The Money Mustache Community
General Discussion => Post-FIRE => Topic started by: SnackDog on March 24, 2015, 11:58:38 AM
-
With wild market fluctuations the last few years, I keep thinking about the implications of a changing net investment vs SWR. If one blindly locked in to 4% in 2009 and adjusted by inflation each year, they could really under spend over 30 years. What I plan to do is watch the investments the first 5-10 years. If 4% (or whatever you choose) of the current year investment is greater than the current year planned withdrawal (which would be initial SWR plus annual inflation adjustments), simply reset to 4% as if you had just retired this year. You are good to go for 30 years at that rate.
Put it another way, anytime your investment grows to the point where the current year withdrawal amount is less than 4% of your investments, simply reset to 4% and year 0.
A work colleague of mine retired in 1995. By 2005, he had spent about $2MM ($200K/year). But he had more principle than he started with in 1995. He was either under-spending or had amazing investment performance. His CPA kept urging him to step up the spending, but he couldn't figure out how. Haha.
Converse is not true. If the indices sink, historically you will survive without reducing below the 4% rate + annual inflation adjustments.
-
Put it another way, anytime your investment grows to the point where the current year withdrawal amount is less than 4% of your investments, simply reset to 4% and year 0.
Isn't that just inviting extra sequence of return risk?
-
Your colleague couldn't figure out how to spend more than $200k a year.. I think we have just hit on the ultimate first world problem..:)
-
One thing to consider is what inflation actually is. There are at least 10 ways to measure it and the ones the Government uses such as the CPI are minimized.
So what is your personal inflation rate? i.e. what do you spend money on?
Fixed mortgage --- no inflation there. (if you don't have one, why not? Great inflation hedge)
Insurance --- yep how much did that go up on average the last 5 years for you? (let's say $200 month)
Utilities -- Hell water alone is running 10-20% in some areas. What is your 5 year trend? (lets say $200 month)
Food -- A biggy for most folks. I'm guessing our food inflation is in the 7-10% range. (let's say $400 month)
Cars/gas --- Maybe not so big for you and hard to predict? (let's say $300 month)
Travel and entertainment? --- totally flexible category
So your "let's say" category might be around 12K per year. What is your core inflation there?
So yeah, figure out your own inflation rate to make an accurate call on your SWR. The hard part is predicting the future. That is why a mortgage is so powerful a hedge. Prices may quadruple on food in the next 30 years (likely) but my mortgage will remain the same at worst or go down with cash out at best.
Then on the income side consider if you are gambling in the market or investing in real estate. You really have no idea what the market will do but you will have a pretty good idea that real estate will provide X amount per year and self adjust for inflation.
Then and lord don't forget the impact of Social Security.
I'm guessing for folks with over 2 million in assets most of this is not worth their time?
-
Isn't that just inviting extra sequence of return risk?
Exactly. This proposal of setting a floor (but no ceiling) on withdrawals sounds like a great way to substantially increase the risk of portfolio failure. Let's take it as given that the future will be no worse than the past and we can therefore use historical data as a perfect predictive tool for the future. So using the standard 4% WR has a 95% chance of succeeding, if you stick with the 4% inflation-adjusted withdrawals for the entirety of the 30-year period. But every time you reset your WR upwards by "marking it to market," you are starting a new plan with a higher than 5/100 chance of failing. So you've taken your 95/100 odds of success and reduced them to something lower. If cfiresim were working right now, we could see exactly what the historical success rate was for this plan, looking at it as a single 30-year period (rather than rolling continuous 30-year periods).
-
I had the same thought.
Part of the reason why 4% worked is that sometimes the portfolio had grown before it hit the crash times, which let it survive (and only couldn't survive if it hit it too early). If you constantly reset upwards, you're not letting it grow as it might need to in order to survive, increasing your odds of failure.
All in all though, the idea of continually monitoring and deciding if you need to cut spending or if you can raise it is a good idea for FIRE. I just wouldn't set a rigid rule on either direction, but be flexible.
-
What several of you are suggesting is that if I retire today and start down my merry way at 4% and adjust the for inflation annually (I plan to just use CPI, since someone asked; I don't care what my personal inflation rate is since I can control it), I will make it 30 years. However, if I reset at year 5, according to the logic I presented, I would have a reduced chance of making it the last 25 years, to say nothing of another 30.
At the same time, if my neighbor retired with the same net worth I had in year 5, he would be good to go at 4% from that day and he would not only make it my last 25 years, but another 5.
What you are missing is that 4% of any principle is enough to start at and will last 30 years, adjusted annually for inflation. You can start or restart that 30 years and 4% any time you want. The math is the same. Under most scenarios, the inflation increases after a few years will have you needing more than the 4% so resetting would give you less, which is no good.
-
What you are missing is that 4% of any principle is enough to start at and will last 30 years, adjusted annually for inflation.
No, what we are saying is that the reason why 4% works as often as it does (93% of the time) is that when the crash happens, often the portfolio has built up enough that your WR is actually below 4% by that point, even though you've adjusted for inflation, because it outgrew inflation, and that outgrowing allows it to last.
Let me make up numbers to illustrate the point.
Lets say a crash happens at year X.
If you FIRE one year before that (at X-1) at 4% WR, you probably won't make it the 30 years. You'll be in the small handful of failure cases. Ditto at X-2. Maybe at X-3.. But maybe at X-4, X-5, etc. your portfolio has gained enough from those 4-5 years that when the crash happens at X, you last through it (in a way you wouldn't if you FIRE'd only 1 year before, at X-1).
But if you do your plan to reset every year, you're constantly risking a crash happening next. Thus -- like Eric said -- you're increasing your sequence of returns risk.
There are failure cases with the 4% rule, and by constantly resetting and potentially inflating your spending to 4% (but never reducing), you're vastly increasing your chance of being a failure case, because no longer do you have to get past the first few years of FIRE, but you have to continually do it.
-
Here's a good article on the subject -Retirement Distributions: How Much Can You Afford? - http://paulmerriman.com/retirement-distributions-how-much-can-you-afford-2/
I also like the free flexible retirement planner because you can run all the different scenarios.
-
I think just the opposite might need to be done. My understanding is the 4% rule is calculated for a 30 year retirement period, while many people here are potentially facing a much longer time period. So I think each year you could recalculate how much you're planning to withdraw, and compare that to 4% of your stashe. If you still have over 30 years left in your expected lifespan, reduce your spending/budget so you're withdrawing no more than 4%, until you get inside that 30 year window.
-
Excellent explanation by ars. If you still find yourself disagreeing, re-read that. As an alternative, since it sounds like what you're wanting is to be able to spend more as your stache grows, is you can just do 4% of current the stache, whether up or down. This lets you spend more as it grows, and also has a 0% failure rate. It can make down years a little lean, however.
-
Interesting question. I agree with ars and the others though, it doesn't work that way. When you retire with a 4% swr you're accepting a ~5% chance that you retired during a bad year that will result in a FIRE failure. By resetting your SWR, you're effectively re-retiring and taking an additional 5% risk of failure.
Even simpler analogy: you're playing Russian roulette with 100 chambers and 5 bullets. When you retire, you pull the trigger once. Each time you reset your SWR to a higher level, you're pulling the trigger one more time. Eventually you find a bullet.
-
What you are missing is that 4% of any principle is enough to start at and will last 30 years, adjusted annually for inflation.
No, what we are saying is that the reason why 4% works as often as it does (93% of the time) is that when the crash happens, often the portfolio has built up enough that your WR is actually below 4% by that point, even though you've adjusted for inflation, because it outgrew inflation, and that outgrowing allows it to last.
Let me make up numbers to illustrate the point.
Lets say a crash happens at year X.
If you FIRE one year before that (at X-1) at 4% WR, you probably won't make it the 30 years. You'll be in the small handful of failure cases. Ditto at X-2. Maybe at X-3.. But maybe at X-4, X-5, etc. your portfolio has gained enough from those 4-5 years that when the crash happens at X, you last through it (in a way you wouldn't if you FIRE'd only 1 year before, at X-1).
But if you do your plan to reset every year, you're constantly risking a crash happening next. Thus -- like Eric said -- you're increasing your sequence of returns risk.
There are failure cases with the 4% rule, and by constantly resetting and potentially inflating your spending to 4% (but never reducing), you're vastly increasing your chance of being a failure case, because no longer do you have to get past the first few years of FIRE, but you have to continually do it.
Bingo on the increased sequence of return risk. Suppose that the 4% SWR produces a 95% chance of success over your time period. Since the cause of failure happens at the start of the time period, chance of success can be estimated as 0.95^y, where "y" is the number of resets + 1.
Thus:
0 resets -> 95%
1 reset -> 90%
2 resets -> 86%
3 resets -> 81%
That said, the approach that you are considering might work if you went with a 3.5% SWR (98% chance of success). There you get:
0 resets -> 98%
1 reset -> 96%
2 resets -> 94%
3 resets -> 92%
You are even more free to do this if you reduce yourself to a 3% SWR.
-
I like it.. My strategy looks a lot like yours but I have not written it down like that.. But I too was planning 2.1% to 4%.
We have measured our spending to be slightly under $30k per year so our $1.5M stash gives us about a 2% WR... Of course thats a 3 week roadtrip as a vacation.. i.e not with the foreign travel.. so I can see that spiking up significantly.
Is your house paid off? or are you renting.. what size of stash are you working with?
-
IMHO, the best way to avoid "under spending" is to establish your maximum happiness for the dollar ratio far before FIRE and to simply continue that lifestyle after you retire. If adjusting up based on returns is on your radar, then to me that's a sign that either you still believe more spending creates more happiness, or you're perhaps tempted to pull the trigger too soon and hope to make up the shortfall with investments. Neither are particularly healthy retirement mindsets.
-
You can start or restart that 30 years and 4% any time you want. The math is the same.
I'm not sure (but can't yet prove without some more thought) that that's true. I'm reaching back really far into my math days (did undergrad and grad in math), but I recall expectation (probability) being different between a sequence of events, and a sequence of events given a prior set of events (which you are describing). Might be the Bernoulli theorems/equations???
That is, you can't just reset the clock, as it were, on a sequence and calculate the expectation going forward in the same way, and certainly not for dependent events. Since market events almost certainly have some dependency built in, the events preceding where you are -- that is, where you are in the sequence -- matters. That doesn't imply you can foresee the events (time the market), but the is dependency there (unlike, say, flips of a coin where each flip has nothing to do with prior flips).
I think this is what others are arguing from an intuitive standpoint.
-
IMHO, the best way to avoid "under spending" is to establish your maximum happiness for the dollar ratio far before FIRE and to simply continue that lifestyle after you retire.
+1
-
I'm not sure (but can't yet prove without some more thought) that that's true. I'm reaching back really far into my math days (did undergrad and grad in math), but I recall expectation (probability) being different between a sequence of events, and a sequence of events given a prior set of events (which you are describing).
Right, that's exactly what downtownshuter's apt Russian roulette analogy demonstrates and what electriceagle described. It is true that each time you pull the trigger, you have a 95/100 chance of not being shot: even if you've done it 95 times and not been shot, on the 96th trigger pull, you still have a 95/100 chance of not being shot. But your chances of repeatedly pulling the trigger over and over again and not being shot are not 95/100; if you do it three times, your chances are (95/100) x (95/100) x (95/100) = 85.74%. The more times you repeat it, the lower your chances of not being shot every single time in the sequence.
-
I'm not sure (but can't yet prove without some more thought) that that's true. I'm reaching back really far into my math days (did undergrad and grad in math), but I recall expectation (probability) being different between a sequence of events, and a sequence of events given a prior set of events (which you are describing).
Right, that's exactly what downtownshuter's apt Russian roulette analogy demonstrates and what electriceagle described. It is true that each time you pull the trigger, you have a 95/100 chance of not being shot: even if you've done it 95 times and not been shot, on the 96th trigger pull, you still have a 95/100 chance of not being shot. But your chances of repeatedly pulling the trigger over and over again and not being shot are not 95/100; if you do it three times, your chances are (95/100) x (95/100) x (95/100) = 85.74%. The more times you repeat it, the lower your chances of not being shot every single time in the sequence.
I don't think that's a valid analogy, and the math is certainly wrong. The analogy describes a simple binary scenario - yes or no, and then you can repeat. But the valid comparison for this analogy would be successive 30-year periods, with the withdrawal rate recalculated every 30 years, not a reevaluation of the withdrawal rate after, say, 5 years.
You could certainly test this type of strategy – if not in a simulator like cFIREsim, then using Excel spreadsheets and a lot of work.
-
I'm not sure (but can't yet prove without some more thought) that that's true. I'm reaching back really far into my math days (did undergrad and grad in math), but I recall expectation (probability) being different between a sequence of events, and a sequence of events given a prior set of events (which you are describing).
Right, that's exactly what downtownshuter's apt Russian roulette analogy demonstrates and what electriceagle described. It is true that each time you pull the trigger, you have a 95/100 chance of not being shot: even if you've done it 95 times and not been shot, on the 96th trigger pull, you still have a 95/100 chance of not being shot. But your chances of repeatedly pulling the trigger over and over again and not being shot are not 95/100; if you do it three times, your chances are (95/100) x (95/100) x (95/100) = 85.74%. The more times you repeat it, the lower your chances of not being shot every single time in the sequence.
I don't think that's a valid analogy, and the math is certainly wrong. The analogy describes a simple binary scenario - yes or no, and then you can repeat. But the valid comparison for this analogy would be successive 30-year periods, with the withdrawal rate recalculated every 30 years, not a reevaluation of the withdrawal rate after, say, 5 years.
You could certainly test this type of strategy – if not in a simulator like cFIREsim, then using Excel spreadsheets and a lot of work.
The reason why it doesn't have to be separate 30 year periods is that a crash in the middle of the 30-year period wouldn't necessarily be doom for someone whose stache had built up enough to ride it out. But the person who reset the year before that crash couldn't necessarily do so.
So it's more like "hope there's no sequence of returns risk -- crash or large inflation -- during any point of my retirement" rather than the current "hope there's no sequence of returns risk -- crash or large inflation -- at the beginning of my retirement"
-
The way the OP set up it up, it's rolling 30-year periods (the clock restarts with each "reset"), so it's a valid analogy if we assume that the chances of success at the start of each reset are 95% (which we can't really assume because we don't know the actual chances of success, but it's roughly what we do when we talk about a given SWR having an X% chance of success), no?
Is the math wrong? What's the correct math? (I guess I need a refresher on probabilistic math too.)
-
Is the math wrong? What's the correct math? (I guess I need a refresher on probabilistic math too.)
My simplistic answer would probably be to recalculate the Trinity calculations every year, using all of the prior years available to give an ever-refined SWR based on all the history we have. That way, your new calculated SWR takes into consideration all the prior events. It seems this is what Pfau and others are already doing, and why he's pointing to the current SWR being more in the 3.x% range nowadays given the latest market events. And, of course, you're always still left with the inevitable "past performance is no guarantee of future results."
-
It seems this is what Pfau and others are already doing, and why he's pointing to the current SWR being more in the 3.x% range nowadays given the latest market events.
I don't think that's quite right, because inclusion of the most recent data (the long bull market) in the historical dataset should increase the history-based success rate, not decrease it. Pfau is just looking at the indicators in the current market climate and predicting below-historical-average returns on that basis.
But I was just asking about the math still keeping with the Russian roulette analogy. (I've learned that when beltim says something, he's usually right.) If there are 100 chambers and 5 bullets, how do you calculate the odds of not getting shot in a sequence of three trigger pulls?
-
It seems this is what Pfau and others are already doing, and why he's pointing to the current SWR being more in the 3.x% range nowadays given the latest market events.
I don't think that's quite right, because inclusion of the most recent data (the long bull market) in the historical dataset should increase the history-based success rate, not decrease it. Pfau is just looking at the indicators in the current market climate and predicting below-historical-average returns on that basis.
But I was just asking about the math still keeping with the Russian roulette analogy. (I've learned that when beltim says something, he's usually right.) If there are 100 chambers and 5 bullets, how do you calculate the odds of not getting shot in a sequence of three trigger pulls?
You had that math right, 0.95^3. But I think he was arguing the analogy isn't correct, so that math is the wrong math to use.
-
You had that math right, 0.95^3. But I think he was arguing the analogy isn't correct, so that math is the wrong math to use.
Ah, got it. Thanks.
-
You had that math right, 0.95^3. But I think he was arguing the analogy isn't correct, so that math is the wrong math to use.
Ah, got it. Thanks.
Depends if you re-roll the cylinder. If you don't, the answer is (1-5/100)x(1-5/99)x(1-5/98) = 0.8560 chance of survival. If you pull the trigger 96 times, you are 0% chance of survival, not 0.95^96.
Sorry for the anal retentive response.
-
Good point Pom. Clearly the second isn't right (not EVERY 30 year period had a failure, so your success rate shouldn't be 0), but I think the first is not quite right either, as they're not completely independent.
Somewhere in the middle, perhaps.
-
Yes, I was assuming that you re-roll the cylinder after each trigger-pull. Isn't that how Russian roulette is played? I'm by no means an expert, but it seems like that's got to be the way it works -- I sure as hell wouldn't pull the trigger after 95 trigger pulls with no shot fired if the cylinder doesn't get re-rolled :-)
Also, now I see beltim's point -- the Russian roulette analogy would require consecutive fully-elapsed 30 year periods for the analogy to correctly apply.
-
I had my theorem name wrong when trying to remember -- it's Bayes, not Bernoulli (hey I got the first letter right!!). I suggest this is the appropriate methodology for looking at a sequence of data and "starting over." It is a conditional probability problem, and the theorem shows that the probability of an outcome can/will change depending on the prior conditions known.
It's a bit paradoxical, but if you step out of the sequence, you must then acknowledge the updated, prior conditions to get a new, accurate computation. Bayes' theorem would be the appropriate computation for playing Russian Roulette without spinning the chamber each time (for example, "What are the odds I will be shot (with a 6-chamber revolver) on the fourth attempt given that the trigger has already been pulled twice without firing?"):
http://en.wikipedia.org/wiki/Bayesian_inference
http://en.wikipedia.org/wiki/Bayes'_theorem
In probability theory and statistics, Bayes' theorem (alternatively Bayes' law or Bayes' rule) relates current probability to prior probability.
It is important in the mathematical manipulation of conditional probabilities.[1] Bayes' rule can be derived from more basic axioms of probability, specifically conditional probability.
...
Bayesian inference is a method of statistical inference in which Bayes' rule is used to update the probability for a hypothesis as evidence is acquired. Bayesian inference is an important technique in statistics, and especially in mathematical statistics. Bayesian updating is particularly important in the dynamic analysis of a sequence of data.
Bayes' theorem is stated mathematically as the following equation:[6]
P(A|B) = (P(B | A) * P(A)) / P(B)
where A and B are events.
P(A) and P(B) are the probabilities of A and B independent of each other.
P(A|B), a conditional probability, is the probability of A given that B is true.
P(B|A), is the probability of B given that A is true.
-
The way the OP set up it up, it's rolling 30-year periods (the clock restarts with each "reset"), so it's a valid analogy if we assume that the chances of success at the start of each reset are 95% (which we can't really assume because we don't know the actual chances of success, but it's roughly what we do when we talk about a given SWR having an X% chance of success), no?
Is the math wrong? What's the correct math? (I guess I need a refresher on probabilistic math too.)
The OP set up rolling 30 year periods, but the math on the analogy is only valid for consecutive year periods. Put another way, there are 0 historical cases of a 4% withdrawal rate failing in the first 5 years. Thus, we know that a 5% failure rate for the first period is wrong, and any analogy that uses that calculation is therefore wrong.
The question then becomes, is it accurate to use the 95% success rate if we start a new 30 year clock with a 4% withdrawal rate. Intuitively, I think it does, as long as you also allow for the possibility of your withdrawals decreasing. If you only have the possibility of increasing withdrawals, I don't think the 95% success rate is valid anymore – you'd have to recompute your success rate, based on what the market did over the last 5 years.
-
But that's what 4%, and the OP's post is based on, only increasing.
I agree one should be flexible, including decreasing spending. But that's not what the proposed rule is.
-
But that's what 4%, and the OP's post is based on, only increasing.
I agree one should be flexible, including decreasing spending. But that's not what the proposed rule is.
Well, right. But I'm interested in trying to figure out the success rate of that strategy would be. I know it's not just .95^(# of resets) for the reasons I described above. I also know it's not a flat 95% because the sequence of returns matters. I'm trying to figure out if there's an easy way to find the actual probability.
-
The OP set up rolling 30 year periods, but the math on the analogy is only valid for consecutive year periods. Put another way, there are 0 historical cases of a 4% withdrawal rate failing in the first 5 years. Thus, we know that a 5% failure rate for the first period is wrong, and any analogy that uses that calculation is therefore wrong.
The question then becomes, is it accurate to use the 95% success rate if we start a new 30 year clock with a 4% withdrawal rate. Intuitively, I think it does, as long as you also allow for the possibility of your withdrawals decreasing. If you only have the possibility of increasing withdrawals, I don't think the 95% success rate is valid anymore – you'd have to recompute your success rate, based on what the market did over the last 5 years.
Yes, you are correct, I realized the logic error (as I noted in reply # 28). The proper analog to Russian roulette would be consecutive, fully-elapsed 30-year periods (where each 30 year period starts off with a portfolio having a 95% chance of success). So that's not the proper analogy for this situation.
The OP's proposed rule would require resets upwards and not downwards (i.e., it sets a spending floor but not a spending ceiling).
Doubledown's "Bayes" method sounds like the right approach if the OP's proposed rule applied for the duration of a single 30-year period (and you can easily see the historical success rate of this approach in cfiresim by setting a variable spending plan with a floor but no ceiling).
But the OP's actual proposal is rolling 30-year periods, starting the clock over at each reset (but I don't think it really makes sense set it up that way or to view it from that perspective).
-
Alright, so I have a dataset for the S&P 500 that goes back to 1928. Using cFIREsim for all 30 year periods since 1928 gives a success rate of 51/58 or 88%. The failure years are:
1929
1965
1966
1967
1968
1969
1973
Let's restrict ourselves to only 30 year periods that start after a 5 year period with annualized returns in excess of 5% (to model cases where the OP might increase spending to a new 4%). That's 36 out of 53 periods ending in 1984. By comparing those to the starting failure years, we can get historical information on the success rate of this strategy. Notably, 1965, 1966, 1967, and 1968 all came after 5-year periods where greater than 5% annual returns. So, in the historical data set of 36 periods, 4 resulted in failure using the OP's idea (put into practice by 5% annual gains over the preceding 5 years). Note that the inflation rate in each of the failure times was less than 5%. So that's a historical success rate of 32/36 or 89%, statistically indistinguishable from the original, full data set.
-
Yes, you are correct, I realized the logic error (as I noted in reply # 28). The proper analog to Russian roulette would be consecutive, fully-elapsed 30-year periods (where each 30 year period starts off with a portfolio having a 95% chance of success). So that's not the proper analogy for this situation.
Yes, I saw that, sorry about that - I was just quoting the most relevant point in order to frame my comment appropriately.
Doubledown's "Bayes" method sounds like the right approach if the OP's proposed rule applied for the duration of a single 30-year period (and you can easily see the historical success rate of this approach in cfiresim by setting a variable spending plan with a floor but no ceiling).
But the OP's actual proposal is rolling 30-year periods, starting the clock over at each reset (but I don't think it really makes sense set it up that way or to view it from that perspective).
I think your analysis of Bayesian statistics is right. I'm trying to figure out the right stats for what the OP is actually suggesting - basically, how the probability of the 4% WR succeeding varies depending on the previous x years of market returns. I think it's an interesting question, particularly in that it has applications beyond the OP's question. The market returns in the few years before retiring certainly affect your probability of success, and it would be interesting to know how. Unfortunately, I suspect there isn't enough data in the historical record to get a statistically significant answer.
-
Unfortunately, I suspect there isn't enough data in the historical record to get a statistically significant answer.
I would tend to agree. I had thought cfiresim's variable spending plan options allowed you to set floors and/or ceilings based on a percentage of the running portfolio value (which would allow us to model the historical success rate of the proposed plan over a desired period length, like, say, 60 years), but looking at the options it seems that only defined values are available as floor/ceiling options. I would imagine that cfiresim could be revised to create this option (though I know bo_knows already has a lot on his plate at the moment).
-
This sounds like a terrible plan. It guarantees that you will "retire" into a bear market without an adequate buffer. If you happen to live through any semi major crash in your retirement life time you are guaranteeing that you are going to be one of those few failures.
-
Alright, so I have a dataset for the S&P 500 that goes back to 1928. Using cFIREsim for all 30 year periods since 1928 gives a success rate of 51/58 or 88%. The failure years are:
1929
1965
1966
1967
1968
1969
1973
When doing cFIREsim and calculating our success rate (over and over) before deciding at last to FIRE, I couldn't help but imagine what it'd be like retiring in those years 65-69. Many people back then retired with pensions, so they were likely okay. Others, given the life expectancy at the time, maybe didn't need a 30 year stash. Still others, like my dad's parents, retired with only their savings. My grandparents, despite their timing, seemed to do great.
In large part their success was because of my Grandma. She managed the stash. When I was still in single digits, she explained to me how she picked stocks: "I buy what I know, and look for value". She explained what a balance sheet was and that she made sure any stock she picked had a healthy balance sheet. She kept on the look-out for the 'must have' items her friends raved about. Then, rather than buying those items, she bought the stock (after a full review of the public books).
Later, when I read Berkshire Hathaway annual reports and interviews with Warren Buffet I couldn't help but think of my Grandma.
Of course now-a-days, indexing is the pretty clear way to go. But she did great.
As for resetting your spend rate to 4% every year the stash ticks up? Shudder! You're just asking for problems. Our FIRE needs to last 60 years if everything goes to plan. Upticking to 4% would terrify me.
-
Many people back then retired with pensions, so they were likely okay. Others, given the life expectancy at the time, maybe didn't need a 30 year stash.
I remember people who retired at that time. They didn't have pensions. They were poor.
-
Many people back then retired with pensions, so they were likely okay. Others, given the life expectancy at the time, maybe didn't need a 30 year stash.
I remember people who retired at that time. They didn't have pensions. They were poor.
You are correct, at least as far back as 1979 (and probably further), and if we're only counting private workers.
http://www.ebri.org/publications/benfaq/index.cfm?fa=retfaq14
Looks like 38% of private sector workers had pensions (28% with only pensions and 10% with both db and dc plans), down to 14% today (3% with only pensions, 11% with both) today. Of course, you have to add public workers in, so that might push it over the 50% mark to "most" retiring with pension.
Still, the point remains, somewhere around half, even back then, didn't have pensions. The numbers are much lower today.
-
Alright, so I have a dataset for the S&P 500 that goes back to 1928. Using cFIREsim for all 30 year periods since 1928 gives a success rate of 51/58 or 88%. The failure years are:
1929
1965
1966
1967
1968
1969
1973
Let's restrict ourselves to only 30 year periods that start after a 5 year period with annualized returns in excess of 5% (to model cases where the OP might increase spending to a new 4%). That's 36 out of 53 periods ending in 1984. By comparing those to the starting failure years, we can get historical information on the success rate of this strategy. Notably, 1965, 1966, 1967, and 1968 all came after 5-year periods where greater than 5% annual returns. So, in the historical data set of 36 periods, 4 resulted in failure using the OP's idea (put into practice by 5% annual gains over the preceding 5 years). Note that the inflation rate in each of the failure times was less than 5%. So that's a historical success rate of 32/36 or 89%, statistically indistinguishable from the original, full data set.
Ok, I've now focused on this post and taken the time to digest the methodology you used here (which was frankly brilliant).
If we put aside for a moment the fact that there probably just isn't enough data to allow us to draw statistically significant conclusions (which I agree is a real problem), then what your analysis tells us is that retiring after a large market run-up (on an annualized basis) does *not* result in decreased chances of portfolio success. That is an incredibly counterintuitive result.
-
if you only ever take 4% then you'll always have a stash.
as long as you adjust on the way down as well as the way up.
There is 0% chance of failure. of course you have to make sure that 4% of the bottom of the cycle is enough to cover your expenses...
-
if you only ever take 4% then you'll always have a stash.
as long as you adjust on the way down as well as the way up.
There is 0% chance of failure. of course you have to make sure that 4% of the bottom of the cycle is enough to cover your expenses...
Yes, cFIREsim has an option like this, to reduce your spending but set a "floor" (to make sure the amount doesn't go below the minimum you need).
-
Okay, so I've been silently reading along... The OP's suggestion sounded reasonable... and the counter arguments to it have convinced me it isn't.
But... I've been probably over saving. I probably over state my expenses and I a think I am being ultra conservative with everything I estimate.
So my follow up here is: what *is* a viable method to re-evaluate and reset spending? (And I know: this is such a pissy first world problem. If I die with a big pile of cash -- that's not a terrible problem to have.)
-
Okay, so I've been silently reading along... The OP's suggestion sounded reasonable... and the counter arguments to it have convinced me it isn't.
But... I've been probably over saving. I probably over state my expenses and I a think I am being ultra conservative with everything I estimate.
So my follow up here is: what *is* a viable method to re-evaluate and reset spending? (And I know: this is such a pissy first world problem. If I die with a big pile of cash -- that's not a terrible problem to have.)
Depends on how conservative you want to be - in other words, what success rate is acceptable to you.
And also how flexible you are with your spending, and if you're willing to take a cut from year to year if the market is down.
IMO, just run cFIREcalc each year with your current parameters. You can use the "investigate" function to determine your max SWR at the present time based on the success rate you feel comfortable with (i.e. plug in 90% or 95% or 100% or whatever) and see what is the max spending you could do for the year, and cap at that (or lower, again, if you want to feel comfortable).
Setting rigid "rules" seems silly - pick what makes you comfortable and adjust based on that.
-
Alright, so I have a dataset for the S&P 500 that goes back to 1928. Using cFIREsim for all 30 year periods since 1928 gives a success rate of 51/58 or 88%. The failure years are:
1929
1965
1966
1967
1968
1969
1973
Let's restrict ourselves to only 30 year periods that start after a 5 year period with annualized returns in excess of 5% (to model cases where the OP might increase spending to a new 4%). That's 36 out of 53 periods ending in 1984. By comparing those to the starting failure years, we can get historical information on the success rate of this strategy. Notably, 1965, 1966, 1967, and 1968 all came after 5-year periods where greater than 5% annual returns. So, in the historical data set of 36 periods, 4 resulted in failure using the OP's idea (put into practice by 5% annual gains over the preceding 5 years). Note that the inflation rate in each of the failure times was less than 5%. So that's a historical success rate of 32/36 or 89%, statistically indistinguishable from the original, full data set.
Ok, I've now focused on this post and taken the time to digest the methodology you used here (which was frankly brilliant).
If we put aside for a moment the fact that there probably just isn't enough data to allow us to draw statistically significant conclusions (which I agree is a real problem), then what your analysis tells us is that retiring after a large market run-up (on an annualized basis) does *not* result in decreased chances of portfolio success. That is an incredibly counterintuitive result.
Beltim, did the data you used for historical S&P 500 returns include reinvestment of dividends? And was it for returns in nominal terms?
The result seemed so counterintuitive as to almost defy explanation. But given that the starting years for the failure cases are so clustered around the second half of the 1960s (meaning that almost all the failures resulted from the same single 34-year period of poor real market performance), do you think it would be a useful exercise to do the same analysis using a more failure-prone WR (like maybe 5%)? Perhaps once there are more failure cases in the dataset, the analysis will show that the market returns in the short period leading up to retirement did have a statistically meaningful effect on the probability of success (though there's probably still not enough data in the overall historical record for the answer to really have statistical significance).
-
Alright, so I have a dataset for the S&P 500 that goes back to 1928. Using cFIREsim for all 30 year periods since 1928 gives a success rate of 51/58 or 88%. The failure years are:
1929
1965
1966
1967
1968
1969
1973
Let's restrict ourselves to only 30 year periods that start after a 5 year period with annualized returns in excess of 5% (to model cases where the OP might increase spending to a new 4%). That's 36 out of 53 periods ending in 1984. By comparing those to the starting failure years, we can get historical information on the success rate of this strategy. Notably, 1965, 1966, 1967, and 1968 all came after 5-year periods where greater than 5% annual returns. So, in the historical data set of 36 periods, 4 resulted in failure using the OP's idea (put into practice by 5% annual gains over the preceding 5 years). Note that the inflation rate in each of the failure times was less than 5%. So that's a historical success rate of 32/36 or 89%, statistically indistinguishable from the original, full data set.
Ok, I've now focused on this post and taken the time to digest the methodology you used here (which was frankly brilliant).
If we put aside for a moment the fact that there probably just isn't enough data to allow us to draw statistically significant conclusions (which I agree is a real problem), then what your analysis tells us is that retiring after a large market run-up (on an annualized basis) does *not* result in decreased chances of portfolio success. That is an incredibly counterintuitive result.
Beltim, did the data you used for historical S&P 500 returns include reinvestment of dividends? And was it for returns in nominal terms?
The result seemed so counterintuitive as to almost defy explanation. But given that the starting years for the failure cases are so clustered around the second half of the 1960s (meaning that almost all the failures resulted from the same single 34-year period of poor real market performance), do you think it would be a useful exercise to do the same analysis using a more failure-prone WR (like maybe 5%)? Perhaps once there are more failure cases in the dataset, the analysis will show that the market returns in the short period leading up to retirement did have a statistically meaningful effect on the probability of success (though there's probably still not enough data in the overall historical record for the answer to really have statistical significance).
At some point of course the years you retire will matter more, but I think one potential takeaway is that 4% seems safe enough in those circumstances, even when nearing a market peak.
-
At some point of course the years you retire will matter more, but I think one potential takeaway is that 4% seems safe enough in those circumstances, even when nearing a market peak.
Yes, I agree that's an important takeaway. I just find it extremely interesting that the data seems to say that retiring after a steep market run-up had zero effect on historical chances of success, which might mitigate the concern expressed in this thread (http://forum.mrmoneymustache.com/ask-a-mustachian/firecalc-and-cfiresim-both-lie/) regarding the presumed greater likelihood of hitting your retirement-commencement-trigger portfolio value at a time with higher-than-average probability of portfolio failure.
-
Ok, I've now focused on this post and taken the time to digest the methodology you used here (which was frankly brilliant).
If we put aside for a moment the fact that there probably just isn't enough data to allow us to draw statistically significant conclusions (which I agree is a real problem), then what your analysis tells us is that retiring after a large market run-up (on an annualized basis) does *not* result in decreased chances of portfolio success. That is an incredibly counterintuitive result.
Hey, thanks!
Beltim, did the data you used for historical S&P 500 returns include reinvestment of dividends? And was it for returns in nominal terms?
Yes on both - it included dividends, and was in nominal terms. However, the inflation in each 5 year period below that was less than 5% annualized.
The result seemed so counterintuitive as to almost defy explanation. But given that the starting years for the failure cases are so clustered around the second half of the 1960s (meaning that almost all the failures resulted from the same single 34-year period of poor real market performance), do you think it would be a useful exercise to do the same analysis using a more failure-prone WR (like maybe 5%)? Perhaps once there are more failure cases in the dataset, the analysis will show that the market returns in the short period leading up to retirement did have a statistically meaningful effect on the probability of success (though there's probably still not enough data in the overall historical record for the answer to really have statistical significance).
I like the idea, but I wonder about the relevance of using a 5% withdrawal rate to look into the success rate of a 4% withdrawal rate.
-
Unfortunately, I suspect there isn't enough data in the historical record to get a statistically significant answer.
I would tend to agree. I had thought cfiresim's variable spending plan options allowed you to set floors and/or ceilings based on a percentage of the running portfolio value (which would allow us to model the historical success rate of the proposed plan over a desired period length, like, say, 60 years), but looking at the options it seems that only defined values are available as floor/ceiling options. I would imagine that cfiresim could be revised to create this option (though I know bo_knows already has a lot on his plate at the moment).
I was playing around with cFIREsim and I think the "Retire Again & Again Method" is a pretty good simulation. Set to a 4% withdrawal rate, and increases any time the market goes up 3% (a crude method for inflation), with a floor in constant dollars of 4% of the initial starting value, the success rate falls from 107/115 to 105/115 or, from 93% to 91.3%.
-
I like the idea, but I wonder about the relevance of using a 5% withdrawal rate to look into the success rate of a 4% withdrawal rate.
It wouldn't be relevant for telling us the impact of market performance in the years leading up to retirement on the specific historical success rate of a 4% withdrawal rate, but it would help to answer the question of whether there is correlation between market performance in the years leading up to retirement and historical portfolio success rates in general. The 4% SWR was intentionally chosen for its status as sustainable in the worst case (or near-worst case) scenario, so by design I think it doesn't produce enough historical failures to properly test the correlation you were testing for.
I was playing around with cFIREsim and I think the "Retire Again & Again Method" is a pretty good simulation. Set to a 4% withdrawal rate, and increases any time the market goes up 3% (a crude method for inflation), with a floor in constant dollars of 4% of the initial starting value, the success rate falls from 107/115 to 105/115 or, from 93% to 91.3%.
The "Retire Again & Again Method" is probably the cfiresim option that most closely approximates the OP's strategy, but it doesn't reflect it exactly because it's ratcheting up the withdrawals based on crossing a specified threshold for a percentage increase in the real value of the portfolio rather than the amount of the withdrawals as a percentage of the real value of the portfolio. I didn't follow why you set the spending increase to 3% -- what does inflation have to do with the spending increase threshold? Inflation is already reflected in both the constant-dollar withdrawal floor and the running portfolio value. Wouldn't it be a closer approximation of the OP's strategy to use the lowest threshold for spending increases possible, like 0.01%? (Doing so results in the same 91.3% you reported, but I think that's because this approach still suffers from the problem of insufficient number of failure cases for statistically significant conclusions.)
Either way, we can use this method to try to approximate the OP's strategy over a full period (consisting of successive individual rolling 30-year periods) of a given length. For a full 60-year retirement, for example, the "Retire Again & Again Method" with the threshold set at 3% reduces the chances of success from 82.35% to 74.12%. Lowering the threshold down to 0.01% doesn't change this result.
-
I didn't follow why you set the spending increase to 3% -- what does inflation have to do with the spending increase threshold? Inflation is already reflected in both the constant-dollar withdrawal floor and the running portfolio value. Wouldn't it be a closer approximation of the OP's strategy to use the lowest threshold for spending increases possible, like 0.01%? (Doing so results in the same 91.3% you reported, but I think that's because this approach still suffers from the problem of insufficient number of failure cases for statistically significant conclusions.)
Yes, I just picked an arbitrary amount greater than 0, figuring that one wouldn't bother adjusting the withdrawal size for tiny amounts (3% is probably still to small for this).
-
I like the idea, but I wonder about the relevance of using a 5% withdrawal rate to look into the success rate of a 4% withdrawal rate.
It wouldn't be relevant for telling us the impact of market performance in the years leading up to retirement on the specific historical success rate of a 4% withdrawal rate, but it would help to answer the question of whether there is correlation between market performance in the years leading up to retirement and historical portfolio success rates in general. The 4% SWR was intentionally chosen for its status as sustainable in the worst case (or near-worst case) scenario, so by design I think it doesn't produce enough historical failures to properly test the correlation you were testing for.
Yeah, this is a good idea. I initially thought I could simply ramp up the previous 5, 10, or 20 years return and see what retirement success rates were when one retired after great market runs. Then I realized that would still suffer from the same problem of not enough failures.
And, honestly, an analysis of the failures that started in the 60s pretty clearly shows that the problem wasn't really low market returns – the problem was the crazy high inflation of the 70s. If you look at 20 year retirement period with a withdrawal rate of 6%, the success rate is about 76%. And one of the big blocks of failures is starting periods in the 60s. But periods with not great returns, but low inflation (i.e. periods ending in 2009 or later) work out pretty well.
-
Yes, I just picked an arbitrary amount greater than 0, figuring that one wouldn't bother adjusting the withdrawal size for tiny amounts (3% is probably still to small for this).
Yeah. After posting, though, I realized you can set the threshold to exactly 0, which (when combined with the constant-dollar floor set to an amount equal to 4% of the original portfolio) actually replicates the OP's proposed strategy exactly.
-
And, honestly, an analysis of the failures that started in the 60s pretty clearly shows that the problem wasn't really low market returns – the problem was the crazy high inflation of the 70s. If you look at 20 year retirement period with a withdrawal rate of 6%, the success rate is about 76%. And one of the big blocks of failures is starting periods in the 60s. But periods with not great returns, but low inflation (i.e. periods ending in 2009 or later) work out pretty well.
Right, but I don't know if I would state it that way. All you're really saying is that the problem was really low market returns in real terms.
What your analysis demonstrated is that (historically) retiring after a great market run should not have caused you to fear above-average chances of portfolio failure, which struck me as completely counterintuitive and is probably something most of us (or none of us) ever realized. However, that is looking at market prices in isolation (and I wonder if the results of the analysis would differ if the 5-year annualized return data you used represented real returns instead of nominal, but my hunch is that wouldn't make a huge difference). Looking at market prices as they relate to earnings (or another market valuation metric), on the other hand, tells a different story, as most recently discussed in this thread: http://forum.mrmoneymustache.com/investor-alley/using-market-valuation-measurements-to-affect-swr/
-
So my follow up here is: what *is* a viable method to re-evaluate and reset spending?
I see an obvious answer to this question: don't re-evaluate at all until you have a better idea of your time horizon. If you're 90 and still have a million dollars left, you can probably crank it up. If you're 30 you probably can't, just because over long time periods your crystal ball is less certain.
That's the real problem here, as beltim and brook discussed above. It's not like the 91% success vs the 95% success rates mean anything, when 6 of the 8 failures are caused by the same crazy inflation rates in near-consecutive years. That was really just one historical event, and none of us have any idea of the real likelihood of that kind of event happening again. Which is why I think anything over about 80% success is just wishful thinking.
do you think it would be a useful exercise to do the same analysis using a more failure-prone WR (like maybe 5%)? Perhaps once there are more failure cases in the dataset, the analysis will show that
I don't think it's useful, no. If the difference in failure rates is only due to how effectively you capture the stagflation of the 1970s, or other similar grouped sequences of bad timing years, then your percentage outcomes don't really tell you anything meaningful about future failure rates. We could have another bad run like the 70s or the 30s at any moment, and what you really need to know is what the likelihood of that kind of bad run happening again is. Historical returns don't help you with that estimate, they only provide false precision by turning those two events into a whole bunch of failure years.
I think cranking the SWR up to 5% is functionally equivalent to doing the same analysis on monthly returns, instead of annual returns. You've just oversampled the same data to get more data points, but it's only going to give you false precision.
-
I think cranking the SWR up to 5% is functionally equivalent to doing the same analysis on monthly returns, instead of annual returns. You've just oversampled the same data to get more data points, but it's only going to give you false precision.
I don't think this is quite right, though it may be functionally equivalent (or at least analogous) to doing the same analysis on a dialed-down threshold for market returns in order to get more data points (the opposite of what beltim suggested in post # 52). I do agree that this entire exercise has limited utility because of the false precision issue, but isn't that equally true of SWR research in general?
(This is related to the recent discussion in this thread (http://forum.mrmoneymustache.com/investor-alley/paying-off-mortgage-early-how-bad-is-it-for-your-fi-date/msg602024/#msg602024) (primarily between skyrefuge and me, interspersed between a lot of noise about the broader topic of investing vs. paying off a mortgage in general).)
-
I'm not entirely sure that retire again and again is implemented correctly on c-firesim - really looks like the spending only ever resets once (not again and again).
Apparently this very idea was discussed here a year and a half ago:
http://forum.mrmoneymustache.com/investor-alley/new-withdrawal-method-retire-again-again/ (http://forum.mrmoneymustache.com/investor-alley/new-withdrawal-method-retire-again-again/)
-
I'm not entirely sure that retire again and again is implemented correctly on c-firesim - really looks like the spending only ever resets once (not again and again).
I think you're right. Based on the "spending level" graph in the cfiresim output (which I hadn't paid attention to before), it looks like there's only one spending reset. So the 91.3% success rate referenced above is overstating the historical success rate of the OP's proposed strategy (perhaps grossly so?).
Apparently this very idea was discussed here a year and a half ago:
http://forum.mrmoneymustache.com/investor-alley/new-withdrawal-method-retire-again-again/ (http://forum.mrmoneymustache.com/investor-alley/new-withdrawal-method-retire-again-again/)
This appears to be the very idea (via its expression a few months earlier in a Bogleheads forum thread started by the same poster) that gave rise to the "Retire Again & Again" spending plan option in cfiresim. So is there just a glitch in cfiresim? It seems to be implementing a spending plan of "Retire Again (But Not Again)" instead of "Retire Again & Again."
-
I'm not entirely sure that retire again and again is implemented correctly on c-firesim - really looks like the spending only ever resets once (not again and again).
I think you're right. Based on the "spending level" graph in the cfiresim output (which I hadn't paid attention to before), it looks like there's only one spending reset. So the 91.3% success rate referenced above is overstating the historical success rate of the OP's proposed strategy (perhaps grossly so?).
Apparently this very idea was discussed here a year and a half ago:
http://forum.mrmoneymustache.com/investor-alley/new-withdrawal-method-retire-again-again/ (http://forum.mrmoneymustache.com/investor-alley/new-withdrawal-method-retire-again-again/)
This appears to be the very idea (via its expression a few months earlier in a Bogleheads forum thread started by the same poster) that gave rise to the "Retire Again & Again" spending plan option in cfiresim. So is there just a glitch in cfiresim? It seems to be implementing a spending plan of "Retire Again (But Not Again)" instead of "Retire Again & Again."
Could be just the spending graph, I suppose. But the spending graph has looked right on other options. I was expecting spending to look more like stair-steps and less like one single jump (often very early in the scenario) followed by a plateau.
-
I do agree that this entire exercise has limited utility because of the false precision issue, but isn't that equally true of SWR research in general?
The SWR dataset is embarrassingly small, given that we are trying to forecast 30 years based on 80? years of sequentially-correlated data. I wonder whether the difference between 5% and 15% failure rates is even statistically significant.
That said, SWR still provides value in framing a way to think about retirement spending.
As far as RAA, this approach guarantees that you retire into a declining market. Not good.
-
I wonder whether the difference between 5% and 15% failure rates is even statistically significant..
Most sources say no difference. Anything above about 80% success is just giving you a false sense of security.
The real question here is what the relevant decades of the future world economy look like. If your early retirement decades look like the 1930s or 70s, your 3% or 5% plan might fail. If they look like any other decades you'll be fine. I'm not sure it makes sense to count up the number of failure rate years in those past decades as a proxy for how likely those types of economic conditions are to reoccur.
-
Step 1. Own your home.
Step 2. Have no debt.
Step 3. Have ~ $1 M in investments churning out 3-5% in dividends per year.
Step 4. Spend the dividends ($30k - $50k) per year and keep capital intact until old age, so you can afford home care to pay for expenses.
Step 5. Forget about the SWR.
That should do it :)
-
Step 1. Own your home.
Step 2. Have no debt.
Step 3. Have ~ $1 M in investments churning out 3-5% in dividends per year.
Step 4. Spend the dividends ($30k - $50k) per year and keep capital intact until old age, so you can afford home care to pay for expenses.
Step 5. Forget about the SWR.
That should do it :)
So your spending could fluctuate by as much as 66% in a given year, based on dividends?
..yeah, I think I'll stick with a more traditional ER plan.
-
This thread is interesting.
It's made me realize that my understanding of the 4% rule was flawed. It never occurred to me that I could continue withdrawing 4% of the starting value of my portfolio every year, no matter what happened, i.e. retire with $1M, withdraw $40,000 the first year, in year two portfolio value drops 40% to $600,000, still withdraw $40K... My understanding now is that, historically, most times, even if I did that, probably my nest egg would survive for 30 years.
To me, that seems like pretty reckless behavior, and I have no intention of doing that, but it's nice to know that even if I did, probably I wouldn't run out of money.
My wife and I are planning on being flexible and definitely reducing withdrawals, probably down to just dividends, from our stash during serious downturns in the stock market, and getting at least part time jobs doing something low stress and relatively enjoyable (because we know it's just temporary). When stocks go on sale during market crashes, we're planning on making enough money to at least max out our IRAs for a year or two, so that we can buy shares at a discount. Conversely, as our stash grows, I won't have any fear about spending 5% or even more, some years, because I'll know that when the markets head south, I'll be ready and able to reduce expenses accordingly, either by working and/or using geographic arbitrage and moving to a lower COL part of the country/world.
-
Yes, I was assuming that you re-roll the cylinder after each trigger-pull. Isn't that how Russian roulette is played? I'm by no means an expert, but it seems like that's got to be the way it works -- I sure as hell wouldn't pull the trigger after 95 trigger pulls with no shot fired if the cylinder doesn't get re-rolled :-)
Also, now I see beltim's point -- the Russian roulette analogy would require consecutive fully-elapsed 30 year periods for the analogy to correctly apply.
Think of each year in the last 100 years as a chamber. In this analogy, the bullets would be distributed throughout the gun in the same places as a retirement start date for a 4% SWR failure scenario. So you would not re-roll the chamber. You would proceed in order year-to-year (chamber-to-chamber) until you hit a bullet or no longer need your portfolio (which sounds like you died either way). Now technically you wouldn't necessarily have portfolio failure if you hit a bullet, say in year 25 and you were only planning to need your money for 40 years, and the sequence of returns you hit at year 25 just happens to allow your portfolio to last for 15 years. So the math gets a little complicated there. You could apply a damping function to each successive year to lower the effect of it on your portfolio success.
-
Alright, so I have a dataset for the S&P 500 that goes back to 1928. Using cFIREsim for all 30 year periods since 1928 gives a success rate of 51/58 or 88%. The failure years are:
1929
1965
1966
1967
1968
1969
1973
Let's restrict ourselves to only 30 year periods that start after a 5 year period with annualized returns in excess of 5% (to model cases where the OP might increase spending to a new 4%). That's 36 out of 53 periods ending in 1984. By comparing those to the starting failure years, we can get historical information on the success rate of this strategy. Notably, 1965, 1966, 1967, and 1968 all came after 5-year periods where greater than 5% annual returns. So, in the historical data set of 36 periods, 4 resulted in failure using the OP's idea (put into practice by 5% annual gains over the preceding 5 years). Note that the inflation rate in each of the failure times was less than 5%. So that's a historical success rate of 32/36 or 89%, statistically indistinguishable from the original, full data set.
Ok, I've now focused on this post and taken the time to digest the methodology you used here (which was frankly brilliant).
If we put aside for a moment the fact that there probably just isn't enough data to allow us to draw statistically significant conclusions (which I agree is a real problem), then what your analysis tells us is that retiring after a large market run-up (on an annualized basis) does *not* result in decreased chances of portfolio success. That is an incredibly counterintuitive result.
I think this is a nice analysis. But I don't draw the same conclusion. I assume Beltim did not adjust for inflation here (no mention of real returns). Given that annualized returns during that 1928+ time period were somewhere around 10% (7% real), 5% returns are not indicative of a bubble, and may not even have kept up with normal historical returns. Beltim doesn't say if each individual year in the 5-year periods had a positive return >= 5%. It could be read that way or just that the 5 year period had a CAGR of 5% or more. So unless I'm not understanding what analysis was done, I'm not sure that we can make the conclusion you state here. And I'm not really sure what it does say. I would be interested in hearing more detail about the methodology and other findings from this analysis. Maybe even posting the spreadsheet used so we can see what happened in the scenarios that were excluded, etc.
-
Beltim, did the data you used for historical S&P 500 returns include reinvestment of dividends? And was it for returns in nominal terms?
Yes on both - it included dividends, and was in nominal terms. However, the inflation in each 5 year period below that was less than 5% annualized.
So it would theoretically have been possible for there to be 5 years of 6% returns each year, and inflation to have been 4% each year?
-
And, honestly, an analysis of the failures that started in the 60s pretty clearly shows that the problem wasn't really low market returns – the problem was the crazy high inflation of the 70s. If you look at 20 year retirement period with a withdrawal rate of 6%, the success rate is about 76%. And one of the big blocks of failures is starting periods in the 60s. But periods with not great returns, but low inflation (i.e. periods ending in 2009 or later) work out pretty well.
Right, but I don't know if I would state it that way. All you're really saying is that the problem was really low market returns in real terms.
What your analysis demonstrated is that (historically) retiring after a great market run should not have caused you to fear above-average chances of portfolio failure, which struck me as completely counterintuitive and is probably something most of us (or none of us) ever realized.
I think the bolded section is the core of what I think is a misunderstanding of the analysis. If I understand what Beltim did, a 5% CAGR in nominal terms is in no way a great market run. It's actually pretty poor historically speaking. So, if I understand correctly, this did not tell us much about retiring when equity prices are substantially overvalued historically. A better analysis would be to look at what the failure rate was for retirements starting in a year when CAPE (or a similar metric) was, say, over 20. There aren't a lot of those years with 30 years following that.
1901
1902
1903
1906
1929
1930
1937
1962
1964
1965
1966
1967
1968
1969
I bet you would find more than 4 failures here.
-
And, honestly, an analysis of the failures that started in the 60s pretty clearly shows that the problem wasn't really low market returns – the problem was the crazy high inflation of the 70s. If you look at 20 year retirement period with a withdrawal rate of 6%, the success rate is about 76%. And one of the big blocks of failures is starting periods in the 60s. But periods with not great returns, but low inflation (i.e. periods ending in 2009 or later) work out pretty well.
Right, but I don't know if I would state it that way. All you're really saying is that the problem was really low market returns in real terms.
What your analysis demonstrated is that (historically) retiring after a great market run should not have caused you to fear above-average chances of portfolio failure, which struck me as completely counterintuitive and is probably something most of us (or none of us) ever realized.
I think the bolded section is the core of what I think is a misunderstanding of the analysis. If I understand what Beltim did, a 5% CAGR in nominal terms is in no way a great market run. It's actually pretty poor historically speaking. So, if I understand correctly, this did not tell us much about retiring when equity prices are substantially overvalued historically. A better analysis would be to look at what the failure rate was for retirements starting in a year when CAPE (or a similar metric) was, say, over 20. There aren't a lot of those years with 30 years following that.
1901
1902
1903
1906
1929
1930
1937
1962
1964
1965
1966
1967
1968
1969
I bet you would find more than 4 failures here.
It looks like there are 5 failures: 1965-1969 using cFIREsim's default portfolio. So 9/14=64% success rate after CAPE>20 retirements. A bunch of the earliest ones get pretty close to depleting the portfolio, but survive. CAPE is 27 right now. The only time it's been that high with 30 years after that is 1929 (which succeeded, but spent about 20 years hovering around 50% of the initial portfolio value, requiring 8% withdrawals year after year).
I don't think it's correct to say that elevated market values are not associated with higher failure rates.
-
I think the bolded section is the core of what I think is a misunderstanding of the analysis. If I understand what Beltim did, a 5% CAGR in nominal terms is in no way a great market run. It's actually pretty poor historically speaking. So, if I understand correctly, this did not tell us much about retiring when equity prices are substantially overvalued historically. A better analysis would be to look at what the failure rate was for retirements starting in a year when CAPE (or a similar metric) was, say, over 20.
It wasn't a misunderstanding - that was what was explicitly stated in the second half of the post you quoted from:
Right, but I don't know if I would state it that way. All you're really saying is that the problem was really low market returns in real terms.
What your analysis demonstrated is that (historically) retiring after a great market run should not have caused you to fear above-average chances of portfolio failure, which struck me as completely counterintuitive and is probably something most of us (or none of us) ever realized. However, that is looking at market prices in isolation (and I wonder if the results of the analysis would differ if the 5-year annualized return data you used represented real returns instead of nominal, but my hunch is that wouldn't make a huge difference). Looking at market prices as they relate to earnings (or another market valuation metric), on the other hand, tells a different story, as most recently discussed in this thread: http://forum.mrmoneymustache.com/investor-alley/using-market-valuation-measurements-to-affect-swr/
The linked thread does almost exactly the analysis you suggested, showing that retirement start years with high CAPEs have historically been much more failure-prone.
-
I think the bolded section is the core of what I think is a misunderstanding of the analysis. If I understand what Beltim did, a 5% CAGR in nominal terms is in no way a great market run. It's actually pretty poor historically speaking. So, if I understand correctly, this did not tell us much about retiring when equity prices are substantially overvalued historically. A better analysis would be to look at what the failure rate was for retirements starting in a year when CAPE (or a similar metric) was, say, over 20.
It wasn't a misunderstanding - that was what was explicitly stated in the second half of the post you quoted from:
Right, but I don't know if I would state it that way. All you're really saying is that the problem was really low market returns in real terms.
What your analysis demonstrated is that (historically) retiring after a great market run should not have caused you to fear above-average chances of portfolio failure, which struck me as completely counterintuitive and is probably something most of us (or none of us) ever realized. However, that is looking at market prices in isolation (and I wonder if the results of the analysis would differ if the 5-year annualized return data you used represented real returns instead of nominal, but my hunch is that wouldn't make a huge difference). Looking at market prices as they relate to earnings (or another market valuation metric), on the other hand, tells a different story, as most recently discussed in this thread: http://forum.mrmoneymustache.com/investor-alley/using-market-valuation-measurements-to-affect-swr/
The linked thread does almost exactly the analysis you suggested, showing that retirement start years with high CAPEs have historically been much more failure-prone.
That poster used a slightly more aggressive portfolio and found slightly poorer outcomes than I did. So given that you've seen these analyses, do you still think that retiring at the top of overvalued markets is not any more risky than retiring in a more normally valued or undervalued market?
-
I don't think it's correct to say that elevated market values are not associated with higher failure rates.
No-one said that elevated market values are not associated with higher failure rates, just that a run-up in market prices was not associated with higher failure rates.
I think your last few posts are all consistent with everything that was said above - a run-up in prices (in nominal terms) in the years leading up to retirement was not indicative of a greater likelihood of portfolio failure. The answer might be similar for real returns (that analysis hasn't been done), which is not the same thing as valuation (using CAPE or similar measures).
-
That poster used a slightly more aggressive portfolio and found slightly poorer outcomes than I did. So given that you've seen these analyses, do you still think that retiring at the top of overvalued markets is not any more risky than retiring in a more normally valued or undervalued market?
Our posts are crossing, but no, I never said that I thought retiring in an overvalued market is not riskier. You're conflating "run-up in prices" with "overvaluation" -- they're not the same thing (as was explicitly recognized and discussed in the earlier discussion in this thread).
-
I don't think it's correct to say that elevated market values are not associated with higher failure rates.
No-one said that elevated market values are not associated with higher failure rates, just that a run-up in market prices was not associated with higher failure rates.
I think your last few posts are all consistent with everything that was said above - a run-up in prices (in nominal terms) in the years leading up to retirement was not indicative of a greater likelihood of portfolio failure. The answer might be similar for real returns (that analysis hasn't been done), which is not the same thing as valuation (using CAPE or similar measures).
OK, I see the distinction. In that case, I would not equate 5% CAGR (Beltim's analysis) to a "run up" in prices. Especially nominal terms. That's more a sideways market. In fact, a 5% nominal CAGR would make me more comfortable about retiring then, because for the last 5 years, I know the market has been underperforming the historical average (by about 50%) and reversion to the mean will increase my expected long term total returns. So for those years closer to the 5% rate you should see more success. For those years much higher than the 5% rate (15% say) you should see more failures. N gets pretty small for some of those.
-
In that case, I would not equate 5% CAGR (Beltim's analysis) to a "run up" in prices. Especially nominal terms. That's more a sideways market. In fact, a 5% nominal CAGR would make me more comfortable about retiring then, because for the last 5 years, I know the market has been underperforming the historical average (by about 50%) and reversion to the mean will increase my expected long term total returns. So for those years closer to the 5% rate you should see more success. For those years much higher than the 5% rate (15% say) you should see more failures. N gets pretty small for some of those.
Yeah, that's a good point. Beltim mentioned the possibility of doing the same analysis using a higher threshold for the annualized return in the years leading to retirement in post # 52, but the problem we keep coming back to is insufficient number of failure cases. That's why I had suggested doing the analysis using a higher (more failure-prone) WR, but no consensus was reached on whether that would be a useful exercise.
-
In that case, I would not equate 5% CAGR (Beltim's analysis) to a "run up" in prices. Especially nominal terms. That's more a sideways market. In fact, a 5% nominal CAGR would make me more comfortable about retiring then, because for the last 5 years, I know the market has been underperforming the historical average (by about 50%) and reversion to the mean will increase my expected long term total returns. So for those years closer to the 5% rate you should see more success. For those years much higher than the 5% rate (15% say) you should see more failures. N gets pretty small for some of those.
Yeah, that's a good point. Beltim mentioned the possibility of doing the same analysis using a higher threshold for the annualized return in the years leading to retirement in post # 52, but the problem we keep coming back to is insufficient number of failure cases. That's why I had suggested doing the analysis using a higher (more failure-prone) WR, but no consensus was reached on whether that would be a useful exercise.
I think doing it for 10% real (which is the important part right) and 12% real would have a small number of years, but would be more illustrative. You would probably have a bunch of failures for the small number of years, which would tell you that it is more risky than you were thinking. Or so I would project. 2010-2014 was 13.5% real. I'd be pretty nervous about retiring with a 4% SWR now.
-
My analysis was only to evaluate the success rate of a retirement strategy where you adjust the withdrawal rate - as was discussed in the first post. I used a period of five years somewhat arbitrarily, and the 5% rate was deliberately relatively low so that I would have more data points. The data don't indicate anything about the success rate after a huge market runup - because they weren't intended to. What the data do show is that you could adjust your withdrawal rate after the market goes up, and the new, higher withdrawal rate doesn't seem to make much of a difference in the chance of your money lasting a new 30 year period.
-
My analysis was only to evaluate the success rate of a retirement strategy where you adjust the withdrawal rate - as was discussed in the first post. I used a period of five years somewhat arbitrarily, and the 5% rate was deliberately relatively low so that I would have more data points. The data don't indicate anything about the success rate after a huge market runup - because they weren't intended to. What the data do show is that you could adjust your withdrawal rate after the market goes up, and the new, higher withdrawal rate doesn't seem to make much of a difference in the chance of your money lasting a new 30 year period.
That makes sense. BG just interpreted it differently. I think you would want to have longer than a 5 year "run up" period then, since the risk would probably only show up in any meaningful way with more time. 10 or 15 years even.
-
Well, to be fair, the analysis may have been prompted by the OP's question and intended to evaluate the success rate of the OP's proposed withdrawal strategy, but it was also done knowing that it would have broader application. As beltim said:
I'm trying to figure out the right stats for what the OP is actually suggesting - basically, how the probability of the 4% WR succeeding varies depending on the previous x years of market returns. I think it's an interesting question, particularly in that it has applications beyond the OP's question. The market returns in the few years before retiring certainly affect your probability of success, and it would be interesting to know how. Unfortunately, I suspect there isn't enough data in the historical record to get a statistically significant answer.
It's the problem of insufficient historical data that we keep bumping into. Beltim chose a relatively low threshold for returns and a relatively short lead-up period to get more data points; increasing either threshold will give us less data points to work with. That is why I suggested running the test with a failure-prone WR.
Forummm, with respect to your point about a 5% (or higher) nominal run-up being par for the course, using that threshold did cut out 38% of the 30-year periods from the original data set, so it wasn't that common in the historical record. But you're right that it isn't very meaningful, given that it's nominal rather than real. And again, the problem is that correcting for it exaggerates the problem of insufficient number of failure cases.
-
Redoing the analysis for a 5-year period with 10% annualized nominal returns, there are just 19 examples to choose from (this seemed low, but the annualized return over that period was 8.2%, so it makes sense that it would be less than half).
Now, here's the truly weird thing. All 19 of those periods succeeded at using a 4% withdrawal rate for a 30 year period. I think we're getting into statistical insignificance here, but it's interesting nonetheless.
-
Now, here's the truly weird thing.
(https://d2pi0bc9ewx28h.cloudfront.net/503t4so/Gr6m7qw4q/untitled-1.gif)
-
Now, here's the truly weird thing.
(https://d2pi0bc9ewx28h.cloudfront.net/503t4so/Gr6m7qw4q/untitled-1.gif)
If we hit that bullseye, the rest of the dominoes should fall like a house of cards. Checkmate.
-
Redoing the analysis for a 5-year period with 10% annualized nominal returns, there are just 19 examples to choose from (this seemed low, but the annualized return over that period was 8.2%, so it makes sense that it would be less than half).
Now, here's the truly weird thing. All 19 of those periods succeeded at using a 4% withdrawal rate for a 30 year period. I think we're getting into statistical insignificance here, but it's interesting nonetheless.
But that's not surprising, because we already know that the start years of all the failure cases (of which there are very few to begin with) are clustered around the late 1960's, when nominal returns were not high. The problem is not enough failure cases in the original data set. Wouldn't using a more failure-prone WR (and thereby increasing the number of failure cases in the full data set) be a better approach? Then we can do the analysis using meaningful real returns (and also longer lead-up periods) with statistically more significant (though probably still not actually statistically significant) results.
-
Redoing the analysis for a 5-year period with 10% annualized nominal returns, there are just 19 examples to choose from (this seemed low, but the annualized return over that period was 8.2%, so it makes sense that it would be less than half).
Now, here's the truly weird thing. All 19 of those periods succeeded at using a 4% withdrawal rate for a 30 year period. I think we're getting into statistical insignificance here, but it's interesting nonetheless.
But that's not surprising, because we already know that the start years of all the failure cases (of which there are very few to begin with) are clustered around the late 1960's, when nominal returns were not high. The problem is not enough failure cases in the original data set. Wouldn't using a more failure-prone WR (and thereby increasing the number of failure cases in the full data set) be a better approach? Then we can do the analysis using meaningful real returns (and also longer lead-up periods) with statistically more significant (though probably still not actually statistically significant) results.
To answer the question of "does a run up in prices negatively correlate to a safe withdrawal rate" maybe. This shows, though, the danger of using small sample sizes. 19 starting years seems like a reasonable number of data points, but the result suggested - that waiting to retire until a 5 year period of 10% annual returns is SAFER than if you don't - is of course absurd.
So if 19 years isn't enough to give a good answer, would 30? Doubtful. So I don't think this analysis will work even at a higher withdrawal rate.
A better analysis setup would be to look at the maximum safe withdrawal rate for each year, and compare that to whatever factor you wanted (CAPE, past 5 year return, past 10 year return, etc.). Anyone know where I could get that data?
-
Someone's done it in a different presentation that I would have but it's pretty cool nonetheless:
(https://www.kitces.com/wp-content/uploads/2014/07/Shiller-CAPE-vs-30-Year-SWR.png)
-
Someone's done it in a different presentation that I would have but it's pretty cool nonetheless
That is pretty cool. As discussed in the other active thread on this topic (linked to earlier in this thread), though, CAPE seems to have lost its predictive value in recent decades (whether temporarily or permanently is an open question).
Do you know what parameters were used for the SWR in that chart?
-
Someone's done it in a different presentation that I would have but it's pretty cool nonetheless
That is pretty cool. As discussed in the other active thread on this topic (linked to earlier in this thread), though, CAPE seems to have lost its predictive value in recent decades (whether temporarily or permanently is an open question).
Do you know what parameters were used for the SWR in that chart?
The chart I got from here: https://www.kitces.com/blog/shiller-cape-market-valuation-terrible-for-market-timing-but-valuable-for-long-term-retirement-planning/
which may have gotten it from here: https://www.kitces.com/wp-content/uploads/2014/11/Kitces-Report-May-2008.pdf
Both are good reads. But it looks like the portfolio they looked at was 60% equities/40% bonds.
Regarding the bolded part, I think you're misinterpreting the data, or I'm looking at the wrong thread. The other thread I'm looking at discusses how current measures of the CAPE are directly comparable to previous measures of the CAPE due to accounting changes. I didn't see anyone argue that CAPE wasn't a good long-term (10 years) predictor of equity returns.
-
Both are good reads.
Thanks - will take a look. I usually love Kitces' stuff.
Regarding the bolded part, I think you're misinterpreting the data, or I'm looking at the wrong thread. The other thread I'm looking at discusses how current measures of the CAPE are directly comparable to previous measures of the CAPE due to accounting changes. I didn't see anyone argue that CAPE wasn't a good long-term (10 years) predictor of equity returns.
This is the thread I was referring to, where skyrefuge pointed out that the past couple of decades are looking kinder to high CAPEs as far as impact on portfolio success:
http://forum.mrmoneymustache.com/investor-alley/using-market-valuation-measurements-to-affect-swr/
-
This is the thread I was referring to, where skyrefuge pointed out that the past couple of decades are looking kinder to high CAPEs as far as impact on portfolio success:
http://forum.mrmoneymustache.com/investor-alley/using-market-valuation-measurements-to-affect-swr/
Oh, I see. Thanks for the clarification. I thought you were saying CAPE was losing its power to predict returns, not safe withdrawal rate. Interestingly, in the second link I gave, Kitces makes the point that the correlation between CAPE and SWR is stronger than CAPE and 10-year forward returns. Interesting stuff!
-
Redoing the analysis for a 5-year period with 10% annualized nominal returns, there are just 19 examples to choose from (this seemed low, but the annualized return over that period was 8.2%, so it makes sense that it would be less than half).
Now, here's the truly weird thing. All 19 of those periods succeeded at using a 4% withdrawal rate for a 30 year period. I think we're getting into statistical insignificance here, but it's interesting nonetheless.
So let's say that 1901-1905 has a CAGR of 11% (made up number). Then your analysis would look at whether retiring in 1906 with a 4% withdrawal rate would fail?
-
Redoing the analysis for a 5-year period with 10% annualized nominal returns, there are just 19 examples to choose from (this seemed low, but the annualized return over that period was 8.2%, so it makes sense that it would be less than half).
Now, here's the truly weird thing. All 19 of those periods succeeded at using a 4% withdrawal rate for a 30 year period. I think we're getting into statistical insignificance here, but it's interesting nonetheless.
So let's say that 1901-1905 has a CAGR of 11% (made up number). Then your analysis would look at whether retiring in 1906 with a 4% withdrawal rate would fail?
Exactly right.
-
The chart I got from here: https://www.kitces.com/blog/shiller-cape-market-valuation-terrible-for-market-timing-but-valuable-for-long-term-retirement-planning/
which may have gotten it from here: https://www.kitces.com/wp-content/uploads/2014/11/Kitces-Report-May-2008.pdf
Both are good reads. But it looks like the portfolio they looked at was 60% equities/40% bonds.
Those were both excellent reads. I'd be curious to see how the chart differs for other allocations, including 100% equities.
I thought you were saying CAPE was losing its power to predict returns, not safe withdrawal rate. Interestingly, in the second link I gave, Kitces makes the point that the correlation between CAPE and SWR is stronger than CAPE and 10-year forward returns. Interesting stuff!
But I'm confused about your point here in drawing this distinction. CAPE has gotten worse at predicting long term stock prices over the past couple of decades, too -- using CAPE's relationship to its historical average as the yardstick, the stock market has been "overvalued" for the past twenty years. Yes, as the Kitces articles described, CAPE has historically been even more highly correlated with SWRs than with long term equity returns, but it's CAPE's correlation with equity returns that explains its correlation with SWRs (because SWRs are largely determined by equity returns).
-
Unless you plan on giving up retirement and picking up work again in 30 years, then 30 years isn't the proper time period for analysis.
If you plan on being retired until you die, then years until death is the appropriate metric to use, and of course it decreases by one each year, so Bayesian statistics are appropriate.
As you near death, the risk goes down somewhat because there are fewer years left for things to go wrong. On the other hand your withdrawal rate is going to be based on the value at a market peak so your future projections are going to have a greater tendency to the downside and less future growth than the typical year, which increases your risk of failure.
I calculated some numbers from the market peak in December 1999 until December 2014. Starting with 1,000,000 in the S&P 500 and withdrawing 40,000. I adjusted spending up each year by the CPI and reinvested all dividends. At the end you have $551,121 and are spending $57,158. At that level of spending vs. assets (>10%) there's a very good chance you'll run out of money if you still have 30+ years to go. With a 3% withdrawal rate you'd still have about $900,000 and spending $42,000.
-
Unless you plan on giving up retirement and picking up work again in 30 years, then 30 years isn't the proper time period for analysis.
I doubt there is anyone here who doesn't already recognize this. But, for the type of historical analysis we've been doing in this thread (and any other history-based, SWR-type research), it makes more sense to use 30-year periods than, say, 60-year periods, because there are a lot more of them in the historical record. No-one is (or, at least, no-one should be) concluding that any answer applicable to a 30-year retirement is equally applicable to a 60-year retirement. Instead, we're using the results of 30-year period analyses as useful data points from which to draw informed conclusions. (For example, if we conclude that a 4% WR is not advisable for a 30-year retirement, then we know it sure as hell isn't advisable for a 60-year retirement.)
If you plan on being retired until you die, then years until death is the appropriate metric to use, and of course it decreases by one each year, so Bayesian statistics are appropriate
But then there's the "garbage in, garbage out" problem. A Bayesian statistical analysis would require making assumptions about future returns, right?
-
I thought you were saying CAPE was losing its power to predict returns, not safe withdrawal rate. Interestingly, in the second link I gave, Kitces makes the point that the correlation between CAPE and SWR is stronger than CAPE and 10-year forward returns. Interesting stuff!
But I'm confused about your point here in drawing this distinction. CAPE has gotten worse at predicting long term stock prices over the past couple of decades, too -- using CAPE's relationship to its historical average as the yardstick, the stock market has been "overvalued" for the past twenty years. Yes, as the Kitces articles described, CAPE has historically been even more highly correlated with SWRs than with long term equity returns, but it's CAPE's correlation with equity returns that explains its correlation with SWRs (because SWRs are largely determined by equity returns).
Has it? CAPE has been off for long periods of time before. It's possible that it's gotten worse as a predictor, but I haven't seen that data.
-
Has it? CAPE has been off for long periods of time before. It's possible that it's gotten worse as a predictor, but I haven't seen that data.
I guess I haven't seen that data either. It is true that CAPE has historically had a high standard deviation from its mean, spending long periods of time well above or below the mean. I was using the fact that CAPE seems to have lost some of its power to predict SWRs to conclude that it must have also lost some of its power to predict returns (because how else could the former be explained?), but maybe that reasoning is flawed? As the Kitces articles pointed out, it's really CAPE's predictive power with respect to the first decade or so of retirement that explains its high predictive power with respect to SWRs (since it's the first decade or so of returns that are the most critical for 30-year portfolio success), so maybe a single anomaly in the early part of the past couple of decades could explain how CAPE has gotten worse than its historical average at predicting SWRs while still allowing it to be as good as (or even better than) it has historically been at predicting long term returns?
-
Has it? CAPE has been off for long periods of time before. It's possible that it's gotten worse as a predictor, but I haven't seen that data.
I guess I haven't seen that data either. It is true that CAPE has historically had a high standard deviation from its mean, spending long periods of time well above or below the mean. I was using the fact that CAPE seems to have lost some of its power to predict SWRs to conclude that it must have also lost some of its power to predict returns (because how else could the former be explained?), but maybe that reasoning is flawed?
But what's your basis for this?
-
But what's your basis for this?
That retirements commenced in the past 20 years clearly seem to be on track for success to a greater extent than 30-year retirements commenced at all the historical periods in the past with high CAPEs. (Of course, we won't know for sure until each of the rolling 30-year periods have fully elapsed, but that's why I've been saying "seems to".) But now that I'm articulating the logical steps I've been taking to draw these conclusions, I'm starting to see that there may be a flaw in the reasoning, but can't pinpoint it. Perhaps it is possible for CAPE to fail to accurately predict SWRs over the past couple of decades without its predictive performance dropping below the historical average (after all, CAPE hasn't had a perfect correlation with SWRs in the past)?
-
But what's your basis for this?
That retirements commenced in the past 20 years clearly seem to be on track for success to a greater extent than 30-year retirements commenced at all the historical periods in the past with high CAPEs.
But other than the two years reference in skyrefuge's post here http://forum.mrmoneymustache.com/investor-alley/using-market-valuation-measurements-to-affect-swr/msg597846/#msg597846 do you have other data?
-
But other than the two years reference in skyrefuge's post here http://forum.mrmoneymustache.com/investor-alley/using-market-valuation-measurements-to-affect-swr/msg597846/#msg597846 do you have other data?
I haven't run the numbers on any other years, but I did eyeball some charts showing CAPE and S&P 500 returns, respectively, over the past twenty years and they appeared consistent with the idea that high CAPEs weren't predictive of poor long-term forward returns (of course, the closer we get to the present, the less "long-term" data there is in the historical record).
-
But other than the two years reference in skyrefuge's post here http://forum.mrmoneymustache.com/investor-alley/using-market-valuation-measurements-to-affect-swr/msg597846/#msg597846 do you have other data?
I haven't run the numbers on any other years, but I did eyeball some charts showing CAPE and S&P 500 returns, respectively, over the past twenty years and they appeared consistent with the idea that high CAPEs weren't predictive of poor long-term forward returns (of course, the closer we get to the present, the less "long-term" data there is in the historical record).
I guess it just looks pretty good to me, and I haven't seen the data formatted in a way to show if it's getting better or worse. Maybe I can generate that. In the meantime, this is what I'm looking at:
(http://www.hussmanfunds.com/rsi/CAPEh.png)
-
I guess it just looks pretty good to me, and I haven't seen the data formatted in a way to show if it's getting better or worse . . . In the meantime, this is what I'm looking at
Yeah, from that chart, it's hard to tell whether or not CAPE's predictive power with respect to returns has been getting worse, but it certainly doesn't seem to have been getting glaringly worse.
Maybe I can generate that.
Would love to see it, if you're up to compiling the data!
-
The chart I got from here: https://www.kitces.com/blog/shiller-cape-market-valuation-terrible-for-market-timing-but-valuable-for-long-term-retirement-planning/
which may have gotten it from here: https://www.kitces.com/wp-content/uploads/2014/11/Kitces-Report-May-2008.pdf
Both are good reads. But it looks like the portfolio they looked at was 60% equities/40% bonds.
Those were both excellent reads. I'd be curious to see how the chart differs for other allocations, including 100% equities.
They were interesting. But I thought 100% equities was in the table in the 2008 article. That said, It looked like his graphs didn't match the percentiles exactly. It looked like there were several times where the SWR was below 4 or below 4.5 for 60/40 but he said it never was. Not sure why that is.
-
But I thought 100% equities was in the table in the 2008 article.
There's the table showing SWRs ranked by CAPE with varying equity exposure (Figure 8), but I'd like to see the chart with line graphs showing start-year CAPEs and SWRs (like Figure 6), but based on 100% equity exposure, so that you can see the correlation between them. I'm guessing it's even higher than the correlation where SWR is based on a 60/40 allocation.
That said, It looked like his graphs didn't match the percentiles exactly. It looked like there were several times where the SWR was below 4 or below 4.5 for 60/40 but he said it never was. Not sure why that is.
Not sure I follow you here.
-
I calculated some numbers from the market peak in December 1999 until December 2014. Starting with 1,000,000 in the S&P 500 and withdrawing 40,000. I adjusted spending up each year by the CPI and reinvested all dividends. At the end you have $551,121 and are spending $57,158. At that level of spending vs. assets (>10%) there's a very good chance you'll run out of money if you still have 30+ years to go. With a 3% withdrawal rate you'd still have about $900,000 and spending $42,000.
You may want to look up raddr's y2k retiree for the realtime analysis of this. :)
-
All of this is very interesting but I wonder how many people will be retiring if the market does a 2008 meltdown later this year.
I don't honestly think it will but I can't help but think the reason many folks have FIRED is because of the incredible market run up.. a 20% downturn may trigger a massive case on OMY syndrome..:)
I'll do my best not to go back to work though..;)
-
I guess it just looks pretty good to me, and I haven't seen the data formatted in a way to show if it's getting better or worse . . . In the meantime, this is what I'm looking at
Yeah, from that chart, it's hard to tell whether or not CAPE's predictive power with respect to returns has been getting worse, but it certainly doesn't seem to have been getting glaringly worse.
Maybe I can generate that.
Would love to see it, if you're up to compiling the data!
Here's an attempt. The recent data certainly has a different slope, but the same trend. I wonder if we got our hands on the Philosophical Economist's data set on recalculated earnings whether the recent data would correct that.
-
I guess it just looks pretty good to me, and I haven't seen the data formatted in a way to show if it's getting better or worse . . . In the meantime, this is what I'm looking at
Yeah, from that chart, it's hard to tell whether or not CAPE's predictive power with respect to returns has been getting worse, but it certainly doesn't seem to have been getting glaringly worse.
Maybe I can generate that.
Would love to see it, if you're up to compiling the data!
Here's an attempt. The recent data certainly has a different slope, but the same trend. I wonder if we got our hands on the Philosophical Economist's data set on recalculated earnings whether the recent data would correct that.
My eyeballed regression line looks a lot steeper for past 1985 data. Both lines look like a pretty decent predictor.
-
My eyeballed regression line looks a lot steeper for past 1985 data. Both lines look like a pretty decent predictor.
Agree completely. Also I flipped the axes for this one:
-
My eyeballed regression line looks a lot steeper for past 1985 data. Both lines look like a pretty decent predictor.
Agree completely. Also I flipped the axes for this one:
Thanks for putting this together. I like this. It's nominal though. Looks like we're expecting around 3-4% real for the next 10 years with 100% equities. Along the lines of what I've been guessing. There will probably be at least 2 significant drawdowns during that time period though.
-
Thanks, beltim - this is great! It reminds me of the Philosophical Economics article on the dangers of reading too much into historical curve-fits, though:
http://www.philosophicaleconomics.com/2013/12/valuation-and-returns-adventures-in-curve-fitting/
I wouldn't extrapolate from these regression lines to predict a narrow band (or even a broad band) of expected future returns.
-
Yes, there are too many factors involved to try to rely on anything predicting the future with any accuracy. I think there's a pretty strong case to be made that something changed with the economy or the metrics enough that CAPE is measuring something different than it used to.