This is going to be a hit or miss post, since I've been out and about traveling and won't be caught up for some time. Some offhand thoughts:
"If Bengen 'concluded that a 4% drawdown rate resulted in certain survival', he was wrong" and "The article is dominated by references to Wade Pfau observations. He too is a very strong advocate of Monte Carlo simulations to help arriving at retirement decisions. "
The NYTimes article has a graphic with three other spending models by Pfau. All three show 100% survival over thirty
years (worst case shows money remaining for all models). That includes a model with a constant (inflation adjusted) drawdown amount.
Yet the simple Monte Carlo tools advocated (based on mean and standard deviation inputs) intrinsically contradict this - they are built on the premise that failure is always possible (since they say that a portfolio can lose value year after year after year after ...). Does that mean that Pfau, like Bengen, was also wrong in concluding certain survival?
A problem is that by design, these simple tools are unable to conclude that survival is certain. Regardless of inputs. If you build a conclusion (failure is always possible) into a tool, you've rigged the results. You can't use these tools to "prove" that 100% success is impossible. They're unable to say anything but.
It's fine to use random number generators (aka Monte Carlo) to "run" models many times and see what outcomes might result. The problem is not in how models are used (trial and error - random numbers), but with the models themselves. Unfortunately these tools conflate the creation of the models with the Monte Carlo running of the models to generate a range of possible outcomes.
Don't confuse a criticism of these tools with a criticism of Monte Carlo simulations.These tools create simplistic models that usually assume each year's market's performance is independent and that returns are normally distributed (bell curve).
But data suggest that stock market performance is a leading indicator of business cycles. Thus stock market performance is itself cyclic (not independent from year to year) albeit with an upward bias.
"stocks as a whole move in advance of the economy" =
AAII Journal, Aug 2003As to the bond market, the trivial Monte Carlo models assume that nominal returns are independent of inflation. The Fischer hypothesis suggests the opposite.
"The Fisher hypothesis is that, in the long run, inflation and nominal interest rates move together."
http://moneyterms.co.uk/fisher-effect/The first paragraph by Pfau in his Forbes column says that the models need to include correlations - something that's antithetic to simplistic free Monte Carlo tools that assume independence of inputs in building their models.
His penultimate paragraph states simply that: "the results of Monte Carlo simulations are only as good as the input assumptions, ... Monte Carlo simulations can be easily adjusted to account for changing realities for financial markets."
It's certainly easy from a mechanical perspective to adjust the models (e.g. by changing the mean return for bonds). What's not easy at all is figuring out what adjustments to make. That gets right back to the results being "only as good as the input assumptions", or as I wrote before, GIGO.
Again quoting Pfau: "Many financial planning assumptions are based on historical returns; however, these historical returns may not be relevant in the future."
https://www.onefpa.org/journal/Pages/MAR17-Planning-for-a-More-Expensive-Retirement.aspxAt best, even if a model is good and analysis sound, all you're going to get is a sense of whether you're saving enough (i.e. what MikeM wrote). It's of less help during retirement because, as hank noted, extraordinary events happen.
I'm wondering who the unnamed "professionals in this field" are. Or even what "this field" is. But for the record - I've never taken a statistics course in my life. I'm just an individual investor like most people here, albeit one who did once ace a course in writing and research.