Fundamentally Wade’s work, in his newest book, is a great summary of a lot of research so far in this nascent field of withdrawing from assets to supplement retirement income.
“How Much Can I Spend in Retirement?: A Guide to Investment-Based Retirement Income Strategies (The Retirement Researcher’s Guide Series)” by Wade D Pfau, available at Amazon.
I would rate his work 5 stars, but give it a 4 for the reasons below. Wade is a thought leader in the field having made many contributions. The main reason for the slight downgrade are the flaws baked into the current paradigm of measuring time while performing the “calculation” (deterministic or stochastic). This isn’t Wade’s issue, but rather the fault of the entire profession and academic community as a result of those calculations evolving from the deterministic thinking process before Bengen’s and Monte Carlo applications emerged.
However, my work with other researchers (Frank and Brayman: Journal of Financial Planning: “Combining Stochastic Simulations and Actuarial Withdrawals into One Model,” November 2016), has changed my paradigm from measuring progress as years IN retirement (how long ago did you begin taking funds out of your assets); to measuring progress as years REMAINING for retirement. In other words, do you count the years UP (the IN paradigm), or count the years DOWN (the remaining paradigm).
I do wish to thank Wade for calling out my work (with others) in this field of research on pages 32, 174, 198 and 223. Wade did an outstanding job summarizing the state of research and thinking today. But, as I argue below, a paradigm shift will resolve many of the issues he raises and simplify decision making for retirees and their advisers in the future.
Since the “Iterative Paradigm” is not described anywhere, here is a brief description. The paradigm begins by recognizing first, that each age has its’ own associated expected period life table age (which further depends on the use of strategic use of table percentiles as described below in the Mitchell reference). The difference between those two ages determines the time that is used in the simulation (how long a period do you use?). There is no real need to speculate on the time period at any given age since statistics help answer the question already. The second recognition is that, as one ages, the expected period life table age also ages, albeit slower than an annual rate. Thus, there is a rolling period affect as one ages. The withdrawal period length slowly declines in the number of years used in each simulation set based on age. This paradigm solves the issue about how do you transition between time periods that you should use, as Wade explained, any 85 year old wouldn’t use a 30 year period. The fixed period paradigm has no good answer to the question which time period do you use for what age. You transition in the iterative paradigm by redoing the simulation each year during reviews and incorporating new data as it becomes available each year. What is currently missing, is software that takes this iterative re-calculative approach showing what the spending answer is for each age (and portfolio balances) through rolling calculations using rolling time periods. What this means is that the future time-adjusted income and portfolio balances can’t be calculated yet (even under the current paradigm as discussed below). Until software becomes available, doing an annual review where everything is recalculated is the solution to answer the question “How much can I prudently spend this year.” Software today can’t tell us what the future time-adjusted income or balances may be. All it tells us is what is prudent spending for the coming year. The process is described in this presentation discussing the findings of research that contrasts the fixed period paradigm with the iterative paradigm.
Although this review may appear long – I thought it necessary to point out the areas that differ depending on which paradigm the income solution is viewed from.
There are significant differences that emerge from what paradigm you operate under. The following discuss these differences.
Note: It is the summation of all of the points that lead to the change in paradigm, each point combined with all the other points I make below. All of these lead to the paradigm shift as a result of being an “actuarial advocate.”
- Fixed periods: Distinguishing the answer from the derivation of the answer. Monte Carlo (even Dynamic Programming method pp224) use a single period over which the iterations of the simulation occur. What results (first graph illustrating Monte Carlo (MC) appears in Exhibit 3.2, with all subsequent graphs using MC in his book, and elsewhere, looking the same, is a single point of origin on the left side of the graph, fanning out into iterations showing ending values based on the percentiles chosen by the graph-er running the simulation to illustrate. How many “answers,” or solutions to the questions, how much can you spend at any given time, or what is your portfolio balance (remaining real wealth) do you see on such graphs? More than one over time? No – there is only ONE solution on any, and all, of those such graphs! The only answer is the singular point on the left side of the graph! All the other points you see illustrated represent derivation of that single solution. The problem becomes trying to interpret the terminal values, essentially a subjective approach, with common initial inputs to the simulation. This makes the emphasis, or focus, backwards. Instead, a common terminal value that we (JFP 2011 paper Wade referenced pp223) suggested is through the use of setting the percentage of failing (or successful) simulations as the common evaluation input value which, when put into use, would derive various origination values (solutions), that can be more objectively compared. Use of success rates (or failure rates) also helps refine when portfolio balances may be in danger of depletion (pp300) as one moves between 90% success to 75% success during the course of any given year should markets decline sufficiently to warrant closer monitoring. This approach focuses attention on objectively evaluating the different solutions (graph’s left side) instead of subjectively evaluating terminal values resulting from common inputs (the right side values float all over as you see comparing the Exhibit’s in Wade’s).
2. Fixed Periods: Trying to resolve all issues through the selection of one, single, fixed, period. This issue is best shown on page 47 where Wade discusses the probabilities of lifetime graphically. He also discusses the actuarial approach throughout (e.g., pp 149, 174-175, 195, 294-297). This is not a safety-first or probability-based view. It is a component of iterative solutions being modeled in that the concept of outliving longevity is a rolling period problem with which a rolling period modeling helps understand and manage this risk better (Frank, Brayman 2016 referenced in the beginning paragraph above). The fundamental question is when do you want to spend your money? Spending earlier in retirement while you’re most likely to be alive, or later in retirement when you’re not? Statically, this appears to be a one-time answer. But, this question applies to each year you’re alive. Strategic use of the life tables (Wade’s discussion of these in Chap 2 is outstanding) because of a feature of the shape of any and all life tables due to age. “Manipulation of the planning horizon is studied here as a means of protecting the eldest retirees. As retirees age the uncertainty of their remaining lifespan (coefficient of variation) increases” (“A Modified Life Expectancy Approach to Withdrawal Rate Management”, Mitchell, SSRN search 1703948, 2010). The manipulation of the planning horizon was applied in research where the longevity percentile was extended slowly (1% per year, starting from expected longevity at age 60) towards the right side of the life table, or older ages, so that this uncertainty was more certain to be considered conservatively as a person ages. (“Transition Through Old Age in a Dynamic Retirement Distribution Model” Frank, Mitchell, & Blanchett, JFP December 2012). To be fair, Wade discusses the actuarial approach using the Required Minimum Distribution (RMD) tables to represent life tables. However, the RMD tables have a taxation agenda baked into them, as well as an exponential effect on withdrawal rate as time periods shorten (as does also happen with any life table). A simple method to address the exponential effect was researched and applied in the 2012 paper referenced in this paragraph earlier. Use of actual life tables (with that choice depending on individual circumstances) is better with the RMD tables being a separate taxation issue (money needs to be taken out of retirement plans for taxation, but doesn’t need to be spent, but could be saved, based on retirement spending determination; i.e., these are two separate things). Because of the static single fixed period paradigm, software has not been developed to calculate the actual spending and capital balances one may have at each given age (as demonstrated in Frank, Brayman 2016). Hopefully, this shortcoming can be resolved soon because it would enable better management of money to see the effects of when spending occurs or how spending affects bequests. By the way, there is always money for bequests under the actuarial method because, as long as you’re still alive at any given age, you always have years remaining require funds to spend over those years (e.g., setting success rates to 90% means 90% of iterations push those funds into those later years at every age). Modeling solutions means all years in the life table (e.g., out to age 120) have already been calculated is so chosen to do so. It’s a matter of “revealing” those solutions if so interested. In other words, choosing a later age to evaluate should not mean getting a different spending solution today, unless you don’t like the spending or wealth remaining; in which case you’d change spending a little bit all along the way to get that future result you wish. MC simulations today don’t allow very well for such refinement of spending and wealth visibility, due to issues discussed in point #4 below.
3. Fixed Periods: How do you transition between them under the static paradigm?
Most periods illustrated are the 30 year period which looks at the time frame of retirement without any real reference to age. Not all retirees today are 65 (age 95 minus 30). Retirees today have many ages. What time frame should be used now for their current age? Of course, other time frames may also be illustrated as Wade did in Exhibit 5.23. When the time frame changes, so does the withdrawal rate – shorter periods have higher rates. How does one transition between these different ages and implied withdrawal rates? And what time frame should be applied to each age? Planners and people use different time frames for the same exact current age! You wouldn’t have such a divergence of time frames using life tables – the main differences coming from which table was used applicable to what circumstances each retiree fit in, a much more rational approach. Wade’s Exhibit 5.23 illustrates this changing rate with changing time frame, and allocation, very well. Note that the “optimal” allocation also changes with time (although the allocation columns aren’t labeled to make that clear). Allocation is discussed more in #5 since #4 needs to be understood first.
4. Time: Counting the years UP? Or, counting the years DOWN?
There are cash flow and balance errors that emerge from how time is used in the retirement arena (Frank, Brayman 2016). Here is the time count methodology for the single fixed period paradigm … note that each year is precisely 1 year in length between each year IN retirement (suggesting counting FROM the single point in time when you actually retired or started withdrawals). This is counting the years UP and is illustrated in all of Wade’s exhibits (retirement duration, or year in retirement).
T1, T2, T3, T4, … T28, T29, T30. But, what about years beyond year 30 if one lives that long?
The actuarial method and the paradigm the comes from it (and all the other points here), count the years remaining for retirement … this is counting the years DOWN (and not as pessimistic as one would think due to the table percentile adjustment that extends time with age as discussed in point #2 above which always has more years considered than the person presently may outlive at older ages). Using the tables, the transition between each age (TN) is less than one year, so there are times when the calculated method for the following year would be the same as the previous year*. I show the current age with the subscript as the time difference from the table between the current age and the table age for the time frame simulated; and that difference depends on which life table one uses (when I say life table, I’m also including cohort tables, i.e., my use here is all inclusive of all tables generically – the effect is the same, so the point doesn’t change – only the time difference for simulation period would change – and that again would be chosen by client specifics). This may sound complicated, but in application (see last summary paragraph below), this is much less complicated than the rules and tables approach currently in practice.
6524, 6623, 6722, 6822*, … 938, 948, 957, etc. out to the very end of the tables.
*where use of finite numbers in a simulation period catches up to less than one year incremental changes between finite ages in the tables.
So why is the time frame important? Because, with a single fixed period, the last year of a simulation set is ONE year long. However, simple reference to any 95 year old still living in any life tables shows that they likely have more than one year to plan for retirement income. This means that iterations for T30 for both cash flow and portfolio balances are off; cash flow is too high which makes capital balances too low. T29 is also off because it is 2 years long (year 29 and year 30). The same for each and every year in a 30 year simulation period. The sum of those cash flows and portfolio balances errors accumulate over time. This is what makes those ending values (the derivation of the solution) in point #1 above less useful than presently interpreted. Setting a fixed value to measure ending values of iterations is important (percentage of failing iterations), but only for that simulation set to get that single year’s (age’s) solution/answer. Iterative simulations should be consecutively performed and serially connected and solutions from each simulation set graphed for decision making during the measuring and monitoring years of retirement. This comparison between paradigm effects is in Frank and Brayman 2016.
Note: The question of what time frame to use for retirement planning was also discussed in this post: “You’d never guess what factor is most important for retirement income!” The time factor (how long a period you calculate for), combined with the dynamics of how life tables (any of them) work as one ages, is the critical difference between the two paradigms (fixed vs iterative)!
5. Allocation glide-paths – Start conservative and glide more aggressive? Or start more aggressive and glide more conservative?
Exhibit 5.23 illustrates the question nicely where allocation differences (and withdrawal rates) are lined up with the time frame (1st column). This is where point #4 above comes into play. Under the present day paradigm as described above and here: If you interpret the first column, labeled “Retirement Duration” as years IN retirement, or years since retiring, and you count the years UP from 1 to 40 (exhibit column going from top to bottom: as is the Monte Carlo graphs displaying 1 on the left end of the bottom axis, and counting the years UP; example Exhibit 6.6 where younger ages are implied to start on the left end of the time axis**), the allocation suggested is to glide the allocation from conservative to aggressive. However, if you interpret the same Exhibit 5.23 from the actuarial paradigm with other factors described above, and count the years from bottom of the graph UP, e.g., from 40 to 1 representing time remaining for retirement, the allocation glidepath goes from aggressive to conservative, a more intuitive result. An opposite allocation glidepath may be possible if the retiree doesn’t need to spend as much and the wealth is allowed to grow beyond spending needs, and thus a more aggressive allocation may be warranted because the time frame for spending isn’t as important a factor because of the excess wealth (i.e., bequest motive greater). Thus, specific retiree circumstances matter.
**The ages in the far right column of Exhibit 5.23, as depicted, represent the actuarial method via the RMD charts and thus properly represent in concept that longer simulation periods correspond to younger ages. Use of a life table would actually stretch the ages out to longer distribution periods; e.g., 40 years of distribution time frame would correspond to someone approximately in the early 50’s (depending on which table used), and time frames less than 10 would correspond to someone in their 90’s and 100’s, also depending on both table used and whether the coefficient of variation adjustment is used for adjusting the table’s percentile (point #2 Mitchell discussion).
Summary.
With so many different approaches existing today, an additional risk becomes “rule selection risk,” (pp216) or stated differently, have you selected the correct rule, and when do you know if and when you should switch rules?
Rather, why not simply redo the calculation/simulation each year updating the data inputs for specific portfolio characteristics (and possible alternative portfolio allocations) with either historical or projected data, updated longevity data being consistent using the same table, or switching life tables if the situation has changed to warrant a switch, and any other factors that should be updated during such an annual review of retirement health (much like is done annually for medical health when retired). Why not model life as we live it, rather than performing a fixed single period simulation and then refer back to that original simulation’s starting point, as is implied through the use of “years IN retirement” graphic displays prevalent in the profession today.
Wade has done a great job summarizing and explaining the retirement income problem as it stands today. A great place to start for those interested in having a deeper understanding of what presently goes into the determination of retirement income. I encourage you to read it to at least get an understanding of what goes into retirement income planning as understood today (2017).
Parts of this review first appeared on Amazon.
