Portfolio Tracking Withdrawal Method (with comparison)

[squonk2]

I just retired at 59.5 and need a withdrawal method that satisfies these requirements:

(1) It must not allow our portfolio to ever be fully depleted, no matter the market returns or our lifespan. Therefore it must use, basically, a percentage of portfolio formula with zero floor. The effective percentage may be dynamically adjusted below or above the expected rate of return to grow or shrink the portfolio as desired.

(2) It must not allow the portfolio to grow exponentially. Therefore it must increase the withdrawal rate with the size of the portfolio above some point.

(3) Prioritize maximizing spending over legacy. Ideally, the late portfolio would be just large enough to sustain our living expenses with its total returns. Gifting should be done earlier.

(4) Accommodate an arbitrary withdrawal profile--in particular, one with a high withdrawal rate during the earlier years and smaller withdrawals later, which is the reverse of what most withdrawal methods produce. This requirement also accounts for uneven guaranteed income such as delayed Social Security benefits.

(5) It must not depend on longevity estimates. We plan to live forever, in theory.

(6) Provide a portfolio fuel gauge so that we always know where we stand and whether corrective action is needed.

(7) Income volatility is acceptable. However, the plan should mitigate downturns as much as possible as long as the other requirements are met.

(8) It should be as simple as possible, but no simpler.

Necessity is the mother of invention. Here is what I came up with, in three steps:

1. Create a yearly withdrawal plan. Project a year-by-year budget spanning the go-go years and into the slow-go phase. Use best estimates of expenses, guaranteed income, and inflation. Withdrawal Plan amounts are total expenses minus guaranteed income:

(a) Plan(year) = Expenses(year) - Income(year)

In our case, planned withdrawals are high at first and fall to a much lower level in later years after the mortgage is paid off, deferred taxes are mostly paid, and Social Security benefits are coming in.

2. Use the withdrawal Plan to estimate the Minimum sustainable portfolio value year-by-year. The curve may slope downward from the initial value and flatten out in later years, like a pinewood derby track. At the endpoint, we want the portfolio to be indefinitely sustainable with roughly constant withdrawal amounts. The Minimum portfolio value then is that year’s withdrawal Plan amount divided by a conservative nominal rate of return, perhaps somewhere between 4% and 6% depending on one’s asset allocation and tolerance for income uncertainty.

(b) Minimum(last_year) = Plan(last_year) / Return

An amortization formula is then used to calculate Minimum portfolio values for prior years, back to the year of retirement:

(c) Minimum(year) = Minimum(year + 1) / (1 + Return) + Plan(year)

Next we add a safety margin, say 33%, and that's our portfolio Target track:

(d) Target(year) = Minimum(year) x 1.33

We made sure that our actual portfolio was greater than Target at the time of retirement. Otherwise, we would have delayed retirement or look for ways to reduce expenses.

The band between Minimum and Target is regarded as the safe zone. Having a wide safe zone helps mitigate market volatility. Ideally, our portfolio would stay near the Target. But, of course, it will wander according to market returns.

3. The key component for meeting the requirements is a course-correcting withdrawal formula. If the portfolio falls below the safety zone, withdrawals are reduced proportionally. If the portfolio rises above the safety zone, a bonus proportional to the excess is added to the withdrawal. This is the withdrawal formula:

(e) If Portfolio < Minimum: Withdrawal = Plan x Portfolio / Minimum.
(f) If Portfolio > Minimum and Portfolio < Target: Withdrawal = Plan.
(g) If Portfolio > Target: Withdrawal = Plan + (Portfolio - Target) x 25%.

Comparing the current portfolio value to the established target provides a ready indicator, or fuel gauge, for the portfolio. When the portfolio is in the “green” safe zone, we go on with our merry lifestyle. If the portfolio falls into the “red” zone below Minimum, we spend more carefully. When the portfolio grows above Target, we can spend more, shore up our cash reserve, and donate extra to charitable causes. (Another financial health indicator is the size of cash reserve, in terms of months of expenses.)

The plan may be updated at any time with improved estimates of spending, guaranteed income, inflation, and rate of return.

I ran 1000 trials of Monte Carlo with random returns and randomized mean and standard deviation (within what I consider to be reasonable ranges). Spending was reduced below Plan only about 2% of the time, and never below 85% of Plan. Most years, it produced happy bonuses. Your results may vary. Perhaps it's counter-intuitive, but disregarding safe withdrawal rates and focusing instead on maintaining a moderate, sustainable portfolio seems to best optimize withdrawals while avoiding extreme outcomes.

The main drawback I see that might be a turn-off to some is the amount a planning required at the beginning. But after that, it is quick and easy to manage with spreadsheet software. Most of the management effort is in gathering account balances periodically. My wife likes the calculator and uses it, and she isn’t a spreadsheet aficionado. She especially likes to know that we are comfortably on track to continue our lifestyle and not run out of money.

Our comprehensive financial plan also covers cash flow, cash reserve, tax strategy, and asset allocation, but that is beyond the scope of this discussion.

That’s it. I would appreciate comments, especially if anyone sees flaws in my plan.

[Edit 2023.7.26: Added missing formulae.]

*****
[Update 2023.7.26]

Part 2: Backtest comparison

Here we compare the performance of the Portfolio Tracking Withdrawal Method to that of two popular methods: constant real value withdrawals and percentage of portfolio withdrawals. To facilitate comparison, the following conditions are set:

Tracking: $40k withdrawal plan, $1M constant Target, $750k Minimum. Withdrawals are calculated using the withdrawal formula given the current portfolio value.
Constant: $40k fixed withdrawal.
Percentage: Withdrawal is 4% of portfolio.

In each case the starting portfolio is $1M, and all amounts are in real (inflation-adjusted) value.

Real returns for a 60/40 asset allocation are extracted from the Bogleheads SIMBA database: https://www.bogleheads.org/wiki/Simba%2 ... preadsheet.

First we examine an unusually stressful period, 1962 to 1981, and chart the value of the portfolio and withdrawal amounts generated by the sequence of returns:



There is not a vast difference here between the methods. The Tracking method preserves a little more of the portfolio than the Constant method and provides a little more income than the Percentage method. The Constant withdrawal may seem optimal until one considers what it would be like to experience that situation. Could a retiree, or should they, keep hanging on while the portfolio heads toward a cliff, not knowing whether a turnaround is imminent? Both the Tracking method and the Percentage method at least offer a little reassurance that the portfolio can't go all the way to zero.

Now we look at a particularly prosperous period, 1980 to 1999:



This scenario better illustrates the difference between the plans. The Tracking method keeps the portfolio under control and generates higher income earlier, when it could be most beneficial. That also helps reduce RMD tax risk if a substantial portion of the portfolio is in tax-deferred accounts.

*****

[dcabler]

Welcome to the forum!

It has some similarities to other withdrawal methods I looked at over the years (I'm also an engineer).
- Gummy's sensible withdrawals is one where it's a mashup of SWR and a way to get a bonus in good years
- There's less similarity to Hebeler's autopilot mode except he also has something of a feedback mechanism (he's a former Boeing engineer)


Other commentary
- I personally think that SWR is such a bad idea in practice that I don't even consider comparing it to anything anymore.
-You ran this with monte carlo, but most people who run monte carlo don't take into account serial correlations of the sequence of returns. Did you? Alternatively, did you run this also with historical sequences and how did that look, especially in periods with sustained low performance?
- Some graphs with results would be cool to see.
- Some of your requirements seem to be at odds with others. #2 and #4 for example

Have you looked at other methods available on this forum such as VPW, ABW & TPAW?
ABW, in particular, also lets you create a profile of spending. And either of these methods could be modified to, essentially, satisfy "live forever" requirement by either using an extremely large number of remaining years or by using life expectancy tables for remaining number of years. There's also McClung's book and ERN's series on withdrawals...

Cheers.

[squonk2]

Thank you!

I spent a lot of time looking at all the proposed withdrawal methods I could find, and none satisfied my requirements. Requirements (2) and (4), and the others, are important to me, maybe not to everyone. I have McClung's book and read it all the way through, parts of it multiple times.

I did some back-testing, but only with a limited amount of of historical data I dug up on 60/40 returns from 2000 to 2019. The results I got were enough to satisfy me: never did the withdrawal amount fall below my plan. I had used a 6% expected nominal rate of return to construct my withdrawal plan. I'll leave a more detailed analysis to someone else if they're interested. I rather hope someone with a rich collection of historical data and experience in how to do it properly will give it a try. I'd like to see the analysis with both 4% and 6% relating withdrawal Plan to Minimum portfolio. Note, though, that most analyses look at probability of failure. With a percentage of portfolio formula there can be no failure, as in completely depleting the portfolio. The question is what range of withdrawal adjustments should we reasonably expect.

[dcabler]

Thank you!

I spent a lot of time looking at all the proposed withdrawal methods I could find, and none satisfied my requirements. Requirements (2) and (4), and the others, are important to me, maybe not to everyone. I have McClung's book and read it all the way through, parts of it multiple times.

I did some back-testing, but only with a limited amount of of historical data I dug up on 60/40 returns from 2000 to 2019. The results I got were enough to satisfy me: never did the withdrawal amount fall below my plan. I had used a 6% expected nominal rate of return to construct my withdrawal plan. I'll leave a more detailed analysis to someone else if they're interested. I rather hope someone with a rich collection of historical data and experience in how to do it properly will give it a try. I'd like to see the analysis with both 4% and 6% relating withdrawal Plan to Minimum portfolio. Note, though, that most analyses look at probability of failure. With a percentage of portfolio formula there can be no failure, as in completely depleting the portfolio. The question is what range of withdrawal adjustments should we reasonably expect.

If you're AA is more or less conventional, then there is significantly more data available on the SIMBA spreadsheet, which is downloadable on this forum. Only caveat is that it is annual-only data. But there's a ton of return series there, many of which goes back to the early 1920's or even before that...

Cheers.

[GeoMetry]

For a portion of your portfolio you could consider a Single Premium Immediate Annuity (SPIA) its simple and it lasts for as long as you live. It could provide the foundational income then you could spend the rest of your portfolio at any time with out any concerns. Alternatively since SPIAs get cheaper as you get older you could just set aside enough money to buy the SPIA later and if you are still alive at that future date you could decide then if you want to buy the SPIA.

I'm guessing you have no interest in a SPIA but its was just what I thought of based on your requirements.

[squonk2]

For a portion of your portfolio you could consider a Single Premium Immediate Annuity (SPIA) its simple and it lasts for as long as you live. It could provide the foundational income then you could spend the rest of your portfolio at any time with out any concerns. Alternatively since SPIAs get cheaper as you get older you could just set aside enough money to buy the SPIA later and if you are still alive at that future date you could decide then if you want to buy the SPIA.

I'm guessing you have no interest in a SPIA but its was just what I thought of based on your requirements.

Good suggestion. I decided, though, to skip it. We expect that in our later years SS will probably cover about all our essential expenses. With careful management of our portfolio we should be able to fund discretionary expenses and more. If push comes to shove, we will downsize.

[squonk2]

Welcome to the forum!

It has some similarities to other withdrawal methods I looked at over the years (I'm also an engineer).
- Gummy's sensible withdrawals is one where it's a mashup of SWR and a way to get a bonus in good years
- There's less similarity to Hebeler's autopilot mode except he also has something of a feedback mechanism (he's a former Boeing engineer)


Other commentary
- I personally think that SWR is such a bad idea in practice that I don't even consider comparing it to anything anymore.
-You ran this with monte carlo, but most people who run monte carlo don't take into account serial correlations of the sequence of returns. Did you? Alternatively, did you run this also with historical sequences and how did that look, especially in periods with sustained low performance?
- Some graphs with results would be cool to see.
- Some of your requirements seem to be at odds with others. #2 and #4 for example

Have you looked at other methods available on this forum such as VPW, ABW & TPAW?
ABW, in particular, also lets you create a profile of spending. And either of these methods could be modified to, essentially, satisfy "live forever" requirement by either using an extremely large number of remaining years or by using life expectancy tables for remaining number of years. There's also McClung's book and ERN's series on withdrawals...

Cheers.

Oh, serial correlation... I built serial correlation into my model. But when it came down to quantifying it I ran into trouble. Looking at historical data, serial correlation was, at least for me, very hard to pin down at any interval--sometimes positive, sometimes negative, often generally zero. I just could not come up with a confident expectation of what it will be like in the future, especially as we diversify beyond traditional assets. So I ended up just setting the constant to zero. And I figured if I run enough MC trials, at least a few of them will happen to appear with serial correlation.

[livesoft]

I think this plan is great! But after a few years (maybe even months or weeks?) you will change the plan anyways, so I recommend that you don't get attached to it.

[squonk2]

If you're AA is more or less conventional, then there is significantly more data available on the SIMBA spreadsheet, which is downloadable on this forum. Only caveat is that it is annual-only data. But there's a ton of return series there, many of which goes back to the early 1920's or even before that...

Cheers.

Thank you for the pointer. I might use that data sometime.

There are two questions in my mind:

The first is a matter of function. Does the machine operate as intended? Does the feedback mechanism stabilize the otherwise unstable system? That is why I applied the 2000+ historical data. It confirmed to me that the method does control the range of outcomes much better than an open-loop system.

The second question is a matter of tuning. How should I set the dials to get satisfactory performance from the machine? This is where MC was more useful. I could try different adjustments and see how it affected the probability and range of spending cuts balanced with size of leftover funds. I settled on some settings that seemed good enough for me. At this point, the case for more back-testing is a matter of tuning, rather than function, and it depends on what we are after. It seems to try to answer the question, how should the dials be set to handle the worst situation that's ever happened? Should I use a rate of 6%, 4%, 2%? I am not keen on trying to protect against the absolute worst case. Otherwise I'd be spending my retirement building a bunker in the backyard. The difficulty here is that expectations are different for everyone. So I'm not sure what I, or anyone else, would gain from more back-testing.

[squonk2]

I think this plan is great! But after a few years (maybe even months or weeks?) you will change the plan anyways, so I recommend that you don't get attached to it.

Lol, I'm with you.

[Ben Mathew]

Much of this can be done by amortizing the portfolio: see amortization based withdrawal (ABW). You can choose upward sloping, downward sloping, or flat withdrawal profiles by choosing withdrawal growth g>0, g<0, or g=0 respectively. You can schedule extra expenses in the early part of retirement to fund the "go-go" years. To capture the perpetual life assumption, you can input a high max age.

2. Use the withdrawal Plan to estimate the Minimum sustainable portfolio value year-by-year. The curve may slope downward from the initial value then flatten out in later years, like a pinewood derby track. The starting point is the portfolio value at retirement. The last year’s value is that year’s withdrawal Plan divided by a conservative nominal rate of return, perhaps somewhere between 4% and 6% depending on one’s asset allocation and tolerance for income uncertainty.

I don't see how this part would work. Dividing a level withdrawal amount by the rate of return will give the portfolio balance required to fund those withdrawals at a constant level forever (perpetual annuity). Even though you are aiming for perpetual withdrawals, those withdrawals are not constant, and so that formula would not apply. If the planned withdrawals are declining at a constant rate, you can adjust the formula for it. If the withdrawals are more complex, you can use a spreadsheet to do the amortization (using the PMT function in Excel). The wiki on ABW formulas may be helpful.

[dcabler]

Welcome to the forum!

It has some similarities to other withdrawal methods I looked at over the years (I'm also an engineer).
- Gummy's sensible withdrawals is one where it's a mashup of SWR and a way to get a bonus in good years
- There's less similarity to Hebeler's autopilot mode except he also has something of a feedback mechanism (he's a former Boeing engineer)


Other commentary
- I personally think that SWR is such a bad idea in practice that I don't even consider comparing it to anything anymore.
-You ran this with monte carlo, but most people who run monte carlo don't take into account serial correlations of the sequence of returns. Did you? Alternatively, did you run this also with historical sequences and how did that look, especially in periods with sustained low performance?
- Some graphs with results would be cool to see.
- Some of your requirements seem to be at odds with others. #2 and #4 for example

Have you looked at other methods available on this forum such as VPW, ABW & TPAW?
ABW, in particular, also lets you create a profile of spending. And either of these methods could be modified to, essentially, satisfy "live forever" requirement by either using an extremely large number of remaining years or by using life expectancy tables for remaining number of years. There's also McClung's book and ERN's series on withdrawals...

Cheers.

Oh, serial correlation... I built serial correlation into my model. But when it came down to quantifying it I ran into trouble. Looking at historical data, serial correlation was, at least for me, very hard to pin down at any interval--sometimes positive, sometimes negative, often generally zero. I just could not come up with a confident expectation of what it will be like in the future, especially as we diversify beyond traditional assets. So I ended up just setting the constant to zero. And I figured if I run enough MC trials, at least a few of them will happen to appear with serial correlation.

You might want to take a look at how Jim Otar approached serial correlation in his Monte Carlo engine.

Cheers.

[squonk2]

You might want to take a look at how Jim Otar approached serial correlation in his Monte Carlo engine.

Cheers.

I'd like that. Where can I find it?

[squonk2]

I don't see how this part would work. Dividing a level withdrawal amount by the rate of return will give the portfolio balance required to fund those withdrawals at a constant level forever (perpetual annuity). Even though you are aiming for perpetual withdrawals, those withdrawals are not constant, and so that formula would not apply. If the planned withdrawals are declining at a constant rate, you can adjust the formula for it. If the withdrawals are more complex, you can use a spreadsheet to do the amortization (using the PMT function in Excel). The wiki on ABW formulas may be helpful.

Right, good question. The simple calculation I gave is just for the last year portfolio value. It is not used to calculate the rest of the curve. It's just a crazy estimate of something to aim for to set the direction I want to go. It's practically a certainty that everything will change between now and then, so I see no point in trying to be very accurate about it.

The formula I actually used for calculating the portfolio Target for years prior is this:

Minimum(year) = Minimum(year + 1) / (1 + Return) + Plan(year)

and

Target(year) = Minimum(year) * 1.33

I had to make sure that the reverse-extrapolated first-year portfolio Target was no greater than my current portfolio value. That's how I determined I could retire now.

----------------------------------------------------------------------------------------------------------
Alice: "Would you tell me, please, which way I ought to walk from here?"
Cheshire Cat: "That depends a good deal on where you want to get to"
Alice: "I don't much care where."
Cheshire Cat: "Then it doesn't much matter which way you walk."
-Lewis Carroll

[dcabler]

You might want to take a look at how Jim Otar approached serial correlation in his Monte Carlo engine.

Cheers.

I'd like that. Where can I find it?

Start here.
http://www.retirementoptimizer.com/

It's in this whitepaper:
http://www.retirementoptimizer.com/arti ... rticle.pdf

Cheers.

[dcabler]

If you're AA is more or less conventional, then there is significantly more data available on the SIMBA spreadsheet, which is downloadable on this forum. Only caveat is that it is annual-only data. But there's a ton of return series there, many of which goes back to the early 1920's or even before that...

Cheers.

Thank you for the pointer. I might use that data sometime.

There are two questions in my mind:

The first is a matter of function. Does the machine operate as intended? Does the feedback mechanism stabilize the otherwise unstable system? That is why I applied the 2000+ historical data. It confirmed to me that the method does control the range of outcomes much better than an open-loop system.

The second question is a matter of tuning. How should I set the dials to get satisfactory performance from the machine? This is where MC was more useful. I could try different adjustments and see how it affected the probability and range of spending cuts balanced with size of leftover funds. I settled on some settings that seemed good enough for me. At this point, the case for more back-testing is a matter of tuning, rather than function, and it depends on what we are after. It seems to try to answer the question, how should the dials be set to handle the worst situation that's ever happened? Should I use a rate of 6%, 4%, 2%? I am not keen on trying to protect against the absolute worst case. Otherwise I'd be spending my retirement building a bunker in the backyard. The difficulty here is that expectations are different for everyone. So I'm not sure what I, or anyone else, would gain from more back-testing.

It's not about trying to protect against the worst case or building a bunker. Goodness knows a number of people on this forum ignore the Great Depression or even a 1966-69 starting year. At least in the engineering field I just retired from, verification included both directed tests and randomized testing. It's more about gaining insight.

Since you don't currently have serial correlation (aka momentum) enabled, this is what you're going to likely miss out on by not seeing long, bad sequences or long good sequences of returns. You haven't mentioned it but perhaps you already did this separately with made-up sequences, which is fine. But I can't imagine only relying on monte carlo without also having looked at some real sequences.

Cheers.

[dcabler]

Much of this can be done by amortizing the portfolio: see amortization based withdrawal (ABW). You can choose upward sloping, downward sloping, or flat withdrawal profiles by choosing withdrawal growth g>0, g<0, or g=0 respectively. You can schedule extra expenses in the early part of retirement to fund the "go-go" years. To capture the perpetual life assumption, you can input a high max age.

2. Use the withdrawal Plan to estimate the Minimum sustainable portfolio value year-by-year. The curve may slope downward from the initial value then flatten out in later years, like a pinewood derby track. The starting point is the portfolio value at retirement. The last year’s value is that year’s withdrawal Plan divided by a conservative nominal rate of return, perhaps somewhere between 4% and 6% depending on one’s asset allocation and tolerance for income uncertainty.

I don't see how this part would work. Dividing a level withdrawal amount by the rate of return will give the portfolio balance required to fund those withdrawals at a constant level forever (perpetual annuity). Even though you are aiming for perpetual withdrawals, those withdrawals are not constant, and so that formula would not apply. If the planned withdrawals are declining at a constant rate, you can adjust the formula for it. If the withdrawals are more complex, you can use a spreadsheet to do the amortization (using the PMT function in Excel). The wiki on ABW formulas may be helpful.

One other thing that people often forget is that you don't have to target a final year portfolio value of $0. I have one friend who has set his on something like a perpetual portfolio. Amortization, but the future is always N years away with a certain terminal value....

Cheers.

[squonk2]

I really appreciate your suggestions. I love this forum. Bogleheads are awesome!

[squonk2]

Update:

I added the missing formulae in the original post.

I, and perhaps others, would like to see a comparison of how this method behaves during good times and bad versus some other methods such as constant real withdrawal or percentage of portfolio. I have some ideas. I'll see what I can do.

[squonk2]

I added to the original post a backtest comparative analysis vs. constant dollar and constant percentage withdrawal methods. Scroll to the bottom of the (getting sorta long) post.

[kjmh53]

All this is "Interesting", but when you are 90-95% invested in IRAs and 401K's for retirement you are subject to the required RMD percentages. It's certainly important to be aware of your drawdown, but in our case, most of this is out of our control.

Still good to keep an eye on.

[marcopolo]

All this is "Interesting", but when you are 90-95% invested in IRAs and 401K's for retirement you are subject to the required RMD percentages. It's certainly important to be aware of your drawdown, but in our case, most of this is out of our control.

Still good to keep an eye on.

Just because you have to take RMDs out of your IRA/401K doesn't mean you need to withdraw it from your portfolio and spend it.

[kjmh53]

"Just because you have to take RMDs out of your IRA/401K doesn't mean you need to withdraw it from your portfolio and spend it."

True Statement. I retired earlier this year and have been doing a "Test Drive" using RMD "Based" withdrawals. I have made it a point to put a fixed percentage in Savings. The goal was to come as close to Full Income replacement as possible and keep an eye on the "tax" impacts. The Guardrails approach appears to have some advantages "if" you can be flexible with your spending - hence the automatic savings.

Thanks for the input....