SWR of 6.21% for 26 years

[hocus]

SalaryGuru (Responding to BigMoneyJim):

What you are trying to do is take the current withdrawal rate of Mr. ER 2000 and back test it against history.  If that were valid, then you would conclude that Mr. ER 2000 was 100% safe when he started with is 4.1% inflation adjusted withdrawals, but is now less than 100% safe with that same withdrawal strategy.   But if he's no longer 100% safe, then what did 100% safe mean 4 years ago?

I think that the point being made here by SalaryGuru is an important one. SWR analysis is a risk assessment tool. The idea is to give you in advance of the day you turn in your resignation an idea of what may happen in the years to come. If we start using the tool in such a way that we cannot trust the risk assessments it provides to remain stable, the value of the tool is greatly diminished.

Say that you retired on January 1, 2000, and that you possessed confidence at the time that 4 percent was the true SWR. Say that Bernstein is right that the true SWR was actually 2 percent. Say that you are living on $60,000 per year. Say that you discover your mistake in at the end of 2005. That means that, at the time you begin looking into the belt-tightening idea, you have already spent $150,000 more than you would have had you used a more accurate SWR methodology. You would have to do two things at this point--reduce your spending by $30,000 per year and make up for the $150,000 shortfall. That's a lot of belt-tightening.

You could return to the workforce poorer than you were the day you left it and five years older. That idea possesses little appeal to me.

You could change your investment allocation to bring your SWR back up to the levels you were seeking in the first place. In all likelihood, however, it would be a fall in stock prices that would cause you to question the conventional methodology. That means that you would be selling your stocks at the worst possible time, when their prices were lowest. Again, the idea possesses little appeal.

I think that it makes more sense to properly assess your risks before turning in the resignation. SalaryGuru is being stubborn about the point he is making, but I think he is being stubborn for a very good reason. He is saying that it does not make sense to have confidence in the 4 number and not in the 6 number. Confidence in the 6 number is a logical consequence of belief in the same things that justify confidence in the 4 number.

SalaryGuru has more confidence in the 4 number than I do. We are not in agreement on some core issues. But I think that he is making a strong point on the logicial consequence issue. He is right when he says that those who do not feel comfortable with the 6 number should be asking themselves whether there is something wrong with the 4 number as well.

 

[dory36]

... My answer is that we should make every effort to understand the implications and limitations of all the models available to us.  I think once you've done that, you decide that if you start with a generous post-retirement budget (one you can cut back on if you need it), plan for a long life, discount your social security benefits, choose a reasonable allocation plan, and keep your initial withdrawal rate a little below 4%, you are probably going to be okay.  

I hereby award SG a box of dryer sheets.

I think the issue of being able to cut back if necessary is particularly important.

Everyone's situation is different, but if you are on the edge with 4%, I suspect you need to start figuring out a way to increase the portfolio or else cut back somehow.

I am not sure what would constitute an ideal ratio, but I think we could cut back from 4% to 3% without much problem, and probably get close to 2% without feeling like we'd better see if Wal-Mart needs a new greeter.

I'm guessing that at least 20-30% of spending ought to be completely discretionary for someone retiring early.

[anti-debt-rant]This would especially apply to folks whose car, house, credit card, whatever payments make up a big part of your expenses. You can perhaps make a marginal dollar or two by borrowing low and using your investment genius to parlay that into a fortune, but you will be at the mercy of events, should anything in your life or the market hit a speed bump.[/rant]

Dory36

PS: Someone decades ago told me that any long term planning effort was like measuring with a micrometer, marking with chalk, and cutting with an axe.

 

[hocus]

SalaryGuru (in response to my "drunken friend" post):

What you are saying is that you believe that the future may be worse than the worst case of the past.

raddr did an analysis over on the NoFeeBoards.com site of the odds that we will see returns in the next 30-year period equal to the returns assumed in the conventional methodology studies. Here's a link (scroll to the bottom for the relevant point).

http://nofeeboards.com/raddr/gordon.htm

Raddr says that the odds of a retiree of today obtaining those sorts of returns are about 1 in 100. When defenders of the conventional methodology say that "the historical data shows a 4 percent withdrawal to be 100 percent safe," what they should be saying is that "the historical data shows that there is a 1 in 100 chance that a 4 percent withdrawal is 100 percent safe." There is a world of difference in the content of those two statements.

 

[hocus]

I have trouble with your equating PE10 to that overvaluation.

I don't want anyone to get the idea that I believe that PE10 is the only tool that may be used to incorporate the effect of changes in valuation levels into the analysis. Bernstein did not use PE10, and the number he came up with was 2. JWR1945 uses PE10 and he came up with numbers in that same general ballpark. I believe that both Bernstein and JWR1945 are on the right track. My beef is with the conventional methodology, which makes no adjustment whatsoever for changes in valuation levels.

Here's a link to a post from the SWR Research Group board where JWR1945 sets forth a table comparing the PE10 at the start of a 30-year historical sequence and the highest surviving withdrawal rate that ultimately applies for that sequence (we refer to this number as the historical database rate, or HDBR).

http://nofeeboards.com/boards/viewtopic.php?t=2245

Here's the chart:

HDBR50 and HDBR80 Ordered by PE10

Lowest Valuations
Code:

Year  PE10  HDBR50  HDBR80
1921    5.1   9.6   11.6
1922    6.2   8.4   10.3
1924    8.0   7.9    9.5
1923    8.1   7.6    9.0
1933    8.7   5.8    8.4
1980    8.8   9.2   10.3
1975    8.9   7.0    8.3
1978    9.2   7.9    9.1
1979    9.2   8.4    9.5
1932    9.3   6.0    8.1
1925    9.6   7.3    8.4
1942   10.1   6.1    9.1
1943   10.1   6.3    9.2
1949   10.2   7.8   11.1
1948   10.4   7.7   10.9
1950   10.7   7.3   10.2
1944   11.0   6.1    8.6
1976   11.1   6.5    7.2
1926   11.3   6.9    7.7
1935   11.4   5.3    7.4


Middle Valuations
Code:

Year  PE10  HDBR50  HDBR80
1947   11.4   6.8   9.4
1977   11.4   6.8   7.4
1951   11.8   7.1   9.4
1945   11.9   5.9   8.2
1954   12.0   6.9   9.0
1952   12.5   6.9   9.0
1934   13.0   4.8   6.3
1953   13.0   6.7   8.6
1927   13.1   6.6   7.3
1938   13.5   4.9   6.6
1974   13.5   5.6   5.9
1958   13.7   5.8   6.8
1941   13.9   5.0   7.0
1939   15.5   4.6   6.0
1946   15.6   5.2   6.7
1955   15.9   5.8   6.9
1940   16.3   4.6   6.1
1971   16.4   4.9   5.0
1931   16.7   5.1   5.6
1957   16.7   5.3   6.0


[/b]Highest Valuations[/b]
Code:

Year  PE10  HDBR50  HDBR80
1936   17.0   4.4   5.5
1970   17.0   4.8   4.9
1972   17.2   4.8   4.8
1959   17.9   5.0   5.4
1956   18.2   5.2   5.9
1960   18.3   5.0   5.3
1961   18.4   5.0   5.3
1973   18.7   4.7   4.5
1928   18.8   5.7   5.8
1963   19.2   4.9   5.1
1967   20.4   4.5   4.5
1962   21.1   4.7   4.8
1969   21.1   4.3   4.2
1968   21.6   4.3   4.2
1937   21.6   3.9   4.6
1964   21.6   4.6   4.6
1930   22.3   4.8   4.8
1965   23.2   4.3   4.2
1966   24.0   4.2   4.0
1929   27.0   4.6   4.3

I see a strong correlation between the PE10 that applies at the start of a sequence and the HDBR that can only be determined 30 years later. It's important to understand that, in the event that there is any correlation whatsoever, the conventional methodology numbers are wrong. The conventional methodology makes no adjustment whatsoever for changes in valuation levels. It churns out the same SWR no matter how high or low current valuations happen to go.

 

[hocus]

If you tell me those dissapointing returns are going to happen next year or you tell me they are going to be dispersed over the next 30 years, I am very skeptical.  I have never seen any reason to believe such predictions and I've seen a lot of reasons to be skeptical.

It's important for people to understand that JWR1945 and I are not engaging in any speculation about things that may or may not happen in the future. We don't have any better insight into what is going to happen than anyone else.

We are starting our SWR analysis from a different foundational premise than are those using the conventional methodlogy. We believe that in the past changes in valuation levels have always affected the withdrawal rates that were safe. We believe that this rule will continue to hold in the future. The result of our adopting that premise is that we come up with different numbers.

There is no need for there to be a nuclear war or a depression or stagflation or any other econonmic calamity for the conventional methodology numbers to fail. All that is needed for those numbers to fail is for the premise that changes in valuation have no effect to turn out to have been wrong.

The conventional methodology was developed at a time when a lot of people believed in something that I refer to as "the Stocks-for-the-Long-Run Paradigm." I believe that that paradigm is flawed. Stocks really do provide wonderful returns when purchased at some valuation levels. There are other valuation levels from which the returns they provide are not so hot, however.

There have been studies showing that short-term timing does not work. But the historical data indicates that timing does work if you are a long-term buy-and-hold investor. SWR analysis is concerned with long-term results. So it doesn't matter for purposes of SWR analysis whether short-term timing is possible or not. What matters is whether long-term timing is possible.

The data that we have looked at indicates that it is. I think that our findings on that question are exciting, and will lead to all sorts of interesting discussions in future months and years.

 

[gummy]

It's interesting to demonstrate (by actual historical precedent) what happened over some BAD 30-year period (like 1930 - 1959).

Starting in January, 1930, a $1M portfolio invested in the S&P 500 would have survived 30 years (to the end of 1959) with a withdrawal rate of 4.08% or $40,800 (increasing when inflation was positive).

By end of 1932 your portfolio would be worth $291,083 (after three terrible years with annual returns of -24.9%, -43.5% and -8.2%).
Note: You'd still be withdrawing $40,800 since inflation was negative during those years so your withdrawals didn't increase

In 1933 (with a small inflation rate) your withdrawals would increase to $41,008 and (starting 1933 with that $291,083 portfolio) your withdrawal rate for 1933 would have been $41,008/$291,083 or 14.1% of your starting portfolio !!

Nevertheless, your portfolio would have survived the 30 years (to 1959).
http://home.golden.net/~pjponzo/SWR-example.gif

 

[John Galt]

The post by gummy bears study.

John Galt

 

[BigMoneyJim]

I told myself I was going to step out of the middle of this discussion, but there's too much to not respond to.

First I'll say that I have learned a lot about SWR and FIREcalc in a short period of time and I thank everyone. My assessment is apparently unique, but I am now satisfied with my analysis and have 15-30 years to reconsider before I actually begin withdrawals.

Second, I have been arguing the semantics and the technicality of the tools, which even if we agree upon what they are and how they work we could still debate for years on what the results mean for ER. But I'm a tool guy and like to fully understand the tools.


I now believe that FIREcalc supports a 4.1% over 30 years "100% safe" rate for Mr. ER 2000 but doesn't support a 6.21% safe rate for Mr. ER 2004 over 26 years, even though casual observance says they are the same.

Here's why: FIREcalc assumes 1929 could start today. It assumed that in 2000, and it assumes it today. If you believe 6.21% is safe today, then (oversimplified, and using the worst case example) you're saying you don't believe 1929-1955 can happen after 2000-2004, but FIREcalc doesn't take into account the current status of the market, so FIREcalc's simulations don't support 6.21% over 26 years as 100% safe.

My point is that saying otherwise is adding information to FIREcalc's results that isn't supported by FIREcalc's data set and method. Adding information may or may not be useful, but it's important to me to note the analysis methodology difference.

I am making no point as to the validity of any of it, nor am I saying FIREcalc is authoritative or more pertinent than any other analysis.

Anyhoo, like I say I learned a lot and have formed my opinions, so I'll try to exit stage left and go back to posting silly things in "Other Topics" and babbling about my debt in "Young Dreamers".

 

[wabmester]

Jim, the limited historical data suggests that there is strong RTM to the upside after crashes. You're basically suggesting that the RE2000 guy has to live through the Great Depression twice to invalidate the 6.21%, and such a string of bad luck hasn't occurred in FIREcalc's data set.

If gummy were to overlap his worst-case 14% SWR with stock and inflation data, you'd see that the only way that 14% worked out was to have some very good years immediately following the bad years.

In a nutshell, that RTM is a core faith of FIREcalc users. Personally, I don't think we have enough data to suggest that RTM will always pull us out of the FIRE unsinged.

 

[gummy]

... good years following the bad years
Amen.
Let's pray that that always happens



http://home.golden.net/~pjponzo/SWR-4.htm

 

[sgeeeee]

. . . I now believe that FIREcalc supports a 4.1% over 30 years "100% safe" rate for Mr. ER 2000 but doesn't support a 6.21% safe rate for Mr. ER 2004 over 26 years, even though casual observance says they are the same. . .

BMJ,

Have you given any thought to how Mr. ER2000 will be able to withdraw an inflation adjusted 6.21% per year for the next 26 years and be safe while Mr. ER2004 will not be able to do the same thing? Do you believe there is an "ER initial year god" that changes the financial mathematics for individuals based on the year they first retired?

If they both have the same amount of money in 2004 -- invested with the same allocations, etc. and they both withdraw the same amount of money each year . . . how is that one survives the next 26 years while one does not?

I realize that many of you are uncomfortable with this conclusion. But faulty logic that leads to contradictory results can't be the best way to deal with your discomfort. Maybe it would be better to come to grips with this result. Either you feel comfortable with ER 2004 using a 6.2% withdrawal rate, or you question the universal 4% SWR.

 

[BigMoneyJim]

Me:

I'll try to exit stage left and go back to . . .

Okay, I take it back. This discussion is too much fun to not keep going. Just let me know if I'm being aggressive, offensive or otherwise an ass because I don't want to alienate anybody here.

Jim, the limited historical data suggests that there is strong RTM to the upside after crashes. You're basically suggesting that the RE2000 guy has to live through the Great Depression twice to invalidate the 6.21%, and such a string of bad luck hasn't occurred in FIREcalc's data set.

Agreed. But to nitpick, RTM is based on market movement analysis. The 4.1% holy grail of historical safe withdrawal rates is based on statistical analysis of historical samples with no regard to current market status. 6.21% 2004-2029 is a combination of the two analyses. (And--statistically--only supported by 3-5 historical sample periods which isn't a sufficient sample for pure statistics.)

Combining the two analyses is fine, but the underlying proofs are not additive.

In a nutshell, that RTM is a core faith of FIREcalc users. Personally, I don't think we have enough data to suggest that RTM will always pull us out of the FIRE unsinged.

FIREcalc has no memory or concept of RTM. It doesn't adjust it's SWR based on market levels at retirement time. If FIREcalc users are RTM believers, that's fine, but Mr. ER 2000's to-date performance is in the bottom 5% of the successful 30-year sample periods; if you believe in RTM you believe you'll be fine; if you believe in statistics you may be concerned at your position in range; if you're into valuations you may be panicking or reducing expenses.

Statistics (FIREcalc 4.1%), market movement (RTM, salaryguru 6.21%) and valuation (*****, jwr ~2%) are all useful tools. It's interesting that they're pointing in three different directions now, but that's useful, too, even if worrying or frustrating.

Do you believe there is an "ER initial year god" that changes the financial mathematics for individuals based on the year they first retired?

If they both have the same amount of money in 2004 -- invested with the same allocations, etc. and they both withdraw the same amount of money each year . . . how is that one survives the next 26 years while one does not?

Obviously they will both have the same realized result. FIREcalc takes historical consecutive-year periods and runs the port/withdrawal rate through each period. It doesn't take into account what the market is like when we retire or whether leading performance of the sample period is comparable to leading performance of the retirement date; if it did we'd have very very few samples to work with, if any. Statistically, given the samples, a 4.1% rate survives all sample 30-year periods. A 6.22% rate fails 29.5% of the sample 26-year periods. Where's the discreapancy? In year 4, the real withdrawal rate is 6.22% or thereabouts in less than 5% of the 30-year period samples when starting with 4.1% SWR; in those 5% of samples the 6.22% rate succeeds 100% of the time. Add in the other 95% of 30-year samples and a 6.22%-of-year-4 port rate would fail 29.5% of the time in the back-end 26 years.

In a nutshell, FIREcalc allows for the worst case historical scenario for each calculated period. Since stocks declined 2000-2004, FIREcalc doesn't care and still allows for the worst case beginning now and going forward. If you or anyone else wants to consider the market level, that's fine but it's not statistically represented in all 100+ historical cases.

The historical statsitical math is okay. The statistical analysis is an unbiased report on how the withrawal rate fares in past periods. Once you decide to apply market fluctuation reasoning you are throwing away a portion of the historical samples becaue only a few of the samples match the market fluctuations in the time period you are specifying.

I realize that many of you are uncomfortable with this conclusion. But faulty logic that leads to contradictory results can't be the best way to deal with your discomfort. Maybe it would be better to come to grips with this result. Either you feel comfortable with ER 2004 using a 6.2% withdrawal rate, or you question the universal 4% SWR.

I probably have 15-30 years before I ER, so I don't have to live with the SWR worry for some time to come. The results are only contradictory because there are at least 3 different analytical methods being employed in this discussion of SWR. Statistically 100% of historical 30-year periods support 4% SWR, <5% of historical 30-year periods support 6.2% SWR of year 4's value for the remaining 26 years (amt's hypothesis). Should that scare anyone? No, it's just a tool to help plan, and understanding the tool is important to deduce real strategy.

I guess one of the faces behind this discussion is "can Mr. ER 2004 feel confident with 6.21% for 26 years?" The counter question to that is "does Mr. ER 2000 feel comfortable continuing his current withdrawal rate from 2004 forward?" I don't think anyone here really feels that either is comfortably safe, but I hope this discussion of the SWR tools is constructive and not destructive.

 

[cute fuzzy bunny]

SG -

I took the def of "correlation" from Oxford.

Monte Carlo: Simple monte carlo, by a college text, says "The direct modeling of a random process". Sophisticated monte carlo is defined as "methods recast deterministic problems in probabilistic terms". The simulators I've seen were simple. I've never seen a sophisticated one that applied financial correlation formulae to this concatenation, because I've never seen financial correlation formulae of any kind that apply to how the stock market behaves. If you have seen such formulae that actually work, please point me to them.

I think we've agreed that any correlation that might exist is not predictable or meaningful or actionable. In that case, I will graciously concede its presence to satisfy, because its in effect meaningless either way.

As far as the next 30 years being worse than the depression. The market action leading up to and going through the first downward slide in the depression is remarkably similar to the recent action. They werent engaged with any meaningful enemies then and our home turf was a fairly safe place to live and be. We were not particularly dependent on anything (ie oil) to go about our business and live our lives. Our economy was entirely self directed with very few import/export issues and very little intertwining with other countries and their economies.

Its not so good now.

Sounds like the next 30 years has an EXCELLENT chance of being FAR worse.

But again I'll concede that we simply dont know.

As far as the historical calculators, let me point out two things. Number one is that they all start out with the 1871 data that starts at the end of the civil war and the associated market turmoil. An optimistic place to start...things look pretty good indeed if you start your analysis in May of 1932. I think it was in Bernsteins book that if you lop off the gains at the beginning of this period that bring you back to average valuations that occurred before the civil war, those "growth of $10,000" charts dont look as rosy. Hence neither do a lot of the rest of the charts and analysis. Gummy's presentation of someone surviving horrific experiences starting in 1/1930 are also optimistic since they avoid a large hunk of the crash from 380 to 267ish that occurred just prior to that.

All that having been said, I'm afraid I lost sight of the original discussion.

If you withdraw 6+% from your portfolio going forward, I believe that you will go broke. But thats because...

I believe that if you started with 4% in 2000 and stayed there through today and going forward, where the withdrawal would be 6.21%, there is a good chance you would go broke unless you had a fairly good size portfolio...if you started with under a million in 12/1999, you probably dont have a big enough nut anymore to feel good about bouncing back without a big cut in your SWR. As for me, I sold my mcmansion last year and pumped another $250k into my portfolio, without which my likelihood of survival would be questionable. This is not a damnation of historical calculators, its simply noting that a run using the highest "safe" withdrawal most people agree upon, applied against a pinnacle portfolio at the head of a significant drop, while stocks were at their highest valuations ever...is simply unlikely to succeed.

I believe that anyone that experiences 10+ years of "going nowhere" or losing money in the market, they wont be in it anymore, and anyone who thinks they're brave enough to "tuff it out" is deceiving themselves. After such a period, I'd have my money out and be buying rental properties, or opening a chain of small businesses.

I believe that if you think a 100% number from any calculator says you will make it, you are incorrect.

I believe (as Dory has said and as I've presented in our mortgage debates) that control over your withdrawal rate is crucial, and limiting debt is the crux of that. I just did some calc last night and without hugely effecting my quality of life, I can cut my withdrawal rate to just south of $9k per year and sustain that for at least a couple of years without feeling too badly pinched.

I believe a well balanced portfolio with a good distribution of asset classes is a very good idea and gives you the best chance of a surviving portfolio.

I believe that you should live well and not get too wound up in what one calculator, person, web site, or any other collective organization has to say about it.

I believe you (sg) like a good argument.

(shhh...I want to see if he quotes and disputes that last bit)

 

[haha]

Either you feel comfortable with ER 2004 using a 6.2% withdrawal rate, or you question the universal 4% SWR.  

It's fairly clear to me that most of this debate is between people who could enjoyably live on about 2% of their stash, and those who couldn't. A few people have straight out said this. In other words, we are not so much thinking as hoping.

Earlier in this thread ***** posted a link to what he calls Raddr's research on the Gordon Equation.

http://nofeeboards.com/raddr/gordon.htm

If you want you can skip the text and just look at the scatter plot. This has nothing to do with a choice of mathematical model, one's reasoning, or any other softcore element. It takes a prediction made by the Gordon Equation, and plots it against the actual historical result over the same historical time period used by FireCalc. Looking at the plot, a 4% annualized return is higher than the prediction. Good luck and the standard deviation of about 1.6% could easily get us to 4%. But wait- we have to deal with the negative effects of variability on a portfolio experiencing periodic withdrawals.

The predicted annualized real return for 30 years is about 3%. This was when the S&P yield was somewhat higher than it is today. Hence today's predicted yield would be less. How can an SWR be greater than the average annual return? Only one way- one is helped, rather than hurt, by statistical variability.

Yet the whole purpose of SWR was to demonstrate the folly of Fidelity's silver haired guru telling us that we could withdraw whatever the real market return was. Variability tends to be our enemy, not our friend.

To me anyway, this all adds up to a realization that a 4% SWR using the usual investments is far from an odds-on bet. (No news here to anyone who has read my other posts on this topic!) What the true odds are no one knows. Still, it can be more important to separate likely from unlikely, than to vainly attempt to precisely quantity degrees of likelihood. Warren Buffet, who knows a thing or two about odds, has famously stated that he doesn't even use a calculator to make a go or no-go decision on an investment. Either it is obviously good, or obviously questionable.

Some have said "a 4% SWR is proabably ok, but anyway, if it isn't I can always go back to work." One thing I remember from reading Paul Terhorst many years ago is that it is qute difficult to go back to our previous level of employment. Not to any sort of work; as long as we are healthy there is always grass cutting, delivering travel trailers, running around in a Geo Metro picking up blood samples. But not many of us will be going back to professional jobs. In my own case I couldn't handle the liabilityl exposure, among other impediments. Our skills atrophy. And possibly even more important we have broken a taboo by opting out. Then there is the probablity that our needing to return to work will coincide with a general economic downturn. Stock markets crashing, inflation skyrocketing, and massive and growing unemployment are not the best background to re-starting a career.

So from my point of view, this is a very serious decision. I say this mostly for the benfit of those who have not yet pulled the plug. Those who are regulars here already have their minds made up, and I believe are experienced enough to know what the stakes are.

Mikey

 

[BigMoneyJim]

Self Correction

At least twice in this thread I either implied or stated that salaryguru suggests 6.21% is safe for 26 years. I don't think he claimed that, so it is incorrect for me to put those words in his mouth.

His point is that FIREcalc supports 6.21% 2004-2029 as strongly as it supports 4.19% 2000-2029, and I disagree on the basis that the 6.21% conclusion applies a separate analysis methodology and invalidates much of FIREcalc's supporting data, or at least the method upon which it reports that data.

On the other hand,if one strongly believes that market level correlation is a strong and reliable indicator then FIREcalc's data may appear to obviously support 6.21% 2004-2029, so maybe we're arguing semantics. But if one strongly believes in market correlation one probably wouldn't have gone with 4.19% SWR in 2000 in the first place.

I'm not placing any method above the other but policing (to benefit my learning) the semantics of which methods are being used to support various conclusions.

I keep wanting to say I have the luxury of not having to decide on SWR, but given the choice I'd rather be in the position to worry about SWR.

---

I cross-posted with Mikey. He brings up some very good points. salaryguru is already retired; I am far from it. salaryguru and the other retirees are presumably comfortable with the future prospects and contingency plans. Those who are reading and wondering how much they can safely withdraw are in a much more harried position. For what it's worth, I'm not advocating any particular SWR but just learning the ways others are coming up with the numbers.

 

[sgeeeee]

Re: Self Correction

At least twice in this thread I either implied or stated that salaryguru suggests 6.21% is safe for 26 years. I don't think he claimed that, so it is incorrect for me to put those words in his mouth.

Thanks for noticing that. This discussion reminds me of some discussions I've had with people about politics. When I mention some politician or policy (whether Dem or Rep backed) that seems poor to me, the other person sometimes assumes I am therefore backing the other party. Often their response is to attack some policy or politician from the other side and assume they've "shown me". If I agree with them, they seem to think this evens the score and if I don't, it only goes to prove I'm part of the other party conspiracy. It's happened with both Democrats and Republicans.

I have neither backed the 6.21% withdrawal rate for the recent retiree, nor the 4.1% SWR for the 2000 retiree. I simply point out that they are equivalent today and you either believe they are safe or you don't.

His point is that FIREcalc supports 6.21% 2004-2029 as strongly as it supports 4.19% 2000-2029, and I disagree on the basis that the 6.21% conclusion applies a separate analysis methodology and invalidates much of FIREcalc's supporting data, or at least the method upon which it reports that data. . .

No. You still aren't getting the point. They are the same, BMJ. I am not applying any post FIRECALC analysis. I am simply computing the results from FIRECALC on a year-to-year basis. The analysis is already imbedded in FIRECALC.

Mr. RE2000 retired in 2000 with $1M and a 4.1% inflation adjusted withdrawal rate that FIRECALC told him was 100% safe for 30 years. Today, he has inflation adjusted that withdrawal and lost money in the stock market. Between his withdrawals of over $41,000 per year and his stock market losses he now has aproximately $700,000 in his nest egg. (The amount will vary depending on his allocation, etc.) And his withdrawal rate will be aproximately 6% (or ~$42,000 per year). From his 100% SWR calculated just before he retired in 2000, he thinks he will be safe for the next 26 years.

Mr. ER2004 has a nest egg of exactly the same amount that Mr. ER2000 has today -- $700,000. But he is 4 years older than Mr. ER2000 so only expects his portfolio to last 26 years. (Note that this is exactly the same number of years that Mr. ER2000 has left). Mr. ER2004 has never heard of FIRECALC. He only knows that Mr. ER2000 is a smart guy and wants to follow his example. Mr. ER2004 talks to Mr. ER2000 and decides that he has the same amount of money as Mr. ER2000, invested in the same investments, that needs to last the same number of years going forward. He decides to withdraw exactly the same amount of money as Mr. ER2000. And that translates into approximately a 6% inflation adjusted withdrawal rate.

So, BMJ . . . it seems impossible to me to explain how Mr. ER2000 could not run out of money while Mr. ER2004 could. Yet that is what you conclude will happen.

 

[arrete]

I am so glad gummy has commented. I was going to reocmmend his site in any case. I'm just sorry he didn't mention playing canasta.

There is an incredible amount of info at his site about retirement. You just better be in a mathematical frame of mind. He does lead you through it painlessly, though. Give it a try.

http://home.golden.net/~pjponzo/gummy_stuff.htm

arrete

 

[sgeeeee]

. . . I see a strong correlation between the PE10 that applies at the start of a sequence and the HDBR that can only be determined 30 years later. It's important to understand that, in the event that there is any correlation whatsoever, the conventional methodology numbers are wrong. The conventional methodology makes no adjustment whatsoever for changes in valuation levels. It churns out the same SWR no matter how high or low current valuations happen to go.

This is not true, *****. I appreciate that you are trying to augment the SWR calculation by including valuation considerations in the standard SWR calculators. It is true that conventional SWR calculators do not consider valuation in their calculations. But that does not mean that conventional calculations are optimistic. The conventional SWR looks for the worst case regardless of valuation. If you are correct and high valuation implies a lower SWR should be used, then FIRECALC is going to find the highest valuation periods and identify them as worst case.

 

[sgeeeee]

I took the def of "correlation" from Oxford.

. . .  If you have seen such formulae that actually work, please point me to them.

When you are talking about mathematics it helps to get your definitions from math sources. There are precise methods used to compute correlation. Once data has been plotted it can be fit to a curve and the formula for the curve used to quantify correlation. It is done routinely for all kinds of monte carlo simulations every day.

I think we've agreed that any correlation that might exist is not predictable or meaningful or actionable.  In that case, I will graciously concede its presence to satisfy, because its in effect meaningless either way.

If you believe that, then you certainly should not use a historical simulator. That's the whole reason that people end up using the historical simulation. If you don't think correlations exist or matter, then just take as many arbitrarily chosen years as you want and tack them together to get your full longevity retirement. Then do it over and over again to get as many samples as you want. If you don't think these correlations matter, then there is no need to be limited by the number of years that we have consecutive data for.

. . .I believe you (sg) like a good argument.

But I can't help but notice that you have avoided most of my points while still insisting on having the last word.


(shhhh . . . let's see if he notices)

 

[hocus]

If you are correct and high valuation implies a lower SWR should be used, then FIRECALC is going to find the highest valuation periods and identify them as worst case.

It is possible to determine from looking at the historical database what effect changes in valuation levels have had in the past and then make the necessary adjustment to reflect that factor. But the bubble level valuation levels have never been experienced before. It is not possible to "find" them by looking at the data in the database without making an adjustment.

The conventional methodology does not call for any adjustment for this critical effect. That's why Bernstein says that the results of the conventional methodology studies are "at times misleading." Do you think that Bernstein is wrong about this?

 

[Peter]

Re: Self Correction

So, BMJ . . . it seems impossible to me to explain how Mr. ER2000 could not run out of money while Mr. ER2004 could. Yet that is what you conclude will happen.

That's not what I got from BMJ's post.

What I understood, and agree with, is that both 4% in 2000 and 6% in 2004 are equally risky (or safe). The point is that the historical record does not contain cases where there are two substantial slumps in equity prices a few years apart. Therefore, the 2000 case assumes that there will be a few years to recover from early losses. The 2004 case makes no such assumption, and allows for a possible slump early on. Of course, if this happened, Mr ER2000 would be in the unfortunate position of encountering something that Firecalc had not assumed from existing historical data -- a double slump.

Whether you believe that a double slump is likely or not is another question. But if it happened, 4% withdrawals starting in 2000 would probably lead to trouble.

Peter

 

[raddr]

Hi SG,

As usual you make a lot of good points.   I will nitpick this one though:

From a mathematical perspective, the number of independent data sets is of little concern to me.  If I want to empirically answer the question, "What is the minimum number of heads I will see if I flip a coin 30 times in a row?"  I don't need to start 100 different series of 30 tosses (ie. 3000 tosses) to answer the question.  I can simply flip the coin  130 times and look at the 100 different series of 30 that are represented in those tosses.  

The 100 series you get with the rolling 130-coin toss are much different than what you would get with 100 separate 30-coin toss experiments.  In the latter case you will get a normal binomial distribution of heads (or tails) but you won't with the former because of the data dependency.  IOW, since each 30-coin sequence shares data with 30-60 others the distribution will be skewed to the extent that the 130-flip series is skewed.

As an example, consider the very real possibilty that the 130-toss series consisted of 55% heads, which would be expected to happen about 13% of the time.  By definition you'll get heads an average of 55% of the time in each of the overlapped 30-toss data sets.

On the other hand, the chances of getting an average of 55% heads in the 100 separate 30-coin experiments is virtually nil since you are tossing the coin 3000 times instead of 130.

This is easy to test with a spreadsheet and a statistics package.  As an experiment, I did the overlapped 130-toss experiment ten times, each time analyzing the distribution of the 100 point datasets with a test of normal distribution called "Chi-square test for discrete variables".  This generates a p-value  which if > 0.05 indicates a "normal" distribution.  Well, the p-value was > 0.05 only 3 times out of ten.  The chances of this happening at random are about one in 20 million.  On the other hand, 100 independent sets of 30 will fall inside the "normal" range at least 8 or 9 times out of 10 tries.

Interestingly, in looking at some of the histograms of the data, it appears that the "tails" of the distributions are often truncated with the overlapped data.  This, of course, is precisely the area that we are interested in with SWR analysis which looks for outliers.  Thus the need for as many independent data points as possible for meaningful analysis at the periphery of the data.

Perhaps a real statistics expert such as gummy can comment here and explain it better than I did.
 

 

[BigMoneyJim]

Re: Self Correction

Thanks for noticing that. This discussion reminds me of some discussions I've had with people about politics. When I mention some politician or policy (whether Dem or Rep backed) that seems poor to me, the other person sometimes assumes I am therefore backing the other party. Often their response is to attack some policy or politician from the other side and assume they've "shown me". If I agree with them, they seem to think this evens the score and if I don't, it only goes to prove I'm part of the other party conspiracy. It's happened with both Democrats and Republicans.

If that's how I came accross then I consider it a failure on my part. Oh well, not the first time, not the last time.... (I have also seen what you describe.)

I have neither backed the 6.21% withdrawal rate for the recent retiree, nor the 4.1% SWR for the 2000 retiree. I simply point out that they are equivalent today and you either believe they are safe or you don't.

I can agree on that. I seem to be approaching it from a different angle, but that's where I arrive. "100% safe" doesn't look that way when your port nosedives 4 years into retirement. And "100% safe" is really a misrepresentation. "Would have survived any 30-year span beginning 1871-1973" is more accurate. And even in the historical periods with similar or worse beginning 4-year performances I suspect retirees suddenly had a taste for rice and beans instead of steak.

No. You still aren't getting the point. They are the same, BMJ. I am not applying any post FIRECALC analysis. I am simply computing the results from FIRECALC on a year-to-year basis. The analysis is already imbedded in FIRECALC.

Perhaps then we are arguing about different things. For what it's worth, I like FIREcalc as a reference tool, but shooting for the a 95%-100% historical success rate seems risky to me. I guess you're not arguing the supporting statistics but the problem of a "100% safe" plan that has now put Mr. ER 2000 in a position that is quite intuitively not 100% safe? I agree with the problem but apparently took a different route to get there.

So, BMJ . . . it seems impossible to me to explain how Mr. ER2000 could not run out of money while Mr. ER2004 could. Yet that is what you conclude will happen.

That was not my conclusion; of course they both end up with the same final balance. My point was that FIREcalc shows both have 29.5% historical failure rates at 6.2%/26yr even though Mr. ER 2000 chose a "100% safe" withdrawal rate in 2000. My conclusion was that the statistics & FIREcalc support 4.1%-30yr yet don't support 6.2%-26yr now, and I thought others were arguing they support both. The problem isn't in the math; the problem is pushing "100% safe" to the bleeding edge of historical support for 100% safe rates, or taking "100% safe" as everlasting gospel. Or perhaps the problem is looking to the past to conclude the future.

There are additional problems with pushing FIREcalc to the brink of 99.99999% or 95% safe such as when money is withdrawn in relation to market movements within the year, the quite limited sample set, massive overlapping of historical samples and the disregard of world/US economic, political and technological situations. (The first three being statistical problems with FIREcalc data, the rest clearly out of scope of FIREcalc. Of course we can't fix the sample problems during our lifespans.)

As somebody else mentioned, perhaps the best thing about historical SWR is that it clearly shows WR's higher than 4.1% of the starting portfolio were historically dangerous.

 

[BigMoneyJim]

Re: Self Correction

That's not what I got from BMJ's post.

What I understood, and agree with, is that both 4% in 2000 and 6% in 2004 are equally risky (or safe).

You know what's funny? You're right--well, mostly--and it's what salaryguru said in the first place. I just didn't realize I was arguing nearly the same conclusion until now.

I disagree that it's necessarily equally risky, so I can still argue it's not exactly the same thing, and I can argue that FIREcalc doesn't say 6% in 2004 is 100% safe. : (Although I think I've exhausted all ways of saying it.)

The point is that the historical record does not contain cases where there are two substantial slumps in equity prices a few years apart. Therefore, the 2000 case assumes that there will be a few years to recover from early losses. The 2004 case makes no such assumption, and allows for a possible slump early on. Of course, if this happened, Mr ER2000 would be in the unfortunate position of encountering something that Firecalc had not assumed from existing historical data -- a double slump.

Whether you believe that a double slump is likely or not is another question. But if it happened, 4% withdrawals starting in 2000 would probably lead to trouble.

Precisely.

 

[cute fuzzy bunny]

When you are talking about mathematics it helps to get your definitions from math sources. There are precise methods used to compute correlation. Once data has been plotted it can be fit to a curve and the formula for the curve used to quantify correlation. It is done routinely for all kinds of monte carlo simulations every day.

Okey dokey. A college math text describes correlation as "measures the degree to which two variables are linearly related". Give me two variables related to the stock markets, and show me the correlation formula that linearly relates them. SHOW ME.

If they did exist, they would be incorporated into a sophisticated monte carlo program. To my knowledge, they dont exist, and therefore dont find themselves in a monte carlo. Disagree? SHOW ME.

If you believe that, then you certainly should not use a historical simulator. That's the whole reason that people end up using the historical simulation. If you don't think correlations exist or matter, then just take as many arbitrarily chosen years as you want and tack them together to get your full longevity retirement. Then do it over and over again to get as many samples as you want. If you don't think these correlations matter, then there is no need to be limited by the number of years that we have consecutive data for.

No. I use a historical simulator to examine how a model portfolio with certain dollars and asset classes would have performed historically, with the conditional acceptance that this gives me some idea of how it would have performed historically. I realize that this may give me some indication of how it will perform in the future, but also realize that in the complete, utter and total absence of any mathematical application or surety to a system that is dominated in the short to medium haul by absolutely non mathematical influences, its merely a tool for modelling, and far from certainty. I unconcede the correlations bit, because you are alternately saying there are clear correlations that cause market behavior, then saying they cant be (easily) measured or explained. If they are there, they are measurable or explicable. If they arent measurable or explicable, they arent there. PROVE ME WRONG.

But I can't help but notice that you have avoided most of my points while still insisting on having the last word.

Actually, no. I have responded to every one of your "points" in the course of your discussion. You haven't responded to any of my invitations to show me examples of correlations that most professional mathematicians who have performed extensive analysis of the market say do not exist. You havent shown me examples of the correlation formulas that are incorporated into monte carlo simulations, or a monte carlo simulation that uses formulae to calculate where the market can or should go based on these correlative factors.

I dont think anything I've posted has any bearing on "having the last word". I've just continued to ask for something to validate your opinion. I've tried to ascertain what point you are trying to make by repeating back paraphrased segments of your posts. My failure to do so successfully indicates you are a bad explainer or I'm a bad understander. I'll accept fault, so please tell me what your point(s) are exactly again? Maybe someone else can explain them to me?

My "last word" has been to concur with what I think a lot of other folks feel...that historic calculators, monte carlo simulations, guesses on future returns, and so forth are simply fodder for discussion, and another tool in the box. Certainly not a definitive guide for 100% success or failure.

So...What are your points, which are opinions, suppositions, and facts, and for those, what sources of information back up or validate those points.

Help me help you.