When you Google “retirement planning” you typically get results about 401ks or investing. On the search topic of retirement, you get few results on how to plan for, transition into, and/or prudently sustain income once retired. The term “retirement” brings to mind, and Google finds results on, the savings and investing part. However, the use of the terms “retirement plan,” or “retirement planning,” don’t actually have much to do with the planning part at all! Instead those terms only relate to savings or investing, i.e., an accumulation bias. The “planning” part has to do with questions how much you need to save and how do you spend it down to last the rest of your life? Those questions have little to do with investing, but more with how to save and then use investment resources for overall goals in life, with retirement being the primary focus for most people and other goals as subsets of current lifestyle and retirement lifestyle.
In other words, the term retirement plan totally misses much of the issues. The true application of the term retirement plan, meaning the planning part, has gone through phases historically. I’ll call them paradigms here, since they have to do with how the planning is visualized and computed, for the purposes of this article. This is an important distinction to understand for both consumers and advisers since this planning focus is where the trends in the profession are currently trending, but not yet well supported by software.
We are presently operating in the 2nd paradigm, built upon the 1st paradigm, for retirement planning computations. I’m suggesting there’s a 3rd paradigm, improving upon its’ predecessors, yet to come. However, should paradigms develop into dogmas, or at least rules of thumb, then resistance sets in, and advancements never come, or they come very slowly.
Below are the observations from a researcher, advisor and planner on the frontline, since 1994, nearing 30 years observing. researching, publishing, applying, and advising real people, and living through the paradigms described below.
Upfront: What’s different?
The fundamental paradigm shift into the 3rd paradigm is through how longevity is viewed and applied to the process of aging over time, because longevity is not static with age and aging itself is not static either. They’re both dynamic and this is not captured in today’s static paradigm. That longevity application has a profound effect on cash flows and balances by age. I’ll illustrate this below.
The 3rd paradigm does 2 things … 1) Allows more visibility on cash flows, balances and allocation by age, and 2) provides clearer calculated decision points for spending adjustments scaled both by age and allocation.
The transition between the present 2nd paradigm, discussed briefly below, into the 3rd paradigm is to expand the “box” so “outside the box” becomes inside, to include current thinking (inside the box) by utilizing the stochastic process, i.e., Monte Carlo, more thoroughly in modeling rather than one simulation at a time. What does that mean?
Working years are not one single time period followed by another single time period in retirement. Both phases for people are part of a continuous rolling series of time periods that seamlessly transition from savings to spending at a chosen retirement age, ages chosen ages are different from other retirees. We age through time year by year, but today’s simulations do not account for this slow transition of time periods or between time periods.
The markets and economy don’t changed based on when each person retires; no, they continue to flow over time too. That is also life, with many phases and transitions over time. The same economic and market events happen at different ages for different people because everyone is a different age at the same time (unless you have identical birthdays by day and year). People also retire at different ages. People understand aging and my clinical experience also shows they understand rolling longevity (illustrated below); in fact, they are more comfortable with their plan once they see that longevity bow wave is incorporated.
The present, 2nd paradigm, has a static view where simulations are run over single periods, with single allocations, and where periods and/or allocation are changed one at a time in a new simulation. The problem with the present paradigm is there are no smooth transitions as one ages, nor is there a transition between working and retirement years using the same process to model when that transition may be feasible. That feasibility is again doing single period simulations at different ages to “find” that feasible transition age.
Modeling should reflect life too as a continuum with phases and transitions; yet built upon a single unified and cohesive modeling program that applies to everyone regardless of their age. In other words, each person working or retired simply “moves” through a unified model that models aging and optimum allocations by age as well. The point of retirement is simply a transition from contributing to withdrawing; a point when the sum of the income sources total the sum of the desired target lifestyle expenses. Retirement transition planning then becomes determining when that transition point may occur, or what meaningful adjustments change that target retirement age outcome.
There are many different allocations even if they all seem to be a 60/40 (stock/bond) allocation (or any stock/bond mix other than 60/40). The issue is that few allocations are optimized to be efficient allocations; and worse all those allocations assume they are when they are in truth not. Optimized and efficient allocations leads to the money working as hard as it can to in turn provide optimized prudent and sustainable supplemental income from the portfolio for retirement.
Below is a description of where we came from in terms of retirement planning, where we are in terms of paradigm, and how to evolve and improve insights from financial planning software programming.
Paradigm shifts are often subtle changes in points of view and thinking. This rolling time periods shift in thinking will be explained below, along with the three major paradigms and how software programming, has been, and may continue to be improved. Application of new statistical and mathematical concepts are also discussed. Only by stepping outside of a simulation mindset to get a more three-dimensional perspective on retirement planning can deeper insights come, much like the Gaia mission is doing mapping the Milky Way in 3D. Modeling the data cloud from hundreds of separate simulations, to step outside of any one simulation, provides the much needed 3D view for retirement planning.
What are the three paradigms?
The First Paradigm – The Past
The first paradigm was deterministic in nature. I remember, as many older planners do during their study period to become a Certified Financial Planner PractitionerTM, that retirement planning was done via the HP-12C handheld calculator. One would calculate a rate of return over a given time period to determine how much need to be saved for retirement. It also would do a spend down over a set time period where all the money may be spent by the end of that time frame. It was a deterministic approach using averages, which is a flawed approach.
William Bengen’s seminal paper began the transition from the deterministic paradigm in that it was an initial look at historical returns applied to retirement spend-down of invested money. 1 His paper wasn’t a stochastic application; it was a form of back tested historical evaluation looking back on history to determine how long withdrawals may last, in number of years, by varying the withdrawal rate. This isn’t the stochastic, or in layman’s terms, Monte Carlo we know today. Unfortunately, Bengen’s work developed into a heuristic, or mental shortcut, with other heuristics added to it; rather than to further develop through more thoughtful simulation applications. It’s unfortunate because the nuances and significance of different time periods in his work were underappreciated at best, and ignored at worst. Those time period differences will become apparent below in the longevity “bow wave” discussion.
The issue with heuristics is that they force one by one computations and simulations because they’re short cuts, rather than computing actual values for each and every age, longevity time frame, and allocation possibilities, as discussed below in the third paradigm. In other words, heuristic short cuts also lead to solution shortcuts – now what most people really desire when it comes to something as important as sustainable and prudent retirement income.
However, Bengen’s work brought about insights into the issues of the first paradigm, that resulted in debate, that led to the transition into what I call the second paradigm in which we are today. Advancement and progress between paradigms has been glacial as can be seen from the start of the 2nd paradigm with Bengen’s work in 1994! Business extinctions happen protecting outdated thoughts and processes and clients don’t get answers to their real and deeper questions and concerns. The many applications of advances in other professions over the past two and half decades and more have not happened in the financial planning profession … because they’re not being incorporated into the planning profession.
The Second Paradigm – The Present
The second paradigm is deterministic in nature with a Monte Carlo overlay. It is a hybrid. It is still a “calculation” based approach since it is over a set period with a single set of factors inputted, instead of a model-based approach I describe below. There is a big difference between a simulation, and a model as will be expanded on below. The second paradigm in other words, is a simulation with heuristic, or shortcut, inputs as well – a hybrid method that doesn’t yet model aging. It computes solutions one by one requiring other simulations to compute other possibilities for time spans and/or allocations – a “hunt and peck” methodology seeking answers one age at a time with one portfolio allocation at a time.
Fundamental problem with Monte Carlo over single time periods is the ever-growing uncertainty of possible outcomes in later years that loosely imply age, and reliance on what is known as probability of failure interpreted as plan failure, when it really means iteration failure rate of the simulation’s iterations, whether there are 1,000 simulations, or 5,000 or any number of iterations. One simulation has many iterations within, and a single simulation in not a model, it’s essentially a single calculation repeated multiple times over the same time period with the same allocation and longevity statistics.
Essentially, today’s paradigm, or the second paradigm, is taking a given time period along with a given set of portfolio statistical characteristics, or single factors at one time, and running a Monte Carlo process to derive a “probability of failure.” What that probability of failure really means is the result of how many simulation iterations fail. Probability of failure (or success) is a misnomer today since it implies success or failure of the overall plan, as is how that is commonly interpreted today … thus, implying that the retiree runs out of money. However, this ignores the fact that Social Security, and pensions for those who have them, or properly structured immediate annuity incomes, do not run out. This approach also often results in trying to adjust, or modify, cash flows within the simulation, as opposed to between simulations (discussed below) in order to get “a better” result.
The academic point of view is also based on first or second paradigm views using single periods and allocations, and it is still rare today to read any research of multiple periods and allocations being studied, compared and contrasted to represent changes over time (age).
The fundamental perspective of the second paradigm is to try and solve the length of the simulation period, starting with some age early in retirement, typically age 60 or 65, and and end age typically age 95 plus or minus 5 years.. As will be seen in the third paradigm discussion below, 1) why try to guess as to the end age in the first place when statistics of longevity tables can do that much better than any guess can, and 2) the longevity bow wave discussion below demonstrates that the resolution of that end age continually, and slowly, shifts with age and thus the proper approach would be to continually roll subsequent simulations into the next time period as a retiree ages. Keeping the end age statistically in front of the retiree regardless of attained age, essentially means dollars to fund those future statistical years is also available, through application of a 90% iteration success rate, barring catastrophic spending (10% iteration failure rate – not to be confused by the misapplied term “probability of failure” used today … plan failure should not be confused with a simulation’s iteration failure).
Today’s second paradigm for software programming is redundant as seen in the series of questions below. One needs to rerun the simulation with those different inputs for those different questions (or worse, change those inputs within the same single factor simulation). Software programming should focus on clearer answers to give people better focus, rather than simply accepting greater uncertainties (wider iteration dispersion) over time from redundant simulations that rerun iterations rather than looking deeper into time period and allocation differences using a control variable so outcomes can be directly compared to each other with that common factor (iteration failure rate is that control variable and common factor; not the other inputs which are the variables one wishes to make comparisons).
Hint: the single, one and only, uniquely sensitive variable across age and allocation that signals spending health IS the iteration failure rate. It is misused and ignored in the current paradigm, but is the doorway to the 3rd paradigm. Because drawdown rates change with age (time period), rule of thumb guardrail adjustments aren’t consistent for those time periods because they haven’t been researched for those older age time periods. Instead, using iteration failure rates, for example 25%, self adjust for age, relative to the baseline failure rate of 10%, as the time periods slowly reduce with age. Why use rules of thumb when each year’s decision points can be specifically calculated? Specifically it is the iteration failure rate that signals health of spending, not the drawdown rate itself or some percentage change in drawdown rate rule of thumb. In fact, the ifr also signals a prudent optimal allocation by age as well.
Running yearly cash flows and yearly balances within a single simulation does not provide the same cash flows and balances that come from cash flows and balances between simulations. The later represents what advisers and people actually do as they age. However, this today is simply redoing the calculation each time when it is time a review is desired. It is NOT looking into the future with better focus; only telling us about today when the simulation is run, how we’re doing today.
The second paradigm today simply answers today’s questions and does little to answer questions about the future. Here are questions people have about retirement that generate anxiety about their futures. What about other time frames (not just 30 years)? Or why pick a “random” end age such as age 90, or 95, or even 100; when there are statistics that should be used, to determine that “end age” for simulation purposes (for people who continue to live as they continue to age)? If one uses a “fixed” age, and recognizing that age will be different for a 60 year old than an 80 year old, when and how does one transition from the younger age to the older age to end the simulation period? Why not establish a transitioning protocol from the very beginning? What about other portfolio characteristics; other than the statistical values used in the simulation? Other portfolios have different statistical characteristics; shouldn’t those be evaluated too? And thus, at what later age should the portfolio be adjusted from the initial allocation, to a later allocation, and even later, as one ages? What’s a prudent gross cash flow in the first place? What does the statistical range of spending throughout retirement look like, regardless of age now, or in the future? How can statistical spending characteristics of retirees be compared by select spending categories and cast by age into the future for each retiree by model? (Hint: Keelan metalogs).
Single simulations, with single factors as inputs, don’t address the question “Compared to what?” Simulation failure rates do not represent retirement income failure. Single simulations at one age do not represent simulations at a later age. In sum, how may the uncertain future be brought into better focus? Today’s software comes nowhere near answering deeper questions, which has been an issue for a long time now (Kitces).
The Third Paradigm – What Should Be
These questions lead to what I’m suggesting 2 would be the third paradigm that would shift and transition current thought and approaches from simply applying a stochastic process over deterministic inputs: a set time period, and set allocation, that currently exist in the second paradigm today. Aging is also viewed in a set and static manner today.
In other words, what is needed is a transition to model retirement across all the future ages a retiree might experience, based on strategic use of longevity table percentiles by age, as well as applying the stochastic process to the statistical characteristics of many optimized portfolios so the software suggests what allocation should be considered by age.
This approach begins with having an idea of what the optimal and sustainable annual gross cash flow may be, by age, in the first place, and then making adjustments for taxes, etc. for a net cash flow. Balances are affected by the gross spending from the overall portfolio, while retiree spending is net after taxes.
Below is an example from prototype software illustrating cash flows and balances based on optimal, or prudent and sustainable, spending may look like computed at each age:
Adjustments can be made to that initial model, to see what a minimum spending, or floor level, may look like in the model. A maximum spending, or ceiling level, may also be evaluated where more insight may be gained by seeing at what age(s) cash flows begin to drop off (a spending floor is set), or portfolio balances begin to balloon (a spending ceiling is set). Below is an adjustment to the above example, but now with a set ceiling, or constrained spending (constrained to 5% drawdown rate):
Notice how the cash flows and balances, in the lower graph set above, now resemble more today’s single period longer term wide Monte Carlo simulation cone shaped dispersion patterns between the simulation percentiles (between 75th or consistently poor returns, and 25th percentiles or consistently good returns) … as compared to the upper graph set. This is because the ceiling, when set at a “low value,” soon becomes the floor spending as well. Since floor and ceiling spending are the same, the ever-widening dispersion at later ages results. 2nd paradigm graphs common now show an ever widening uncertainty between upper and lower iterations, rather than more focused by age range of future visibility for better decision making today as shown in the upper graph. Wider dispersion equals more and more future uncertainty which behaviorally makes clients less comfortable – the opposite of what the profession should provide!
This ever widening dispersion, or cone shape, is also what creates the illusion sequence goes away over time and is simply because spending is constrained. Is that the retiree’s goal to restrain spending? Inadvertently this means over saving for retirement in the first place as well as tilting the goal to heir bequests. Is that the retiree’s goal? You can see how a deeper discussion about the present versus the future come about with modeling versus single simulations. Sequence risk is always present at all ages since it happens concurrently (at the same time) for everyone invested in the markets regardless of age (and any firm investing in any market). It’s not sequence risk; rather it’s better thought of as Downturn Risk.
Modeling provides focus and insights that measure and monitor sustainable and prudent consumption as long as one lives. Modeling also provides insights into meaningful measures and adjustments to take when the markets take a downturn.
These two sets of example graphs demonstrate the tradeoff between spending (upper graph) versus balance retention in the lower graph (for heirs? or one’s own higher spending at later ages). In other words, in the spending constrained (spending ceiling) graph, the outcome is higher balances, as well as possibly a shift to greater equity allocations, resulting in more for later spending or for heirs, or both. Is that the retiree’s goal? Different answers for different retirees depending on their individual and unique druthers; that should be each retiree’s choice after discussion.
A later illustration below expands on these concepts and shows how age, shifting longevity, and allocation characteristics slowly transform “the picture” of cash flows and balances as one ages.
For now, suffice it to say there’s a direct linkage between spending, balances, and allocations by age that can be modeled for better clarity for advisers and clients. That linkage may be seen in the below illustrations.
Note the cash flow graph on the left has a “Conceptual” line which represents the concept of level consistent spending, and a “Real People” line which represents evidence of spending patterns (declines) as one ages (Blanchett). The prototype software demonstrates it is possible to compare all of those by age for better clarity for discussion and retiree decision making as they age.
Consumption smoothing of cash flows is possible when done age by age showing cash flows after lumpy expenditures for future spending post lumpy expenditures. How might a deferred annuity income at a later age compare to portfolio income at that age? Note that purchase of a deferred annuity also means that purchase amount is not presently available in the meantime – so what is the lost income between now and the deferred age of that purchase, compared to not having purchased the deferred annuity in the first place?
Modeling cash flows between simulations focuses more clarity on what future cash flow ranges may look like through using simulation percentiles to illustrate upper (good markets), median, and lower (poor markets) cash flows as well as future possible portfolio balances. 3, 4 Note too, that Social Security and/or pensions would be included as part of those cash flows. Impacts on survivor cash flows (reductions in income from Social Security rules, or poor pension options or choices) are also seen with more clarity too.
There is no plan “failure” other than more refined degrees of funding compared to client defined desires such as spending floors, ceilings, or adjustable spending relative to consumer spending statistics. Failure is an old term of the 2nd paradigm, referring to simulation iterations, and not part of the vocabulary of the 3rd.
Modeling in this paradigm would also provide more clarity to sporadic cash flows, such as buying a new car later in retirement – Cash or loan? – how are cash flow and balances affected by using either a loan to purchase, or a lump sum cash withdrawal for purchase.
Today’s planners are unaware of what they could ask for as far as better modeling software for income, or cash flows, and balances. Cash flows and goals can be integrated better. Rather than running redundant simulations, all the simulations should be run one time and the data archived for programming use (easily done if software programming was focused on doing this task). Planning software programming could then be refocused on iteration percentiles representing consistently good markets, median markets, or consistently poor markets so as to better visualize the range of possible cash flows and balances outcomes by age, out to the end ages of longevity tables, some of which go to 120. In other words, planning software would then focus on what advisers and retirees are really interested in by age in the future in a more narrow and better connected series of cashflows and balances.
This approach results in cash flows and balances by age, even out to the end of longevity tables (some go to age 120). A strategic use of good, median, and poor market simulation percentiles narrows the range of outcomes in the future to those more likely to be experienced. Future reviews continue to narrow those uncertain futures with future updates to statistical data. Many may think why 120 when most don’t live that long. Computing well past ages today illustrates that there is money out to those older ages, barring catastrophic spending. It also demonstrates the need to keep the estate plan up to date, not only from the point of view of the retiree(s), but also from the point of view of the estate planning lawyer as laws, or even the interpretation of those laws, change over time. This removes the worry people have of outliving their money and helps them focus on real needs and wants.
The Longevity Statistics of Aging … The “Longevity Bow Wave”
Rolling use of longevity table statistics based on attained age and strategic use of longevity percentiles results in the retiree(s) money outliving them, since there are always statistical life in the tables based on attained age, at least to 120, which means using the tables means money to fund those statistical years to come is also available, barring catastrophic spending.
An analogy: there is always a bow wave in front of a moving boat. The boat never catches the far end of the bow wave. The bow wave always exists in front of the boat until the boat stops!
Here’s an illustration of the “longevity bow wave” illustrating a strategic use of longevity percentiles. The last page shows the “big picture” of how aging changes the time frame over which simulations need to be run.
Online Sharefile Graphic source – Longevity “Bow Wave.” (click to zoom)
It is the application of the statistical longevity bow wave illustrated above, that results in the cash flow and balances graphed below.
These illustrations are quite different from Monte Carlo simulation illustrations of today because the illustrations here incorporate both “probability of the portfolio” with “probability of the person” (longevity). Recognize as well, there are ten yearly transitions between each illustrated 10 year “snapshots,” but not illustrated specifically, so that the transitional points don’t clutter the illustration. Each of those 10 transitional points represent aging, one year at a time, between the snapshots illustrated.
What is the fundamental difference between the second paradigm we’re presently under, and the third paradigm?
The present, or second, paradigm counts time up numerically. The paradigm’s point of view is “years in retirement,” or “years since retirement.” Result: short time periods first with longer time periods at the end. The third modeling paradigm counts time periods down to model aging. Long periods come first with shorter periods later … as seen in the longevity bow wave illustration above. The third paradigm’s point of view represents aging.
The point of measurement is always the retiree’s present attained age and looks forward into statistical life through a strategic use of the life table percentiles. For example, expected longevity by definition is the 50th percentile. A 60-year-old may use the 50th percentile from the life tables. A 70-year-old may use the 40th percentile where 40% statistically outlive that table derived age. An 80-year-old may use the 30th percentile where 30% statistically outlive that table derived age. This Xth percentile use may be used strategically one year at a time to slowly make it more and more unlikely the statistical age might be outlived. As long as a retiree is alive, there are statistical years ahead that need funding. The statistical bow wave continues ahead of the retiree regardless of age. This is in lieu of the present paradigm’s practice of “picking” some random age or time frame over which to cast that single factor simulation.
An example of aging, and modeling aging, is graphically illustrated below using prototype software. It illustrates why using a set age as a reference is a poor planning process. It illustrates the “bow wave” of aging as long as a retiree is alive, there are statistical years ahead of them requiring supplemental income. Any life table may be used, since all are subsets of the same population, to determine longevity percentiles that are illustrated for each age via percentages who outlive end of simulation determined ages, which set the time period the simulation covers, but in reality the main point is the actual end age keeps moving out as the retiree ages in a rolling time frame and simulation period protocol.
Note that allocation aggressiveness goes down with age, which is really what Kitces/Pfau had in their data on the glidepath research if they hadn’t arranged the data counting years up, but instead of counting years down. Bengen’s own research showed higher drawdown rates for shorter periods. What was missing in the 2nd paradigm was how to order the data; short to long, or long to short?
So how should time periods be considered? This is a key question and the answer is critical to the transition from the 2nd to the 3rd paradigms. This counting years method is another paradigm prevalent in the profession at the moment whereby viewing years SINCE retirement, rather than years remaining in retirement. This counting-years-up methodology is akin to saying “the further back I compare myself to now, the better I am.” The “Red Zone” perspective results from the disconnect that presently exists between saving years and spending down years.
The current paradigm yielding the red zone view can’t connect the transition between the two phases, between the working accumulation years and the retirement spending years, because neither phase is presently age based, but rather is based on a flawed paradigm emerging from counting years up described above … which leads to a view that sequence risk of returns exists only at the time of retirement.
The missing link between working-saving years, and retirement-spending years.
An age-based longevity approach eliminates the so-called retirement red zone … volatility and returns sequences actually exists at all ages. Establishing statistical measures using simulation percentiles to signal good or poor spending, and balance health and sustainability, also come from an age-based model protocol as discussed in the 3rd paradigm. In other words, establishing a longevity, age-based, methodology where rolling time periods based on attained ages each year, is used from the first year (young age) of saving for retirement all the way through to the end of the longevity tables. This methodology connects the saving with the spending phases smoothly and transitions between the phases at any age – simply because the longer time periods are kept in front of the shorter time periods throughout the rolling age longevity based methodology, regardless of age.
Example: “If I were to retire today at age (insert age here: e.g., 40, or any age) , how much would I need to save for it to last throughout retirement as long as I live? How much extra do I need to save to bridge me to when my Social Security (and/or pension) starts? At what age do my savings, Social Security (and/or pension) accumulate sufficiently to retire and sustain my desired lifestyle through retirement. What are the meaningful levers I have to adjust those results, not only until retirement, but also during retirement? What is the dynamic effect on lifestyle and retirement feasibility simply by saving more?”
How about ordering data according to attained age and the longevity, and even longevity percentile, associated with it? As soon as one brings in longevity statistics (incorporated into below illustration) so one is counting years forward while also considering corresponding age-based time periods, the time period numbering protocol for simulations on the time axis reverses, from counting lower to higher, to counting higher to lower because one is using time period lengths corresponding to the time period each age has derived from life tables. In other words, years since retirement puts the long-period simulations at the end of the data series, BUT long-period data series should be placed at the beginning because those are the longer periods, with shorter periods corresponding to older ages, as seen in the age timeline graphed below.
Cash flows and balances by age (not time since retirement, or in retirement) … an illustration
The below aging illustration also shows spending and balances in today’s dollars, using real returns, with desired upside spending, or suggested downside spending, ranges depicted using yearly modeling percentile ranges between 75th and 25th percentiles (consistently good and consistently poor markets) with the median 50th percentile in between. Drift into less than 50th percentiles would signal spending adjustments to retain more shares (selling shares at lower prices means selling more shares to net the same dollars). Rerunning the model program for attained age would suggest to what gross spending amount may be desired. Note that the percentile spread in the prototype software is probably wider than a more pragmatic narrower spread such as the 60th and 40th percentiles. Note also that the “cover” over later ages 1) moves slowly towards older ages to reveal both spending and balances at those older statistically derived ages, 2) may also be uncovered to show retirees those older ages during plan discussions (not uncovered in graphs), and 3) shows as long as there is statistical life possible, cash flows and balances model to “end of the life table” barring catastrophic spending. So the question on longevity estimations commonplace today – why even try to guess how long one might live? – when one can use statistics strategically to keep the longevity “bow wave” continually ahead so one outlives their money, barring catastrophic expenses.
The illustrated concept is to follow the prudent spending percentile range year by year as one ages. Cash flows are gross income before taxes. Different income levels and States would be programmed for net income spending after applicable taxes. Floor or ceiling concepts are not illustrated; only the prudent income as a starting level in the first place. Lumpy, floor or ceiling spending effects would ripple through later ages if applied
In sum, the graphic modeling that most closely resembles the third paradigm below combines both statistical aging (Probability of the Person(s)), along with portfolio statistics (Probability of the Portfolio) as well as consumer spending statistics (Blanchett).
People are more interested in their future than they are in their past on the topic of retirement income for their future; for this year and all the other future years. This future oriented interest exists at all ages and includes pre-retirees saving for retirement as well as retirees spending during retirement. The age-based paradigm, and modeling based on it, smoothly transition people from working to retirement at any age.
Incorporating the strategic use of longevity percentiles concept graphed above results in cash flows and balances graphed below.
Cash Flows and Balances While Aging … with the Longevity “Bow Wave” Incorporated
The illustrated age-based model below recognizes aging, where start ages and end ages shift slowly in a rolling year by year manner as one ages, and that shifting time frame changes simulation results for each age (simply because of the changing time frame for each simulation period), and thus insights for decision making also changes as one ages.
Note in the graph below that the short time periods are at the end of the time series, not at the beginning as viewed by the current 2nd paradigm of counting years up or years in retirement. Also note the changing allocations over time, getting more defensive as one ages (as noted above/below how the data was arranged under a counting-the-years-up paradigm in the Kitces/Pfau study. If their data were to be arranged with short periods last representing aging longevity, the same allocation glide path to more defensive is observed).
Note below how the time periods slide from left to right simply because the shorter periods are at the end of the aging cycle. The prior age modeling has the later age modeling within the boundaries of the 75th and 25th simulation percentiles simply because each age and allocation simulation has been previously run. Just as in the longevity bow wave illustration above, recognize as well, that between each of the 10 year “snapshots” there are 10 annual computational points represented for both cash flows and balances, but not illustrated specifically, so that the transitional points don’t clutter the illustration. Each of those 10 transitional points represent aging, one year at a time, between the snapshots illustrated.
Many people wish to retain their principal throughout retirement. However, how much principal do you need when you’re in your 80s or 90s? Answer: enough to last the rest of your statistical life (which turn into bequests eventually).
Notice the cash flows, unlike today’s 2nd paradigm simulations, that later age projections are near the projections for those ages in the earlier age projections. That is not what today’s 2nd paradigm projections show because they are not modeling aging, but instead are single period simulations. This helps refine modeling cash flow tax impacts as well.
Summary so far …
Planning today has a static perspective with static time periods and considers aging within simulations as sufficient. Cash flows within simulation attempts to age a retiree within that simulation too. But aging occurs between simulations in real life because advisers rerun simulations over time for people; advisers don’t refer to the old simulation to evaluate spending now later on – they do a new one. Thus modeling cash flows between simulations more closely captures aging as shown above. Also, since advisers have more than one client, all at different ages, modeling all ages and allocation characteristics for a “data cloud” of results, software programming power can be more efficiently utilized for modeling present and future optimized cash flows, rather than performing redundant simulations just for today’s age and allocation one at a time for each retiree one at a time.
The third paradigm embraces aging specifically with rolling time periods, each of different lengths of time based on attained ages, uses the longevity table percentiles strategically, and replicates retiree cash flows and balances as they age, so as to model cash flows and balances between simulations since that is what actually happens in real life.
I hope to see the dawn of the third paradigm for retirement planning. We’re firmly in the second paradigm and I hope to see the day for the third. The third paradigm is modeling which brings more clarity and focus on the questions relating to planning for supplementing retirement income from investments, Social Security, and/or pensions in a combined manner. One can look at the “Cash Flows and Balances by Age” illustration above and have a better focus and clarity on the range of possible future outcomes. The ripple effects of small spending changes in the plan can also be seen with better clarity by age as well. As long as a retiree is alive, they have statistical years ahead of them in the “longevity bow wave” to fund and thus balances continue to exist into those years as well. Once the retiree passes, those balances are a moot point for retiree income (though still needed if there is a surviving spouse), and that is when balances become bequests. What might those balances be by age? The third paradigm provides sharper focus on the range of possible answers to that question.
In other words, the first paradigm in retirement income determination was rote arithmetic. The second paradigm added Monte Carlo for returns only, to a rote process (while retaining a rote single factor paradigm) in recognition that market returns are not a single average, but a sequence of random unknown returns. The third paradigm is a deeper embrace of statistics and statistical processes to model in sharper focus retirement cash flows and balances by age as one ages; not to some single randomly selected age, but using the longevity percentiles more strategically so the odds of outlining that rolling age also means the odds of outlining retirement money are low too.
These applied statistical principles also allow for more insightful preretirement planning as well, so that people may have deeper insights and better understanding of how to understand and sustain their lifestyle, to and through retirement. Retirement planning should include a seamless transition between accumulation years while working, and spending years while retired. Today’s 2nd paradigm has a troublesome disconnect between the two phases which is attempted to be explained away by a “Retirement Red Zone” and temporary sequence risk exposures. Both come out of the same disconnected joining of these two phases; and both are eliminated when an age-based approach is used for both phases.
Better Focus … An analogy
The Hubble telescope, a tool that simply changed location of the telescope as a tool, brought deeper insights by changing location from earth’s surface to space. The Hubble could look deeper into the past with much better clarity. The 3rd paradigm is simply repositioning the same Monte Carlo tool to look deeper into the future, in contrast to the capability today Monte Carlo provides with just one simulation at a time, thus resulting in better clarity into answering real people’s deeper questions, anxieties, and concerns! The 3rd paradigm application of Monte Carlo is to do “all” the simulations at once up front, and using that output more efficiently where programming looks over the optimum simulation result by age and strings those cash flows and balances together from attained age all the way to the end of the longevity table in use (some go out to age 120).
I suggest deeper insights into retirement would come from a different paradigm, i.e., programming, that changes simulations from single factor simulations to all the factors simulations through the use of “data clouds” through which choices may be better compared by age (time periods that correspond to longevity derived periods). The result is cash flow and balance modeling, rather than one simulation at a time, age by age, when one reaches a later age.
Modeling is the same tool, i.e., Monte Carlo, however the tool is deployed more strategically to bring more focus on a narrower range of future outcomes through strategic use of simulation percentiles as opposed to those percentiles radiating out of one single simulation over one single time period. This is possible by using iteration failure rates as a control variable so that all simulations can be compared to each other through the use of that control variable. The paradigm shift includes the realization that iteration failure rates are not probability of failure as presently interpreted.
The third paradigm, when it comes, will bring more focus and clarity to retirement planning by age. At the moment, the profession is stuck in the one-by-one simulation paradigm over a single fixed time frame, of whatever randomly chosen time period. It will take “Gordion Knot,” with outside-of-the-box thinking and approaches, different from present profession’s and software programmer’s thought processes. There is no “silo” within either side of the profession to advance modeling as a methodology since silos are compartmentalized within, and between programming, planning, investing and advising professions.
Silos aren’t only within and between different functional expertise’s, but also within and between organizations, inhibit and prevent advancements through cross pollination of ideas. A cross silo process should be developed to address moving the multi-profession disciplines of planning and programming into the statistical and age-based modeling paradigm; and the transition process should begin with advisors asking the programmers to develop and support the statistical age-based modeling paradigm.
To date, no one is thinking in terms of the third paradigm – which goes deeper into answering real retiree questions about their future at various future ages, as well as what heirs may be looking at, if that’s a retiree goal too, by future age. Modeling shows how one can spend a dollar once – so is that dollar spent during early ages, or retained for future ages. This means there’s a direct connection between spending and balances at all ages and modeling illustrates the effect of spending on balances at all ages. Just as in the spending years, where each spending year is simulated separately from the others over its’ unique time period (comparing each allocation to determine the optimal cashflow), each accumulation year is also simulated separately from the others over its’ unique time period and summed over selected simulation percentiles around the median percentile with allocation comparisons too (of research interest on this accumulation point – are the “end of year one” simulations the same over all working year simulations, or do they too slowly adjust). Final note on the accumulation years: simulations simulate the fact that each year’s worth of savings grows to a different sum because they have different time frames to grow in the first place … i.e., the first contribution grows more than the last simply because of time, even with the very same rate of return … so why not simulate each year separately too (Probability Management SIPmath demonstrates that statistics sum too)? (Those who might argue that monthly contributions would grow differently than annually – true; however, the volatility of returns would wash over that nuance much like tides washing away foot prints in the sand, so an annual modeling approximation with percentile evaluation would work).
What is missing in today’s programming and approaches is a more focused method to measure and monitor spending health (cash flows and balances) through retirement, including a method to evaluate and compare how future spending effects the present or how present spending ripple effects into the future. Also missing is the accumulation years and transition into the spending years, let alone at what age that transition may be expected and insights into what meaningful adjustments may change that outlook, in other words a “retirement feasibility timeline,” where the ripple effect of adjustments by age are shown graphically without having to separate simulations with different assumptions.
So what’s to do?
The need for better retirement income planning – solved with a age-based model focus.
Science, and its’ application, in all the professions have advanced over the past decades. The science, and software programming to support it, for retirement income planning has advanced little in comparison. Dr’s didn’t invent lasers, but they use them; the same with robotics and many other advancements. Other professions have done the same by looking outside their own organizational and corporate silos into other disciplines that at first blush appear unrelated. Part of the issue is that advisers, and consumers, are unaware that they should ask for more modernized software programming based on modern statistical applications.
It is time to model aging and spending, in not only thought, but practice supported by software programming, and bring the modern versions of science and mathematics of statistics 5,6 into the sphere of retirement planning understanding that the retirement timeframe is not a fixed or static, single factor, timeframe, but rather a connected series of rolling time frames of aging. Modeling should reflect those connected rolling time frames of aging. Retirees age through life with prior spending decisions affecting subsequent spending and balances.
The same can be said about utilizing consumer spending statistics (Blanchett) by age and category so retirees can have a statistical point of reference by age what they may need to plan for looking ahead (the boat’s bow wave). “The results of the analysis suggest that although the retiree consumption basket is likely to increase at a rate that is faster than general inflation, actual retiree spending tends to decline in retirement in real terms. This decrease in real consumption averages approximately 1 percent per year during retirement. A ‘retirement spending smile’ effect is noted. This finding has important implications when estimating retirement withdrawal rates and determining optimal spending strategies.”7
What’s missing for the third paradigm to come to fruition is a much fuller embrace of statistics and their strategic percentile use, by age, for each set of statistics relevant to deeper insights with a better focus on a narrower range of possible outcomes into retirement measurements and monitoring, not only up to and at the moment of retirement (when might that be?), but throughout retirement regardless of age as well.
A true model goes from the early ages, up to and through the transition into retirement, and using the same modeling methodology, throughout all the statistically possible ages in retirement as well.
How about broadening and deepening the retirement accumulation and spending discussions with modeling that illustrates the above concepts in simple illustrations … a picture says 1,000 words.
All of the above is only possible IF the necessary group of people across various professions involved get together, outside their organizational silos, 8 to make it happen. Otherwise, the hunt and peck planning of the 2nd paradigm as it exists today will continue ad infinitum.
Concepts and statistics are all available now. Who is going to bring it all together as a useful tool for advisers? Can the dogma of the 2nd paradigm be overcome to shift to the 3rd? Where is the next paradigm shift going to start? Who (what combination of groups) will start it? When?
New insights come from changing old thoughts and understanding.
Nobody has yet applied the concepts and principles of the 3rd paradigm (though Ken Steiner, a retired actuary, has been a proponent, albeit through a deterministic actuarial process). Nobody has yet disproved the age-based application of longevity statistics and strategic use of their percentiles either. The dogma of the 2nd paradigm is too imbedded and strongly held. So the issues with the second paradigm persist, primarily the second paradigm has no cohesive methodology that connects and smoothly transitions from the working accumulation phase with the retirement spending phase. The age based methodology discussed above in the third paradigm does just that since it uses the same methodology from the first savings year while working, through the last spending year in retirement (and also provides a sharper focus on potential bequests to heirs at any age as well). The same statistical principles also provide a more mathematically robust method to calculate key decision points, through a strategic application of iteration failure rates, to measure and monitor spending health when markets are stressed; referring to the present conditions rather than a rule of thumb approach measured from the past.
People can more easily adjust their behavior when they can more clearly visualize and see their future as well as how decisions today improve or deteriorate that future.
This is post is suggesting that with more advanced applications as discussed above, software addressing retirement cash flows and balances using statistics from many applicable areas would be an improvement on what is available today (and even more advanced and sophisticated than the prototype programming illustrated in the above examples). Applying more sophisticated models based on a fuller application of statistics would bring more modern capabilities with a better focus to both advisers and their clients. It is past time the financial planning profession advance their thoughts, sciences, and applications as well, just as most other professions have done the past 3 decades when the second paradigm shift began and the dogma of the first paradigm shifted.
The profession is slowly being left behind while the applicable sciences useful to the profession advance. Worse, left behind because the applicable sciences are not integrated into a model relevant to aging and the issue of measuring and monitoring the health of retiree spending. Processes, not products, support the modeling approach since processes are adaptable, while products once bought by a retiree are not. I’ll be bold and state that the nature of planning won’t advance without a sharper focus on modeling aging itself, simply because that is retirement.
The longer to adopting a new paradigm, the further behind the profession falls as the speed of change accelerates. 9
Research and application evidence:
1 Bengen, William P. 1994. “Determining Withdrawal Rates Using Historical Data.” Journal of Financial Planning October 1994.
3 Frank, Larry R., and Shawn Brayman. 2016. “Combining Stochastic Simulations and Actuarial Withdrawals into One Model.” Journal of Financial Planning 29 (11): 44–53.
4 Mitchell, John B. 2010. “A Modified Life Expectancy Approach to Withdrawal Rate Management.” Presented at Academy of Financial Services annual meeting, Denver, Colorado. www.ssrn.com/abstract=1703948 . The working paper for reference 3 above.
7 Blanchett, David. 2014. “Exploring the Retirement Consumption Puzzle.” Journal of Financial Planning 27 (5): 34–42. Consumer spending trends with age.
8 Heath, Dan. 2020. “Upstream: The Quest to Solve Problems Before They Happen.” On how to get organization silos to work better together, both within and between organizations. There are many silos, or “firehouses,” within, and between, organization that need to come together to form a “new way” of thinking and operating.
A possible 4th Paradigm to emerge from the adoption of the 3rd where artificial intelligence, mimicking the human manual programming, statistical data gathering, and data processing using AI’s subset of machine learning to perform the desired cash flow and balances, and multiple simulations to form the data by age as described above in the 3rd paradigm, i.e., applying the broader concepts of artificial intelligence “big data.” The 4th Paradigm gathers the statistical data and synthesizes it into the programming the human programmers continue to refine programming for human advisers to continue to apply insights for human decision making. In other words, statistical data gathering and initial processing of consumer spending, portfolio statistics, longevity statistics, etc., so humans don’t need to perform those data gathering functions. Humans free up time to focus on final data refinements and programming updates and refinements. The programmer also refines programming hand in hand with adviser feedback working with clients and changing tax laws, etc. The role of the advisor would be more focused on bigger picture customizing, reviewing, planning contingency options, and overall advising of the plan with the clients as they age. In other words, a stronger move towards advising and planning for people. But, it’s tempting to skip so the profession has to go through and refine the third before the fourth paradigm is possible.
Note: Characteristics and assumptions archived separately.