Lots of over analyzation going on, Im guilty myself of that.
As people who have decided not only to retire extremely early, but also to participate in internet forums to engage in discussions regarding the planning of our extremely early retirements, we are all guilty of over-analyzing (which, in my view, is a good thing--as Sol once said, I'm only planning on retiring once, and I don't want to fuck it up).
Is there a way to use cfiresim to figure out the precise oh-shit test failure rate for specific inputs? (In other words, figure out what percentage of historical cases would have, at any point during their trajectories, had a then-current success rate of less than X%, where X equals the "oh shit percentage.) Dr. Doom claimed that, using his numbers, the failure rate for an 80% oh-shit test was "greater than forty percent," but this may have been ball-park approximating using the cfiresim results. I can't figure out a way to extract a precise answer from cfiresim, if there is one.
Ok, I've now gone back and re-read Dr. Doom's Withdrawal Series. I think there's a flaw in the method he used determined the oh-shit test failure rate that produced an artificially pessimistic result.
Partly (or maybe solely, if there's no longer anyone else following along with this conversion that I'm now having with myself) for my own sanity, let's recap what I'm trying to figure out here, since it's getting very confusing given that we're talking about determining the probability of the occurrence of an outcome having a certain probability of the occurrence of another outcome. So, to recap:
The "oh-shit percentage" is the historical portfolio success rate that will cause you to decide: "oh shit, that success rate is too low; I need to [go back to work/cut my expenses/etc.]" As you progress through retirement, it's easy to determine if you ever breach the oh-shit percentage: at any given time, you can re-run cfiresim using your then-current numbers, and see whether or not it spits out a success rate below the oh-shit percentage.
What I am trying to determine is the historical probability that you will find yourself in a situation where you have a historical success rate below your designated oh-shit percentage (which tells you the historical probability that you will have to do whatever it is that you are using breach of the oh-shit percentage as a trigger for--return to work, cut your expenses, etc.).
Dr. Doom claimed that the historical probability of breaching his designated 80% oh-shit percentage using his particular cfiresim inputs was "greater than forty percent." I believe he determined this > 40% oh-shit test failure rate as follows:
- first, he used cfiresim to determine the stash size that would (as of the beginning of the period) result in a success rate equal to the oh-shit percentage. In Dr. Doom's case, using his specific inputs, a stash size 20% lower than his actual original stash size (i.e., a stash size of $484k, which is 20% lower than his original stash size of $605k) would result in an 80% success rate.
- next, he ran a cfiresim simulation using his specific inputs and examined the dip analysis to find out the percentage of historical cases that, at any point in their trajectories, fell below the reduced stash size determined in the first step. In Dr. Doom's case, the dip analysis told him that in over 40% of the historical cases, the stash size dipped below $484k.
However, I think this approach overstates the oh-shit test failure rate. This method takes the stash size that would generate a success rate equal to the oh-shit percentage at the beginning of the period and incorrectly uses it as a proxy for the stash size that would generate a success rate equal to the oh-shit percentage
at any point throughout the entirety of the period. But this is not accurate -- the success rate will vary as the remaining life to the end of the retirement period changes.
Using Dr. Doom's numbers, over 40% of the historical cases had dips that caused the stash to shrink below $484k. However, some of those dips will have occurred very early in the period (e.g., cases where the market crashed over 20% immediately following retirement) and others will have occurred later in the period. Let's say in one of those historical cases, the dip below $484k occurred 8 years into retirement; if Dr. Doom would have re-executed cfiresim at the 8-year mark plugging in his stash size then-equal to $484k, cfiresim would
not have reported a success rate of only 80% (his designated oh-shit percentage). Instead, it would have reported a higher success rate, because his then-applicable retirement period would have been 8 years shorter (and his money would have needed to last 8 years fewer).
That said, I can't figure out a way to use cfiresim to calculate the
actual oh-shit test failure rate from the data it currently provides. I know Bo-Knows is in the process of updating cfiresim; I wonder how much work it would be to modify cfiresim to calculate the failure rate for you for a specified oh-shit percentage.