Short Takes: Podcasts, 2022 Returns, and more

I haven’t had many people ask me whether I’d consider hosting a podcast, but it’s come up enough to make me think about it.  I have some solid reasons for not doing a podcast: it’s way more work than I’m willing to do, and my voice isn’t good.  To illustrate the best reason, though, consider this hypothetical exchange:

MJ: Welcome to the podcast, Dr. G.

Guest: I’m happy to be here.

MJ: Let’s get right to it.  Please describe your research interests.

Guest: I work on retirement decumulation strategies, safe withdrawal rates, and risk levels of equities.

MJ: From what data do you draw your conclusions?

Guest: I use worldwide historical returns of stocks and bonds.

MJ: How do you deal with the challenge that we don’t have enough historical return data to directly draw statistically significant conclusions?

Guest: Uh … I perform simulations drawing from the available pool of data.

MJ: So, you create seemingly plausible return histories to extrapolate from the small pool of available data.

Guest: Yes, … but I use methods widely accepted in the literature.

MJ: Isn’t it true that you have to make assumptions about the distribution of market returns, such as autocorrelations and the size of tails, to be able to perform simulations?

Guest: Well, yes, but I preserve the properties of the original data as much as possible.  In some simulations I draw blocks of returns selected at random.

MJ: Yes, I can see that you’re trying to preserve autocorrelations that way, but it still destroys long-term autocorrelation and the tendency for long-term valuation-based reversals.  For example, consider one block that ends in the depths of the great depression, and another that ends at the peak of the year 2000 tech boom.  Your simulation will make no distinction between these two states for the next block it draws.

Guest: There’s a limit to what I can do with the small amount of data available.

MJ: Yes, that’s my point.

Guest: In other simulations, I treat the expected return over the next year as a random variable that drifts over time.

MJ: Yes, and you assume a particular probability distribution for this drift.

Guest: I have to assume some kind of distribution.

MJ: Aren’t there many other possible sets of assumptions we could make about market return distributions that would ultimately lead to completely different conclusions about retirement decumulation strategies, safe withdrawal rates, and risk levels of equities?  In fact, aren't all your conclusions primarily attributable to your underlying return distribution assumptions rather than the actual historical return data?

Guest: [Fuming] Do you have a better idea?

MJ: Perhaps not.  We could try multiple approaches and see if they give similar results.  For example, we could use a model where the current stock market price-to-earnings ratio affects the distribution of the upcoming year’s return.  Another idea is to model corporate earnings growth separately from investor sentiment as expressed by the price-to-earnings ratio.

Guest: Good luck with that. [storms out]

Successful podcasters let their guests make arguments without challenging their ideas in any serious way.  Listeners might enjoy some fireworks, but few interesting guests would want to participate in a podcast like my example above.  My first guest might be my last.

Here are my posts for the past two weeks:

My Investment Return for 2022

Bullshift

Here are some short takes and some weekend reading:

Justin Bender goes through model portfolio returns for 2022.  The losses in long-term bonds aren’t pretty.

The Blunt Bean Counter has some advice on doing a 2022 financial clean-up and a 2023 financial tune-up.