JWR1945 wrote: Norbert Schlenker wrote:
We have a very strong case that past sequences were lucky sequences. Raddr showed this in the early days of the FIRE board.
That's worth reading. Could you provide a url?
You are here! I am talking about our FIRE board at this NoFeeBoards website. Look at pages 1, 2 and so forth. I doubt that you have to go beyond page 3.
In addition, we have an excellent search feature at this site.
John, I used the excellent search feature to look for posts by raddr in the FIRE forum with the word variance. No hits. With "deviation", I got five hits but (a) they don't address the issue of "lucky sequences" and (b) raddr seems pretty scathing about your methodology at first glance. I certainly hadn't noticed yet that you tossed a couple of inconvenient years in order to get a better r^2, but raddr filled me in right away. Are you sure you want me to read this stuff?
Actually, mean reversion does not have a strong influence on Historical Surviving Withdrawal Rates. Mean reversion tends to be a long-term effect.
That may be. Your problem is that you are calculating SWRs and confidence limits using an unstated assumption that there is no mean reversion. You're trying to tell people that what is safe is so far below 4% because return sequences could be really unlucky, when mean reversion implies that really unlucky return sequences are really unlikely.
Historical Surviving Withdrawal Rates are dominated by the returns during the first eleven years (this was one of my early findings).
I believe you're talking through your hat. Speaking loosely now, the classic 4% SWR is based on an assumption that the future will look like the past. You believe that's wrong. Hell, I believe it's wrong too. As far as I can tell, everyone else, even intercst, admits that it's wrong.
So what do you do instead? You run questionable regressions. Raddr's point re much too high estimates for r^2s and confidence because of sequence overlap cannot be dismissed here. You then take these regression coefficients and build equations, posting endless tables of numbers in numerous threads which can mean nothing to anyone.
The worst is yet to come. It turns out that everything, all these numbers, depends on knowing what returns will be for the next 11 (14?) years. You have "90% confidence in these limits" as long as you know this unknowable fact.
You and hocus are tearing strips off people for assuming an unknowable fact - that the future will be like the past. Then you turn around and build something new that depends on a different unknowable fact - returns for the next 11 or 14 years. What's the freaking point?
Great minds think alike. Fools seldom differ.