SWR of 6.21% for 26 years

[dory36]

Let me try to offer a different example to show why this is mathematically accurate (with a clarification). I'm ignoring whether you think the future will be worse than the past -- that's a different dogfight. However, if you'll follow the logic below, it doesn't matter what you believe about future vs past *to accept the conclusion of this example.

The clarification: keep in mind that you're now looking at 27 years and not 30 year survival.

Imagine Bill and Bob.

Bill retires on 12/31/00, taking his 4.1%. Bob works 3 more years.

On 12/31/03, Bob retires, and by coincidence he has exactly as much to start his retirement as Bill has left in his portfolio.

The withdrawal that Bill is taking has now risen, due to portfolio losses, to 6.21% of his remaining portfolio.

Now regardless of who retired when, if Bob starts withdrawing the identical amount of money from a portfolio that, as of 12/31/03, has the identical value, then they will both have the identical amount on 12/31 of every following year -- even though Bob's starting withdrawal happened to be 6.21% of his 12/31/03 portfolio.

How could it be otherwise?

Dory36

(Anyone who protests that Bill and Bob will have different tax situations will be fined a dryer sheet!)

[IDunno]

Quote:

Now regardless of who retired when, if Bob starts withdrawing the identical amount of money from a portfolio that, as of 12/31/03, has the identical value, then they will both have the identical amount on 12/31 of every following year -- even though Bob's starting withdrawal happened to be 6.21% of his 12/31/03 portfolio. How could it be otherwise?

You are 100% correct (OK, in my opinion ).

However, the question then becomes "Is this a safe withdrawl rate?" If so, the 6.21% SWR is fine for Bob retiring in 2003, but Bill was stuck with a 4.19% SWR in 2000. We know Bill couldn't have safely taken out 6.21%. What is the difference between Bill's situation and Bob's? The only difference is what we know the history of the stock market for the past few years.

So I believe that the 6.21% SWR theory is flawed, *unless* we can assume that the decline in the stock market for the past few years indicates that better years are to come (which can easily be debated either way).

Going back to the flaw in the original logic that came up with the 6.21% SWR, let's exaggerate a bit. Let's say that Bill retired 28 years ago with $1,000,000 and his 4.21% SWR rate, and has $20,000 left today. According to the theory that came up with a 6.21% SWR rate at the beginning of this thread, Bob can retire today with $20,000 and have a SWR of 200+% (since Bill is taking out $42,100 per year, that's what Bob takes) for 2 years. I believe this correlates exactly with the original scenario, just with a difference in the number of years (28 versus 3) and the fact that it is based on a hypothetical stock market rather than the real 2000-2003 results.

I don't have an opinion (yet!) on whether or not past results can help predict future results (the basis for why 6.21% may be OK today). But I do believe that the original basis for the 6.21% SWR is flawed (although I didn't think it was flawed about 6 months ago, when I too had the brilliant idea that the 6.21% SWR would be OK -- and almost posted here about it).

[charlie]

TH,

I was siding with salaryguru's position that there
is some correlation year-to-year on stock market
returns. I know it is difficult to measure this
correlation and impossible to predict short or
even intermediate term returns but according to
Bernstein/Bogle et. al., the long term trend is
very predictable via the Gordon Equation. My
point is that there are "long" cycle forces like
the business cycle and inflation that continue
over a period of years and, IMHO, influence
market returns. The dog may walk in ramdom
directions but as long as he is constrained by
the leash of GDP growth, his average direction
will be north. Thus, I believe that the logic
behind FIREcalc is valid and the coin-toss analogy
is not valid.

But what do I know?

Cheers,

Charlie (aka Chuck-Lyn)

[Peter]

Maybe I'm missing something here, but it seems to me that there's no mystery about different SWR estimates at different times.

Looking at the historical record, there were times when a given SWR resulted in a large remaining portfolio after some given number of years. At other times, the same starting portfolio, with the same SWR, would be totally exhausted after the same number of years.

Obviously if you start in a 'good' year, you can draw more than if you start in a 'bad' year. The problem is that you only know about 'good' and 'bad' in retrospect.

And while it's tempting to say that the decline in values over the last few years suggests that now may be 'good', there is no guarantee that this will turn out to be the case.

Peter

[cute fuzzy bunny]

Charlie -

You know a lot, and as you know the gordon equation and I are agreeable.

That having been said, while the gordon equation is strong at 30+ year measurements, in less than that its no predictor. We've had 3 periods of 20 year "hard times", and those were all before we had the potential for attacks on our own soil. Its been a long time since that happened. While history and common sense tells us that there is a good chance that all may go well, a gordon equation fitting scenario of 27 years straight down or sideways, followed by 3 years of 30-40% upticks still "straightens out the dog".

The question is: will the average portfolio survive long enough to make it past those 27 years and will the average retired investor leave their stocks in play for that same period of time waiting for the eventual uptick. I think the answer to either question is a strong "unlikely". Coupled I think its a clear "no".

The other thing is that Bernstein accorded a second formula/scenario to go along with the Gordon equation. The trend of a civilization/market over its term of existence.

As a civilization has become more developed and settled, its risks (and market returns) become progressively lower. I remember some kind of bell curve in the "four pillars" that showed returns swelling up (along with risks) as a market emerged, then a long slow slide as risk became reduced further.

I would imagine the greeks and romans, living on top of the world, as they grew more obese, less interested in the goings on of their government, and more interested in the amusements of the moment...they never expected it was all about to end. Sound familiar? Same for the Bernstein example I used in my "book report" post where the pre-WWI londoner feels the world is his oyster.

Something in the next 30 years is going to bang up the markets. Whether its a depression-type scenario or a 65-84 scenario, those scared a generation away from the stock market, WWIII breaking out in the middle east, or a dirty bomb detonated over NYC or Washington. Or maybe just the effects of trickling out a few billion dollars at a time trying to fight an unstructured enemy with structured approaches strong in hindsight, followed by billions more investigating why we failed and whose fault it was.

But as usual I wander far afield. The original point was that a good year in the market doesnt dictate a bad one is forthcoming, nor vice versa. Its reasonable to say that a good one will come eventually, but its unreasonable to say that any past pattern becomes a predictor of the future. Hindsight says that over long hauls it should all work out, but the length of the long haul is past the lifespan of many of us. And while the two guys with the same size portfolio making the 6.21% withdrawal will be in exactly the same boat, there isnt anything saying that either of them will get to the finish line with a positive bank balance.

[BigMoneyJim]

I may make a fool of myself here, but I learn more when I do so:

This kind of comparison we're talking about here--that since 4.1% was safe 4 years ago, and that's 6.21% today given the past 4 years and therefore 6.21% is safe--may be like the Stevie Wonder logic disconnect:

• God is love
• Love is blind
• Stevie Wonder is blind
• Therefore, Stevie Wonder is God

The SWR is calculated based upon yearly (?) samples from market history.

The simplest counter to amt's proposal is that I go to FIREcalc, enter a portfolio of $1mil, a yearly withdrawal of $62,100, lifespan of 26 years and leave all other entries alone and it tells me the success rate based on historical data is 70.5%. It doesn't matter that I used $1mil here, what matters are the percentages, and according to FIREcalc that withdrawal rate would survive any historical 26-year period 70.5% of the time.

I think the logic disconnect here is the assumption that year 4 of this 30-year period is comparable to year 4 in all the sample historical 30-year periods; it is not; in fact it's unique. If we had a hundred or so historical samples where years 1-4 performed exactly as 2000-2004 then maybe we could use those as a guide, but we don't have that data.

Even more simple: If you are confident enough in the historical analysis method that you believe 4.1% is 30-year safe, then why do you suddenly disagree when the same method and the same calculator tells you 6.21% is not 26-year safe?

Here's a project for someone who loves this stuff: Look through the historical data and find 30-year periods where the surviving portfolio of 4.1% is 6.21% in year 4. (I believe this means the overall return from years 1-4 equals the return of 2000-2004.) Find how many periods there are like that and decide if it's statistically significant. If it is, calculate the 26-year survival rate for years 5-30 in those sample periods and tell us the number of samples and the results.

EDIT: Oh yeah, I echo what Bob Smith said. When getting into the numbers and trying to predict the future it's easy to get all wound up and even worried that what we've done isn't good enough. Now that I'm readying to switch from debt reduction to accumulation I find myself wanting to save half my salary until age 65 (I'm 34) to be sure I'm safe, but the truth is that through controling expenses and thoughtful investing we are way ahead of where we were: non sum qualis eram. We can't account for all contingincies, but we're making hay while the sun shines and that's the best we can do, because "the best laid plans o' mice an' men. . ."

Man plans and God smiles.

John Galt

[cute fuzzy bunny]

Amen brother.

[Michael]

Quote:

Many are the plans in a man's heart, but it is the LORD's purpose that prevails.

Proverbs 19:21 NIV

[sgeeeee]

amt is correct mathematically. If you don't like his conclusion, then you have to reject all conclusions from the FIRECALC program if you are going to be consistent. There are reasons why you may choose to reject historical simulation results (past performance is no guarantee of future results) but rejecting amt's results while accepting a SWR result is inconsistent.

Regarding correlations of performance and inflation over time: Companies make investments today that pay off in days, months and years to come. Companies make investments today that cost them dearly in days, months and years to come. Governments make laws and invest in ways that affect industry in months and years to come. Thus, there is correlation (in fact, cause and effect) between performance today and in the future. It is not direct and is too complex to quantify in an equation, but it is real. If this were not true, then an investor would always be ill-advised to invest in stocks or bonds. When people on this board argue in one breath that they are waiting to invest in real estate or TIPS or . . . till the market improves, but then argue that there is no correlation of performance over time, they are being inconsistent. If there is no correlation, then these investments are as likely to go up as down over any time period.

Similarly, if financial results of today are unrelated and uncorrelated to results of the past, then there is no reason to use a historical simulator. You would be better off just randomly selecting one year's performance after another to assemble a retirement portfolio performance. You could make up an almost infinite number of histories to examine to determine a SWR.

A lot of people who use historical simulators don't seem to be comfortable with the implications pointed out by amt. But if you reject his conclusions, then you must believe that FIRECALC produced invalid results for the retiree who chose to retire in March 2000. Because the retiree who quit then with a 4.1% SWR is currently withdrawing 6.21% from his/her existing portfolio today and still has the same expectation for the longevity of that portfolio.

If you don't feel comfortable using this result in retirment, then don't do it. Like I said, there are reasons to question any result from a historical simulator. But you are wrong if you are looking for a mathematical flaw in the reasoning.

[charlie]

My brain hurts from thinking about all this. It really
does not matter much to me since I plan to spend
what I need each year and trust that the Lord
will provide in the future. This arrangement has
worked OK for me for the past 70 years.

Cheers,

Charlie (aka Chuck-Lyn)

[sgeeeee]

Quote:

. . . The problem with the conventional methodology is that it looks at the entire data set for its data, but then uses the results from only a single data point to generate its conclusion. The valuation levels that produced numbers close to 4 were the valuation levels that applied in 1929 and 1966. There are only 2 data points showing that a 4.1 withdrawal worked at the highest pre-bubble valuation levels, not 130.

That's not really accurate, *****. The historical simulation SWR analysis is based on the assumption that the future is not likely to be any worse than the worst case that ever happened in the past. Since there will only be one worst case in the past, the simulator looks for it and finds it (based on your asset allocation, expense ratios, etc.). By definition there can't be more than one worst case, so the simulator finds it and determines what it would take to retire then. Again, by definition, if your portfolio would survive the worst case, it will survive cases that aren't as bad. But in fact, FIRECALC and other historical simulators I am familiar with go ahead and evaluate the performance of your portfolio for every single available case and let you know how much over-performance your asset allocation and other inputs would have resulted in for each of those cases.

Now. . . for almost all realistic investment allocations, it turns out that the same couple of starting retirement dates proves to be worst case. That hardly seems surprising. In fact, it would be very unsettling if it were otherwise. If there were dozens of equally bad financial histories that RE's needed to worry about, it would imply that the probability of facing tough times in retirement was very high and that almost no strategy would be safe.

Quote:

. Two data points is just not enough to support reasonable claims that a 4 percent withdrawal at those valuation levels is safe. What JWR1945 and I do with our data-based SWR model is make adjustments for changes in valuation, which permits us to make use of the entire data set in generating our SWR findings. My recollection is that the number JWR1945 tentatively came up with for retirements starting at valuation levels equal to the highest pre-bubble valuations was 3.3, not 4.1. We still have work to do in confirming those tentative findings, but it does not appear to me that the 4.1 number is going to stand even for the pre-bubble period. In the bubble of the late 1990s, valuations went far higher than where they were in 1929 or 1966. So the SWR we get for retirements beginning at bubble-level valuations is much lower yet.

There are serious mathematical problems that need to be addressed before you should come to this result. To start, you are going to have to complete a principle component analysis and prove that your valuation metric is correlated to SWR in a statistically significant way. Then quantify the SWR vs valuation distribution. And finally apply that to the valuations that existed in early 2000. When you finish all that, you could conceivably produce a probability that your SWR is higher or lower than the tradional SWR by a given amount. You still have problems with this analysis, however, since the correlation you find in the first part of this analysis would not be valid outside of the range of valuations observed in the past. Your analysis is empirically derived, so cannot be legitimately extrapolated outside the range of empirical data.

Quote:

. . . The root question is--Do changes in valuation affect the determination of what withdrawal rate is safe or do they not? The data we have assembled reveals a strong correlation between the PE10 (the valuation assessment tool endorsed by Robert Shiller) that applies at the retirement start date and the highest surviving withdrawal rate that ultimately applies for the 30-year historical return sequence that follows (we refer to the latter number as the historical data base rate, or HDBR.)

Although PE10 is a better valuation metric than simple PE for certain types of analysis, there are serious problems using it to determine modifications to existing SWR. If you do ever get around to performing legitimate mathematical analysis and testing it, you need to keep in mind that a valid valuation metric predicts both the future trend as well as the time frame. This is important for the retirement SWR calculation because the portfolio longevity will varry significanlty for different people. A completely accurate valuation for a 10 year portfolio may well be completely inaccurate for a 40 year portfolio. Your current analysis ignores the importance of time in the valuation entireley.

[BigMoneyJim]

Quote:

amt is correct mathematically. If you don't like his conclusion, then you have to reject all conclusions from the FIRECALC program if you are going to be consistent.

But FIREcalc tells me that 6.21% is only 70.5% safe for a 26-year period.

I still think there's a logic error in saying that today's year-4 withdrawal rate for Mr. ER 2000 is a safe 26-year rate for Mr. ER 2004 given the same 2004 balance.

. . .
(lots of thinking and some tinkering with FIREcalc and Excel and a calculator)
. . .

Okay, I think I get it now. To oversimplify it, the argument supporting amt's supposition is that 1929-1933 didn't happen twice in a row or twice with no interim recovery. So if we were safe with the original model then we're safe with the remaining portion of that model.

If I disagree with amt's supposition then I'm basically saying 1929-1933 (or 2000-2004) could happen twice with no interim recovery. If I say that then why did I agree with the model in the first place since it doesn't allow for that happening during the 30-year period?

For 30-year period beginnings, 2000-2004 was better than 1930-1934 and 1929-1933 and falls between the large gap between those and 1937-1940. (Using amt's example of 4.1% of port at start of 2001 is 6.21% of port at start of 2004 and the yearly withdrawal as a percentage of the Yr 3 column in FIREcalc's detail; that's probably a backward way of looking at it.)

Ow, my brain hurts now. I can't decide if this boosts my opinion of amt's supposition or decreases my confidence in the historical comparison method.

EDIT: I think I counted wrong and should've used the year 4 column from the FIREcalc examples, but the oversimplified conclusions remain the same.

[wzd]

One problem with firecalc, that has been discussed elsewhere, is the fact that recent years data results are not used as the beginning of a period in the test runs. How do you know that the 4+% withdrawal started in 2000 is not one of the failure cases? In fact, this discussion supports concluding that it is!

One way to add to firecalc is to find a way to complete the runs started in recent years. Some have suggested cycling back to the beginning of the data. I have been a proponent of taking the adjusted balance, withdrawal (adjusted for inflation), etc. and running firecalc for the remaining years in the period....sounds a lot like this discussion, doesn't it!

IMHO, one possibility to consider strongly is that the portfolio of the retiree from year 2000 is in the 5% of the failures. And even if you run at 100% survival rates, there is no guarentee, and you could still be in a failure scenerio. Perhaps retiring in 2004 and withdrawing 6.21% is simply matching that failure scenerio.

Wayne

[Bob_Smith]

Quote:

Perhaps retiring in 2004 and withdrawing 6.21% is simply matching that failure scenerio.

Wayne, I think you are correct. We know intuitively that 6.21% runs a high risk of failure going forward. If we accept that, then we must also accept that 4.21% in 2000 runs a high risk of failure. It seems to me that anything exceeding dividend yield (plus around 1% or so) is automatically suspect.

SG, your analysis is incredibly articulate and thorough. In fact, the brain power displayed by so many in this thread makes me very reluctant to jump in!

[cute fuzzy bunny]

Just to point out, the 1929 scenario occurred when our market was still somewhat emerging and speculative. The 1964-84 scenario was similarly still in a higher risk/return scenario. The firecalc runs for long durations (like we calculate) dont include full periods for times since 1984 when we would be closer to current dynamics.

Then again, our real rate of return for the last 100 years adjusted for inflation and taxes is a little over 4%. If Bernstein is right and our real rate of return over long hauls is 3-3.5%, then thats our SWR at the high end. With a strong stomach to take long down periods and a sizeable enough portfolio and/or low enough withdrawal rate to survive "bad times".

As far as correlation in returns or anything else from one year(s) to the next year(s), if anyone wants to bet what investment vehicles will return what for the next 5 years, I'll be willing to put my money into those vehicles. If they return the predicted rates (or more), I'll give them half the profit. Otherwise they pay me a fee of 25% of my current portfolio.

Any takers?

I dont want to seem irritating here, but I thought it was already well accepted science among us that "predicting" the market or having a "plan" that produced or predicted results was simple snake oil.

[hocus]

If you do ever get around to performing legitimate mathematical analysis and testing it....

I developed the data-based SWR tool in early 1996 and have been using it and refining it ever since. I did not do any formal statistical testing of the concept in the mid-90s. JWR1945 has helped me with the statistical analysis with the work he has done during the two years of the Great SWR Debate, particularly with the work he has put forward at the SWR Research Group board at NoFeeBoards.com.

I am not able to respond to your hard-core statistical points, SalaryGuru, because I lack the background needed to respond effectively. I very much would like to have the questions you raise put to rest, however. I plan to take the Data-Based SWR concept public in a big way in coming months and I am going to make a big fool of myself if it turns out that the claims I am putting forward are not backed by a reasoned analysis of the historical data. I would much appreciate it if you would be willing to take a look at the research that JWR1945 has put forward at the SWR board, and to post any questions you have about the methodology he uses at that board for his review. I very much want to get this right.

I held back from going public with all of this for a long time because I wanted to be absolutely certain before I did so. It is one thing to be revealed as a fool on an internet discussion board, it is something else to be revealed as a fool in an article in Money magazine or in a radio interview or in a book you put forward into the public arena. I am quite sincere in my request of you and any other community member who has doubts about the analytic validity of the data-based SWR tool that they please come forward with any lingering questions or concerns. It's OK with me if you do that here. It's also OK with me if you do it at the SWR board.

My goal from the start has been not so much to win a battle of ideas as to determine for certain which idea is correct. It does not seem posssible to me that both the conventional SWR methodology (which assumes that changes in valuation levels have no effect) and the data-based SWR methodology (which assumes that changes in valuation have a significant effect) could be analytically valid. One of these assumptions is wrong, and it seems to me that the best way to determine which one is wrong is to look at the historical data and see whether changes in valuation have had an effect in the past.

I find your post encouraging. You obviously are not agreeing with my SWR claims. What I like about the post, however, is that you are taking my claims seriously. You are not attacking me as a person, you are attacking the logic of my claims. It is that sort of feedback that I seek when I post at boards like this. So I am grateful for the post. I hope that you will pursue the questions you have put forward in more depth either here or at the other board.

I hope that no one has gotten the idea that I am unwilling to re-examine old positions of mine. I am very much willing to do so. I had 90 percent confidence in the data-based SWR concept on May 13, 2002, and I have 99 percent confidence today because of new things that I have learned in the course of the debate. But there is always that 1 in 100 chance that I messed up somewhere. Any poster who puts something forward which leads me to conclude that that is in fact the case has done me a huge favor. It's because discussion boards provide access to that sort of feedback that I see them as an important communications medium of the future.

[sgeeeee]

Quote:

. . . Ow, my brain hurts now. I can't decide if this boosts my opinion of amt's supposition or decreases my confidence in the historical comparison method. . . .

Well. . . at least you're asking the right question.

None of us can be sure of the accuracy of the historical SWR for the future. And I know of no mathematical arguments to change that.

[sgeeeee]

Quote:

Just to point out, the 1929 scenario occurred when our market was still somewhat emerging and speculative. *The 1964-84 scenario was similarly still in a higher risk/return scenario. *The firecalc runs for long durations (like we calculate) dont include full periods for times since 1984 when we would be closer to current dynamics. Then again, our real rate of return for the last 100 years adjusted for inflation and taxes is a little over 4%. *If Bernstein is right and our real rate of return over long hauls is 3-3.5%, then thats our SWR at the high end. *With a strong stomach to take long down periods and a sizeable enough portfolio and/or low enough withdrawal rate to survive "bad times". As far as correlation in returns or anything else from one year(s) to the next year(s), if anyone wants to bet what investment vehicles will return what for the next 5 years, I'll be willing to put my money into those vehicles. *If they return the predicted rates (or more), I'll give them half the profit. *Otherwise they pay me a fee of 25% of my current portfolio. Any takers? I dont want to seem irritating here, but I thought it was already well accepted science among us that "predicting" the market or having a "plan" that produced or predicted results was simple snake oil.

Irritating? . . . No, but I am confused. TH, there is a correlation of tomorrows performance results with the performance yesterday. That does not imply that the correlation is 1 (ie completely predictive) or even that I can quantify it.

But you seem to acknowledge this on the one hand, while rejecting it on the other. You seem comfortable with Bernstein's predictions for the market 30 years out. How could any prediction be made about performance in the future if that performance were not correlated to today's financial situation? Don't confuse correlation with detailed predictability.

[BigMoneyJim]

Quote:

Well. . . at least you're asking the right question.

Thanks. That's good to hear. This is my first serious foray into the SWR topic; before now I've taken it as a guideline that has been considered and approved by others whose opinions I've come to trust.


After sleeping on it I woke up with this thought: if we accept that periods like 2000-2004 can be eliminated from the sample pool for the next 26 years then our number of historical samples has decreased significantly.

And if it's okay to project backwards from now to 2000, then why did Mr. ER 2000 not project backwards a few years to take what Mr. ER 1996's current SWR is?

I just looked it up: if Mr. ER 2000 projected back 4 years he would find that Mr. ER 1996 who started with a 4.1% withdrawal is at a 2.21% withdrawal rate in 2000.

If Mr. ER 2004 looks at Mr. ER 2000 and takes his 6.21% withdrawal rate, then why shouldn't Mr. ER 2000 have looked at Mr. ER 1996 and taken the 2.21% withdrawal rate?

I'm not awake enough to calculate how many sample periods we've eliminated, my my intuition says we've eliminated between a quarter and a half of the samples.

However, if we're using the past 4 years to eliminate historical 4-year periods from our 26-year period, then aren't we bound to take only 26-year samples in which the previous 4 years behaved similarly to 2000-2004? If that's the case then our sample size has dropped to 4 or so historical periods depending on how closely you want to match the year 1-4 overall return.