Using a 5y moving average to smooth withdrawals in VPW?

Discuss all general (i.e. non-personal) investing questions and issues, investing news, and theory.
Post Reply
mortal
Posts: 439
Joined: Mon Sep 15, 2008 11:15 pm

Using a 5y moving average to smooth withdrawals in VPW?

Post by mortal » Sun Jul 02, 2017 4:36 pm

Howdy folks,

Lately I've looking at various variable withdrawal strategies. The series I link to seems to have a pretty thorough analysis, but the cape strategy they favor seems to be pretty complex. I also don't know how to calculate or where to find cape for a diversified portfolio consisting of domestic (including small / mid cap value) as well as international. It's just too complicated in my opinion to be practical.

So, what's the alternative? I'd like to know what effect using a 5 year moving average would have in calculating a variable withdrawal rate? My intuition is that this would smooth out withdrawals, stretching out fluctuations over a longer time period and giving you more time to adjust to new spending levels. I've heard that prior attempts to smooth vpw failed, and it's very much a 'pay me now or pay me later' type of situation. I understand. I'm entirely ok withdrawing lower amounts for a longer time period provided the individual yearly adjustments aren't as severe or volatile.

My problem is that I don't know how to run the calculations or model it in a spread sheet. Is my intuition right here? What do you guys think?

GAAP
Posts: 137
Joined: Fri Apr 08, 2016 12:41 pm

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by GAAP » Mon Jul 03, 2017 12:11 pm

There's been a fair amount of discussion around smoothing VPW in this forum. The creator of the method doesn't like the idea at all -- smoothing is essentially incompatible with the expressed goal of spending down the entire portfolio in a specific time.

If, after more research, you still want to go down that path, then I suggest you play with the parameters in the underlying PMT function calculation -- smoothing the number of periods is probably the easiest way. Trying to smooth based upon portfolio performance sounds remarkably uncertain to me -- especially in a severe downturn.

mortal
Posts: 439
Joined: Mon Sep 15, 2008 11:15 pm

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by mortal » Mon Jul 03, 2017 1:07 pm

The creator of the method doesn't like the idea at all -- smoothing is essentially incompatible with the expressed goal of spending down the entire portfolio in a specific time.
Perhaps. I'm not heart set on a given outcome besides not running out of money. If I'm left with a quarter of my starting balance at the end of the planning period, that's ok (though I understand it could be harder to model). I stress again, I do not mind making adjustments. I think they're necessary for the health of the portfolio, but I think most people would a large amount of volatility. Some folks would say that's evidence they should have less in equities. I disagree with this, especially considering I'll likely retire early.

I've looked into coding my own office function. Sadly those functions seem to be written in some dialect of visual basic (yech). Still, it's better than the in cell convolutions I'd have to do otherwise.

John Z
Posts: 266
Joined: Sun Nov 14, 2010 5:42 pm

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by John Z » Mon Jul 03, 2017 1:16 pm

As I was approaching retirement I was researching withdrawal methods and VG Managed Payout Funds hit the streets. I believe they based the withdrawal amount over the last 3 years of performance so I was leaning in that direction: Starting with 4% and then varying the percentage each year based on the last 3 years' returns. But after a couple of years I simplified it to withdrawing, on average, 1% per quarter based on value at quarter end. I've found that withdrawals happen in clumps (new roof, new AC, new vehicle) so it's not a smooth 1% per quarter but I keep a running total of what the withdrawal amount should be ongoing so I know if I'm above or below the 1% average per quarter.

I did discover that the withdrawal amount is only part of the retirement equation. WHERE, WHEN and WHAT FUNDS to take the money from makes a difference which is brilliantly discussed in McClung's book Living Off Your Money where the first 3 chapters are a free download. There are 11 Chapters that cover not only best withdrawal methods but portfolio make up and just about all aspects of financial retirement.

http://livingoffyourmoney.com/

GAAP
Posts: 137
Joined: Fri Apr 08, 2016 12:41 pm

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by GAAP » Mon Jul 03, 2017 1:40 pm

FWIW, my plan is to use a typical spreadsheet function that works in Excel or LibreOffice: PMT(Rate, NPER, PV, FV, Type).

For Rate I use the expected real return of my portfolio, based upon a 50-year historical period and modified by my asset allocation.

I set NPER to 1, Present Value to my portfolio value, Future Value to a negative portfolio value (multiply by negative 1), and Type to 1. This effectively says I want to withdraw the amount at the start of the year that leaves we with no net change in portfolio value at the end of the year after net returns (assuming I guessed right on real returns, of course). This sets my floor -- the minimum value I would withdraw. I multiply that value by 25% to determine the ceiling. I do not adjust the withdrawal amount the following year, unless that amount bumps against the ceiling or floor limits -- keeping it within that band. I do not expect to spend every dime, and do expect a terminal bequest amount to result.

In the past, I played with averaging life expectancy from the RMD tables and 120-age, which increases the reductions in withdrawals as time progresses, but doesn't really reduce the income variability that it sounds like you're concerned about.

Smoothing effectively trades some maximum portfolio utilization for reduced income volatility. To me, that is another way of saying that your actual withdrawal rate will vary, leading to a need to determine how much variance is appropriate. You could use other ways to determine the range than mine -- perhaps you would be more comfortable allowing VPW to fluctuate between a 40-year period and a 50-year period, for example.

I don't think you need visual basic to achieve what you want. The core of VPW is essentially the same calculation used for mortgages and car loans, without all the extra monthly periods. I don't use visual basic for any of my planning.

User avatar
siamond
Posts: 3684
Joined: Mon May 28, 2012 5:50 am

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by siamond » Mon Jul 03, 2017 3:22 pm

mortal wrote:So, what's the alternative? I'd like to know what effect using a 5 year moving average would have in calculating a variable withdrawal rate? My intuition is that this would smooth out withdrawals, stretching out fluctuations over a longer time period and giving you more time to adjust to new spending levels. I've heard that prior attempts to smooth vpw failed, and it's very much a 'pay me now or pay me later' type of situation. I understand. I'm entirely ok withdrawing lower amounts for a longer time period provided the individual yearly adjustments aren't as severe or volatile.

My problem is that I don't know how to run the calculations or model it in a spread sheet. Is my intuition right here? What do you guys think?
Previous attempts at convincing the VPW author to add an optional smoothing method to his spreadsheet did fail. Longinvest is very attached to the elegant simplicity of his model, and not terribly sensitive to the fact that its innate volatility presents a clear practical issue for many users and notably those eager to keep a good chunk of equity exposure in their Asset Allocation.

Ok, be it, but there are solutions indeed, which were discussed at length in this thread, and illustrated through backtesting. I ended up using one of those approaches for myself since then, and I know that several other folks on this forum do something similar. Since this thread was quite long, discussed both Guyton-Klinger and VPW/PMT, and ended up with rather 'animated' discussions, maybe I can summarize the findings here, in following posts, since this question pops up quite often. I need to refresh a bit my memory first though...

User avatar
siamond
Posts: 3684
Joined: Mon May 28, 2012 5:50 am

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by siamond » Mon Jul 03, 2017 4:06 pm

To begin, let's discuss short-term volatility. When the portfolio hiccups (up or down) for a year or two, and then comes back near where it was, which a fairly common occurrence, there is really little reason to change the annual withdrawal by much. Let's call such issues 'short-term volatility'. There are at least two ways to deal with that, keeping in mind that we can't know in hindsight if this is indeed a short-term issue, or a more sustained market change.

First way is essentially what the OP suggested, use a moving average of the past X years (I would suggest 3 years instead of 5 though). The idea is simple, you let VPW makes its normal recommendation, and archive the number each year. Then for your withdrawal of the current year, simply average the VPW recommendations for year N-2, year N-1 and current year. To bootstrap the process, just use the regular VPW recommendation for the first two years, or average them if you wish for year 2.

If you're inflation-savvy, then it would be a tad better to do the math in inflation-adjusted terms, but this really wouldn't change the outcome very much, unless we come back to a troubled period like the 70s.

Such approach has the virtue of simplicity, and works reasonably well. It will eliminate a lot of silly ups and downs for sure. One might argue it would be a tad slow to react to a long and sustained crisis (Japan-like) though, or overreact to a long bull market (like the 90s). But still, this will work out fine, I backtested it at length, and this is a reasonable approach.

User avatar
Sheepdog
Posts: 4888
Joined: Tue Feb 27, 2007 3:05 pm
Location: Indiana, retired 1998 at age 65

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by Sheepdog » Mon Jul 03, 2017 4:26 pm

I use a continuous moving average not just a 5 year. This is my process, which long time members here have read it before.

My 18 year withdrawal rate has certainly oscillated, but that was expected. For me, planning a big vacation in Europe or Asia, new autos, or buying and maintaining a lake cottage (all of which are in our history), requires discipline if you don't have excess assets which I don't. My withdrawal plan, a plan where I expected variable needs year to year, has worked. It requires an annual review of past spending and future needs to control withdrawals. My goal was for an AVERAGE annual withdrawal percentage of 4.6%, NOT adjusted for inflation, not a specific set annual withdrawal percentage. (Actually, I started with a 4.5%l plan, but increased it later as I was doing so well. I may do it again.) The withdrawals will be variable year to year, low in some years in order to pay for the large expenses in others, yet keep within the average of multiple years. It requires limiting spending after down market years. (Can you say 2008 and 2009?) That has worked for our lifestyle....good living, new autos on a regular basis, nice vacations and cruises when income is good, plus we want to enjoy sports and entertainment, etc. every year.

The plan was not to schedule any relatively expensive item (for me that means over $12K) unless that purchase will not increase my multi-year average much above my planned average withdrawal. For example, at the end of 2016, I had an excess of $28,945 available from the moving average. That means, I did not spend $28,945 in previous years from the 4.6% annual budget. I can use that for anything special without ruining a budget or cause excess reduction of assets.

Historically, my actual annual withdrawals have ranged from 3.11% to 7.52%, but the 18 year 4.59% moving average is very close to the original plan. The result has been that I have not wanted for anything and my investment total is worth more than when I retired.
Last edited by Sheepdog on Mon Jul 03, 2017 4:32 pm, edited 1 time in total.
People should not say everything they think. They should think about everything they say.

stlutz
Posts: 4206
Joined: Fri Jan 02, 2009 1:08 am

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by stlutz » Mon Jul 03, 2017 4:31 pm

First, I do recommend the very long thread on the subject:

viewtopic.php?t=120430

One of the main points that longinvest makes is that VPW should only be a part of your retirement income strategy. You really should have an income floor which can be provided by: Social Security, pensions, inflation-adjusted annuities, and some combination of TIPS, iBonds, and CDs (i.e. fixed income investments with limited inflation risk). With that taken care of, VPW really handles more discretionary expenses.

I am a fan of longinvest's point of view about averaging withdrawals. The fundamental point is to make withdrawals based on current reality, and not where the market was a year ago or 5 years ago. As a former BH poster used to say, "Markets don't consult the history books before deciding which way to move." If that's the case, why would you base current year consumption on last year's consumption as opposed to on how much money you actually have right now?

Sure, the market roared back pretty quickly in the second half of 2009 after the '08 crash. Was that necessarily the case? And will that necessarily be the case next time there is a significant market decline? I would say "no" in both cases. If so, where the market was in 2007 doesn't really matter if it's 2009. And conversely, if my portfolio has gone up in value a lot the past few years, I should feel free to take that round-the-world cruise as opposed to not spending money because the market was a lot lower back in 2013.

If you can't adjust your spending downward if the market goes down, then you may need to reconsider your asset allocation, whether to annuitize more etc.

One challenge in implementing VPW is the fact that expenses don't always come when you plan them to, whether that be a medical issue or needing a new roof, new AC, and a new sewer line all at the same time. But, if you spend more than the PMT function says you should in one year, then next year your recalculate your new withdrawal amount based on how much money you have then. Again, you base everything on the facts as they are right now.

The bigger challenge is around longevity. VPW is built around spending all of the money by the end year that you've planned for. If you live longer than that, well then the strategy fails. Ways around there are to buy a longevity annuity (which are pretty cheap all things considered), plan your VPW to out to age 110 instead of 85 or 90 (which in most cases defeats the strategy of spending the money while you're alive), or if you have another pot of money available that you were hoping to bequeath to your heirs or the like.

In short, VPW is part of a withdrawal strategy as opposed to being a strategy in and of itself. I think the desire to smooth things out probably suggests that you are asking it to do too much on it's own.

User avatar
siamond
Posts: 3684
Joined: Mon May 28, 2012 5:50 am

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by siamond » Mon Jul 03, 2017 4:40 pm

Another approach towards short-term volatility was provided (surprise!) by longinvest himself in this post. Let me duplicate the key formula.
PlainWithdrawal = portfolio x VPW percentage
ModifiedWithdrawal =
If PlainWithdrawal > LastYearWithdrawal : MAX( PlainWIthdrawal / 1.3, LastYearWithdrawal + ( PlainWIthdrawal - LastYearWithdrawal ) * 0.1 )
If PlainWithdrawal < LastYearWithdrawal : MIN( PlainWIthdrawal x 1.3, LastYearWithdrawal - ( LastYearWithdrawal - PlainWIthdrawal ) * 0.1 )

The slow 10% adjustment should give you the smoothing you seek, while the 30% band should protect the portfolio against catastrophic events.
While backtesting such approach (see details in the other thread), I ended up with the conclusion that the 10% parameter (0.1) was actually doing too much smoothing and 25% seemed more appropriate. While the second parameter (30%) didn't seem terribly useful, and actually introduced some occasional weird hiccup, and I think this simpler formula is perfectly fine:
PlainWithdrawal = portfolio x VPW percentage
ModifiedWithdrawal =
If PlainWithdrawal > LastYearWithdrawal : LastYearWithdrawal + ( PlainWIthdrawal - LastYearWithdrawal ) * 0.25
If PlainWithdrawal < LastYearWithdrawal : LastYearWithdrawal - ( LastYearWithdrawal - PlainWIthdrawal ) * 0.25
From the various tests I ran, I believe this approach is superior to the moving average. Less need for past years data, it is fairly reactive without overreacting, and it's really quite simple.

User avatar
siamond
Posts: 3684
Joined: Mon May 28, 2012 5:50 am

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by siamond » Mon Jul 03, 2017 4:55 pm

Now, there is a more complicated consideration. The short-term issues are the most glaring. But the more I ran backtesting cycles, the clearer it became that there is also a long-term volatility issue, some 'big waves' you can see on the trajectory of withdrawals, which are rather disturbing, and not quite necessary in hindsight. Trouble is there are other scenarios (e.g. Japan-like crisis) when things go downwards for decades, with no end in sight. And that's quite the conundrum.

Long story short, if one replaces the fixed rate of return in the PMT formula underlying the VPW logic by an expected return computation refreshed every year, then the problem of long-term volatility is largely mitigated while ALSO properly adjusting to catastrophic Japan-like scenarios. The core principle is simple, if the current value of one's portfolio falls by 50% in one year, then the expected returns are significantly higher than usual, and it makes little sense to stick to a constant rate. And the converse can be said for a crazy high like in the end of the 90s. Now if both price and (smoothed) earnings fall apart, then it is indeed time to tighten the belt...

I am not going to provide more details here, this is an approach which requires more mastery of the inner workings of VPW (which are actually extremely simple!) AND a leap of faith that some expected returns models (e.g 1/PE, bond yields) do have a solid kernel in truth in there, AND that the robustness of the PMT formula is such that the probabilistic nature of expected returns is actually quite fine for such use case. Interested folks can read more starting from this post and/or send me PMs.

One last thought: interestingly enough, this approach nicely mitigates long-term volatility, but ALSO does a decent job for short-term volatility. And is very adaptive towards an uncertain future. And doesn't require any math based on inflation or last year withdrawal or whatever. So... I personally settled on this. Time will tell!

User avatar
siamond
Posts: 3684
Joined: Mon May 28, 2012 5:50 am

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by siamond » Mon Jul 03, 2017 5:10 pm

stlutz wrote:In short, VPW is part of a withdrawal strategy as opposed to being a strategy in and of itself. I think the desire to smooth things out probably suggests that you are asking it to do too much on it's own.
No, this isn't quite true - or at least not for everybody. All the backtests I ran were based on real-life scenarios, with a basis of fixed income (SSA/Pension/etc) in addition to portfolio withdrawals. Yes, this makes withdrawal volatility less impactful, because what really matters is the variability of your spending budget (the sum of fixed income plus the withdrawal), but the outcome remains very difficult to accept imho. See details and numerous examples in the G-K/VPW thread. This post is a good starting point.

This being said, it certainly depends A LOT of one's personal circumstances and one's psychology/behavior. Personally, I strongly believe in providing options. We're all different in our needs and goals and behavior. The base VPW (as part of an overall retirement plan with fixed income) is perfectly fine for some people. While other people have different needs/requirements, and this does include forms of smoothing. There is no one-size-fits-all when it comes to retirement... So I tried to summarize the options I know of in the previous posts. I hope this helps the people who feel the problem is very real.

longinvest
Posts: 2547
Joined: Sat Aug 11, 2012 8:44 am

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by longinvest » Mon Jul 03, 2017 5:30 pm

Mortal,
mortal wrote:I'm entirely ok withdrawing lower amounts for a longer time period provided the individual yearly adjustments aren't as severe or volatile.

Using a balanced portfolio (e.g. 50/50 stocks/bonds) will deliver a much smoother ride than a 70/30 or 80/20 portfolio when looking (rightfully) at the nominal withdrawal path. Looking at a CPI-adjusted withdrawal path is a mistake, as CPI is only an indirect measure of inflation, and it is quite volatile (often more so than bonds).

I simply don't remember the last time I checked how my paycheck fluctuates in CPI-adjusted terms. Why would I expect to start doing so after retirement? All I care about is that I have enough money to buy food and pay for shelter, transportation, clothing, and other things. When broccoli costs more, I buy cauliflower (or another vegetable) instead. Long term inflation is a real concern, though. So, we do care about inflation, but not necessarily about short-term CPI fluctuations which significantly affect historical returns.

Let me repeat this: looking at a CPI-adjusted chart without taking into account the volatility and imprecision of CPI to measure short-term personal inflation is a huge mistake.

Here's the backtested nominal VPW withdrawal path of a $1,000,000 50/50 portfolio using a 35-year horizon, starting in 1966, one of the worst ever retirement years:

Image

The following chart shows what happens when we combine VPW with base income, as recommended in our wiki. I added a $20,000 Social Security (SS) inflation-indexed pension to the above backtested VPW withdrawal path. The black line represents total income (VPW + SS):

Image

This is what a retiree would have seen deposited into his bank account, annually. Total income kept increasing (with very minor setbacks in very bad years). Better: the long-term increase actually beat inflation.

This definitely qualifies as good enough for me.


As for smoothing VPW using averaging techniques (or market-timing using metrics such as P/E ratios), while maintaining a high-stock allocation, here's my opinion:

People fail to see how dreadful it was to be fully invested in stocks during 1973-1975. Yes, I know, inflation was high and it was not the best of times for nominal bonds either, but stocks fared much worse than bonds! The nominal dip in the 50/50 portfolio VPW path seen above was entirely due to stocks. Bonds actually had positive nominal returns all along in the 1970s. One would have slept much better with a 50/50 portfolio than with a 100/0 portfolio during this period. Nobody seems to see this, as they usually look at CPI-adjusted charts and often use highly-volatile long Treasuries for bond returns.

It is very easy, mathematically, to average VPW withdrawals (or find a metric that backtests well) to smooth them and hide the 40% drop experienced by a 100% stock portfolio in 1973-1974 while inflation was up 20%, but this means that the retiree would have been taking an unsustainable amount of money out of the portfolio at the worst of times, hurting the portfolio in the process.

Of course, with 20/20 hindsight, we now know that it was "safe" to withdraw a smoothed amount in the mid 1970s, as we know that the portfolio would have been able to recover from the harm done to it by the outsized withdrawals. But, I doubt that any sane retiree would have actually withdrawn such an unsustainable amount at the time, when looking at his portfolio balance during a deep crisis, without knowing for sure what will happen in the future. It would be quite scary to apply a similar smoothing approach in a Japan-like situation where stocks dropped and never recovered within the next almost 30 years.


In summary: It is my opinion that combining lifelong non-portfolio inflation-indexed base income* with VPW applied on a balanced portfolio is a simple yet very robust approach to retirement funding. If too high, the volatility of total income can easily be reduced by (a) increasing the amount of lifelong non-portfolio income or (b) adding more bonds to the portfolio. As for inflation worries, with bonds, there now exist bonds which carry no inflation risk: Treasury Inflation Protected Securities (TIPS).

* Early retirees can use a non-rolling TIPS ladder to safely fill the gap between retirement and the start of pension payments.
Last edited by longinvest on Mon Jul 03, 2017 10:47 pm, edited 4 times in total.
Bogleheads investment philosophy | Lifelong Portfolio: 25% each of (domestic/international)stocks/(nominal/inflation-indexed)bonds | VCN/VXC/VAB/ZRR

longinvest
Posts: 2547
Joined: Sat Aug 11, 2012 8:44 am

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by longinvest » Mon Jul 03, 2017 5:49 pm

The simplicity of VPW is very important. I cannot imagine explaining to my wife how to go and lookup CPI numbers and calculate complex ratios (or use a spreadsheet) after my death. But, I would have no trouble explaining to her how to use the following table which would be easy to include in my will:

50/50 portfolio allocation

Year Age Withdrawal Percentage
2016 65 4.8%
2017 66 4.9%
2018 67 4.9%
2019 68 5.0%
2020 69 5.1%
2021 70 5.2%
2022 71 5.3%
2023 72 5.4%
2024 73 5.5%
2025 74 5.7%
2026 75 5.8%
2027 76 6.0%
2028 77 6.1%
2029 78 6.3%
2030 79 6.5%
2031 80 6.8% *** Buy as much inflation-indexed SPIA as necessary (but no more) to insure a sufficient lifelong income floor (including Social Security and other pensions) and, thus, eliminate longevity risk ***
2032 81 7.0%
2033 82 7.3%
2034 83 7.6%
2035 84 7.9%
2036 85 8.3%
2037 86 8.8%
2038 87 9.3%
2039 88 10.0%
2040 89 10.7%
2041 90 11.6%
2042 91 12.7%
2043 92 14.0%
2044 93 15.8%
2045 94 18.1%
2046 95 21.4% *** Possibly limit this and future portfolio withdrawals to 20%, so as to never fully deplete the portfolio ***
2047 96 26.3%
2048 97 34.5%
2049 98 50.8%
2050 99 100.0%

DIsclaimer: My wife and I are younger than the above example table indicates. I'm still in my 40's. I'm not retired.

Acronyms:
CPI: Consumer Price Index
SPIA: Single Premium Immediate Annuity
VPW: Variable Percentage Withdrawal
Bogleheads investment philosophy | Lifelong Portfolio: 25% each of (domestic/international)stocks/(nominal/inflation-indexed)bonds | VCN/VXC/VAB/ZRR

aristotelian
Posts: 3597
Joined: Wed Jan 11, 2017 8:05 pm

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by aristotelian » Tue Jul 04, 2017 8:19 am

OP,
Can you clarify, it appears you are talking about variable withdrawal (constant % based on variations of total portfolio size) like a nonprofit endowment. Variable percentage withdrawal would vary the %, usually according to age, as life expectancy declines (as above poster laid out). I believe you are talking about the first but you use the term VPW in the thread title.

I work for a nonprofit that does 5% based on 3 year moving average, invested over 90% in stocks. We have to live with major fluctuations that might be unacceptable to an individual in retirement, but so far the strategy has worked to grow the endowment over time.

User avatar
Top99%
Posts: 266
Joined: Sat Apr 22, 2017 9:30 am
Location: Austin, TX

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by Top99% » Tue Jul 04, 2017 9:15 am

longinvest wrote:The simplicity of VPW is very important. I cannot imagine explaining to my wife how to go and lookup CPI numbers and calculate complex ratios (or use a spreadsheet) after my death.
Great post as always Longinvest. And a huge thank you for your efforts in creating VPW. VPW along with the approach described in the "delay Social Security to 70 / spend more money at 62" thread have had a profound positive impact on our plans and one of the reasons I am now working part time. :sharebeer
Adapt or perish

stlutz
Posts: 4206
Joined: Fri Jan 02, 2009 1:08 am

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by stlutz » Tue Jul 04, 2017 12:04 pm

Using a balanced portfolio (e.g. 50/50 stocks/bonds) will deliver a much smoother ride than a 70/30 or 80/20 portfolio when looking (rightfully) at the nominal withdrawal path. Looking at a CPI-adjusted withdrawal path is a mistake, as CPI is only an indirect measure of inflation, and it is quite volatile (often more so than bonds).
After agreeing with your earlier, I'm afraid I have to disagree with you on this point, longinvest.

When running a historical simulation, you need to adjust your results for inflation. Yes it is true that the CPI is different from my personal inflation rate, and frankly, I don't know if one's personal rate of inflation is even measurable. However, that's a lot different from just saying that 10% is functionally equal to 0%.

When I backtest the VPW approach, starting in 1967 does long look good in real terms regardless of the asset allocation. Which years were the worst can vary based on stock/bond allocation, but the fact is that your withdrawal in 1982 is much lower than it was in 1967, regardless of your bond weighting. That was a bad time for pretty much everything. Bonds did not in fact provide a very good portfolio cushion. Ignoring inflation isn't a particularly realistic way to deal with this problem.

Looking at '67ff in inflation-adjusted terms is a pretty good "worst case" scenario to test for a withdrawal plan. That doesn't mean that one should create the portfolio that would have worked best then, as a future worst case scenario would unfold differently.

All of that said, one would not actually factor the CPI into the withdrawals they take when actually executing VPW. Rather, the PMT formula provides you a withdrawal amount and that's what you can take. CPI doesn't come into play at all.

longinvest
Posts: 2547
Joined: Sat Aug 11, 2012 8:44 am

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by longinvest » Tue Jul 04, 2017 6:21 pm

Stlutz,
stlutz wrote:
Using a balanced portfolio (e.g. 50/50 stocks/bonds) will deliver a much smoother ride than a 70/30 or 80/20 portfolio when looking (rightfully) at the nominal withdrawal path. Looking at a CPI-adjusted withdrawal path is a mistake, as CPI is only an indirect measure of inflation, and it is quite volatile (often more so than bonds).
After agreeing with your earlier, I'm afraid I have to disagree with you on this point, longinvest.

When running a historical simulation, you need to adjust your results for inflation. Yes it is true that the CPI is different from my personal inflation rate, and frankly, I don't know if one's personal rate of inflation is even measurable. However, that's a lot different from just saying that 10% is functionally equal to 0%.
I did not say that all simulations should be done in nominal terms; I said that smoothness should be evaluated in nominal terms. It's not the same thing.

You'll note that the VPW backtesting spreadsheet actually provides both nominal and CPI-adjusted simulations on the same chart. This allows for evaluating the different aspects of a withdrawal path. A CPI-adjusted simulation is fundamental to evaluating the sustainability of the withdrawal path. What I am saying is that a nominal simulation is more appropriate to evaluate its smoothness.

Here's an example. If I gave you $50 each month, during the next 5 years, would you say that I am giving you a volatile stream of payments? Most people would say that, on the contrary, I was giving you a smooth stream of payments. But, if you decided to buy gas for your car using this monthly $50, the quantity you would be able to buy would fluctuate a lot. Does this make the $50 unsmooth? Personally, I say no; I say that gas prices are volatile, not the monthly 50$.

In real life, lots of prices that we pay everyday are quite volatile. Food, gas, etc. Yet, other prices are actually quite stable, such as mortgage payments. Note that mortgage payments are often a significant part of the budget of younger families.

CPI is just a wide-ranging average. Many of our expenses are often fixed for a period by contract. Regardless of how CPI fluctuates, mortgage payments don't change until the end of the contract. When the price of houses fluctuates, it mostly affects the ability of new buyers to buy them; people who already own a house don't get higher mortgage payments due to higher house prices. Yet, I'm sure CPI is partially affected by house price variations.

Inflation is insidious. It's not the impact of one year of inflation that kills one's purchase power, it's the cumulative effect of inflation over a few years. We often instinctively dodge many short-term effects of inflation through smart choices. The problem is that after a while, our options run out and we are impacted all at once by the cumulative increase in prices. For example, when we run out of affordable vegetables, we start paying for the higher price vegetables. The jump in prices that we experience at that point can be significant, because it represents the accumulation of many small increases.

So, just to try to make myself clear, again. I am not saying that it's smart to ignore inflation, or that inflation has no impact. I am saying that short-term price fluctuations should not be accounted into the smoothness of a stream of payments. That's all.
stlutz wrote:When I backtest the VPW approach, starting in 1967 does long look good in real terms regardless of the asset allocation. Which years were the worst can vary based on stock/bond allocation, but the fact is that your withdrawal in 1982 is much lower than it was in 1967, regardless of your bond weighting. That was a bad time for pretty much everything. Bonds did not in fact provide a very good portfolio cushion. Ignoring inflation isn't a particularly realistic way to deal with this problem.
Let's compare. Here's the 50/50 portfolio's VPW withdrawal path for a 1966 retirement (without Social Security):

Image

Here's the 100/0 portfolio's VPW withdrawal path for a 1966 retirement (without Social Security):

Image

First, let's look at the red lines which represent the CPI-adjusted VPW withdrawal paths. We can see that both streams did badly, in inflation-adjusted terms, in the 1975-1985 period, delivering something around $30K per year (sometimes a little more, sometimes a little less). This tells us that neither was more sustainable than the other.

Second, let's look at the blue lines which represent the nominal VPW withdrawal paths. It should be easy to see that the 50/50 line is much smoother than the 100/0 one.

You might have been able to say the same thing about the red lines, but it's not as clear cut, because inflation significantly affects the paths. One has trouble seeing that the balanced portfolio delivered a smooth stream of payments.

Let's make an exercise (in nominal terms). Imagine a retiree 100% invested in stocks. In January 1973, he gets a $70K VPW payment. He's happy and feels rich with his $1.1M portfolio. He tells his wife: "See how we're doing good; it's our 8th year of retirement and our initial $1M portfolio is now even bigger!" Unfortunately, 2 years later, in January 1975, he gets a $40K withdrawal and has to tell his wife: "We're not millionaires anymore. Markets are bad, we've lost half of a million dollars. Our portfolio is now $600K."

Imagine another retiree invested in a balanced portfolio, this time. She has quite a different experience during the same period. In January 1973, she gets a $60K VPW payment. She's happy and feels rich with her $1.1M portfolio. She tells her husband: "See how we're doing good; it's our 8th year of retirement and our initial $1M portfolio is now even bigger!" Unfortunately, 2 years later, in January 1975, she gets a $45K withdrawal and has to tell her husband: "We're not millionaires anymore. Markets are bad, we've lost quarter of a million dollars. Our portfolio is now $850K."

If we were to look purely at the numbers, some would say that the first retiree should be happier. (I'm only considering the 1973 and 1975 withdrawals, as a simplification). $70K + $40K = $110K in withdrawals is bigger than $60 + $45 = $105K. Yet, I personally would prefer to tell my wife that we lost quarter of a million and that we have to cut our budget by 25%, than tell her that we lost half of a million and that we have to cut the budget by 40%.

In the years that followed, 1976 to 1985, both the balanced and the 100% stocks portfolios delivered (mostly) increasing VPW payment streams in nominal terms, sometimes beating inflation a little, sometimes losing to inflation a little. Unfortunately, the harm had already been done in 1973-1975; payments remained low in inflation-adjusted terms.

Note that any mathematical smoothing of the 100% stocks path implies taking even more than $40K from a $600K remaining portfolio in 1975. Personally, I would definitely not have the temerity to do that; I would much prefer taking a (mathematically) unsmoothed $45K from a $850K portfolio.
stlutz wrote:Looking at '67ff in inflation-adjusted terms is a pretty good "worst case" scenario to test for a withdrawal plan. That doesn't mean that one should create the portfolio that would have worked best then, as a future worst case scenario would unfold differently.

All of that said, one would not actually factor the CPI into the withdrawals they take when actually executing VPW. Rather, the PMT formula provides you a withdrawal amount and that's what you can take. CPI doesn't come into play at all.
I think that we mostly agree, but that there was a misunderstanding. I was simply trying to explain that CPI-adjusted charts are misleading when trying to assess the smoothness of a stream of payments. I was also trying to explain the importance of a relatively stable portfolio, when one wants to extract a relatively stable stream of withdrawals from it.
Last edited by longinvest on Tue Jul 04, 2017 10:32 pm, edited 4 times in total.
Bogleheads investment philosophy | Lifelong Portfolio: 25% each of (domestic/international)stocks/(nominal/inflation-indexed)bonds | VCN/VXC/VAB/ZRR

longinvest
Posts: 2547
Joined: Sat Aug 11, 2012 8:44 am

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by longinvest » Tue Jul 04, 2017 7:47 pm

Let me add that I'm not alone in generally considering smoothness in nominal terms, separately from inflation.

Typically, retirees living on the dividends paid by their 100% (selected) stocks portfolio are proud to claim that their dividend stream is much smoother than the income of retirees taking a percentage of portfolio withdrawal (out of a 100% stocks portfolio, of course). What interests them is that many dividend stocks pay relatively fixed dividends (in nominal terms) which are increased every year, or two, or three. They perceive this income as stable and increasing. As long as it grows fast enough to offset inflation over time, they're happy. They don't seek perfect CPI-indexed matching increases (and decreases) on every quarterly payment.

We live in a nominal dollar world. CPI is just an indirect measure of inflation. We don't perceive the impact of short-term CPI-U variations in real life as much as we perceive a nominal loss or gain on our quarterly brokerage statement.

In other words, there's no psychological gain to reduce a mostly-imperceptible volatility (that is due to CPI), when the problem we're trying to solve is a psychological aversion to income volatility (perceived in nominal term). Don't you agree?

Of course, when people look at CPI-adjusted charts, they want to see nice smooth lines on them. But, they often don't realize that this does not match what they perceive in real life, outside of the colorful charts.
Last edited by longinvest on Tue Jul 04, 2017 9:53 pm, edited 3 times in total.
Bogleheads investment philosophy | Lifelong Portfolio: 25% each of (domestic/international)stocks/(nominal/inflation-indexed)bonds | VCN/VXC/VAB/ZRR

stlutz
Posts: 4206
Joined: Fri Jan 02, 2009 1:08 am

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by stlutz » Tue Jul 04, 2017 8:26 pm

Thanks for the clarifications, longinvest--that is very helpful!! :sharebeer

AlohaJoe
Posts: 2835
Joined: Mon Nov 26, 2007 2:00 pm
Location: Saigon, Vietnam

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by AlohaJoe » Tue Jul 04, 2017 10:43 pm

mortal wrote:it's very much a 'pay me now or pay me later' type of situation.
This is not really an accurate summary. When you smooth income you are generating a kind of risk. The risk can be removed by either reducing withdrawals or by the market performing better. It is entirely possible that you smooth income and never have to "pay later". Actually, that's what happens the vast majority of the time. That said, there are possible scenarios where it does eventually catch up to you. If you follow a traditional 4% safe withdrawal rate strategy and run out of money that's the classic example of (extreme) income smoothing eventually having to "pay later".

I've come to believe that you can't really destroy risk in investing. But you can trade one kind of risk for another. And everyone has different risk tolerances for all the different kinds of risk out there. Income volatility is one kind of risk. Portfolio volatility is another. Longevity a third. There are dozens of kinds of financial risks and we have to figure out which ones we care about most and what tradeoffs we'll make when balancing the various risks.

Income smoothing trades one risk for another -- income volatility risk goes down but "lack of mean reversion" risk goes up.

Here's a series of withdrawals (in nominal dollars), showing "plain" VPW versus one that uses a 3-year rolling average. The red boxes highlight areas where plain-VPW had large single-year income drops that were quickly reversed. You can also see that any kind of income smoothing also is slower to adjust upward -- but few people are worried about that kind of risk. This is also an example where there was no "pay later".

Image

To get a sense for what income smoothing has been like historically, here's the same for 1920-1985 in five year increments (i.e. 1925, 1930, 1935, 1940...). It is an external link, since I don't want to clog up Bogleheads with 13 more charts:

http://imgur.com/a/Yrt6l

I looked at 3 different income smoothing schemes -- the one siamond mentioned above that longinvest created, rolling averages, and one suggested by Ken Steiner. Here's what they look like for someone who retired in 1921.

Image

I also looked at rolling averages using 2 through 7 years of data. I did not look at using varying rates in the PMT calculation as siamond suggests, though. I wrote it all up in a blog post:

https://medium.com/@justusjp/income-smo ... 0062771466

There is a spectrum of income smoothing. The traditional "4% Safe Withdrawal Rate" is one extreme end (where it is totally smooth, at least on an inflation-adjusted basis) and VPW or a constant percentage is the other end. When I looked at the historical record I came away convinced that moderate amounts of smoothing can paper over short-term market volatility without introducing meaningful amounts of other risks.

I didn't really write about these in my blog posts but here are a few other things to consider:
  • Even if you've convinced yourself that income smoothing is safe enough for you...will you really follow the plan when markets tank? Even among relatively level headed Bogleheads, there are many instances of fear and even panic during relatively normal/common market movements.
  • If you're willing to use VPW then you know that markets can crash and stay down for 2-3 years pretty easily. So you have built a retirement plan with a fair amount of flexibility. Spending you can cut, purchases you can delay, gratification that go unsatisfied. Is smoothing really buying you that much in the real world? Does income smoothing just mean you cut $3,000 instead of $6,000 this year?
  • Most models (like all of my charts above, the VPW spreadsheet, and many other places) assume annual withdrawals. That is, in January 2018 you'll sell $90,000 of stocks & bonds, put it in your checking account, and then spend it over the course of the year. Maybe some people do that? But I think more people make their withdrawal quarterly or even monthly. In that scenario then I think the case for income smoothing becomes stronger, because we know that intra-year market volatility is more pronounced than annual volatility.

User avatar
siamond
Posts: 3684
Joined: Mon May 28, 2012 5:50 am

Re: Using a 5y moving average to smooth withdrawals in VPW?

Post by siamond » Wed Jul 05, 2017 10:07 am

Thank you, AlohaJoe, for a great study and for bringing back this topic to the OP's question.

Would be great to see your analysis extended to the adjusted smoothing formula I suggested above, and to see the charts both in nominal terms (for the psychological reasons longinvest and yourself mentioned) and in inflation-adjusted terms (to better capture the real goal, protect one's purchasing power and standard of living). Both perspectives are useful.

A technical note: standard-deviation analysis on a growing series of numbers is unfortunately rather skewed, and the nominal analysis is of course especially prone to that. I believe the human eye tends to adjust by comparing the trajectory to a much smoother line in our mind, with our uncanny ability to identify patterns out of noise. Translating that in a formula could be captured by looking at the std-deviation of the difference between the numbers and their regression. Trouble is this might have to be a polynomial regression. And this gets complicated... Fortunately, an inflation-adjusted trajectory is much easier to analyze.

Oh, and I agree that smoothing small variations is quite inconsequential in real life. It's really when markets get in a frenzy (deep crisis AS WELL AS big bull markets) that it gets really important to have a proper target in mind to guide one's decisions. And that's good, because this is the time where we get the most emotional. This may not prevent us from overreacting, but at least, we'll have solid guidelines to look at.

Post Reply