Any tool used for assessment of the safety of a retirement portfolio that does not take valuation levels at the start of retirement into account is fundamentally flawed. This factor always affects the result. A tool that does not take it into account cannot produce the correct result. That sort of study has not taken into account all the inputs that matter. Without taking into account all the inputs, it is not reasonable to expect the analysis to produce the correct output.
How exactly do you propose to take valuations into account? I don't see anything quantitative here, and I don't see how to get anything quantitative. The most suggestive plot in this regard is the PE10 vs SWR plot that BenSolar linked to on another thread. But even that doesn't seem to lend itself to any hard and fast rule, and it is not clear how to extrapolate it out to the extremes of the 2000 bubble peak. I could imagine reasonable functional fits to the data which would give anywhere from 4% down to 0% SWRs out there, but these would all involve purely empirical functions, nothing with any theoretical basis behind them, so there's no way to choose among them. Not to mention statistical confidence problems. (I might play with such fits for fun if I get a chance, but wouldn't believe anything that came out.)
I guess it makes sense to say, as a rule of thumb, that it may be wise to tweak down the withdrawal rate when starting from high valuations, but I just don't see how to quantify this into anything really useful.
Of course I also believe Gummy is right that the whole SWR idea based on absolute fixed withdrawals is only really useful for guessing how big a pot one might have to accumulate in advance. As an actual withdrawal strategy, it is too rigid and fails in very unforgiving ways.
Cheers,
Bpp