sgeeeee said:
No. . . Firecalc tests every case in its data base in order to find the worst case. Only that worst case defines SWR. The other data is ignored for that result. That is what FIRECalc does, and that is a worst case simulator.
It is only a worst case simulator if you look at the 100% SWR. For any lesser percentage, it will be looking at a number of bad cases, only one of which will be the
worst.
If you are not looking at the worst case (100% SWR), then the "resetting withdrawal amount up after a good year" (RWAUAAGY) has at least one hidden assumption in it. The probabilities in Firecalc are like assuming you take your money and are transported back in time to a random starting point -- the probabilities are the average over this collection of starting points. However, in the RWAUAAGY approach, you are only being time traveled back to start at time points that are just after a good year in the market.
IF the financial time series (inflation and markets) that go into Firecalc are statistically independent from year to year, then the choice of totally random starting points or just the collection of starting points after good years is unimportant. But if the time series are
NOT independent from year to year (that is, there is correlation), then the choice of which collection of starting points makes a difference.
In an extreme correlation case (just for illustration), if every strongly good year was followed by a strongly bad year, you might be in a situation where retiring at a fixed SWR+inflation method gives 90% success when considering all sequences of years. But retiring just after a bad year might be 100% and just after a good year 80%. Averaged out, we get 90%. The extra information (the market's performance last year) has predictive value if there is "serial correlation" in the financial time series.
Of course, real time series (anti-) correlations are not so strong, but correlations do exist. This may be one source of the intuition expressed earlier that RWAUAAGY is OK after
several good years in a row -- that whatever correlations in the financial fluctuations that may be there have died out -- that this year isn't that correlated with 5 years ago (say).
I'll add that the use of historical time series in Firecalc is implicitly a rejection of the idea of statistical independence between years. Otherwise, a Monte Carlo approach, choosing a random sequence of years (rather than the actual sequence of years), could be used to give more statistical power by averaging over more sample cases -- because there are a lot more random sequences of years than actual sequences.