MichaelDickens

Wiki Contributions

Comments

A generalized strategy of ‘mission hedging’: investing in 'evil' to do more good

To be clear, my model is exactly the same as your model, I just changed one of the parameters—I changed the AI portfolio's overall expected return from 4.7% to 1.3%.

It's not intuitively obvious to me whether, given the 1.3%-return assumption, the optimal portfolio contains more AI than the global market portfolio. I know how I'd write a program to find the answer, but it's complicated enough that I don't want to do it right now.

(The way you'd do it is to model the correlation between the AI portfolio and the market, and set your assumptions such that the optimal value-neutral portfolio (given the two investments of "AI stocks" and "all other stocks") equals the global market portfolio. Then write a utility function that assigns more utility to money in the short-timelines world and maximize that function where the independent variable is % allocation to each portfolio. You can do this with Python's scipy.optimize, or any other similar library.)

A generalized strategy of ‘mission hedging’: investing in 'evil' to do more good

50 randomly-chosen stocks are much better diversified than 50 stocks that are specifically selected for having a high correlation to a particular outcome (e.g., AI development).

This paper provides some more in-depth explanation of what I was talking about with the math. It's fairly technical, but it doesn't use any math beyond high school algebra/statistics.

The key point I was making is that, if markets are efficient, then you shouldn't expect a 5% (or even 4.7%) geometric mean return from the AI portfolio. Instead, you should expect more like 1.3%. I might have messed up some of the details, but I'm confident that the geometric return for an un-diversified portfolio in an efficient market is meaningfully lower than the global market return. This is not to say that mission hedging is a bad idea, just that this is an important fact to take into account.

A generalized strategy of ‘mission hedging’: investing in 'evil' to do more good

Thanks for making this model extension!

I believe the most important downside to a mission hedging portfolio is that it's poorly diversified, and thus experiences much more volatility than the global market portfolio. More volatility reduces the geometric return due to volatility drag.

Example case:

  • Stocks follow geometric Brownian motion.
  • AI portfolio has the same arithmetic mean return as the global market portfolio.
  • Market standard deviation is 15%, AI portfolio standard deviation is 30%.
  • Market geometric mean return is 5%.

In geometric Brownian motion, arithmetic return = geometric return + stdev^2 / 2. Therefore, the geometric mean return of the AI portfolio is 5% + 15%^2/2 - 30%^2/2 = 1.6%. If we still assume a 20% return to AI stocks in the short-timelines scenario, that gives 1.3% return in the long-timelines scenario. And the annual return thanks to mission hedging is -1.1%.

(I'm only about 60% confident that I set up those calculations correctly. When to use arithmetic vs. geometric returns can be confusing.)

Of course, you could also tweak the model to make mission hedging look better. For instance, it's plausible that in the short-timeline world, money is 100x more valuable instead of 10x, in which case mission hedging is equivalent to a 24% higher return even with my more pessimistic assumption for the AI portfolio's return.

A generalized strategy of ‘mission hedging’: investing in 'evil' to do more good

It seems to me that for mission hedging to work, there needs to be a strong positive relationship between production and stock price. That is, when (say) a fossil fuel company produces more oil, its stock price goes up. That might happen, but it might not. Several things need to happen:

  1. The increased quantity is not offset by a decrease in price
  2. The increased revenue translates into higher profit (this might not happen if, e.g., increased revenue inuces more competition, or induces increased costs for the oil company)
  3. Higher profit translates into a higher stock price

Step 3 seems very likely to happen in the long run, but steps 1 and 2 seem more uncertain to me, and I don't have a great understanding of the relevant economics. Do we have good reason to expect increased production to translate into stock returns? Or do we at least understand the circumstances under which it will or will not translate?

(Alternatively, we could look at the relationship between, say, oil production and the price of oil futures. This is a simpler relationship, but I'd guess the two numbers are basically uncorrelated. They will move together if demand changes, and will move oppositely if supply changes.)

Metaculus Questions Suggest Money Will Do More Good in the Future

It was an accident. I should have made a post, not a question.

Metaculus Questions Suggest Money Will Do More Good in the Future

I mistakenly submitted this as a question instead of as a post. Is there any way to convert it to a post?

What important questions are missing from Metaculus?

The question is intended to look at tail risk associated with stock markets shutting down. Transformative AI may or may not constitute such a risk; for example, the AI might shut down the stock market because it's going to do something far better with people's money, or it might shut down the market because everyone is turned into paperclips. So I think it should be unconditional.

Looking for more 'PlayPumps' like examples

In a number of cases, this reduction in hospital admissions and emergency room visits resulted in a cost savings in excess of $10,130, the cost of the average wish. In other words, Make-A-Wish helped, and helped in a cost-effective way.

This doesn't follow. The $10,130 cost savings went into hospital budgets, not into buying bednets, so it doesn't particularly matter that this money was saved.

Also, it seems implausible that Make-A-Wish could meaningfully reduce hospital admissions, so I'm inclined to disbelieve this study.

What important questions are missing from Metaculus?

Just to be clear, you specifically mean to exclude not-yet-EAs who set up DAFs in, say, 2025?

Yes, the intention is to predict the maximum length of time that foundations and DAFs created now (or before now) can continue to exist.

It might be interesting to have forecasts on the amount of resources expected to be devoted to EA causes in the future [...]

Agreed.

What important questions are missing from Metaculus?

I have a doc on my computer with some notes on Metaculus questions that I want to see, but either haven't gotten around to writing up yet, or am not sure how to operationalize. Feel free to take any of them.

Giving now vs. later parameter values

  • "In 2030, I personally will either donate at least 10% of my income to an EA cause or will work directly on an EA cause full time"
    • attempting to measure value drift
    • or maybe ask about Jeff Kaufman or somebody like that because he's public about his donations
      • or make a list of people, and ask how many of them will fulfill the above criteria
  • "According to the EA Survey, what percent of people who donated at least 10% in 2018 will donate at least 10% in 2023?"
    • Not sure if it's possible to derive this info
    • According to David Moss in Rethink Priorities Slack, it's probably not feasible to get data on this
  • "When will the Founders Pledge's long-term investment fund make its last grant?"
  • "When the long-term investment fund run by Founders Pledge ceases to make grants, will it happen because the fund is seized by an outside actor?"
    • by a government, etc.
  • "When will the longest-lived foundation or DAF owned by an EA make its last grant?"
    • EA defined as someone who identifies as an EA as of this prediction
    • the DAF must already exist and contain nonzero dollars
  • question about Rockefeller/Ford/Gates foundation longevity
  • best achievable QALYs per dollar in 2030 according to ACE, etc.
  • "Will the US stock market close by 2120?"
    • A stock market is considered to have closed if all public exchanges cease trading for at least one year
    • Could also ask about any developed market, but I think it makes most sense to ask about a single country

Open research questions

  • "By 2040, there will be a broadly accepted answer on how to construct a rank ordering of possible worlds where some of the worlds have a nonzero probability of containing infinite utility."

    • "broadly accepted" doesn't mean everyone agrees with its prescriptions, but at least people agree that it's internally consistent and largely aligns with intuitions on finite-utility cases
  • "In 2121, it will be broadly agreed that, all things considered, donations to GiveDirectly were net positive."

    • attempt at addressing cluelessness
    • "broadly agreed" is hard to define in a useful way. it's already broadly agreed right now, in spite of cluelessness
      • maybe "broadly agreed among philosophers who have written about cluelessness" but this might limit your sample to like 4 people
  • "By 2040, there will be a broadly accepted answer on what prior to use for the lifespan of humanity." see https://forum.effectivealtruism.org/posts/XXLf6FmWujkxna3E6/are-we-living-at-the-most-influential-time-in-history-1

    • alternate formulation: Toby Ord and Will MacAskill both agree (to some level of confidence) on the correct prior
  • "By 3020, a macroscopic object will be observed traveling faster than the speed of light."

    • relevant to Beyond Astronomical Waste

Finance

  • "What annual real return will be realized by the Good Ventures investment portfolio 2022-2031?"
    • Can be calculated by Form 990-PF, Schedule B, Part II, which gives the gain of any assets held
    • Might make more sense to look at Dustin Moskowitz's net worth
      • But that doesn't account for spending
  • "Will the momentum factor have a positive return in the United States 2022-2031?"
    • Fama/French 12-2 momentum over a total market index
    • As measured by "Momentum Factor (Mom)" on Ken French Data Library
    • Gross of costs
  • "Will the Fama-French value factor (using E/P) be positive in the United States 2022-2031?"
    • Fama-French value over a total market index (not S&P 500), measured with E/P, not B/P
    • French "Portfolios Formed on Earnings/Price"
    • Factor is considered positive if the low 30% portfolio (equal-weighted) outperforms the high 30% portfolio.
    • E/P chosen due to being less subject to company structure than B/P
  • "What annualized real return will be obtained by the top decile of momentum stocks in the United States 2022-2031?"
    • same definitions as previous question
  • "What will be the magnitude of the S&P 500's largest drawdown 2022-2031?"
    • magnitude = percent decline from peak to trough
Load More