In our recommendations we have repeatedly advised people not to depend on numbers alone. Rather, we have advised people to look for cause and effect. We have searched out sensitivities to reduce our dependence on the numbers themselves.
some of us (e.g. me) tend to be a little literal with regard to numbers. if jwr's switching strategy shows that switching % stock allocation at a pe10 of 15 (or whatever) would have been optimal in the past, we would tend to apply that to the future. any intelligent person would realize that you can't predict optimal future approaches using optimized past approaches. I think that what jwr means about not using his numbers "alone" is that he agrees that optimal (or even good) future switching thresholds are unknown and unknowable. But there is some (unknown) future switching threshold that would be optimal. In principle, one should hold less of an asset class as it becomes over valued (or fully valued, or less undervalued). IOW if expected returns are lower hold less of that asset class. I think that this is broadly consistent with results you would get from mean variance optimizer analysis. In most circles this is probably noncontroversial. Perhaps some would hold that an 80% stock allocation is always optimal based on historical swr from 1871-2000 regardless of current valuation. I think jwr is responding to those individuals. bernstein does mention considering "the price of tomatoes" when making allocation policy (as an alternative to relying strictly on mvo outputs). bernstein makes no effort to quantify this for the s&p or other asset classes- as far as I know.
this generating numbers that are not to be relied upon has confused me and possibly others. I think that all this effort has been directed at supporting another poster's point that an 80% stock allocation isn't always optimal for a retiree-