Withdrawals again... Making the Best of the Worst?

[siamond]

PREAMBLE

Sounds that several of us have been torturing themselves pretty bad about more or less variable forms of Withdrawal Methods. I plan to early retire in a couple of years, so count me in the club of the tortured...

A few thoughts to start with:
- intuitively, I like(d) the idea of variable methods, but I'm not sure to like the outcome of most, due to a) too much variability in income/spend b) this goes pretty bad when you start from a really bad year (try 1965 - even when approximating a diversified portfolio), you might be in a drought for a loooong time.
- the classic 4% constant withdrawal method may be very flawed, but fact is it works really nicely when you start from such really bad year (i.e. a market top, followed by a dip)
- but the classic model just makes no sense whatsoever if you start from a good year (i.e. in a market dip), your portfolio goes to the stratosphere, and you get no benefit out of it

After torturing myself (I spare you the giant Excel sheet), I questioned my own goals, and established the following:
- I like the idea of a spending raise , I dislike the idea of a spending cut (heck, in my working life, it worked like that - yes, I know, I am fortunate)
- I want to retire, stay zen, relax... I don't want the vagaries of the market to negatively impact my day-to-day life - back to the previous point
- I want to retire with a given portfolio value, and aim (roughly) at keeping it unchanged (or a tad higher) in real dollars (various reasons, we never know how long life will extend; legacy to heirs, N+1th level of backup for unexpectedly-high LTC, etc)
- And although I have some doubts about the role of inflation once I'll be retired, I'd rather play safe and have my income/spending evolve with the inflation ride

[siamond]

THE CONCEPT

After thinking about various variable methods and related Excel-based torture, an idea suddenly struck me when driving to work...

The Classic fixed-withdrawal method (inflation adjusted) has this totally weird characteristic of giving you a very different result between starting one year or another (try 1973 vs 1975!). Which in itself tells me this is quite flawed if used as is. 1973 is a starting point where the protection against the 'Worst' is the most effective strategy, because the worst is indeed going to happen... 1975 is WAY different.

So why not reset the withadrawal value (in inflation adjusted dollars) every time the current portfolio value turns out higher than the (inflation adjusted) initial portfolio? Not the WR rate, but the absolute value of the annual withdrawal. Just compute the current portfolio value multiplied by your usual rate (4%, 3.5%, whatever you feel safe with). That is, if and only if this actually gives you a raise in real dollars (usually it does, but sometimes the inflation goes faster than the market).

In other words, you give yourself a raise every now and then, when you reach a point where the Classic fixed-withdrawal method gives you a better number if you were to retire again. You make the Best out of the Worst.

To refine the idea a tiny bit, we can define a threshold (say 5%). When the current portfolio exceeds the initial (or targeted) portfolio by such threshold, then reset the withdrawal value. If and only if this gives you a raise. Very simple to implement.

Ok, before you read the following, let's stop here. This idea doesn't exactly seem like rocket science. I suspect I re-invented the wheel. Did I? Is there a known strong defect of what I'm suggesting?

[siamond]

ASSESSMENT

Since I grew wary about great intuitive ideas about variable methods (I had a few... All very flawed...), I played around with my own big ugly Excel sheet, and then decided to also assemble a variation of the neat sheet longinvest created to assess his nifty VPW method. I called my own method "BOW" (Best Of Worst).

Since longinvest's Excel is much easier to interpret than mine, with neat comparisons with existing methods, I added my own method and compared... Here it is:
https://docs.google.com/file/d/0B0svRQG ... sp=sharing

(longinvest, if you read this, please check the change log, I believe many of my changes have a broader applicability)

Playing with multiple scenarios, backtesting in a systematic manner, more or less diverse portfolio, shorter or longer retirement, etc... For once, I couldn't break the idea. And it fulfills all the requirements I stated in my first post. And this is wicked simple. And you may safely stop to adjust at any time if you grow wary of Excel in your old years.

Here is an example, starting in 1978 with a basic portfolio (Bonds/Stock 40/60; US/Int'l 70/30), and a 3.5% initial WR. Blue line is the Classic constant-% method. Green line is the BOW method. Yes, this is a very favorable example. Less favorable examples simply get closer to the classic model, and at worse, are exactly equal. It can only get better... If it does!





And just for fun, a comparison between the VPW model (blue line) and the BOW model (green line). For sure, the VPW method allows to get more money out of the initial portfolio, but it also breaks many of my stated goals. And doesn't leave much at the end. In all fairness, as somebody pointed out, VPW is just NOT designed for my goals, simple as that. Still, it's eye-opening to compare spend/portfolio trajectories.





Overall, after running many simulations, I would observe that:
1. the resulting portfolio tends to grow a tad on the long run, but never a lot
2. the resulting portfolio may go through significant dips, well, that's the nature of the market, but this is somewhat compounded by the main risk of the method (see point 4).
2. the withdrawals are always equal or better than the classic method (quite obvious), but sometimes getting REALLY better (2x isn't uncommon)
3. although it feels as safe as the classic method, this is not entirely true. Because you tend to settle on the 'best of worst', this makes your selection of a safe withdrawal rate all the more critical.

About the last point, this tends to imply that you'd better start at 3.5% and just let the magic happen (raises!). Aside from this caveat, this seems to work remarkably well! I think?

[siamond]

REFINEMENTS
Ok, making a giant leap of faith that this is a good idea...

One small refinement is to make the formula dependent on a target ('second base') portfolio value, typically higher than the initial portfolio. After all, in most (not all!) situations, your portfolio will increase, notably if you don't increase the withdrawals (inflation adjusted) until then. So if you have some tolerance for risks (e.g. an early retiree with some backup options), you might start by $1.5M and aim at $2M. And then settle on the 'second base'.

Another small refinement is to slightly adjust your withdrawal rate every time you give yourself a raise. You could get more conservative as your income/spending gets better (that is... if it does!). More precisely, define a target WR rate (ah, the 'second base' analogy again), and adjust by 0.1 point whenever you get a 'raise'. Then you can start conservative (say 3.5%), and aim at a very conservative 3.2% or 3.0% over time. And if the world nearly comes to an end (Japan-like crisis or worse), and you never get a raise, well, you will probably do something beyond this algorithm at some point in time.

I implemented all of the above in my own Excel sheet, this is fairly trivial, and works well.

Now, back to VPW. Say you reach a point in life where you feel too flush with money (is that possible? ). Maybe you downsized your house and suddenly increased your portfolio value, or something like that. I would then suggest that you create (just by virtue of some Excel math) a split in your portfolio, and you manage the spending on the first part with the BOW algorithm to satisfy your regular needs in a safe manner. And you manage the rest with the nifty VPW method, essentially aiming at spending all of the 2nd chunk of the portfolio value, accepting market vagaries to influence how much you spend, donate, etc. And that's now perfectly fine, a sort of extension of my personal goals, actually, because this spend (donations to heirs or charities before death, special luxury car or vacation, etc) is by its very nature much more easy to postpone or trigger, while you're safe & steady on your primary source of income/spend.

[Taylor Larimore]

Bogleheads:

Below is a link to the updated (1926-1009) Trinity study showing actual success rates using various withdrawal percentages and with different stock/bond allocations:

Portfolio Success Rates: Where to Draw the Line by Philip L. Cooley, Ph.D.; Carl M. Hubbard, Ph.D.; and Daniel T. Walz, Ph.D.

Best wishes.
Taylor

[siamond]

Thank you, Taylor. You are a man of much less words than me! And undoubtedly much more wisdom...

I skimmed through the article (will read more carefully tomorrow), then the appendix, and surprisingly enough, didn't see a single reference to the basic idea I am suggesting. Something must be amiss here. Am I saying something downright silly? If yes, why?

[siamond]

I added the missing charts and comments on this post: http://www.bogleheads.org/forum/viewtop ... 3#p1771079

Sorry, I got a bit distracted last night.

[Clearly_Irrational]

By the way Gummy has a bunch of different articles on withdrawal methods and spreadsheets to go with them. Here's one example: http://www.gummy-stuff.org/sensible_withdrawals.htm

[siamond]

By the way Gummy has a bunch of different articles on withdrawal methods and spreadsheets to go with them. Here's one example: http://www.gummy-stuff.org/sensible_withdrawals.htm

Thanks. I read the article last night. The author is a funny guy. Yet another variation on the theme of variable withdrawals, with a floor based on the initial withdrawal amount. Which in my humble opinion is ok (-ish), but not exactly great, because of the initial withdrawal amount is chosen rather 'randomly' (depending on the state of the market when you retire), plus we have significant ups and downs of actual revenue/spend every year. Some people may like it, I don't think I would.

So I would say... fagedabawdit!

[siamond]

Well, can't say I'm eliciting a lot of feedback here (you guys must be so wary of discussing the N+1th withdrawal scheme... still... would love a tiny bit of feedback... please!).

REFINEMENTS (MORE!)

I kept thinking about it, and running simulations. And I came up with another refinement. The gist of what I suggested is to sync up on the year where your current portfolio is at a high point and "retire again", resetting the constant-% process (if this gives you a 'raise'). And what I suggested was to adjust over time (thinking to the future).

Well, maybe we can also think about the past when we retire the first time. If you take your portfolio value at that point, and compare it to the past few years (say 10) using inflation-adjusted math, and you notice that there were peaks higher than where you currently are, then just bootstrap the process by taking your safe-WR of choice (say 3.5%) and apply it to said peaks. Take a 2002 example, the market dropped quite a lot, but also peaked quite a lot a couple of years before. Then retire based on the peak, not the valley.

Now of course, if you do so, the more 'exuberant' the peak, the more you'd better be extra conservative on your 'safe' withdrawal rate. I wouldn't start at 4% in such a case. And by extension, if you retire at the end of 1999, I would suggest you should be extra conservative too.

ASSESSMENT (MORE)

I think it all shows that the primary challenge with my proposal is to select a 'truly safe' withdrawal rate. It's not a good idea to go too low (say 2%!), as otherwise, you'll make your portfolio grows quite a lot before your withdrawal amount increases enough to settle on a stable rhythm (without speaking of eating catfood in the mean time!). But you REALLY want to be conservative enough, as the fact of settling on 'the best of the worst' is actually making the probability of portfolio failure significantly higher.

So the WR % second base' idea (cf. my 1st round of refinement thoughts) does seem useful. Also, based on what I just wrote, when retiring, you may want to try to assess if you're at a peak (market high) or a valley (way under market high) or in-between - only a very rough assessment, that's enough, but still, ponder about it before picking a 'safe' WR %.

[Sbashore]

Well, can't say I'm eliciting a lot of feedback here (you guys must be so wary of discussing the N+1th withdrawal scheme... still... would love a tiny bit of feedback... please!).

REFINEMENTS (MORE!)

I kept thinking about it, and running simulations. And I came up with another refinement. The gist of what I suggested is to sync up on the year where your current portfolio is at a high point and "retire again", resetting the constant-% process (if this gives you a 'raise'). And what I suggested was to adjust over time (thinking to the future).

Well, maybe we can also think about the past when we retire the first time. If you take your portfolio value at that point, and compare it to the past few years (say 10) using inflation-adjusted math, and you notice that there were peaks higher than where you currently are, then just bootstrap the process by taking your safe-WR of choice (say 3.5%) and apply it to said peaks. Take a 2002 example, the market dropped quite a lot, but also peaked quite a lot a couple of years before. Then retire based on the peak, not the valley.

Now of course, if you do so, the more 'exuberant' the peak, the more you'd better be extra conservative on your 'safe' withdrawal rate. I wouldn't start at 4% in such a case. And by extension, if you retire at the end of 1999, I would suggest you should be extra conservative too.

ASSESSMENT (MORE)

I think it all shows that the primary challenge with my proposal is to select a 'truly safe' withdrawal rate. It's not a good idea to go too low (say 2%!), as otherwise, you'll make your portfolio grows quite a lot before your withdrawal amount increases enough to settle on a stable rhythm (without speaking of eating catfood in the mean time!). But you REALLY want to be conservative enough, as the fact of settling on 'the best of the worst' is actually making the probability of portfolio failure significantly higher.

So the WR % second base' idea (cf. my 1st round of refinement thoughts) does seem useful. Also, based on what I just wrote, when retiring, you may want to try to assess if you're at a peak (market high) or a valley (way under market high) or in-between - only a very rough assessment, that's enough, but still, ponder about it before picking a 'safe' WR %.

It seems to me you're over analyzing this. I've been retired since 2005. I take my withdrawals based on nominal values, depending on my portfolio growth to give me raises. I track my internal rate of return and spend a lesser percentage in lean years and more in better years. No matter what excruciatingly exact methodology you develop, you will adjust your behavior depending on circumstances, if you are at all typical.

[siamond]

It seems to me you're over analyzing this. I've been retired since 2005. I take my withdrawals based on nominal values, depending on my portfolio growth to give me raises. I track my internal rate of return and spend a lesser percentage in lean years and more in better years. No matter what excruciatingly exact methodology you develop, you will adjust your behavior depending on circumstances, if you are at all typical.

I totally realize that. Please note that although I'm of the 'wordy' type, my proposal is actually EXTREMELY straightforward to implement. As to the refinement ideas, they are entirely optional, and perhaps indeed a tad on the over-thinking side, just because it's fun to analyze...

Really, the essence of what I am suggesting is to 'retire again' when times are good. This is probably not far from what you just explained (WITHOUT the lean times though - not needed - key difference). But this formalizes the thinking, with (I believe) a solid theoretical basis behind it (all the Trinity-like studies). Then... yes, for sure, we human beings are adaptive, and real-life will be a tad different. Still... I am not retired yet... So I need a plan I believe in!

[nisiprius]

Forgive me, Siamond, I'm not going to look at the specifics in detail, I just sort of tune out on "safe withdrawal rates."

A specific critique of your proposal is that, in effect, the traditional methods count on building up a reserve during good periods that can be drawn down during the bad periods. If you spend away all the buildups during the good periods, you are eroding the safety margin. The studies, for whatever they're worth, assume a random starting point--not the starting point you get after spending down a lucky windfall.

What I can say is that there have been a number of withdrawal systems proposed that have flexible or variable withdrawals, e.g. Guyton's; see http://www.bogleheads.org/wiki/Safe_Withdrawal_Rates for some notes and links. But I think trying to refine withdrawal systems is basically misguided. I would point you to two observations; first, the original Trinity study, which stated

The word planning is emphasized because of the great uncertainties in the stock and bond markets. Mid-course corrections likely will be required, with the actual dollar amounts withdrawn adjusted downward or upward relative to the plan. The investor needs to keep in mind that selection of a withdrawal rate is not a matter of contract but rather a matter of planning.

The second is Taylor Larimore's Safe Withdrawal Rates: Complexity versus Simplicity:, which I am surprised he didn't quote himself.

We simply withdrew what we needed and kept an eye on our portfolio balance. Most years our balance went up and we spent the money on vacations, luxuries and charity. When our balance went down we tightened our belt and economized.

This is what most people do and it works.

As a rule of thumb 4% isn't bad, and asking "how often would it have failed in the past" is good as a thought experiment and a reality check.

Beyond, that, you can't make a safe silk purse out of a risky sow's ear. If you want safety, and high "efficiency" in the sense of having a high withdrawal rate and using up all your money yourself, without leaving any "on the table" (or "for the kids"), there's an easy way to do it, which is to pay an insurance company to do the job of pooling risk and averaging out life expectancies--that is to say, buying a fixed annuity. If you want to go it alone, then you have to strike a balance between low withdrawals, a wide margin of safety, and the likelihood of leaving unspent money; and high withdrawals and high risk. There's no magic formula that's going to give you high withdrawals and low risk.

To put it bluntly, it seems likely that something like 3-4% is safe, and that beyond that it's just people thrashing around because they don't like the 3-4% number and want to convince themselves they can spend more. Well, I think either one should setting your spending low enough to have a fat margin of safety under any withdrawal system, or one should go buy a fixed annuity.

I just doubt that there's much benefit to setting up a numeric system for doing in Excel what Taylor Larimore does by intuition.

[Sbashore]

Nisiprius put it better than I ever could. I wondered if I should have defined "lean" when I wrote it. When my internal rate of return YTD goes below zero either because of spending or market conditions I spend a little less. As far as eliminating "lean" times with a withdrawal strategy, that's not going to happen, unless you spread the risk around, like Nisiprius said with an annuity. I know what you mean by needing a strategy to believe in. I need that too. I think we're trying get you to believe in a realistic one.

[siamond]

Let me quote what cut-throat said on another thread:

PS. as a side note, your honest feedback on my best-of-worst concept would be very welcome.

Actually, I proposed a similar strategy once, and it was listed under 'Bob's Financial Website' (Which is no longer Operational). The strategy is sound and does increase your spending if your portfolio increases. However it does not reduce your spending if your portfolio decreases, which is the Real problem of Retirees.

This is the Problem I have with all Constant Dollar, Inflation Adjusted withdrawal methods. In my opinion, we don't behave this way as humans. When a recession hits, human nature causes us to 'Pull in our Horns'. We generally Pull our horns in more than needed. I really don't think that someone that is retired would continue to spend his inflation adjusted dollars into the teeth of a severe recession. And if you did, you greatly increase your chances that your portfolio will fail.

So, why not have a methodical plan of withdrawal, just as you have a methodical plan of Investment and Asset Allocation? It takes a lot of the 'emotion' out of Investment and Spending. This is why I much prefer the VPW method. It lets you increase spending as you age and if your portfolio increases, however it will put the brakes on spending when and if your portfolio shrinks too fast. This also pretty much guarantees that you won't run out of money. Yes you have to be flexible with your spending, but I view that as a requirement of quitting working.

Thanks for the feedback, good to hear that you came up with a similar idea. There is a statement well hidden in the updated Trinity study that Taylor shared, which also seems to hint at the same base concept, by the way.

I am definitely with you about the need for a methodical plan of withdrawal (even if we have different goals for what it should do). Even if I do hear the wisdom expressed by Taylor and Nisiprius and Sbashore that real life doesn't unfold according to plan, and that adaptability is essential, I'd rather not completely wing it... And not bet the farm either on a starting point (the value of the portfolio the year you retire, quite random in truth, depending on how low/high the market is). While keeping a plan simple enough to implement.

The more I play around with the BOW model (which some folks started to dub the "Retire Again & Again" model), the more I believe that it works remarkably well for 95% of the scenarios if you choose sensible parameters to start (and share my goals). Now the few remaining scenarios are 1916, 1929, 1937, 1965, those few really tricky years. For which some form of spending cut is indeed required & desirable (to avoid starting with an extremely low WR rate in all cases). I have yet to find the right formula for it without creating annoying side-effects on the 95% 'good years' (so... maybe this is indeed where we should just let the human brain react to exceptional situations)...

[longinvest]

Hi siamond,

Thanks for the nice comments on the VPW spreadsheet.

ASSESSMENT (MORE)

I think it all shows that the primary challenge with my proposal is to select a 'truly safe' withdrawal rate. It's not a good idea to go too low (say 2%!), as otherwise, you'll make your portfolio grows quite a lot before your withdrawal amount increases enough to settle on a stable rhythm (without speaking of eating catfood in the mean time!). But you REALLY want to be conservative enough, as the fact of settling on 'the best of the worst' is actually making the probability of portfolio failure significantly higher.

So the WR % second base' idea (cf. my 1st round of refinement thoughts) does seem useful. Also, based on what I just wrote, when retiring, you may want to try to assess if you're at a peak (market high) or a valley (way under market high) or in-between - only a very rough assessment, that's enough, but still, ponder about it before picking a 'safe' WR %.

Cut-Throat and you are giving me a lot of feedback on VPW (privately or through proposed improvements and BestOfWorst). Thank you.

I see that the main objective of Best of Worst is to avoid the constant revenue variations (mostly the negative ones), from year to year, when using VPW. Cut-Throat also makes suggestions seeking to avoid such drops. I don't think that an increase (over inflation) is a real objective of a retiree; it would mean that the retiree plans to spend more and travel later rather than sooner in his retirement. On the other hand, getting inflation increases seems important.

I'll propose an improvement to VPW (in its thread) that is inspired from this.

As for Best of Worst, I think that it still has a fundamental flaw. Here's what happens if you start with a 4% withdrawal, 40% Bonds / 60% Stocks portfolio in 1966. THe portfolio balance ends-up negative before the target age:


The problem of picking a "super safe" initial withdrawal rate is that you are targeting the worst possible unknown future scenario. Somehow, what you are proposing is simply Constant-Dollar Withdrawal with an upside if things turn out better than the worst-ever expected scenario. As such, BoW is still completely exposed to the traditional bad sequence of return problem, as shown in 1966 above.

Constant-Dollar Withdrawal schemes are like planning for the worst possible retirement. To make an analogy, it's like deciding not to ever drive a car and travel, because you might die in an accident doing so. Of course, you minimize the risks of dying in an accident, but you possibly reduce the enjoyment you get from life, too.

VPW started from the idea that I wanted to make sure I will be able to enjoy the money I am saving, earlier rather than later in my retirement, yet have some safeguards that protect me from depleting my portfolio early. To make an analogy, I prefer to take some chances: travel a little, drive my car, and enjoy life (within my own risk boundaries), but also adapt according to circumstances. I won't take the plane and travel to a country experiencing war. If a war starts during my vacation, I'll cut my vacations and come back home immediately. That's what VPW seeks to do: reduce withdrawals when the portfolio drops.

Maybe the biggest flaw of VPW is that withdrawals are constantly changing (in constant-dollars). Adding some stability seems desirable. I'll post an improvement suggestion in the VPW thread: http://www.bogleheads.org/forum/viewtop ... 0&t=120430

[Red Rover]

http://www.marketwatch.com/story/the-ul ... 2013-08-14

[siamond]

As for Best of Worst, I think that it still has a fundamental flaw. Here's what happens if you start with a 4% withdrawal, 40% Bonds / 60% Stocks portfolio in 1966. THe portfolio balance ends-up negative before the target age:


The problem of picking a "super safe" initial withdrawal rate is that you are targeting the worst possible unknown future scenario. Somehow, what you are proposing is simply Constant-Dollar Withdrawal with an upside if things turn out better than the worst-ever expected scenario. As such, BoW is still completely exposed to the traditional bad sequence of return problem, as shown in 1966 above.

Hi longinvest,

Thanks a lot for the thoughtful feedback. First off, now that I have more thinking/modeling under my belt, I now understand much better that BoW and VPW have very different goals, and it all depends on your objectives (hence my very first post). And there are situations where combining both ('virtually' splitting your portfolio) could actually make a lot of sense. I also finally understood that it does not make much sense to inject weird compromises in one model or the other to try to accommodate a goal completely at odds with the model itself (e.g. my silly attempt at some form of capital protection in your model!). Yes, I know, cut-throat and yourself told me right away that I was misguided, but well, I needed to understand by myself...

Next, about your comment for 1965/66... Yeah, I've been pondering about it for a while now (and couple of other especially bad starting years). I don't believe this breaks the model at all, it just shows the limit of doing something entirely mathematical. If one goes through a 1965/66-like situation, a sensible human being would end up doing something beyond the model to adjust to exceptional situations... That being said, if you start at 3.8% with a properly diversified AA, you do survive 1965/66. And starting at 3.8% actually works fine for more regular years and does much better that the classic 'Bengen' CDW model at 4%, as the 'retire again -and get a raise!' mechanism usually quickly kicks in.

I tried to do an XBOW model that would be more adaptive for extreme crisis, struggled a lot (so hard to make an algorithm react in time to a severe crisis and not overreact to a perfectly recoverable situation - at least without compromising too much on my top-level goals), and ended up deciding this was non-sense...

Now something I had not quite factored in in all of this until a few days ago is Social Security (or a Pension or equivalent). Obviously, this is a sort of ultimate protection against portfolio dramatic failure, guaranteeing a living, albeit very constrained. Less obviously, if your portfolio is at least in a half-decent shape when SS kicks in, you might choose to reinvest the corresponding new income (while optionally adding a few bucks to your withdrawal, e.g. IWR times SS-amount, hence a small raise every year). Such re-investment provides a nice buffer against crazy market conditions down the road, I just finished the corresponding modelling in my own Excel sheet. For somebody like me planning for an early retirement, this works especially well.

Overall, after killing my eyes and my Excel skills for a while... I am EXTREMELY happy with the BOW (aka Retire Again & Again) model. This is very simple, and fulfills my goals perfectly. And if somehow I do get 'too wealthy' (it happens, e.g. start in 1975), then I'll probably use VPW to spend/donate the extras. It's the BOBW - Best Of Both Worlds!

[longinvest]

Hi siamond,

I think that I have not clearly explained my perception in my previous message.

First, the basics:

At first glance, BoW's stated objective of an increasing payment could seem attractive. Yes, we're used to get annual salary increases, so it would feel nice to get the same during retirement.

But, I think that this is an illusion. Most people think of salary increase in nominal terms, not real terms. I would propose that a retiree would be happy to get nominal payment increases that match inflation. This objective gives us the well known Constant-Dollar Withdrawal (CDW) method.

CDW suffers from two important flaws, in my opinion. First, it can completely fail and leave you with no payments at all within a period of 30 years with a low 4% initial withdrawal increased yearly for inflation. Second, it will, in most cases, let you die with a huge pile of unspent money (because 4% tries to capture the worst possible future). This is nice if you have major bequest motives, but otherwise, it seems awfully wasteful.

The deeper stuff:

Of course, you present BoW as an improved CDW: one where the second flaw is mostly eliminated. That's probably a valid view in theory, but I believe that it is flawed in practice.

First, BoW does not address CDW's first (and most important) flaw: if you select too high of an initial withdrawal rate, BoW will fail. Now, there are many papers by brilliant researchers on how to try selecting a Safe Withdrawal Rate (SWR), which is simply an initial withdrawal rate that won't fail within 30 years. The problem is that there is no way to know with complete certainty what the SWR is today for any specific asset allocation (of Stocks and Bonds).

Now, assume that you were able to find a working crystal ball for the next 30 years: what would you select as a starting withdrawal rate? Would you select:

1. the maximun SWR, or
2. a lower starting withdrawal rate?

I think that most people would rationally select the first amount: the maximum SWR.

If they do and decide to use BoF, how will BoF behave? If BoF increases withdrawals, the plan will fail where CDW did not fail. If BoF does not increase withdrawals, then BoF behaves exactly as CDW. So, under this view, BoF is actually worse (or, at least, no better) than CDW.

But that's not the only flaw I see. If I put the theory aside and look at practice, one could say: don't worry, with BoF I can confidently start with a lower withdrawal rate; it will increase if the markets are good. So, I'll have less risk than with CDW.

I would dispute this view. I think that for most normal retirees (people retiring in their 60s), delaying consumption is of very little utility. Living in near misery during one's 60s in order to get high payment raises in their 80s or 90s is not rational. Studies show that people in their 80s and 90s have diminished mobility and lower needs. Getting higher payments at that point won't allow you to build the memories you would have accumulated travelling with your grand-children in your 60s. It is just too late to get the higher payments.

To sum it up, BoW does not protect from CDW's major failure flaw. Its supposed protection against too-low payments only show up too late, once the utility of increased payments has mostly disappeared. But, at first sight, it did look nice in theory (increasing payments).

[Peter Foley]

The following is an approach that you might consider. I copied it from a link on the Wiki.

Floor and Ceiling - The withdrawal amount is calculated as a percentage of the portfolio’s total value. The withdrawal amount is allowed to rise or fall with the performance of the portfolio. However, the withdrawal will neither fall below a “floor” nor rise above a “ceiling”. Both the “floor” and “ceiling” are adjusted upwards for inflation.

As an example, the initial floor can be set at 3.6% and rise with inflation. And the initial ceiling can be set at 5% and rise with inflation. The withdrawal amount can be set initially at a fixed percentage like 4%. On a $1,000,000 portfolio, the withdrawal will not fall below an inflation-adjusted amount of $36,000. Nor will the withdrawal rise above an inflation-adjusted amount of $50,000.

Another option is to use the 5 year reset approach:

5 Year Plan - In this method, the retirement is broken into five-year periods. Independently separate withdrawal rates can be assigned to each five-year period. In the first year of each period, the withdrawal amount is calculated by multiplying the total amount remaining in the portfolio by the withdrawal rate assigned to that five-year period. Then in years two through five of each five-year period, the withdrawal amount is increased by the previous year’s inflation rate.
However, in the event that the returns during first five years are poor, whereby the total portfolio declines by 10% or more, the withdrawal amount gets reset. The new withdrawal amount is obtained by multiplying the value of the now-reduced portfolio by 4%. The withdrawal amount is then increased by the previous year’s inflation rate for the remainder of the first five-year period. There is no “4% reset” in subsequent five-year periods.

[siamond]

*** EDITED: following request to post more readable tables with simpler 75/25 portfolio returns and no 'dampening' ***

Hi siamond,

I think that I have not clearly explained my perception in my previous message.

[...]

To sum it up, BoW does not protect from CDW's major failure flaw. Its supposed protection against too-low payments only show up too late, once the utility of increased payments has mostly disappeared. But, at first sight, it did look nice in theory (increasing payments).

Hey longinvest, many thanks for the detailed feedback. I really appreciate it. Now I understand the concerns you have. I actually had similar concerns before I started to look harder to it!

Ok, let's go through your various points, and to do so, let's put the theory in practice with a bunch of backtesting.
- I used a simple portfolio, with 25% US bonds, 75% US stocks. Using Simba returns, and before 1971, using Shiller data.
- I tested 45 cycles from 1925 to 1969, assuming an early retirement at 55 lasting till I'm 100 (hey, you never know!). Such long retirement periods are real hard on CDW, by the way, quite a few 30 years periods work ok, but fail the 45 years test. Average inflation was around 3.0% in such time period.
- I used a portfolio of $1M, with an initial WR of 3.2% for BoW, and 4% for CDW.

I defined portfolio failure as ending at less than $250k. I had 0 failures with BoW and 9 failures with CDW (aka 'Classic'). If I had started at 3.4% or above, failures do start to happen.

Now let's look at various metrics:



First thing to observe is that of course you get much more out of your portfolio (while protecting its basis for your heirs) than CDW, and the portfolio balance does stay in check. Yes, I agree that getting raises in real dollars isn't necessarily a goal for retirees, but getting the opportunity to settle on higher returns when it is 'safe' to do so (aka 'retire again') does seem pretty good to me. Notably when it comes with a guarantee to not drop your spend again in the future - barring an unprecedented crisis.

Second thing to observe is that solid 'raises' usually do come in the first 10 years. Not always, but often enough, as the average shows. I chose to compare with CDW at 4% to show that BOW usually catches up reasonably quickly (and much more safely than CDW at 4%, cf. those 9 failures!). I agree with you that early retirees want money to spend, but it seems that BOW delivers. Of course, not if you retire in 1936 or 1965...

So yes, the best way to use BoW is indeed to start with a conservative WR. You may have a few lean years to start with (same would be true with CDW at 3.2%), but then you have a real solid chance to start seeing the money trickling in. I would also observe that early retirees are quite likely to get some extra income from side activities. I know I plan to! Still, you are certainly correct that the best spend is tilted towards the latter years.

I do agree with you that the business of choosing a 'safe' WR isn't perfect (this is where VPW is so beautiful indeed). Hopefully, you'd agree I'd been conservative in my testing (cf. including the 1929 crisis, cf. the very long retirement period), but yes, the future might be even worse. I did suggest a mitigation, which is to lower your WR as you get 'raises', to settle around 3.0% or something arch-conservative like that. Not perfect, but even safer. Finally, I'd come back to the fact that algorithms aren't perfect, and human beings adjust themselves to exceptional situations. BoW is NOT the ultimate protection against market craziness. But it doesn't seem half bad...

To finish, just for kicks, here is the same table with CVPW, with an initial WR of 4%. You will note that the average first 10 years beat BoW, but not with such a large difference... While leaving no money whatsoever at the end. Yup, 45 failures out of 45 with my $250k criterion (I'm joking, this is by design, of course).



What do you think?

[longinvest]

To finish, just for kicks, here is the same table with CVPW, with an initial aggressive WR of 4%. You will note that the average first 10 years barely beat BoW... While attempting (and succeeding) to leave no money whatsoever at the end. Yup, 45 failures out of 45 with my $250k criterion (I'm kidding, this is by design, of course).

CVPW is not anything official; it is a custom version of VPW added for you and Cut-Throat! Could you compare to the real VPW?

[siamond]

To finish, just for kicks, here is the same table with CVPW, with an initial aggressive WR of 4%. You will note that the average first 10 years barely beat BoW... While attempting (and succeeding) to leave no money whatsoever at the end. Yup, 45 failures out of 45 with my $250k criterion (I'm kidding, this is by design, of course).

CVPW is not anything official; it is a custom version of VPW added for you and Cut-Throat! Could you compare to the real VPW?

Using CVPW was interesting here, to start from a withdrawal amount equal to CDW. I gave it a try with the 'real' VPW, and... hm... interesting... I suspect you knew what would happen!



Let's keep in mind that averages are quite deceiving here, as it misses the variability of returns & spend. Please check the std-deviation numbers I included.

Anyhoo, my point was really not to compare BoW and VPW, they really fulfill very different goals, I believe.

My point was more that, although I agree with most of the comments you made about BoW, the practical consequences seem very reasonable. That is, if one would trust that backtesting is indicative enough of what might happen in the future, which of course, isn't guaranteed at all.

[siamond]

By the way, if some of you want to play with BoW, the cFIREsim author was kind enough to implement the "Retire Again & Again" model (fancier name for BoW!) in his backtesting tool.

(cFIREsim is fairly similar to Firecalc, with a friendlier display of the end results, and with an active programmer behind it, ready to add more features as needs be)

Check it out here: http://www.cfiresim.com/input.html. Just don't forget to use a conservative initial spend to start with...

[cjking]

Re-edited to say: I have now scanned the rest of the thread, so I see that my objections below are understood.

My original response, based only on a quick scan of the OP:-

If I've understood the OP method correctly, it sounds very dangerous.

As I understand it, assuming the withdrawal rate is 4%, you take that (without inflation adjustments) until you hit a year where your inflation-adjusted balance is higher than when income was last set, then you start taking 4% of the new balance.

I will ignore that the 4% isn't inflation-adjusted, as although that lowers risk somewhat I don't think it makes any qualitative difference to the arguments.

Suppose you know 4% has worked from 95 out of 100 historical starting dates. (The exact proportion, as long as it's non-zero, doesn't matter.) The point about different starting dates is that the stock-market is more expensive on some dates than others, the starting dates that tend to lead to failure are the ones where the stock-market is expensive/near a peak.

With this strategy, you effectively re-retire whenever the stock-market goes up. So you effectively seek out a date from which failure is guaranteed, then stick with that.

In addition to not taking inflation-adjustments, the period might be slightly shorter when you re-retire. That will also reduce risk somewhat. I still think the strategy sounds dangerous though.

(I read the OP very quickly so it's possible I've misunderstood the proposal though.)

[cjking]

I've briefly tried out the spreadsheet. Couldn't get it to work for long ago dates, but confirmed what I suspected, that you need to use a starting rate low enough to be a reasonable withdrawal rate for the worst year you encounter.

For year 2000 in USA, my estimate of an indefinitely sustainable withdrawal rate (83%/PE10) was in the region of 2% and for Japan in 1989 it would be in the region of 1%. We don't want indefinitely sustainable, so we can go a little higher. But a starting rate of 3.5% in 2000 gives a rate in the region of 8% by 2010-2012, that is much to high. I reckon 2.1% - 2.2% is now the optimal value indicated by US history. Of course, had you set your rate before 2000, you could have justified something much higher. But if you take Japan into account, maybe you need to go lower.

Edit: just to clarify, I looked at 100% equities.

[siamond]

cjking, yes, using 4% as the initial WR with this method would indeed increase risks in an unacceptable manner. The intent is indeed to use a lower WR to start with, and then wait for safe 'raises' to occur. If this WR is conservative enough to allow to go through really bad crisis, then you're good to go, and the whole thing will auto-adjust whether things go bad or go well.

As shown by the backtesting I just posted, with a very poorly diversified portfolio (US stock/bonds), this requires to go down to 3.2% to be truly safe (with a better portfolio with international stock/etc, you could do significantly better, but it's hard to backtest before 1971). So this may not seem like much money to withdraw, but remember... those raises are (usually) coming... Unless you retired in 1965 or one of those truly nasty years, and then... Good thing you didn't retire at 4%, whatever the method.

If you check my long ramblings, you'll also notice that I suggested that, as you get 'raises', you get wealthier (and older!), and this is probably a good opportunity to get more conservative and 'glide' towards an even more conservative WR %. I have that baked in my latest Excel sheet, works well.

Now as your point of retiring at 2% or so to be TRULY safe... You're basically saying that the future might be MUCH MORE grim. Yup, maybe. Don't know. Luckily, the SS money should act as one more level of cushion when reaching 70 (allowing an even lower WR rate on your main capital if you wish to do so), so I am not getting overly paranoid, but... who knows.

In any case, assuming one is reasonably wise in picking the initial WR (yeah, BIG IF), the whole thing is actually much safer than the regular constant-dollar withdrawal (CDW) concept, and the smooth auto-adaptive behavior seems pretty attractive to me.

[longinvest]

Using CVPW was interesting here, to start from a withdrawal amount equal to CDW. I gave it a try with the 'real' VPW, and... hm... interesting... I suspect you knew what would happen!

I just knew that VPW requires a pretty conservative portfolio and a long retirement to start with as low a rate as 4%. You had an aggressive 75% stocks / 25% bonds portfolio, so I knew the shape of withdrawals was probably quite different.

Anyhoo, my point was really not to compare BoW and VPW, they really fulfill very different goals, I believe.

My point was more that, although I agree with most of the comments you made about BoW, the practical consequences seem very reasonable. That is, if one would trust that backtesting is indicative enough of what might happen in the future, which of course, isn't guaranteed at all.

BoW's main idea is attractive. It might suit some people like you, as long as you stay careful during retirement and monitor what happens in the market. Failure, like in 1966, can be disastrous in real life, unless you have combined BoW with a safe revenue floor (like Social Security, pensions, and SPIAs). But, even then, the psychological impact might not be good.

[cjking]

Now as your point of retiring at 2% or so to be TRULY safe... You're basically saying that the future might be MUCH MORE grim.

Depends what you mean by the "future", as far as I'm concerned year 2000 is the past, and possibly the worst starting date, but hasn't been factored in yet because the full 30 years isn't over. If you want to factor it in (in the same way you presumably used older data to get 3.2%) then maybe see what the highest safe rate is for 20-year retirements. If starting with 3.2% in 2000 leads to a 2010 withdrawal rate that's higher than the 20-year safe rate, then you could experiment to see what year-2000 withdrawal rate gives you the 20-year safe rate in 2010.

It's perfectly possible that experiment will show 3.2% is still OK, but I think it's worth a look.

[manwithnoname]

All of the modeling of future income to some mathematical formula means bupkis in the real world of living on sufficient income in retirement since the projections do not take into account real risks. I advise investors to have 1.5-2.0 X the amount of assets they will need in retirement over and above SS and pension plans. If you will need 40k in income in addition to SS and pensions then you should have an investment portfolio of $1.5-2.0 M generating a 4% return to cover inflation, longevity risk and investment risk. Any income not used for living expenses in a year should be saved and invested for future years when expenses will exceed income.

[cjking]

My suggestion for a tweak to this strategy is to have a context-specific start rate consisting of the initial yield on the bond allocation and a yield in the region of 1/PE10 on the equity allocation. This should give you a higher starting yield for those start dates when that's safe, overcoming the objection that the start rate is often too low. (100%/PE10 may not be exactly optimum, but choose whatever percentage p/PE10 back-tests best.)

Edit: the above would only make sense for setting an initial withdrawal rate, obviously you couldn't reuse the custom rate in different circumstances.

[Clearly_Irrational]

My suggestion for a tweak to this strategy is to have a context-specific start rate consisting of the initial yield on the bond allocation and a yield in the region of 1/PE10 on the equity allocation. This should give you a higher starting yield for those start dates when that's safe, overcoming the objection that the start rate is often too low. (100%/PE10 may not be exactly optimum, but choose whatever percentage p/PE10 back-tests best.)

That's an interesting idea. I wonder if that would be a good method for picking a successful SWR.