USA Today article on SWR

[ERD50]

Download link:
https://www.dropbox.com/s/cwprtn6y8ouyajj/SPY_Withdraw_by_Guyton_rules.xls

I'll eschew editorial comment right now. Other than to note that it was interesting to see the action of the Capital Preservation Rule and the Prosperity Rule.

Dowload it. Play with the parameters as you desire.

I welcome any comments about errors that may be in the sheet.

Very nicely done, THANKS! I just have a minute to look at it, gotta run, but part of my point is that (if I read it correctly - not a good speed-reader), even with these adjustments, the portfolio drops to ~ 1/2 by year six!

Maybe I'm reading too much into the comments on this forum, but I get the impression that many here think they will avoid a dip like this by adjusting their spending. That is a scary dip. And spending does drop down to 2.4%, a big drop.

Will play with it more later.

-ERD50

 

[BigNick]

That article has a calculation error in only the second paragraph:

If you crave safety, a one-year Treasury bill yields just 0.1%, or $21 a year. [on $250,000 capital]

Umm, no. That's $250 a year or $21 a month. I don't think I'll take financial advice from someone who's so sloppy with their calculations, thanks.

 

[pb4uski]

That article has a calculation error in only the second paragraph:

Umm, no. That's $250 a year or $21 a month. I don't think I'll take financial advice from someone who's so sloppy with their calculations, thanks.

I think it is more properly a proofreading error rather than a calculation error. Your conclusion seems to me to be a bit extreme for a proofing error where year should be month.

 

[photoguy]

Just wondering -- what's the issue with Monte Carlo sampling? My understanding (not a statistics expert) is that as long as the probability distribution used for sampling correctly fits the historical data, then it's probably a better way of simulating returns than just using strictly historical data.

This is a HUGE caveat. The MC probability distribution will be based on some "model" of how the market works and it is guaranteed to be wrong. Given the huge amount of debate about simple things like whether a value premium exists or if the higher historical US stock returns are due to survivorship bias or structural advantages, I wouldn't trust an MC approach much.

 

[Independent]

One of the plusses with financial debates about strategies, such as Covered Calls and Indexed Universal Life, is that once you create a spreadsheet that models a strategy using actual historical data, it's pretty easy to revise it to create a spreadsheet for a different strategy.

Since I already have developed several, and since I already have a spreadsheet that I use to guide my own portfolio withdrawals, it only took a couple of hours to create a spreadsheet to backtest Guyton-Klinger rules using the S&P500 historical returns.

It also shows a side-by-side comparison of the standard SWR withdrawal rules.

We don't care about the good times -- what we all worry about is what happens in the bad times. So 1973 is a good year to start the backtest from.

It takes a few parameters. The SWR for the G-K rules, the SWR for standard rules, and the cap for the CPI adjustment.

Download link:
https://www.dropbox.com/s/cwprtn6y8ouyajj/SPY_Withdraw_by_Guyton_rules.xls

I'll eschew editorial comment right now. Other than to note that it was interesting to see the action of the Capital Preservation Rule and the Prosperity Rule.

Dowload it. Play with the parameters as you desire.

I welcome any comments about errors that may be in the sheet.

Thanks. I worked on this today, too. The rules are so complex that I had trouble figuring out what order to apply them. The only way to communicate is to post a spreadsheet complete with formulas.

I've got one question, you mentioned that it was interesting to see the action of the Prosperity Rule. I'm having trouble finding that.

In the version I downloaded (5.5% initial withdrawal) the PR flags as a "yes" in 1996. Both the other rules are "n". So I expected to see column E pick up column Q instead of column R (i.e. $6,143 instead of $5,584).

Any thoughts?

 

[urn2bfree]

One of the plusses with financial debates about strategies, such as Covered Calls and Indexed Universal Life, is that once you create a spreadsheet that models a strategy using actual historical data, it's pretty easy to revise it to create a spreadsheet for a different strategy.

Since I already have developed several, and since I already have a spreadsheet that I use to guide my own portfolio withdrawals, it only took a couple of hours to create a spreadsheet to backtest Guyton-Klinger rules using the S&P500 historical returns.

It also shows a side-by-side comparison of the standard SWR withdrawal rules.

We don't care about the good times -- what we all worry about is what happens in the bad times. So 1973 is a good year to start the backtest from.

It takes a few parameters. The SWR for the G-K rules, the SWR for standard rules, and the cap for the CPI adjustment.

Download link:
https://www.dropbox.com/s/cwprtn6y8ouyajj/SPY_Withdraw_by_Guyton_rules.xls

I'll eschew editorial comment right now. Other than to note that it was interesting to see the action of the Capital Preservation Rule and the Prosperity Rule.

Dowload it. Play with the parameters as you desire.

I welcome any comments about errors that may be in the sheet.

Very interesting, Thanks!
It looks like the rules allow for more certain survival than the standard, but most years taking about 40% less than traditionally planned. In this very bad case scenario.

 

[ERD50]

This is a HUGE caveat. The MC probability distribution will be based on some "model" of how the market works and it is guaranteed to be wrong. ...

Agreed. There is an interaction between things like inflation, bond/CD returns, and the stock market. There have been patterns of boom/bust. We have actual data - how do I know if a model replicates those patterns/interactions? And why model it when we have - (redundantly, for effect) - actual data?

-ERD50

 

[ERD50]

It takes a few parameters. The SWR for the G-K rules, the SWR for standard rules, and the cap for the CPI adjustment.

Download link:
https://www.dropbox.com/s/cwprtn6y8ouyajj/SPY_Withdraw_by_Guyton_rules.xls

....

I welcome any comments about errors that may be in the sheet.

I have not checked thru formulas, but a couple Q:

1) Why the 6% CPI cap on 'standard WR'? Firecalc does not do that, AFAIK.

2) I dropped the 'standard WR' to 3.15% so the portfolio lasts 30 years. From year 7 on, The 'standard' method allows a higher WR, and it levels out at > 150% of the G-K rule spend rate for most of the remaining 30 years.

Not a desirable outcome for me. Others may feel differently.

-ERD50

 

[rayvt]

So I expected to see column E pick up column Q instead of column R (i.e. $6,143 instead of $5,584).

Right you are! I just knew there was an error somewhere, but I couldn't find it for the live of me. Thanks.

Fixed. Plus fixed a couple other errors. Plus enhanced it so that the start date is now a parameter that you can enter. Plus extended the data back to 1960.

New version has been uploaded. Same url: https://www.dropbox.com/s/cwprtn6y8ouyajj/SPY_Withdraw_by_Guyton_rules.xls

Summary stats, $100K, Jan 1973 start, Jan 2013 end, SWR's of 5.5% for G-K, 4.0% for standard, CPI-W 6% cap
1973 is a great start for disaster testing, because you get immediately hit with 15% loss and 26% loss in the first two years.

                 G-K             Standard
Final Value     $450,244        $642,742
Lowest Val       $55,431         $57,092
Total Draws     $622,355        $408,647
 
Last yr draw    $28,058         $16,886
Last % draw        6.2%            2.6%
Highest Draw    $30,623         $16,886
Lowest Draw      $4,473          $4,000
 
High % Draw     8.8%            7.7%
Low % Draw      3.1%            1.6%

 

[NanoSour]

Great job on this RAYVT. Appreciate the work.

 

[Independent]

Right you are! I just knew there was an error somewhere, but I couldn't find it for the live of me. Thanks.

Fixed. Plus fixed a couple other errors. Plus enhanced it so that the start date is now a parameter that you can enter. Plus extended the data back to 1960.

New version has been uploaded. Same url: https://www.dropbox.com/s/cwprtn6y8ouyajj/SPY_Withdraw_by_Guyton_rules.xls

Thanks, you've really done a lot in a short period.

I can't help posting a format that I like. I tried a number of different initial withdrawals. I also used $1 million initial portfolio to make the numbers look more typical. Notice that I removed the 6% inflation cap for the "standard" SWR run.

These are CPI adjusted values for:
Initial Withdrawal,
Lowest Withdrawal,
Average Withdrawals for 10 year periods,
Ending Portfolio at 10 Year Intervals

Initial	40,000	40,000	45,000	50,000	55,000
Minimum	40,000	18,921	21,185	21,185	23,304
---	---	---	---	---	---
A 1-10	40,000	27,330	30,999	33,327	36,660
A 11-20	40,000	25,361	26,209	26,209	26,653
A 21-30	40,000	49,861	46,822	50,754	46,481
A 31-40	40,069	68,157	68,084	63,428	61,055
---	---	---	---	---	---
P-10	384,555	525,886	486,028	460,441	424,045
P-20	432,012	1,143,270	1,006,801	928,373	805,702
P-30	356,943	1,598,370	1,371,427	1,184,284	998,222
P-40	67,730	1,658,395	1,316,499	1,088,980	824,464

The message here is that for this severe scenario, the G-K rules are very aggressive about dropping the withdrawals, getting into the range of half the standard SWR 4%. They leave quite a bit on the table at the end of 30 or 40 years.

 

[siamond]

Rayvt, I finally got enough time to look at your spreadsheet. I'll echo the compliments, this is amazing work and extremely useful. Many thanks for doing that.

I tried to play around, see the outcome if I start in 1961, 1965, 1970, 1975, etc. And I paid close attention to the delta on the absolute monthly withdrawal between the two models. Most of the time, it works beautifully, the delta is always positive and would often be a really good life improvement (that is, if more money makes you happier!). Now for sure, the capital at the end is typically lower for the G-K model, but that's to be expected and perfectly ok in my mind as long as it stays over the inflation adjusted starting point.

It actually almost works too beautifully in numerous cases, where I'd be uncomfortable spending that much money instead of savings some for the future.

I do agree though that some scenarios are worrying because the monthly withdrawal drops way too much for comfort.

This all re-inforces my gut feeling about the model, it should include a floor and a ceiling on the withdrawals. The floor being your minimum spend if you get real thrifty, the ceiling being the maximum spend if you want to enjoy life but not get overboard. Both being inflation-adjusted, of course.

Once again, many thanks or your great work.

 

[Fred123]

This is a HUGE caveat. The MC probability distribution will be based on some "model" of how the market works and it is guaranteed to be wrong. Given the huge amount of debate about simple things like whether a value premium exists or if the higher historical US stock returns are due to survivorship bias or structural advantages, I wouldn't trust an MC approach much.

Well, every retirement calculator uses some sort of model (implicit or explicit) to attempt to predict the future from past data, and, of course, every model won't exactly predict the future. MC calculators make no assumption about how the market "works." They make assumptions about how the returns are distributed, which is entirely different. And the distribution assumption can be tested with various statistical tests.

With FIRECALC, the model is that the only possibilities for time series of future returns are those that occurred in the past. While this might be a good starting point, it certainly seems unlikely that any of these *exact* time series will replay in the future. It seems to me that a better model would add some randomness to these past time series.

Another problem with strictly deterministic calculators like FIRECALC is that they are biased towards older data. Say, for instance, you're looking at the outcomes over a 35 year period. Returns from, say, 1950 are included in 35 time series while data from the 2008 bust is only included in 5 of the FIRECALC time series. And, even worse, none of the time series include 2008 data at the beginning of the 35 year period, where the returns are more important in determining the ultimate chances of success. This isn't true of calculators that use MC techniques.

 

[explanade]

Thanks, you've really done a lot in a short period.

I can't help posting a format that I like. I tried a number of different initial withdrawals. I also used $1 million initial portfolio to make the numbers look more typical. Notice that I removed the 6% inflation cap for the "standard" SWR run.

These are CPI adjusted values for:
Initial Withdrawal,
Lowest Withdrawal,
Average Withdrawals for 10 year periods,
Ending Portfolio at 10 Year Intervals

Initial	40,000	40,000	45,000	50,000	55,000
Minimum	40,000	18,921	21,185	21,185	23,304
---	---	---	---	---	---
A 1-10	40,000	27,330	30,999	33,327	36,660
A 11-20	40,000	25,361	26,209	26,209	26,653
A 21-30	40,000	49,861	46,822	50,754	46,481
A 31-40	40,069	68,157	68,084	63,428	61,055
---	---	---	---	---	---
P-10	384,555	525,886	486,028	460,441	424,045
P-20	432,012	1,143,270	1,006,801	928,373	805,702
P-30	356,943	1,598,370	1,371,427	1,184,284	998,222
P-40	67,730	1,658,395	1,316,499	1,088,980	824,464

The message here is that for this severe scenario, the G-K rules are very aggressive about dropping the withdrawals, getting into the range of half the standard SWR 4%. They leave quite a bit on the table at the end of 30 or 40 years.

I don't understand this table. The far right column starts with a $55k withdrawal or 5.5% but ends with a higher balance than the first column, which is a $40k withdrawal?

 

[Independent]

I don't understand this table. The far right column starts with a $55k withdrawal or 5.5% but ends with a higher balance than the first column, which is a $40k withdrawal?

I can give a very rough explanation.

It may start at $55k, but it rapidly drops, bottoming out at $23,304 in the 11th through 13th years.
So the $55k portfolio grows faster, you can see the difference at the end of 10 and 20 years.
After that, the $55k withdrawals grow, finally exceeding $40k in the 24th year. But, by then the investment returns on the earlier differences dominate the remaining years.

That's awfully rough. All I can say beyond that is to look at the detail in the spreadsheet. (Note that for the level SWR I removed the spreadsheet 6% cap on inflation, and I've stated all the numbers in CPI-adjusted amounts.)

 

[photoguy]

Well, every retirement calculator uses some sort of model (implicit or explicit) to attempt to predict the future from past data, and, of course, every model won't exactly predict the future.

I see the point that you're making but I wouldn't say that an approach like Firecalc (which replays historical data) is a model. In my mind, a model has to relate inputs to outputs and this doesn't really apply to firecalc.

MC calculators make no assumption about how the market "works." They make assumptions about how the returns are distributed, which is entirely different.

I was a bit loose with my wording. I should have written model of returns and not market. But to have a useful model of returns it has to cover at some basic level of detail how the market works.

And the distribution assumption can be tested with various statistical tests.

Yes it can be tested with statistical tests. But one of two things is going to happen: (1) the researchers are going to fit the parameters of their model to maximize the chance of getting the observed data. In which case you are facing the same limitations as a data driven approach like Firecalc (only a few data points drive the simulation) or (2) the researchers are going to put in assumptions which diverge from history and can be biased.

With FIRECALC, the model is that the only possibilities for time series of future returns are those that occurred in the past. While this might be a good starting point, it certainly seems unlikely that any of these *exact* time series will replay in the future. It seems to me that a better model would add some randomness to these past time series.

I totally agree with you that Firecalc is limited (quite severely in many ways) and although it is a good starting point, one shouldn't read too much into it. However, the addition of randomness via MC simulation is not free. It comes at the cost of using an artificial model which could be substantially biased.

 

[explanade]

But is it realistic to have the withdrawals drop more than 50% like that?

Or is that attempting to simulate a severe market downturn?

 

[rayvt]

I plotted the withdrawals & portfolio values for each method on a graph and noticed some things.

First of all, it's only been recently that people have started talking about SWR down in the 3%-3.5% range. Up until just a few years ago the number was 4%. So if you started withdrawals in 1973 you'd have used 4%. There would have been ro reason to consider 3.5% or 3.15%.

Second off,
Regardless of withdrawal amounts, the portfolio got whacked hard in the next 2 years -- 15% loss then 27% loss. The portfolio value got cut almost in half. PANIC TIME! I can't see anybody going into the third year and even maintaining the same draw, let alone making a CPI increase.

De facto, people will reduce their draw. The question is, how will they decide the new amount? Standard SWR rules say nothing about that, so you're operating on gut and guess. Now you're in uncharted waters and flailing about and panicing.

With G-K rules, there is a pre-defined strategy for reducing the draw, and a pre-defined strategy for bumping it back up when the storm abates.

Third thought,
If you don't put a cap on the maximum CPI adjustment, then the combination of low returns and high inflation (Hello, Jimmy Carter!) will kill the portfolio. I think that's why Guyton set a maximum on the CPI adjustment. You lose purchashing power in the short run but preserve your portfolio (and the ability to maintain withdrawals) in the long run.

 

[ERD50]

I plotted the withdrawals & portfolio values for each method on a graph and noticed some things.

First of all, it's only been recently that people have started talking about SWR down in the 3%-3.5% range. Up until just a few years ago the number was 4%. So if you started withdrawals in 1973 you'd have used 4%. There would have been ro reason to consider 3.5% or 3.15%.

I can't agree with this at all - unless by "people" you mean the journalists using generic 'rules of thumb' like 4% and 'your retirement budget will be 80% of your working income'. Many of us on the forum have discussed, and are (trying to? aiming for?) implementing WR in the 3-3.5% range.

OK, FIRECALC wasn't around in 1973 but using it today will show you that 4% won't survive in 5% of the historic cases. You need to get down to 3.59% for 100%.

Second off,
Regardless of withdrawal amounts, the portfolio got whacked hard in the next 2 years -- 15% loss then 27% loss. The portfolio value got cut almost in half. PANIC TIME!

True, the hit is there even with zero WD.

I can't see anybody going into the third year and even maintaining the same draw, let alone making a CPI increase.

De facto, people will reduce their draw. The question is, how will they decide the new amount? Standard SWR rules say nothing about that, so you're operating on gut and guess. Now you're in uncharted waters and flailing about and panicing.

I'm not sure that starting with a low WR and maintaining it is any more a gut/guess than starting high and lowering it? We are looking at each against a historic drop. Sure, the math is unarguable - if you do something like % of portfolio, you never deplete your portfolio. But is it a meaningful distinction if your spending drops near $0, or to poverty levels for a long spell?

With G-K rules, there is a pre-defined strategy for reducing the draw, and a pre-defined strategy for bumping it back up when the storm abates.

And that is the discussion - how would these rules perform, how would I need to adjust my spending, is this preferable to low and steady? Likely an individual decision, one factor being flexibility to lower your WDs.

Third thought,
If you don't put a cap on the maximum CPI adjustment, then the combination of low returns and high inflation (Hello, Jimmy Carter!) will kill the portfolio. I think that's why Guyton set a maximum on the CPI adjustment. You lose purchashing power in the short run but preserve your portfolio (and the ability to maintain withdrawals) in the long run.

But if we cap the 'standard' constant spending model (CSM)- it makes the G-K look better in comparison as far as the spending drops. You created a drop in spending in the CSM that isn't there. The point of FIRECALC is that you can see if you can weather these storms and still maintain your standard of living. That is the comparison, not some artificial 6% cap.

I hope I don't sound negative on G-K, I'm not. I do want to investigate it and understand it as an option. It might look attractive for some. But so far, those kinds of drops in spending would be unsettling for me. But my understanding may not be all that complete yet, or maybe there are some tweaks I would make for myself?

I'll try to follow up later on some alternate ideas with FIRECALC.


-ERD50

 

[Independent]

But is it realistic to have the withdrawals drop more than 50% like that?

Or is that attempting to simulate a severe market downturn?

I'm not sure who you're asking.

It's "realistic" in the sense that rayvt's calculations use the actual S&P 500 returns and the actual CPI changes that people retiring in 1973 experienced.

Using those numbers, the rules in the G-K paper result in withdrawals dropping almost 50%.

I'm sure that some people would say it's unrealistic to think that such a sequence of events could happen again, but I'm not one of those people.

 

[Independent]

Third thought,
If you don't put a cap on the maximum CPI adjustment, then the combination of low returns and high inflation (Hello, Jimmy Carter!) will kill the portfolio. I think that's why Guyton set a maximum on the CPI adjustment. You lose purchashing power in the short run but preserve your portfolio (and the ability to maintain withdrawals) in the long run.

I think it makes more sense to look at purchasing power than nominal dollars.

The Guyton approach gives different real spending adjustments if the nominal return is -5.0% with an inflation rate of 8%, than a nominal return of -10.3% with an inflation rate of 2%. But those two scenarios have the same real returns. In the first, he cuts real withdrawals by 1-1/1.08 = 7.4%. In the second, he cuts real withdrawals by 2.0%.

If I were developing a set of rules, I'd want to react the same to those two cases.

 

[Willers]

Right you are! I just knew there was an error somewhere, but I couldn't find it for the live of me. Thanks.

Fixed. Plus fixed a couple other errors. Plus enhanced it so that the start date is now a parameter that you can enter. Plus extended the data back to 1960.

New version has been uploaded. Same url: https://www.dropbox.com/s/cwprtn6y8ouyajj/SPY_Withdraw_by_Guyton_rules.xls

[/CODE]

Thank you very much for your work on this.

 

[scrinch]

However, the addition of randomness via MC simulation is not free. It comes at the cost of using an artificial model which could be substantially biased.

There's no such thing as an unbiased future model. Applying what happened in 1950 or 1973 or 2010 to a model of the future is biased too. Just because it happened once doesn't mean it will happen again just like that. So bias the future the way you think it might look. That's what modeling is all about. If you're an optimist, build your model that way. If you're a pessimist, build it that way. Make it reflect the future with all the features and uncertainty you think it might hold. That's what a model is for...to apply some math to your view of the future.

 

[siamond]

New version has been uploaded. Same url: https://www.dropbox.com/s/cwprtn6y8ouyajj/SPY_Withdraw_by_Guyton_rules.xls

So I developed my own version...

In the main tab, I introduced various comparisons to constant dollars (i.e. initial withdrawal amounts with inflation adjustment). This led me to introduce a min-CPI factor, as the regression in 2011 was making my comparisons looks weird, plus it does seem logical to me to reduce withdrawals in a period of recession (I mean, at least by the deflation factor).

Then I created a "main2" tab, where I modeled new additional rules designed to address what I perceive as unrealistic effects (e.g. shrink or expand your spend beyond reasonable). In other words, I introduced a concept of a floor (bare minimum you can live on) and a ceiling (spend you shouldn't exceed for your own sanity!). I dubbed up the Sanity Rule(s)!

Then I played around with various starting points, and it became clear that the 5.50% assumption is too dangerous to start with, once you apply such sanity rules... Starting in the 70s would bring your portfolio dangerously low, staying on the floor makes it hard to recover. Starting in the 80s is also very interesting, then the ceiling applies numerous times.

Another thing I'd like to point out is that the portfolio evolution is currently calibrated on SPX only, and well, that's an 11% return in the period of interest, which seem a tad optimistic when pondering about the future... I wanted to play around with a more realistic stock/bond split, but the bond return values in Sheet2 do not seem quite correct?

Anyhoo, here is my modified Excel sheet. Rayvt, I use orange marks to show my various changes. If you would be kind enough to general the logic of my changes, it would be appreciated, I think I understood your logic, but maybe I missed a subtlety here and there.

https://docs.google.com/file/d/0B0svRQGBG_eaTWZjSTFZSERkanc/edit?usp=sharing

Overall, personally, I'm rather convinced that such rules, although complicated (Excel to the rescue!), are very well worth using. The plain 4% rule just gets too disconnected from the reality of your portfolio in many cases, otherwise.

 

[rayvt]

ERD50: No, the 4% rule has been "The Number" in the broad financial community ever since Bengen's 1994 paper and the Trinity study. William Bengen - Wikipedia, the free encyclopedia While it is true that not everybody agreed with it, the broad community did. It's only in the last few years that the general consensus is starting to think that 4% is too high.

IMHO, the right SWR to start with when doing initial backtests is what was accepted at the time (notwithstanding that 1972 is 20 years before Bengen's paper). AFAIK, Guyton started from that 4% number when he invented his Withdrawal Decision Rules.
It's tricky to avoid the hindsight bias fallacy, and IMHO -- for the purposes of comparing these two withdrawal strategies -- using a SWR of 3%-3.5% is hindsight bias.

OTOH, capping the CPI adjustment to 6% is also hindsight bias! The only time inflation has been much above 6% is one period of a few years centered around 1977.

So IMHO the proper comparison of plain SWR vs. G-K is 4% and 5.5%, with no inflation caps.

::sigh:: Alas, my OCD is starting to show. I just put together this spreadsheet to try to get an understanding of what the actual differences there would be between these two withdrawal methods, not to get bogged down in arcane details. And, as you indicated, to see how the G-K rules performed on historical data.

siamond: Yes, I had similar thoughts about some sort of Sanity Rule ceiling. Once you survive the 1970's and pull out of the 1980's pothole, the G-K withdrawal grows sky high, like 50% more than the inflation-only growth of the standard SWR method. There's a lot of refinements you could do, but: a) it makes the model much more complex, and b) it doesn't matter. What we really want to know is "does the portfolio have a good chance to survive?" The added complexity doesn't help answer that -- the simple original rules give us the answer to that.

As far a putting a "bare minimum" floor, I'm against it. It's all well and good to say, "I insist on taking at least $X", but reality says that when the portfolio gets to zero, you won't be making ANY withdrawals.

Heh, using $1,000,000 starting value. Yes, that's more likely, but it doesn't matter. Using $10K or $100K is the same, just add another zero. In another thread where I used a $1M initial value some people got all freaked out at seeing final values in the tens of millions, and annual draws in the half-million range. The freaking out interfered with the discussion, so I keep the starting numbers low so that the finals don't get that big.

BTW, in playing around with the parameters, I discovered that the SWR (for standard SWR method ) can be extremely sensitive at the cusp. A difference of 0.2% (4.2% to 4.4%) means the difference in the final value of $2.2M vs. $88K-headed-for-the-cliff. Pick a too-high SWR and you'll be eating Alpo when you're 92.

--------------
I didn't stress this, but also in the sheet is data for the S&P returns where you are in/out based on the 10 month SMA, like Faber's QTAA method. Using that instead of buy&hold is a more substantial improvement than fiddling with the decimal points of the SWR.

==================
Anyway ...... I uploaded a new version. This one lets you individually set the CPI caps, and it also has charts of the annual draws and portfolio values. Same URL: https://www.dropbox.com/s/cwprtn6y8ouyajj/SPY_Withdraw_by_Guyton_rules.xls This will be my last update unless somebody finds an error that I need to correct.