A Year's Safe Withdrawal Rate #2

Research on Safe Withdrawal Rates

Moderator:hocus2004

JWR1945
***** Legend
Posts:1697
Joined:Tue Nov 26, 2002 3:59 am
Location:Crestview, Florida
A Year's Safe Withdrawal Rate #2
I am posting this again because the number of replies count in the original thread is in error. There have been more posts than indicated.

A Year's Safe Withdrawal Rate

Now that we are beginning to design retirement portfolio strategies, it makes sense to define a special term, a Year's Safe Withdrawal Rate. It is an exceedingly powerful concept. Failure to differentiate between a Year's Safe Withdrawal Rate and the traditional uses of the term Safe Withdrawal Rate has caused us no end of troubles. IMHO, if we were to substitute the term, a Year's Safe Withdrawal Rate, for hocus's use of the words Safe Withdrawal Rate, we would have a whole lot better idea of what he has been saying all along.

I will start with the traditional historical sequence method of calculating a Safe Withdrawal Rate. Using that method, you can calculate a separate Safe Withdrawal Rate for every start year in the database. I know. I did it for the years 1921-1980 using Captain Bill's (dory36's) FIRECalc calculator. I have reported those results. It is meaningful to identify a Safe Withdrawal Rate for 1959, for example. It would be a Year's Safe Withdrawal Rate. It would be the one for 1959.

In the traditional study that uses the historical sequence method, only one number is identified. It is the lowest of all of the Safe Withdrawal Rates examined. It is not even necessary to calculate a Safe Withdrawal Rate for each and every year. It is only necessary to increase the withdrawal rate until you encounter a one failure, and then to back off minimally until there are no failures. That final rate is reported. It is the lowest from among the Safe Withdrawal Rates for individual years. It is a lower bound.

If one assumes that there is no relationship whatsoever between stock prices and Safe Withdrawal Rates, that single reported number would be sufficient. But there are ways to relate portfolio safety and valuations. That means that a Year's Safe Withdrawal Rate is meaningfully defined in terms of valuations. It is not necessary to confine yourself to using the lower bound. In a very real sense, the traditional Safe Withdrawal Rates based on historical sequences are not valid for estimating any individual Year's Safe Withdrawal Rate. And since we normally associate such calculations with financial projections, the results of traditional Safe Withdrawal Rate studies have minimal utility going forward. In a very real sense, traditional Safe Withdrawal Rate studies have not incorporated valuations. They have looked at a collection of years and there is a range of valuations associated with those years. They have not extracted the relationship between the valuation and the Safe Withdrawal Rate of each individual year. We can do that now.

It has been possible all along to incorporate valuations into Monte Carlo calculators. One specifies the mean and standard deviation (or standard deviations) of his investments as inputs. The Gordon equation translates a measure of valuation into an estimate of the mean. It can be modified for use with several measures of valuation.

The Monte Carlo approach will always allow for a continuous range of probabilities, whereas the historical sequence method is limited by having a discrete number of years. It is common to set a failure rate a 5% (or 95% probability of success) when extracting Safe Withdrawal Rates from a Monte Carlo model. It is best to form a similar estimate when using the historical sequence method. It is seldom done, but it is applicable. The general idea is to estimate the probability distribution in the 5% to 10% region by curve fitting, weighted heavily in favor of the 20% to 25% of the years that failed at the lowest withdrawal rates. When something of this nature is done, it can be meaningful to talk about a Safe Withdrawal Rate (at the 95% level of safety) that can increase as well as decrease as the number of years vary. With this variation, which always occurs at levels of safety other than 100%, it is meaningful to talk about an average of Safe Withdrawal Rates as the number of years vary.

Being able to estimate each Year's Safe Withdrawal Rate gives us a powerful tool for designing strategies for retirement portfolios. We can do that now. When we exclude the 1881-1920 anomaly, we can relate P/E10 closely to a Year's Safe Withdrawal Rate. We can use the Gordon Model with inputs based on stock market valuations to produce a complementary estimate of a Year's Safe Withdrawal Rate. We have already identified the mechanism for failure at times of high valuations. We are entering an era when we will Specify and Design retirement portfolio strategies.

Have fun.

John R.

JWR1945
***** Legend
Posts:1697
Joined:Tue Nov 26, 2002 3:59 am
Location:Crestview, Florida
hocus wrote this post.

If we were to substitute the term, a Year's Safe Withdrawal Rate, for hocus's use of the words Safe Withdrawal Rate, we would have a whole lot better idea of what he has been saying all along.

I'm an optimist, JWR1945. But you are really an optimist! I think optimism is good, so I offer no objection to what you say here. I do feel a need to note, however, that there are alternate explanations of the "goings on" that we have witnessed over the past 14 months.

My sense is that the problem is the one that Bernstein refers to in Chapter 2. He says that most investors understand on an intellectual level that there is a connection between valuation levels and returns, but that most have difficulty accepting this on an emotional level. I think that apreciating that point is key to understanding SWRs. SWRs are largely about numbers and data, but not entirely so. An important part of the equation is the emotions that the numbers and data generate. The reasons why I have become an "expert" on SWRs is not that I am good with numbers and data; I obviously am not. I believe that I have been able to see things relating to SWRs that most others miss because I have a special facility in understanding human emotions. To understand SWRs fully, you need to possess a keen grasp of the emotions that the various numbers evoke, and my sense is that I am better able to grasp some of that stuff than some others, including some of those with the greatest facility with numbers.

In the early days of this, I thought that the key was proving the case. I thought that, since the SWR was the product of a calculation, the way to make my case was to show that the calculation was wrong. When I found the Bernstein book, which shows conclusively that the calculation done according to the conventional methodology produces wrong results, and that did not convince a good number of people, I saw that I had been wrong to put so much focus on the numbers. The cause of the strong reactions to the subject is not a concern over how the calculations are done. It is a concern over what the product of those calculations mean.

The number is used by people, and people have emotions. People use money to achieve their life goals, and people care about those life goals. This number is telling them that the ways they invested in the 1990s were not the best ways. That's extremely disconcerting. It wasn't a few people advocating the investment strategies that were followed in the 1990s. It was lots and lots of people. You were hearing the same story everywhere you turned. This number is telling us that the advice we heard for all those years was fundamentally flawed. It is not true that stocks are always the best investment class if you are investing for the long term. That's the most important message that this number is telling us. That's in conflict with the message that most money experts were telling us in the 1990s. So accepting that the new number is the right number means accepting a big change in how you go about evaluating investment options, and it is often difficult for human beings to accept dramatic changes in how they think about things that affect their capcity to realize their life goals.

I think that that is the primary explanation for the reaction we have seen to the idea of discussing the realities of SWRs. I think that people sense the dramatic change in investing styles inplied in acceptance of the new numbers, and they are extremely uncomfortable with the implications. I think that will change with time, but only with time. I don't think this is something that you can force on people. I think we have an obligation to present it to people, and then to allow them to make the decision as to how much of the realities they can let in at any given moment. You can't force people to accept in one moment dramatic changes in a way of thinking that they have developed over the course of a good number of years, in my view.

All that said, I certainly would be glad to see your "A Year's SWR" idea bring a few people around. It is certainly true that a core compliant that I have with the conventional methodology is that it does not tell you the SWR that applies for any particular year. That's a big, big problem with the conventional methodology, in my view. You need to know the number that applies in a particular year to make effective use of the tool. And the conventional methodology does not tell you what it is.

If one assumes that there is no relationship whatsoever between stock prices and Safe Withdrawal Rates, that single reported number would be sufficient

That's right. It is an implicit premise of the conventional methodology that there is no relationship. Researchers who employ the conventional methodology are putting their names behind this premise. They are endorsing it by using this methodology to calculate SWRs. Buying into a false premise can often get you into trouble, and that is what is happening to the researchers who use this methodology. They are buying into a false idea without giving much (if any) thought to it, and that false idea is ruining all their efforts that follow. You can try and try and try to get all the calculations right, but if you started from a false premise, you end up generating bad numbers as your final product.

In a very real sense, the traditional Safe Withdrawal Rates based on historical sequences are not valid for estimating any individual Year's Safe Withdrawal Rate.

Thanks for saying that. That's exactly what I am saying.

The only really important difference between your position and mine that I can make out is that I think that what you say above invalidates the methodology. The purpose of SWR analysis is to calculate a number for a particular year, the year the retirement commences. If the conventional methodology is not capable of generating the correct number for that year, it is invalid. That's my take.

You have got to know the number that applies for your particular retirement, or most of the tool's value is lost to you. The conventional methodology was not designed in such a way as to reveal this number. So it doesn't do the job it purports to do. It doesn't tell you the SWR that applies for you. It tells you something else, but it does not tell you the SWR that applies to someone who retires at a particular point in time, as all retirees do.

In a very real sense, traditional Safe Withdrawal Rate studies have not incorporated valuations.

Again, thank you for saying that. I think that's right, but in the interests of being a little more precise, I am going to add a wrinkle.

The studies do incorporate valuations in a way. They don't so it in a way that allows you to know the SWR for the year you retire. The conventional studies are invald for purposes of calculating the SWR. But there is valuation stuff mixed in the data somewhere. There were various valuation levels experienced during the historical period and those changes are reflected somehow in the data. Isn't that so?

My sense is that the conventional methodology does indeed incorporate valuations, just not in a way that allows for calculation of the SWR. The conventional methodology mixes together valuation inputs for all the various time periods. There's data in there relating to low valuation years and medium valuation years and high valuation years, and it all gets aggregated into one big bowl of data soup. I don't think that is the proper way to determine the SWR for any particular year (which is the purpose of the exercise). The impact of the data from low valuation years is cancelling out the impact of the data from high valuation years. The number being produced is a number that I believe is probably close to an accurate number for the medium valuation years, but not for the high valuation years or the low valuation years.

You seem to be saying that the conventional methodology gets the number wrong in the bubble period and in the pre-bubble low valuation years, but that it gets it right in the pre-bubble high valuation years. Perhaps. It's clear that it gets it wrong in the two scenarios in which you say it gets it wrong. I am not personally convinced that it gets it right in the pre-bubble high valuatiion years. I am open to being convinced on this point. But I would like to see data supporting the idea before buying into it.

I think that you may be confusing the concept of a safe withdrawal rate and the concept of a lucky withdrawal rate. We know that the 4 percent number survived for a retirement that began in the year 1929. That does not prove that this was a safe number for a retirement that began in 1929. There are two ways that a retirement based on a particular number can survive. It may be that the number is a safe number, and it may be that the number is a lucky number. A safe number will almost always work. A lucky number may work only once or twice and fail in all other circumstances.

How many data points do we have indicating that the 4 percent number is a safe number for the 30-year period beginning in 1929 and extending to 1959? We have one. There were no other occasions prior to the bubble period at which we were at that valuation level. So we don't really have much reason to say that the 4 percent number is safe at that valuation level. It's a question we need to explore, in my view, not one that we already know the answer to.

There is a lot of evidence on the table pointing to a conclusion that the 4 percent number is not safe for a retirement beginning at the valuation level in effect in 1929. The SWR is a calulation that is always done at the beginning of a retirement, never at the end. It is not reasonable to say that, because a number worked one time, it is therefore to be deemed "safe." We need to check this out, in my view.

I lean toward saying that the conventional methodology is invalid even if it turns out that the 4 percent number really is safe for valuation levels equal to those in effect in 1929. Even if that is so, it is still true that the conventional methodology produces bad numbers in bubble periods and in low-valuation pre-bubble valuation levels. That's enough for me to say that analysts should not use this approach. But one of the questions that I would like to tie down at some future date is this question of whether the 4 percent number is truly safe at 1929 valuation levels. My sense is that it is not. A lot of people (including you) seem to be coming down on the other side on this question, but each time I find myself starting to waver on this point, I pull out some of the data we have looked at and again wonder if that 4 percent number is truly safe at all pre-bubble valuation levels. This is the outstanding question that I would most like to get a better fix on at this point.

Say that it turns out that the 4 percent number is not truly safe at 1929 valuation levels. What would that tell us about what that 4 percent nunmber signifies? It could be that it is a meaningless number, it's just a number that pops out when you employ an invalid methodology. Part of me believes that that is the case. But part of me tells me that perhaps there is some significance to the number. The conventional methodology does incorporate valuation, it just does so improperly. So I wonder if the valuation effect is somehow being reflected in the results of the conventional studues, even if in an improper way.

This is why I have come to believe that the 4 percent number may be a rough average of the SWRs for the various years. I do not say that this is so for sure. It's just a theory. And I do not say even in the theory that the 4 percent number is the exact average of the various SWRs. My sense is that it might be a rough average. The number 4 does not appear to me to be a randomly generated number. The data we have seen suggests to me that it is not too far from what the true average SWR is. Perhaps the average SWR is 3 percent or perhaps it is 5 percent. I am not saying that it is exactly 4. I am saying that the conventional methodology may be telling us that 4 is a number in the neighborhood of the average SWR. That's not an entirely insiginficant information bit.

The "average" question is not a big deal. Whether 4 turns out to be the average SWR or not will not have a big effect on the question of whether the convetnional methodology is valid or not, and that is the most important question before us. If I am right about this, it makes my case stronger. If I am not right, it does not undermine my case too much. It would still be true that we have shown that the conventional methodology turns out wrong numbers in bubble years and in pre-bubble low-valuation years. So it is a mistake to place too much focus on the question.

The reason that I brought it up the other day is that I find it helpful at times to puzzle over the question as to what that number 4 really signifies. Perhaps it signifies nothing. Perhaps it is the correct SWR for pre-bubble high valuation years. But perhaps neither of those two things is so. I think that it is too early in the game to be making definitive statements on this question.

When something of this nature is done, it can be meaningful to talk about a Safe Withdrawal Rate (at the 95% level of safety) that can increase as well as decrease as the number of years vary.

Is it possible to say "by moving your safety percentage up by x amount, you move your SWR down by y amount?" I understand that the relationship between the two numbers is not the same in all circumstances, that the extent to which a change in one causes a change in the other varies from circumstance to circumstance. But is it true that this relationship can be determined in a given set of circumstances? My understanding has been that it is possible to do this, but it would help me a bit if I could obtain verification on this point.

If you know that the SWR at 90 percent safety is x, it is fair to say that the SWR at 80 percent safety is something higher? Is it generally possible to calculate how much higher? Do the data and statistical techniques available permit this? I think that we could provide clues on some of the issues noted above if we could do analyses of the extent to which changes in the confidence percentage cause changes in the SWR.

We are entering an era when we will Specify and Design retirement portfolio strategies.

Extremely encouraging words. I have in the past employed a slightly different terminology in making reference to the concept you refer to above. I call it "getting to the good stuff."

JWR1945
***** Legend
Posts:1697
Joined:Tue Nov 26, 2002 3:59 am
Location:Crestview, Florida
hocus
This is why I have come to believe that the 4 percent number may be a rough average of the SWRs for the various years. I do not say that this is so for sure. It's just a theory. And I do not say even in the theory that the 4 percent number is the exact average of the various SWRs. My sense is that it might be a rough average.

For the sake of clarity, you will have to abandon the choice of the word average in this context. That word has a precise, mathematical definition. It is different from what intend to say. I might have selected the word estimate or I might have referred to it as a result produced by a model or calculation.

What you have described is the degree of certainty or the likelihood that the number 4% is a good number. From raddr's sensitivity studies, there is a good chance that a better calculation would have produced a smaller number. That is, if the estimate is 4% only because the actual historical sequence was a lucky sequence, then an ideal calculation would have produced a smaller number.

When you have used the word average, you may have been referring to raddr's results or something like it or the combination of just about everything that you have ever seen...an all-of-the-above type of answer. In that sense, the word average would have been used in accordance with its mathematical meaning. If that is the case, it is necessary to refer to such details to avoid confusion.

Unfortunately, even if we were to calculate an estimate based on raddr's methodology, we would have to concern ourselves as to whether his assumptions were right. It is possible that there is an undiscovered reason that causes history to exclude everything except lucky sequences.

I think that you can see that there is no end to this. This can lead to a hopeless diversion. It is much, much better to have your thoughts...which you have written explicitly...brought out into the open and placed along side of your basic discussion. You have some good points. I don't like seeing them lost because of semantics.

Have fun.

John R.

JWR1945
***** Legend
Posts:1697
Joined:Tue Nov 26, 2002 3:59 am
Location:Crestview, Florida
hocus
If you know that the SWR at 90 percent safety is x, it is fair to say that the SWR at 80 percent safety is something higher?

Yes, it is always true. Yes, we can try to estimate the probability distribution and get better answers. But, no, that is not what I was getting at.

Suppose that you have 10 data points. You can make estimates from them. You can estimate the 20% and 30% probability levels. For the 20% level, you would use the larger of the two smallest data points. For the 30% (three data points), you would use the largest of the three smallest points.

Now suppose that you add another 10 data points and that nothing has changed in a statistical sense. It is possible that all 10 of those points could be the higher than the old 30% level. You would have twenty data points but only two out of twenty would now correspond to the old 20% level and only three out of twenty would now correspond to the old 30% level. Your new calculations would replace the old estimate of 20% based on ten data points with 10% based on twenty data points. Your would replace your old estimate of 30% based on ten data points with 15% based on twenty data points.

It could have been different. All of the new data points might have been lower than the original 20% level. Then we would have 12 data points out of 20 at, or below, the old 20% level. That would push everything up to 60%. That would be quite a change.

Because you are always talking about a limited number of years and you are always talking about a very small number of data points below a particular threshold, the threshold number can jump around quite a bit as you add data points.

That is what I had in mind. If we are talking about 2 or 3 years out of 130 years, my percentages can jump around quite a bit. If the two events (portfolio failures) occur early on, I might start referring to 2 out of 50 years, which is quite a bit different from 2 out of 130 years. But if the events (portfolio failures) did not occur until the 1960s, I would have been saying zero failures until then.

In terms of yearly safe withdrawal rate results, the fact that failures occur in bunches means that this kind of pronounced jumping around of the statistics is very likely.

Have fun.

John R.

JWR1945
***** Legend
Posts:1697
Joined:Tue Nov 26, 2002 3:59 am
Location:Crestview, Florida
hocus quoting me:

If we were to substitute the term, a Year's Safe Withdrawal Rate, for hocus's use of the words Safe Withdrawal Rate, we would have a whole lot better idea of what he has been saying all along.

I'm an optimist, JWR1945. But you are really an optimist! I think optimism is good, so I offer no objection to what you say here. I do feel a need to note, however, that there are alternate explanations of the "goings on" that we have witnessed over the past 14 months.

Maybe peteyperson was right after all.

I will back down just a bit. The term should be sufficient for those wishing to communicate and gain a mutual understanding.

It can never be sufficient if a person is just engaged in a game of gotcha.

Have fun.

John R.

JWR1945
***** Legend
Posts:1697
Joined:Tue Nov 26, 2002 3:59 am
Location:Crestview, Florida
hocus wrote this post:

When you have used the word average, you may have been referring to... raddr's results or something like it or the combination of just about everything that you have ever seen...an all-of-the-above type of answer. In that sense, the word average would have been used in accordance with its mathematical meaning.

The theory came to me as a result of everything I have seen, including raddr's work.

If that is the case, it is necessary to refer to such details to avoid confusion.

I aim to state things as clearly as possible. I think that the problem is that others are often jumping to unwarranted conclusions.

I did not say that I thought that the people doing the research were deliberately trying to determine the average of all the SWRs. When I put forward the first post on this question, BenSolar came back at me asking if that is what I was saying and I responded with something to the effect of "I certainly would not want to leave anyone with that impression."

The communication problem stems from the fact that many are still making use of the conventional methodology as if it were a reasonable means of determining the SWR. I am saying "the methodology is invlaid" and they are saying "it's not, it says that the SWR is 4 percent." And I am saying "how do you know that 4 percent is the right number?" And they are coming back and saying "that's what the studies say it is!"

I understand perfectly well that that is what the studies say it is. I do not focus on that too much because I consider the methodology used in the studies to be invalid. The argument that "you must accept the number provided in the studies in any argument you make that the methodology is invalid" is a circular one. If I accepted the numbers produced by the methodology, I would accept the methodology. It's the fact that the methodology produces bad numbers that causes me to have such serious doubts about the methodology itself.

The phrase "safe withdrawal rate" means something. Researchers can't just come along and say, "it's too much trouble to calculate what that is, so we are going to calculate something altogether different and call it that." If they want to calculate something different, they should call it by some other name. If they want to purport to calculate the SWR, they should set up a methodology that permits them to calculate the SWR with at least a reasonable degree of accuracy.

Unfortunately, even if we were to calculate an estimate based on raddr's methodology, we would have to concern ourselves as to whether his assumptions were right. It is possible that there is an undiscovered reason that causes history to exclude everything except lucky sequences.

I understand. This is where the point that you have made, that researchers should have flexibility to do things in a multitude of ways, comes into play. I am impressed by raddr's research, but it would not necessarily be unreasonable for a researcher to have a different assessment of it and to elect not to factor it in to his own work. I am OK with that. There are certain grey areas where you have to give researchers leeway to do things one way or another. So long as they explain what they have done in the texts of their studies, that is not a problem.

My core point is that the issue of valuation is not one of these "grey area" type issues. Valuation affects returns as a matter of "mathematical certainty." It is a critical factor in a SWR calculation, as critical as volatility. To ignore the effect of changes in valuation altogether is as bad as to ignore volatility altogether. A methodology that ignores factors known to be of critical importance is invalid, in my view.

I think that you can see that there is no end to this. This can lead to a hopeless diversion.

I don't recommend that we spend much effort on this question of whether 4 is the average SWR. It's not a terribly important point one way or the other. It's something that I puzzle over from time to time in trying to figure out where this is all headed. That's all. I think it is entirely possible that 4 is not the average SWR.

The other matter I do think is of some importance. I think it would be good at some point to determine whether 4 is the SWR in all pre-bubble high valuation years. I lean strongly toward thinking that it is not, but I am not certain. I think that the case for invalidity stands up regardless of how this question is resolved because it is clear that the conventional methodology produces the wrong number at bubble-level valuations and in pre-bubble low valuation years. But I would like to figure out at some point whether it also produces the wrong number in pre-bubble high valuation years.

I don't say that we should turn our attention to this matter right away. There are lots of angles to pursue profitably. However, at some future date, I would like to make more progress on that question. It's a question that has been put on the table by some of the work we have done, and it would be good to tie it up (to the extent possible) somewhere down the line.

JWR1945
***** Legend
Posts:1697
Joined:Tue Nov 26, 2002 3:59 am
Location:Crestview, Florida
JWR1945 (me):

Maybe peteyperson was right after all.

Mr. Peteyperson was the first to proclaim that I am an optimist.

Have fun.

John R.

hocus
Moderator
Posts:435
Joined:Mon Dec 02, 2002 12:56 am
Thanks for making the fix, JWR1945. I am going to repost in my name the one that you put up for me because I want it to show up under my screen-name if someone does a search for just my posts.

If we were to substitute the term, a Year's Safe Withdrawal Rate, for hocus's use of the words Safe Withdrawal Rate, we would have a whole lot better idea of what he has been saying all along.

I'm an optimist, JWR1945. But you are really an optimist! I think optimism is good, so I offer no objection to what you say here. I do feel a need to note, however, that there are alternate explanations of the "goings on" that we have witnessed over the past 14 months.

My sense is that the problem is the one that Bernstein refers to in Chapter 2. He says that most investors understand on an intellectual level that there is a connection between valuation levels and returns, but that most have difficulty accepting this on an emotional level. I think that apreciating that point is key to understanding SWRs. SWRs are largely about numbers and data, but not entirely so. An important part of the equation is the emotions that the numbers and data generate. The reasons why I have become an "expert" on SWRs is not that I am good with numbers and data; I obviously am not. I believe that I have been able to see things relating to SWRs that most others miss because I have a special facility in understanding human emotions. To understand SWRs fully, you need to possess a keen grasp of the emotions that the various numbers evoke, and my sense is that I am better able to grasp some of that stuff than some others, including some of those with the greatest facility with numbers.

In the early days of this, I thought that the key was proving the case. I thought that, since the SWR was the product of a calculation, the way to make my case was to show that the calculation was wrong. When I found the Bernstein book, which shows conclusively that the calculation done according to the conventional methodology produces wrong results, and that did not convince a good number of people, I saw that I had been wrong to put so much focus on the numbers. The cause of the strong reactions to the subject is not a concern over how the calculations are done. It is a concern over what the product of those calculations mean.

The number is used by people, and people have emotions. People use money to achieve their life goals, and people care about those life goals. This number is telling them that the ways they invested in the 1990s were not the best ways. That's extremely disconcerting. It wasn't a few people advocating the investment strategies that were followed in the 1990s. It was lots and lots of people. You were hearing the same story everywhere you turned. This number is telling us that the advice we heard for all those years was fundamentally flawed. It is not true that stocks are always the best investment class if you are investing for the long term. That's the most important message that this number is telling us. That's in conflict with the message that most money experts were telling us in the 1990s. So accepting that the new number is the right number means accepting a big change in how you go about evaluating investment options, and it is often difficult for human beings to accept dramatic changes in how they think about things that affect their capcity to realize their life goals.

I think that that is the primary explanation for the reaction we have seen to the idea of discussing the realities of SWRs. I think that people sense the dramatic change in investing styles inplied in acceptance of the new numbers, and they are extremely uncomfortable with the implications. I think that will change with time, but only with time. I don't think this is something that you can force on people. I think we have an obligation to present it to people, and then to allow them to make the decision as to how much of the realities they can let in at any given moment. You can't force people to accept in one moment dramatic changes in a way of thinking that they have developed over the course of a good number of years, in my view.

All that said, I certainly would be glad to see your "A Year's SWR" idea bring a few people around. It is certainly true that a core compliant that I have with the conventional methodology is that it does not tell you the SWR that applies for any particular year. That's a big, big problem with the conventional methodology, in my view. You need to know the number that applies in a particular year to make effective use of the tool. And the conventional methodology does not tell you what it is.

If one assumes that there is no relationship whatsoever between stock prices and Safe Withdrawal Rates, that single reported number would be sufficient

That's right. It is an implicit premise of the conventional methodology that there is no relationship. Researchers who employ the conventional methodology are putting their names behind this premise. They are endorsing it by using this methodology to calculate SWRs. Buying into a false premise can often get you into trouble, and that is what is happening to the researchers who use this methodology. They are buying into a false idea without giving much (if any) thought to it, and that false idea is ruining all their efforts that follow. You can try and try and try to get all the calculations right, but if you started from a false premise, you end up generating bad numbers as your final product.

In a very real sense, the traditional Safe Withdrawal Rates based on historical sequences are not valid for estimating any individual Year's Safe Withdrawal Rate.

Thanks for saying that. That's exactly what I am saying.

The only really important difference between your position and mine that I can make out is that I think that what you say above invalidates the methodology. The purpose of SWR analysis is to calculate a number for a particular year, the year the retirement commences. If the conventional methodology is not capable of generating the correct number for that year, it is invalid. That's my take.

You have got to know the number that applies for your particular retirement, or most of the tool's value is lost to you. The conventional methodology was not designed in such a way as to reveal this number. So it doesn't do the job it purports to do. It doesn't tell you the SWR that applies for you. It tells you something else, but it does not tell you the SWR that applies to someone who retires at a particular point in time, as all retirees do.

In a very real sense, traditional Safe Withdrawal Rate studies have not incorporated valuations.

Again, thank you for saying that. I think that's right, but in the interests of being a little more precise, I am going to add a wrinkle.

The studies do incorporate valuations in a way. They don't so it in a way that allows you to know the SWR for the year you retire. The conventional studies are invald for purposes of calculating the SWR. But there is valuation stuff mixed in the data somewhere. There were various valuation levels experienced during the historical period and those changes are reflected somehow in the data. Isn't that so?

My sense is that the conventional methodology does indeed incorporate valuations, just not in a way that allows for calculation of the SWR. The conventional methodology mixes together valuation inputs for all the various time periods. There's data in there relating to low valuation years and medium valuation years and high valuation years, and it all gets aggregated into one big bowl of data soup. I don't think that is the proper way to determine the SWR for any particular year (which is the purpose of the exercise). The impact of the data from low valuation years is cancelling out the impact of the data from high valuation years. The number being produced is a number that I believe is probably close to an accurate number for the medium valuation years, but not for the high valuation years or the low valuation years.

You seem to be saying that the conventional methodology gets the number wrong in the bubble period and in the pre-bubble low valuation years, but that it gets it right in the pre-bubble high valuation years. Perhaps. It's clear that it gets it wrong in the two scenarios in which you say it gets it wrong. I am not personally convinced that it gets it right in the pre-bubble high valuatiion years. I am open to being convinced on this point. But I would like to see data supporting the idea before buying into it.

I think that you may be confusing the concept of a safe withdrawal rate and the concept of a lucky withdrawal rate. We know that the 4 percent number survived for a retirement that began in the year 1929. That does not prove that this was a safe number for a retirement that began in 1929. There are two ways that a retirement based on a particular number can survive. It may be that the number is a safe number, and it may be that the number is a lucky number. A safe number will almost always work. A lucky number may work only once or twice and fail in all other circumstances.

How many data points do we have indicating that the 4 percent number is a safe number for the 30-year period beginning in 1929 and extending to 1959? We have one. There were no other occasions prior to the bubble period at which we were at that valuation level. So we don't really have much reason to say that the 4 percent number is safe at that valuation level. It's a question we need to explore, in my view, not one that we already know the answer to.

There is a lot of evidence on the table pointing to a conclusion that the 4 percent number is not safe for a retirement beginning at the valuation level in effect in 1929. The SWR is a calulation that is always done at the beginning of a retirement, never at the end. It is not reasonable to say that, because a number worked one time, it is therefore to be deemed "safe." We need to check this out, in my view.

I lean toward saying that the conventional methodology is invalid even if it turns out that the 4 percent number really is safe for valuation levels equal to those in effect in 1929. Even if that is so, it is still true that the conventional methodology produces bad numbers in bubble periods and in low-valuation pre-bubble valuation levels. That's enough for me to say that analysts should not use this approach. But one of the questions that I would like to tie down at some future date is this question of whether the 4 percent number is truly safe at 1929 valuation levels. My sense is that it is not. A lot of people (including you) seem to be coming down on the other side on this question, but each time I find myself starting to waver on this point, I pull out some of the data we have looked at and again wonder if that 4 percent number is truly safe at all pre-bubble valuation levels. This is the outstanding question that I would most like to get a better fix on at this point.

Say that it turns out that the 4 percent number is not truly safe at 1929 valuation levels. What would that tell us about what that 4 percent nunmber signifies? It could be that it is a meaningless number, it's just a number that pops out when you employ an invalid methodology. Part of me believes that that is the case. But part of me tells me that perhaps there is some significance to the number. The conventional methodology does incorporate valuation, it just does so improperly. So I wonder if the valuation effect is somehow being reflected in the results of the conventional studues, even if in an improper way.

This is why I have come to believe that the 4 percent number may be a rough average of the SWRs for the various years. I do not say that this is so for sure. It's just a theory. And I do not say even in the theory that the 4 percent number is the exact average of the various SWRs. My sense is that it might be a rough average. The number 4 does not appear to me to be a randomly generated number. The data we have seen suggests to me that it is not too far from what the true average SWR is. Perhaps the average SWR is 3 percent or perhaps it is 5 percent. I am not saying that it is exactly 4. I am saying that the conventional methodology may be telling us that 4 is a number in the neighborhood of the average SWR. That's not an entirely insiginficant information bit.

The "average" question is not a big deal. Whether 4 turns out to be the average SWR or not will not have a big effect on the question of whether the convetnional methodology is valid or not, and that is the most important question before us. If I am right about this, it makes my case stronger. If I am not right, it does not undermine my case too much. It would still be true that we have shown that the conventional methodology turns out wrong numbers in bubble years and in pre-bubble low-valuation years. So it is a mistake to place too much focus on the question.

The reason that I brought it up the other day is that I find it helpful at times to puzzle over the question as to what that number 4 really signifies. Perhaps it signifies nothing. Perhaps it is the correct SWR for pre-bubble high valuation years. But perhaps neither of those two things is so. I think that it is too early in the game to be making definitive statements on this question.

When something of this nature is done, it can be meaningful to talk about a Safe Withdrawal Rate (at the 95% level of safety) that can increase as well as decrease as the number of years vary.

Is it possible to say "by moving your safety percentage up by x amount, you move your SWR down by y amount?" I understand that the relationship between the two numbers is not the same in all circumstances, that the extent to which a change in one causes a change in the other varies from circumstance to circumstance. But is it true that this relationship can be determined in a given set of circumstances? My understanding has been that it is possible to do this, but it would help me a bit if I could obtain verification on this point.

If you know that the SWR at 90 percent safety is x, it is fair to say that the SWR at 80 percent safety is something higher? Is it generally possible to calculate how much higher? Do the data and statistical techniques available permit this? I think that we could provide clues on some of the issues noted above if we could do analyses of the extent to which changes in the confidence percentage cause changes in the SWR.

We are entering an era when we will Specify and Design retirement portfolio strategies.

Extremely encouraging words. I have in the past employed a slightly different terminology in making reference to the concept you refer to above. I call it "getting to the good stuff."

hocus
Moderator
Posts:435
Joined:Mon Dec 02, 2002 12:56 am
Here is a repost of the other one that you put up for me, JWR1945. I'm leaving up the ones that you put up so that people will know at what point in the conversation my posts showed up. I'm putting the new ones up so that people will find the posts if they search only for hocus posts.

When you have used the word average, you may have been referring to... raddr's results or something like it or the combination of just about everything that you have ever seen...an all-of-the-above type of answer. In that sense, the word average would have been used in accordance with its mathematical meaning.

The theory came to me as a result of everything I have seen, including raddr's work.

If that is the case, it is necessary to refer to such details to avoid confusion.

I aim to state things as clearly as possible. I think that the problem is that others are often jumping to unwarranted conclusions.

I did not say that I thought that the people doing the research were deliberately trying to determine the average of all the SWRs. When I put forward the first post on this question, BenSolar came back at me asking if that is what I was saying and I responded with something to the effect of "I certainly would not want to leave anyone with that impression."

The communication problem stems from the fact that many are still making use of the conventional methodology as if it were a reasonable means of determining the SWR. I am saying "the methodology is invlaid" and they are saying "it's not, it says that the SWR is 4 percent." And I am saying "how do you know that 4 percent is the right number?" And they are coming back and saying "that's what the studies say it is!"

I understand perfectly well that that is what the studies say it is. I do not focus on that too much because I consider the methodology used in the studies to be invalid. The argument that "you must accept the number provided in the studies in any argument you make that the methodology is invalid" is a circular one. If I accepted the numbers produced by the methodology, I would accept the methodology. It's the fact that the methodology produces bad numbers that causes me to have such serious doubts about the methodology itself.

The phrase "safe withdrawal rate" means something. Researchers can't just come along and say, "it's too much trouble to calculate what that is, so we are going to calculate something altogether different and call it that." If they want to calculate something different, they should call it by some other name. If they want to purport to calculate the SWR, they should set up a methodology that permits them to calculate the SWR with at least a reasonable degree of accuracy.

Unfortunately, even if we were to calculate an estimate based on raddr's methodology, we would have to concern ourselves as to whether his assumptions were right. It is possible that there is an undiscovered reason that causes history to exclude everything except lucky sequences.

I understand. This is where the point that you have made, that researchers should have flexibility to do things in a multitude of ways, comes into play. I am impressed by raddr's research, but it would not necessarily be unreasonable for a researcher to have a different assessment of it and to elect not to factor it in to his own work. I am OK with that. There are certain grey areas where you have to give researchers leeway to do things one way or another. So long as they explain what they have done in the texts of their studies, that is not a problem.

My core point is that the issue of valuation is not one of these "grey area" type issues. Valuation affects returns as a matter of "mathematical certainty." It is a critical factor in a SWR calculation, as critical as volatility. To ignore the effect of changes in valuation altogether is as bad as to ignore volatility altogether. A methodology that ignores factors known to be of critical importance is invalid, in my view.

I think that you can see that there is no end to this. This can lead to a hopeless diversion.

I don't recommend that we spend much effort on this question of whether 4 is the average SWR. It's not a terribly important point one way or the other. It's something that I puzzle over from time to time in trying to figure out where this is all headed. That's all. I think it is entirely possible that 4 is not the average SWR.

The other matter I do think is of some importance. I think it would be good at some point to determine whether 4 is the SWR in all pre-bubble high valuation years. I lean strongly toward thinking that it is not, but I am not certain. I think that the case for invalidity stands up regardless of how this question is resolved because it is clear that the conventional methodology produces the wrong number at bubble-level valuations and in pre-bubble low valuation years. But I would like to figure out at some point whether it also produces the wrong number in pre-bubble high valuation years.

I don't say that we should turn our attention to this matter right away. There are lots of angles to pursue profitably. However, at some future date, I would like to make more progress on that question. It's a question that has been put on the table by some of the work we have done, and it would be good to tie it up (to the extent possible) somewhere down the line.

hocus
Moderator
Posts:435
Joined:Mon Dec 02, 2002 12:56 am
Unfortunately, I left out the italics in the post that appears two above this one. I will let it be since you posted a version with the italics up higher in the thread.

All of the posts from the original thread now appear here as they were originally posted and in the order in which they were originally posted. They also will all appear if someone does a search by author to retrieve those posts put up by a particular community member.

On top of all that, that pesky post count now comes closer to saying what it should say.

Another day of productive work at the SWR Research Group! I suggest we reward ourselves for our hard work by getting an early start on the weekend!

hocus
Moderator
Posts:435
Joined:Mon Dec 02, 2002 12:56 am
Oh no!

I now realized that, by deleting the original thread, I have messed up the "Views" count. Here is what you have to do if you want to possess an accurate understanding of the historical record. The original thread had 101 views at the time it was deleted, and this new one now has 25. So if you add 76 to whatever number of views is reported at any given time, you will know the actual number.

I think it is more important to have the "Response Post" count right than the "Views" count because there might be some who would not check out the thread at all if they did not know that it had generated a good number of responses. I presume that not too many focus on the "Views" count too much.

Anyway, I better stop pushing buttons before I do some real damage.

JWR1945
***** Legend
Posts:1697
Joined:Tue Nov 26, 2002 3:59 am
Location:Crestview, Florida
hocus
The other matter I do think is of some importance. I think it would be good at some point to determine whether 4 is the SWR in all pre-bubble high valuation years. I lean strongly toward thinking that it is not, but I am not certain. I think that the case for invalidity stands up regardless of how this question is resolved because it is clear that the conventional methodology produces the wrong number at bubble-level valuations and in pre-bubble low valuation years. But I would like to figure out at some point whether it also produces the wrong number in pre-bubble high valuation years.

No, it is not. I have attached a table that shows each year's safe withdrawal rate for the twenty years with the highest valuations. The table identifies the year, the P/E10 value in January of that year and the Safe Withdrawal Rate for that year. (The portfolio was 80% stocks and 20% commercial paper. The method was historical sequences with a lifespan of 30 years. I used FIRECalc.)

The Safe Withdrawal Rate ranged from 4.0% to 5.8% in these high valuation non-bubble years. If we limit ourselves to those years with P/E10 levels of 20 and above, the range is 4.0% to 4.8%. Those are the ten years with the highest valuations. Notice that 1930 had a 4.8% safe withdrawal rate. It is the fourth from the top out of sixty start years (1921-1980).

Year PE10 SWR (80%)
1936 17.0 5.4%
1970 17.0 4.8%
1972 17.2 4.8%
1959 17.9 5.4%
1956 18.2 5.8%
1960 18.3 5.2%
1961 18.4 5.2%
1973 18.7 4.6%
1928 18.8 5.8%
1963 19.2 5.0%
1967 20.4 4.4%
1962 21.1 4.8%
1969 21.1 4.2%
1968 21.6 4.2%
1937 21.6 4.6%
1964 21.6 4.6%
1930 22.3 4.8%
1965 23.2 4.2%
1966 24.0 4.0%
1929 27.0 4.2%

Here might be another way of describing these numbers. The lowest safe withdrawal rate among the other 40 years (which I have calculated but not shown) is 5.6%. Among the top third (the 20 years out of 60) with the highest valuations, the safe withdrawal rates ranged from 4.0% to 5.8%. Among the top sixth (the 10 years out of 60) with the highest valuations, the safe withdrawal rates ranged from 4.0% to 4.8%.

It does not require extreme precision to separate the safe withdrawal rates of most of the high valuation years from the single value of 4.0%.

Claims that a single number is appropriate are intimately tied to the idea that it is impossible to relate any year's safe withdrawal rate with any measure of valuation. We have proved by example (actually, by counter-example) that such an assumption is false.

Is that what you are asking? Valuations always matter. Assertions that they do not are false. We have produced prima facie evidence. The technical argument is clear-cut. That argument is over...except emotionally.

Have fun.

John R.

JWR1945
***** Legend
Posts:1697
Joined:Tue Nov 26, 2002 3:59 am
Location:Crestview, Florida
From hocus:
The other matter I do think is of some importance. <b>I think it would be good at some point to determine whether 4 is the SWR in all pre-bubble high valuation years.</b> I lean strongly toward thinking that it is not, but I am not certain. I think that the case for invalidity stands up regardless of how this question is resolved because it is clear that the conventional methodology produces the wrong number at bubble-level valuations and in pre-bubble low valuation years. But I would like to figure out at some point whether it also produces the wrong number in pre-bubble high valuation years.

My answer is emphatic: No, it is not.

I wish to drive home this point. Not only can we discern that different safe withdrawal rates depending upon valuations, we can also discern the different safe withdrawal rates from among years with similar (high levels of) valuations. We can point to reliable cause and effect relationships.

Here is a table with the 20 years in 1921-1980 with the highest valuations. It lists the year, the valuation (P/E10), the dividend yield in the year indicated and the safe withdrawal rate for that year.

Year PE10 Yield SWR (80%)
1936 17.0 3.50% 5.4%
1970 17.0 3.50% 4.8%
1972 17.2 2.98% 4.8%
1959 17.9 3.15% 5.4%
1956 18.2 3.78% 5.8%
1960 18.3 3.21% 5.2%
1961 18.4 3.26% 5.2%
1973 18.7 2.67% 4.6%
1928 18.8 4.43% 5.8%
1963 19.2 3.28% 5.0%
1967 20.4 3.41% 4.4%
1962 21.1 2.93% 4.8%
1969 21.1 3.01% 4.2%
1968 21.6 3.08% 4.2%
1937 21.6 4.17% 4.6%
1964 21.6 3.00% 4.6%
1930 22.3 4.47% 4.8%
1965 23.2 2.92% 4.2%
1966 24.0 2.93% 4.0%
1929 27.0 3.46% 4.2%

Note: there are special circumstances related to the 1937 dividends.

Not only does a year's safe withdrawal rate depend upon valuations, it depends upon the dividend yield. Thus, we can discern between 1930 with its very high valuation (P/E10 = 22.3) with its 4.8% safe withdrawal rate and 1966, which also has a very high valuation (P/E10 = 24.0), with its 4.0%. The steady income from 1930's higher dividend yield (4.47% versus 2.93%) sustained a higher safe withdrawal rate than for 1966.

Notice that 1962 and 1966 had identical dividend yields (2.93%). Yet, 1962 has a higher safe withdrawal rate (4.8%) than 1966 (4.0%). We can recognize the reason by comparing valuations. Stocks in 1966 had higher valuations than in 1962 (P/E10s of 24.0 versus 21.1).

We are not restricted to using any single answer for the safe withdrawal rates in the pre-bubble era, even at high valuations. In that sense the conventional methodology fails.

The conventional methodology allows us to calculate different safe withdrawal rates for different years. That part of the conventional methodology remains valid. The conventional methodology failed in the sense that it gave up on finding out how to order the safe withdrawal rates of different years. In that sense, the conventional methodology is invalid.

We do not wish to throw out everything in the studies because they still have some valid sections: that is, they contain the correct (historical) safe withdrawal rate results for different years. (I am using the word study in its broadest sense. I am including SWR calculators that produce the results for each year as part of such studies. Specifically, I include dory36's (Captain Bill's) FIRECalc calculator under this general heading.)

The technical case supporting your previously made assertions is rock solid. It is no longer arguable. (At least, in terms of this specific group of assertions.)

Have fun.

John R.

hocus
Moderator
Posts:435
Joined:Mon Dec 02, 2002 12:56 am
We do not wish to throw out everything in the studies because they still have some valid sections:

Please don't get the idea that I want to "throw out" anything from the existing studies. I love the existing studies. I agree with Bernstein that the Trinity study was a "breakthrough" piece of work. I have been recommending the REHP study for a long time, and have made profitable use of it myself. I would like to take this idea that I want to "throw out" the existing studies off the table once and for all.

What I want to do is not to throw out anything that is in the existing studies, but add to what is there. The existing studies do a great job of taking into consideration three critical factors--nominal return, inflation, and volatility. There is a fourth factor that they fail to take into consideration--changes in valuation levels. I am arguing for the inclusion of something that is not in the existing studies.

that is, they contain the correct (historical) safe withdrawal rate results for different years.

This is where you and me part company. You are sneaking in by way of a parenthetical the old BenSolar idea of an hSWR. I reject that concept. I don't mean to sound harsh, but I view it as a nonsense concept. I don't see how there could ever even be such a thing as a hSWR.

The phrase "Safe Withdrawal Rate" means something. It is a number that provides an assessment of the probabilities of various future possibilities, based on what has happened in the past. That's my working definition. Those are my words, but I believe that the definitions used in the studies are in accord with those words.

The phrase "Safe Withdrawal Rate" always makes reference to things in the past and to things in the future. To know what is safe, you must look to the past --that's where all the data is. It's possible that one thing you will be considering is a valuation change that took place just yesterday, but yesterday is the past. If it weren't the past at the time you calculated the number, you wouldn't have data for that factor. SWR analysis always requires taking into consideration things that happened in the past.

It also always requires making reference to the future. The phrase is not "Safe Withdrew Rate," it is "Safe Withdrawal Rate." The purpose of the exercise is to assess future possibilities. Looking blindly at things that happened in the past without making any effort to use those things in ways that help you assess future possibilities is not SWR analysis. It is something else. SWR analysis always requires looking not only backwards, but forwards.

You are buying a bike for your child and you want to be sure that it is safe. You find a study that a bike manufacturer performed published on the internet. You read it to determine whether you can feel comfortable buying your child the bike or not.

The study shows what happened to 100 bike riders, 33 who drove blue bikes, 34 who drove green bikes (I used 34 here to make the total come to 100) and 33 who drove red bikes. The methodology was to see how many got in accidents in the course of a year's time.

The results show zero accidents for blue bikes, zero accidents for green bikes, and 10 accidents for red bikes. The study concludes that there is a 10 percent chance of getting in an accident on any color of bike that you decide to buy.

This is nonsense. The data shows clearly that red bikes are more dangerous than green bikes or blue bikes. The methodology is mixing up all the data into one big bowl of data soup and reporting results that mask the true findings. That's how the conventional SWR methodology hanfles valuation, it mixes everything together rather than showing you the effect of changes in valuation.

Most do not agree with me on this point, but you do. You are saying that it has been proven that valuation has an effect on safety. You are saying that blue bikes are clearly safer than red bikes. So we are OK on all that, and that is the most important part of this debate. So we are in agreement on the most important questions.

I still am more than a little uneasy with what you are saying on a different question, however. You are saying that because a 4 percent withdrawal worked from the time 1929 to 1959 that it is therefore reasonable to declare it safe at the valuation level that applied in 1929. I have serious doubts about that.

You look deeper into the data for the bike study and you find that the researchers identified different levels of accidents. Level One accidents had no damage to the bike or the person, Level 4 accidents required hospitalization, and the other two levels were in the middle.

The researcher also identified different shades of color for the bikes studies. The 33 red bikes were not all the same shade of red. Some were bright red. Some were medium red. Some were light red.

Looking at the data, you see that there is a correlation between the level of severity of the accidcent experiened and the brightness of the red coloring on the bike. The worst accidents of all were those that happened with the reddest bikes.

Still, the worst accident of the 10 examined in the study was only a Level 2 accident. There was some damage to the bike and a few minor bruises, but no hospitalization. Is it fair to say from looking at this data that "You can be 100 percent certain that nothing worse will happen to you than perhaps getting a few bruises if you ride the reddest of all bikes."

I say no. You only looked at 10 accidents, and, of the 33 red bikes you looked at, only one was of the brighest shade of red, and that one got in the worst accident of those examined. It seems to me that what the data is telling you is that red bikes are dangerous and the bright shade of red bikes are the most dangerous of all.

It does not seem right to me to conclude from all this that bright red bikes are perfectly safe, that the odds are that you won't get in any accident and that, if you do, it is sure to be a minor one. My sense from considering this data is that there is a bad accident just waiting to happen for someone riding a bright red bike.

Have we seen one in the data? No. But we only looked at 10 accidents, and only one for a bike that is a bright shade of red. What does one data point tell you?

It does tell you something, It shows a correlation between redness and accidents, and that's an important thing to know. So this study is of great value in assessing bike safety. The data here is trying to tell you a very important message.

The conclusion that the researchers came to is a dangerously irresponsible one, however. They are saying that there is no risk of a worse accident just because in that one case the accident that turned up was not so bad. It is a conclusion directly at odds with the message that the data itself is trying to convey, in my view.

I believe that the historcal data is trying to tell us that a 4 percent withdrawal at the valuation level in effect in 1929 was a withdrawal with a good bit of risk attached to it. Some hypothetical investor who tried it got away with it. But getting away with something does not mean that the something you got away with was a safe practice. It means that it was either safe or lucky. In this case, given what the data says, it appears to me that the 4 percent withdrawal at 1929 valuation levels was lucky, not safe.

My sense is that the data is screaming out to us that the 4 percent number is not safe at 1929 valuation levels. You had a post one time that said "It's Safe Now" because we had dropped to pre-bubble valuation years. I don't buy it. I don't think that the data is saying that 4 percent is a safe withdrawal at high-valuation pre-bubble years.

I am not as locked in on this view as I am on the more important question that changes in valuation levels should be taken into account in SWR analysis. But I think that showing statistically that a 4 percent withdrawal is not safe at 1929 valuation levels would strengthen my case a bit. So, if it is possible to show that with a statistical analysis, I would like to do so.

We have proven two very important things, that the conventional methodology gets the number wrong on the high side in bubble years and that it gets the number wrong on the low side in pre-bubble years. That's enough for me to declare the methodology invalid. But it would be one more nail in the coffin to show that it also gets the number wrong on the high side in pre-bubble years.

When I use the phrase "one more nail in the coffin," I am not trying to suggest "throwing away" the old studies. I am arguing that the old studies need always to include in addition to what they include now some adjustment for the effects of changes in valuation. I am saying that it is a critical factor that is ignored in the conventional methodology, and that valid methodologies must consider all factors with a critical bearing on the question being examined.

JWR1945
***** Legend
Posts:1697
Joined:Tue Nov 26, 2002 3:59 am
Location:Crestview, Florida
hocus quoting me, then responding:
that is, they contain the correct (historical) safe withdrawal rate results for different years.

This is where you and me part company. You are sneaking in by way of a parenthetical the old BenSolar idea of an hSWR. I reject that concept. I don't mean to sound harsh, but I view it as a nonsense concept. I don't see how there could ever even be such a thing as a hSWR.

The phrase "Safe Withdrawal Rate" means something. It is a number that provides an assessment of the probabilities of various future possibilities, based on what has happened in the past. That's my working definition. Those are my words, but I believe that the definitions used in the studies are in accord with those words.

The phrase "Safe Withdrawal Rate" always makes reference to things in the past and to things in the future. To know what is safe, you must look to the past --that's where all the data is. It's possible that one thing you will be considering is a valuation change that took place just yesterday, but yesterday is the past. If it weren't the past at the time you calculated the number, you wouldn't have data for that factor. SWR analysis always requires taking into consideration things that happened in the past.

It also always requires making reference to the future. The phrase is not "Safe Withdrew Rate," it is "Safe Withdrawal Rate." The purpose of the exercise is to assess future possibilities. Looking blindly at things that happened in the past without making any effort to use those things in ways that help you assess future possibilities is not SWR analysis. It is something else. SWR analysis always requires looking not only backwards, but forwards.

and later

I still am more than a little uneasy with what you are saying on a different question, however. You are saying that because a 4 percent withdrawal worked from the time 1929 to 1959 that it is therefore reasonable to declare it safe at the valuation level that applied in 1929. I have serious doubts about that.

I agree and you have given us an excellent example. I did not intend to bring back the old idea of a Historical Safe Withdrawal Rate (hSWR). In fact, that is why I put the word historical in parenthesis. My intention was to state that those results from past years are still valuable for helping us to predict what will be safe in future years.

This is all in the context of this thread. We always want to make predictions about the future. The previous studies, at least in the sense of the calculators that they have produced, allow us to look at individual years in the past so as to assist us in sorting out possible cause and effect relationships. They are helpful in the sense that having a database is helpful. The overall objective, of course, is always directed toward making predictions.

This is quite subtle, but I do not think that what happened in 1929-1959 is appropriate for valuations and the other conditions that happened in 1929. What happened in 1929-1959 provides a single data point of what can happen under such circumstances. What applies for identical valuations, etc., for 1929 is best described in terms of an underlying probability distribution (which we do not know in detail). We search to identify and understand which effects...including valuations and many others...are helpful in making predictions.

Have fun.

John R.

hocus
Moderator
Posts:435
Joined:Mon Dec 02, 2002 12:56 am
My intention was to state that those results from past years are still valuable for helping us to predict what will be safe in future years.

I agree that the data used in the conventional methodology studies has value. My argument is that it is not by itself conclusive as to what the SWR is. It is perfectly reasonable to look at the historical sequence data as part of a determination of the SWR, in my view.

Here's what Bernstein says on Page 73: "Unfortunately, although the Discounted Dividend Model informs us well about expected returns, it tells us nothing about future risk. We are dependent on the pattern of past returns to inform us of the potential risks of an asset. And, in this regard, I believe that the historical data serve us well."

I like that way of putting it. The historical sequence data tells one side of the story. And the Gordon Equation or some other means of incorporation of an assessment of the effect of changes in valuation levels tells the other side. You put the two together, and you have an incredibly powerful tool. You exclude either side of the equation, and you have a tool that misleads. Reporting well on part of the story that one type of data is telling is not good enough. You need to report the whole story, and changes in valuation levels are an important part of the story.

We always want to make predictions about the future.

I don't like the word "predictions." I don't view the SWR tool as a tool for making predictions.

Say that a valid SWR analysis tells us that the SWR for a given year is 3 percent. Say that retirees taking a 3 percent withdrawal for retirements beginning in that year go bust. Does that mean that there was necessarily something wrong in the way that the analysis was performed?

I say no. The SWR tool does not aim to make predictions as to what will work or not work. It makes assessments of the probabilities of various future possibilities.

This is the point that I was trying to get across in my weather-man example from a while back. If the weather man says that it is going to be sunny tomorrrow and then it rains, was the methodology he used invalid? I don't think that it necessarily was invalid. He may have used a perfectly valid methodology and still came out with an assessment that didn't work out.

If you are getting your weather reports from an astrologer, it is fair to call them "predictions." If those reports end up not working out, it is fair to say that the methodology (looking to the positions of the stars) was proven invalid, at least in this one case. The weather-man is not even aiming to make a "prediction," in my view, so I don't think it is fair to say that a "prediction" failed just because some event other than the expected one turned up in reality.

When you are making an assessment of probabilities, it should always be clear to the person using the assessment that things may not work out as expected. That's why an SWR analysis should always include a caveat that the assessment will work only "if the future is not worse than the past." A SWR analyst is not pretending to know anything about the future. He is telling you what will happen in the future in the event that the future is like the past.

I don't believe that Bernstein is saying that a 2 percent withdrawal is guaranteed to work in the future. He is saying that a 2 percent withdrawal is guaranteed to work in the future, assuming that the future is no worse than the past. The two concepts are not the same.

SWR analysis aims to identify the highest withdrsawal rate that will work with certainty if the future is not worse than the past. He is saying that that number is 2, not 4. That's an important thing to know. There are all sorts of implications that follow from discovery of that reality. I think we need to focus on the significance of our discovery that that number is sometimes 2, sometimes 4, and sometimes 6.

It confuses things to suggest that Bernstein is making a prediction of some sort. He is not using tools that permit him to make predictions. I don't think that he believes that such tools even exist. So it subjects his analysis to an unfair standard to suggest that he is making a prediction, in my view.

I do not think that what happened in 1929-1959 is appropriate for valuations and the other conditions that happened in 1929. What happened in 1929-1959 provides a single data point of what can happen under such circumstances.

I think that we are in agreement here. I am not absolutely sure, but I do not have an objection to the way you state things with these particular words.

What applies for identical valuations, etc., for 1929 is best described in terms of an underlying probability distribution (which we do not know in detail)

I agree. Knowing the SWR "in detail" appears to be a complicated task. It appears to me that there are several different ways of going about doing what needs to be done, and that none is 100 percent satisfactory. So I don't think that we are going to be able to say at the end of this "this is the one right way to do it." Perhaps people will reach that point at some date in the distant future. But I think that developing an understanding of SWRs at that level of precision is a long ways off.

One of the benefits I see in having the conventional methodology declared invalid is that I think it would prompt more researchers to examine the questions that most need to be examined at this point in development of the concept. Once you get the idea that it can be done without consideration of changes in valuation off the table, the obvious next question is, "What is the answer when you <i>do</i> include valuation in the analysis?"

That's the question at which I think analysts should be directing most of their energies. We have done important work, but there is a whole lot more than needs to be done. The more analysts who are involved in this sort of work, the sooner good insights will be developed.

My view is that it is a waste of time at this point for any researcher to be developing a study that fails to consider valuation altogether. That methodology has been discredited, so researchers' energies would be better directed elsewhere.

It is good to leave open lots of possibilities for researchers to examine because you never know what someone is going to come up with. But once a particular methodology has been shown to <i>always</i> produce the wrong answer, it is time to stop wasting time on it. I don't see how any good could ever come of doing studies that presume that changes in valuation have no effect after it has been conclusively proven that changes in valuation always do have an effect. What would the point be?

I view the discovery of the significance of changes in valuation as roughly equivalent in importance to the discovery of the significance of volatility. There was a time when perfectly smart and reasonable people determined their withdrawal rates just by looking at long-term real returns. Peter Lynch did this not so long ago. But once it was revealed that volatility affects the result, it became silly to continue doing analyses that failed to account for the volatility factor. It's silly to pretend that something we all know is so really is not so.

So it is with changes in valuation, in my view. Changes in valuation levels affect the answer to the question being posed. So all future research should take that factor into account. Future studies should include all the stuff that is in the existing studies, of course. All of that stuff has an effect too. But there is no excuse at this point for putting out a new study that fails to account for the effect of changes in valuation levels.

JWR1945
***** Legend
Posts:1697
Joined:Tue Nov 26, 2002 3:59 am
Location:Crestview, Florida
First, let me confirm that you are right about 1929.

hocus
I am not as locked in on this view as I am on the more important question that changes in valuation levels should be taken into account in SWR analysis. But I think that showing statistically that a 4 percent withdrawal is not safe at 1929 valuation levels would strengthen my case a bit. So, if it is possible to show that with a statistical analysis, I would like to do so.

When I did my latest study, From Intrinsic Value, I included a table and this is the last portion of it:

Year PE10 Yield SWR
1964 21.6 3.00% 4.6%
1930 22.3 4.47% 4.8%
1965 23.2 2.92% 4.2%
1966 24.0 2.93% 4.0%
1929 27.0 3.46% 4.2%

In that analysis I identified the relationship between the dividend yield and (what I have erroneously called) the Safe Withdrawal Rate for 1929. It includes a single factor that should be added to the dividend yield to calculate the Safe Withdrawal Rate. I identified that factor as being between 0.74% and 1.36% and I provided a theoretical justification for this range of numbers.

The easiest thing to do is to match the lowest "SWR" of 4.0 in 1966 with its initial dividend of 2.93%. That would put the adjustment at 1.07% or very close to 1%. If one chooses to do that, the 1929 result would be 4.5% (rounded). However, if one looks at many years, the "correct" adjustment could easily be 0.7% (rounded) so that the 1929 result should have been reported as 4.2% (rounded). We also have an upper bound (much less likely) of 4.8% (by adding 1.36% to the dividend yield of 1929). Thus, the Safe Withdrawal Rate for 1929 is most likely to be in the range of 4.2% to 4.5% with an upper bound of 4.8%. My best estimate is 4.5%. There is reasonable rationale to support estimates from 4.2% to 4.8%.

NOTE: There is an additional adjustment because the P/E10 of 1929 was 27.0, not 24.0. It is adequately covered by the range of the 0.74% to 1.36% factor.

I should have stated that the Safe Withdrawal Rate for 1929 is 4.5%. Not doing so was my error and I thank you for correcting me.

The number that I have listed as the Safe Withdrawal Rate (or SWR) in the table in not a Safe Withdrawal Rate. It is a value from a historical database. The two ideas are quite distinct and I apologize for being so slow on the uptake in recognizing that distinction. Consistent with our standard definition (along with a specialized term, the Year's Safe Withdrawal Rate), no historical result is ever a safe withdrawal rate. There must be a calculation based on history (with limited exceptions to the reference to history, such as a direct application of mathematical theorems). There can never be a change based upon subsequent events.

There can be some specialized usages that would involve the 1929 historical sequence. But they would not be 1929 safe withdrawal rates.

For example, there could be a Safe Withdrawal Rate for the year 1940 for the historical sequence that started in 1929. That is, every year up until 1940 would come from the actual historical sequence that started in 1929. The most obvious and clearly significant factor would be the portfolio's balance in 1940. But none of the years from 1940-1959 would be taken from historical data. Everything would be an estimate or prediction stated in terms of probabilities based upon mathematical computations involving only pre1940 information.

Thanks for sticking with me for so long. You have identified a critical distinction that I have glossed over in the past.

Have fun.

John R.

JWR1945
***** Legend
Posts:1697
Joined:Tue Nov 26, 2002 3:59 am
Location:Crestview, Florida
BTW, the Safe Withdrawal Rate for 1966 is most likely to be in the range of 3.7% to 3.9% with an upper bound of 4.3%. My best estimate is 3.9%. There is reasonable rationale to support estimates from 3.7% to 4.3%.

The value in the historical database is 4.0%

Have fun.

John R.

hocus
Moderator
Posts:435
Joined:Mon Dec 02, 2002 12:56 am
Thanks for sticking with me for so long. You have identified a critical distinction that I have glossed over in the past.

I'm not so sure about the part suggesting that I have been sticking with you. My impression is that it has been the other way around. But no matter. If I understand what you are saying here properly, this looks to be a most exciting development.

It sounds to me as if you have put forward a reasonable data-based approach to SWR calculation that includes a valuation adjustment and that gives you a number of 2.2 percent at the top of the bubble; significantly higher numbers than the conventional methodology in pre-bubble low valuation years; a higher number in the most important pre-bubble high valuation year (1929); and a slightly lower number in one other pre-bubble high valuation year (1966). Is that a fair summary?

If it is, then it sounds to me as if your next step would be to put out your own SWR study. You have developed an approach that at least to my eyes makes sense. The appropach is data-based and it factors in the effect of changes in valuation levels. That's the sort of SWR analysis that I think more people need to be hearing about.

I would like to see you put out a formal study using this approach and see what the reaction is. Then I would like to see others put out formal studies making use of alternate reasonable approaches. The more we get out there, the greater the clash of ideas will be. The greater the clash of ideas, the more we will learn.

I don't yet personally agree that the true SWR in the year 1929 was greater than 4. But it appears to me that you are making a reasonable case. It might be that you will persuade me over time as I give more thought to the matter. Or it might be that someone else will do an alternate reasonable study that will persuade me that some other number is the true number for 1929. Either way, your study will have played a role in bringing me to that conclusion.

Are you considering doing your own study, JWR1945? I think that would be a most encouraging development. I think it would be the best thing to come of all of this thus far.

JWR1945
***** Legend
Posts:1697
Joined:Tue Nov 26, 2002 3:59 am
Location:Crestview, Florida
I plan to write a summary thread in about two weeks. I may have a separate thread about the rationale, the reasoning behind the numbers, because that part will always be correct.

The biggie in what I have done is that I have identified a way of looking at Safe Withdrawal Rates that always makes sense. It clarifies your thinking. You automatically look at the right things.

Have fun.

John R.