FIREcalc and long withdrawal periods

[cute fuzzy bunny]

Forgive me if this has already been processed somewhere, and redirect me to that discussion. I looked but didnt see anything pertinent.

I was thinking today (in case anyone smelled smoke) and it occurred to me that there is one concern on using firecalc with very long retirement periods, such as those endured by an early retiree. After all, by IRS standard age ranges I should live at least another 40-45 years, excluding any improvements in medical technology.

The chief problem is that for ranges this long, you get no complete periods (ie, ones that count) in the modern era. All of the 40-45 year full periods will be back in the time of JFK and prior. Perhaps not representative of modern times. I know the calc also does partial periods, but those are less helpful the less data is in them. I know this is where monte carlo simulations are supposed to improve except that those dont demonstrate real trends that have occurred.

So I did this: instead of running firecalc with a 40 year period, I ran it with a 20 year period (which gives me more modern examples), took the very worst terminal portfolio size from that and ran it again for another 20 years. Gave me 100% on the first run and 80.something% on the second run. With a few tweaks to my portfolio mix I was able to get it to 100% and 95%. That was useful to learn. My prior runs with a 40 year period gave me 98% with the port as it was, but this was clearly not worst case.

Was this completely silly or am I poking something interesting, preferably with a stick?

Was this completely silly or am I poking something interesting, preferably with a stick?

You're right-on. I've done it myself only I used a 30 yr time frame instead of your 20. It pays to be cautious.

[dory36]

At first blush, I think what you are doing will produce useful results. It seems to me that the new ground it will cover is, as you said, the shorter periods after the mid-1960s. If there are dangers lurking in the recent past, this should catch them.

One of our regular contributors has even suggested a major rewrite of Firecalc that would do pretty much what you are describing. (Frankly, a major rewrite is currently awaiting a period of major boredom. The period of major boredom that led to Firecalc in the first place was the last year of my employment, and it was also the last time I've been bored, so I can't promise anything soon! )

Dory36

[sgeeeee]

Quote:

. . . Was this completely silly or am I poking something interesting, preferably with a stick?

I've done the same thing. And I've played around with 10+30 year runs as well as 30+10 year runs. I don't trust a 40 year run for the reasons you've mentioned, so I thought I would try to look at it a different way. That has some value.

Now, here's the problem. It is probably worse than worst case. Take the 20+20 year runs. What happens is the first time you run the simulator, FIRECalc finds the worst 20 years in history. Then, when you run it again, it finds those same 20 years all over again. It's like Groundhog day for the 1929-1949 or the 1965-1985 years.

[sgeeeee]

I went back into my files to see if I could find the results from my previous 40 year safe withdrawal runs. I did. But then it occured to me that the study I did is flawed for another reason.

What I had done was take a 50/50 allocation (stocks/TIPS@2%) and ran a 40 year SWR simulation as my baseline. Then I ran the same allocation for 30 years. I took the worst case terminal value and adjusted the initial withdrawal to be the 30 year initial withdrawal value divided by the terminal value -- and I ran that for a 10 year run. I kept making gueses at initial withdrawal rate for the 30 year run till I found the optimum. I did the same thing for 20+20 and for 10+30 year runs.

The problem is that withdrawal rate for the second half of the simulation is not adjusted for inflation. So now I'm convinced that my own study is seriously flawed. Maybe I misunderstood what you did. How did you adjust withdrawal rate for your second simulation?

[wabmester]

Quote:

Was this completely silly or am I poking something interesting, preferably with a stick?

Sort of interesting. As far as I can tell from your description, you found the worst 20-year period and basically appended it to itself for an imaginary worst-case 40-year period that is undoubtably worse than any real 40-year period we've seen.

Have you tried 4 iterations with 10-year periods? That should be frightening.

[Telly]

We have talked about this before, but feel free to do so again May come up with even a better idea, who knows?

From one year ago this month: http://www.early-retirement.org/cgi-...137134;start=2

Scroll down to about the third post, and read from there.

[dory36]

Quote:

Why not just 40 iterations of the worst year OR make it a true ground hogs day and pick the worst day and relive it over and over for 40 years?

That sounds painfully close to the way it was at my last job!

Dory36

[wabmester]

Quote:

Why not just 40 iterations of the worst year OR make it a true ground hogs day and pick the worst day and relive it over and over for 40 years?

Now that would be silly. *Unless you were modeling the Japanese stock market, in which case it would be pretty realistic.

I agree with you about past events having no influence on future events from a statistical standpoint. * The underlying assumption with this simulation is that there are some rational forces acting on the market, so you should be able to extrapolate past behavior to some extent.

However, the value of this simulation decreases with increasing payout periods. * For somebody with a 40-year payout, there are only three independent data sets to extrapolate from, which is basically useless as a predictor.

It makes some sense to try to add more virtual data points to find a more realistic worst-case, and cloning an existing data point to generate a virtual is as good as any other technique, I suppose.

So, if you believe that there are rational forces on the market *and* you can come up with a long enough period that the irrational forces average out, then you might be able to improve the model by adding some virtual data points. * My gut tells me that 10-20 years is when the irrational behavior starts to average out. *Probably closer to 20 (generational).

[wzd]

Try the spreadsheet to get inflation adjustment information. It at least has inflation adjusted terminal portfolio values, so your second run can then use that portfolio value. The only part I am not sure of is non-inflation adjusted withdrawals. When using inflation adjusted numbers for portfolio, etc., the non-inflation adjusted input parameters need to be reduced.

Wayne

[cute fuzzy bunny]

With regards to the inflation adjustment on the withdrawal, I simply made the adjustment manually with a calculator.

I did then do four ten year iterations. The worst one is the period starting in 1929, so thats always the terminal size line I took from.

Hence, most of what firecalc does is redundant. Nobody would need a portfolio lasting 75+ years, so all the tool really needs to do is use the data starting in 1929. Perhaps the equities to bond mix would change that, but I did my runs at 55 and 75% stock. Doubt too many would go a lot lower or a lot higher than that and skew the results.

Second worst was that run during the mid 60's on.

So the optimal tool and allocation would be a set of assets that would survive the great depression in multiple iterations, still produce the desired SWR, and not give up too much in terminal portfolio size. Perhaps really optimal would be iterations of the depression alternated with that nasty 1964-1973 period.

Thats *really* what you want to do with this sort of modelling, right? Survive the worst case scenario without giving up anything serious like withdrawal rate or terminal port size? I heard someone say that they thought an 80% result was good enough. I dont think so. I think that result says that you'll make it through the relatively good times but a bad stretch will finish you.

Now I have to go read over that year old thread and see what they came up with.

Now this has me thinking about options...you could put in your data and instead of running through thirty to forty year stretches from the old days, you could check off specific good/bad time periods that economists have identified, use that data to produce sequences or iterations of events and see if your modelling works out.

By the way, the optimum I found was somewhere between 50 and 55% equities. My current port is about 54% equity and 46% bond, although my equity portions dont track the TSM, and my bond holdings are a spectrum of the firecalc choices. That combination combined with my withdrawal rate and initial portfolio size survived four trips through the great depression.

All I wanted to do was see how well the stash did coming out of a long bad spell but without the unnatural assistance of of an improbable "bubble"

I picked 1965-1995 then looped back for 20 yrs. I know that is quite unrealistic in itself but, I don't expect a once-every-2-generations bubble to come save my ass in the future. I thought it prudent to run what one might call a somewhat worse than what the numbers would have us expect case scenario

[wabmester]

Quote:

If you want to create new data points, just randomly choose years and create as many 40 year intervals as you want.*

That's essentially what a monte carlo simulator does. *You can just do a bunch of random walks biased by the tendency of the market to rise (somebody can correct me here, but I think the market tends to go up two years for every down year).

What you lose by doing that is realism. *There is momentum in the market. * There tend to be bull runs followed by bear retreats. * And that has a big impact on returns when you're withdrawing money to live on.

Of course, you're right that the future can be totally different. * And I suspect it will be. * There are a lot worse possibilities than a period like the Great Depression. * We didn't have flying nukes, global warming, or Al Quaeda in 1929.

The only true SWR is the one you get by assuming you and your spouse both live to be 120, converting all of your assets into the things you'll need for the next 80 years, and then rationing them over that period.

As I think TH has said, these simulations are just one arrow in your quiver. *I'm not betting the farm on the results.

[cute fuzzy bunny]

Thats about right. I like being able to sample real periods of action/reaction, and i'm not comfortable with un-current economic periods being my only measure. But sequencing worst case periods and seeing how the portfolio survivability behaves is at least comforting and consumptive of a little time I had that needed consuming

I did another run alternating the depression 10 year period with the 64-73 period, two each. I fell two years short, running out of money after 38 years.

I can live with even that, since I'm not factoring in SS, selling my house, or my better half's portfolio (which will be bigger than mine by the time she reaches her desired retirement age in 13 years).