Various mixed thoughts.
I generally agree with Felix's position that people associate "risk" and "volatility" too strongly.
A tangential point I'd make, which the Anarkulova research sort of touches on, is the role of Social Security. Obviously Social Security isn't a bond, but rather, more like an annuity. However, a huge inflation-resistant annuity can arguably replace much of the function of bonds in a retirement portfolio.
Obviously it's difficult to guess how that will change over time, but you could say the same about a bond allocation. (I do think that the "Social Security might not exist" thing gets overstated. Even if no policy changes are made and the trust depletes, Social Security can cover a much bigger portion of most people's expenses than they realize, and there's
a lot of political friction to entirely eliminating it.)
A major feature of the research is its observation that reliance on American market data results in an effectively-tiny sample size: a century and change from a single country simply isn't very many economic cycles, and United States data in particular might have a lot of survivorship bias.
I agree with this line of thought, but I'm skeptical of how usefully their kitchen-sink-of-country-data addresses the issue. I have three concerns here:
1-The extra data might not be usefully representative. The United States has a strange position in worldwide financial markets: it has an extraordinarily outsized bond market, its currency has a lot of global coupling and an unusual scheme for monetary policy, and its domestic stock market is much more diversified than those of small economies. It's not obvious that blending in "what would happen if your domestic stock allocation behaved like the Latvian stock market" improves the prediction of future American markets.
2-Depending on the question being asked, anti-survivorship bias might reduce actionability. I think this is highlighted by the study's implicit ultra-low safe withdrawal rate. How much utility is there, really, in modeling smooth trajectories through situations where the world is cast into complete chaos, and the institutions are all toppled? What makes me especially skeptical is that none of those past calamities happened in a societies where a huge fraction of the older folks were leaning hard into broad stock market allocations. Even if we want to pursue a "retirement planning as apocalypse insurance" approach, I can't help but think that "WWII but everyone has a 401(k)" might imply different market dynamics than actual WWII. (And this is a general question mark around historical data, including older data from the US. Different asset classes can behave differently in situations where their relationships with investors are different.)
3-Nearly all of the datasets end at the close of 2019, but they begin across a wide sprawl of years dating back to the 1890s, all the way up into 2018. Maybe I'm just missing something, but I can't help but wonder if it systematically over-weights characteristics of the post-GFC economic cycle, despite diversifying these characteristics across different country's exposures to them.
Another qualm I have with the study is, I feel like it's a bit muddled as to what it's trying to simulate. It's doing enough income/savings-rate simulation to perturbate the results, but I'm not sure that it's doing enough to be a useful aggregate-whole-lives simulation versus pure wealth-building and retirement-withdrawal simulations. It seems like it might be non-usefully muddying the waters. For example, it uses an interesting empirically-derived model for income volatility, but an arbitrary and (IMO) weakly-justified model for savings rate (basically "Vanguard says this is what people throw at 401k accounts").
(I'd further state that
all of these kinds of studies have representativeness issues for individuals due to withdrawal patterns - and especially ability to tolerate reduction - being so diverse.)
In conclusion, ¯\_(ツ)_/¯