The number one suggestion from readers for future projects in our Safe Withdrawal Rate Series: look into dynamic withdrawal rates, especially the Guyton-Klinger (GK) withdrawal rate rules. The interest in dynamic rate rules is understandable. Setting one initial withdrawal amount and then stubbornly adjusting it for CPI inflation regardless of what the portfolio does over the next 50-60 years seems wrong (despite the extremely simple and beautiful withdrawal rate arithmetic we pointed out last week).
So, here we go, our take on the dynamic withdrawal rates. Jonathan Guyton and William Klinger proposed a dynamic strategy that starts out just like the good old static withdrawal rate strategies, namely, setting one initial withdrawal amount and adjusting it for inflation. However, once the withdrawal rate (expressed as current withdrawal rate divided by the current portfolio value) wanders off too far from the target, the investor makes adjustments. Also, notice that this works both ways: You increase your withdrawals if the portfolio appreciated by a certain amount relative to your withdrawals and you decrease your withdrawals if the portfolio is lagging behind significantly. Think of this as guardrails on a road; you let the observed withdrawal rates wander off in either direction, for a while at least, but the guardrails prevent the withdrawal rate from wandering off too far, see chart below. It’s all pretty intuitive stuff, though, as we will see later, the devil is in the details.
The Wall Street Journal calls this methodology “A Better Way to Tap Your Retirement Savings” because it allows higher (!) withdrawal rates than the traditional 4% rule. As you probably know by now, we’re no fans of the 4% rule and if people claim that we can push the envelope even further by just applying some “magic dynamic” we are very suspicious. Specifically, we believe that the GK methodology has (at least) one flaw and we like to showcase it here.
Guyton-Klinger basics
See a nice summary here and the original paper here. An interesting link with lots of calculations, examples and an Excel Spreadsheet with sample calculations is here. cFIREsim also simulates the GK method! In any case, the Guyton-Klinger method has four major ingredients, of which three are essential and the fourth seems to be there mostly for “cosmetic” reasons:
- Forego the CPI-adjustment in withdrawals when the nominal portfolio return was negative. Even when doing the CPI-adjustment following a positive return, cap it at 6%, which seems somewhat arbitrary to us.
- (Guard Rail 1) If the withdrawal rate (current withdrawal amount divided by current portfolio value) is greater than 1.2 times the initial withdrawal rate then cut the withdrawal amount by 10%.
- (Guard Rail 2) If the withdrawal rate (current withdrawal amount divided by current portfolio value) is smaller than 0.8 times the initial withdrawal rate then increase the withdrawal amount by 10%.
- Some pretty convoluted mumbo-jumbo on the withdrawal mechanics, e.g., which assets to draw down first, a process they call the Portfolio Management Rule. To us, this seems like a slightly infantile description of a portfolio rebalance back to target weights, i.e., draw down the assets with the highest returns first because they are the ones with the largest overweights relative to the target weights. Why not just do a simple rebalance to target weights then? There are only two possibilities: a) There is no gain from their procedure relative to a plain rebalance, then why do it the complicated way? b) There is an advantage relative to a simple rebalance but given the ad-hoc nature of their rules, we would argue that any advantage is likely a fluke. In fact, by GK’s own admission (Table 2 in their paper), their portfolio management rule doesn’t add anything when targeting a 90% probability of success and adds only marginally when targeting a 95% probability of success.
The way we model the dynamic rule is a simplified (decluttered) version of Guyton-Klinger:
- Run simulations at a monthly frequency, rather than annual, to be consistent with our other research on the topic and, of course, for the plain and simple reason that once we are retired we don’t like a whole year worth of withdrawals sitting around in cash every January. We hate leaving money on the table, as you may know from our post on emergency funds.
- Since we don’t have all the different equity asset class returns going back to 1871 we simply assume that there is one single equity index (U.S. Large Cap) and one single bond asset (10-year Benchmark U.S. Treasury Bond) as in our previous research, again consistent with our earlier research based on a simple Stock-Bond portfolio
- We discard GK’s convoluted portfolio management rule. We have only two assets (stocks and bonds) and simply assume that the portfolio is rebalanced back to the target weights every month. It’s simpler to model and calculate in our number-crunching software: a simple matrix algebra operation, i.e., we multiply the Tx2 matrix of stock/bond returns with a 2×1 vector of asset weights. Done! No need to carry around time-varying portfolio weights.
- If the 12-month trailing (real) return was negative, then forego the inflation adjustment, i.e., shrink the real withdrawal by the CPI-rate that month. If the 12-month trailing return was positive, then do the CPI-adjustment. We don’t use the Guyton-Klinger 6% cap on the CPI-adjustment, which seems pretty arbitrary and also causes a big loss of purchasing power in the 1970s.
- If the withdrawal rate (current withdrawal amount divided by current portfolio value) is greater than (1+g) times the initial withdrawal rate then cut the withdrawal amount by x. It’s the same setup as in Guyton-Klinger.
- If the withdrawal rate (current withdrawal amount divided by current portfolio value) is smaller than (1-g) times the initial withdrawal rate then increase the withdrawal amount by x. Again, the same as in Guyton-Klinger.
Our take on Guyton-Klinger captures the main ingredients: the guardrails and a decision rule for making vs. skipping the CPI-adjustments, without the baggage of their complicated and likely useless portfolio management rule.
Results
Let’s start with the good news. The number one reason we like the GK-rule: If done right it’s (almost) impossible to run out of money with the GK rule (in very stark contrast to the non-trivial probabilities of depleting the portfolio under the naive static withdrawal rule, see our previous research). You heard that right! Our simulations show that if we set the initial withdrawal not too crazy high and we use a tight enough guard rail parameter (g=20%) and aggressive enough adjustment parameter (x=10%) then even under adverse market conditions (e.g., the January 1966 retirement cohort) we won’t run out of money. (side note: this requires to do the guardrail adjustments throughout retirement, while GK stop doing the adjustments 15 years before the end of the retirement horizon, in which case you do face the risk of running out of money)
Now for the bad news. We identified one reason to be skeptical, very skeptical, about the Guyton-Klinger rule:
Under Guyton-Klinger you may have to curb your consumption. By a lot more than you think!
Let’s make this more fun and let me first present the GK simulation results in a very deceptive way to make the dynamic GK rules appear much better than they really are. Let’s see who can spot the deception…
Let’s present a 1966 case study, the last time in recent history when the 4% rule failed (though you may remember our 2000-2016 case study, where we showed that the 4% rule also looks pretty shaky for the January 2000 retirement cohort). If Guyton-Klinger can succeed here it will succeed almost anywhere! Throughout, we assume an 80%/20% Stock/Bond portfolio and the same return assumptions as outlined in part 1 of this series. We consider 4 different withdrawal strategies:
- The good old 4% rule: set the initial monthly withdrawal rate to 0.333% (=4% p.a.) and then adjust the withdrawals by CPI regardless of the portfolio performance. This method depletes the portfolio after 28 years.
- Guyton-Klinger with +/-20% guardrails and 10% adjustments and a 4% p.a. initial withdrawal rate
- Same as 2, but with a 5% initial withdrawal rate
- Same as 2, but with a 6% (!) initial withdrawal rate
The time series chart of the real, CPI-adjusted portfolio value (normalized to 100 in January 1966) is below:
Holy Mackerel!!! GK beats the 4% rule and it’s not even close. The GK-4% has surpassed the initial $100 (adjusted for CPI!) after 26 years while the old 4% has gone bankrupt after 28 years. The 5% rule is almost back to normal and the 6% rule is hanging in there pretty well, too. Talking about withdrawal percentages, let’s look at those as well, see picture below:
Amazing! Look at the 5% Guyton-Klinger rule. By construction, it stays between 4% and 6% (=5% times 1+0.2 and 1-0.2, respectively), so it never falls below 4% due to the guardrails. Moreover, it has a higher initial withdrawal and a higher final value! It appears to beat the static 4% withdrawal rate in every dimension we care about. It looks like the occasional 10% cuts in withdrawals haven’t hurt us too much. Amazing! Have we just found a Safe Withdrawal Rate Nirvana? Let’s nominate Guyton and Klinger for the Nobel Prize! Economics or Peace? Heck, both, of course, and in the same year to save them the travel expenses to Stockholm!
But before you open the champagne bottles, let’s bring us all back to planet earth. I just scammed you all! To be sure, the numbers are 100% correct, but the way I presented them was false advertising, even borderline fraudulent.
Where was the deception I mentioned above?
Pay close attention to what I didn’t show you yet! I never showed you the actual inflation-adjusted withdrawal amounts. Who cares about percentages of the portfolio value when the portfolio value is a moving target? I want to know the dollar amounts. It’s called “Show me the money” and not “Show me the percentages,” after all. So, how much in CPI-adjusted dollars can I withdraw under the different rules and, specifically, by how much do I have to curb my consumption during retirement due to the withdrawal cuts once we hit the guard rails? That’s displayed in the chart below:
What a disappointment! That’s where the Guyton-Klinger skeletons are hidden. Sure, when your initial withdrawal rate is 5% you never drop below a 4% withdrawal rate (due to the guardrail), but it’s 4% of a much-depleted portfolio value, not 4% of the initial value. That subtle distinction makes a huge difference. For example, the average withdrawal values for GK under the 4/5/6% initial withdrawal rates are only 2.74%, 3.02%, and 3.22% of the initial portfolio value, respectively. Well, it’s no longer a surprise that we have a higher final value than under the static 4% rule because we withdrew so much less! The advertised 5% withdrawal was only 3.02% withdrawal. What a scam!
Talking about skeletons, here’s more data from the GK horror show: The decline of withdrawals from peak to bottom is a staggering 59%, 66%, and 69%, respectively. Ouch! If you thought that the $1,000,000 portfolio can afford you a $50,000 per year lifestyle using GK, you better plan for a few sub-$20k years and an entire decade (!) of sub-$25k p.a. withdrawals. Suddenly the Guyton-Klinger method doesn’t look so hot anymore.
How is it possible to experience such massive declines in the withdrawals? The GK-rules hide this drop behind the +/-20% guardrails and +/-10% withdrawal adjustments (not to mention the distraction in the form of the asinine “portfolio management rule”) that make it sound like we only suffer relatively minor and temporary decreases in purchasing power. But the 0.2 guardrail is on top of the drop in the portfolio. If the portfolio is down by 50% and you hit the lower guardrail, the drop in the withdrawal is (1-0.5)x(1-0.2)=0.4 = 60% under the initial withdrawal. Hence, the large reduction in withdrawals! Skipping the CPI-adjustment in some of the years also erodes the purchasing power.
The claim that we can afford a higher initial withdrawal rate than under the fixed withdrawal rules is a pretty blatant case of false advertising. In fact, this claim has about the same ring to it as the good old “You can afford that big McMansion” or “You can afford that suped-up brand new car.” A 5% initial withdrawal rate may seem nice in the beginning but reality will catch up eventually. The higher you set the initial withdrawal rate the more of a drop in your consumption pattern you might suffer if the market doesn’t cooperate.
Conclusion
We actually have a lot more material and have to defer all of that to a future post. We’re already past 2,000 words and have only scratched the surface. We prepared another case study (the dreaded January 2000 retirement cohort), more comprehensive historical simulations (including the likelihood of a significant long-lasting drop in purchasing power for different CAPE regimes), and like to show several other smaller flaws in the GK methodology. Probably next week!
To wrap up today’s post, the initial question was: Is the Guyton-Klinger method overrated? False advertising sounds more appropriate. The GK-type rules seem to imply that they can offer higher initial withdrawal rates and better long-term success rates. True, but all that comes at the cost of potentially massive reductions in withdrawals (50%+ below the initial).
Oh well, what did we all expect? The GK rules can’t square the circle by offering higher withdrawal rates and lower failure rates. If we wanted to be sarcastic, we’d point out that GK won’t cure athletes’ feet, either. If you want to use GK yourself, make sure you’re aware of the downside (literally!), i.e., be prepared to curb consumption by 50% if things don’t work out. And that’s not just for a year or two, but potentially for a decade or more! That may be doable if your initial withdrawal is $80k or $100k and there is enough downside cushion. But for the folks with a tighter budget, GK would imply a significant probability of returning to work during early retirement!
Update (2/20/2024):
A reader pointed out in the comments section that you would have gotten different results if you didn’t adjust as rapidly downward (10%) but instead transformed the 10% GK adjustment to a monthly value of 1.10^(1/12)=0.00797=0.797%. Does that change anything in this case study? Would this change make the GK simulations look more appealing? No. It’s like squeezing a balloon. You will indeed soften the drop in withdrawal amounts. You will undoubtedly lower the volatility of monthly withdrawal changes. But because you delay the inevitable and necessary fall in consumption, you also create some new headaches!
In the two charts below, I present you the withdrawal amounts (for a $100 initial portfolio). I use the same scale for easier comparison. True, because of the rapid triggering of the GK rule, the fall is very swift in the baseline simulation (top) and more gradual in the new method (bottom). But I also noticed that because the withdrawal cut under the new method was less reactive, the portfolio depletion was worse, and thus, the withdrawal cut lasted a few years longer. And the eventual recovery in withdrawals was also slower. So, there is no escaping the mathematical realities of Sequence Risk when using this more gradual GK adjustment parameter. The trough consumption level is about the same using the new method. Shockingly, the volatility of retirement withdrawal levels would have been higher (worse) than in the baseline for the 5% and 6% GK rule. That’s because the spending trough lasted longer, and the recovery was more gradual.
The volatility of annualized spending levels over the 30-year retirement, assuming a $1,000,000 initial portfolio (rounded to the closest $100):
- Baseline: $7,600, $9,600, 12,200 for GK 4%, 5%, 6%.
- New method with gradual GK adjustments: $7300, $10,400, $14,400 for GK 4%, 5%, 6%.
Very much appreciate your posts! I do have a question on your statement about the withdrawal amount in this post from the following excerpt:
“But the 0.2 guardrail is on top of the drop in the portfolio. If the portfolio is down by 50% and you hit the lower guardrail, the drop in the withdrawal is (1-0.5)x(1-0.2)=0.4 = 60% under the initial withdrawal.”
Here, you state that in this example if a portfolio drops 50% and hits the lower guard rail, the subsequent withdrawal amount is adjusted by BOTH the new portfolio amount multiplied by (1-0.5), and also the reduction in spending percentage (1-0.2). But it seems that per the reference you provided (http://finalytiq.co.uk/guyton-klinger-sustainable-withdrawal-rules/), it is hard for me to reconcile this with the following statement from that reference:
“The capital preservation rule: If the current withdrawal rate rises above 20% of the initial rate, then current spending is reduced by 10%.”
Here, it only refers to “current spending”, which one could reasonably interpret as the current spending level, which suggests the reduction relative to this level would not depend on reducing the -withdrawal rate- by an amount proportional to the product of the portfolio drop and the reduction in spending percentage together. Indeed, in the original paper, the capital preservation rule mentions only a drop in the current spending level to be acceptable for the current time period, and then uses this same spending level relative to the new portfolio value at the end of the current time period/beginning of the next time period to determine the appropriate course of action at the beginning of the next time period.
To this effect, when the time period length shrinks to zero, it seems that your claim is wholly true, as the effects of the capital preservation rule for small time period intervals rapidly converge on a basis for subsequent withdrawals that weighs both the portfolio value drop and the percentage drop equally. However, for period intervals that are larger (e.g. a year), it would take considerable time to converge on this result as it would require a sustained drop to last a long number of periods.
I bring this up because in cfiresim, the Guyton-Klinger method is modeled to update on a yearly time period, and as such it does not show the drastic drop in withdrawals that you show, even for 1966. I think that is because you are using a monthly drop, which with 12 samples in a year allows your result to converge faster on a much lower withdrawal value compared to the 1 sample a year in cfiresim. I believe this is because the larger time periods act as a filter, which result in withdrawal rates that seem substantially higher. To be fair, it is clear these higher rates would reduce the probability of success compared to the fine-grain higher-sample configuration of smaller time periods, and I have not quantified that–perhaps that would be a good place to investigate for further research.
The market never drops by 50% over-night. GK very much allows for a drop of 10% in several consecutive years. In the 1965 cohort you’d observe 6 (!!!) consecutive 10% drops., hence the drop in consumption from $40k to $21k. So you do see a significant drop in consumption.
The difference between my GK rule and the one in cfiresim is that, indeed, I do the calculations monthly, so the drop can happen faster. Also, I use the inflation-adjustments rule, as described in the original GK paper: Do no CPI-adjustment if the portfolio return in the previous period was negative. Thus, you see already some decline in real withdrawals before ever hitting the guardrail.
But also notice that in my calculations you recover much faster than in the cfiresim calculations. In my chart, the GK4% line is almost back to $40k again after 30y. While in the cfiresim numbers you’re still at $23k even after 30 years.
https://i0.wp.com/earlyretirementnow.com/wp-content/uploads/2017/02/swr-part9-chart3.png?resize=863%2C549&ssl=1
So, pick your poison: Do you want the GK rule to look like my chart (deeper drop, faster recovery) or in cfiresim (drop not quite a deep, but almost no recovery after 30y)?
So, the monthly frequency doesn’t make the GK rule look worse. It makes it look worse in one dimension, but makes it look better than cfiresim in another dimension.
Hope this explains the differences.
The author also used an 80% stock portfolio and 20% bonds. Most retirees won’t do this. If he had used a 60/40 portfolio he would not have found such drastic swings. I consider this a case of looking for the absolutely most extreme outlier, which I guess was his point. But it puts the guyton Klinger model in a bad light by using a portfolio balance I don’t think most retirees would use in the worst year ever
Most early retirees use a higher than 60% equity weight (normally around 75-80) because the portfolio has to last so much longer than for the average traditional retiree. That would be impossible with 40% bonds and an essentially zero real Treasury yield right now.
But if you ask me nicely I can certainly run the simulation again for the 60/40 case and prove to you that GK would have the qualitative features as I showed here. So what part would you like to see for the 60/40 case?
what if you make a bottom line, what happens? I mean tried to make a model when the condition of cutting the expenses hits a minimum I keep spending that minimum. Looks like works fine on this way.
It’s like squeezeing a balloon. You have less severe drawdowns, for sure. But you now risk depleting the money. This is especially true over very long horizons (50+ years).
Thanks for the responde. It makes sense.
Do you know another methods about controling drawdowns that we can use to make sustainable withdraw rates? What have been discussed most?.
I’m brazilian investor looking for a FI. We dont have many materials about this here.
Cheers from Brazil
I like the Glidepath the most. (SWR Series parts 19-20)
I also like the CAPE-based rule (Part 18)!
Cheers and “Obrigado” to beautiful Brazil!
GK rules should be implemented with the adjustments being made to the dollar amount that was the previous years safe withdrawal rate. The percentage isn’t adjusted or applied each year this would be safer but provide much more volatile income. The scaling relative to portfolio growth (including negative growth) is provided by the Capital Preservation and Prosperity Rules (i.e. the guard rails).
That’s exactly how I model the GK rule in my simulations. And that creates a lot of problems as I laid out in the two posts Part 9 and Part 10.
OK we got it’s false advertisement, but I was expecting you had a proposition to make it better in the end. Where’s the big Ern methodology for a variable speding rule?
Well, we can’t put a square item into a round hole, so we can use leverage sparingly and improve results slightly.
In terms of variable spending rules, I’ve written about CAPE-based rules.
What I would find more appealing would be a variable borrowing rule. Thus, keep the spending constant but vary the amount of borrowing (and potentially repayment of the loan). Essentially a follow-up to Part 49.
I will write a separate blog post on that. Can’t solve all issues in one post! 🙂
I find this an informative blog overall, but I have some issues with how this post was handled. You basically took Guyton and Klinger’s work, modified into something different, and then accused them of “false advertising”.
You remove the portfolio management rule (a simple rebalancing withdrawal strategy) along with the diversified equity portfolio claiming that it has negligible impact on performance, but looking at the final chart the difference between the two at a 99% confidence level is nearly an entire percent of initial withdrawal rate. That is huge!
I have a bigger issue with the change of the guardrail calculation. Clearly it will operate very differently when handled monthly instead of yearly. It will be much more volatile, contributing to your numbers that showed a massive reduction in real withdrawal. If sticking with the original rules, the withdrawal amount would not have fallen nearly as quickly (and it would have taken longer to recover as well). My guess is that if they were operating monthly, the percentages would have been recalibrated (something like 25 or 30% would be needed to trigger a 10% change).
I’m curious to see how the actual original rules for the guardrails would have faired under this situation. Based on their numbers, the 6% would likely have failed but the 5% should have survived. The real withdrawal would have been depressed for many years (but by significantly less than what you have calculated).
The concept in general is valid. You can start with a higher initial withdrawal percentage or have a higher chance of avoiding failure (or both) at the expense of more variable spending and in most cases a smaller final balance. It all depends on your tolerance for spending variance. Given that your goal seems to be retiring as soon as possible, I can see where this kind of trade off has less value (although I would argue that after a few years if disaster has not struck the real withdrawal rate can probably increase at least some). Other folks have a different set of preferences and will potentially be in a position where they can tolerate more spending variance. They may also want to spend as much as they can in the early years while they can truly enjoy it. In those cases it seems like some kind of variable spending strategy would make a lot of sense.
1: I’m not sure what chart you mean when you write “but looking at the final chart the difference between the two at a 99% confidence level is nearly an entire percent of initial withdrawal rate” The portfolio management approach is ineffective. You will not be able to generate any alpha with such a naïve rule. Believe me, I’ve worked for 10 years in tactical asset allocation, and this approach will not add consistent alpha. It’s laughable that anyone would claim this method will beat a static asset allocation approach. That’s why, over time, Guyton-Klinger have moved away from emphasizing that approach. It reminds me of the dreaded bucket approach discussion (see Part 55, item 3) where people also falsely claim that with some cheap asset allocation rules you can suddenly beat the market. You can’t. Anyone who claims otherwise is delusional.
PS: after looking at the GK article again, I presume you refer to Table 7 (the final table, not final chart, but whatever) and the fact that the “Multi-Class Equities” SWR is higher by 0.7 percentage points (e.g. 5.4% vs. 4.7%). That has nothing to do with the portfolio management approach. You’ll also get a much higher SWR with a STATIC allocation to small-cap and value stocks. There’s no excess return from the portfolio management. It’s the small-cap premium before 1980 and the Value premium before 2005. Has nothing to do with the GK approach.
Again, it’s a very sneaky, false-advertising approach by GK. A casual reader like you might think that the 0.7% extra SWR is due to GK’s method. But it’s certainly not.
2: monthly vs. annual makes no difference in SWR simulations. See Part 47 for that insight.
Also note that I withdraw 1/12 of the annual amount per month, not the annual amount.
Admittedly, the GK Rules are not terrible, but it’s false advertising to not show how badly the purchasing power would have been depleted. Where is that in the original GK paper? Can you show me? I showed you my simulations. Can you show me yours and prove they are better?
So I referred to both the portfolio management rule as well as the diversified equity portfolio. It does not really matter how much each contributed as you removed both.
I read your part 47, and I don’t think I made myself very clear. I think by checking the 20% guardrail every month, it will get triggered significantly more often that it would yearly. You have talked about the spectrum between a static rate and a percentage of portfolio (fully dynamic). By applying the same percentage check monthly as opposed to yearly, I believe you are making the guardrail solution more dynamic than the authors intended.
You are correct that I am a casual reader and I don’t have the ability to generate these simulations myself. I also agree that it would have been useful (and thorough) if they had shown worst case real withdrawal values. I am simply under the impression (perhaps wrongly) that by applying the original rules laid out in their paper, it would not be as low as your showing above (but with more likelihood of failure).
On a completely different note, I have been using a combination of relative strength and moving average to move around ETFs with a fair amount of success the last few years. Is your first paragraph suggesting that this kind of “tactical asset allocation” is doomed to failure?
I greatly appreciate you taking the time to respond to my post on a 3 year old article.
You are correct to question the effect of monthly withdrawals- (and presumably monthly rebalancing) – The benefits of rebalancing more frequently are dubious and some studies suggest that rebalancing more frequently LOWERS returns and INCREASES volatility, thus triggering more (unnecessary) adjustments in withdrawal rates. You can run simulations in cfiresim of a Guyton and Klinger approach with yearly withdrawals but with a basement withdrawal amount (indexed to inflation) that still allows a larger initial SWR-(around 4.7%) and survives many many years with the basement amount more than adequate to avoid a return to work– even in the worst historical case (1966)
Not true. I ran GK on cfiresim and the 1966 cohort displays qualitatively and quantitatively the exact same features as my monthly simulations.
For example, start with $47,000 initial withdrawals. The 1966 cohort will have to cut withdrawals to only $20,231.96 in 1980 and stays there until 1991. The spending somewhat recovers to 22,255.15 in 1992. Do you consider $20k a comfortable retirement budget? That’s a massive 57% cut in spending. About in line with the monthly simulations in my post.
That is why I said you put in a lower limit – (a floor-or basement lowest level) and you never suffer that kind of downgrade in spending.
Yeah, I tried that too. Start with $47k, with a $32k floor (>30% below initial). In 1966 you drop to that $32k level almost immediately. And your average spending over a 30y retirement was $35k (i.e., >25% below the initial). So, my point again: false advertising. People believe that they can miraculously increase their SWR with GK. But you can’t. $35k on average is just 3.5%. I never doubted that 3.5% is an appropriate SWR, even at the market peaks.
You can think all want, but I know that this is an irrelevant issue. I’ve simulated GK with annual data and you get qualitatively the same and quantitatively almost identical results, i.e., if the monthly simulations generate 2 downgrades (as the 4% ad 5% rates would have in the 1966 cohort) then the annual simulations generate 2 downgrades in back-to-back years. Your point is completely pointless, no matter how much you believe in it.
And again: the portfolio management issue is irrelevant. The entire outperformance of the more complicated index portfolio is due to the (historical) outperformance of small and value biases. And, as I have written about on my blog, there is no guarantee that this will continue, but that’s a completely separate issue.
In fact, reiterating the “false advertising” issue, you realize the irony, right? Your misunderstanding that this 70bps extra WR is due to the portfolio management rule, proves my point of false advertising.
And again, I could have added the static Small+Value bias and do my simulations and shown that in that portfolio, too, GK is false advertising because of the massive drawdowns in real spending. Albeit from a higher initial level due to the historical SCV bias.
Also, forgot to answer the other question: I think that a momentum strategy with some additional bells and whistles is likely not a bad idea.
So I have been doing something that roughly replicates the multi factor ETF OMFL for the last couple of years. The performance has been similar as well (3-4 percent better than the market including during down years). I have roughly another 6 years to monitor before retirement. My question would be how much do I incorporate this into my initial withdrawal rate? I’m already intending to go 100% equities because the improvement in down years over the market is roughly similar to adding 20 or 30 percent bonds in the mix. I just don’t know how to mitigate the risk of my methodology failing. I don’t know if 10 years of implementation plus another 10 – 15 years of back testing is enough sample to rely on.
As an economist, I like that idea. There should be some benefit from timing different styles/sectors over the business cycle.
However, there are a few caveats: 1) the model is only as good as your economic forecasting tool. and 2) the model is only useful if past correlations stay intact. Styles that did well during past bear markets may no longer do so in the future.
But again, in general I like the idea and at my previous job I actually helped set up a custom-built product like this (though not for the retail market).
Hi ERN, thanks for the article!
While I agree GK allows for deep withdrawal drawdowns, you’ve made a mistake in your simulation, and I see it clearly only because I did my own version in Python with monthly data too.
Scaling from annual to monthly, you withdraw 0.33% per month, but adjust your withdrawals with full annual 10% number in each month, which is not in line with the original paper and introduces withdrawals volatility which was never there.
Instead, you should scale 10% down to 1.1^(1/12)-1 ≈ 0.797% per month. I found the results are more in line with the original paper this way.
That’s not a mistake. It’s intentional. If your portfolio is down by 20% you don’t just pussyfoot around and adjust by only 0.797% every month.
In any case, I reran my simulations and the new assumption makes no material difference in this 1966 case study. The volatility of consumption over the 30-year retirement horizon is actually higher with your adjustment method when using the 5% and 6% GK rule, but admittedly, slightly lower with the 4% GK rule.
Also, I found this quote amusing: “I found the results are more in line with the original paper this way.”
There are no simulations in the GK paper of the type I have performed, i.e., case studies and simulating over all cohorts (see Part 10). And if you had done simulations the way I am doing them you’ll find that the results of the 10% vs. 0.797% monthly are qualitatively very similar. I will post an update in the post and show you the results. Stay tuned.
In the future, maybe be more careful before accusing others of mistakes. 😉
Mistake is mine, nothing changed indeed. 😅
The thing is I studied a different market, had cash, gold and more bonds in my portfolio and also implemented their ‘portfolio management rule’ for monthly withdrawals and annual rebalancing. Looking at percentiles of real withdrawal amounts across entire retirement periods there are small differences between two approaches to withdrawal adjustments favoring the slower one. The downside of slower adjustments though are slightly increased failure rates.
Anyway, I forgot worst cases were almost the same because they are prolonged in time.
> There are no simulations in the GK paper of the type I have performed, i.e., case studies and simulating over all cohorts (see Part 10).
Sure, I meant in line with the speed of withdrawal adjustments in the original paper because they used annual data and we use monthly.
Thanks for confirming. Yes, we can come up with examples where you just trigger the adjustment but then the market recovers again. Then the 10% adjustment overreacts, while the 0.8% monthly adjustment is more cautious and correct in that case.
But the big market events look like the charts I showed, of course.
Glad we’re on the same page!