & seeding micro charities based on what‘s been learned about what’s most effective.

Hi there!

I think the OP uses the term "risk" to denote only potential outcomes in which an intervention ends up being neutral (and thus the money that was used to fund it ends up being functionally "wasted"). But in the domains of anthropogenic x-risks and meta-EA, many impactful interventions can easily end up being harmful because, for example, they can draw attention to info hazards, produce harmful outreach campaigns, produce dangerous experiments (e.g. in machine learning or virology), shorten AI timelines, intensify competition dynamics among AI labs, etcetera.

In the for-profit world, a limited liability company will generally not be worth to its shareholders less than nothing, even if it ends up causing a lot of harm. Relatedly, the "prospecting for gold" metaphor for EA-motivated hits-based giving is problematic, because it's impossible to find a negative amount of gold, while it is possible to accidentally increase the chance of an existential catastrophe.