September 14, 2021
Welcome to a new installment of the Safe Withdrawal Rate Series, dealing with Bucket Strategies. This is one approach that’s often considered a viable solution to the dreaded Sequence Risk Problem. Simply keep buckets of assets with different risk characteristics designated to cover expenses during different time windows of your retirement. Specifically, keep one or more buckets with low-risk assets to hedge the first few years of retirement. And – poof – Sequence Risk evaporates, just like that! Sounds too good to be true, right? And it likely is. Long story short, while there are certain parts of the bucket strategy that can indeed partially alleviate the risk of retirement bust, bucket strategies are by no means a solution to Sequence Risk. Let’s take a look at the details…
A favor to ask
Before I get started today, I wanted to direct you to my buddy Victor Vulpescu’s podcast. He interviewed me recently on safe withdrawal strategies, sequence risk, glidepaths in retirement, and much more. I was really impressed with the preparation he put into his Q&A session with me. Very thoughtful and challenging questions – one of the most finance-savvy podcasters I have ever encountered. I would really like to drive some traffic to his content because it’s a true undiscovered gem:
- Podcast: https://victorvulpescu.ro/podcast/
- My episodes directly on Youtube, chopped into 3 parts because we chatted for much longer than planned!
His podcast is obviously a smaller, less-known one, but he has been able to land a lot of heavy-hitters and some of my favorite speakers in the industry:
- Rick Ferri
- Paul Merriman
- Meb Faber
- Larry Swedroe
- Lars Kroijer, and more
Now, back to today’s topic. Let me start by looking at one form of the bucket strategy that will certainly eliminate Sequence Risk. Though, it comes at a cost…
LDI: The Ultimate Bucket Strategy! And why it’s likely not suitable for today’s early retirees
If you wanted to take the bucket strategy to its extreme, you could certainly just build a CD or bond or TIPS ladder and thus keep one bucket for each retirement year (or quarter or month) to target a mostly risk-free payout on that specific date. Well, it depends on what you call risk-free. If the ladder is implemented with CDs or nominal bonds then you guarantee nominal payouts only but leave yourself exposed to inflation shocks. If you like to hedge inflation risk as well, you would be better served with TIPS.
This matching of assets to specific future expected liabilities at specific times is also appropriately called Liabilities-Driven Investing, or LDI. That’s what some pension funds are doing. You allocate your assets so as to exactly offset your future liabilities, year-by-year; sometimes by investing in bonds that mature at the time you need to service your future liabilities. Sometimes it’s sufficient to simply calculate the average duration of your liabilities and target that same duration in your bond portfolio.
One major headache with this approach right now: To guarantee stable and predictable payouts immune from any Sequence Risk we might have to reduce our retirement budget quite substantially because yields on safe, inflation-adjusted bonds are so low right now! Pushing -2% for shorter-term TIPS, around -1% for 10-year TIPS, and still negative at -0.36% for the 30-year TIPS bonds as of 9/9/2021, see the screenshot from the Federal Reserve below.
Does that mean we have a negative safe withdrawal rate? Not really. We can translate real interest rates into Safe Withdrawal Rates using the Excel PMT function. For example, over a 30-year retirement horizon, with a real rate of -0.5%, and the final value target of 25% of the initial, we compute our safe withdrawal rate as:
=PMT(-0.005 , 30 , -1 , 0.25 , 1)
The inputs in the Excel PMT function are:
- -0.005 for a -0.5% real rate per annum
- 30 for the horizon (years)
- -1 for the inital portfolio value (modeled here as a -1 cash flow to indicate we pay $1 into the portfolio today)
- 0.25 for the final value target (what we take out of the portfolio)
- 1 to indicate that we withdraw at the beginning of the year, as is usual in the safe withdrawal context (Trinity Study, my own simulations, etc.)
Since bond rates are so low we get some spectacularly low withdrawal rates. Only 2.2% in the numerical example above! That’s substantially lower than all of the historical retirement cohorts, even those that retired right before the worst Sequence Risk episodes, e.g., the Great Depression or the 1970s. Protection from Sequence Risk comes at a steep price!
How about other assumptions? In the chart below I plot the (real) withdrawal rates (assuming CPI-adjustments in withdrawals) as a function of the Real Rate (x-axis) and for different retirement horizons and final portfolio targets. For the current range in real rates, somewhere in the -1% to +/-0% region, we get Safe Withdrawal Rates of around 3% for a 30-year horizon when completely exhausting the portfolio, 2-2.5% when planning to leave a 25% bequest, around 1.5% for a 60-year horizon when depleting the money and well below 1% to slightly above 1% when planning a 60-year retirement and a 25% bequest target. Ouch, that’s getting close to the 0.5% number that Financial Sumotroll, or whatever his name was, “calculated”. I’d rather just try my luck with a Stock/Bond portfolio and a 3.25-3.50% safe withdrawal rate that would hedge even against a repeat of the Great Depression!
But don’t get me wrong! If we ever move back to an environment like the late 1990s when real bond rates were well above 3%, we could easily push the 30-year withdrawal rates to 5% or even above! Not sure we will ever reach that point again during my lifetime, but I wouldn’t discard the idea completely.
Some further related reading: check out a post I wrote in 2019 on some of the other challenges of translating pension fund mechanics to personal retirement: Part 32: You are a Pension Fund of One (or Two). In a lot of ways, personal retirement is often more complicated than running a pension fund for thousands of beneficiaries.
OK, of course, this is not the Bucket Strategy that most people in the retirement community have in mind. This brings me to the next idea…
Using a bucket approach to design a strategic asset allocation in retirement
Unless you are a successful internet entrepreneur and/or retirement blogger who can afford to keep 100% of his/her/their portfolio in stocks, most retirees will need to design a less adventurous retirement portfolio with some diversifying assets. And then find a way to stick with that allocation. How do you design the portfolio weights then? Well, a bucket approach isn’t a bad idea to start. Maybe keep 3-4 years in cash (money market accounts, short-term CDs, etc.), another 6-7 years in intermediate bonds (10-year maturity), and the rest in stocks. Nice! That gets you an allocation of around 60% stocks, 25% bonds, and 15% cash.
Imagine now that you start with a $1,000,000 portfolio, $40,000 annual withdrawals, and a 60%/25%/15% Stocks/Bonds/Cash allocation. You take out $40k from the cash bucket at the beginning of year 1. You get income/returns of $300, $4,000, and $36,000 in the three respective buckets, which means that at the beginning of Year 2, your portfolio is completely out of whack compared to the 60/25/15 target asset allocation. If we rebalance, we’d need to take roughly 36k and 4k out of the stock and bond bucket, respectively, and transfer that into the cash bucket to bring the portfolio back to the 60/25/15 target.
In the second year, if you now take out $40k from the cash bucket again you may think that it came out of the cash bucket, but effectively the money came out of your stock and bond bucket! Money is fungible! Just like in the bucket strategies diagram below, you may think that you took money out of the cash bucket, but of course, the bond bucket continuously spills into the cash bucket and the equity bucket continuously spills into the bond bucket. So, in this sense, the bucket strategy is largely window dressing if you rebalance regularly. The rebalancing will continuously take money out of the Sequence-Risk-sensitive buckets, invalidating the alleged benefits of the bucket strategy! Sorry to pop your
bucket, uhm, bubble!
Just to be sure, what I still like about the bucket strategy is the fact that it might give retirees more confidence. Personally, I’d prefer to build up my confidence with my own approach: Play with different asset allocations and see how they would have performed in historical simulations. At least I can study and quantify the pros and cons of different asset allocations and see what performs best during past Sequence Risk events. But, whatever floats your boat; if a bucket strategy makes you sleep better at night, go ahead!
But the bucket approach doesn’t really give you much guidance in how to design your exact bucket sizes. Should your cash bucket be 1 year or 5 years’ worth of living expenses? Should the bond allocation be 5 years or 10 years or 20 years? The bucket approach looks more like a cheap sales gimmick from the Financial Planner industry and/or a “sexy” topic to write about for retirement bloggers. A financial beauty contest, really. But I can’t learn much from looks. Show me the numbers! Show me the simulations!
So, if regular rebalancing just amounts to a fixed asset allocation and the buckets do not really protect you from Sequence Risk at all, should we then just deplete the buckets? This brings me to the next section…
Buckets without rebalancing
What would happen if we exhausted the short-term buckets without rebalancing? Now we can indeed claim that the bucket strategy will leave the high-risk/long-term bucket shielded from withdrawals and thus Sequence Risk. This however brings up one additional headache. If we start with the bucket strategy like the one introduced above and we subsequently exhaust the $150,000 in the cash bucket after about 3.75 years and then the $250,000 in the bond bucket after another roughly 6.25 years, will the average retiree be comfortable with a 100% stock portfolio after 10 years? Your average CFP financial planner would get a heart attack when they hear that. Traditionally, financial planners tend to shift the equity portion down not up during retirement. The same is true for all target-date fund families I am aware of, including Fidelity and Vanguard.
But then again, research has shown that glidepaths in retirement that shift up the equity portion over time can alleviate Sequence Risk. Not eliminate, just alleviate! Michael Kitces wrote about it. Part 19 and Part 20 in my SWR Series deal with this issue. So, I like to present some additional simulations to study the efficacy of a strategy involving a bucket of safe assets, which brings me to the next section…
Bucket Strategy Simulations: ERN-style
To be sure, I looked at the topic briefly in Part 25, and I showed some simulation results. But I never published any dedicated post about this important topic. Which is one of the purposes of today’s post. Let’s assume we use the following baseline model assumptions:
- A retiree with a 30-year horizon.
- A 25% bequest target. As mentioned earlier, I like this parameterization because it applies both to a traditional retiree with a modest bequest target, but also an early retiree who wants to bridge the first 30 years of retirement until other cash flows kick in, e.g., Social Security, Pensions, inheritance, etc.
- A 75% Stock, 25% Bond portfolio
- No other supplemental cash flows along the way
- The exact Dollar amount in the portfolio doesn’t really matter because we can scale this up/down to any initial value. But to make this more relatable, I assume that the initial portfolio is $2,000,000.
- No cash bucket in this baseline scenario!
This parameterization is pretty straightforward to implement in the Safe Withdrawal Rate Google Sheet (see Part 28 for details). Add the asset allocation in the main parameter sheet, set all the supplemental cash flows to zero, and check the results.
The main results are the failsafe withdrawal rates and $ amounts (scaled to a $2,000,000 initial portfolio). I like to distinguish those failsafe numbers by decade as well, to get an understanding of when those historical worst-case cohorts faced an ugly Sequence Risk problem. Quite intriguingly, the 1920s, including September 1929 was not the worst-case scenario. The failsafe rate was even lower in the 1960s due to the protracted bad performance during the 60s, 70s, and early 80s in both stocks and bonds. This was a worse episode than the Great Depression, from a Sequence Risk perspective because a) it took longer to recover and b) bonds were not very good in diversifying the equity bear market, due to the inflation shock and extreme interest rate hikes under Fed Chairman Volcker.
Next, let’s implement the safe cash bucket. I assume that our retiree sets aside a lump sum of money to guarantee a $5k/month or $60k a year, inflation-adjusted cash flow. I was debating whether I should set that cash flow to $6k, just above the $71,683 annual failsafe, but I decided to round that number down to $5k/month. So, we cover almost the entire worst-case monthly retirement budget for a number of years. This lump-sum amount of money we have to set aside is a function of the monthly cash flow ($5k), the length of time we want to cover with the cash bucket, and the prevailing real interest rate (using the PV function in Excel!). I use cash buckets worth 1 through 25 years in 1-year steps and real interest rates of -1%, 0%, and +1%.
Let’s look at how the failsafe withdrawal amount would have changed if you had deployed this cash bucket approach right before the Great Depression. That failsafe cohort was pretty much consistently the September 1929 retirement date. I plot the results in the chart below. Specifically, the safe withdrawal amounts as a function of the number of years worth of expenses in the cash bucket (x-axis), normalized to 0% at the baseline, i.e., without the cash bucket. And each line corresponds to a different real interest rate.
Even with a -1% real rate for the safe, inflation-adjusted cash bucket, we would have significantly benefited from the bucket strategy. You could have increased the withdrawal amounts by about 10%. Say, from 3.6% ($72,000 p.a.) to almost 4.0% ($80,000 p.a. in a $2m portfolio). Pretty impressive! The peak performance was at 6 years’ worth of expenses. But even larger cash buckets, even all the way to 18 years you’d still benefit from the cash bucket (i.e., the blue line is above 0%). Astonishing, because that would involve setting aside almost $1.2m or almost 60% of the portfolio! If we find ourselves with higher real rates than today, 0% or 1%, cash bucket allocations become even more attractive.
One word of caution about these simulations: We are indeed doing a little bit of financial simulation “sausage-making”: I model the cash bucket using today’s real interest rates, while the stock and bond returns are the historical numbers from the 1920s and 1930s. Could I have used the interest rates prevailing in 1929? Sure, but keep in mind that back then we didn’t even have TIPS. I could have used the 1929 interest landscape (both short-term and long-term rates far above today’s). Using the higher past short-term interest rates to study the efficacy of the cash bucket seems inappropriate when you get essentially 0% in a money market account today – before inflation! I found my approach the more intuitive way of implementing the cash bucket: use past real stock and bond returns but factor in the cash bucket through a preset real interest rate. But just to be sure, later I also include some simulations using the actual historical short-term interest rates.
On to the next Sequence Risk cohorts. The 1960s had an even slightly lower safe withdrawal rate in the baseline. The bad news is that with a real interest rate too low (-1%), the cash bucket is useless. Even for higher real rates, the bucket would have been ineffective unless you pile at least 10, or better 15 to 20 years’ worth of expenses into your cash bucket. The reason is that the bad sequence risk cohorts during the mid-1960s didn’t see any noticeable bear market until 8-17 years into their retirement (1973, 1980, 1982). To cover the entire bad market event you’d need enough cash to cover your expenses until at least the mid-1980s when the market had swiftly recovered from its 1982 bear market bottom. But that big chunk of the portfolio sitting around and earning negative interest rates drags the blue line below zero. Even for the 0% or +1% real rate, who would have planned for a cash bucket that large in the 1960s? Who would do this today?
Next, let’s take a look at the 1970s, which also had some gnarly bad retirement cohorts right before the 1973-75 recession and bear market. See the chart below. Here we find a positive impact from all three interest rate assumptions and the maximum boost in the failsafe between 11 and 13 years, depending on the real interest rate. The 1973 cohort, right before that bear market had a slightly easier job than the mid-1960s retirees because the overall length of the sequence risk event was shorter. Also, the bear market was bad enough and swift enough, that even with a -1% real yield, you would have greatly benefitted from the cash bucket.
To summarize so far: the cash bucket certainly works both in a deflationary recession (Great Depression) and an inflationary recession (the 1970s) if you keep around 6 to maybe 10+ years worth of safe assets even with a negative real yield. The cash bucket poses more of a challenge when the stock market moves sideways for a number of years and then crashes 10+ years into retirement. If real rates are too low, as they are today, there will be no amount of money in the cash bucket that can ever significantly shield you from Sequence Risk.
Modeling the Bucket Strategy through a Glidepath
The idea of phasing out a component of the portfolio over a certain timeframe should sound familiar. It’s a glidepath! Well, it’s not exactly identical to the glidepath because with a strict bucket strategy there would be no rebalancing between the cash/bond buckets and the equity bucket(s) along the way. On the other hand, the glidepath would stoically rebalance to the (changing) weights along the path every month. Depending on the relative bucket returns this will almost certainly create flows across buckets. Specifically, if the equity market drop is swift and deep enough, you will likely shift funds from the cash/bond allocation into stocks throughout the Bear Market. This form of Dollar-Cost Averaging is in contrast to the strict bucket approach where you withdraw from the short-term bucket and leave the long-term buckets untouched.
Unfortunately, the Google Sheet is not yet set up for studying this glidepath approach over the entire long list of possible retirement cohorts (1,700+ cohorts!). Rather, the current version only allows me to study glidepaths as part of the “Case Study” for one specific retirement cohort at a time. It’s OK for our purposes because I really need to look at the three different cases in the 1920s, 1960s, and 1970s that generated their respective failsafe withdrawal rates: September 1929, November 1965, and January 1973. In each case, I study three different glidepath assumptions:
- Move 30% into cash/MM, and keep the remaining 70% in a 75/25 portfolio. Transition this 52.5/17.5/30.0 portfolio back to a 75/25/0 portfolio over 10 years. And again, the 30% shift into cash correscponds to $60k p.a. times 10 years.
- Same as above, but don’t bother about cash or money market. Simply start with a 52.5% stock and 47.5% bond portfolio and shift that back to a 75/25 portfolio again over 10 years.
- In line with the simulations from Parts 19 and 20 of the series, start with 60/40 and shift that into a 100/0 portfolio over 10 years.
- For comparison, I also include the results for the 10-year cash bucket, $5,000/month, 0% real rate cash bucket.
In the table below are the final portfolio values, adjusted for CPI, after 30 years of withdrawals. Notice that in the 1929 and 1965 case study I use the 3.60% withdrawal rate, i.e., $72k p.a. out of a $2m portfolio. In 1973 I raise that initial withdrawal rate to 4% or $80k p.a. because in that cohort the 4% Rule not just preserved money but even left you with $670,844 after 30 years.
We can just quickly check that with our 0% real interest rate and a 10-year cash bucket we would have done substantially better than in the base case. Instead of assuming a 0% real cash rate, if we apply actual cash returns during that time we would have done even better in 1929 and 1965, no doubt because actual average real interest rates from short-term bonds were quite a bit higher than 0%. For the 1965 cohort, the 52.5/17.5/30.0 glidepath did slightly worse than the 10Y cash cushion with 0% real interest rates. And that makes sense, too, because cash yields didn’t keep up with rising inflation rates in the mid-70s, and thus the actual real returns of a cash bucket would have been negative. Nevertheless, the glidepath with cash was a success even for the 1973 cohort because the final value at $1.4m+ was more than twice that in the baseline.
Also, it comes as no surprise that during the Great Depression, you would have done even better by just using 10-year Treasury Bonds instead of cash and you would have done spectacularly well with the 60/40/0 to 100/0/0 glidepath. On the other hand, the 1960s and 1970s retirees would have lagged behind the glidepath utilizing 30% cash. But they would have still done better with the bond glidepaths than with the static 75/25 allocation. So the bond-only glidepaths still worked, relative to the static 75/25 baseline. Just not as well as the cash glidepath.
What about the Type 1 Error: Have a cash bucket when it wasn’t even necessary?
Finally, I also like to study how much money we leave on the table if we implement the cash bucket at the wrong time. When after the fact it turns out that it wasn’t even necessary. Imagine someone had started retirement a year before the worst-case scenario in September 1928 or January 1972 or two years before the 1965 episode in November 1963. (I didn’t use November 1964 because that looked almost like November 1965). So these are all retirement cohorts with elevated equity valuations, but the market still had 1-2 years to run.
It turns out, all cohorts would have easily met and even substantially exceeded their $500,000 final value target. Quite intriguingly, the 60/40->100/0 glidepaths do quite well. In 1928 and 1972 even better than the static asset allocation (so not even a Type 1 error, after all). In 1963, you lag behind the static allocation a little bit (~$1.27m vs. $1.39m), but that’s small compared to the outsized outperformance for the other cohorts. The 60/40 glidepath would have also significantly outperformed the glidepath starting with 30% cash in all three cases.
What about Fritz Gilbert’s Bucket Strategy?
This post wouldn’t be complete without mentioning my buddy Fritz’s blog The Retirement Manifesto because he has written extensively on the bucket strategy approach. Here are the links to his excellent posts on this topic:
- How to Build A Retirement Paycheck From Your Investments
- InfoGraphic – The Bucket Strategy
- Your Bucket Strategy Questions, Answered!
Fritz and I don’t disagree at all. In essence, Fritz will end up with a 63/27/10 allocation which would roughly correspond to my simulations with roughly 4-5 years of a safe asset bucket and the remaining part of the portfolio in the 75/25 bucket. (a slight mismatch of the weights but close enough).
Fritz also insists that his plan involves rebalancing for as long as the market cooperates and using the cash bucket drawdown only when the market tanks. This will line up exactly with my simulations if you study the historical failsafe cohorts. And even the pre-market peak cohorts that regularly rebalance the portfolio will exactly end up at the market peak before the drawdown with their 63/27/10 intact and then start the cash bucket drawdown when the market retreats in 1929 or 1965 or 1973. So, to answer Fritz’s question below “Not sure if you’d be able to model that, but I’d be interested.”: I did model this approach (or a close-enough variation of it) for the 3 cohorts that retired at their respective market peaks. And just to be sure, the approach “works” if you define “works” as slightly reducing sequence risk. We can never eliminate Sequence Risk. Fritz’s bucket approach and my glidepath approach are not really that different. In fact, his 27% bond and 10% cash drawn down will be very close to my 40% bonds drawn down over time.
Hope this helps clarify Fritz’s question and the questions from the many readers of my blog that also follow his content! 🙂
There you go, finally, I finished the post. After teasing everybody and getting folks excited about the bucket strategy post, I hope that it was worth waiting for another 5 days! 🙂
What I’ve learned from today’s post is that a cash bucket faces challenges when interest rates are low but it certainly works if you deploy it right at the peak of the market. Depending on what kind of bear market we go through, though, the cash bucket might work better or worse than diversifying with a bond portfolio. If you fear an upcoming inflation spike the cash pile will be the better option. If we face another demand-side recession where the Federal Reserve will try to push down the bond term structure again, then government bonds will be better.
I must say that I still like the 60/40 to 100/0 glidepath the best. It wasn’t too far behind the cash glidepath in an inflationary environment, but it does a lot better in a deflationary setting and/or in the Type 1 error scenario where we still have a few years to go in the current bull market.